Optimization control method and system for source-network-load-storage integrated market mechanism, and electronic equipment

Abstract

The objective of the invention is to provide an optimization control method and system for a source-network-load-storage integrated market mechanism, and an electronic device, and the method comprises the steps: calculating a market price and transaction electric quantity through a market clearing model, and carrying out the optimization of a source-network-load-storage collaborative optimization model with the minimum system operation cost as a target; comprehensive optimization of power generation cost, energy storage degradation cost, network line loss and market deviation penalty is realized; the initial solution generated by multi-region ADMM distributed optimization and reinforcement learning ensures the global scheduling feasibility and calculation efficiency; the conditions of extreme load, low wind and light output, market price fluctuation and the like can be covered, and the robustness of the system to uncertainty is enhanced. Multi-agent collaborative optimization and real-time control are realized; and the source network load storage collaborative optimization model is combined with a multi-region multi-agent reinforcement learning algorithm to realize collaborative control of a generator, energy storage and a load side, and power balance constraint, energy storage SOC constraint, node voltage constraint and load adjustable constraint are considered.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power system optimal control, in particular to a source-grid-load-storage integrated market mechanism optimal control method and system and electronic equipment. BACKGROUND

[0002] With the rapid increase of new energy installed capacity, the volatility and uncertainty of renewable energy such as wind power and photovoltaic have put higher requirements on the stable operation of power systems. At the same time, the rapid development of energy storage systems makes it possible to integrate source-grid-load-storage, providing a technical basis for achieving multi-objective optimization such as economy, reliability, renewable energy consumption and energy storage life extension. The current traditional power market and dispatch mechanism often uses static and layered optimization methods, which are difficult to simultaneously consider the real-time, risk management and economy of the system.

[0003] The existing technology has the following problems: the uncertainty of new energy output leads to high risk: the traditional dispatch method is usually based on historical average or static prediction, which is difficult to capture the tail risk of load and renewable energy output, and is prone to system power imbalance or market revenue deviation. Market clearing and dispatch are disconnected: the existing market mechanism lacks effective coordination between actual dispatch, making it difficult to unify the bidding strategy and the physical constraints of the power grid, limiting the market efficiency and economy. Insufficient optimization of energy storage and load: traditional dispatch lacks consideration of energy storage life degradation, load adjustability and equipment constraints, and has limited real-time control capability. The closed loop of layered optimization is imperfect: the market layer, dispatch layer and control layer are optimized independently, lacking end-to-end risk quantification, rolling optimization and revenue settlement closed loop, and the system lacks robustness to extreme events and deviations.

[0004] Therefore, in order to solve the above technical problems, it is necessary to propose a new technical means. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a source-grid-load-storage integrated market mechanism optimal control method and system and electronic equipment, which calculates market price and transaction electricity through a market clearing model, and realizes the comprehensive optimization of generation cost, energy storage degradation cost, network line loss and market deviation penalty in a source-grid-load-storage collaborative optimization model with the goal of minimizing system operation cost. The initial solution generated by multi-region ADMM distributed optimization and reinforcement learning ensures the global dispatch feasibility and computational efficiency; it can cover extreme load, low wind and light output and market price fluctuations, enhancing the system's robustness to uncertainty. Realize multi-agent collaborative optimization and real-time control. The source-grid-load-storage collaborative optimization model combines multi-region multi-agent reinforcement learning (MARL) algorithm to realize the collaborative control of generators, energy storage and load side, considering power balance constraints, energy storage SOC constraints, node voltage constraints and load adjustability constraints.

[0006] The application provides an optimization control method of a source-network-load-energy storage integrated market mechanism, comprising the following steps:

[0007] S1, collecting operation data of a source side, a network side, a load side and an energy storage side; based on the operation data, an uncertainty scenario generation model is established to generate a scenario set containing load prediction information, renewable energy output prediction information and market price prediction information;

[0008] S2, according to the scenario set, a market clearing model containing a market subject benefit function and a power grid constraint condition is constructed;

[0009] The market clearing model receives bidding information of the source side, the load side and the energy storage side, and calculates market clearing prices and transaction power results;

[0010] S3, according to the market clearing prices, the transaction power results and the operation data, a source-network-load-energy storage collaborative optimization model is established; the source-network-load-energy storage collaborative optimization model takes the minimum system operation cost as the target, and the constraint conditions include power balance constraints and device operation constraints; and regional power distribution results and energy storage operation plans are output;

[0011] S4, according to the regional power distribution results and the energy storage operation plans, a rolling optimization algorithm is used to calculate energy storage power output and load adjustment power;

[0012] S5, according to the power output and the load adjustment power determined in step S4, the source-network-load-energy storage power grid system is controlled to work.

[0013] Further, in step S1, the following steps are specifically included:

[0014] S1-1, collecting operation data of the source side, the network side, the load side and the energy storage side, including load actual output, renewable energy actual output, energy storage state, historical market transaction data and network operation indexes, for forming a system historical state set , the historical state set includes historical data and historical operation state ;

[0015] S1-2, based on the historical state set , an uncertainty scenario generation model is established, the input is historical data and historical operation state , and the output is a future prediction scenario :

[0016]

[0017] S1-3. Generate a scenario set containing load forecasting information, renewable energy output forecasting information, and market price forecasting information using the uncertainty scenario generation model. .

[0018] Furthermore, step S2 specifically includes:

[0019] S2-1, Market entity i obtains historical operating status. Including load forecasting Renewable energy output forecast Energy storage status Historical market transaction data And network constraint metrics, used to describe the state of the environment;

[0020] S2-2, The market entity will record its historical operating status. Input to an Actor network, which generates a policy function. Generate the main bidding action ,in For the reported supply or demand, This corresponds to the quoted price;

[0021] S2-3, Perform the main bidding action. Input market clearing operator Calculate the transaction volume results for each market participant. With market clearing price ;

[0022] S2-4, By adjusting the market clearing price Multiply by the transaction volume result Calculate revenue, minus production and supply costs. minus penalties for breach of contract After deducting the equivalent lifetime cost of energy storage degradation Calculate instant rewards:

[0023] .

[0024] Furthermore, step S2 also includes: S2-5, transferring historical operating status. Main bidding action and The data is stored in the experience replay buffer, and the Critic and Actor network parameters are updated using formulas for subsequent mini-batch network parameter updates. In the Critic network update, the target value is calculated. , As a discount factor, For the next historical running state, ( The target Critic network outputs for the next historical operating state, with the target network parameters provided; the Critic network updates the parameters by minimizing the mean square error loss , which approximates the true principal bid action; the Actor network updates the policy parameters according to the gradient signal output by the Critic , where is the policy function generated by the Actor network, and the gradient signal output by the Critic is used to guide the Actor network to update, is the gradient of the Actor network parameters.

[0025] Further, the construction method of the source network load storage collaborative optimization model comprises:

[0026] S3-1, obtaining a global system state , which contains the load prediction of each regional level scheduling agent , the renewable energy output prediction , the energy storage state , the historical market transaction data , and the network operation index ;

[0027] S3-2, each regional level scheduling agent generates an initial solution based on the global system state using the reinforcement learning Actor network ;

[0028] S3-3, applying power balance constraints and device operation constraints to the initial solution to generate a near feasible solution;

[0029] S3-4, taking the near feasible solution as the initial value, calling a mathematical programming solver for fine-tuning optimization, with the goal of minimizing the system comprehensive operation cost, to solve the global variables :

[0030]

[0031] where F(x) is the system comprehensive operation cost, including generation cost , energy storage life degradation cost , network line loss , node voltage deviation , and market deviation penalty , and Ω represents the feasible region, which is determined by the power balance, capacity constraints, energy storage dynamics, line flow constraints, and market commitment constraints;

[0032] ​​S3-5. Decompose the global optimization problem into subproblems by region and solve them using ADMM in parallel iterative steps, where global variables... Let z be the local decision vector of the subproblem, z be the coupling variable representing the power of the boundary nodes, y be the multiplier used for consistency constraints, ρ be the penalty parameter controlling ADMM convergence, and the coupling matrix be... The boundary constraint c explicitly describes the linear relationship between the subproblems in each region and the global variable. The ADMM update formula is:

[0033] Local subproblem update:

[0034]

[0035] Global variable update:

[0036]

[0037] Multiplier Update:

[0038]

[0039] S3-6 continues until the local subproblem updates converge, then outputs the power allocation results for each region. Energy storage operation plan and load adjustment power .

[0040] Furthermore, specific methods for establishing uncertainty scenario generation models include:

[0041] Generative models are constructed using variational autoencoders (VAEs) to learn from historical observations. Probability distribution of system operating state under certain conditions ;

[0042] Historical observation Mapped to the latent variable space by the encoder And perform reparameterization to generate latent variable samples. Use a decoder to generate samples of future system states. ;

[0043] The model is trained by maximizing the lower bound of evidence.

[0044]

[0045] in This is used to ensure that the generated samples are close to the real state. This is used to regularize the latent variable distribution to a standard Gaussian to ensure sample diversity and continuity;

[0046] Sampling from the trained generative model Future operating status covering possible variations of load, renewable energy output and market price;

[0047] for each future operating state calculating system operating cost and loss and calculating the conditional value at risk CVaR at a confidence level

[0048]

[0049] wherein is the loss threshold at the confidence level quantile; passing the scenario set S and its corresponding risk indicator CVaR as inputs to a multi-agent reinforcement learning scheduler of the dispatch layer.

[0050] Further, the power balance constraint is satisfied:

[0051]

[0052] wherein,

[0053] denotes the unit active power output optimization result of the region k at time t;

[0054] denotes the energy storage charging and discharging power of the region k at time t;

[0055] denotes the renewable energy generation power of the region k at time t;

[0056] denotes the power load demand of the region k at time t;

[0057] denotes the power load demand of the region k at time t;

[0058] The equipment operation constraint includes an energy storage operation constraint, a load adjustment constraint and a network security constraint;

[0059] The energy storage operation constraint is satisfied:

[0060]

[0061]

[0062] wherein,

[0063] denotes the state of charge of the energy storage at time k;

[0064] denote the charging efficiency and the discharging efficiency, respectively;​

[0065] respectively represent the charging power and discharging power stored at time k;

[0066] denotes the length of time interval;

[0067] denotes the rated power of the energy storage device;

[0068] respectively represent the upper and lower threshold values of the state of charge of the energy storage;

[0069] The load adjustment constraint satisfies:

[0070]

[0071] wherein,

[0072] denotes the load-side adjustable power of region k at time t;

[0073] denotes the minimum and maximum values of the load adjustment;

[0074] The network security constraint satisfies:

[0075]

[0076] wherein,

[0077] denotes the voltage amplitude of node i;

[0078] denotes the branch power transmission between node i and node j;

[0079] , denotes the upper and lower limits of the node voltage operation;

[0080] denotes the maximum transmission capacity of the branch.

[0081] Correspondingly, the present application also provides a system, comprising a memory, a processor, and a program stored in the memory, wherein the processor executes the program to realize the method in the above

[0082] Correspondingly, the present application also provides an electronic device, comprising:

[0083] a processor;

[0084] a memory for storing processor-executable instructions;

[0085] The processor is configured to invoke instructions stored in the memory to perform the method in the above.

[0086] Advantages of the present application: through the present application, the market price and transaction power are calculated through the market clearing model, and the system operation cost is minimized as the target in the source network load storage collaborative optimization model, realizing the comprehensive optimization of power generation cost, energy storage degradation cost, network line loss and market deviation penalty. The initial solution generated by multi-region ADMM distributed optimization and reinforcement learning ensures the global scheduling feasibility and calculation efficiency; it can cover extreme load, low wind and light output and market price fluctuation, etc., and enhance the robustness of the system to uncertainty. Realize multi-agent collaborative optimization and real-time control. The source network load storage collaborative optimization model combines multi-region multi-agent reinforcement learning (MARL) algorithm to realize the collaborative control of generators, energy storage and load side, considering power balance constraints, energy storage SOC constraints, node voltage constraints and load adjustable constraints. BRIEF DESCRIPTION OF DRAWINGS

[0087] The present application will be further described below in conjunction with the drawings and examples:

[0088] Figure 1 The flowchart of the present application. DETAILED DESCRIPTION

[0089] The present application will be further described below in conjunction with the drawings and examples:

[0090] The present application provides a kind of source network load storage integrated market mechanism optimization control method, it is characterized by: including the following steps:

[0091] S1, the operation data of source side, network side, load side and energy storage side are collected;Based on the operation data, an uncertainty scenario generation model is established to generate a scenario set containing load prediction information, renewable energy output prediction information and market price prediction information;The system can comprehensively obtain the historical and real-time operation data of source side, network side, load side and energy storage side, and establish a probability model based on historical state to generate future uncertainty scenarios. Through the scenario set, load fluctuation, renewable energy output change and market price change are covered, providing sufficient uncertainty information for subsequent market clearing and dispatching optimization, and improving the response capability of the system to extreme events and risks.

[0092] S2, according to the scenario set, a market clearing model containing market subject benefit function and power grid constraint condition is constructed;By constructing a market clearing model containing market subject benefit function and power grid constraint, reasonable market clearing price and transaction power can be calculated, and economic coordination of source, network, load and storage is realized. Ensure that market decision-making is carried out under the constraint of power grid safety, improve the transparency of power transaction, and at the same time provide clear power instruction for dispatching layer.

[0093] The market clearing model receives bid information of the source side, the load side and the energy storage side, calculates a market clearing price and a transaction power result;

[0094] S3, according to the market clearing price, the transaction power result and the operation data, a source-grid-load-storage collaborative optimization model is established; the source-grid-load-storage collaborative optimization model takes the minimum system operation cost as the target, the constraint conditions include power balance constraints and device operation constraints; the power distribution results of each region and the energy storage operation plan are output; the minimum system operation cost is taken as the target, the power balance constraints and the device operation constraints are comprehensively considered, and the coordinated distribution of power generation, energy storage and controllable load in each region is realized. The power distribution results of each region and the energy storage operation plan are output, the unity of system economy and physical constraints is realized, and a feasible reference scheme is provided for rolling optimization and real-time control

[0095] S4, according to the power distribution results of each region and the energy storage operation plan, a rolling optimization algorithm is used to calculate the energy storage power output and the load adjustment power; the rolling optimization algorithm is used to dynamically calculate the energy storage power output and the load adjustment power, which can improve the adaptability of the system to prediction errors on the premise of ensuring power balance and energy storage safety, realize short-term regulation and flexible scheduling. Real-time control strategies are provided for load response and energy storage management, and the accuracy of scheduling execution is improved; wherein the rolling optimization algorithm uses existing algorithms such as rolling window algorithm, which is prior art and will not be described here;

[0096] S5, according to the power output and the load adjustment power determined in step S4, the source-grid-load-storage power grid system is controlled. Through the above method, the market price and the transaction power are calculated by the market clearing model, and the minimum system operation cost is taken as the target in the source-grid-load-storage collaborative optimization model, so as to realize the comprehensive optimization of generation cost, energy storage degradation cost, network line loss and market deviation penalty. The initial solution generated by multi-region ADMM distributed optimization and reinforcement learning ensures the global scheduling feasibility and calculation efficiency; it can cover extreme load, low wind and light output and market price fluctuation, and enhance the robustness of the system to uncertainty. Multi-agent collaborative optimization and real-time control are realized. The source-grid-load-storage collaborative optimization model combines multi-region multi-agent reinforcement learning (MARL) algorithm to realize the collaborative control of generators, energy storage and load side, and considers power balance constraints, energy storage SOC constraints, node voltage constraints and load adjustable constraints.

[0097] In this embodiment, step S1 specifically includes:

[0098] S1-1, collecting operation data of the source side, the grid side, the load side and the energy storage side, including actual load output, actual renewable energy output, energy storage state, historical market transaction data and network operation indicators, for forming a system historical state set , the historical state set Including historical data and historical operating states ; by comprehensively collecting source-side (wind power, photovoltaic), network-side (power transmission and distribution state), load-side (actual power consumption), and energy storage-side (SOC / SOH state) operating data, a complete system historical state set is formed. The historical state set contains historical output, load, energy storage state, market transaction data, and network operation indicators, providing high-quality input data basis for subsequent uncertainty modeling, ensuring that scenario generation can truly reflect system operation rules and historical fluctuation characteristics;

[0099] S1-2, based on the historical state set establishing an uncertainty scenario generation model , the input is historical data and historical operating states , and the output is future predicted scenarios :

[0100]

[0101] Based on the historical state set, a probability generation model (such as VAE or Bayesian network) is constructed, which can learn the probability distribution of system states under historical observation conditions. By modeling historical data and historical operating states, the non-Gaussian characteristics and tail risks of load, renewable energy output, and market prices can be captured, providing a reliable statistical basis for future state generation and realizing the quantitative representation of uncertainty;

[0102] S1-3, generating a scenario set containing load prediction information, renewable energy output prediction information, and market price prediction information through the uncertainty scenario generation model . Use the trained uncertainty scenario generation model to generate future multi-step prediction scenarios, covering load prediction, renewable energy output prediction, and market price prediction information. This scenario set can reflect the uncertainty and potential extreme situations of system operation, providing diverse and controllable future state inputs for market clearing, collaborative optimization, and risk assessment, and realizing the robustness and reliability of subsequent optimization and decision-making

[0103] In this embodiment, step S2 specifically includes:

[0104] S2-1, the market subject i acquires historical operating states , including load prediction , renewable energy output prediction , energy storage state , historical market transaction data , and network constraint indicators, which are used to describe environmental states;

[0105] S2-2, the market subject combines historical operating states an input Actor network, which generates a policy function , generates an agent bid action wherein is the reported supply or demand, is the corresponding quoted price;

[0106] S2-3, inputting the agent bid action into a market clearing operator , calculates a trading power result for each market agent and a market clearing price ;

[0107] S2-4, calculates a revenue by multiplying the market clearing price by the trading power result , subtracts production and supply costs , subtracts a compliance deviation penalty , deducts a storage degradation equivalent life cost , and calculates an immediate reward:

[0108] .

[0109] Specifically, in step S2, further comprising: S2-5, storing the historical operating state , the agent bid action and into an experience replay buffer, and updating the Critic and Actor network parameters by a formula for subsequent small-batch updating of network parameters; in the Critic network update, calculating a target value , is a discount factor, is the next historical operating state, ( is the target Critic network output of the next historical operating state, provided by the target network parameters ; the Critic network updates the parameters by minimizing the mean square error loss to approximate the real agent bid action; the Actor network updates the policy parameters according to the gradient signal output by the Critic, and the gradient calculation is wherein is the policy function generated by the Actor network, and the gradient signal output by the Critic is used to guide the Actor network update, The Actor network parameter gradient; the historical running state and the subject bidding action are stored in the experience replay buffer, the storage and reuse of the historical interaction information are realized, and the data diversity and training stability when small batch training are ensured. The Critic network approximates the value function output of the real subject bidding action by minimizing the mean square error loss, so that the policy evaluation can accurately reflect the actual market income situation. The Actor network updates the policy parameters using the gradient signal output by the Critic network, so that the subject optimizes the bidding action and improves the expected value of future rewards. The training mechanism can realize multi-round iteration optimization, balance exploration and utilization, and enhance the robustness and adaptability of the strategy.

[0110] In the embodiment, the construction method of the source network load storage cooperative optimization model comprises:

[0111] S3-1, acquiring a global system state , wherein the load prediction of each regional dispatching agent , renewable energy output prediction , energy storage state , historical market transaction data , network operation index ; each regional dispatching agent generates an initial solution based on the global state using the reinforcement learning Actor network, the initial solution includes generator output, energy storage power and controllable load adjustment amount, which provides a heuristic solution for subsequent optimization and reduces the search space of the solver;

[0112] S3-2, each regional dispatching agent generates an initial solution based on the global system state using the reinforcement learning Actor network ; the initial solution is subjected to power balance constraints and device operation constraints to generate a near feasible solution, so as to ensure that the initial solution meets the basic physical constraints of the power system and the safety limits of the device, and to improve the convergence speed and stability of the mathematical programming solution;

[0113] S3-3, the initial solution is subjected to power balance constraints and device operation constraints to generate a near feasible solution;

[0114] S3-4, taking the near feasible solution as the initial value, calling a mathematical programming solver for fine-tuning optimization, the target is to minimize the system comprehensive operation cost, and solving the global variable :

[0115]

[0116] Wherein F(x) is the system comprehensive operation cost, including generation cost , energy storage life degradation cost , network line loss , node voltage deviation and market deviation fines Ω represents the feasible region, which is jointly determined by power balance, capacity constraints, energy storage dynamics, line power flow constraints, and market commitment constraints. The mathematical programming solver is called with the near-feasible solution as the initial value to minimize the overall system operating cost F(x), including generation cost, energy storage degradation cost, network line loss, node voltage deviation, and market deviation penalty, so as to achieve global cost optimization.

[0117] S3-5. Decompose the global optimization problem into subproblems by region and solve them using ADMM in parallel iterative steps, where global variables... Let z be the local decision vector of the subproblem, z be the coupling variable representing the power of the boundary nodes, y be the multiplier used for consistency constraints, ρ be the penalty parameter controlling ADMM convergence, and the coupling matrix be... The boundary constraint c explicitly describes the linear relationship between the subproblems in each region and the global variable. The ADMM update formula is:

[0118] Local subproblem update:

[0119]

[0120] Global variable update:

[0121]

[0122] Multiplier Update:

[0123]

[0124] S3-6 continues until the local subproblem updates converge, then outputs the power allocation results for each region. Energy storage operation plan and load adjustment power .

[0125] In this embodiment, the specific method for establishing an uncertainty scenario generation model includes:

[0126] Generative models are constructed using variational autoencoders (VAEs) to learn from historical observations. Probability distribution of system operating state under certain conditions ;

[0127] Historical observation Mapped to the latent variable space by the encoder And perform reparameterization to generate latent variable samples. Use a decoder to generate samples of future system states. ;

[0128] The model is trained by maximizing the lower bound of evidence.

[0129]

[0130] wherein for ensuring the generated samples are close to the real state, for regularizing the latent variable distribution to be standard Gaussian to ensure the sample diversity and continuity; using the decoder to generate future system state samples from the latent variable, including various possible changes of load, renewable energy output and market price, covering extreme events and tail risk, and improving the representation ability of uncertainty; the VAE model is trained by maximizing the evidence lower bound (ELBO), so as to realize that the generated samples are close to the real system state, and the latent variable is regularized to be a standard Gaussian distribution, thereby ensuring the diversity and continuity of the generated samples;

[0131] sampling from the trained generative model a future operation state covering the possible changes of load, renewable energy output and market price;

[0132] calculating the system operation cost and loss for each future operation state and calculating the conditional value at risk CVaR at a confidence level

[0133]

[0134] wherein is the loss threshold of the confidence level quantile point; the scenario set S and the corresponding risk indicator CVaR are input into the multi-agent reinforcement learning scheduler of the scheduling layer; the generated scenario set and the corresponding CVaR are input into the multi-agent reinforcement learning (MARL) scheduler of the scheduling layer, which is used for subsequent rolling optimization and risk-constrained scheduling, thereby improving the decision robustness and economy of the system in an uncertain environment.

[0135] In the embodiment, the power balance constraint is satisfied as follows:

[0136]

[0137] wherein,

[0138] represents the unit active power optimization result of the region k at the time t;

[0139] represents the energy storage charging and discharging power of the region k at the time t;

[0140] represents the renewable energy power generation of the region k at the time t;

[0141] ​​represents the power load demand of region k at time t;

[0142] represents the power load demand of region k at time t;

[0143] The device operation constraints include energy storage operation constraints, load adjustment constraints, and network security constraints;

[0144] The energy storage operation constraints satisfy:

[0145]

[0146]

[0147] wherein,

[0148] represents the state of charge of the energy storage at time k;

[0149] respectively represent the charging efficiency and the discharging efficiency;

[0150] respectively represent the charging power and the discharging power of the energy storage at time k;

[0151] represents the length of the time interval;

[0152] represents the rated power of the energy storage device;

[0153] respectively represent the upper and lower threshold values of the state of charge of the energy storage;

[0154] The load adjustment constraints satisfy:

[0155]

[0156] wherein,

[0157] represents the load-side adjustable power of region k at time t;

[0158] represents the minimum and maximum values of the load adjustment;

[0159] The network security constraints satisfy:

[0160]

[0161] wherein,

[0162] represents the voltage amplitude of node i;

[0163] denotes the branch power transmission between node i and node j;

[0164] , denotes the upper and lower limits of the node voltage operation;

[0165] denotes the maximum transmission capacity of the branch.

[0166] Correspondingly, the present application also provides a system, comprising a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the method in the above

[0167] Correspondingly, the present application also provides an electronic device, comprising:

[0168] a processor;

[0169] a memory for storing processor-executable instructions;

[0170] wherein the processor is configured to invoke the instructions stored in the memory to execute the method in the above.

Claims

1. An optimized control method for an integrated source-grid-load-storage market mechanism, characterized in that: Includes the following steps: S1. Collect operational data from the source side, grid side, load side, and energy storage side; An uncertainty scenario generation model is established based on operational data to generate a scenario set that includes load forecasting information, renewable energy output forecasting information, and market price forecasting information. S2. Based on the scenario set, construct a market clearing model that includes the revenue function of market participants and the power grid constraints; The market clearing model receives bid information from the source side, load side, and energy storage side, and calculates the market clearing price and the transaction volume results; S3. Based on the market clearing price, transaction volume results, and operational data, establish a source-grid-load-storage collaborative optimization model; the source-grid-load-storage collaborative optimization model aims to minimize system operating costs, and the constraints include power balance constraints and equipment operation constraints; output the power allocation results and energy storage operation plan for each region. S4. Based on the power allocation results of each region and the energy storage operation plan, the energy storage power output and load adjustment power are calculated using a rolling optimization algorithm; S5. Based on the power output and load adjustment determined in step S4, the power control system of source grid, load and storage is operated.

2. The optimized control method for the integrated source-grid-load-storage market mechanism as described in claim 1, characterized in that: S1-1. Collect operational data from the source side, grid side, load side, and energy storage side, including actual load output, actual renewable energy output, energy storage status, historical market transaction data, and network operation indicators, to form a system historical status set. The set of historical states Including historical data and historical operating status ; S1-2, Based on the historical state set Establish an uncertain scenario generation model Input is historical data and historical operating status Output future prediction scenarios : S1-3. Generate a scenario set containing load forecasting information, renewable energy output forecasting information, and market price forecasting information using the uncertainty scenario generation model. .

3. The optimized control method for the integrated source-grid-load-storage market mechanism according to claim 2, characterized in that: S2-1, Market entity i obtains historical operating status. Including load forecasting Renewable energy output forecast Energy storage status Historical market transaction data And network constraint metrics, used to describe the state of the environment; S2-2, The market entity will record its historical operating status. Input to an Actor network, which generates a policy function. Generate the main bidding action ,in For the reported supply or demand, This corresponds to the quoted price; S2-3, Perform the main bidding action. Input market clearing operator Calculate the transaction volume results for each market participant. With market clearing price ; S2-4, By adjusting the market clearing price Multiply by the transaction volume result Calculate revenue, minus production and supply costs. minus penalties for breach of contract After deducting the equivalent lifetime cost of energy storage degradation Calculate instant rewards: 。 4. The optimized control method for the integrated source-grid-load-storage market mechanism according to claim 3, characterized in that: S2-5, Record historical operating status Main bidding action and The data is stored in the experience replay buffer, and the Critic and Actor network parameters are updated using formulas for subsequent mini-batch network parameter updates. In the Critic network update, the target value is calculated. , As a discount factor, For the next historical running state, ( The target Critic network output for the next historical running state is determined by the target network parameters. Provided; the Critic network minimizes the mean square error loss. To update parameters The Actor network approximates the actual bidding actions of real entities; it updates its policy parameters based on the gradient signals output by the Critic network. The gradient is calculated as follows: ,in The policy function generated by the Actor network, and the gradient signal output by the Critic. Used to guide Actor network updates. The gradient of the Actor network parameters.

5. The optimized control method for the integrated source-grid-load-storage market mechanism according to claim 4, characterized in that: The method for constructing the source-grid-load-storage collaborative optimization model includes: S3-1. Obtain global system status This includes load forecasting for each regional dispatch agent. Renewable energy output forecast Energy storage status Historical market transaction data Network operation indicators ; S3-2, Regional-level scheduling agents are based on the global system state. Initial solutions are generated using reinforcement learning Actor networks. ; S3-3, Regarding the initial solution A near-feasible solution is generated by applying power balance constraints and equipment operation constraints. S3-4. Using a near-feasible solution as the initial value, call the mathematical programming solver for fine-tuning and optimization, with the goal of minimizing the overall system operating cost, and solve for the global variables. : Where F(x) is the overall operating cost of the system, including the cost of power generation. Energy storage lifespan degradation costs Network line loss Node voltage deviation and market deviation fines Ω represents the feasible region, which is jointly determined by power balance, capacity constraints, energy storage dynamics, line power flow constraints, and market commitment constraints. S3-5. Decompose the global optimization problem into subproblems by region and solve them using ADMM in parallel iterative steps, where global variables... Let z be the local decision vector of the subproblem, z be the coupling variable representing the power of the boundary nodes, y be the multiplier used for consistency constraints, ρ be the penalty parameter controlling ADMM convergence, and the coupling matrix be... The boundary constraint c explicitly describes the linear relationship between the subproblems in each region and the global variable. The ADMM update formula is: Local subproblem update: Global variable update: Multiplier Update: S3-6 continues until the local subproblem updates converge, then outputs the power allocation results for each region. Energy storage operation plan and load adjustment power .

6. The optimized control method for the integrated source-grid-load-storage market mechanism according to claim 2, characterized in that: Specific methods for establishing uncertainty scenario generation models include: Generative models are constructed using variational autoencoders (VAEs) to learn from historical observations. Probability distribution of system operating state under certain conditions ; Historical observation Mapped to the latent variable space by the encoder And perform reparameterization to generate latent variable samples. Use a decoder to generate samples of future system states. ; The model is trained by maximizing the lower bound of evidence. in This is used to ensure that the generated samples are close to the real state. This is used to regularize the latent variable distribution to a standard Gaussian to ensure sample diversity and continuity; Sampling from the trained generative model Future operating status Potential changes in load, renewable energy output, and market prices; For each future running state Calculate system operating costs and losses and at confidence level The following conditions are used to calculate the Value at Risk (CVaR): in Confidence level The loss threshold of the quantile; the scenario set S and its corresponding risk index CVaR are used as input and passed to the multi-agent reinforcement learning scheduler of the scheduling layer.

7. The optimized control method for the integrated source-grid-load-storage market mechanism according to claim 5, characterized in that: The power balance constraint satisfies: in, This represents the optimized active power output of the units in region k at time t; This represents the energy storage charging and discharging power of region k at time t; This represents the renewable energy power generation capacity of region k at time t; This represents the electricity load demand of region k at time t; This represents the electricity load demand of region k at time t; The equipment operation constraints include energy storage operation constraints, load adjustment constraints, and network security constraints. The energy storage operation constraints satisfy: in, This represents the state of charge of the stored energy at time k; These represent charging efficiency and discharging efficiency, respectively. These represent the charging power and discharging power of the energy storage at time k, respectively. Indicates the length of the time interval; Indicates the rated power of the energy storage device; These represent the upper and lower thresholds of the energy storage state of charge, respectively. The load adjustment constraint satisfies: in, This represents the adjustable power on the load side of region k at time t; This indicates the minimum and maximum adjustable load values; The network security constraints satisfy: in, This represents the voltage magnitude at node i; This represents the branch power transfer between node i and node j; , Indicates the upper and lower limits of the node voltage operation; This indicates the maximum transmission capacity of the branch.

8. A system comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, it implements the method as described in any one of claims 1-8.

9. An electronic device, characterized in that: include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in memory to execute the method of any one of claims 1 to 8.

Images