A power station time scale game optimization scheduling method and system adaptive to power spot market fluctuation

Abstract

This invention discloses a time-scale game-theoretic optimization scheduling method and system for power plants adapted to fluctuations in the electricity spot market, belonging to the fields of power system operation and control and electricity market trading technology. The method first generates a set of uncertain scenarios for electricity prices and output, constructs a two-layer game model between energy storage participants and market operators to obtain the day-ahead benchmark charging and discharging power curve; performs rolling optimization at preset intervals, corrects power commands through an intraday optimization model that introduces a dynamic deviation penalty term; performs high-frequency smoothing processing on the commands and triggers a deviation correction mechanism; and can also update model parameters through adaptive learning. This invention can solve the shortcomings of existing energy storage scheduling methods in the electricity spot market, such as inaccurate electricity price prediction, lack of market game perspective, and single time scale, thereby improving the returns of energy storage power plants in the electricity spot market and reducing performance risks.

Description

Technical Field

[0001] This invention relates to the fields of power system operation and control and power market trading technology, specifically to a power plant time-scale game-theoretic optimization scheduling method and system that adapts to fluctuations in the electricity spot market. Background Technology

[0002] With the deepening of power market reforms, the electricity spot market has become a major means of resource allocation. Spot market electricity prices exhibit strong temporal and spatial volatility and uncertainty. Energy storage power stations, as a flexible regulatory resource, can profit from arbitrage through "low charging and high discharging." However, existing dispatching methods have the following problems: (1) Traditional methods often adopt "one-time optimization before the day", which cannot cope with the deviation of intraday electricity price forecast and sudden load changes, resulting in actual implementation deviating from the optimal strategy, and even losses due to deviation assessment; (2) Existing models mostly regard energy storage as a price taker, ignoring the feedback effect of large-scale energy storage participation in the market on the clearing price. That is, the energy storage behavior itself will change the electricity price, and there is a lack of pricing strategies based on master-slave game or non-cooperative game. (3) In the face of extreme fluctuations in spot prices, traditional scheduling methods lack robustness and are prone to overcharging and over-discharging of batteries or failing to capture transient high price opportunities. (4) It is difficult to achieve a dynamic balance between maximizing arbitrage profits, minimizing battery life loss, and meeting grid dispatch instructions.

[0003] Therefore, an energy storage optimization scheduling method that can adapt to spot market fluctuations, integrate multi-timescale rolling corrections, and introduce game theory mechanisms has broad application prospects. Summary of the Invention

[0004] The purpose of this invention is to provide a power plant time-scale game-theoretic optimization scheduling method and system that adapts to fluctuations in the electricity spot market, and to solve the problems of low returns, high assessment risks, and damaged battery life caused by inaccurate electricity price prediction, lack of market game perspective, and single time scale in existing energy storage scheduling methods in the electricity spot market.

[0005] To achieve the above objectives, the present invention employs the following technical solutions.

[0006] A time-scale game-theoretic optimization scheduling method for power plants that adapts to fluctuations in the electricity spot market includes the following steps: S1. Obtain historical electricity spot market price data, new energy output data, and load data; generate a probability distribution of electricity price and output for future periods, and generate a set of uncertain scenarios containing various fluctuation characteristics based on the probability distribution; S2. Construct a two-layer game model between energy storage participants and electricity market operators; wherein, the upper layer is the market operator, which sets the market clearing price with the goal of minimizing system operating costs; the lower layer is the energy storage participants, which formulate day-ahead charging and discharging declaration plans with the goal of maximizing net profits; solve the equilibrium solution of the two-layer game model to obtain the day-ahead benchmark charging and discharging power curve and declaration strategy for the future preset period considering the impact of energy storage behavior on electricity price feedback; S3. Perform rolling optimization at preset time intervals; read the current actual operating status, the latest updated electricity price forecast information, and the uncertainty scenario set for the remaining period; construct an intraday optimization model, with the joint objective of maximizing expected returns and minimizing risks, to optimize the day-ahead benchmark charging and discharging power curve; during the optimization process, introduce a day-ahead plan tracking deviation penalty term to seek a balance between capturing arbitrage opportunities and reducing market deviation assessment costs, and output the intraday corrected charging and discharging power command; S4. Receive grid regulation signals, and under the premise of meeting the physical safety constraints of the energy storage system, perform high-frequency smoothing on the intraday corrected power command to generate the final control command and send it to the energy storage power control unit for execution; at the same time, monitor the deviation between the actual electricity price and the predicted value in real time. If the deviation exceeds the preset threshold, the rolling correction mechanism in step S3 is triggered.

[0007] Further, in step S2, the two-layer game model is a master-slave game model; the solution process of the master-slave game model includes: transforming the optimization problem of the lower-layer energy storage participants into KKT optimality conditions; linearizing the nonlinear complementary constraints in the KKT conditions using complementary relaxation conditions; substituting the linearized constraints into the optimization problem of the upper-layer market operation organization, transforming the two-layer programming problem into a single-layer mixed integer linear programming problem for solution; wherein, the objective function of the lower-layer energy storage participants includes battery life depreciation cost, and the battery life depreciation cost model adopts a linearized approximation model based on the rainflow counting method, which segments the battery charge and discharge depth and cycle life curve into linearized segments, converts charge and discharge behaviors at different depths into equivalent full cycle counts, and calculates the life depreciation cost per unit time in conjunction with battery replacement costs, which is directly included in the objective function of the lower-layer followers to suppress excessively frequent charge and discharge behaviors in pursuit of short-term electricity price differences.

[0008] Further, in step S3, the intraday optimization model is an intraday correction model based on sub-Bruker optimization; the intraday correction model based on sub-Bruker optimization specifically involves: constructing a fuzzy set of uncertainties in electricity price and output, wherein the fuzzy set is defined by the empirical distribution of the uncertainty scenario set and its confidence boundary; the optimization objective is: , Where P is the decision variable, i.e., charging and discharging power. Let Q be the feasible region, and let Q be any probability distribution within the fuzzy set D. Let R be a random variable, and let R be the arbitrage profit. The lifetime loss cost is the optimization objective of the model, which is to find the optimal scheduling strategy in the worst-case scenario across all possible probability distributions.

[0009] Furthermore, in step S3, the calculation formula for the day-ahead plan tracking deviation penalty term is as follows: , in, The power at time t is obtained through intraday optimization. Here, T represents the day-ahead baseline power, and T represents the end time of the intraday optimization. The penalty coefficient is dynamically adjusted; the dynamically adjusted penalty coefficient is dynamically set based on the time remaining until the previous day's reporting deadline, the current spot market deviation assessment rules, the current state of charge of the energy storage system, or market liquidity: the closer to the settlement point or the more stringent the market assessment, the more severe the penalty coefficient. The larger the value, the better.

[0010] Further, in step S1, a probability prediction model is used to generate the probability distribution of electricity price and power output for future periods. The probability prediction model uses a generative adversarial network and generates a typical scenario set containing multiple fluctuation characteristics based on Monte Carlo simulation or generative adversarial network. The process of generating the uncertain scenario set also includes a scenario reduction step: the initial large number of scenarios generated are reduced into a small number of representative typical scenarios through a clustering algorithm, and each typical scenario is assigned an occurrence probability to reduce the computational complexity of subsequent optimization.

[0011] Further, in step S4, the high-frequency smoothing process uses a low-pass filter; the grid regulation signal includes an automatic grid generation control command or a high-frequency regulation signal; when an automatic grid generation control command is received, the grid frequency regulation requirement is prioritized, and the power deviation caused by responding to the automatic generation control command is recorded as the power to be compensated; in the next intraday rolling optimization cycle, the power to be compensated is used as a correction amount for the initial SOC or corresponding equality constraints are added to ensure energy conservation and plan fulfillment rate over a long period.

[0012] Furthermore, this method is also applicable to scenarios where multiple energy storage participants aggregate to participate in the market. In step S2, the two-layer game model is extended to a non-cooperative game model with one leader and multiple followers. The upper-level leader is the market operation agency, and the lower-level leaders are agents of multiple independent energy storage participants. By iteratively solving the Nash equilibrium and the Stackelberg equilibrium of the market clearing price of each energy storage participant, the optimal bidding strategy of each participant considering the competitive relationship between energy storage is obtained, thus avoiding the price collapse effect caused by multiple energy storage participants charging and discharging at the same time.

[0013] Furthermore, it also includes the following steps: S5. Adaptive learning and parameter update: After each scheduling cycle, compare the deviations between the actual electricity price, actual power output and predicted and planned values; use the deviation data to update and correct the probabilistic prediction model in step S1 and the boundary parameters of the uncertain fuzzy set in step S3 online, so that the scheduling strategy can adaptively evolve with changes in market fluctuation characteristics.

[0014] A power plant time-scale game-theoretic optimization scheduling system adapted to fluctuations in the electricity spot market includes: a data acquisition and prediction module for executing step S1; a game-theoretic optimization calculation module for executing steps S2 and S3; and a real-time control execution module for executing step S4. The game optimization calculation module includes: a day-ahead game optimization unit, used to construct a two-layer game model between energy storage participants and electricity market operators; and an intraday rolling optimization unit, used to perform rolling optimization at preset time intervals.

[0015] Furthermore, it also includes an adaptive learning module for performing step S5.

[0016] The advantages of this invention are: This invention introduces a master-slave game mechanism, which fully considers the feedback impact of large-scale energy storage participation on the clearing price, avoids the strategy bias caused by treating energy storage as a price taker, and enables energy storage power stations to formulate application strategies that are more in line with market realities. In the intraday optimization, a dynamically adjusted daily plan tracking deviation penalty item was introduced, realizing an intelligent scheduling strategy that allows for flexible adjustments in the early stage to capture arbitrage opportunities and strict tracking in the later stage to reduce assessment risks, effectively balancing arbitrage returns and assessment costs. The intraday correction model based on split-brush optimization can ensure the optimality of the scheduling strategy in the worst-case scenario when there are uncertainties in electricity price and output, thus improving the robustness of the scheduling method. By directly incorporating battery life loss costs into the optimization objective function, excessively frequent charging and discharging behavior can be effectively suppressed, battery life can be extended, and the total life cycle cost of energy storage power stations can be reduced. It supports scenarios where multiple energy storage power stations participate in the market together. By constructing a non-cooperative game model with one leader and many followers, it can avoid the price collapse effect caused by multiple energy storage stations charging and discharging at the same time, and improve the overall profitability of multiple energy storage systems. The addition of adaptive learning and parameter update steps enables continuous optimization of the prediction model and uncertainty parameters based on actual market operation data, allowing the scheduling strategy to adaptively evolve with changes in market volatility characteristics and possessing long-term applicability.

[0017] In summary, the multi-timescale game-theoretic optimization scheduling method and system for power plants proposed in this invention, which adapts to fluctuations in the electricity spot market, can effectively cope with the strong volatility and uncertainty of electricity prices in the electricity spot market. It solves the problem that traditional day-ahead one-time optimization cannot adapt to intraday market changes, and significantly improves the returns obtained by energy storage power plants in the electricity spot market while reducing performance risks. Attached Figure Description

[0018] Figure 1 The flowchart shows the power plant time-scale game-theoretic optimization scheduling method adapted to fluctuations in the electricity spot market according to the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] Example 1 Please refer to Figure 1 This invention discloses a multi-timescale game-theoretic optimization scheduling method adapted to fluctuations in the electricity spot market, comprising the following steps: Step 1: Data Acquisition and Generation of Typical Scenario Sets Market price information covering specific regions is obtained from provincial power trading centers or the State Grid Corporation of China; renewable energy output forecast data is obtained from meteorological departments or professional forecasting agencies; and load data can be obtained from the power grid company. The acquired data is preprocessed by mapping it to the [0,1] interval using the min-max normalization method, and then converting it to a standard normal distribution by calculating the mean and standard deviation.

[0021] We use non-parametric kernel density estimation to fit the true probability distribution, assuming the historical dataset is... The probability density estimation formula at any point x is: , in, is the kernel function; h is the bandwidth parameter, which controls the smoothness of the curve.

[0022] Generative Adversarial Networks (GANs) are used to generate typical scene sets. The standard goal of a GAN is to find a Nash equilibrium, and its loss function is expressed as follows: , in, Data sampled from historical electricity prices / output; Let be a random vector sampled from a Gaussian distribution; The fake scene generated by the generator; The probability that the discriminator determines x to be true data; To generate scenarios with specific time periods or fluctuation characteristics, this example introduces the conditional information y of the predicted mean, at which point the formula becomes: , Since GANs generate a large number of scenes, they need to be reduced to a small number of typical scenes using clustering algorithms while preserving probabilistic characteristics. This requires calculating the probability of any two scene sequences. and Distance between: Assuming N scenes are generated, each scene The initial probability is usually set to After cluster reduction, the newly generated typical scenarios The probability of a given scene is the sum of the probabilities of all original scenes within its cluster: , in, For the original scene set contained in the k-th cluster, the above calculation process can realize the whole process from probability distribution fitting, high-fidelity scene generation, and construction of typical scene set.

[0023] Step Two: Construction and Solution of the Master-Slave Game Model A day-ahead optimization model based on a master-slave game is constructed for energy storage power stations and electricity market operators. In this two-level programming model, the energy storage power station acts as a follower, and the market operator acts as the leader. The objective of the upper-level model where the leader resides is to minimize the total system operating cost and determine the nodal marginal electricity price. : , in, Let be the active power output of the i-th generating unit at time t. Let be the load shedding amount at time t. This represents the net charging and discharging power of the energy storage power station at time t. The offload value represents the cost of not providing energy during a power outage; Let be the power generation cost function of the i-th generating unit.

[0024] The constraints for optimizing the above objective function include: (1) The energy storage system must meet the power balance condition: , in, and Let be the predicted wind power and solar power output at time t, respectively. Let be the total system load demand at time t; (2) Upper and lower limits of unit output: Each generator unit in the energy storage system has a maximum and minimum output. These upper and lower limits are usually determined by the unit's technical parameters and operating conditions. The active power output of the unit at time t must meet the following constraints: ; (3) Power flow constraints of the line: , in, This represents the transmission capacity limit of the l-th transmission line; is the power transfer distribution factor, representing the impact of node i injection on the power flow of line l.

[0025] In the day-ahead optimization model based on master-slave game theory, the energy storage power station, as a follower, aims to achieve the marginal electricity price at a given node. To maximize net profit: , in, For the battery life loss cost model, a linearized approximation method based on the rainflow counting method is adopted to decompose the charging and discharging power into positive and negative components: , in, and These represent the discharge power and charging power of the energy storage power station at time t, respectively. Then, the battery life loss cost model can be expressed as: , For ease of solution, it is simplified into a linear expression that is proportional to the throughput: , in, This is the lifetime depreciation cost factor per unit throughput.

[0026] The constraint on the optimization objective of the lower-level model is that the SOC of the energy storage system satisfies the dynamic equation: , in, and These are the energy storage discharge efficiency and charging efficiency, respectively; at the same time, the upper and lower limits of the State of Charge (SOC) must also be met, namely: .

[0027] Since the lower-level model is a convex optimization problem, it can be replaced with equivalent constraints using KKT conditions and incorporated into the upper-level model. KKT optimality conditions include: (1) Stationarity condition: , , in, The energy balance constraint is the dual variable of the stationarity constraint, representing the marginal contribution of the electricity at time t to the revenue at time t+1. The dual variable is introduced as a new variable into the upper-level model, replacing the original lower-level optimization process. (2) Original feasibility and dual feasibility: satisfy all inequality constraints and the dual variables are non-negative; (3) Complementary relaxation conditions: , , in, and These are the upper and lower limits of SOC, respectively, and the battery is fully charged ( Continuing to charge would be a waste of resources; if the battery is not fully charged ( If the constraint is not met, then the constraint will not take effect.

[0028] In the KKT condition transformation, for the inequality constraints in the lower-level optimization problem, there exist corresponding complementary relaxation conditions: In this embodiment, 'a' represents a slack variable of the constraint, specifically the upper limit constraint on battery capacity. , then define ; b represents the Lagrange multiplier or dual variable corresponding to the constraint. In this example, the multiplier corresponding to the above-mentioned upper limit of the electric quantity is denoted as . The aforementioned nonlinear conditions are incorporated into the solution of mixed-integer linear programming, introducing binary auxiliary variables. Given a sufficiently large constant M, construct the following set of linear constraints: , Substituting the above conditions as constraints into the upper-level objective function, the original two-level game problem is transformed into a single mixed-integer linear programming problem, which is then automatically adjusted by the Gurobi solver. The value of (0 or 1) is used to solve for the reference charge and discharge power curve and reporting strategy for the next 24 hours.

[0029] Step 3: Intraday Blob Rolling Optimization To construct an intraday correction model based on split-bar optimization, we first collect historical electricity price data and prediction error data from the same time point over the past N days to form an empirical sample set. These samples constitute the empirical distribution. Based on empirical distribution Centered on the Wasserstein distance Let be the radius, construct a confidence sphere for the probability distribution, and the actual market electricity price probability distribution has a probability of (1−α) falling within this sphere; define an uncertainty fuzzy set D, and let the random variable of electricity price be... If the true probability distribution is P, then the expression for the fuzzy set based on Wasserstein distance is: , in, It is an experience distribution based on N historical scenarios; It is the p-order Wasserstein distance, which measures the difference between distributions.

[0030] The goal of intraday correction is to maximize expected return under the worst-case scenario. The optimization objective can be expressed as: , in, This is a dynamic penalty coefficient. and These are the power commands at time t after optimization and correction for the intraday and day-ahead phases, respectively.

[0031] At each intraday scheduling time k (set to 15 minutes in this embodiment), the following rolling optimization process is executed: (1) Read the current SOC value and executed day-ahead plans of the energy storage power station. And the latest market electricity price observations; (2) Using the sliding window mechanism, select the prediction error samples from the most recent period and recalculate the current confidence radius. and update the fuzzy set D; (3) According to the strong duality theory, the above infinite-dimensional min-max problem can be equivalently transformed into a finite convex optimization problem: , The constraints of the above objective function are: , in, This is an auxiliary relaxation variable for the i-th scenario in the Bruker optimization; for a linear cost function, this constraint indicates that it is not necessary to know the true probability distribution, only historical data is needed. And represents the penalty regularization term for uncertainty. The robust optimal strategy can then be determined.

[0032] (4) The optimal charging and discharging power at the current moment is obtained by solving the above steps. The data is then sent to the energy storage converter for execution, and the initial state for the next moment is updated.

[0033] During the optimization process, a day-ahead tracking deviation penalty coefficient is introduced. The electricity spot market typically targets day-ahead reporting plans. With actual implementation The deviation is assessed, and in this embodiment, the assessment rules are set as follows: , in, As a severe penalty coefficient, This is the threshold for exemption from the exam.

[0034] In the intraday optimization goals, the expected future performance costs are approximated as penalties: , Define the dynamic penalty coefficient: , in, For market delivery time, This is either a sigmoid function or a step function; when the delivery time is long... At this point, a larger deviation is allowed to obtain new arbitrage opportunities; as settlement approaches, ,at this time and This approach aims to minimize the risk of hefty fines. The above logic enables an intelligent dispatch strategy that allows for flexible adjustments in the early stages of electricity market settlement and strict monitoring in the later stages.

[0035] Step 4: Real-time control and deviation triggering correction Upon receiving automatic generation control (AGC) commands or high-frequency regulation signals from the grid, and under the premise of meeting the physical safety constraints of the energy storage battery, the power commands corrected intraday are smoothed at the second / minute level to generate the final control commands and send them to the energy storage converter for execution. At the same time, the deviation between the actual electricity price and the predicted value is monitored in real time. If the deviation exceeds the preset threshold, the rolling correction mechanism in step three is triggered.

[0036] Upon receiving an AGC command, priority is given to meeting the grid frequency regulation requirements, and the power deviation caused by responding to the AGC command is recorded as "power to be compensated". In the next intraday rolling optimization cycle, the "power to be compensated" is used as a correction amount for the initial SOC or corresponding equality constraints are added to ensure energy conservation and plan fulfillment rate over long periods.

[0037] At the same time, the deviation between the actual electricity price and the predicted value is monitored in real time. If the deviation exceeds the preset threshold (20% in this embodiment), the rolling correction mechanism in step three is immediately triggered to recalculate the optimal charging and discharging power command for the remaining period in order to adapt to sudden changes in the market.

[0038] Step 5: Adaptive Learning and Parameter Update At the end of each scheduling cycle, the deviations between the actual electricity price, actual executed power, and predicted / planned values ​​are compared. The deviation data is used to train the probabilistic prediction model from step one online, updating model parameters and improving prediction accuracy. Simultaneously, based on the actual prediction error distribution, the boundary parameters of the uncertainty fuzzy set from step three are adjusted to match the robustness level with the actual uncertainty, avoiding overly conservative or aggressive scheduling strategies. Through this adaptive learning process, the scheduling strategy continuously optimizes and evolves in response to changes in market volatility characteristics.

[0039] Example 2 This invention also provides a multi-timescale game-theoretic optimization scheduling system adapted to fluctuations in the electricity spot market, comprising a data acquisition and prediction module, a game-theoretic optimization calculation module, a real-time control execution module, and an adaptive learning module. The data acquisition and prediction module performs the function of step one in Embodiment 1; the game-theoretic optimization calculation module includes a day-ahead game-theoretic optimization unit and an intraday rolling optimization unit, which respectively perform the functions of steps two and three in Embodiment 1; the real-time control execution module performs the function of step four in Embodiment 1; and the adaptive learning module performs the function of step five in Embodiment 1.

[0040] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A power plant time-scale game-theoretic optimization scheduling method adapted to fluctuations in the electricity spot market, characterized in that, Includes the following steps: S1. Obtain historical electricity spot market price data, new energy output data, and load data; generate a probability distribution of electricity price and output for future periods, and generate a set of uncertain scenarios containing various fluctuation characteristics based on the probability distribution; S2. Construct a two-layer game model between energy storage participants and electricity market operators; wherein, the upper layer is the market operator, which sets the market clearing price with the goal of minimizing system operating costs; the lower layer is the energy storage participants, which formulate day-ahead charging and discharging declaration plans with the goal of maximizing net profits; solve the equilibrium solution of the two-layer game model to obtain the day-ahead benchmark charging and discharging power curve and declaration strategy for the future preset period considering the impact of energy storage behavior on electricity price feedback; S3. Perform rolling optimization at preset time intervals; read the current actual operating status, the latest updated electricity price forecast information, and the uncertainty scenario set for the remaining period; construct an intraday optimization model, with the joint objective of maximizing expected returns and minimizing risks, to optimize the day-ahead benchmark charging and discharging power curve; during the optimization process, introduce a day-ahead plan tracking deviation penalty term to seek a balance between capturing arbitrage opportunities and reducing market deviation assessment costs, and output the intraday corrected charging and discharging power command; S4. Receive grid regulation signals, and under the premise of meeting the physical safety constraints of the energy storage system, perform high-frequency smoothing on the intraday corrected power command to generate the final control command and send it to the energy storage power control unit for execution; at the same time, monitor the deviation between the actual electricity price and the predicted value in real time. If the deviation exceeds the preset threshold, the rolling correction mechanism in step S3 is triggered.

2. The power plant time-scale game-theoretic optimal scheduling method for adapting to fluctuations in the electricity spot market as described in claim 1, characterized in that, In step S2, the two-layer game model is a master-slave game model. The solution process of the master-slave game model includes: transforming the optimization problem of the lower-layer energy storage participants into KKT optimality conditions; linearizing the nonlinear complementary constraints in the KKT conditions using complementary relaxation conditions; substituting the linearized constraints into the optimization problem of the upper-layer market operation organization, and transforming the two-layer programming problem into a single-layer mixed integer linear programming problem for solution; wherein, the objective function of the lower-layer energy storage participants includes battery life depreciation cost. The battery life depreciation cost model adopts a linearized approximation model based on the rainflow counting method, which segments and linearizes the battery charge / discharge depth and cycle life curve, converts charge / discharge behavior at different depths into equivalent full cycle counts, and calculates the life depreciation cost per unit time in conjunction with battery replacement cost, and directly includes it in the objective function of the lower-layer followers to suppress excessively frequent charge / discharge behavior in pursuit of short-term electricity price differences.

3. The power plant time-scale game-theoretic optimal scheduling method adapted to fluctuations in the electricity spot market according to claim 1, characterized in that, In step S3, the intraday optimization model is an intraday correction model based on sub-Bruker optimization; specifically, the intraday correction model based on sub-Bruker optimization involves: constructing a fuzzy set of uncertainties in electricity price and output, wherein the fuzzy set is defined by the empirical distribution of the uncertainty scenario set and its confidence boundary; the optimization objective is: ， Where P is the decision variable, i.e., charging and discharging power. Let Q be the feasible region, and let Q be any probability distribution within the fuzzy set D. Let R be a random variable, and let R be the arbitrage profit. The lifetime loss cost is the optimization objective of the model, which is to find the optimal scheduling strategy in the worst-case scenario across all possible probability distributions.

4. The power plant time-scale game-theoretic optimal scheduling method for adapting to fluctuations in the electricity spot market as described in claim 1, characterized in that, In step S3, the calculation formula for the day-ahead plan tracking deviation penalty is as follows: ， in, The power at time t is obtained through intraday optimization. Here, T represents the day-ahead baseline power, and T represents the end time of the intraday optimization. The penalty coefficient is dynamically adjusted; the dynamically adjusted penalty coefficient is dynamically set based on the time remaining until the previous day's reporting deadline, the current spot market deviation assessment rules, the current state of charge of the energy storage system, or market liquidity: the closer to the settlement point or the more stringent the market assessment, the more severe the penalty coefficient. The larger the value, the better.

5. The power plant time-scale game-theoretic optimal scheduling method for adapting to fluctuations in the electricity spot market as described in claim 1, characterized in that, In step S1, a probability prediction model is used to generate the probability distribution of electricity price and power output for future periods. The probability prediction model uses a generative adversarial network and generates a typical scenario set containing multiple fluctuation characteristics based on Monte Carlo simulation or generative adversarial network. The process of generating the uncertain scenario set also includes a scenario reduction step: the initial large number of scenarios generated are reduced into a small number of representative typical scenarios through a clustering algorithm, and each typical scenario is assigned an occurrence probability to reduce the computational complexity of subsequent optimization.

6. The power plant time-scale game-theoretic optimal scheduling method for adapting to fluctuations in the electricity spot market as described in claim 1, characterized in that, In step S4, the high-frequency smoothing process uses a low-pass filter; the grid regulation signal includes an automatic generation control command or a high-frequency regulation signal; when an automatic generation control command is received, the grid frequency regulation requirement is prioritized, and the power deviation caused by responding to the automatic generation control command is recorded as the power to be compensated; in the next intraday rolling optimization cycle, the power to be compensated is used as a correction amount for the initial SOC or corresponding equality constraints are added to ensure energy conservation and plan fulfillment rate over a long period.

7. The power plant time-scale game-theoretic optimal scheduling method for adapting to fluctuations in the electricity spot market as described in claim 1, characterized in that, This method is also applicable to scenarios where multiple energy storage participants aggregate to participate in the market. In step S2, the two-layer game model is extended to a non-cooperative game model with one leader and multiple followers. The upper-level leader is the market operation agency, and the lower-level leaders are agents of multiple independent energy storage participants. By iteratively solving the Nash equilibrium and the Stackelberg equilibrium of the market clearing price of each energy storage participant, the optimal bidding strategy of each participant considering the competitive relationship between energy storage is obtained, thus avoiding the price collapse effect caused by multiple energy storage participants charging and discharging at the same time.

8. The power plant time-scale game-theoretic optimal scheduling method for adapting to fluctuations in the electricity spot market according to any one of claims 1 to 7, characterized in that, It also includes the following steps: S5. Adaptive learning and parameter update: After each scheduling cycle, compare the deviations between the actual electricity price, actual power output and predicted and planned values; use the deviation data to update and correct the probabilistic prediction model in step S1 and the boundary parameters of the uncertain fuzzy set in step S3 online, so that the scheduling strategy can adaptively evolve with changes in market fluctuation characteristics.

9. A power plant time-scale game-theoretic optimization scheduling system adapted to fluctuations in the electricity spot market, characterized in that, include: The data acquisition and prediction module is used to perform step S1 as described in claim 1; A game optimization calculation module is used to execute steps S2 and S3 as described in claim 1; A real-time control execution module is used to execute step S4 as described in claim 1; The game optimization calculation module includes: a day-ahead game optimization unit, used to construct a two-layer game model between energy storage participants and power market operators; The intraday rolling optimization unit is used to perform rolling optimization at preset time intervals.

10. The power plant time-scale game-theoretic optimization scheduling system adapted to fluctuations in the electricity spot market according to claim 9, characterized in that, Also includes: An adaptive learning module is used to perform step S5 as described in claim 8.

Images