Large-model-driven power system intelligent agent optimization scheduling system and large-model-driven power system intelligent agent optimization scheduling method

Abstract

The invention discloses a large-model-driven power system intelligent agent optimization scheduling system, and the system comprises a data collection module which collects the real-time operation data of a power system, and the real-time operation data comprise the power demand, the power grid operation state, and the power grid load change; the power grid prediction model construction module is used for presetting and constructing a plurality of models by adopting a deep learning algorithm on the basis of historical operation data and real-time operation data of the power system, and predicting power demands, power grid operation states and power grid load changes within preset time; and the agent optimization scheduling module is used for defining a plurality of agents, obtaining an optimal solution by adopting a reinforcement learning algorithm according to the real-time acquired data and the prediction result on the premise of meeting the target function of minimizing the total operation cost of the system, and carrying out optimization scheduling on the operation state of each agent. According to the method, the target function is set and the reinforcement learning algorithm is combined to optimize the operation state of the intelligent agent, so that economic operation is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power dispatching and optimization, and in particular to a large model-driven electric power system intelligent agent optimization dispatching system and method. Background Art

[0002] With the rapid development of modern power systems and the transformation of energy structures, power grid dispatching faces unprecedented challenges in improving energy efficiency, ensuring power supply security, and promoting the integration of new energy sources into the grid. The dispatching of traditional power systems mainly relies on manual experience and some preset rules to start and stop generators, distribute loads, and other operations. With the increasing complexity of power systems, especially the introduction of renewable energy, the volatility and uncertainty of power supply have increased, and traditional dispatching methods are difficult to meet the requirements of efficiency, economy, and reliability. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture.

[0004] The present invention provides a large model driven power system intelligent agent optimization dispatching system, comprising the following:

[0005] Data acquisition module: collects real-time operation data of the power system, including power demand, grid operation status and grid load changes;

[0006] Power grid prediction model building module: Based on the historical and real-time operation data of the power system, it uses deep learning algorithms to build multiple models in advance to predict the power demand, power grid operation status and power grid load changes within a preset time;

[0007] Intelligent agent optimization scheduling module: define multiple intelligent agents, and based on the real-time collected data and prediction results, use the reinforcement learning algorithm to obtain the optimal solution and optimize the operation status of each intelligent agent under the premise of satisfying the objective function of minimizing the total cost of system operation. The intelligent agents include the power generation group agent, the load side agent and the energy storage system agent.

[0008] Preferably, the agent optimization module comprises:

[0009] Agent definition unit: used to define multiple agents, including power generation group agent, load side agent and energy storage system agent. Each agent has independent operating status and optimization objectives.

[0010] Data receiving unit: used to receive the real-time operation data provided by the data acquisition module and the prediction results provided by the power grid prediction model building module;

[0011] Objective function setting unit: used to set the objective function of minimizing the total cost of system operation and serve as a constraint condition for the agent's optimal scheduling;

[0012] Reinforcement learning algorithm unit: Using reinforcement learning algorithm, based on real-time data and prediction results, and on the premise of meeting constraints, the operating status of each intelligent agent is optimized and scheduled to obtain the optimal solution.

[0013] Preferably, the reinforcement learning algorithm unit includes:

[0014] State space definition subunit: used to define the state space of each agent, including the power generation of the power generation group agent, the load demand change of the load side agent, and the charging and discharging state of the energy storage system agent;

[0015] Action space definition subunit: used to define the action space of each agent, including power adjustment of the power generation group agent, load management strategy of the load side agent, and charging and discharging decision of the energy storage system agent;

[0016] Reward function design subunit: preset reward function to evaluate the effect of each agent taking actions in different states. The reward function is associated with the objective function of minimizing the power generation cost and the total system operation cost.

[0017] Strategy learning subunit: Adopts reinforcement learning algorithm, learns the optimal strategy of each agent based on real-time data and prediction results, and satisfies constraints, so as to achieve optimal scheduling of the running status of the agent;

[0018] Optimal solution generation subunit: Based on the learned strategy, the operating status of each intelligent agent is optimized and scheduled to generate the optimal solution.

[0019] Preferably, the power grid prediction model building module includes:

[0020] Data preprocessing unit: preprocesses the historical operation data and real-time operation data of the power system, including data cleaning, missing value filling and normalization processing;

[0021] Power grid prediction model training unit: Based on the pre-processed historical operation data, multiple models are constructed using deep learning algorithms;

[0022] Prediction function implementation unit: Based on multiple trained models, real-time operation data is input into multiple models to predict power demand, grid operation status, and grid operation status;

[0023] Prediction result output unit: outputs the prediction results as input data for agent optimization scheduling.

[0024] Preferably, the data preprocessing unit includes:

[0025] Data cleaning subunit: cleans the historical and real-time operation data of the power system to remove noise data, abnormal values ​​and duplicate data;

[0026] Missing value filling subunit: Process the missing values ​​of the cleaned data and use interpolation to fill in the missing data;

[0027] Normalization processing subunit: normalize the data and convert the data of different dimensions to the same scale range to eliminate the dimensional differences between the data;

[0028] Dataset generation subunit: divides the preprocessed data into training set, validation set and test set, and outputs them to the multi-model construction unit.

[0029] The present invention provides a large model driven power system intelligent agent optimization dispatching method for executing the large model driven power system intelligent agent optimization dispatching system, comprising the following steps:

[0030] S1: Collect real-time operation data of the power system, including power demand, grid operation status and grid load changes;

[0031] S2: Based on the historical and real-time operation data of the power system, a deep learning algorithm is used to build multiple models to predict the power demand, grid operation status and grid load changes within a preset time.

[0032] S3: Define multiple intelligent agents, and based on the real-time collected data and prediction results, use the reinforcement learning algorithm to obtain the optimal solution and optimize the operation status of each intelligent agent under the premise of satisfying the objective function of minimizing the total cost of system operation. The intelligent agents include the power generation group intelligent agent, the load side intelligent agent and the energy storage system intelligent agent.

[0033] The present invention discloses a large-model driven power system intelligent agent optimization dispatching system and method, which has the following beneficial effects: by quickly predicting power demand and operating status and generating a real-time optimization dispatching plan, manual intervention is reduced and dispatching efficiency is improved; by setting an objective function and optimizing the operating status of the intelligent agent in combination with a reinforcement learning algorithm, the power generation cost, energy storage cost and load management cost can be effectively reduced to achieve economic operation; by optimizing dynamic power balance constraints, the supply and demand balance is ensured to avoid grid overload or power waste; the participation of the energy storage system intelligent agent improves the adaptability to load fluctuations and enhances the stability and reliability of the grid; by designing a distributed optimization model of the power generation group intelligent agent, the load side intelligent agent and the energy storage system intelligent agent, each of which operates independently and makes collaborative decisions, the global optimization of system resources is achieved, and the system incoordination caused by single-objective optimization is avoided; BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A method flow chart of a large model-driven power system intelligent agent optimization dispatching method provided by the present invention; DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0038] refer to Figure 1 The present invention provides a large model driven power system intelligent agent optimization dispatching system, including the following:

[0039] Data acquisition module: collects real-time operation data of the power system, including power demand, grid operation status and grid load changes;

[0040] Power grid prediction model building module: Based on the historical and real-time operation data of the power system, it uses deep learning algorithms to build multiple models in advance to predict the power demand, power grid operation status and power grid load changes within a preset time;

[0041] Intelligent agent optimization scheduling module: define multiple intelligent agents, and based on the real-time collected data and prediction results, use the reinforcement learning algorithm to obtain the optimal solution and optimize the operation status of each intelligent agent under the premise of satisfying the objective function of minimizing the total cost of system operation. The intelligent agents include power generation group agents, load side agents and energy storage system agents.

[0042] The invention provides a large-model driven power system intelligent agent optimization and dispatching system, which can quickly predict power demand and operating status, and generate real-time optimization dispatching schemes, reduce manual intervention, and improve dispatching efficiency; by setting the objective function and optimizing the operating status of the intelligent agent in combination with the reinforcement learning algorithm, it can effectively reduce the power generation cost, energy storage cost, and load management cost, and realize economic operation; by optimizing the dynamic power balance constraint, it ensures the balance between supply and demand, and avoids grid overload or power waste; the participation of the energy storage system intelligent agent improves the adaptability to load fluctuations, and enhances the stability and reliability of the grid; by designing the distributed optimization model of the power generation group intelligent agent, the load side intelligent agent, and the energy storage system intelligent agent, each of which operates independently and makes collaborative decisions, it realizes the global optimization of system resources, and avoids the system incoordination caused by single-objective optimization;

[0043] In a preferred embodiment, the agent optimization module includes:

[0044] Agent definition unit: used to define multiple agents, including power generation group agent, load side agent and energy storage system agent. Each agent has independent operating status and optimization objectives.

[0045] Data receiving unit: used to receive the real-time operation data provided by the data acquisition module and the prediction results provided by the power grid prediction model building module;

[0046] Objective function setting unit: used to set the objective function of minimizing the total cost of system operation and serve as a constraint condition for the agent's optimal scheduling;

[0047] Specifically, the total cost objective function of the system operation is:

[0048] C total =C g +C s +C l ,

[0049] Among them, C g is the total electricity generation cost;

[0050] C s The operating cost of the energy storage system;

[0051] C1 is the load adjustment cost.

[0052] The constraints of the total system operation cost objective function include power balance constraints, power generation unit constraints, and energy storage unit constraints;

[0053] The power balance constraint is:

[0054]

[0055] P d is the load demand;

[0056] P loss is the power loss of the transmission grid;

[0057] P sj is the power output of the energy storage system;

[0058] N s is the number of energy storage systems;

[0059] N g is the number of power generation units;

[0060] Pgi is the output power of the i-th power generation unit.

[0061] The power generation unit limits are:

[0062]

[0063] Energy storage unit limits are:

[0064]

[0065] SOC j (t+1): energy storage state at time t+1;

[0066] SOC j (t): energy storage state at time t;

[0067] P sj (t): power output of the energy storage system at time t;

[0068] Δt: time step;

[0069] C j : The capacity of the energy storage system.

[0070] When charging (P sj <0)SOC j Increase; when discharging (P sj >0)SOC j Reduce, SOC j The change of is limited by the current power output and system capacity and must satisfy [SOC j min , SOC j max ] range. [SOC j min , SOC j max ]The range is set according to battery type, life protection requirements and operational safety.

[0071] Reinforcement learning algorithm unit: Using reinforcement learning algorithm, based on real-time data and prediction results, and on the premise of meeting constraints, the operating status of each intelligent agent is optimized and scheduled to obtain the optimal solution.

[0072] In a preferred embodiment, the reinforcement learning algorithm unit includes:

[0073] State space definition subunit: used to define the state space of each agent, including the power generation of the power generation group agent, the load demand change of the load side agent, the charging and discharging state of the energy storage system agent, and the grid operation state;

[0074] Action space definition subunit: used to define the action space of each agent, including power adjustment of the power generation group agent, load management strategy of the load side agent, and charging and discharging decision of the energy storage system agent;

[0075] Reward function design subunit: used to preset the reward function, which is used to evaluate the effect of each agent taking actions in different states. The reward function is associated with the objective function of minimizing the power generation cost and the total system operation cost;

[0076] Specifically, the reward function is:

[0077] R t =-C total +λ1·Penalty balance +λ2·Penalty constraints

[0078] C total : Total system operating cost, including power generation cost, energy storage cost and load adjustment cost;

[0079] Penaltybalanc e : Power balance constraint breach penalty to ensure supply and demand balance;

[0080] Penalty constraints : Penalties for violating agent constraints (such as SOC range, power upper and lower limits);

[0081] λ1, λ2: weight coefficients used to balance the optimization objectives and constraints.

[0082] Strategy learning subunit: Adopts reinforcement learning algorithm, learns the optimal strategy of each agent based on real-time data and prediction results, and satisfies constraints, so as to achieve optimal scheduling of the running status of the agent;

[0083] Optimal solution generation subunit: Based on the learned strategy, the operating status of each intelligent agent is optimized and scheduled to generate the optimal solution.

[0084] Specifically,

[0085] Enter the current state S t To the trained policy network, output action A t A t =πθ(St)

[0086] Action A t Including: Power adjustment of the power generation group intelligent body ΔP gi ; Charging and discharging decision of energy storage intelligent body ΔSOC j ; Demand response adjustment of load agents.

[0087] Check whether the generated action meets the constraints (such as power balance, energy storage SOC range, etc.). If the constraints are violated, adjust the action based on the penalty function: A t ′=At-λ.Penalty, execute the adjusted action and update the system status: S t +1=f(S t , A t ′) Where f is the system state transfer function, which represents the impact of the current action on the system state.

[0088] According to the next moment state S t+1 and updated real-time data, repeatedly executes the strategy network generation action, and outputs the optimal solution. The optimal solution includes: the optimal output power of each power generation unit, the optimal charging and discharging strategy, and the optimal load response strategy.

[0089] This embodiment effectively reduces dependence on traditional power generation units, improves the absorption capacity of renewable energy, reduces carbon emissions, and promotes the use of clean energy through the charge and discharge optimization of the energy storage system intelligent body and the demand response strategy of the load-side intelligent body.

[0090] In a preferred embodiment, the power grid prediction model building module includes

[0091] Data preprocessing unit: preprocess the historical operation data and real-time operation data of the power system, including data cleaning, missing value filling, and normalization processing;

[0092] Power grid prediction model training unit: Based on the pre-processed historical operation data, multiple models are constructed using deep learning algorithms;

[0093] Prediction function implementation unit: Based on multiple trained models, real-time operation data is input into multiple models to predict power demand, grid operation status, and grid operation status;

[0094] Prediction result output unit: outputs the prediction results as input data for agent optimization scheduling.

[0095] Specifically, a long short-term memory network is used to build a power demand forecasting model to predict the power load demand within a preset time. A graph neural network is used to build a power grid operation status forecasting model to predict the power grid operation status within a preset time, such as node voltage, line flow, etc.; a gated recurrent unit is used to build a power grid load change forecasting model to predict the dynamic change trend of the power grid load within a preset time.

[0096] In a preferred embodiment, the data preprocessing unit comprises:

[0097] Data cleaning subunit: cleans the historical and real-time operation data of the power system to remove noise data, abnormal values ​​and duplicate data;

[0098] Missing value filling subunit: Process the missing values ​​of the cleaned data and use interpolation to fill in the missing data;

[0099] Normalization processing subunit: normalize the data and convert the data of different dimensions to the same scale range to eliminate the dimensional differences between the data;

[0100] Dataset generation subunit: divides the preprocessed data into training set, validation set and test set, and outputs them to the multi-model construction unit.

[0101] refer to Figure 1 The present invention provides a large model driven power system intelligent agent optimization dispatching method for executing a large model driven power system intelligent agent optimization dispatching system, characterized in that it includes the following steps:

[0102] S1: Collect real-time operation data of the power system, including power demand, grid operation status and grid load changes;

[0103] S2: Based on the historical and real-time operation data of the power system, a deep learning algorithm is used to build multiple models to predict the power demand, grid operation status and grid load changes within a preset time.

[0104] S3: Define multiple intelligent agents. Based on the real-time collected data and prediction results, and under the premise of satisfying the objective function of minimizing the total cost of system operation, use the reinforcement learning algorithm to obtain the optimal solution and optimize the operation status of each intelligent agent. The intelligent agents include the power generation group agent, the load side agent and the energy storage system agent.

Claims

1. A large model driven power system intelligent agent optimization dispatching system, characterized in that: These include: Data acquisition module: collects real-time operation data of the power system, including power demand, grid operation status and grid load changes; A power grid prediction model building module: based on the historical operation data of the power system and the real-time operation data, a deep learning algorithm is used to preset and build multiple models to predict the power demand, power grid operation status and power grid load changes within a preset time; Intelligent agent optimization scheduling module: define multiple intelligent agents, and based on the real-time collected data and prediction results, use the reinforcement learning algorithm to obtain the optimal solution and optimize the operation status of each intelligent agent under the premise of satisfying the objective function of minimizing the total cost of system operation. The intelligent agents include the power generation group agent, the load side agent and the energy storage system agent.

2. According to the large model driven power system intelligent agent optimization dispatching system of claim 1, it is characterized in that: The intelligent agent optimization module comprises: Agent definition unit: used to define multiple agents, including power generation group agent, load side agent and energy storage system agent. Each agent has independent operating status and optimization objectives. Data receiving unit: used to receive the real-time operation data provided by the data acquisition module and the prediction results provided by the power grid prediction model building module; Objective function setting unit: used to set the objective function of minimizing the total cost of system operation and serve as a constraint condition for the agent's optimal scheduling; Reinforcement learning algorithm unit: Using reinforcement learning algorithm, based on real-time data and prediction results, and on the premise of meeting constraints, the operating status of each intelligent agent is optimized and scheduled to obtain the optimal solution.

3. A large model driven power system intelligent agent optimization dispatching system according to claim 2, characterized in that: The reinforcement learning algorithm unit comprises: State space definition subunit: used to define the state space of each agent, including the power generation of the power generation group agent, the load demand change of the load side agent, and the charging and discharging state of the energy storage system agent; Action space definition subunit: used to define the action space of each agent, including power adjustment of the power generation group agent, load management strategy of the load side agent, and charging and discharging decision of the energy storage system agent; Reward function design subunit: preset reward function to evaluate the effect of each agent taking actions in different states. The reward function is associated with the objective function of minimizing the power generation cost and the total system operation cost. Strategy learning subunit: Adopts reinforcement learning algorithm, learns the optimal strategy of each agent based on real-time data and prediction results, and satisfies constraints, so as to achieve optimal scheduling of the running status of the agent; Optimal solution generation subunit: Based on the learned strategy, the operating status of each intelligent agent is optimized and scheduled to generate the optimal solution.

4. A large model driven power system intelligent agent optimization dispatching system according to claim 1, characterized in that: The power grid prediction model building module includes: Data preprocessing unit: preprocesses the historical operation data and real-time operation data of the power system, including data cleaning, missing value filling and normalization; Power grid prediction model training unit: Based on the pre-processed historical operation data, multiple models are constructed using deep learning algorithms; Prediction function implementation unit: Based on multiple trained models, real-time operation data is input into multiple models to predict power demand, grid operation status, and grid operation status; Prediction result output unit: outputs the prediction results as input data for agent optimization scheduling.

5. A large model driven power system intelligent agent optimization dispatching system according to claim 2, characterized in that: The data preprocessing unit comprises: Data cleaning subunit: cleans the historical and real-time operation data of the power system to remove noise data, abnormal values ​​and duplicate data; Missing value filling subunit: Process the missing values ​​of the cleaned data and use interpolation to fill in the missing data; Normalization processing subunit: normalize the data and convert the data of different dimensions to the same scale range to eliminate the dimensional differences between the data; Dataset generation subunit: divides the preprocessed data into training set, validation set and test set, and outputs them to the multi-model construction unit.

6. A method for executing a large model driven power system intelligent agent optimization dispatching system according to any one of claims 1 to 5, characterized in that: The steps include: S1: Collect real-time operation data of the power system, including power demand, grid operation status and grid load changes; S2: Based on the historical and real-time operation data of the power system, a deep learning algorithm is used to build multiple models to predict the power demand, grid operation status and grid load changes within a preset time. S3: Define multiple intelligent agents, and based on the real-time collected data and prediction results, use the reinforcement learning algorithm to obtain the optimal solution and optimize the operation status of each intelligent agent under the premise of satisfying the objective function of minimizing the total cost of system operation. The intelligent agents include the power generation group intelligent agent, the load side intelligent agent and the energy storage system intelligent agent.

Images