Abstract
With the urgency of mitigating global warming, the low-carbon transformation of power grid systems has emerged as a pivotal industry upgrade for sustainable development. We proposed a novel deep neural network-based approach for investment decision planning in the low-carbon transformation of power grids, which aimed to address multidimensional key indicators related to power grid transformation and provided reliable electricity industry layouts and investment plans for power system investment decisions. To achieve this, three targeted investment branch models were established, encompassing investment behavior, electricity production and consumption, and predictions of new capacity investment. These models effectively tackled challenges associated with power distribution, electricity price scheduling, power carbon quotas, and the feasibility of low-carbon power generation technologies. Subsequently, a global investment decision planning model was constructed, employing spatiotemporal neural networks and recurrent neural networks, which integrated the aforementioned branch models and incorporated existing low-carbon transformation data. A comparative analysis was conducted, examining the predicted results against actual values from three perspectives: power generation portfolio, grid economy, and overall investment decision plans. The results demonstrated the effectiveness of our method in accurately predicting future installed capacity of diverse low-carbon power generation technologies, sustainability indices, and investment returns. Notably, our method achieves an impressive forecasting accuracy of over 90% compared to actual values of investment decision planning over the past 4 years.
1 Introduction
The imperative of global low-carbon energy transformation has emerged as the most critical challenge of the 21st century. Over the past five decades, developed nations have witnessed a five-fold increase in Gross Domestic Product, leading to a substantial surge in energy consumption from 155.22EJ to 556.63EJ, with energy consumption accounting for two-thirds of global carbon dioxide emissions [1, 2]. This alarming emission of carbon dioxide has significantly contributed to the current predicament of global warming, thereby necessitating worldwide environmental policies to focus on carbon emission reduction. Understanding the intricate relationship between economic development and energy consumption, along with promoting energy efficiency through low-carbon means, is pivotal in achieving sustainable development [3, 4]. Electricity plays a paramount role in modern economies, with its share in national energy production and consumption progressively increasing in tandem with the varying levels of economic development across countries [5, 6]. As per capita household income continues to rise and the reliance on electricity-driven transportation and digital connectivity grows incessantly, electricity demand is predicted to nearly sextuple in the next decade [7]. Energy stands as an indispensable requirement for sustainable development, underscoring the critical objective of selecting low-carbon and green energy sources in national energy industry transformations [8].
As a pivotal component of the global energy landscape, the power sector has dedicated substantial endeavors to effectuate the shift towards low-carbon power, yielding commendable advancements [9, 10]. Concurrently, numerous developing nations have progressively implemented standardized protocols for managing low-carbon power energy and formulated corresponding legislation and policies to foster low-carbon environments [11]. Looking ahead, the successful attainment of low-carbon transformation within the power system necessitates the greater integration of renewable energy sources [12, 13], alongside substantial investments in transmission networks and energy storage infrastructure. Nevertheless, the optimization of these investment strategies mandates the utilization of sophisticated mathematical models [14]. Throughout the low-carbon transformation process, the traditional power grid must contend with diverse sources of uncertainty, encompassing demand growth, the deployment of distributed energy resources, and the penetration rates of renewable energy. These uncertainties impose computational burdens on the planning efforts for low-carbon transformation [15]. Moreover, energy storage facilities assume a pivotal role in the transition towards a low-carbon power grid, encompassing batteries, pumped hydro storage, flywheels, and other technologies. To ensure their efficacy, appropriate planning weights must be assigned to these low-carbon storage solutions, taking into account their operational intricacies and thereby amplifying the overall complexity. In conjunction with long-term uncertainties, investment decisions pertaining to low-carbon grid transformation pose substantial challenges [16].
Recent cases of low-carbon transformation in the power sector indicate a gradual shift in investment decisions from government policy-driven approaches to decentralized corporate investments. The factors influencing power generation companies’ investment decisions have significant implications for the long-term low-carbon transformation of the electricity industry. Key factors include the grid’s risk preference in terms of profit and loss, as well as preferences for optimization technologies [17, 18]. These preferences can lead to investment paths that deviate from conventional estimates. Scholars have endeavored to simulate the uncertainties and risks associated with grid investments in low-carbon transformation, proposing a risk-neutral assumption and estimating that investment decision weights tend to align with grid companies’ risk preferences [19]. Furthermore, researchers have developed an upper-level planning model for the grid that considers reducing carbon dioxide emissions, offering quantitative and directional suggestions for carbon emission policies and low-carbon transformation in the grid [20]. Addressing electricity price and capacity expansion issues, certain research teams have introduced multistage stochastic optimization models to explore the trade-off among low-carbon electricity, electricity prices, and expected profit risks. They also conducted sensitivity analyses on risk weighting factors and budget amounts [21]. In order to aid investment decisions with dynamic risk quantification, some researchers have employed deterministic models to assess the success rates of low-carbon transformation cases in practice [22]. Others have utilized stochastic models to enhance cost control and quantify investment risks in low-carbon transformation [23, 24]. Researchers have proposed distributed renewable energy system planning based on machine learning techniques, leveraging advanced artificial intelligence for key planning methods in distributed low-carbon grids [25]. The study unveils the remarkable performance of machine learning techniques in analyzing data pertaining to low-carbon grids.
The utilization of deep neural networks in facilitating the transition towards low-carbon grids has demonstrated profound efficacy. To simulate carbon emissions for diverse transition scenarios and forecast the weights crucial for sustainable low-carbon grid development, researchers introduce an input–output-based Bayesian neural network method [26]. In this investigation, a convolutional neural network (CNN) model is introduced to enable swift reconstruction of distribution networks. It leverages a switch dataset crafted upon system load models and corresponding optimization combinations. By extracting information from the collected data and making switch decisions for various network states, the CNN contributes to enhanced performance. Validation of the proposed CNN model for distribution network reconstruction is conducted on the IEEE 33-bus system [27]. Furthermore, the researchers in reference [28] develop an input–output-based Bayesian neural network (IO-BNN) method to gauge the potential for carbon reduction across diverse scenarios. Through the integration of the input–output (IO) framework and Bayesian neural network (BNN) within the comprehensive framework, IO-BNN uncovers the influence of major economic sectors and energy consumption factors on carbon emissions while predicting future carbon emissions under multiple scenarios. Deep neural networks possess the capability to effectively tackle high-dimensional problems and encompass various uncertainties inherent in the progression towards low-carbon grids, thereby furnishing valuable data support for sustainable grid investment decisions [26, 29].
The analysis of decision-making processes for low-carbon transformation in the grid is confronted with the challenge of handling abundant multidimensional data, which limits the comprehensive consideration of all global influencing factors and diminishes the reference value of decision-making solutions. To address this challenge, our study explores the utilization of collaborative deep neural networks to assimilate information from diverse decision factors, enabling a holistic assessment of the feasibility and sustainability of low-carbon transformation. This approach aims to provide a comprehensive investment plan for grid transformation, optimizing the combination of low-carbon power generation for high efficiency. The subsequent sections of this article are organized as follows: Section 2 outlines the implementation principles and processes of targeted branch models, along with the integration of deep neural networks in the low-carbon transformation of the grid. Section 3 analyzes the power generation portfolio and the economic outcomes of the grid, thereby validating the performance of the deep neural network model. Finally, Section 4 concludes the research process and highlights the innovative aspects of this study.
2 Method
2.1 Investment behavior modeling
Investment decisions in the power sector involve multiple stakeholders, including power generation companies, the government, electricity consumers, grid companies, and the environment. However, the implementation of low-carbon policies in the electricity market imposes constraints on investment decisions, necessitating consideration of low-carbon transformation. To capture the reality of investment decisions, we draw upon optimization methods discussed in the literature [30, 31]. Leveraging collaborative spatio-temporal neural networks, we optimize the scale of investment and resource allocation for low-carbon transformation in the grid, thereby promoting the harmonization of diverse low-carbon power technologies defined by power generation companies in the transmission and storage components of the electricity system. Moreover, we employ Fick’s law to model the relationship between the efficiency of clean energy generation and the cost of electricity [32], integrating it with optimal resource allocation to plan the distribution of low-carbon electricity post-transformation. The proposed model satisfies the following mathematical equations.
$$ \begin{equation} \frac{\partial{C}_i}{\partial t}={D}_i\frac{\partial^2{C}_i}{\partial{x}^2} \end{equation} $$
(1)
where |$i=1,2,...,n$|, |$t>0$|, and|${x}_{i-1}<x<{x}_i$|. |${C}_i$| and |${D}_i$|, respectively, represent the carbon quota and new production capacity of the ith low-carbon power solution in the market, and |$x$| represents the completion time of low-carbon power technology. When planning investment strategies, initial conditions and boundary conditions need to be set in advance. The power generation mix, carbon emissions, electricity prices of the first iteration satisfy the following equation expressions.
$$\begin{equation} {C}_1\left({t}_0,x\right)={C}_{1,0}(x);\left[0\le x\le{x}_1\right] \end{equation}$$
(2)
$$\begin{equation} {C}_i{l}_{x_i}=\frac{C_{i+1}{l}_{x_i}}{K_{i,i+1}} \end{equation}$$
(3)
$$ \begin{equation} {\left.{D}_n\frac{\partial{C}_n}{\partial x}\right|}_{x=H}={k}_L\left({K}_{pf}{C}_f-{C}_n{l}_{x=H}\right) \end{equation} $$
(4)
where |${l}_{x_i}$| represents the carbon emission coefficient, |${D}_n$| represents the levelized electricity price cost, |${k}_L$| represents the electricity market price fluctuation coefficient, |${K}_{pf}$| represents the electricity price control parameter, which is generally a fixed value, and |${C}_f$| represents the power generation derating coefficient. The wealth of data pertaining to power generation and grid consumption provides valuable insights into the technical and economic landscape of the power industry on an annual basis. The investments made in new power capacity and the modernization of aging power systems play a critical role in determining the annual allocation of resources for new power plants and the strategic planning of operational power plants for the subsequent year. Moreover, the allocation of carbon quotas and subsidies in support of low-carbon policies acts as a crucial mechanism to ensure the profitability and stability of the power system following its transition to a low-carbon framework. These boundary conditions serve as fundamental elements for modeling investment behavior in the context of the grid’s low-carbon transformation. The intricate interplay among various investment factors exerts a significant influence on investment decisions, as depicted in Fig. 1.
Figure 1
Grid investment factor correlation.
Open in new tabDownload slide
2.2 Modeling electricity production and consumption
The power generation phase involves crucial considerations such as the selection of clean energy sources, optimization of generation equipment, and effective cost control in investment decision-making for power producers. Prior to establishing electricity prices, power generation companies submit their electricity demand to grid companies and engage in bidding and production planning processes. When making investment decisions, they prioritize energy-saving dispatch requirements based on the electricity market and low-carbon policies, subsequently determining the dispatch electricity price for power plants. To ensure generation volume and carbon allocation quotas, we employ the Arrhenius approximation modeling method [33] to transform multidimensional independent variables, including electricity demand, bidding principles, regional economic disparities, and retail electricity prices, into dependent variables. This approach maximizes the range of price dispatch areas. The mathematical expression for this methodology is as follows:
$$\begin{equation} {D}_i(T)={D}_i\left({T}_{\mathrm{ref}}\right)\exp \left(-\frac{E_{A,i}}{R}\left(\frac{1}{T}-\frac{1}{T_{\mathrm{ref}}}\right)\right) \end{equation}$$
(5)
$$\begin{equation} {r}_T=\Big\{{\displaystyle \begin{array}{c}f(T),{T}_1\le T\le{T}_2\\{}0,{T}_1>T\;\mathrm{or}\;T>{T}_2\end{array}} \end{equation}$$
(6)
$$ \begin{align}& f(T)=\nonumber\\&\frac{\left(T-{T}_1\right){\left(T-{T}_2\right)}^2}{\left({T}_{\ast }-{T}_1\right)\left[\left({T}_{\ast }-{T}_1\right)\left(T-{T}_{\ast}\right)-\left({T}_{\ast }-{T}_2\right)\left({T}_{\ast }+{T}_1-2T\right)\right]} \end{align} $$
(7)
where |${E}_{A,i}$|represents the activation function in Arrhenius modeling, R represents the carbon quota coefficient, |${r}_T$| represents the electricity demand guarantee rate within the bidding period |$T$|, |${T}_1$| represents the lower limit of the boundary condition, |${T}_2$| represents the upper limit of the boundary condition, and |${T}_{\ast }$| represents any random period between |${T}_1$| and |${T}_2$|, and the other model parameters are shown in Table 1. The results reveal that the majority of the free variables pertain to interdependent co-action parameters, while certain parameters are directly derived from research findings documented in other references. In determining the boundary conditions and scope for retail power production prices, we will judiciously modify the mutation coefficient considering regional economic disparities and effectively train the factors of the investment strategy for electricity price scheduling in a multi-layer configuration. The specific parameters employed in the Arrhenius approximate modeling method are comprehensively presented in Table 1.
Table 1
Open in new tab
Arrhenius model parameters
| Parameter | Value | Units |
|---|---|---|
| |$R$| | 7.32 | Carbon quota coefficient |
| |$a$| | 76 | Retail price of power production |
| |${K}_{\mathrm{pf}}$| | |$4\times{10}^{-4}$| | Regional economic differences |
| |${T}_{\mathrm{ref}}$| | 287 | Electricity price dispatch coefficient |
| |${E}_{A, pp}$| | |$1.45\times{10}^6$| | Investment decision weight |
| |${K}_L$| | |$3\times{10}^{-5}$| | Economic difference coefficient |
| Parameter | Value | Units |
|---|---|---|
| |$R$| | 7.32 | Carbon quota coefficient |
| |$a$| | 76 | Retail price of power production |
| |${K}_{\mathrm{pf}}$| | |$4\times{10}^{-4}$| | Regional economic differences |
| |${T}_{\mathrm{ref}}$| | 287 | Electricity price dispatch coefficient |
| |${E}_{A, pp}$| | |$1.45\times{10}^6$| | Investment decision weight |
| |${K}_L$| | |$3\times{10}^{-5}$| | Economic difference coefficient |
Table 1
Open in new tab
Arrhenius model parameters
| Parameter | Value | Units |
|---|---|---|
| |$R$| | 7.32 | Carbon quota coefficient |
| |$a$| | 76 | Retail price of power production |
| |${K}_{\mathrm{pf}}$| | |$4\times{10}^{-4}$| | Regional economic differences |
| |${T}_{\mathrm{ref}}$| | 287 | Electricity price dispatch coefficient |
| |${E}_{A, pp}$| | |$1.45\times{10}^6$| | Investment decision weight |
| |${K}_L$| | |$3\times{10}^{-5}$| | Economic difference coefficient |
| Parameter | Value | Units |
|---|---|---|
| |$R$| | 7.32 | Carbon quota coefficient |
| |$a$| | 76 | Retail price of power production |
| |${K}_{\mathrm{pf}}$| | |$4\times{10}^{-4}$| | Regional economic differences |
| |${T}_{\mathrm{ref}}$| | 287 | Electricity price dispatch coefficient |
| |${E}_{A, pp}$| | |$1.45\times{10}^6$| | Investment decision weight |
| |${K}_L$| | |$3\times{10}^{-5}$| | Economic difference coefficient |
In the actual electricity market, grid companies typically determine dispatch plans based on energy-saving dispatch principles within the scope of policy formulation. Therefore, in our study, we assume that all low-carbon power generation satisfies the electricity demand. We consider the number of low-carbon power generation companies actually producing power as |$SQ(t)$| and the power generation plan based on electricity demand as |$Q(t)$|. By taking into account actual bidding and the carbon emissions forecast for coal-fired power generation, we establish the low-carbon power generation bidding index. This satisfies the following mathematical equations.
$$ \begin{equation} SQ(t)=Q(t)\ast{\frac{D^{\mathrm{total}}(t)-S{Q}_{\mathrm{low}-\mathrm{carbon}}(t)}{\sum_i{\sum}_jQ(t)}}{\ast}\frac{{\mathrm{bid}}_{\mathrm{ave}}(t)}{{\mathrm{bid}}_j(t)} \end{equation} $$
(8)
where |${D}^{\mathrm{total}}$| represents the total electricity demand in year t, |$S{Q}_{\mathrm{low}-\mathrm{carbon}}(t)$| represents the power generation plan after low-carbon transformation, |${\mathrm{bid}}_{\mathrm{ave}}(t)$| represents the average cost of electricity bidding, and |${\mathrm{bid}}_j(t)$| represents the expected profit from electricity bidding. In addition to the aforementioned factors, power generation companies also face the constraints imposed by carbon emission policies, which necessitate their prioritization of actual carbon emissions and carbon quotas in the planning of electricity production and supply distribution [34]. In our study, we make the assumption that electricity production and supply distribution adhere to the prescribed carbon quota limits. Each power generation company is considered a bidding participant, thereby establishing an investment return ratio model that encompasses both the power generation and consumption ends.
2.3 New capacity investment forecast
New capacity investment serves as a crucial aspect of the horizontal dimension in the low-carbon transformation of the power industry. Among the range of investment options, commercially viable low-carbon technologies that merit consideration include solar power generation, tidal power generation, wind power generation, nuclear power generation, biomass power generation, supercritical technology power generation, and combined heat and power generation. When modeling supply–demand dynamics within the electricity market, the mathematical relationship between power shortages and new capacity is expressed as follows: [35].
$$ \begin{equation} \frac{D_{i}^{\mathrm{pre}-\mathrm{total}}\left(t+3\right)}{Q^{\mathrm{ser}}(t)+{Q}^{\mathrm{com}}(t)-{Q}^{\mathrm{ret}}(t)}>{\omega}_i \end{equation} $$
(9)
In the equation, |${D}_{i}^{\mathrm{pre}-\mathrm{total}}$| represents the electricity market forecast demand for the ith low-carbon power generation company. All electricity demand data undergo linear regression to ensure the reliability of preliminary trend predictions. |${Q}^{\mathrm{ser}}$|, |${Q}^{\mathrm{com}}$|, and |${Q}^{\mathrm{ret}}$|, respectively, denote the maximum power generation capacity that can be generated by power plants currently in normal operation, power plants under construction, and power plants reaching the end of their lifespan after 3 years. |${\omega}_i$| represents the lower limit of electricity demand set during the bidding period.
Power generation technology plays a pivotal role in shaping investment decisions. Investors rely on data furnished by power generation companies, which encompasses investment return ratios, risk analysis, and anticipated technological advancements associated with various power generation technologies. To compute the investment utility of different technologies, we employ a standard utility function [36]. This function incorporates risk preferences and adaptive coefficients among power generation technologies into the investment utility function, as depicted in the subsequent mathematical equations.
$$ \begin{equation} {U}_{i,j}(t)=\left(1-{e}^{-{\gamma}_i\ast{W}_{i,j}(t)}\right)\ast{\varphi}_i(t) \end{equation} $$
(10)
Where, |${U}_{i,j}(t)$| represents the investment utility of power generation company i in low-carbon power generation technology j, |${W}_{i,j}(t)$| represents the predicted investment return rate of power generation company i in low-carbon power generation technology j, and |${\gamma}_i$| is the Arrow-Pratt risk aversion coefficient. This coefficient is used to measure the risk preference of power companies in electricity production and supply–demand. A higher value of this coefficient indicates a higher investment risk for the power company. |${\varphi}_i(t)$| represents the technological maturity preference of power generation company in low-carbon power generation technology j. Investment strategies exhibit diverse weightings for technological maturity preference. The range of preference for technological maturity is typically established by considering the ratio between the technology’s share in the investment portfolio and the national share of the company for that technology, typically falling within the range of 0.95–1.05 [37]. In evaluating the profitability and profit margin of newly built power plants during mid-term calculations, the preference weight for the current technology can be flexibly adjusted as required, utilizing the following mathematical equations.
$$\begin{equation} {\varphi}_i(t)={\varphi}_{i,j}(t)+\varDelta \varphi \end{equation}$$
(11)
$$ \begin{equation} {W}_{i,j}(t)=\sum_{t\hbox{'}=t}^{T+t}\frac{p^e(t)\ast{Q}^{ser}+{p}^c(t)\ast{Q}^{com}-{p}^r(t){Q}^{ret}}{{\left(1+r\right)}^{t^{\prime}-t}} \end{equation} $$
(12)
When companies embark on technological investments, they encounter a multitude of uncertainties stemming from market demand, competitive landscape, technological advancements, regulatory policies, and more. The fluctuations in these factors can significantly impact the returns of such investments. To enhance the accuracy of new capacity investment forecasts and effectively evaluate the risks and returns associated with technological investments, the utilization of Monte Carlo simulation and uncertainty modeling approaches has proven beneficial [38, 39]. These methods involve generating a large number of random samples to simulate various uncertain future scenarios. In the context of technological investments, companies can define multiple future conditions, such as market size, competitors’ actions, and technological development pace, and subsequently simulate each scenario accordingly. Through this process, companies can calculate the benefits of technological investments for each scenario, encompassing financial indicators (e.g. investment return rate, net present value) and nonfinancial indicators (e.g. market share growth, brand value enhancement). Simulations also enable companies to assess the risks associated with technological investments, including the likelihood of investment losses under different scenarios.
2.4 Investment decision planning model
The investment decisions of grid enterprises are confronted with both technical uncertainty and economic uncertainty, necessitating a diversified model for planning low-carbon transition investments. Technical uncertainty primarily arises from the inherent variability of renewable energy generation capacity, which depends on natural conditions, making it a supplementary rather than a primary power supply option. On the other hand, economic uncertainty stems from the long-term reliance of the electricity market on fuel prices, leading to significant regional disparities in electricity prices resulting from low-carbon power generation. To ensure a seamless transition in investment decisions concerning low-carbon technologies and the fuel power market, we address the inherent shortcomings in investment decision characteristics caused by different technology types at the decision-making level. As previously mentioned, we divide the grid’s low-carbon transition investment decisions into three distinct decision branches: investment behavior, power production and consumption, and new capacity investment forecasts. Each branch is assigned separate investment decision weights for effective investment planning. The investment behavior decision branch, guided by environmental sustainability, prioritizes carbon emission reduction. The power production and consumption branch, guided by economic benefits, focuses on investment return rates. Lastly, the new capacity investment branch, guided by the feasibility of low-carbon power generation technologies, centers on the practicality of employing low-carbon technologies. The network structure of investment decision planning is visually depicted in Fig. 2.
Figure 2
Neural network framework for global investment decision-making planning in power grid low-carbon transformation.
Open in new tabDownload slide
In order to achieve effective coordination of investment decision planning weights, we employ a recurrent neural network as the underlying framework, leveraging time convolutional networks and spatial convolutional networks for extracting correlated features among the independent decision branches [40]. Following the feature extraction process, the investment behavior branch is integrated into the power production carbon emission prediction layer of the recurrent neural network, thereby providing predictive weights for investment decision planning in terms of environmental sustainability. The power production and consumption prediction branch involve approximating modeling and carbon quota feature extraction, which are then combined with the economic investment return rate prediction layer of the recurrent neural network, resulting in predictive weights for investment decision planning in relation to economic benefits. As for the new capacity investment forecast branch, after conducting uncertainty modeling and selecting low-carbon technology features from different new capacity data, they are integrated with the low-carbon technology efficiency prediction feature of the recurrent neural network, yielding predictive weights for investment decision planning concerning low-carbon power generation technology [41]. By leveraging the independent decision weights from the three branches and the collaborative feature extraction of the recurrent neural network in the economic and technological aspects, the weights of low-carbon grid transformation in investment decision planning can be accurately predicted. This provides essential data support for grid enterprise investments and serves as a valuable reference for investment planning.
3 Results and discussion
The transition period for the grid to embark on long-term low-carbon transformation is influenced by several factors, as discerned through investment behavior analysis, including power generation portfolio, grid economy, and transformation stability. Achieving a well-balanced power generation portfolio enables companies to mitigate the risks associated with fluctuating renewable energy generation. Transitioning from cogeneration to wind power generation, for instance, can result in a 10% reduction in carbon emissions, with the potential for a 30% decrease in total carbon emissions over the next decade. From the perspective of the grid economy, prudent low-carbon investment decisions can lead to a 15% reduction in retail electricity prices and a 12% increase in total revenue for companies. However, it is crucial to consider the technological disparities between developed cities and third- and fourth-tier cities, as this may still give rise to regional disparities in electricity prices. Additionally, by adopting careful decision planning to avoid risks, companies can provide a stable development outlook for the low-carbon transformation of the grid. The analysis and discussion of the results will now be presented based on these three aspects.
3.1 Impact of generation mix on low-carbon transition
The low-carbon transformation of the grid economy profoundly influences the power generation technology mix employed by power generation companies. Economic data indicators for different low-carbon power generation technologies are presented in Table 2, revealing that the combination of various power generation technologies can compensate for the limitations of individual economic indicators. To evaluate the long-term implications of different power generation technology combinations on electricity’s low-carbon transformation, we have established five comparative groups and assigned distinct investment risk levels and risk aversion coefficients to each group, as depicted in Table 3. Leveraging decision analysis neural networks, we can forecast the corresponding shifts in power generation composition, carbon emissions, electricity prices, and investment return ratios for the low-carbon transformation of electricity from 2020 to 2050.
Table 2
Open in new tab
Technical and economic data for different generation technologies
| Standard capacity (MW) | Average capital investment (million USD/MW) | Annual operation and maintaining costs (million USD/MW) | Construction period (year) | Lifespan (year) | |
|---|---|---|---|---|---|
| Solar | 60 | 1.70 | 0.02 | 3 | 30 |
| Wind | 60 | 1.25 | 0.04 | 3 | 25 |
| Hydropower | 500 | 1.16 | 0.02 | 7 | 80 |
| Biomass | 40 | 1.64 | 0.07 | 1 | 20 |
| Nuclear | 1000 | 2.10 | 0.01 | 4 | 50 |
| CHP | 700 | 0.72 | 0.05 | 2 | 35 |
| Standard capacity (MW) | Average capital investment (million USD/MW) | Annual operation and maintaining costs (million USD/MW) | Construction period (year) | Lifespan (year) | |
|---|---|---|---|---|---|
| Solar | 60 | 1.70 | 0.02 | 3 | 30 |
| Wind | 60 | 1.25 | 0.04 | 3 | 25 |
| Hydropower | 500 | 1.16 | 0.02 | 7 | 80 |
| Biomass | 40 | 1.64 | 0.07 | 1 | 20 |
| Nuclear | 1000 | 2.10 | 0.01 | 4 | 50 |
| CHP | 700 | 0.72 | 0.05 | 2 | 35 |
Table 2
Open in new tab
Technical and economic data for different generation technologies
| Standard capacity (MW) | Average capital investment (million USD/MW) | Annual operation and maintaining costs (million USD/MW) | Construction period (year) | Lifespan (year) | |
|---|---|---|---|---|---|
| Solar | 60 | 1.70 | 0.02 | 3 | 30 |
| Wind | 60 | 1.25 | 0.04 | 3 | 25 |
| Hydropower | 500 | 1.16 | 0.02 | 7 | 80 |
| Biomass | 40 | 1.64 | 0.07 | 1 | 20 |
| Nuclear | 1000 | 2.10 | 0.01 | 4 | 50 |
| CHP | 700 | 0.72 | 0.05 | 2 | 35 |
| Standard capacity (MW) | Average capital investment (million USD/MW) | Annual operation and maintaining costs (million USD/MW) | Construction period (year) | Lifespan (year) | |
|---|---|---|---|---|---|
| Solar | 60 | 1.70 | 0.02 | 3 | 30 |
| Wind | 60 | 1.25 | 0.04 | 3 | 25 |
| Hydropower | 500 | 1.16 | 0.02 | 7 | 80 |
| Biomass | 40 | 1.64 | 0.07 | 1 | 20 |
| Nuclear | 1000 | 2.10 | 0.01 | 4 | 50 |
| CHP | 700 | 0.72 | 0.05 | 2 | 35 |
Table 3
Open in new tab
Risk aversion coefficient and technology preference of power generation portfolio investment
| Investment risk level | The Arrow–Pratt risk aversion coefficient | Low carbon technology preference coefficient | |
|---|---|---|---|
| Group 1 | High risk | 2.5 | 0.2 |
| Group 2 | Medium to high risk | 2.0 | 0.4 |
| Group 3 | Medium risk | 1.5 | 0.6 |
| Group 4 | Low to medium risk | 1.0 | 0.8 |
| Group 5 | Low risk | 0.5 | 1.0 |
| Investment risk level | The Arrow–Pratt risk aversion coefficient | Low carbon technology preference coefficient | |
|---|---|---|---|
| Group 1 | High risk | 2.5 | 0.2 |
| Group 2 | Medium to high risk | 2.0 | 0.4 |
| Group 3 | Medium risk | 1.5 | 0.6 |
| Group 4 | Low to medium risk | 1.0 | 0.8 |
| Group 5 | Low risk | 0.5 | 1.0 |
Table 3
Open in new tab
Risk aversion coefficient and technology preference of power generation portfolio investment
| Investment risk level | The Arrow–Pratt risk aversion coefficient | Low carbon technology preference coefficient | |
|---|---|---|---|
| Group 1 | High risk | 2.5 | 0.2 |
| Group 2 | Medium to high risk | 2.0 | 0.4 |
| Group 3 | Medium risk | 1.5 | 0.6 |
| Group 4 | Low to medium risk | 1.0 | 0.8 |
| Group 5 | Low risk | 0.5 | 1.0 |
| Investment risk level | The Arrow–Pratt risk aversion coefficient | Low carbon technology preference coefficient | |
|---|---|---|---|
| Group 1 | High risk | 2.5 | 0.2 |
| Group 2 | Medium to high risk | 2.0 | 0.4 |
| Group 3 | Medium risk | 1.5 | 0.6 |
| Group 4 | Low to medium risk | 1.0 | 0.8 |
| Group 5 | Low risk | 0.5 | 1.0 |
The investment disparities stemming from power generation portfolio choices have a substantial impact on the installed capacity of power plants, as evidenced by the findings presented in Fig. 3. In the high-risk comparative group, the preference for renewable energy sources, such as solar power and wind power, is more pronounced at the level of technological preferences. However, its influence on hydroelectric power, biomass power, and nuclear power technologies is relatively modest. By 2050, it is estimated that the installed capacity of solar power and wind power will rise by 38.65% and 28.33%, respectively, in the high-risk and medium-risk comparative groups. Conversely, in the low-risk grouping, there is a greater inclination towards nuclear power and biomass power generation, leading to a 32.47% decrease in the proportion of renewable energy power generation technologies. In the medium-risk and medium-low-risk groupings, all low-carbon power generation technologies maintain an equal share and growth level.
Figure 3
Impact of installed capacity of low-carbon transformation power plants with different power generation mixes.
Open in new tabDownload slide
Moreover, to further validate the indicators pertaining to the maturity, uncertainty factors, development potential, and economic returns of diverse low-carbon power generation technologies, we employed the analytic hierarchy process. For instance, investments in solar power and wind power have a significant impact on the sustainability and stability of renewable energy supply, while cogeneration investments are highly susceptible to fluctuations in fuel and carbon prices. This transformation is intricately linked to the development potential and economic returns of the respective technologies. By evaluating the transformation cycles of different low-carbon technologies based on sustainability indicators, we can enhance the adjustment of investment decision-making strategies. To facilitate comparison, we have normalized all the data given the multitude of technological indicators, employing a sustainability index as the measurement criterion. A higher sustainability index signifies greater feasibility, investment return rates, and development potential of the corresponding low-carbon power generation technology. The outcomes are visually depicted in Fig. 4.
Figure 4
Sustainability index analysis of different low-carbon power generation technologies.
Open in new tabDownload slide
Based on the insights provided in Fig. 4, it becomes apparent that renewable energy sources are poised to attain maturity at an earlier stage in their development within the next three decades. This suggests that these technologies have the potential to yield stable investment returns ahead of schedule. In contrast, technologies like cogeneration exhibit a slower rate of progress in terms of their sustainability index development. It is projected that they will only meet the prerequisites for sustainable development within the next 30years. Consequently, these technologies, which are contingent upon fuel and carbon prices, will be assigned a relatively lower decision weight in the investment allocation process of the power generation portfolio. Although hydropower is a renewable energy source, its technological maturity falls behind that of biomass power generation. This disparity can be largely attributed to the inherent instability of hydropower, stemming from seasonal variations such as summer floods and winter low-water periods. These fluctuations have a detrimental impact on the stability of hydropower systems. Consequently, the enhancement of hydropower technology undergoes brief annual testing phases, resulting in a prolonged technological cycle and a delayed attainment of maturity. In stark contrast, biomass power generation offers the advantage of year-round opportunities for technological innovation, facilitating more rapid advancements in comparison to hydropower.
3.2 The economic impact of the grid
With the increasing adoption of low-carbon power generation, power companies have the opportunity to reduce generation costs through investments in new capacity, leading to a mitigated impact on regional electricity retail prices. To assess the influence of generation costs and power generation structure, we conducted a comparative analysis of electricity pricing costs for various low-carbon power generation technologies, as depicted in Fig. 5. Clear observations emerge, as generation costs exhibit a consistent downward trend over time, reflecting the evolution of these technologies. Notably, when considering the disparities between renewable energy generation and other low-carbon power generation technologies illustrated in Fig. 4, it becomes apparent that while electricity pricing costs demonstrate a declining trajectory in the forecast, the future deployment scale of cogeneration power generation technologies may diminish due to the economic returns offered by the power generation portfolio. In practical investment decision-making, driven by the augmented presence of biomass power generation and nuclear power generation in power companies’ grid economic planning, the share of low-carbon power generation below a risk aversion coefficient of 1.5 has expanded by nearly 42%. Based on projections from the recurrent neural network encompassing carbon quotas and power generation demand, it is anticipated that by 2050, the share of low-carbon power generation will continue to surge by approximately 70%. However, renewable energy power generation technologies are poised to emerge as the predominant power generation technologies by that time.
Figure 5
Electricity price calculation costs of different low-carbon power generation technologies.
Open in new tabDownload slide
Within the context of investment planning in grid economics, a greater emphasis on low-carbon power generation, aligned with environmental policies, can yield substantial policy subsidies and tax reductions, thereby augmenting the returns on grid investment decisions. The findings delineated in Fig. 4 underscore the gradual ascendance of renewable energy generation technologies as the prevailing force in the low-carbon transition. Consequently, it becomes imperative to analyze the projected investment decision returns for solar power generation, wind power generation, and hydroelectric power generation. Moreover, the risk aversion coefficient serves as a metric to gauge the risk preferences of power companies in relation to electricity production and supply–demand dynamics. A lower value of this coefficient corresponds to higher returns on grid investment decisions. Based on the levels of the risk aversion coefficient, we will employ the medium risk threshold as a demarcation and examine the decision returns below the medium-risk level, as prognosticated by the investment decision planning model. The outcomes are visually depicted in Fig. 6.
Figure 6
Forecast of income from power grid economic investment decisions.
Open in new tabDownload slide
According to the findings presented in Fig. 6, it is projected that retail electricity prices will experience a consistent upward trend by 2025, within the risk aversion coefficient range of 1.5–0.5. This suggests that the power industry can anticipate a growth in investment returns over the next three decades, coupled with a decrease in the risk aversion coefficient associated with investment decisions. Notably, solar power emerges as the most promising option, with a significant projected increase of 23.43% in investment returns. As a result, prioritizing investments in industries aligned with this technology becomes crucial for future power-related endeavors. In contrast, hydroelectric power and wind power face greater uncertainties and risks, necessitating a careful assessment of the investment return proportions in power investment decision-making.
3.3 Investment decision-making and planning in low-carbon transition
In order to facilitate a comprehensive low-carbon transformation in power investments, encompassing the entirety of the power generation portfolio and its associated economic benefits, it is crucial to effectively anticipate industry development uncertainties and potential risks. Of paramount importance for long-term stability in the low-carbon transition and investment planning weighting is the holistic forecast of carbon emissions stemming from the low-carbon power generation portfolio. To assess the performance of the deep neural network in investment decision planning, simulation scenarios were devised for various low-carbon power generation technologies, taking into account investment behavior, power production and consumption, as well as new capacity investments, as depicted in Fig. 7. The distinct nodes within the figure represent the economic investment-return relationships, while sustainability potential predictions were also conducted to gauge the efficacy of the simulation scenarios.
Figure 7
Simulation scheme for forecasting investment decisions of different low-carbon power generation technologies.
Open in new tabDownload slide
The analysis presented in Fig. 7 reveals a conspicuous pattern in the investment decision scenarios generated from renewable energy technology forecasts, manifesting a consistent and predictable variation within the sustainability analysis model. This observation underscores the stable financial and profitability capabilities inherent in these investment decision scenarios, thereby indicating their promising development potential over the ensuing three decades. Conversely, the sustainability analysis model for biomass power generation and nuclear power generation scenarios displays intermittent changes, suggesting reduced stability in terms of investment and returns. Power companies should exercise prudence when evaluating the feasibility of these two simulation scenarios. Additionally, the cogeneration power generation scenario demonstrates commendable stability during the initial decade but exhibits significant fluctuations in the subsequent two decades. Consequently, investors should judiciously consider the merits of the power generation portfolio to offset concerns regarding stability during the later stages of the investment plan.
Figure 7 presents a graphical representation of the correlation between different nodes and economic investment income within the low-carbon grid framework. In this context, nodes represent distinct components or factors that influence economic investment. These nodes could include variables such as renewable energy deployment, policy incentives, technological advancements, regulatory frameworks, market conditions, and other relevant factors. It helps identify critical nodes that have a pronounced impact on investment income and highlights the complex dynamics involved in decision-making processes related to economic investment. It is important to note that the specific nodes and connections in Fig. 7 may vary depending on the study’s focus and methodology. Therefore, the figure should be interpreted within the context of our research, providing valuable insights into the relationship between nodes and economic investment income in the low-carbon grid.
Furthermore, we conducted comprehensive assessments of the deep neural network’s prediction accuracy and recall rate in investment decision planning. To substantiate the efficacy of the predictive data, we employed grid low-carbon transformation data from 2020 to 2023 as a control group, evaluating the degree of correspondence, error rate, and investment controllability within the investment decision planning framework. The outcomes are summarized in Table 4. In terms of correspondence, with the exception of solar power generation and cogeneration, all other low-carbon power generation technologies maintained a matching degree exceeding 80%. The matching degree for solar power generation is relatively low due to the pronounced influence of environmental factors and heightened uncertainty. Constrained by fuel prices and market dynamics, cogeneration exhibited a matching degree of only 72%. Regarding the error rate, it consistently remained below 0.25. In terms of investment controllability, all investment decision projection scenarios demonstrated a controllable range above 80%, adhering to established investment risk control standards. The prediction accuracy hovered around 90%, the recall rate approximated 0.9, and the F1 score surpassed 0.8. These findings underscore the exceptional performance of the deep neural network in forecasting investment decisions for low-carbon transformation in the power sector. It furnishes valuable indicators and potential for predicting investment returns, thereby serving as a crucial reference for investment decision-making.
Table 4
Open in new tab
Product outer packaging design material matching verification
| Matching degree | Error rate | Investment controllable rate | Precision | Recall | F1 score | |
|---|---|---|---|---|---|---|
| Solar | 61% | 0.16 | 89% | 89.32% | 0.91 | 0.83 |
| Wind | 85% | 0.22 | 85% | 92.15% | 0.92 | 0.86 |
| Hydropower | 89% | 0.18 | 86% | 90.64% | 0.91 | 0.85 |
| Biomass | 91% | 0.19 | 82% | 89.94% | 0.91 | 0.83 |
| Nuclear | 83% | 0.21 | 80% | 90.31% | 0.90 | 0.81 |
| CHP | 72% | 0.25 | 81% | 91.46% | 0.93 | 0.80 |
| Matching degree | Error rate | Investment controllable rate | Precision | Recall | F1 score | |
|---|---|---|---|---|---|---|
| Solar | 61% | 0.16 | 89% | 89.32% | 0.91 | 0.83 |
| Wind | 85% | 0.22 | 85% | 92.15% | 0.92 | 0.86 |
| Hydropower | 89% | 0.18 | 86% | 90.64% | 0.91 | 0.85 |
| Biomass | 91% | 0.19 | 82% | 89.94% | 0.91 | 0.83 |
| Nuclear | 83% | 0.21 | 80% | 90.31% | 0.90 | 0.81 |
| CHP | 72% | 0.25 | 81% | 91.46% | 0.93 | 0.80 |
Table 4
Open in new tab
Product outer packaging design material matching verification
| Matching degree | Error rate | Investment controllable rate | Precision | Recall | F1 score | |
|---|---|---|---|---|---|---|
| Solar | 61% | 0.16 | 89% | 89.32% | 0.91 | 0.83 |
| Wind | 85% | 0.22 | 85% | 92.15% | 0.92 | 0.86 |
| Hydropower | 89% | 0.18 | 86% | 90.64% | 0.91 | 0.85 |
| Biomass | 91% | 0.19 | 82% | 89.94% | 0.91 | 0.83 |
| Nuclear | 83% | 0.21 | 80% | 90.31% | 0.90 | 0.81 |
| CHP | 72% | 0.25 | 81% | 91.46% | 0.93 | 0.80 |
| Matching degree | Error rate | Investment controllable rate | Precision | Recall | F1 score | |
|---|---|---|---|---|---|---|
| Solar | 61% | 0.16 | 89% | 89.32% | 0.91 | 0.83 |
| Wind | 85% | 0.22 | 85% | 92.15% | 0.92 | 0.86 |
| Hydropower | 89% | 0.18 | 86% | 90.64% | 0.91 | 0.85 |
| Biomass | 91% | 0.19 | 82% | 89.94% | 0.91 | 0.83 |
| Nuclear | 83% | 0.21 | 80% | 90.31% | 0.90 | 0.81 |
| CHP | 72% | 0.25 | 81% | 91.46% | 0.93 | 0.80 |
3.4. Limitation
Although we have successfully established the correlation between diverse low-carbon power generation technologies and economic indicators using deep neural networks, incorporating power generation companies’ preferences into practical investment decisions becomes imperative for a more comprehensive evaluation of the actual efficiency and cost of low-carbon policies in power grids. Furthermore, different countries implement corresponding policies to support the transition of enterprises towards low-carbon grids, such as tax incentives and equipment subsidies. However, this particular variable remains volatile, leading us to assign uniform values to policy support programs in this study. Lastly, our research focuses solely on currently commercialized low-carbon power generation technologies, neglecting the potential for future technological innovations. To address this, future studies can employ agent-based models to investigate the investment behavior of power generation companies and governments in technological research and development, thereby capturing significant technological breakthroughs that may emerge in the future.
4. Conclusion
This study presented a novel approach utilizing a deep neural network for investment decision planning in the low-carbon transformation of the power grid, with the objective of providing valuable references for the power system’s low-carbon transition. Firstly, we employed investment behavior modeling to simulate the relationship between clean energy generation efficiency and power costs, enabling optimal resource allocation and planning the power distribution problem post low-carbon transformation. Secondly, approximate modeling methods were utilized to establish power production and consumption models, considering factors such as electricity demand planning, bidding principles, regional variations in retail electricity prices, and the decentralization of electricity price scheduling. Uncertainty models were employed to predict the risks and returns associated with new capacity investments in low-carbon power generation technologies, market scale, and technological advancements. Lastly, a collaborative investment decision planning model was developed, incorporating spatiotemporal neural networks and recurrent neural networks, to predict the weights of various indicators in investment decision scenarios for the power grid’s low-carbon transformation.
The results and analysis revealed that the maturity of different low-carbon power generation technologies was influenced by various uncertainties. Optimal combinations of power generation portfolios contribute to stability and returns in power investments, reducing the risks associated with renewable energy generation decisions. Economic investment planning for the power grid necessitated a comprehensive consideration of the correlation between risk aversion coefficients and the share of low-carbon power generation, thereby exploring the maximization of investment returns within acceptable risk levels. Finally, the test results demonstrated the deep neural network’s high accuracy, significant matching degree, and low error rate in predicting investment decisions for the power sector’s low-carbon transformation. This approach exhibited tremendous potential in providing indicators for investment decision-making and predicting investment returns.
The low-carbon transformation of power systems exhibits variations among diverse cities and regions. In pursuit of a more flexible investment decision model, our research methodology focuses on acquiring supplementary low-carbon transformation investment data from alternative regions and training the model through meticulous feature analysis. It is imperative to acknowledge that not all power grid companies presently adhere to the low-carbon transformation prerequisites. Neglecting this boundary condition during the research process may lead to excessively sanguine outcomes. Furthermore, in forthcoming investigations, we intend to explore transfer learning strategies to facilitate the model’s ability to extrapolate to dissimilar cities or regions, thereby augmenting its overall performance.
Author contributions
Min Wang (Conceptualization [equal], Data curation [equal], Formal analysis [Equal], Funding acquisition [equal], Investigation [equal], Methodology [equal], Project administration [equal], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review and editing [equal]), Yixiao Wang (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Resources [equal], Software [equal], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal]), Bobo Chen (Methodology [equal], Project administration [equal], Resources [equal], Software [equal], Supervision [equal], Validation [equal], Visualization [equal], Writing—original draft [equal], Writing—review and editing [equal]), and Yunhui Chen (Data curation [equal], Formal analysis [equal], Funding acquisition [equal], Investigation [equal], Validation [equal], Visualization [equal], Writing—original draft [equal]).
Funding
None declared.
References
[1]
Pearson PJG Watson J
The unfolding low-carbon transition in the UK electricity system
.
Proc Natl Acad Sci USA
2023
;
120
:e2206235120. https://doi.org/10.1073/pnas.2206235120.
OpenURL Placeholder Text
[2]
Miller CA Iles A Jones CF
The social dimensions of energy transitions
.
Sci Cult
2013
;
22
:
135
–
48
[3]
Kabeyi MJB Oludolapo AO
Geothermal wellhead technology power plants in grid electricity generation: a review
.
Energy Strat Rev
2022
;
39
:
100735
[4]
Jin Z Wang J Tang Z
The effects of participation in global value chains on energy intensity: evidence from international industry-level decomposition
.
Energy Strat Rev
2022
;
39
:
100780
[5]
Liu Y Feng C
Promoting renewable energy through national energy legislation
.
Energy Econ
2023
;
118
:
106504
[6]
Solarin SA Bello MO Bekun FV
Sustainable electricity generation: the possibility of substituting fossil fuels for hydropower and solar energy in Italy
.
Int J Sustain Dev World Ecol
2021
;
28
:
429
–
39
[7]
Khan I Zakari A Dagar V
World energy trilemma and transformative energy developments as determinants of economic growth amid environmental sustainability
.
Energy Econ
2022
;
108
:
105884
[8]
Griffiths S Sovacool BK Kim J
Industrial decarbonization via hydrogen: a critical and systematic review of developments, socio-technical systems and policy options
.
Energy Res Soc Sci
2021
;
80
:
102208
[9]
Chen H Wang C Ye M
An uncertainty analysis of subsidy for carbon capture and storage (CCS) retrofitting investment in China's coal power plants using a real-options approach
.
J Clean Prod
2016
;
137
:
200
–
12
[10]
Choulga M Janssens-Maenhout G Super I
Global anthropogenic CO2 emissions and uncertainties as a prior for earth system modelling and data assimilation
.
Earth Syst Sci Data
2021
;
13
:
5311
–
35
[11]
Ngan HW
Electricity regulation and electricity market reforms in China
.
Energy Policy
2010
;
38
:
2142
–
8
[12]
Wu Y Yu Z Ngan HW
Sustaining China′ s electricity market development
.
Energy Policy
2014
;
73
:
30
–
7
[13]
Zhao X Lyon TP Song C
Lurching towards markets for power: China’s electricity policy 1985–2007
.
Appl Energy
2012
;
94
:
148
–
55
[14]
Moreno R Street A Arroyo JM
Planning low-carbon electricity systems under uncertainty considering operational flexibility and smart grid technologies
.
Philos Trans R Soc A Math Phys Eng Sci
2017
;
375
:
20160305
[15]
Velásquez C Watts D Rudnick H
A framework for transmission expansion planning: a complex problem clouded by uncertainty
.
IEEE Power Energy Magazine
2016
;
14
:
20
–
9
[16]
Haas J Cebulla F Cao K
Challenges and trends of energy storage expansion planning for flexibility provision in low-carbon power systems–a review
.
Renew Sust Energ Rev
2017
;
80
:
603
–
19
[17]
Huang Y Masrur H Lipu MSH
Multi-objective optimization of campus microgrid system considering electric vehicle charging load integrated to power grid
.
Sustain Cities Soc
2023
;
98
:
104778
[18]
Walls MR
Combining decision analysis and portfolio management to improve project selection in the exploration and production firm
.
J Pet Sci Eng
2004
;
44
:
55
–
65
[19]
Jano-Ito MA Crawford-Brown D
Investment decisions considering economic, environmental and social factors: an actors' perspective for the electricity sector of Mexico
.
Energy
2017
;
121
:
92
–
106
[20]
Zhang D Ma L Liu P
A multi-period superstructure optimisation model for the optimal planning of China's power sector considering carbon dioxide mitigation: discussion on China's carbon mitigation policy based on the model
.
Energy Policy
2012
;
41
:
173
–
83
[21]
Pisciella P Vespucci MT Bertocchi M
A time consistent risk averse three-stage stochastic mixed integer optimization model for power generation capacity expansion
.
Energy Econ
2016
;
53
:
203
–
11
[22]
Flores-Quiroz A Strunz K
A distributed computing framework for multi-stage stochastic planning of renewable power systems with energy storage as flexibility option
.
Appl Energy
2021
;
291
:
116736
[23]
Wang B Wang X Wei F
Multi-stage stochastic planning for a long-term low-carbon transition of island power system considering carbon price uncertainty and offshore wind power
.
Energy
2023
;
282
:
128349
[24]
Haugen M Blaisdell-Pijuan PL Botterud A
Power market models for the clean energy transition: state of the art and future research needs
.
Appl Energy
2024
;
357
:
122495
[25]
Fu X Wu X Zhang C
Planning of distributed renewable energy systems under uncertainty based on statistical machine learning
.
Prot Contr Mod Pow Systems
2022
;
7
:
41
[26]
Zhang B Hu W Xu X
Hybrid data-driven method for low-carbon economic energy management strategy in electricity-gas coupled energy systems based on transformer network and deep reinforcement learning
.
Energy
2023
;
273
:
127183
[27]
Yu Y Yang M Zhang Y
Fast reconfiguration method of low-carbon distribution network based on convolutional neural network
.
Front Energy Res
2023
;
11
:
1102949
[28]
Zhou B Li Y Ding Y
An input-output-based Bayesian neural network method for analyzing carbon reduction potential: a case study of Guangdong province
.
J Clean Prod
2023
;
389
:
135986
[29]
Qiu J Zhao J Wen F
Challenges and pathways of low-carbon oriented energy transition and power system planning strategy: a review
.
IEEE Trans Netw Sci Eng
2023
;
1
–
21
. https://doi.org/10.1109/TNSE.2023.3344729.
OpenURL Placeholder Text
[30]
Wang Y Peng S Guan X
Collaborative logistics pickup and delivery problem with eco-packages based on time–space network
.
Expert Syst Appl
2021
;
170
:
114561
[31]
Morgan DR Styles D Lane ET
Packaging choice and coordinated distribution logistics to reduce the environmental footprint of small-scale beer value chains
.
J Environ Manag
2022
;
307
:
114591
[32]
Zhang F Wang B Gong Z
Development of photovoltaic-electrolyzer-fuel cell system for hydrogen production and power generation
.
Energy
2023
;
263
:
125566
[33]
Anderson TR Hawkins E Jones PD
CO2, the greenhouse effect and global warming: from the pioneering work of Arrhenius and Callendar to today's earth system models
.
Endeavour
2016
;
40
:
178
–
87
[34]
Ma CQ Ren YS Zhang YJ
The allocation of carbon emission quotas to five major power generation corporations in China
.
J Clean Prod
2018
;
189
:
1
–
12
[35]
Niu W Feng Z Cheng C
Optimization of variable-head hydropower system operation considering power shortage aspect with quadratic programming and successive approximation
.
Energy
2018
;
143
:
1020
–
8
[36]
Sellak H Ouhbi B Frikh B
Towards next-generation energy planning decision-making: an expert-based framework for intelligent decision support
.
Renew Sust Energ Rev
2017
;
80
:
1544
–
77
[37]
Chen LJ Zhu L Fan Y
Long-term impacts of carbon tax and feed-in tariff policies on China's generating portfolio and carbon emissions: a multi-agent-based analysis
.
Energy Environ
2013
;
24
:
1271
–
93
[38]
Zhang C Wu X Zhao S
Simulation study on the low carbon transition process in Japan’s electricity market
.
Green Technol Sustain
2023
;
1
:
100006
[39]
Luo Y Wang Z Zhu J
Multi-objective robust optimization of a solar power tower plant under uncertainty
.
Energy
2022
;
238
:
121716
[40]
Han X Nishida N Morita M
Compensation method for missing and misidentified skeletons in nursing care action assessment by improving spatial temporal graph convolutional networks
.
Bioengineering
2024
;
11
:
127
[41]
Fang L He B
A deep learning framework using multi-feature fusion recurrent neural networks for energy consumption forecasting
.
Appl Energy
2023
;
348
:
121563
© The Author(s) 2024. Published by Oxford University Press.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com
