Optimization system and method for participation of electric vehicle cluster in electric power standby market

Abstract

The invention discloses an optimization system and method for participation of an electric vehicle cluster in an electric power standby market, and is applied to the field of an electric power auxiliary service market, and the system comprises a market uncertainty modeling module which is used for constructing a joint uncertainty model of market price and standby calling probability, wherein the model comprises a probability distribution model and a conditional probability model; the risk quantitative evaluation module is used for evaluating risk exposure degrees of different decision-making schemes, including establishing a multi-period risk accumulation model, and comprehensively evaluating balance indexes of expected income, risk openness and opportunity cost; the robust optimization decision-making module constructs a decision-making model according to the market uncertainty modeling module and the risk quantitative evaluation module, and generates an optimal strategy that the electric vehicle cluster participates in the standby market; the adaptive learning optimization module is used for optimizing the decision model through reinforcement learning; according to the invention, risk-controllable revenue maximization can be realized, so that the electric vehicle cluster can efficiently participate in the electric power standby market.

Description

Technical Field

[0001] The present invention relates to the field of power auxiliary service market, and in particular to an optimization system and method for electric vehicle clusters to participate in the power backup market. Background Art

[0002] With the acceleration of global energy transition and transportation electrification, electric vehicle ownership is experiencing explosive growth. Large-scale electric vehicle fleets are deeply integrated with the power system through vehicle-to-grid technology, becoming a vital distributed energy resource. Participating in the power reserve market is a key way for electric vehicle fleets to create additional value. The power reserve market primarily includes spinning reserve, frequency regulation reserve, and reserve capacity markets, providing emergency support and balancing services to the power grid, ensuring safe and reliable system operation.

[0003] The power reserve market is plagued by volatile and unpredictable prices, leading to high uncertainty in the returns of electric vehicle clusters participating in the market. Traditional approaches struggle to effectively address market volatility risks. Furthermore, the highly random nature of electric vehicle user charging and travel behavior creates significant uncertainty in the capacity available for backup services, impacting service reliability and revenue stability. Traditional approaches often assume fixed call probabilities, which deviate significantly from reality and can lead to excessive system risk exposure.

[0004] In order to overcome these defects, the present application proposes an optimization system and method for electric vehicle clusters to participate in the power backup market. Summary of the Invention

[0005] The purpose of this application is to provide an optimization system and method for electric vehicle clusters to participate in the power reserve market, aiming to solve the multiple uncertainty challenges faced by electric vehicle clusters in participating in the power reserve market.

[0006] To achieve the above objectives, this application provides the following technical solutions:

[0007] In a first aspect, the present application provides an optimization system for electric vehicle clusters participating in a power reserve market, the optimization system comprising the following modules:

[0008] A market uncertainty modeling module is used to construct a joint uncertainty model of market price and backup call probability, including: a probability distribution model of market price and a conditional probability model based on the backup call probability;

[0009] The risk quantification assessment module is used to evaluate the risk exposure of different decision-making options, including: establishing a multi-period risk accumulation model and comprehensively evaluating the balance indicators of expected returns, risk exposure and opportunity costs;

[0010] A robust optimization decision module, which constructs a decision model based on the market uncertainty modeling module and the risk quantification assessment module to generate an optimal strategy for electric vehicle clusters to participate in the backup market;

[0011] Adaptive learning optimization module, used to optimize decision models through reinforcement learning.

[0012] In a second aspect, the present application provides a multimodal animal abnormal data behavior classification method, the steps comprising:

[0013] Constructing a probability distribution model based on market prices and a conditional probability model based on the probability of standby deployment; wherein the input variables of the conditional probability model include system load, renewable energy output, and model parameters, and generating a multi-scenario tree that describes the joint evolution of market prices and deployment probabilities;

[0014] Construct a multi-period risk accumulation model, calculate the cumulative risk through the risk covariance matrix and period weight vector, and comprehensively evaluate the balance indicators of expected return, risk exposure and opportunity cost;

[0015] By establishing a decision model based on a scenario tree, the optimal strategy for electric vehicle clusters to participate in the reserve market is obtained;

[0016] The decision model is optimized through reinforcement learning, experience replay mechanism and Thompson sampling strategy, the parameters and learning rate of the decision model are adjusted, the conditional risk constraints are updated and the target weights are optimized.

[0017] In a third aspect, the present application provides a computer device comprising a processor and a memory coupled to the processor, wherein the memory stores program instructions for implementing an optimization system for an electric vehicle cluster to participate in a power reserve market; and the processor is used to execute the program instructions stored in the memory to implement a multimodal animal abnormal data behavior classification.

[0018] In a fourth aspect, the present application provides a storage medium storing program instructions executable by a processor, wherein the program instructions are used to execute an optimization system for electric vehicle clusters to participate in a power reserve market.

[0019] This application provides an optimization system and method for electric vehicle clusters to participate in the power reserve market, which has the following beneficial effects:

[0020] (1) Traditional methods cannot effectively capture the temporal correlation and conditional dependence between market prices and call probabilities. In contrast, this application combines kernel density estimation with time series decomposition technology to construct a market price model, uses Bayesian networks to establish a conditional distribution model of call probabilities, and describes the market evolution path through a multi-scenario tree, which can more accurately characterize the complex uncertainty characteristics and dynamic evolution process of the power market.

[0021] (2) By constructing a multi-level risk assessment system, we have broken through the limitations of existing technologies; and by using conditional value at risk (CVaR) to quantify risk exposure in extreme situations, designing a multi-period risk accumulation model to consider long-term risk evolution, and proposing a risk-return balance indicator system to evaluate comprehensive benefits, we have achieved an important breakthrough from single risk constraint to comprehensive risk management.

[0022] (3) Existing methods often directly apply commercial solvers to handle stochastic programming problems. When faced with large-scale scenario sets, the computational burden is heavy and real-time performance is difficult to guarantee. This application innovatively proposes a dimensionality reduction technique based on scenario aggregation and a fast solution algorithm based on Lagrangian relaxation. For practical application scenarios with limited computing resources, it achieves a balance between optimization efficiency and solution quality. In particular, for large-scale problems containing tens of thousands of scenario nodes, the approximate algorithm can maintain high-quality solutions while significantly reducing computational complexity.

[0023] (4) This application innovatively proposes to build an adaptive optimization framework based on reinforcement learning, use experience replay to improve learning efficiency, and use Thompson sampling equilibrium exploration and utilization to achieve continuous optimization and adaptability enhancement of the decision model; this self-learning ability enables the system to adapt to changes in the market environment and the evolution of user behavior, forming a virtuous cycle of closed-loop optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a structural diagram of an optimization system for electric vehicle clusters participating in a power reserve market according to Example 1 of the present application;

[0025] Figure 2 This is a structural block diagram of an optimization system for electric vehicle clusters participating in a power reserve market according to Example 1 of the present application;

[0026] Figure 3 This is a flow chart of a multimodal animal abnormal data behavior classification method according to Example 2 of the present application;

[0027] Figure 4 This is a schematic diagram of the computer device structure of Example 3 of the present application;

[0028] Figure 5 This is a schematic diagram of the storage medium structure of Example 4 of the present application. DETAILED DESCRIPTION

[0029] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0030] The following is an analysis of the solutions in the prior art in combination with relevant technologies.

[0031] Research on electric vehicle clusters participating in the power reserve market has received widespread attention in recent years. Currently, there are three main types of technical solutions closest to the present invention: the first type is based on stochastic programming, such as the stochastic optimization framework for electric vehicle clusters considering price uncertainty proposed by Wang et al. (2020). This method captures the randomness of market prices through scenario generation and constructs a stochastic programming model that maximizes expected returns. The second type is based on robust optimization, such as the work published by Liu et al. (2021) in IEEE Transactions on Smart Grid, which proposed a robust optimization method for electric vehicles participating in the reserve market considering user behavior uncertainty. This method uses an uncertain set to describe the fluctuation range of user behavior and optimizes the decision-making performance in the worst case. The third type is real-time scheduling methods based on model predictive control, such as the two-stage scheduling framework proposed by Zhang et al. (2022) that combines offline optimization and online adjustment to dynamically adjust the reserve capacity to cope with real-time system changes.

[0032] Existing methods focus more on maximizing expected returns, fail to adequately quantify risk exposure in extreme market conditions, and lack effective risk management mechanisms. Second, user behavior modeling is simplified, with most adopting simple probability distributions or deterministic boundaries to describe electric vehicle user behavior, failing to fully capture spatiotemporal correlations and conditional dependencies. Third, the coupling effects of market uncertainty and user behavior uncertainty are not adequately addressed, with existing methods often considering a single type of uncertainty separately and ignoring the interactive impact of multiple sources of uncertainty. Fourth, there is a lack of multi-time-scale coordination mechanisms, resulting in a disconnect between day-ahead decisions and real-time execution, making it difficult to effectively respond to real-time system changes. Finally, the adaptive capabilities are insufficient, making it impossible to continuously optimize decision-making strategies based on historical experience and market feedback.

[0033] This application proposes an optimization system and method for electric vehicle clusters participating in the power reserve market. This system comprehensively considers market price fluctuations, user behavior randomness, and reserve call uncertainty to maximize returns under controllable risk conditions. By constructing accurate uncertainty modeling, comprehensive risk assessment, efficient robust optimization algorithms, and continuous adaptive learning mechanisms, electric vehicle clusters can participate in the power reserve market efficiently and stably, providing flexibility services to the power grid and generating additional revenue without impacting users' normal vehicle usage needs.

[0034] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0035] Example 1

[0036] See also Figure 1 , which is a structural diagram of an optimization system for electric vehicle clusters participating in the power backup market according to Example 1 of the present application; the specific contents include:

[0037] A market uncertainty modeling module is used to construct a joint uncertainty model of market price and backup call probability, including: a probability distribution model of market price and a conditional probability model based on the backup call probability;

[0038] The risk quantification assessment module is used to evaluate the risk exposure of different decision-making options, including: establishing a multi-period risk accumulation model and comprehensively evaluating the balance indicators of expected returns, risk exposure and opportunity costs;

[0039] A robust optimization decision module, which constructs a decision model based on the market uncertainty modeling module and the risk quantification assessment module to generate an optimal strategy for electric vehicle clusters to participate in the backup market;

[0040] Adaptive learning optimization module, used to optimize decision models through reinforcement learning.

[0041] In this embodiment, the market uncertainty modeling module, the risk quantification assessment module, the robust optimization decision module and the adaptive learning optimization module are respectively described in detail.

[0042] The market uncertainty modeling module is used to accurately describe the uncertainty characteristics of the power reserve market. This module first uses kernel density estimation combined with time series decomposition technology to construct a probability distribution model of market prices, which can effectively capture the random fluctuations, seasonal changes, and trend characteristics of prices. Specifically, it includes:

[0043] The formula of the probability distribution model is expressed as:

[0044]

[0045] Among them, p is the price variable; p i is the historical price sample; K is the kernel function; h is the bandwidth parameter, and n is the number of samples.

[0046] Considering that prices have obvious time characteristics, the time series decomposition method is used to decompose the price series into trend items, seasonal items, and random items, and a prediction model is established for each item to improve the accuracy of price distribution prediction. The formula is expressed as:

[0047] P t =T t +S t +R t ,

[0048] Among them, P tRepresents the price sequence at time t, that is, the original price time series data; T t Indicates the trend item, reflecting the long-term trend of price changes; S t represents the seasonal term, which captures the cyclical change pattern of prices, such as intraday, intraweek or seasonal fluctuations; R t represents the random term or residual term, which represents the random fluctuation after removing the trend and seasonality;

[0049] For the probability of reserve capacity call, a conditional probability distribution model based on Bayesian network is designed, which takes into account various influencing factors such as system load, renewable energy output, network parameters, etc. to achieve accurate prediction of call probability. The conditional probability model is expressed as:

[0050] Pr(call|L,W,S)=g(L,W,S,θ),

[0051] Where L is the system load; W is the wind power output; S is the photovoltaic output; and θ is the conditional probability model parameter.

[0052] On this basis, a multi-scenario tree T = {N, E, P, V} describing the joint evolution of market price and standby call probability is generated by combining clustering and dynamic programming, where N is the node set, E is the edge set, P is the node probability set, and V is the node value set.

[0053] As new market data continues to be generated, the scenario probability dynamic update mechanism based on Bayesian reasoning ensures that the model can be continuously optimized as new data is generated; the formula is expressed as:

[0054]

[0055] Among them, P(s i |O) means that after observing new data O, scenario s i The posterior probability of . That is, the scenario s updated based on the new market observation data i Probability of occurrence; P(O|s i ) indicates that in scenario s i Under the condition, the likelihood probability of observing data O. That is, if scenario s i If it holds, what is the probability of observing the current market data O? P(s i ) indicates the situation i The prior probability of the updated scenario s i The probability of occurrence is considered; P(O) represents the marginal probability of observing data O, which serves as a normalization factor. It can be understood as the total probability of observing the current market data under all possible scenarios. An exponential decay weight is also introduced to enhance the influence of recent data on probability updates.

[0056] The market uncertainty modeling module breaks through the limitations of simple random variables or deterministic boundaries in existing methods and provides a more accurate description of uncertainty for subsequent decision-making.

[0057] The risk quantification module is used to comprehensively quantify risk exposure under different decision-making scenarios. Its core technologies include conditional value-at-risk (CVaR) calculation methods, a multi-period risk accumulation model, and a risk-return balance indicator system. CVaR (Conditional Value-at-Risk) can assess potential losses in extreme market conditions, providing a quantitative basis for risk management. Specifically, the CVaR for alternative market participation is calculated using the confidence level α:

[0058]

[0059] Among them, CVaR is the conditional value at risk; VaR α is the risk value at the confidence level α; X is the random return;

[0060] Calculate CVaR using the sample average approximation method:

[0061]

[0062] Among them, X i is the value of the benefit under scenario i; N is the total number of scenarios.

[0063] In addition, considering the risk accumulation problem in long-term market participation, this module proposes a multi-period risk accumulation model expressed as:

[0064] R T =f(R1, R2, ..., R T；ρ ),

[0065] Among them, R T is the risk measure for period T; ρ is the time correlation coefficient matrix; f represents how to combine the risk measures R1, R2, ..., R T Combined together, it forms the cumulative risk R of the entire time period T T ;

[0066] Calculate the cumulative risk based on the time covariance structure:

[0067]

[0068] Where w is the risk weight vector for each period; ∑ is the risk covariance matrix.

[0069] The multi-period risk accumulation model breaks through the limitation of traditional methods that deal with risks in each period independently, takes into account the impact of time correlation on cumulative risk, and is more in line with the actual situation of long-term market participation.

[0070] On this basis, the balance indicators of expected returns, risk exposure and opportunity costs are obtained according to the risk quantification assessment module:

[0071]

[0072] in, is the expected return; CVaR α (P) is the risk measure; OC is the opportunity cost; λ1, λ2, and λ3 are weight coefficients.

[0073] The designed risk-adjusted rate of return is:

[0074]

[0075] Among them, C represents cost, including various costs required for electric vehicle clusters to participate in the market, such as operating costs and management costs; RC represents risk capital, which is the amount of capital required to prepare for potential losses, usually determined based on risk measurement (such as CVaR).

[0076] The risk-return balance indicator system comprehensively considers expected returns, risk exposure, and opportunity costs, providing multi-dimensional evaluation criteria for the robust optimization decision-making module. The risk quantification assessment module combines advanced concepts in financial risk management with the characteristics of power systems to establish a comprehensive risk assessment module suitable for electric vehicle clusters participating in the power reserve market.

[0077] The robust optimization decision-making module builds a decision-making model based on the market uncertainty modeling module and the risk quantification assessment module to generate the optimal strategy for electric vehicle clusters to participate in the backup market.

[0078] The objective function of the robust optimization decision module is:

[0079]

[0080] in, is the market price during period t under scenario s; is the corresponding reserve capacity bid amount; is the call cost; is the call amount; λ is the risk aversion coefficient;

[0081] Construct a decision model with risk constraints, including backup capacity constraints, electric vehicle energy balance constraints, and non-predictive constraints:

[0082] Spare capacity constraints:

[0083] Electric vehicle energy balance constraints:

[0084] Non-predictive constraints:

[0085] The decision model is expressed as:

[0086]

[0087] in, Refers to maximizing expected profit, that is, maximizing the average revenue of electric vehicle clusters participating in the power reserve market. α ≤R max Indicates that the conditional risk value does not exceed the maximum tolerable risk R at the confidence level α max .

[0088] The optimal strategy is obtained based on scenario aggregation and dimensionality reduction technology and Lagrangian relaxation solution. Specifically, it includes:

[0089] Scenario probabilities are combined by clustering:

[0090]

[0091] in,

[0092] The fast solution algorithm of Lagrangian relaxation is used to greatly improve the solution efficiency; the formula is expressed as:

[0093] L(x, λ)=f(x)+λ·(g(x)-b).

[0094] in,

[0095] It can be understood that the robust optimization decision-making module takes the maximization of risk-adjusted returns as its objective function, taking into account various operating constraints such as backup capacity constraints, electric vehicle energy balance constraints, and non-predictive constraints. In order to control risk, a multi-objective optimization algorithm considering risk constraints is designed to maximize returns under risk control conditions. To address the computational challenges of large-scale optimization problems, a dimensionality reduction technique based on scenario aggregation and an approximate solution method based on Lagrangian relaxation are proposed, which significantly improves computational efficiency. By combining stochastic programming with risk constraints, the robust optimization decision-making module not only takes into account the expected performance of decisions, but also ensures that risks are controllable in extreme cases. At the same time, it ensures the feasibility of practical applications through efficient algorithms.

[0096] The adaptive learning optimization module optimizes the decision model through reinforcement learning, improving the system's adaptability and long-term performance. First, a multidimensional time series database is designed to store historical decision variables, market conditions, and outcome indicators, providing a data foundation for the learning process. Then, a reinforcement learning framework based on the Markov decision process is constructed: MDP = {S, A, P, R, γ}, where S is the state space, A is the action space, P is the state transition probability, R is the reward function, and γ is the discount factor.

[0097] Using deep reinforcement learning algorithms such as deep Q-network, we can continuously optimize decision-making strategies based on historical experience. The Q-value update formula for reinforcement learning is:

[0098]

[0099] Where α is the learning rate; s and a are the current state and action respectively; γ is the discount factor, which measures the importance of future rewards relative to immediate rewards (between 0 and 1); r is the immediate reward, which is the immediate return obtained from the environment after performing action a; s′ is the next state.

[0100] Introducing experience replay mechanism to improve learning efficiency: D=(s t , a t , r t , s t+1 ). Randomly sample batches of experience for learning to reduce sample correlation. Implement the model parameter update rule based on error feedback:

[0101]

[0102] Among them, θ t is the model parameter vector at the current moment; θ t+1 is the updated model parameter vector; η is the learning rate, which controls the step size of each parameter update; L(θ t ) is the loss function, which measures the current parameter θ t The error between the model prediction and the true value; The gradient of the loss function with respect to the parameter θ indicates the direction and rate of change of the loss function at the current parameter point

[0103] And adaptive learning rate adjustment based on performance indicators:

[0104]

[0105] In addition, an exploration-exploitation balance mechanism based on Thompson sampling is proposed:

[0106]

[0107] where μ a and σ a are the mean and standard deviation estimates of action a, z a is a standard normal random variable.

[0108] The system also implements online updates and adaptive adjustment mechanisms for decision model parameters, dynamically adjusting the learning rate and exploration strategy based on system performance. The adaptive learning optimization module incorporates artificial intelligence learning technology, enabling the system to self-optimize and adapt to environmental conditions, enabling it to cope with complex situations such as changing market rules and evolving user behavior.

[0109] The four modules proposed in this application form a complete closed-loop system through close collaboration of data and control flows. In terms of data flow, the market uncertainty modeling module provides basic data for the risk quantification and assessment module and the robust optimization decision-making module; the risk quantification and assessment module provides risk metrics for robust optimization decision-making; and the adaptive learning optimization module collects system operation data and uses feedback to optimize the parameters of each module. This system implements full-process management from day-ahead decision-making to real-time execution, including the generation of reserve capacity bidding strategies, real-time scheduling control, and post-event evaluation and analysis.

[0110] In practice, the optimization system can implement multiple modes based on different scenario requirements. For a standard market environment, a conventional stochastic programming framework is used: the system collects three types of data (market data, electric vehicle data, and system operation data), uses an ARIMA-GARCH model to capture price characteristics, establishes a call probability model based on net load and reserve capacity gaps, generates 120 market scenarios, and constructs a stochastic programming model with a risk preference coefficient of λ = 0.5 to solve the optimization decision and execute the bidding instructions.

[0111] In response to extreme market conditions, a multi-level risk constraint and emergency response mechanism is introduced: the system uses extreme value theory to construct price tail distribution, pays special attention to extreme high and low price events, and sets multi-level risk constraints Implement a three-stage decision-making strategy and an emergency response mechanism for extreme prices.

[0112] For large-scale electric vehicle clusters, a distributed computing architecture and hierarchical optimization structure are adopted: For large-scale clusters containing 10,000 electric vehicles, the system adopts a distributed computing architecture, divides the cluster into 50 subclusters, each containing approximately 200 vehicles, and adopts a hierarchical optimization structure, including top-level coordination, middle-level subcluster optimization, and bottom-level single-vehicle management. The decomposition and solution are carried out through the alternating direction multiplier method (ADMM) to achieve efficient optimization.

[0113] To address long-term market participation, the system has designed a multi-timescale learning mechanism, including short-term learning (hourly) to adjust specific bidding strategies, medium-term learning (daily) to optimize risk management parameters, and long-term learning (monthly) to update market models and policy networks. This constructs a multi-period nested incentive system to effectively address seasonal fluctuations in participation and maintain a stable participation rate throughout the year. This achieves coordinated optimization and learning updates across multiple timescales. This flexible design enables the invention to adapt to electric vehicle clusters of varying sizes and market environments, promising broad application prospects.

[0114] It should be noted that in market uncertainty modeling, in addition to kernel density estimation and Bayesian networks, deep generative models such as variational autoencoders or generative adversarial networks can also be used to construct more complex uncertainty distributions. For risk quantification and assessment, other risk metrics such as expected shortfall and lower partial moment risk can be used instead. Regarding robust optimization algorithms, distributionally robust optimization methods can be used to account for the inherent uncertainty of the uncertainty distribution. For adaptive learning, other reinforcement learning algorithms such as deep deterministic policy gradient (DDPG) or soft actor-critic (SAC) algorithms can be used.

[0115] The present invention can also be extended to other fields: first, it can be applied to electric vehicles participating in other electricity markets, such as the day-ahead energy market, the real-time balancing market or the ancillary service market; second, it can be extended to the aggregation optimization of other distributed energy resources, such as distributed photovoltaics, energy storage systems or market participation of controllable loads; third, it can be applied to the uncertainty optimization scheduling of microgrids or regional energy systems; it can also be used for decision-making optimization problems in other fields, such as financial asset portfolio optimization, supply chain risk management and other complex systems with significant uncertainty.

[0116] In summary, this embodiment 1 achieves the decision-making goals of risk control and profit maximization in an uncertain environment through the organic combination of the four core modules: market uncertainty modeling module, risk quantification assessment module, robust optimization decision module and adaptive learning optimization module, providing technical support for the efficient participation of electric vehicle clusters in the power market.

[0117] Example 2

[0118] See also Figure 3 , is a flow chart of a multimodal animal abnormal data behavior classification method according to Example 2 of the present application; the steps include:

[0119] S1: Construct a probability distribution model based on market prices and a conditional probability model based on the probability of standby call; wherein the input variables of the conditional probability model include system load, renewable energy output and model parameters, and generate a multi-scenario tree describing the joint evolution of market prices and call probability.

[0120] In this embodiment, kernel density estimation and time series decomposition techniques are used to construct a probability distribution model of market prices:

[0121]

[0122] Among them, p is the price variable; p i is the historical price sample; K is the kernel function; h is the bandwidth parameter, and n is the number of samples.

[0123] The time series decomposition method is used to decompose the market price into trend items, seasonal items and random items, and forecast models are established for each item. The formula is expressed as follows:

[0124] P t =T t +S t +R t ,

[0125] Among them, P t Represents the price sequence at time t, that is, the original price time series data; T t Indicates the trend item, reflecting the long-term trend of price changes; S t represents the seasonal term, which captures the cyclical change pattern of prices, such as intraday, intraweek or seasonal fluctuations; R t represents the random term or residual term, which represents the random fluctuation after removing the trend and seasonality;.

[0126] A conditional distribution model of the backup call probability is constructed based on the Bayesian network. The input variables include system load, renewable energy output, and network topology parameters. The model is expressed as:

[0127] Pr(call|L,W,S)=g(L,W,S,θ),

[0128] Where L is the system load; W is the wind power output; S is the photovoltaic output; and θ is the conditional probability model parameter.

[0129] Finally, a multi-scenario tree describing the joint evolution of market price and call probability is generated, and the scenario probability is dynamically updated through Bayesian reasoning; the formula is expressed as:

[0130]

[0131] Among them, P(s i |O) means that after observing new data O, scenario s iThe posterior probability of . That is, the scenario s updated based on the new market observation data i Probability of occurrence; P(O|s i ) indicates that in scenario s i Under the condition, the likelihood probability of observing data O. That is, if scenario s i If it holds, what is the probability of observing the current market data O? P(s i ) indicates the situation i The prior probability of the updated scenario s i The probability of occurrence is considered; P(O) represents the marginal probability of observing data O, which serves as a normalization factor. It can be understood as the total probability of observing the current market data under all possible scenarios. An exponential decay weight is also introduced to enhance the influence of recent data on probability updates.

[0132] S2: Construct a multi-period risk accumulation model, calculate the cumulative risk through the risk covariance matrix and period weight vector, and comprehensively evaluate the balance indicators of expected return, risk exposure and opportunity cost.

[0133] In this example, the conditional value at risk (CVaR) is calculated as:

[0134]

[0135] Among them, CVaR is the conditional value at risk; VaR α is the risk value at the confidence level α; X is the random return;

[0136] Calculate CVaR using the sample average approximation method:

[0137]

[0138] Among them, X i is the value of the benefit under scenario i; N is the total number of scenarios.

[0139] Considering the risk accumulation problem in long-term market participation, a multi-period risk accumulation model is constructed:

[0140] R T =f(R1, R2, ..., R T；ρ ),

[0141] Among them, R T is the risk measure for period T; ρ is the time correlation coefficient matrix; f represents how to combine the risk measures R1, R2, ..., R T Combined together, it forms the cumulative risk R of the entire time period T T ;

[0142] Calculate the cumulative risk using the risk covariance matrix and the period weight vector:

[0143]

[0144] Where w is the risk weight vector for each period; ∑ is the risk covariance matrix.

[0145] The multi-period risk accumulation model breaks through the limitation of traditional methods that deal with risks in each period independently, takes into account the impact of time correlation on cumulative risk, and is more in line with the actual situation of long-term market participation.

[0146] On this basis, a risk-return balance indicator system is also designed:

[0147]

[0148] in, is the expected return; CVaR α (P) is the risk measure; OC is the opportunity cost; λ1, λ2, and λ3 are weight coefficients.

[0149] The designed risk-adjusted rate of return is:

[0150]

[0151] Among them, C represents cost, including various costs required for electric vehicle clusters to participate in the market, such as operating costs and management costs; RC represents risk capital, which is the amount of capital required to prepare for potential losses, usually determined based on risk measurement (such as CVaR).

[0152] The risk-return balance indicator system comprehensively considers expected returns, risk exposure and opportunity costs, and the decision in step S3 provides multi-dimensional evaluation criteria.

[0153] S3: By establishing a decision model based on a scenario tree, the optimal strategy for electric vehicle clusters to participate in the backup market is obtained.

[0154] In this embodiment, a stochastic programming model based on a scenario tree is established, the objective function is to maximize the risk-adjusted return, and the constraints include spare capacity constraint, electric vehicle energy balance constraint and non-predictive constraint.

[0155] Among them, the objective function is:

[0156]

[0157] in, is the market price during period t under scenario s; is the corresponding reserve capacity bid amount; is the call cost; is the call amount; λ is the risk aversion coefficient;

[0158] Spare capacity constraints:

[0159] Electric vehicle energy balance constraints:

[0160] Non-predictive constraints:

[0161] The scenario aggregation technology is used to reduce the dimensionality of large-scale scenario sets and merge similar scenarios to reduce computational complexity. The specific formula is:

[0162]

[0163] The Lagrangian relaxation algorithm is used to approximate the solution to the optimization problem, balancing the solution quality and computational efficiency; the formula is expressed as:

[0164] L(x, λ)=f(x)+λ·(g(x)-b).

[0165] S4: Optimize the decision model through reinforcement learning, experience replay mechanism and Thompson sampling strategy, adjust the parameters and learning rate of the decision model, update the conditional risk constraints and optimize the target weight.

[0166] In this embodiment, historical decision variables, market status, and result indicators are stored in a multi-dimensional time series database, supporting incremental storage and batch retrieval.

[0167] Construct a reinforcement learning framework based on the deep Q network (DQN). The Q value update formula of reinforcement learning is:

[0168]

[0169] Where α is the learning rate; s and a are the current state and action respectively; γ is the discount factor, which measures the importance of future rewards relative to immediate rewards (between 0 and 1); r is the immediate reward, which is the immediate return obtained from the environment after performing action a; s′ is the next state.

[0170] Combine the experience replay mechanism and Thompson sampling strategy to optimize the decision model, dynamically adjust model parameters and learning rate, and update conditional risk constraints and optimize target weights based on real-time market feedback.

[0171] In summary, Example 2 of the present application first proposes market price modeling through kernel density estimation and time series decomposition, a conditional call probability model based on Bayesian network, and multi-scenario tree generation and dynamic update technology; secondly, a comprehensive risk quantification assessment system, including conditional risk value calculation, multi-period risk accumulation model and risk-return balance indicator design. Then, a random programming model based on the scenario tree, a risk constraint processing method and an approximate solution technology under limited computing resources are used; finally, an adaptive learning optimization mechanism is utilized, including the design of a historical decision database, a strategy optimization based on reinforcement learning and a parameter adaptive adjustment method, and an online update and parameter adaptive adjustment mechanism.

[0172] Example 3

[0173] See also Figure 4 , which is a schematic diagram of the computer device structure of Example 3 of the present application. The computer device 50 includes a processor 51 and a memory 52 coupled to the processor 51.

[0174] The memory 52 stores program instructions for implementing the above-mentioned optimization system for electric vehicle clusters to participate in the power reserve market.

[0175] The processor 51 is used to execute program instructions stored in the memory 52 to implement a multimodal animal abnormal data behavior classification.

[0176] The processor 51 may also be referred to as a CPU (Central Processing Unit).

[0177] Processor 51 may be an integrated circuit chip with signal processing capabilities. Processor 51 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. A general-purpose processor may be a microprocessor or any conventional processor.

[0178] Example 4

[0179] See also Figure 5, which is a structural diagram of the storage medium of Example 4 of the present application. The storage medium of the embodiment of the present application stores a program file 61 that can implement all the above methods, wherein the program file 61 can be stored in the above storage medium in the form of a software product, including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or a computer, server, mobile phone, tablet and other devices.

[0180] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0181] The above description is only a preferred embodiment of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

[0182] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

[0183] Of course, the present invention may have many other implementations. Based on this implementation, other implementations obtained by ordinary technicians in this field without any creative work are all within the scope of protection of the present invention.

Claims

1. An optimization system for electric vehicle clusters to participate in the power reserve market, characterized in that: The optimization system includes the following modules: A market uncertainty modeling module is used to construct a joint uncertainty model of market price and standby call probability, including: a probability distribution model based on the market price and a conditional probability model based on the standby call probability; The risk quantification assessment module is used to evaluate the risk exposure of different decision-making options, including: establishing a multi-period risk accumulation model and comprehensively evaluating the balance indicators of expected returns, risk exposure and opportunity costs; A robust optimization decision module, which constructs a decision model based on the market uncertainty modeling module and the risk quantification assessment module to generate an optimal strategy for electric vehicle clusters to participate in the backup market; Adaptive learning optimization module, used to optimize decision models through reinforcement learning.

2. The optimization system for electric vehicle clusters to participate in the power reserve market according to claim 1, characterized in that: The market uncertainty modeling module includes: a probability distribution model and a conditional probability model; The probability distribution model is constructed by the kernel density estimation method; the formula is expressed as: Among them, p is the price variable; p i is the historical price sample; K is the kernel function; h is the bandwidth parameter, and n is the number of samples; The time series decomposition method is used to decompose the price series into trend terms, seasonal terms and random terms, and model them separately; the formula is expressed as: P t =T t +S t +R t , Among them, P t Represents the price sequence at time t, that is, the original price time series data; T t Indicates the trend item, reflecting the long-term trend of price changes; S t represents the seasonal term, which captures the cyclical change pattern of prices, such as intraday, intraweek or seasonal fluctuations; R t represents the random term or residual term, which represents the random fluctuation after removing the trend and seasonality; Build a conditional probability model: Pr(call|L,W,S)=g(L,W,S,θ), Where L is the system load; W is the wind power output; S is the photovoltaic output; and θ is the conditional probability model parameter.

3. The optimization system for electric vehicle clusters to participate in the power reserve market according to claim 2, characterized in that: Generate a multi-scenario tree T={N, E, P, V} describing the joint evolution of market prices and standby call probabilities through the market uncertainty modeling module, where N is a node set, E is an edge set, P is a node probability set, and V is a node value set; Update scenario probabilities using Bayesian inference: Among them, P(s i |O) means that after observing new data O, scenario s i The posterior probability of the scenario s after updating based on the new market observation data i Probability of occurrence; P(O|s i ) indicates that in scenario s i Under the condition, the likelihood probability of observing data O, that is, if scenario s i If it holds, what is the probability of observing the current market data O? P(s i ) indicates the situation i The prior probability of the updated scenario s i The probability that O is considered to have occurred; P(O) represents the marginal probability of observing data O, which serves as a normalization factor.

4. The optimization system for electric vehicle clusters to participate in the power reserve market according to claim 3, characterized in that: The risk quantification assessment module includes: Calculate the conditional value at risk of the alternative market participation based on the confidence level α: Among them, CVaR is the conditional value at risk; VaR α is the risk value at the confidence level α; X is the random return; Calculate CVaR using the sample average approximation method: Among them, X i is the value of the benefit under scenario i; N is the total number of scenarios.

5. The optimization system for electric vehicle clusters to participate in the power reserve market according to claim 4, characterized in that: The multi-period risk accumulation model is expressed as: R T =f(R1, R2,..., R T ;ρ), Among them, R T is the risk measure for period T; ρ is the time correlation coefficient matrix; f represents how to combine the risk measures R1, R2, ..., R T Combined together, it forms the cumulative risk R of the entire time period T T ; Calculate the cumulative risk based on the time covariance structure: Where w is the risk weight vector for each period; ∑ is the risk covariance matrix; The balance index of expected return, risk exposure and opportunity cost is obtained according to the risk quantification assessment module: in, is the expected return; CVaR α (P) is the risk measure; OC is the opportunity cost; λ1, λ2, and λ3 are weight coefficients; the risk-adjusted rate of return is: Among them, C represents cost, including various costs required for electric vehicle clusters to participate in the market, such as operating costs and management costs; RC represents risk capital, which is the amount of capital required to prepare for potential losses, usually determined based on risk measurement (such as CVaR).

6. The optimization system for electric vehicle clusters to participate in the power reserve market according to claim 5, characterized in that: The objective function of the robust optimization decision module is: in, is the market price during period t under scenario s; is the corresponding reserve capacity bid amount; is the call cost; is the call amount; λ is the risk aversion coefficient; A decision model with risk constraints is constructed, including reserve capacity constraints, electric vehicle energy balance constraints, and non-predictive constraints. The decision model is expressed as: in, Refers to maximizing expected profit, that is, maximizing the average revenue of electric vehicle clusters participating in the power reserve market, CVaR α ≤R max Indicates that the conditional risk value does not exceed the maximum tolerable risk R at the confidence level α max ; The optimal strategy is obtained based on dimensionality reduction technology of scenario aggregation and Lagrangian relaxation.

7. The optimization system for electric vehicle clusters participating in the power reserve market according to claim 1, characterized in that: In the adaptive learning optimization module, the Q-value update formula of reinforcement learning is: Where α is the learning rate; s and a are the current state and action respectively; γ is the discount factor, which measures the importance of future rewards relative to immediate rewards (between 0 and 1); r is the immediate reward, which is the immediate return obtained from the environment after executing action a; s ′ For the next state.

8. A multimodal animal abnormal data behavior classification method, characterized by the following steps: include: Construct a probability distribution model based on market prices and a conditional probability model based on standby call probability; The input variables of the conditional probability model include system load, renewable energy output and model parameters, generating a multi-scenario tree that describes the joint evolution of market price and call probability; Construct a multi-period risk accumulation model, calculate the cumulative risk through the risk covariance matrix and period weight vector, and comprehensively evaluate the balance indicators of expected return, risk exposure and opportunity cost; By establishing a decision model based on a scenario tree, the optimal strategy for electric vehicle clusters to participate in the reserve market is obtained; The decision model is optimized through reinforcement learning, experience replay mechanism and Thompson sampling strategy, the parameters and learning rate of the decision model are adjusted, the conditional risk constraints are updated and the target weights are optimized.

9. A computer device, characterized in that: The computer device includes a processor and a memory coupled to the processor, wherein the memory stores program instructions for implementing an optimization system for an electric vehicle cluster participating in a power reserve market as described in claim 8; the processor is used to execute the program instructions stored in the memory to implement a multimodal animal abnormal data behavior classification.

10. A storage medium, characterized in that: Program instructions executable by a processor are stored, and the program instructions are used to execute the optimization system for electric vehicle clusters participating in the power reserve market as described in claim 8.

Images