A new energy subject-oriented compliance confidence ladder day-ahead bidding optimization method

Abstract

The application discloses a new energy subject-oriented performance confidence ladder day-ahead bidding optimization method, which comprises the following steps: based on historical output data of a new energy station, statistical characteristic analysis is carried out, the upper limit of a new energy subject declaration is corrected, and the confidence of the new energy subject reportable output is described according to a probability curve; the corrected reportable output is divided into multiple ladder intervals, a ladder declaration curve is constructed according to the confidence level of each interval, and the coupling correlation between the new energy output uncertainty and the dispatchable generating capacity is realized; further, the market clearing result is used to update the strategy network parameters, the adaptive ladder operation control strategy of the new energy subject is obtained through iterative training, and is used for day-ahead power market declaration, so that the generating plan execution accuracy and the new energy station operation stability are improved.

Description

Technical Field

[0001] This invention belongs to the field of power system operation control and intelligent decision-making technology, specifically relating to a day-ahead pricing optimization method for new energy entities based on performance confidence tiers. Background Technology

[0002] With the continuous and rapid growth of installed capacity of new energy sources, represented by wind power and photovoltaics, new energy is gradually shifting from guaranteed consumption to market-based consumption, participating in spot market transactions using a "volume-based bidding" method. Under the "day-ahead clearing + real-time balancing" mechanism of the spot market, new energy power generation entities need to generate power generation plans in advance based on forecast information. However, the output of new energy is significantly affected by natural conditions, and there is often a large deviation between the output curve declared day-ahead and the actual output, which leads to an increase in the deviation of power generation plan execution, affecting the resource utilization efficiency of new energy power generation and the operational stability of the power system.

[0003] In existing technologies, renewable energy entities typically generate application curves using fixed parameters or empirical rules. However, this approach lacks an effective characterization of the uncertainty in renewable energy output, making it difficult to dynamically adjust power generation plans based on different output levels. Furthermore, as the scale of renewable energy grid connection expands, complex coupling relationships exist between the operating states of different power generation entities. Traditional methods based on static game theory or analytical models struggle to accurately characterize the operating characteristics in a multi-entity dynamic interaction environment, and also fail to achieve adaptive control in complex operating scenarios. Therefore, it is necessary to construct a method that can both reflect the uncertainty of renewable energy output through a stepped application curve and achieve adaptive optimization and dynamic adjustment of power generation plans in complex multi-entity operating environments. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing application methods by providing a day-ahead bidding optimization method for renewable energy entities, based on a tiered performance confidence system. This invention introduces a method for representing the exceedance probability of renewable energy output credibility and an adaptive operation control algorithm based on reinforcement learning. This achieves dynamic matching between renewable energy power generation plans and actual output characteristics, improving the resource utilization efficiency and operational stability of renewable energy power generation, enhancing the adaptive adjustment capability of renewable energy entities to output fluctuations, thereby improving the level of renewable energy absorption and the safe operation capability of the power system.

[0005] The objective of this invention is achieved through the following technical solution: a day-ahead pricing optimization method for new energy entities based on performance confidence tiers, comprising the following steps: (1) Data preprocessing and statistical correction of the upper limit of the application: Based on the historical actual output of new energy power plants and the corresponding day-ahead predicted output data, establish the prediction error time series of new energy power plants; calculate the positive correction deviation that meets the reliability requirements and add it to the predicted output; combine the installed capacity constraints to determine the upper limit of the output power of the day-ahead application curve of the unit at each time, so as to improve the accuracy of new energy power generation capacity assessment and the level of adjustable output resource utilization. (2) Confidence representation and tiered application of new energy output: Based on the transcendence probability theory and the revised application upper limit, the actual output ratio of new energy and its confidence probability are statistically represented, and a set of transcendence probability curves of the output ratio of new energy power plants is constructed; the applied output is divided into segments, allowing new energy entities to construct differentiated tiered application curves according to their performance risk preferences, so as to realize the hierarchical representation of the uncertainty of new energy output and the refined configuration of power generation plans. (3) Markov decision modeling for new energy participation in spot trading: Construct a multi-entity reinforcement learning environment for the power market, model each new energy power generation entity as an intelligent agent with autonomous learning capabilities, and use the step-by-step bidding curve in step (2) as the decision action of the intelligent agent; combine the publicly disclosed information of the market trading institution with the individual observation information of the power generation entity to simulate the Markov bidding decision process under information asymmetry in the real market environment, and realize the dynamic correlation modeling between the operating status of multiple entities and the power generation plan; (4) Strategy training and application execution based on improved reinforcement learning algorithm: The output fluctuation characteristics and operating status change characteristics of new energy entities are evaluated by a distributed value network. A risk distortion operator is introduced to realize the adaptive expression of different operating preferences of the entities. The key interaction relationship in the multi-entity environment is dynamically modeled by combining the attention mechanism. The strategy network parameters are updated according to the market clearing results. The adaptive tiered operation control strategy of new energy entities is obtained through iterative training and used for day-ahead power market application.

[0006] Further, the positive correction deviation in step (1) is defined as: ; in, For new energy units r At any moment t Positive correction bias; For new energy units r At any moment t The actual output sequence; For new energy units r At any moment t The predicted output sequence; Quantile0.95{‧} represents the value at the 95th percentile in the sample sequence; max{‧} represents the maximum value in the sample sequence.

[0007] Furthermore, the upper limit of the declared output in step (1) is defined as: ; in, For new energy units r At any moment t The revised upper limit of the declared power output; For new energy units r At any moment t The predicted output value; For new energy units r At any moment t Positive correction bias; For new energy units r The installed capacity; min{‧} represents the minimum value of the calculated sample sequence.

[0008] Furthermore, the actual output ratio in step (2) is defined as: ; in, This represents the actual output ratio of new energy sources after positive deviation correction; For new energy units r At any moment t The revised upper limit of the declared power output; For new energy units r At any moment t The predicted output value.

[0009] Furthermore, the exceedance probability curve in step (2) is: ; Wherein, EPC is the exceedance probability, and its value ranges from [0,1]. X It is a random variable; The preset threshold is used; Pr(‧) represents the probability of the event occurring.

[0010] Furthermore, in step (3), each new energy power generation entity is modeled as an independent intelligent agent, and the participation of new energy in spot trading is modeled as a single-step Markov decision process. That is, in the scenario where new energy units participate in the day-ahead market declaration, each intelligent agent submits only once a trading day a stepped declaration curve containing the time periods of the next day, and the market operator performs a unified clearing based on the results of all declarations.

[0011] Furthermore, the stepped declaration curve in step (3) takes the form of: ; in, For new energy units r At any momentt A tiered application curve; For new energy units r At any moment t No. m The confidence level of the predicted output of the segment; For new energy units r At any moment t No. m The declared output and corresponding declared price of the segment; The total number of segments; based on the output exceedance probability curve of the new energy unit, the corresponding predicted output value is calculated in reverse to realize the mapping from the output confidence level to the output value segment point, thereby forming the actual action of the intelligent agent.

[0012] Furthermore, the publicly disclosed information in step (3) includes the system dispatch load forecast curve for the following day and the average clearing price for historical trading days; the individual observation information is data that the new energy entities themselves do not disclose, including... d The daily new energy power output forecast curve, power output confidence parameters, historical trading day submission curves, and winning bid results are denoted as: ; in, For the observation input of new energy intelligent agents; for d Daily total social load forecast vector; Let d-1 be the market-average clearing electricity price vector; For new energy units r exist d The predicted output vector for the day; For new energy units r exist d The reporting curve for -1 day; For new energy units r exist d -1 day's winning bid results; For new energy units r The output confidence characteristics.

[0013] Furthermore, the reinforcement learning algorithm characterizes payoff uncertainty through distributed value assessment, introduces a risk distortion operator to achieve an endogenous expression of the subject's risk preferences, and combines an attention mechanism to dynamically model key interactions in the game environment; based on the payoff distribution output by the distributed Critic, a risk-distorted subjective value function is constructed, which is calculated as follows: ; in, Let be the subjective value function of the proposed reinforcement learning algorithm; s For the observation input of new energy intelligent agents;a A set of actions for a new energy intelligent agent; The quantile position; It is a monotonically increasing twisting function; For the first k The cumulative probability of each quantile; K This represents the number of quantile fits.

[0014] Furthermore, by continuously updating the network parameters of the intelligent agent strategy based on the market clearing results, the adaptive optimization of the application strategy for new energy entities can be achieved.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention transforms random power output into a performance confidence index by constructing a new energy power output exceedance probability curve, and further discretizes the continuous power output space into multiple confidence steps, thus realizing the transformation from "random power output" to "segmentable decision variables". Compared with the traditional method of directly reporting based on predicted values, this invention can more accurately reflect the uncertainty characteristics of new energy power output, improve the accuracy of new energy power generation capacity assessment and power generation plan formulation, and enhance the matching degree between power generation plan and actual power output.

[0016] (2) This invention designs a stepped declaration curve based on the confidence level of performance, enabling output ranges with different confidence levels to adopt differentiated parameter configuration methods, thereby achieving hierarchical management and refined control of the uncertainty of new energy output. Compared with the existing single parameter configuration method, this method can reduce the deviation of plan execution caused by prediction errors, improve the utilization efficiency of new energy power generation resources, and enhance the adaptability and operational stability of new energy power plants to output fluctuations.

[0017] (3) This invention models the new energy entity as a multi-agent system and uses reinforcement learning algorithms to iteratively optimize the control strategy in a dynamic operating environment. It can capture the dynamic interaction relationships between different entities and realize the transformation from static rule-driven to adaptive learning optimization. Compared with traditional methods based on empirical rules or analytical models, this invention has stronger environmental adaptability and online optimization capabilities. It can dynamically adjust the power generation plan parameters according to changes in operating status, improve the power output regulation capability and operation control accuracy of new energy entities, and enhance the level of multi-entity collaborative operation. Attached Figure Description

[0018] Figure 1 The flowchart below shows the day-ahead pricing optimization method for new energy entities based on the performance confidence ladder of the present invention. Figure 2 This is a schematic diagram of the new energy performance confidence tiered declaration method of the present invention; Figure 3 This is a schematic diagram of the distributed multi-agent proximal policy optimization algorithm with enhanced risk perception and attention according to the present invention. Detailed Implementation

[0019] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of methods consistent with some aspects of the invention as detailed in the appended claims.

[0020] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0021] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0022] The present invention will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0023] See Figure 1 The present invention provides a method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities, which specifically includes the following steps: (1) Data preprocessing and statistical correction of the upper limit of the application: Based on the historical actual output of new energy power plants and the corresponding day-ahead predicted output data, establish the prediction error time series of new energy power plants; calculate the positive correction deviation that meets the reliability requirements and add it to the predicted output; combine the installed capacity constraints to determine the upper limit of the output power of the day-ahead application curve of the unit at each time, so as to improve the accuracy of new energy power generation capacity assessment and the level of adjustable output resource utilization. Furthermore, the positive correction deviation that meets the reliability requirements is defined as: in, For new energy units r At any moment tPositive correction bias; For new energy units r At any moment t The actual output sequence; For new energy units r At any moment t The predicted output sequence; Quantile0.95{‧} represents the value at the 95th percentile in the sample sequence; max{‧} represents the maximum value in the sample sequence; The upper limit of the output power of the reported curves at each time point is defined as follows: in, For new energy units r At any moment t The revised upper limit of the declared power output; For new energy units r At any moment t The predicted output value; For new energy units r The installed capacity; min{‧} represents the minimum value of the calculated sample sequence; The actual output ratio after positive deviation correction is defined as follows: in, This represents the actual output ratio of new energy sources after positive deviation correction; It should be noted that this invention, without altering the nature of the prediction error, selects a 95% confidence level as the cutoff benchmark, effectively covering most normalized fluctuation scenarios and eliminating extremely low-probability positive extreme errors, thus achieving an optimal trade-off between system reliability and scheduling economy. After the above correction process, the originally bidirectional distribution of prediction errors is transformed into a unidirectional distribution, ensuring that in non-extreme scenarios, the prediction error of new energy units at any given time should be non-positive, and the actual output ratio should not exceed 1. This prevents new energy entities from missing the opportunity to obtain high-price day-ahead market benefits from positive deviation electricity from the application source, improving trading flexibility and the full utilization of resources.

[0024] (2) Confidence representation and tiered application of new energy output: Based on the transcendence probability theory and the revised application upper limit, the actual output ratio of new energy and its confidence probability are statistically represented, and a set of transcendence probability curves of the output ratio of new energy power plants is constructed. On this basis, the application output is divided into segments, allowing new energy entities to construct differentiated tiered application curves according to their performance risk preferences, so as to realize the hierarchical representation of the uncertainty of new energy output and the refined configuration of power generation plans. Furthermore, the exceedance probability curve is: Wherein, EPC is the exceedance probability, and its value ranges from [0,1]. X It is a random variable; For a specific threshold; Pr(‧) represents the probability of the event occurring. For example, if the prediction error... for The probability of exceeding the limit is p Then, according to the definition of the transcendental probability curve, we can deduce that ; In this embodiment, the schematic diagram of the proposed new energy performance confidence tiered declaration method is as follows: Figure 2 As shown, the exceedance probability curve can intuitively reveal the tail characteristics and extreme value risks of a random variable distribution through a probability-value mapping relationship. Specifically, each point on the exceedance curve corresponds to a set of "exceedance probability-output ratio" relationships, reflecting the confidence level that the actual output of the new energy unit is greater than or equal to a certain ratio level. For example, when the exceedance probability curve shows "when the output ratio is 30%, the exceedance probability is 90%", it means that in the historical statistical sample of the new energy unit, the probability that the actual output of the new energy is not lower than the revised upper limit of 30% is 90%, that is, the output level has a 90% confidence guarantee. For any given confidence level, the corresponding theoretical output threshold of the new energy can be obtained by inverse mapping through the exceedance probability curve, realizing the transformation of uncertain output into a quantifiable and controllable deterministic parameter.

[0025] (3) Markov decision modeling for new energy participation in spot trading: Construct a multi-entity reinforcement learning environment for the power market, model each new energy power generation entity as an intelligent agent with autonomous learning capabilities, and use the step-by-step bidding curve in step (2) as the decision action of the intelligent agent; combine the publicly disclosed information of the market trading institution with the individual observation information of the power generation entity to simulate the Markov bidding decision process under information asymmetry in the real market environment, and realize the dynamic correlation modeling between the operating status of multiple entities and the power generation plan; Furthermore, in the reinforcement learning environment, each new energy power generation entity is modeled as an independent intelligent agent. These agents are in a non-cooperative game relationship and exhibit significant physical heterogeneity (such as installed capacity and prediction error distribution characteristics) and behavioral heterogeneity (such as risk preference). The participation of new energy in spot trading is modeled as a single-step Markov decision process. Specifically, in the scenario where new energy units participate in day-ahead market declarations, each intelligent agent submits only one tiered declaration curve containing all time periods of the following day within a single trading day. The market operator performs a unified clearing based on all declaration results. Furthermore, the action space in the Markov decision-making process is a stepped application curve for the new energy agent, but it does not directly generate discrete application actions. Instead, it combines the output exceedance probability curve of the new energy unit with the output confidence and price profile to form segmented application actions: in, For new energy units r At any moment t A tiered application curve; For new energy units r At any moment t No. m The confidence level of the predicted output of the segment; For new energy units r At any moment t No. m The declared output and corresponding declared price of the segment; The total number of segments. Based on the output exceedance probability curve of the new energy unit, the corresponding predicted output value can be calculated in reverse, realizing the mapping from the output confidence level to the output value segmentation point, and thus forming the actual action of the intelligent agent; Furthermore, in the Markov decision-making process, the state space is jointly constructed from the day-ahead market observable information of each agent, divided into public observation information and individual observation information. Among these, publicly disclosed information includes the information from the following day (…). d The system load forecast curve for the day, and historical trading days ( d Average clearing electricity price (e.g., -1 day); individual observation information is data that the renewable energy entities themselves do not disclose, including... d Daily renewable energy output forecast curves, output confidence parameters, historical trading day application curves and winning bid results, etc. This can be recorded as: in, For the observation input of new energy intelligent agents; for d Daily total social load forecast vector; Let d-1 be the market-average clearing electricity price vector; For new energy units r exist d The predicted output vector for the day; For new energy units r exist d The reporting curve for -1 day; For new energy units r exist d -1 day's winning bid results; For new energy units r The output confidence characteristics; Furthermore, the Markov decision, with net revenue per kilowatt-hour as the objective reward function, is defined as follows: in, For the reward function of new energy intelligent agents; and They are time points t Market-clearing electricity prices and deviation penalty prices; For new energy units r In the spot market t The winning bid volume is determined by a market clearing model after all entities have submitted their bids. For new energy units r At any moment t The actual power generation capacity.

[0026] It should be noted that the new energy multi-agent takes load forecasting, new energy output forecasting, historical market clearing prices and winning bids as state inputs, and tiered bidding strategies as action outputs. Under a given market clearing mechanism, it executes the bidding and clearing process to obtain the market electricity price and winning bids, and constructs a reward function based on the revenue results.

[0027] (4) Strategy training and application execution based on improved reinforcement learning algorithm: The output fluctuation characteristics and operating status change characteristics of new energy entities are evaluated by a distributed value network. A risk distortion operator is introduced to realize the adaptive expression of different operating preferences of the entities. The key interaction relationship in the multi-entity environment is dynamically modeled by combining the attention mechanism. The strategy network parameters are updated according to the market clearing results. The adaptive tiered operation control strategy of new energy entities is obtained through iterative training and used for day-ahead power market application to improve the accuracy of power generation plan execution and the operation stability of new energy power plants. In this example, the reinforcement learning algorithm is an improvement on the traditional multi-agent proximal policy optimization algorithm. The algorithm principle is as follows: Figure 3 As shown. Based on the return distribution of the distributed Critic output, a risk-distorted subjective value function is constructed, which is calculated as follows: in, Let be the subjective value function of the proposed reinforcement learning algorithm; s For the observation input of new energy intelligent agents; a A set of actions for a new energy intelligent agent; The quantile position; It is a monotonically increasing twisting function; For the first k The cumulative probability of each quantile; K The number of quantiles fitted; It should be noted that when the risk aversion level is high, the risk distortion operator will assign higher weights to the left side of the payoff distribution, prompting the agent to tend to adopt a more conservative low-price bidding strategy. During the training process, if the high-price strategy gradually shows better payoff feedback in the group game interaction, the attention to the payoff on the right side of the distribution will be increased accordingly, thereby weakening the inhibitory effect on high-payoff but high-risk strategies.

[0028] Furthermore, regarding intelligent agents i Encode its observations and actions into feature vectors By mapping and transforming queries, keys, and values, intelligent agents can be characterized. i Other intelligent agents j The strength of the correlation between them in price game and strategy interaction, and its attention weight is expressed as: in, Attention weights; , For a learnable parameter matrix, For feature dimensions; Furthermore, the features of all opponents are weighted and aggregated to obtain a global context vector that incorporates key strategy interaction information, defined as: in, For global context vectors; The opponent's feature vector; Furthermore, by continuously updating the network parameters of the intelligent agent strategy based on the market clearing results, the adaptive optimization of the application strategy for new energy entities can be achieved.

[0029] It should be noted that the above are specific improvements to the reinforcement learning algorithm, which iteratively update the agent's policy and determine whether the policy converges through multiple rounds of training. If it does not converge, the policy is continuously optimized and iterated until a stable optimal tiered application policy output is obtained. This enables new energy entities to make market application decisions that coordinate risk and return under the condition of considering the uncertainty of power output.

[0030] In summary, this invention, considering the uncertainty of renewable energy output, constructs a performance confidence ladder declaration curve based on transcendental probability theory, achieving a hierarchical confidence representation and structured expression of renewable energy declarations. Furthermore, it introduces a multi-agent reinforcement learning method to adaptively optimize the power generation plan generation parameters, thereby improving the consistency between renewable energy power generation plans and actual output, and enhancing the adaptive adjustment capability of renewable energy power plants to random fluctuations. Compared with traditional declaration methods based on single predicted values ​​or static rules, this invention can more fully utilize renewable energy output information, improve the efficiency of renewable energy power generation resource utilization and the accuracy of plan execution, enhance the flexibility and environmental adaptability of operation control strategies, and improve the coordination level between renewable energy entities and the operational needs of the power system. Simultaneously, through continuous iterative optimization of the reinforcement learning algorithm, the adaptive adjustment capability of operation control parameters can be further improved, enhancing the operational stability of renewable energy power plants and the renewable energy absorption capacity, providing technical support for the safe and stable operation of high-proportion renewable energy integration into the power system.

[0031] 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 scope of protection of the present invention.

[0032] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities, characterized in that, Includes the following steps: (1) Data preprocessing and statistical correction of the upper limit of the application: Based on the historical actual output of new energy power plants and the corresponding day-ahead predicted output data, establish the prediction error time series of new energy power plants; calculate the positive correction deviation that meets the reliability requirements and add it to the predicted output; combine the installed capacity constraints to determine the upper limit of the output power of the day-ahead application curve of the unit at each time, so as to improve the accuracy of new energy power generation capacity assessment and the level of adjustable output resource utilization. (2) Confidence representation and tiered application of new energy output: Based on the transcendence probability theory and the revised application upper limit, the actual output ratio of new energy and its confidence probability are statistically represented, and a set of transcendence probability curves of the output ratio of new energy power plants is constructed; the applied output is divided into segments, allowing new energy entities to construct differentiated tiered application curves according to their performance risk preferences, so as to realize the hierarchical representation of the uncertainty of new energy output and the refined configuration of power generation plans. (3) Markov decision modeling for new energy participation in spot trading: Construct a multi-entity reinforcement learning environment for the power market, model each new energy power generation entity as an intelligent agent with autonomous learning capabilities, and use the step-by-step bidding curve in step (2) as the decision action of the intelligent agent; combine the publicly disclosed information of the market trading institution with the individual observation information of the power generation entity to simulate the Markov bidding decision process under information asymmetry in the real market environment, and realize the dynamic correlation modeling between the operating status of multiple entities and the power generation plan; (4) Policy training and application execution based on improved reinforcement learning algorithm: The output fluctuation characteristics and operating status change characteristics of new energy entities are evaluated through distributed value network, risk distortion operator is introduced to realize the adaptive expression of different operating preferences of entities, and attention mechanism is combined to dynamically model the key interaction relationship in multi-entity environment; The strategy network parameters are updated based on the market clearing results. An adaptive tiered operation control strategy for new energy entities is obtained through iterative training and used for day-ahead electricity market declarations.

2. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, The positive correction deviation in step (1) is defined as: ； in, For new energy units r At any moment t Positive correction bias; For new energy units r At any moment t The actual output sequence; For new energy units r At any moment t The predicted output sequence; Quantile0.95{‧} represents the value at the 95th percentile in the sample sequence; max{‧} represents the maximum value in the sample sequence.

3. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, The upper limit of the declared output in step (1) is defined as follows: ； in, For new energy units r At any moment t The revised upper limit of the declared power output; For new energy units r At any moment t The predicted output value; For new energy units r At any moment t Positive correction bias; For new energy units r The installed capacity; min{‧} represents the minimum value of the calculated sample sequence.

4. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, The actual output ratio in step (2) is defined as follows: ； in, This represents the actual output ratio of new energy sources after positive deviation correction; For new energy units r At any moment t The revised upper limit of the declared power output; For new energy units r At any moment t The predicted output value.

5. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, The exceedance probability curve in step (2) is as follows: ； Wherein, EPC is the exceedance probability, and its value ranges from [0,1]. X It is a random variable; The preset threshold is used; Pr(‧) represents the probability of the event occurring.

6. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, In step (3), each new energy power generation entity is modeled as an independent intelligent agent, and the participation of new energy in spot trading is modeled as a single-step Markov decision process. That is, in the scenario where new energy units participate in the day-ahead market declaration, each intelligent agent submits only once a trading day a stepped declaration curve containing the time periods of the next day, and the market operator performs a unified clearing based on the results of all declarations.

7. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, The stepped declaration curve in step (3) is in the form of: ； in, For new energy units r At any moment t A tiered application curve; For new energy units r At any moment t No. m The confidence level of the predicted output of the segment; For new energy units r At any moment t No. m The declared output and corresponding declared price of the segment; The total number of segments; based on the output exceedance probability curve of the new energy unit, the corresponding predicted output value is calculated in reverse to realize the mapping from the output confidence level to the output value segment point, thereby forming the actual action of the intelligent agent.

8. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, The publicly disclosed information in step (3) includes the system dispatch load forecast curve for the next day and the average clearing price of historical trading days; Individual observation information refers to data that the new energy entities themselves do not disclose, including... d The daily new energy power output forecast curve, power output confidence parameters, historical trading day submission curves, and winning bid results are denoted as: ； in, For the observation input of new energy intelligent agents; for d Daily total social load forecast vector; Let d-1 be the market-average clearing electricity price vector; For new energy units r exist d The predicted output vector for the day; For new energy units r exist d The reporting curve for -1 day; For new energy units r exist d -1 day's winning bid results; For new energy units r The output confidence characteristics.

9. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, The reinforcement learning algorithm characterizes payoff uncertainty through distributed value assessment, introduces a risk distortion operator to achieve an endogenous expression of the subject's risk preference, and combines an attention mechanism to dynamically model key interaction relationships in the game environment. Based on the return distribution of the distributed Critic output, a risk-distorted subjective value function is constructed, which is calculated as follows: ； in, Let be the subjective value function of the proposed reinforcement learning algorithm; s For the observation input of new energy intelligent agents; a A set of actions for a new energy intelligent agent; The quantile position; It is a monotonically increasing twisting function; For the first k The cumulative probability of each quantile; K This represents the number of quantile fits.

10. The method for optimizing day-ahead pricing based on performance confidence tiers for new energy entities according to claim 1, characterized in that, By continuously updating the network parameters of the intelligent agent strategy based on the market clearing results, the adaptive optimization of the application strategy for new energy entities can be achieved.

Images