Electric vehicle polymer operation optimization method fusing multi-time scale market mechanism and data-driven intelligent decision

Abstract

The invention is suitable for the technical field of electric power market participation and electric vehicle intelligent scheduling, and provides an electric vehicle polymer operation optimization method and device fusing a multi-time scale market mechanism and data-driven intelligent decision. The method comprises the following steps: proposing an electric vehicle charging station day-ahead electric quantity commitment model based on multi-time scale market collaboration; designing a real-time power dynamic scheduling strategy based on deep reinforcement learning; and establishing an aggregator market participation strategy considering the power grid constraint and the market equilibrium influence. The method can effectively coordinate decision coupling of a day-ahead market and a real-time market, get rid of dependence on accurate prediction of uncertainty variables, and quantify the system-level influence of an electric vehicle aggregator as a flexible resource on an electric power market, thereby improving the adaptability, robustness and profitability of the electric vehicle aggregator in a complex market environment.

Description

Technical Field

[0001] This application belongs to the field of power market participation and intelligent dispatching technology for electric vehicles, and in particular relates to a method and device for optimizing the operation of electric vehicle aggregates by integrating multi-timescale market mechanisms and data-driven intelligent decision-making. Background Technology

[0002] Against the backdrop of accelerated global decarbonization, the large-scale development of electric vehicles has become a key path to promote electricity substitution in the transportation sector, achieve an energy consumption revolution, and reduce carbon emissions. However, the disorderly connection of a large number of electric vehicles to the grid for charging will lead to a sharp increase in local loads and a widening of peak-valley differences, seriously affecting the power quality and economic security of the power grid. At the same time, as a distributed energy storage resource with bidirectional regulation capabilities, electric vehicles can effectively participate in the power system's peak shaving and valley filling if they can be rationally aggregated and dispatched, thereby improving the grid's ability to absorb a high proportion of renewable energy.

[0003] Currently, electric vehicles (EVs) still face multiple challenges in participating in the electricity market: On the one hand, existing dispatch strategies largely rely on accurate forecasts of electricity prices and charging demand, making it difficult to adapt to high-dimensional random disturbances in the market environment and user behavior; on the other hand, the lack of a collaborative optimization mechanism integrating day-ahead and real-time dual-layer markets makes it impossible to effectively coordinate the coupling relationship between day-ahead power commitments and real-time power dispatch. Furthermore, traditional market models fail to fully consider the proactive impact of EV aggregators as a flexible resource on market clearing prices and system operation, making it difficult to support them in formulating participation strategies that are both economical and grid-supporting in complex market environments.

[0004] Therefore, there is an urgent need to develop an electric vehicle aggregation operation optimization method that can integrate multi-timescale market mechanisms and data-driven intelligent decision-making to enhance its ability to cope with uncertainty and participate in market collaborative operation, thereby achieving efficient and robust aggregation and scheduling of large-scale electric vehicle resources. Summary of the Invention

[0005] Against this backdrop, the present invention proposes a method and model that are more in line with practical applications, including methodology, model building, and algorithm improvement, in order to address the complexity brought about by the high dimensionality of electric vehicle charging and the deep coupling of vehicle-road-grid, as well as the huge challenges that this poses to the stable operation of the power grid.

[0006] Firstly, a method for optimizing the operation of electric vehicle aggregations that integrates multi-timescale market mechanisms with data-driven intelligent decision-making is proposed, including: A day-ahead electricity commitment model for electric vehicle charging stations based on multi-timescale market collaboration is proposed. Design a real-time power dynamic scheduling strategy based on deep reinforcement learning; Establish an aggregator market participation strategy that takes into account the impact of grid constraints and market equilibrium.

[0007] In one embodiment, the proposed day-ahead electricity commitment model for electric vehicle charging stations based on multi-timescale market collaboration includes: To address the risks of multi-timescale coupling and price volatility faced by electric vehicle aggregates participating in the electricity market, a mixed-integer linear programming model is established to determine the optimal power purchase commitment in the day-ahead market.

[0008] The objective function consists of six components: the profit obtained from charging electric vehicles using day-ahead dispatched electricity. The cost of charging energy storage systems using day-ahead dispatched electricity. Revenue generated from charging electric vehicles through a real-time market The cost of charging energy storage systems through the real-time market Revenue from supplying electricity to electric vehicles with energy storage systems and battery degradation costs The constructor function is as follows: This model takes into account historical electricity prices, predicted aggregated charging demand, and the arbitrage potential of battery energy storage systems. It aims to hedge against real-time market price uncertainty and provide an economic basis for subsequent real-time dispatch.

[0009] Secondly, a real-time power dynamic scheduling strategy based on deep reinforcement learning is provided, including: The real-time operation problem of electric vehicle aggregates is modeled as a Markov decision process (MDP), where the state vector includes the current time period, day-ahead committed electricity, day-ahead and real-time electricity prices, aggregated charging demand and energy storage system status, and the action vector is used to specify the electricity extracted from the day-ahead market, real-time market and energy storage system. The soft actor-critic (SAC) deep reinforcement learning algorithm is used as the solver, combined with a two-branch neural network architecture with long short-term memory (LSTM) units to capture temporal dependencies in historical states. By training the agent in a model-free, simulation-based manner, it can adaptively generate the optimal power allocation strategy without needing to predict the distribution of electricity prices and demand uncertainties, thereby maximizing the gross profit of the entire day's operation.

[0010] Thirdly, a market participation strategy for aggregators that considers the impact of grid constraints and market equilibrium is provided, including: Electricity market participants are categorized into three types: new energy power generation companies, high-efficiency and energy-saving thermal power companies, and electric vehicle aggregators. Cost models are constructed for various entities, with the thermal power company model including conventional power generation costs and carbon emission costs, and the aggregator model including battery aging costs. By analyzing the impact of aggregators' bidirectional regulatory behavior as "virtual power generators" on market clearing prices, traditional power generator strategies, and system carbon emissions, we can quantify their system support capabilities and guide aggregators to formulate market participation strategies that can simultaneously improve their own economic benefits and grid stability.

[0011] In this embodiment of the application, an electric vehicle aggregate operation optimization framework that integrates a day-ahead and real-time two-layer market coordination mechanism is constructed. In the day-ahead stage, the optimal electricity commitment is determined based on a mixed integer linear programming model to hedge against price risk. In the real-time phase, a dynamic scheduling strategy based on deep reinforcement learning is adopted, which can adaptively coordinate the power allocation of multiple sources without relying on the uncertainty distribution. Meanwhile, a Cournot game equilibrium model that considers grid constraints and carbon emission costs is introduced to characterize the impact of electric vehicle aggregators as "virtual power generators" on market clearing and system stability.

[0012] This method achieves full-chain collaborative optimization from multi-timescale power decision-making and data-driven real-time scheduling to market equilibrium strategy formulation, effectively improving the economy, robustness and grid support capabilities of electric vehicle aggregates in complex power market environments, and providing key technical support for flexible and efficient resource aggregation under high-proportion renewable energy access.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this statement. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic flowchart of the electric vehicle charging load prediction method provided in an embodiment of this application; Figure 2 This is a schematic diagram of an EVCH power supply, control operation (power flow mode), and power load provided in an embodiment of this application; Figure 3 This is a schematic diagram of a two-step training framework for a DRL-SAC agent for power management provided in an embodiment of this application; Figure 4This is a schematic diagram of a neural network architecture with LSTM provided in an embodiment of this application. Detailed Implementation

[0016] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0017] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0018] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0019] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0020] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0022] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Reference Figure 1 The diagram shows the flowchart of the electric vehicle charging load forecasting method of this invention. This method is applied to the operational optimization scenario of electric vehicle agglomerations participating in multi-timescale electricity markets. It aims to achieve efficient collaborative operation and profit maximization of electric vehicle agglomerations in complex market environments by integrating a day-ahead-real-time dual-layer market mechanism with data-driven intelligent decision-making. Due to the wide spatial and temporal distribution of electric vehicles, the strong randomness of user behavior, and the outstanding bidirectional regulation capabilities of charging and discharging, directly incorporating them into the electricity market dispatch system presents significant challenges. This necessitates electric vehicle agglomerators playing an intermediary and bridging role. By aggregating dispersed electric vehicle resources, electric vehicle agglomerators form electric vehicle clusters with a certain scale and regulation capabilities, making them easier for the grid to dispatch and control. Simultaneously, agglomerators can formulate reasonable day-ahead power commitments and real-time power dispatch strategies based on electricity market price signals, grid operating status, and user charging demand to achieve optimized regulation of electric vehicle charging and discharging behavior.

[0024] Figure 2 This is a schematic diagram of an EVCH power supply, control operation (power flow mode), and power load provided in an embodiment of this application.

[0025] Figure 3 This is a schematic diagram of a two-step training framework for a DRL-SAC agent for power management provided in an embodiment of this application.

[0026] Figure 4 This is a schematic diagram of a neural network architecture with LSTM provided in an embodiment of this application.

[0027] Reference Figures 1 to 4 The detailed description of the optimization method for the operation of the electric vehicle aggregation is as follows: The application provides a method for optimizing the operation of electric vehicle aggregates by integrating multi-timescale market mechanisms and data-driven intelligent decision-making, including: In one embodiment, step 101 proposes a day-ahead electricity commitment model for electric vehicle charging stations based on multi-timescale market collaboration.

[0028] To address the risks of multi-timescale coupling and price volatility faced by electric vehicle aggregates participating in the electricity market, a mixed-integer linear programming model is constructed at the day-ahead stage to determine the optimal power purchase commitment. This model comprehensively considers historical electricity prices, projected aggregated charging demand, and the arbitrage potential of battery storage systems, aiming to hedge against real-time market price uncertainty and provide an economic basis for subsequent real-time dispatch.

[0029] Step 1011 establishes the objective function as follows: in: The profit gained from using day-ahead dispatched electricity to charge electric vehicles; The cost of charging the energy storage system using day-ahead dispatched electricity; Revenue generated from charging electric vehicles through a real-time market; The cost of charging the energy storage system through the real-time market; Revenue generated from supplying electricity to electric vehicles via energy storage systems; Let the battery degradation cost be denoted by the electric vehicle charging revenue, which is the day-ahead electricity price. of times; This is the current real-time electricity price.

[0030] The objective function includes the profit obtained from charging electric vehicles using day-ahead dispatched electricity, the cost of charging energy storage systems using day-ahead dispatched electricity, the revenue obtained from charging electric vehicles through the real-time market, the cost of charging energy storage systems through the real-time market, the revenue obtained from supplying energy storage system power to electric vehicles, and the cost of battery degradation. The core constraints of the mixed-integer linear programming model include: day-ahead committed power allocation constraints, electric vehicle charging demand satisfaction constraints, real-time market power consumption upper limit constraints, battery energy storage system state mutual exclusion constraints, battery charging and discharging power limit constraints, and battery state of charge upper and lower limit constraints.

[0031] Step 1012 assumes that the revenue from charging electric vehicles is always based on the day-ahead electricity price. The battery degradation model is as follows: (a+1 times) (1) Where m is a linear approximation of the remaining battery life as a function of the number of charge-discharge cycles experienced by the BSS; This indicates the amount of discharge from the BSS to the EV; This is the maximum capacity of the BSS; That is the cost of BSS.

[0032] Step 1013: Establish core constraints: in and These are the minimum and maximum charging power of the battery, respectively; and These are the minimum and maximum discharge power of the battery, respectively. Let t be the amount of energy stored in the battery at time t; and These represent the minimum and maximum capacity limits for the battery, respectively. The first line of the formula ensures the day-ahead commitment. Used for charging electric vehicles and battery storage systems. The second row of the formula shows the charging demand for electric vehicles. This demand must be met by electricity supplied from the day-ahead market, the real-time market, and battery energy storage systems. We assume that at any given time of day, the electricity consumption in the real-time market must not exceed [a certain threshold]. The electric vehicle charging demand is ∈ (0, 1) times the expected demand, as shown in the third row of the formula. This assumption aims to ensure grid reliability and reduce the impact of real-time electricity price fluctuations on the profitability of the electric vehicle charging business. Furthermore, the fourth row of the formula ensures that the battery storage system can only be in one of three states: charging (…). = 1), discharge ( = 1), or remain idle ( =0 and = 0). Meanwhile, the charging power limitation of the battery energy storage system... and discharge power limit These are maintained by the fifth and sixth lines of equations, respectively. The seventh and eighth lines of equations represent the total power flowing into and out of the battery energy storage system, respectively.

[0033] In one embodiment, in step 201, a real-time power dynamic scheduling strategy based on deep reinforcement learning is designed.

[0034] To address the problem that existing real-time scheduling methods rely on the assumption of uncertainty distribution and are difficult to adapt to dynamic market environments, this invention models the real-time operation problem as a Markov decision process (MDP). The state space includes the current time period, the day-ahead committed electricity, the day-ahead and real-time electricity prices, aggregated charging demand, and the energy storage system state. The action space includes the electricity extracted from the day-ahead market, the real-time market, and the energy storage system. The soft actor-commentator (SAC) deep reinforcement learning algorithm is used to solve the problem.

[0035] Step 2011 establishes its objective function as follows: The SAC algorithm maximizes the expected value of the reward and entropy under policy π. As shown below: In the above equation, and They represent strategies respectively. The resulting state distribution and its state-action edge distribution. Temperature parameter. Then used to control the objective function The proportion of the entropy term. To optimize the above objective function, SAC uses a parameterized... and State value function Soft Q function and policy function .

[0036] The reward function of the above Markov decision process is calculated according to the action type of the battery energy storage system, including: Charging rewards This reflects the profit or loss from storing day-ahead or real-time electricity in energy storage rather than using or selling it directly; Where: Electric vehicle charging revenue is assumed to be the day-ahead electricity price. of times; The current price of BSS electricity; Revenue generated from supplying electricity to electric vehicles via energy storage systems; This refers to the real-time electricity price at the current moment. This refers to the excess charge promised before the current day due to battery discharge. The real-time electricity price for the next moment; Costs related to battery degradation; Charging profits The relationship with potential power is shown below: in: For potential day-ahead charging volume; For potential real-time charging amount; discharge reward This reflects the benefits of using energy storage power for energy arbitrage, the impact on day-ahead power sales, and the cost of battery degradation. in: This represents the maximum amount of electricity that the battery can physically release at the current moment. Idle reward The profit and loss calculation is based on the potential profit and loss when no charging or discharging action is selected. .

[0037] Step 2012, establishing the value function, aims to make... minimize.

[0038] in, The expected value for sampling the experience replay pool; This is the output of the current value network; This represents the expected value of the current strategy.

[0039] Step 2013 aims to establish the soft Q-function so that the following soft Bellman residuals are obtained. minimize.

[0040] in, For the agent in state Execute action The environment then immediately provides the actual score; Discount factor; These are the target value network parameters; The expectation of the next state.

[0041] The value is updated according to the following rules: in, For target network parameters; Main network parameters; For soft update coefficients ∈ [0, 1].

[0042] Step 2014: Establish the policy function The goal is to minimize the KL divergence between the two distributions: in, Represents the partition function; For divergence; It is a soft Q function; This is the entropy regularization coefficient; For expectation operators; This is the partition function.

[0043] Next, we need to calculate The gradient. Before proceeding, the above equation needs to be simplified.

[0044] Step 2015: Multiply the expression by And ignore the logarithmic partition function Because of its influence The gradient has no effect, so we can directly obtain... .

[0045] in, It is not differentiable. Therefore, a reparameterization technique is used to obtain the motion. , in: Indicates action It is a state and external noise It was decided jointly; It is a mean network; This is a standard deviation / variance network. Function Output the mean and variance, while It is noise, sampled from a standard normal distribution. Using this technique, the entire process becomes fully differentiable. It can be rewritten as follows: In MDP modeling, the system state vector includes information such as the current time period, day-ahead commitments, day-ahead and real-time electricity prices, aggregated electric vehicle charging demand, BSS state of charge, and BSS power price; the action vector specifies the electricity allocation scheme extracted from the day-ahead, real-time market, and BSS. The reward function comprehensively considers arbitrage profits, electricity price difference gains and losses, and battery degradation costs, and introduces an "idle reward" mechanism to avoid ineffective operations.

[0046] To improve policy performance, this invention employs a two-branch neural network architecture with Long Short-Term Memory (LSTM) units: the feedforward branch processes the current state features, while the recurrent branch captures historical temporal dependencies through the LSTM; the outputs of the two branches are concatenated and then passed through a fully connected layer to generate the final action. This architecture is abbreviated as DRL-SAC-LSTM.

[0047] The training process adopts Figure 3 The two-step framework is illustrated: In the execution phase, the policy network generates actions and collects experience, storing it in the replay buffer; in the update phase, batch data is sampled from the buffer, and the parameters of the Q-network, value network, and policy network are updated alternately until convergence. Experiments show that the DRL-SAC-LSTM strategy significantly outperforms traditional methods in terms of gross profit, demonstrating superior dynamic adaptability and forward-looking decision-making capabilities.

[0048] In one embodiment, in step 301, an aggregator market participation strategy that takes into account the impact of grid constraints and market equilibrium is established.

[0049] To characterize the bidirectional regulatory capabilities of electric vehicle aggregators as "virtual power generators" and their systemic impact on the electricity market, this invention draws on the Cournot game equilibrium concept to construct an electricity market equilibrium model that includes carbon emission costs.

[0050] Market participants are categorized into three types: ① new energy power generation companies; ② high-efficiency and energy-saving thermal power companies; and ③ electric vehicle aggregators. For thermal power companies, costs include conventional secondary costs and carbon emission costs. For aggregators, net output energy is considered equivalent to electricity generation and is included in the unit battery aging cost.

[0051] Step 3011 constructs a quadratic cost function based on output for both new energy power generation companies and high-efficiency energy-saving thermal power companies, reflecting the nonlinear relationship between output and cost, as shown in the following expression: in, For the real-time power output (MW) of the k-th type of power generation company; , , , where is the cost coefficient of the k-th type of power generation company (all are positive real numbers, determined by equipment characteristics and fuel type); Let be the total cost of conventional power generation for the k-th type of power generation company.

[0052] Step 3012 constructs a quadratic cost function based on output for both new energy power generation companies and high-efficiency energy-saving thermal power companies, reflecting the nonlinear relationship between output and cost, as shown in the following expression: in, For the real-time power output (MW) of the k-th type of power generation company; , , , where is the cost coefficient of the k-th type of power generation company (all are positive real numbers, determined by equipment characteristics and fuel type); Let be the total cost of conventional power generation for the k-th type of power generation company.

[0053] Step 3013 applies only to high-efficiency and energy-saving thermal power companies, considering the carbon emission costs during their power generation process, and constructing a cost model based on the carbon emission factor per unit output: Where e represents the amount of carbon dioxide emitted; f is the amount of carbon dioxide emitted by the fuel used in the power generation process; w is the unit calorific value of the fuel; η is the power generation efficiency; and q is the power generation output of the power producer. This represents the carbon emission price corresponding to electricity producer i; Let be the total carbon emission cost of the k-th type of power generation company.

[0054] Step 3014 establishes that the total power generation cost of an efficient and energy-saving thermal power company is the sum of conventional costs and carbon emission costs: in: The total cost for energy-efficient power generators; This is the cost of conventional power generation; Cost of carbon emissions.

[0055] Step 3015 is for the electric vehicle aggregator of an efficient energy-saving thermal power company: This model realizes revenue through arbitrage operations on the excess electric energy during the charging and discharging process on the premise of meeting the driving requirements of traditional electric vehicles. During period t, the total net output energy of the electric vehicle aggregator is: Where represents the purchase of electricity, represents the sale of electricity, represents the electricity consumption for the daily driving of electric vehicles. can be approximated as the output electricity of the power generation company.

[0056] The electric vehicle aggregator considers the battery aging cost : Where y is the cost caused by battery aging under the unit net output power, and 0 < y < 1, and the general value is 0.6.

[0057] Step 302 constructs the profit maximization objective function of each subject based on the Cournot game framework, and derives the equilibrium condition in combination with the market clearing constraint; Step 3021 Power generation company profit Objective: The power generation company sells electricity at the real-time market price p, and the profit is the revenue from selling electricity minus the total power generation cost. The objective function is as follows: Where represents the unit cost of the electricity generation of power generator i; is the market load demand; is the power generation output of the power generator; is the total output of other power generators.

[0058] Step 3022 Aggregator profit objective: In the spot market, power producers pursue profit maximization, and electric vehicle aggregators can be analogized as one of the power producers. Therefore, the market equilibrium model is: The constraint conditions are: Where represents the upper limit of the generator output; This indicates the charging and discharging capacity of an electric vehicle.

[0059] The market-clearing price is determined by aggregate supply and elastic load demand. Players interact strategically with the goal of maximizing profits: power generation companies determine output, and aggregators determine net discharge. Under the Cournot competition framework, solving for the Nash equilibrium of each player's reaction function yields the aggregator's optimal market participation strategy.

[0060] This model reveals the "peak shaving and valley filling" effect of aggregator charging and discharging behavior on market clearing prices, and quantifies its strategic coupling relationship with traditional power generators, providing a theoretical basis for formulating market strategies that take into account both economic efficiency and grid support capabilities.

[0061] In summary, this invention effectively enhances the adaptability, robustness, and profitability of electric vehicle agglomerations in high-proportion renewable energy power systems through the aforementioned three-step collaborative mechanism—day-ahead power commitment, real-time intelligent scheduling, and market-balanced participation—and achieves efficient and orderly aggregation of large-scale electric vehicle resources and market-coordinated operation.

[0062] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. Furthermore, although the operation of the method of this application is described in a specific order in the accompanying drawings, this does not require that it must be executed in that order, and some steps may be omitted, combined, or broken down.

[0063] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0064] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 this application, and should all be included within the protection scope of this application.

Claims

1. A method for optimizing the operation of electric vehicle aggregates by integrating multi-timescale market mechanisms and data-driven intelligent decision-making, characterized in that, include: A day-ahead electricity commitment model for electric vehicle charging stations based on multi-timescale market collaboration is proposed. Design a real-time power dynamic scheduling strategy based on deep reinforcement learning; Establish an aggregator market participation strategy that takes into account the impact of grid constraints and market equilibrium.

2. The electric vehicle aggregate operation optimization method as described in claim 1, characterized in that, The proposed day-ahead electricity commitment model for electric vehicle charging stations based on multi-timescale market collaboration includes: Construct the objective function of a mixed-integer linear programming model with the goal of maximizing profit; in: The profit gained from using day-ahead dispatched electricity to charge electric vehicles; The cost of charging the energy storage system using day-ahead dispatched electricity; Revenue generated from charging electric vehicles through a real-time market; The cost of charging the energy storage system through the real-time market; Revenue generated from supplying electricity to electric vehicles via energy storage systems; Let the battery degradation cost be denoted by the electric vehicle charging revenue, which is the day-ahead electricity price. of times; This refers to the current real-time electricity price; The objective function includes the profit obtained from charging electric vehicles using day-ahead dispatched electricity, the cost of charging energy storage systems using day-ahead dispatched electricity, the revenue obtained from charging electric vehicles through the real-time market, the cost of charging energy storage systems through the real-time market, the revenue obtained from supplying energy storage system power to electric vehicles, and the cost of battery degradation. The core constraints of the mixed-integer linear programming model include: day-ahead committed power allocation constraints, electric vehicle charging demand satisfaction constraints, real-time market power consumption upper limit constraints, battery energy storage system state mutual exclusion constraints, battery charging and discharging power limit constraints, and battery state of charge upper and lower limit constraints.

3. The electric vehicle aggregate operation optimization method as described in claim 1, characterized in that, The design is based on a real-time power dynamic scheduling strategy using deep reinforcement learning, including: The real-time operation problem is modeled as a Markov decision process, where the state space includes the current time period, the day-ahead committed electricity, the day-ahead and real-time electricity prices, aggregated charging demand and the energy storage system status, and the action space includes the electricity extracted from the day-ahead market, the real-time market and the energy storage system. The Markov decision process is solved using a soft actor-critic deep reinforcement learning algorithm. A dual-branch neural network architecture with long short-term memory units is used as the policy network to capture the temporal dependencies of state information.

4. The electric vehicle aggregate operation optimization method as described in claim 3, characterized in that, The reward function of the Markov decision process is calculated according to the action type of the battery energy storage system, including: Charging rewards This reflects the profit or loss from storing day-ahead or real-time electricity in energy storage rather than using or selling it directly; Where: Electric vehicle charging revenue is assumed to be the day-ahead electricity price. of times; The current price of BSS electricity; Revenue generated from supplying electricity to electric vehicles via energy storage systems; This refers to the real-time electricity price at the current moment. This refers to the excess charge promised before the current day due to battery discharge. The real-time electricity price for the next moment; Costs related to battery degradation; Charging profits The relationship with potential power is shown below: in: For potential day-ahead charging volume; For potential real-time charging amount; discharge reward This reflects the benefits of using energy storage power for energy arbitrage, the impact on day-ahead power sales, and the cost of battery degradation. in: This represents the maximum amount of electricity that the battery can physically release at the current moment. Idle reward The profit and loss calculation is based on the potential profit and loss when no charging or discharging action is selected. 。 5. The electric vehicle aggregate operation optimization method as described in claim 1, characterized in that, The establishment of an aggregator market participation strategy that considers the impact of grid constraints and market equilibrium includes: Electricity market participants are categorized into three types: new energy power generation companies, high-efficiency and energy-saving thermal power companies, and electric vehicle aggregators. Based on the Cournot game equilibrium concept, a power market equilibrium model including the above three types of subjects is constructed: in Let be the spot market electricity price at time t; This represents the power output of the i-th power generator at time t; For the i-th generator at time t The cost of electricity; Let t be the energy that the aggregator discharges into the power grid at time t; Let t be the charging energy purchased by the aggregator from the power grid. At time t, the electrical energy consumed by electric vehicle users for daily driving. Costs associated with battery aging; This study analyzes the "peak shaving and valley filling" effect of aggregator charging and discharging behavior on market clearing prices, as well as the interaction between aggregator charging and discharging behavior and the strategies of traditional power generators.

6. The electric vehicle aggregate operation optimization method as described in claim 2, characterized in that, The core constraints specifically include: The allocation constraints on the committed electricity volume will ensure that the committed electricity volume is reasonably allocated to the charging of electric vehicles and battery storage systems. Electric vehicle charging demand meets constraints, ensuring that the charging demand of electric vehicles is met by electricity provided by the day-ahead market, the real-time market, and battery energy storage systems. The real-time market electricity consumption ceiling constraint restricts the real-time market electricity consumption from exceeding a certain percentage of the expected electric vehicle charging demand; The battery energy storage system has a state mutual exclusion constraint to ensure that the battery energy storage system can only be in one of the charging, discharging or idle states at any given time; Battery charging and discharging power limits ensure that the battery's charging and discharging power do not exceed its rated limits; Battery state of charge (SOC) upper and lower limits are constrained to ensure that the battery's SOC always operates within the allowable safe range.

7. The electric vehicle aggregate operation optimization method as described in claim 3, characterized in that, The adoption of a dual-branch neural network architecture with long short-term memory units as the policy network specifically includes: The feedforward branch is used to process the state features at the current moment; Looping branches capture the temporal dependencies of historical state sequences through long short-term memory units; The fusion layer concatenates the outputs of the feedforward and loop branches, and then processes them through a fully connected layer to finally generate the action probability distribution.

8. The electric vehicle aggregate operation optimization method as described in claim 3, characterized in that, The training process of the soft actor-critic deep reinforcement learning algorithm adopts a two-step framework, including: During the execution phase, the policy network generates actions based on the current state and interacts with the environment to collect experience data and store it in the replay cache. During the update phase, batches of empirical data are sampled from the replay cache, and the parameters of the Q-function network, state-value function network, and policy network are updated alternately until the model converges.

9. The electric vehicle aggregate operation optimization method as described in claim 5, characterized in that, The aforementioned model for constructing a total power generation cost model for high-efficiency and energy-saving thermal power companies, which includes carbon emission costs, specifically includes: Construct a quadratic cost function based on power output to reflect the nonlinear relationship between power generation output and conventional costs; Construct a carbon emission cost model based on the carbon emission factor per unit output; The total power generation cost of an energy-efficient thermal power company is the sum of conventional costs and carbon emission costs.

10. An electric vehicle aggregation operation optimization device that integrates multi-timescale market mechanisms and data-driven intelligent decision-making, characterized in that, The electric vehicle aggregate operation optimization method according to any one of claims 1 to 9 includes: The day-ahead commitment module is used to propose a day-ahead electricity commitment model for electric vehicle charging stations based on multi-timescale market collaboration. The real-time scheduling module is used to design real-time power dynamic scheduling strategies based on deep reinforcement learning. The market participation module is used to establish aggregator market participation strategies that take into account the impact of grid constraints and market equilibrium.

Images