A method and system for electricity price interval prediction and price difference evaluation

Abstract

The application belongs to the technical field of power market analysis and prediction, and discloses a price interval prediction and price difference evaluation method based on policy regulation signals and market game. Firstly, the policy text is quantitatively coded through natural language processing technology, and a policy impact factor is extracted. Secondly, a market game model based on multi-agent reinforcement learning is constructed to simulate the evolution of the power supplier's bidding strategy. Then, the policy factor and the game characteristics are fused with multi-source data and input into a quantile regression neural network or a probability diffusion model to output the complete probability distribution of the electricity price (interval prediction). Finally, based on the predicted node electricity price distribution, the correlation between the electricity prices of different nodes is described by using a Copula function, and the price difference distribution and arbitrage opportunity index under different confidence levels are calculated. The application significantly improves the accuracy of electricity price prediction and the reliability of price difference evaluation in the policy window period by explicitly modeling the difficult-to-quantify policy signals and endogenously incorporating market game into the prediction process.

Description

Technical Field

[0001] This invention belongs to the field of power market analysis and forecasting technology, and in particular relates to a method and system for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory. Background Technology

[0002] With the deepening of electricity market reform, the electricity pricing mechanism has become increasingly complex. Electricity prices are not only influenced by economic factors such as supply and demand, fuel costs, and renewable energy output, but also by significant intervention from government policies at all levels, including adjustments to coal benchmark prices, peak-valley electricity pricing policies, carbon emission quota allocation, market price limits, and the phasing out of renewable energy subsidies. These policy signals are characterized by their suddenness, non-linearity, and difficulty in quantification. Traditional electricity price forecasting methods often simplify them as dummy variables or completely ignore them, leading to systematic biases in forecasting models during policy release windows. On the other hand, the electricity market is essentially a multi-player game. The bidding strategies of power generators, electricity retailers, large users, and energy storage participants influence each other, jointly determining the market clearing price. Existing electricity price forecasting methods often treat the market as a "black box," learning statistical patterns solely from historical price series, lacking explicit modeling of the behavioral characteristics of the game players, and failing to capture sudden price changes caused by shifts in market power and strategic bidding. Especially in nodal price forecasting, the price differences (i.e., price spreads) between different nodes reflect the value of congestion revenue rights (FTRs). Accurate price spread forecasting is the foundation for arbitrage strategies and financial transmission rights trading, but existing methods generally lack sufficient accuracy in predicting price spreads. Therefore, there is an urgent need for a comprehensive electricity price forecasting method that can quantify policy control signals, characterize the game behavior of market participants, and simultaneously output both electricity price range forecasts and nodal price spread assessments.

[0003] Based on the above analysis, the problems and shortcomings of the existing technology are as follows:

[0004] Existing electricity price forecasting methods often treat the market as a "black box," relying solely on historical price series to learn statistical patterns. They lack explicit modeling of the behavioral characteristics of game players, making it difficult to capture sudden price fluctuations caused by changes in market power and strategic bidding. Particularly in nodal price forecasting, the price differences between different nodes (i.e., price spreads) reflect the value of congestion revenue rights (FTRs). Accurate price spread forecasting is fundamental to arbitrage strategies and financial transmission rights trading, but existing methods generally lack sufficient accuracy in predicting price spreads. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory.

[0006] This invention is implemented as follows: a method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory includes:

[0007] Step 1: Policy signal collection and quantification, real-time monitoring of policy documents released by government websites and regulatory agencies;

[0008] Policy content and release time were extracted using NLP technology, and policy impact factors were calculated. and policy uncertainty index ;

[0009] Step 2: Market game state perception, run multi-agent reinforcement learning simulation, simulate the bidding behavior of major generators, extract game features (MPI, BidDev, Intensity), or estimate these features in real time from actual market data;

[0010] Step 3: Multi-source feature fusion and electricity price range prediction;

[0011] By integrating policy characteristics, game theory characteristics, and traditional data, and inputting them into a quantile regression network or diffusion model, the project outputs electricity price quantile predictions for each node over multiple future time periods. ;

[0012] Step 4: Node correlation modeling and price difference assessment;

[0013] Estimate Copula parameters, calculate the joint distribution of price spreads between nodes, and output key price spread indicators (expected price spread, confidence interval, arbitrage probability, FTR value).

[0014] Step 5: Results visualization and decision support, generating visualization results such as electricity price range forecast charts, price spread heatmaps, and arbitrage opportunity maps to assist users in risk management, pricing strategy optimization, and financial transmission rights trading decisions;

[0015] Step Six: Online Model Update. When new data arrives, the model parameters are updated using an online learning method.

[0016] In particular, the attenuation parameter of policy shocks It will be dynamically adjusted based on the actual impact.

[0017] Furthermore, the electricity price range is predicted as follows:

[0018] (1) Problem definition and symbol explanation;

[0019] Assume there is an electricity market Each node predicts the future. The nodal electricity price at each time step (e.g., per hour); at time... Nodes need to be predicted In the future Electricity price at any time The probability distribution and node pairs The price difference between The distribution of ; define the following feature set:

[0020] Basic feature set: load Renewable energy output fuel prices Unit Combination Status

[0021] Market Feature Set: Historical Electricity Price Series Market clearing volume Blocked surplus

[0022] Policy Feature Set: Quantified Vectors of Policy Texts

[0023] Game Theory Feature Set: Characteristics of Market Participants' Bidding Strategies ;

[0024] (2) Policy regulation signal quantification module;

[0025] (3) Market participant game behavior modeling module;

[0026] (4) Electricity price range forecasting backbone network;

[0027] (5) Node price difference assessment module.

[0028] Furthermore, the policy regulation signal quantification module:

[0029] Policy signals are characterized by being textual, discrete, and sudden. This invention proposes a multi-level policy quantification method.

[0030] 1) Policy text coding

[0031] Collect policy texts related to electricity prices (such as price documents from the National Development and Reform Commission, market supervision announcements, carbon emission policies, etc.) and perform semantic encoding using a pre-trained language model:

[0032] (1)

[0033] For the electricity price forecasting task, the original code is reduced in dimensionality and fine-tuned to adapt to the task:

[0034] (2)

[0035] in , These are learnable parameters;

[0036] 2) Decomposition of policy shock effects

[0037] The policy impact is decomposed into direct and indirect effects. Direct effects refer to price adjustments explicitly stipulated by the policy (such as an increase of X cents / kWh in the benchmark price of coal). Indirect effects refer to the price impact of the policy transmitted through market participants' expectations and behaviors. An event study method is used to estimate the policy impact function.

[0038] (3)

[0039] in For the first The release time of the policy, The initial impact strength, The attenuation coefficient is... For indicator functions;

[0040] 3) Measurement of policy uncertainty

[0041] Besides the content of the policy, policy uncertainty also affects electricity prices; keywords related to uncertainty in the text (such as "may") are used. The density of policies (such as "gradual" and "timely adjustment") is used to construct the policy uncertainty index.

[0042] (4).

[0043] Furthermore, the market participant game behavior modeling module:

[0044] In the electricity market, power generators make strategic bids based on the principle of profit maximization, and their behavior can be described by game theory models.

[0045] 1) Multi-agent reinforcement learning game framework

[0046] The major generators in the market are considered as intelligent agents, each intelligent agent... At any moment status This includes: the status of its own generating units, historical price quotes from competitors, market supply and demand, policy signals, etc.; the actions of the intelligent agent. The parameters of the bidding curve (such as slope and intercept) are defined; the environment (market operator) clears the market based on the bids of all agents, determining the node price and revenue; the reward function for each agent is its profit.

[0047] (5)

[0048] in For intelligent agents The collection of generator sets owned, The power generation cost function;

[0049] The equilibrium policy is learned using the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm; agents of network Input the states and actions of all agents, and output the actions of the agents. Value estimation:

[0050] (6)

[0051] actor network Output an action based on its own state:

[0052] (7)

[0053] in To explore noise;

[0054] 2) Game Feature Extraction

[0055] Extracting features reflecting market game states from multi-agent simulations:

[0056] (1) Market power index: reflects the power generator's ability to manipulate prices.

[0057] (8)

[0058] (2) Price deviation: The average deviation between the actual price and the marginal cost

[0059] (9)

[0060] (3) Game intensity index: the degree of drastic change in price between adjacent time periods

[0061] (10).

[0062] Furthermore, the electricity price range prediction backbone network:

[0063] By integrating policy characteristics, game-theoretic characteristics, and traditional characteristics, a probabilistic prediction model is constructed.

[0064] 1) Multi-source feature fusion

[0065] Feature vector concatenation:

[0066] (11)

[0067] Using a Transformer encoder to capture timing dependencies:

[0068] (12)

[0069] 2) Quantile Regression Network

[0070] For each node and prediction step size The network outputs multiple quantiles:

[0071] (13)

[0072] in The quantile levels are (e.g., 0.1, 0.25, 0.5, 0.75, 0.9).

[0073] The training objective is pinball loss:

[0074]

[0075] (14)

[0076] 3) Probabilistic prediction based on diffusion model

[0077] To more flexibly characterize multimodal distributions, a diffusion probability model can be used;

[0078] (1) Gradual addition of noise during the forward process:

[0079] (15)

[0080] (2) Reverse process learning for noise reduction:

[0081] (16)

[0082] (3) The training objective is a simplified form of the variational lower bound:

[0083] (17).

[0084] Furthermore, the inter-node price difference assessment module:

[0085] Based on the predicted distribution of nodal electricity prices, the price difference between nodals is assessed to provide a basis for financial transmission rights trading and arbitrage strategies.

[0086] 1) Joint distribution modeling

[0087] Because nodal prices are strongly correlated, individual nodes cannot be treated independently; a Copula function is used to model the inter-node dependency structure; based on Sklar's theorem:

[0088] (18)

[0089] in For nodes Marginal distribution of electricity prices (obtained from quantile predictions) The Copula function is used; t-Copula is chosen to capture tail correlation:

[0090] (19)

[0091] in For degrees of freedom The t-distribution, This is a correlation matrix;

[0092] 2) Calculation of price spread distribution

[0093] For any pair of nodes Price difference The distribution can be obtained through Copula sampling:

[0094] (20)

[0095] Key spread calculation:

[0096] Price spread expectation:

[0097] Price spread confidence interval:

[0098] Probability of price spread sign: Represents a node Electricity price higher than node probability

[0099] Arbitrage opportunity indicator: When the absolute value of the price difference exceeds the transmission cost. Arbitrage opportunities exist

[0100] (twenty one)

[0101] 3) Valuation of Obstruction Revenue Rights

[0102] Financial transmission rights (FTR) holders receive revenue during congestion; for slave nodes... To the node The FTR, whose unit value is the price difference. The expected value of FTR is:

[0103] (twenty two)

[0104] Conditional Value at Risk (CVaR) measures value under extreme conditions:

[0105] (twenty three).

[0106] Another objective of this invention is to provide a system for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory, comprising:

[0107] The data acquisition and quantification module is used for policy signal acquisition and quantification, monitoring policy documents released by government websites and regulatory agencies in real time; it extracts policy content and release time through NLP technology and calculates policy impact factors. and policy uncertainty index ;

[0108] The state awareness module is used for market game state awareness, runs multi-agent reinforcement learning simulation, simulates the bidding behavior of major generators, extracts game features (MPI, BidDev, Intensity), or estimates these features in real time from actual market data.

[0109] The prediction module is used for multi-source feature fusion and electricity price range prediction. It integrates policy features, game theory features, and traditional data, inputs them into a quantile regression network or diffusion model, and outputs the electricity price quantile predictions for each node over multiple future time periods. ;

[0110] The evaluation module is used for modeling the correlation between nodes and evaluating price spreads; it estimates Copula parameters, calculates the joint distribution of price spreads between nodes, and outputs key price spread indicators (expected price spread, confidence interval, arbitrage probability, FTR value).

[0111] The visualization module is used for result visualization and decision support, generating visualization results such as electricity price range forecast maps, price difference heat maps, and arbitrage opportunity maps to assist users in risk management, pricing strategy optimization, and financial transmission rights trading decisions.

[0112] The update module is used for online model updates. When new data arrives, it updates the model parameters using an online learning approach; this is particularly important for the attenuation parameters of policy shocks. It will be dynamically adjusted based on the actual impact.

[0113] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, the computer program being executed by the processor causing the processor to perform the steps of the method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory.

[0114] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory.

[0115] Another objective of this invention is to provide an information data processing terminal, which is used to implement the electricity price range prediction and price difference assessment system based on policy control signals and market game theory.

[0116] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0117] This invention protects a method and system for predicting electricity price ranges and assessing price spreads based on policy control signals and market game theory. The core of the method lies in constructing a four-stage framework: "policy quantification - game modeling - range prediction - price spread assessment." First, policy texts are quantified and encoded using natural language processing technology to extract policy impact factors. Second, a market game model based on multi-agent reinforcement learning is constructed to simulate the evolution of power generator pricing strategies. Then, policy factors and game characteristics are fused with multi-source data and input into a quantile regression neural network or probability diffusion model to output a complete probability distribution of electricity prices (range prediction). Finally, based on the predicted node price distribution, a Copula function is used to characterize the correlation between node prices, and the price spread distribution and arbitrage opportunity indicators at different confidence levels are calculated. This invention significantly improves the accuracy of electricity price prediction during policy windows and the reliability of price spread assessment by explicitly modeling policy signals that are difficult to quantify and by endogenizing market game theory into the prediction process.

[0118] 1. By quantitatively encoding policy texts and modeling impact effects, the ability to predict policy responses is significantly improved. The method reduces the mean absolute percentage error (MAPE) of electricity price forecasts during the policy release window (3 days before and after) by 32.5% compared to traditional methods, effectively solving the problem of forecast failure caused by policy shocks.

[0119] 2. The multi-agent game theory model can simulate the strategic bidding behavior of power generators, and the extracted game features show a correlation of 0.87 with actual market power. In cases of market power abuse, the model can provide early warning of abnormal electricity price fluctuations 2-3 hours in advance.

[0120] At a 3.90% confidence level, the prediction interval coverage (PICP) reaches 89.5%, and the average interval width (PINAW) is 18.3% narrower than the baseline method, providing a sharper prediction interval while maintaining reliability.

[0121] The constructed policy uncertainty index has a correlation of 0.72 with electricity price volatility, making it a leading indicator of market risk. During periods of concentrated policy activity (such as the release of a large number of electricity reform documents), the model can automatically adjust the forecast interval to accurately reflect market uncertainty. Attached Figure Description

[0122] Figure 1This is a flowchart of the method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory provided in this embodiment of the invention.

[0123] Figure 2 This is a structural block diagram of an electricity price range prediction and price difference assessment system based on policy control signals and market game theory provided in an embodiment of the present invention.

[0124] Figure 3 This is an overall system integration diagram provided in an embodiment of the present invention.

[0125] Figure 4 This is a comparison chart of the accuracy of policy window prediction provided in the embodiments of the present invention.

[0126] Figure 5 This is an example diagram comparing the performance of electricity price range prediction provided in an embodiment of the present invention. Detailed Implementation

[0127] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0128] like Figure 1 As shown in the figure, the method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory provided by this invention includes the following steps:

[0129] S101: Policy signal collection and quantification, real-time monitoring of policy documents issued by government websites and regulatory agencies;

[0130] Policy content and release time were extracted using NLP technology, and policy impact factors were calculated. and policy uncertainty index ;

[0131] S102: Market game state perception, running multi-agent reinforcement learning simulation, simulating the bidding behavior of major generators, extracting game features (MPI, BidDev, Intensity), or estimating these features in real time from actual market data;

[0132] S103: Multi-source feature fusion and electricity price range prediction;

[0133] By integrating policy characteristics, game theory characteristics, and traditional data, and inputting them into a quantile regression network or diffusion model, the project outputs electricity price quantile predictions for each node over multiple future time periods. ;

[0134] S104: Inter-node correlation modeling and price difference assessment;

[0135] Estimate Copula parameters, calculate the joint distribution of price spreads between nodes, and output key price spread indicators (expected price spread, confidence interval, arbitrage probability, FTR value).

[0136] S105: Results visualization and decision support, generating visualization results such as electricity price range forecast maps, price difference heat maps, and arbitrage opportunity maps to assist users in risk management, pricing strategy optimization, and financial transmission rights trading decisions;

[0137] S106: Online model update; when new data arrives, the model parameters are updated using an online learning method.

[0138] In particular, the attenuation parameter of policy shocks It will be dynamically adjusted based on the actual impact.

[0139] This invention constructs an overall predictive framework based on the inherent transmission logic of the electricity market: policy shock—market game—price formation—price spread evolution. Its working principle lies in unifying the modeling of exogenous policy control signals and endogenous market game behavior, and characterizing price range and node price spread risks through probability distributions. First, at the policy signal level, by capturing policy texts released by the government and regulatory agencies in real time, natural language processing technology is used to identify key information such as policy category, intensity, scope of application, and effective time. This constructs policy shock factors and policy uncertainty indices, transforming qualitative policy events into quantifiable time-series features, and introducing an impact decay function to describe the dynamic process of policy impact decreasing over time. Subsequently, at the market level, a multi-agent reinforcement learning simulation environment is constructed to simulate the evolution of bidding strategies by major power generators under constraints. Game characteristic parameters reflecting the degree of market competition and behavioral deviation are extracted, or inverse estimation is performed based on real market transaction and bidding data, thereby characterizing the behavioral driving mechanism of electricity price formation. Building upon this foundation, the system integrates policy characteristics, game theory features, and traditional variables such as load forecasting, fuel costs, and meteorological factors. These are then input into a quantile regression network or a probability generation model based on a diffusion mechanism to directly output the conditional distribution quantiles of electricity prices at each node over multiple future time periods, achieving interval forecasting rather than single-point forecasting. Furthermore, the system uses a Copula function to model the correlation coupling of electricity price distributions at different nodes, obtaining the joint probability distribution of price differences between nodes. This allows for the calculation of key risk indicators such as expected price difference, confidence interval, arbitrage probability, and the value of financial transmission rights.

[0140] Finally, the system uses an online learning mechanism to dynamically update model parameters when new data arrives, especially by adaptively correcting the policy impact attenuation coefficient, so that the model can continuously reflect the latest market structure changes, thereby providing stable, interpretable and quantifiable decision support for risk management, pricing optimization and cross-node arbitrage decisions.

[0141] like Figure 2As shown in the figure, an electricity price range prediction and price difference assessment system based on policy control signals and market game theory provided by this invention includes:

[0142] The data acquisition and quantification module is used for policy signal acquisition and quantification, monitoring policy documents released by government websites and regulatory agencies in real time; it extracts policy content and release time through NLP technology and calculates policy impact factors. and policy uncertainty index ;

[0143] The state awareness module is used for market game state awareness, runs multi-agent reinforcement learning simulation, simulates the bidding behavior of major generators, extracts game features (MPI, BidDev, Intensity), or estimates these features in real time from actual market data.

[0144] The prediction module is used for multi-source feature fusion and electricity price range prediction. It integrates policy features, game theory features, and traditional data, inputs them into a quantile regression network or diffusion model, and outputs the electricity price quantile predictions for each node over multiple future time periods. ;

[0145] The evaluation module is used for modeling the correlation between nodes and evaluating price spreads; it estimates Copula parameters, calculates the joint distribution of price spreads between nodes, and outputs key price spread indicators (expected price spread, confidence interval, arbitrage probability, FTR value).

[0146] The visualization module is used for result visualization and decision support, generating visualization results such as electricity price range forecast maps, price difference heat maps, and arbitrage opportunity maps to assist users in risk management, pricing strategy optimization, and financial transmission rights trading decisions.

[0147] The update module is used for online model updates. When new data arrives, it updates the model parameters using an online learning approach; this is particularly important for the attenuation parameters of policy shocks. It will be dynamically adjusted based on the actual impact.

[0148] The electricity price range prediction and price difference assessment system based on policy regulation signals and market game theory provided in this invention revolves around a closed loop of exogenous policy-driven, endogenous game evolution, probability distribution prediction, node-related coupling, and dynamic updates. Each module forms a collaborative linkage mechanism of data flow and parameter flow. The data acquisition and quantification module performs real-time capture and structured processing of policy documents published by government websites and regulatory agencies. It uses a natural language processing model to perform semantic parsing and time annotation on the policy text, mapping information such as policy category, scope of application, and intensity into policy impact factors and policy uncertainty indices. Furthermore, it constructs a dynamic trajectory of policy impact through a time decay function, forming quantitative features that can be directly used in modeling. Based on this, the state perception module simulates the evolution of bidding behavior of power generation entities under different constraints by constructing a multi-agent reinforcement learning simulation environment, or estimates game behavior parameters based on real market transaction and bidding data. This extracts game characteristic variables reflecting market competition intensity, bidding deviation, and strategy concentration, thereby characterizing the intrinsic mechanism of electricity price formation.

[0149] The prediction module integrates the aforementioned policy and game-theoretic characteristics with traditional variables such as load, weather, and fuel costs, using a multi-source fusion approach. It inputs these variables into a quantile regression network or diffusion probability model, directly outputting the conditional distribution quantiles of electricity prices at each node for multiple future time periods, thus achieving interval prediction and tail risk characterization. The evaluation module further models the relevant structure of electricity price distributions at different nodes based on Copula functions, obtaining the joint probability distribution of price differences between nodes and calculating key indicators such as expected price difference, confidence interval, arbitrage probability, and the value of financial transmission rights, extending from single-node risk to cross-node risk. The visualization module transforms the prediction and evaluation results into multi-dimensional graphical interfaces such as interval prediction maps, price difference heatmaps, and arbitrage opportunity distribution maps, providing users with intuitive decision support. The update module uses an online learning mechanism to incrementally correct model parameters when new data arrives, particularly adaptively adjusting the policy impact attenuation parameters, enabling the system to continuously reflect changes in market structure and forming a dynamic closed-loop optimization mechanism. Furthermore, the present invention also implements the above method steps in the form of computer equipment, computer-readable storage media and information data processing terminals, so that the system functions can run in a standard hardware environment and realize industrial-grade application.

[0150] like Figure 3 As shown in Figure 1, Mathematical Modeling for Electricity Price Range Forecasting and Price Spread Assessment

[0151] 1.1 Problem Definition and Symbol Explanation

[0152] Assume there is an electricity market Each node predicts the future. The nodal electricity price at each time step (e.g., per hour). Nodes need to be predicted In the future Electricity price at any time The probability distribution and node pairs The price difference between The distribution of . Define the following feature set:

[0153] Basic feature set: load Renewable energy output fuel prices Unit Combination Status

[0154] Market Feature Set: Historical Electricity Price Series Market clearing volume Blocked surplus

[0155] Policy Feature Set: Quantified Vectors of Policy Texts

[0156] Game Theory Feature Set: Characteristics of Market Participants' Bidding Strategies

[0157] 1.2 Policy Regulation Signal Quantification Module

[0158] Policy signals are characterized by their textual nature, discreteness, and suddenness. This invention proposes a multi-level policy quantification method.

[0159] 1.2.1 Policy Text Coding

[0160] Collect policy texts related to electricity prices (such as price documents from the National Development and Reform Commission, market supervision announcements, carbon emission policies, etc.) and perform semantic encoding using a pre-trained language model:

[0161] (1)

[0162] For the electricity price forecasting task, the original code is reduced in dimensionality and fine-tuned to adapt to the task:

[0163] (2)

[0164] in , These are learnable parameters.

[0165] 1.2.2 Decomposition of Policy Shock Effects

[0166] The policy impact is decomposed into direct and indirect effects. Direct effects refer to price adjustments explicitly stipulated by the policy (such as an increase of X fen / kWh in the benchmark price of coal); indirect effects refer to the price impact transmitted through the policy's influence on market participants' expectations and behaviors. The event study method is used to estimate the policy shock function.

[0167] (3)

[0168] in For the first The release time of the policy, The initial impact strength, The attenuation coefficient is... This is an indicator function.

[0169] 1.2.3 Measurement of Policy Uncertainty

[0170] Besides the content of the policy, policy uncertainty also affects electricity prices. A policy uncertainty index is constructed using the density of uncertainty keywords (such as "possibly," "gradually," and "timely adjustment") in the text.

[0171] (4)

[0172] 1.3 Market Participant Game Behavior Modeling Module

[0173] In the electricity market, power generators make strategic bids based on the principle of profit maximization, and their behavior can be described by game theory models.

[0174] 1.3.1 Multi-agent reinforcement learning game framework

[0175] The major generators in the market are considered as intelligent agents, each intelligent agent... At any moment status This includes: the status of its own generating units, historical price quotes from competitors, market supply and demand, and policy signals. The actions of the intelligent agent. The parameters of the bidding curve (such as slope and intercept) are defined. The environment (market operator) clears the market based on the bids of all agents, determining the node price and revenue. The reward function for each agent is its profit.

[0176] (5)

[0177] in For intelligent agents The collection of generator sets owned, This is a function for the cost of electricity generation.

[0178] The equilibrium policy is learned using the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. (Agents...) The critic network Input the states and actions of all agents, and output the actions of the agents. Value estimation:

[0179] (6)

[0180] actor network Output an action based on its own state:

[0181] (7)

[0182] in To explore noise.

[0183] 1.3.2 Game Theory Feature Extraction

[0184] Extracting features reflecting market game states from multi-agent simulations:

[0185] (1) Market power index: reflects the power generator's ability to manipulate prices.

[0186] (8)

[0187] (2) Price deviation: The average deviation between the actual price and the marginal cost

[0188] (9)

[0189] (3) Game intensity index: the degree of drastic change in price between adjacent time periods

[0190] (10)

[0191] 1.4 Electricity Price Range Forecasting Backbone Network

[0192] By integrating policy characteristics, game theory characteristics, and traditional characteristics, a probabilistic prediction model is constructed.

[0193] 1.4.1 Multi-source feature fusion

[0194] Feature vector concatenation:

[0195] (11)

[0196] Using a Transformer encoder to capture timing dependencies:

[0197] (12)

[0198] 1.4.2 Quantile Regression Network

[0199] For each node and prediction step size The network outputs multiple quantiles:

[0200] (13)

[0201] in The quantile levels are (e.g., 0.1, 0.25, 0.5, 0.75, 0.9).

[0202] The training objective is pinball loss:

[0203]

[0204] (14)

[0205] 1.4.3 Probabilistic Prediction Based on Diffusion Model

[0206] To more flexibly characterize multimodal distributions, a diffusion probability model can be used.

[0207] (1) Gradual addition of noise during the forward process:

[0208] (15)

[0209] (2) Reverse process learning for noise reduction:

[0210] (16)

[0211] (3) The training objective is a simplified form of the variational lower bound:

[0212] (17)

[0213] 1.5 Inter-node price difference assessment module

[0214] Based on the predicted distribution of nodal electricity prices, the price difference between nodals is assessed, providing a basis for financial transmission rights trading and arbitrage strategies.

[0215] 1.5.1 Joint Distribution Modeling

[0216] Because nodal prices are strongly correlated, individual nodes cannot be treated independently. A Copula function is used to model the dependency structure between nodes. According to Sklar's theorem:

[0217] (18)

[0218] in For nodes Marginal distribution of electricity prices (obtained from quantile predictions) The Copula function is used. The t-Copula is chosen to capture tail correlation.

[0219] (19)

[0220] in For degrees of freedom The t-distribution, This is the correlation matrix.

[0221] 1.5.2 Calculation of Price Spread Distribution

[0222] For any pair of nodes Price difference The distribution can be obtained through Copula sampling:

[0223] (20)

[0224] Key spread calculation:

[0225] (1) Expected price spread:

[0226] (2) Price spread confidence interval:

[0227] (3) Probability of price difference sign: Represents a node Electricity price higher than node probability

[0228] (4) Arbitrage opportunity indicator: When the absolute value of the price difference exceeds the transmission cost Arbitrage opportunities exist

[0229] (twenty one)

[0230] 1.5.3 Valuation of Obstruction Revenue Rights

[0231] Financial transmission rights (FTR) holders earn revenue during periods of congestion. For slave nodes... To the node The FTR, whose unit value is the price difference. The expected value of FTR is:

[0232] (twenty two)

[0233] Conditional Value at Risk (CVaR) measures value under extreme conditions:

[0234] (twenty three)

[0235] 2.1 Test System Construction

[0236] This study focuses on the electricity market of a certain province, encompassing 58 nodes and market participants including 12 power generation groups (total installed capacity of 45GW), 8 electricity sales companies, and 3 energy storage operators. The data covers the period from 2022 to 2024, with a time resolution of 1 hour, totaling approximately 26,280 time points. Policy data sources include 246 electricity price-related policy documents issued by the National Development and Reform Commission, the National Energy Administration, and the provincial Development and Reform Commission, with the time span matching the electricity price data. The system is built using Python, with the following main modules:

[0237] (1) Policy Quantification Module: Fine-tuning of BERT Model Based on HuggingFace Transformers Library

[0238] (2) Game simulation module: The MADDPG multi-agent framework is implemented based on PyTorch.

[0239] (3) Electricity price forecasting module: diffusion model implementation

[0240] (4) Price spread assessment module: Copula parameter estimation and sampling based on Copulalib

[0241] 2.2 Comparison of Model Configurations

[0242] Five comparative models are set up, mainly including: ARIMA-GARCH: traditional time series model; LSTM: stacked LSTM (historical electricity price only); LSTM+features: LSTM input with multi-source features (no policy quantification, no game theory modeling); Transformer+quantiles: Transformer encoder + quantile output (policy quantification, no game theory); the patented method: policy quantification + game theory features + Transformer + quantiles + Copula price difference assessment.

[0243] 2.3 Analysis of Experimental Results

[0244] like Figure 4 (1) Comparison of the accuracy of policy window period forecasts

[0245] We selected 12 major electricity price policy announcements between 2022 and 2024 and compared their predictive performance three days before and after the policy announcements.

[0246]

[0247] The MAPE of this patented method is only 5.7% during the policy window period, which is 56.8% lower than the benchmark model, demonstrating significant adaptability to policy shocks.

[0248] (2) Validation of game features

[0249] Compare the prediction results before and after adding game-theoretic features:

[0250]

[0251] Incorporating game theory features significantly improves prediction accuracy, especially the Market Power Index (MPI), which shows the most significant improvement.

[0252] like Figure 5 (3) Comparison of Interval Prediction Performance

[0253]

[0254] The patented method achieves a PICP value close to the ideal of 90%, the narrowest PINAW, and the diffusion model version performs better at extreme quantiles.

[0255] (4) Price spread prediction and arbitrage effect

[0256] Five key blocking sections were selected to compare the price spread prediction performance:

[0257]

[0258] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0259] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory, characterized in that, Includes the following steps: Step 1: Policy signal collection and quantification, real-time monitoring of policy documents released by government websites and regulatory agencies; Policy content and release time were extracted using NLP technology, and policy impact factors were calculated. and policy uncertainty index ; Step 2: Market game state perception, run multi-agent reinforcement learning simulation, simulate the bidding behavior of major generators, extract game features (MPI, BidDev, Intensity), or estimate these features in real time from actual market data; Step 3: Multi-source feature fusion and electricity price range prediction; By integrating policy characteristics, game theory characteristics, and traditional data, and inputting them into a quantile regression network or diffusion model, the project outputs electricity price quantile predictions for each node over multiple future time periods. ; Step 4: Node correlation modeling and price difference assessment; Estimate Copula parameters, calculate the joint distribution of price spreads between nodes, and output key price spread indicators; Step 5: Results visualization and decision support, generating visualization results such as electricity price range forecast charts, price spread heatmaps, and arbitrage opportunity maps to assist users in risk management, pricing strategy optimization, and financial transmission rights trading decisions; Step Six: Online Model Update. When new data arrives, the model parameters are updated using an online learning method. In particular, the attenuation parameter of policy shocks It will be dynamically adjusted based on the actual impact.

2. The method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory as described in claim 1, characterized in that, The predicted electricity price range: (1) Problem definition and symbol explanation; Assume there is an electricity market Each node predicts the future. The nodal electricity price at each time step (e.g., per hour); at time... Nodes need to be predicted In the future Electricity price at any time The probability distribution and node pairs The price difference between Distribution; definition The following feature set: Basic feature set: load Renewable energy output fuel prices Unit Combination Status Market Feature Set: Historical Electricity Price Series Market clearing volume Blocked surplus Policy Feature Set: Quantified Vectors of Policy Texts Game Theory Feature Set: Characteristics of Market Participants' Bidding Strategies ; (2) Policy regulation signal quantification module; (3) Market participant game behavior modeling module; (4) Electricity price range forecasting backbone network; (5) Node price difference assessment module.

3. The method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory as described in claim 2, characterized in that... The policy regulation signal quantification module: Policy signals are characterized by being textual, discrete, and sudden. This invention proposes a multi-level policy quantification method. 1) Policy text coding Collect policy texts related to electricity prices (such as price documents from the National Development and Reform Commission, market supervision announcements, carbon emission policies, etc.) and perform semantic encoding using a pre-trained language model: （1） For the electricity price forecasting task, the original code is reduced in dimensionality and fine-tuned to adapt to the task: （2） in , These are learnable parameters; 2) Decomposition of policy impact effects The policy impact is decomposed into direct and indirect effects. Direct effects refer to price adjustments explicitly stipulated by the policy (such as an increase of X cents / kWh in the benchmark price of coal). Indirect effects refer to the price impact of the policy transmitted through market participants' expectations and behaviors. An event study method is used to estimate the policy impact function. （3） in For the first The release time of the policy The initial impact strength, The attenuation coefficient is... For indicator functions; 3) Measurement of policy uncertainty Besides the content of the policy, policy uncertainty also affects electricity prices; a policy uncertainty index is constructed using the density of uncertainty keywords (such as "possibly," "gradually," and "timely adjustment") in the text. （4）。 4. The method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory as described in claim 2, characterized in that... The market participant game behavior modeling module: In the electricity market, power generators make strategic bids based on the principle of profit maximization, and their behavior can be described by game theory models. 1) Multi-agent reinforcement learning game framework The major generators in the market are considered as intelligent agents, each intelligent agent... At any moment status This includes: the status of its own generating units, historical price quotes from competitors, market supply and demand, policy signals, etc.; the actions of the intelligent agent. The parameters of the bidding curve (such as slope and intercept) are defined; the environment (market operator) clears the market based on the bids of all agents, determining the node price and revenue; the reward function for each agent is its profit. （5） in For intelligent agents The collection of generator sets owned, The power generation cost function; The equilibrium policy is learned using the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm; agents The critic network Input the states and actions of all agents, and output the actions of the agents. Value estimation: （6） actor network Output an action based on its own state: （7） in To explore noise; 2) Game Feature Extraction Extracting features reflecting market game states from multi-agent simulations: (1) Market power index: reflects the power generator's ability to manipulate prices. （8） (2) Price deviation: The average deviation between the actual price and the marginal cost （9） (3) Game intensity index: the degree of drastic change in price between adjacent time periods （10）。 5. The method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory as described in claim 2, characterized in that, The electricity price range prediction backbone network: By integrating policy characteristics, game-theoretic characteristics, and traditional characteristics, a probabilistic prediction model is constructed. 1) Multi-source feature fusion Feature vector concatenation: （11） Using a Transformer encoder to capture timing dependencies: （12） 2) Quantile Regression Network For each node and prediction step size The network outputs multiple quantiles: （13） in The quantile levels are (e.g., 0.1, 0.25, 0.5, 0.75, 0.9). The training objective is pinball loss: （14） 3) Probabilistic prediction based on diffusion model To more flexibly characterize multimodal distributions, a diffusion probability model can be used; (1) Gradual addition of noise during the forward process: （15） (2) Reverse process learning for noise reduction: （16） (3) The training objective is a simplified form of the variational lower bound: （17）。 6. The method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory as described in claim 2, characterized in that, The inter-node price difference assessment module: Based on the predicted distribution of nodal electricity prices, the price difference between nodals is assessed to provide a basis for financial transmission rights trading and arbitrage strategies. 1) Joint distribution modeling Because nodal prices are strongly correlated, individual nodes cannot be treated independently; a Copula function is used to model the inter-node dependency structure; based on Sklar's theorem: （18） in For nodes Marginal distribution of electricity prices (obtained from quantile predictions) The Copula function is used; t-Copula is chosen to capture tail correlation: （19） in For degrees of freedom The t-distribution, This is a correlation matrix; 2) Calculation of price spread distribution For any pair of nodes Price difference The distribution can be obtained through Copula sampling: （20） Key spread calculation: Price spread expectation: Price spread confidence interval: Probability of price spread sign: Represents a node Electricity price higher than node probability Arbitrage opportunity indicator: When the absolute value of the price difference exceeds the transmission cost. Arbitrage opportunities exist （21） 3) Valuation of Obstruction Revenue Rights Financial transmission rights (FTR) holders receive revenue during congestion; for slave nodes... To the node The FTR, whose unit value is the price difference. The expected value of FTR is: （22） Conditional Value at Risk (CVaR) measures value under extreme conditions: （23）。 7. A system for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory, implementing the method for predicting electricity price ranges and assessing price differences based on policy control signals and market game theory as described in any one of claims 1-6, characterized in that, The electricity price range prediction and price difference assessment system based on policy control signals and market game theory includes: The data acquisition and quantification module is used for policy signal acquisition and quantification, monitoring policy documents released by government websites and regulatory agencies in real time; it extracts policy content and release time through NLP technology and calculates policy impact factors. and policy uncertainty index ; The state awareness module is used for market game state awareness, runs multi-agent reinforcement learning simulation, simulates the bidding behavior of major generators, extracts game features (MPI, BidDev, Intensity), or estimates these features in real time from actual market data. The prediction module is used for multi-source feature fusion and electricity price range prediction. It integrates policy features, game theory features, and traditional data, inputs them into a quantile regression network or diffusion model, and outputs the electricity price quantile predictions for each node over multiple future time periods. ; The evaluation module is used for modeling the correlation between nodes and evaluating price spreads; it estimates Copula parameters, calculates the joint distribution of price spreads between nodes, and outputs key price spread indicators (expected price spread, confidence interval, arbitrage probability, FTR value). The visualization module is used for result visualization and decision support, generating visualization results such as electricity price range forecast maps, price difference heat maps, and arbitrage opportunity maps to assist users in risk management, pricing strategy optimization, and financial transmission rights trading decisions. The update module is used for online model updates. When new data arrives, it updates the model parameters using an online learning approach; this is particularly important for the attenuation parameters of policy shocks. It will be dynamically adjusted based on the actual impact.

8. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of the electricity price range prediction and price difference assessment method based on policy control signals and market game theory as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the electricity price range prediction and price difference assessment method based on policy control signals and market game theory as described in any one of claims 1-6.

10. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the electricity price range prediction and price difference assessment system based on policy control signals and market game as described in claim 7.

Images