Power grid operation control method for realizing source grid load storage coordination

Abstract

The invention discloses a power grid operation control method for realizing source-grid-load-storage cooperation, and relates to the technical field of power grid operation control, and the method comprises the steps: carrying out the ultra-short-term prediction of power grid operation data through LSTM, ARIMA, VAE, MDP and other models, covering the new energy output, power flow change, load demand and energy storage state, and carrying out the fusion of the day-ahead prediction and real-time prediction data, thereby achieving the real-time prediction of the power grid operation data. Optimizing a prediction result by using a weighting method; in the aspect of scheduling optimization, a scheduling optimization model is established according to prediction data, a day-ahead scheduling plan is calculated, the operation cost of a power grid is minimized, and meanwhile, source-side wind and light output, grid-side power flow regulation and control, load-side demand response and a storage-side charging and discharging strategy are optimized; scheduling optimization is introduced, so that the output of a conventional unit is reduced in a wind power high output period, and the new energy consumption rate is improved; in the real-time scheduling stage, dynamic optimization is carried out by adopting MPC in combination with a day-ahead scheduling plan and real-time prediction data; and when the wind power output is excessive, the energy storage equipment is controlled to absorb redundant electric energy.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid operation control, and in particular to a power grid operation control method for realizing source-grid-load-storage coordination. Background Art

[0002] Grid operation control has always been the core of stable operation of energy systems. Traditional power grids rely on large centralized generating units, such as thermal power and hydropower, to provide supply and demand balance through fixed dispatching strategies. The dispatching methods are mostly based on day-ahead planning, combined with some intraday adjustments, and mainly rely on the output changes of controllable units to adapt to load fluctuations.

[0003] Although the introduction of renewable energy power generation has reduced the pressure on traditional power generation to a certain extent, due to the volatility and uncertainty of its power generation, when renewable energy is connected to the grid, it is necessary to rely on deep peak regulation of coal-fired units or even take power rationing measures. In addition, the changes in load-side demand are becoming increasingly complex. During peak hours, the power grid needs to dispatch additional standby units or purchase electricity. When the power demand is low, some units still need to operate due to the minimum stable output limit, resulting in reduced energy utilization efficiency. In order to improve adaptability, some traditional solutions introduce flexible resources such as energy storage systems and demand response, adjust the charging and discharging strategies of energy storage or guide the load side to participate in regulation, and enhance the regulation capacity of the power grid. However, existing methods are mostly based on fixed rules, which are difficult to make real-time adaptive adjustments under complex renewable energy fluctuation scenarios. On the other hand, as an important part of urban energy supply, the combined heat and power (CHP) system's "heat-based electricity" operating mode meets the heating demand while also limiting the flexibility of power dispatch. During the winter heating period, the output of thermal power units is subject to the heat load demand. Even if the wind power output is sufficient, a relatively high level of thermal power generation needs to be maintained, resulting in limited wind power consumption and even large-scale wind power abandonment.

[0004] In this context, there is an urgent need for a power grid operation control solution that can achieve coordination between sources, grids, loads and storage. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides a grid operation control method for realizing source-grid-load-storage coordination to solve the problems that traditional cogeneration CHP limits wind power consumption in a heat-to-power mode, scheduling lags, energy storage is not deeply integrated with CHP, and it is difficult to adapt to new energy fluctuations.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: The embodiment of the present invention provides a power grid operation control method for realizing source-grid-load-storage coordination, which comprises: Step S1, collecting grid operation data, and using a prediction model to make ultra-short-term predictions on the output of new energy on the source side, the change of the power flow on the grid side, the load demand on the load side, and the energy storage status on the storage side, and generating day-ahead prediction data and real-time prediction data; Step S2, based on the forecast data of S1, establish a scheduling optimization model and calculate the day-ahead scheduling plan; Step S3, combining the day-ahead dispatch plan of S2 and the real-time forecast data of S1, adopts model predictive control MPC to perform dynamic optimization, including: new energy output on the source side, power flow regulation on the grid side, load adjustment on the load side, charging and discharging on the storage side, and giving priority to wind power for heating; Step S4, based on the optimization result of S3, reinforcement learning is used to dynamically respond to the load demand on the load side.

[0008] As a preferred scheme of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, the power grid operation data includes source-side new energy output data, grid-side power grid flow data, load-side load demand data, storage-side energy storage status data, as well as meteorological data and electricity price data.

[0009] As a preferred solution of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, the steps of performing ultra-short-term prediction and generating day-ahead prediction data and real-time prediction data are as follows: Based on LSTM model, ARIMA model, variational autoencoder VAE and Markov decision process MDP, ultra-short-term forecasting is performed to generate day-ahead forecast data and real-time forecast data. Long short-term memory network LSTM is used for nonlinear time series prediction to obtain the wind power and photovoltaic output forecast values: and ,in, For the moment Forecasted wind power output, For the moment Forecasted PV output, The autoregressive moving average model ARIMA is used to predict the power flow of the power grid. The prediction formula is: , in, For the moment The flow data, is the order of the autoregressive term, is the autoregression coefficient, for Current trend data, is the moving average term order, is the moving average coefficient, for The random error of time, is the random error at the current moment, Using variational autoencoder VAE for load demand modeling: , , in, is a latent variable, is a nonlinear activation function, is the encoder weight matrix, is the historical load data, is the encoder bias, Reconstruct data for the decoder, is the decoder weight matrix, is the decoder bias, The Markov decision process MDP is used for modeling: , , in, For the moment The energy storage state, For the moment The control actions taken, is the state transfer function, The energy storage state for the next moment, For instant rewards, is the reward function, Combined with day-ahead forecast With real-time prediction , the fusion formula is: , in, is the fused prediction data, is the weight coefficient, is the day-ahead forecast data, For real-time prediction data.

[0010] As a preferred scheme of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, wherein: the dispatch optimization model integrates the coordinated constraints of the source side, grid side, load side, and storage side, optimizes the wind and solar output on the source side, the flow control on the grid side, the demand response on the load side, and the charging and discharging strategy on the storage side, and uses reinforcement learning to calculate the day-ahead dispatch plan to reduce the output of conventional units on the source side during periods of high wind power output.

[0011] As a preferred solution of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, wherein: the step of establishing a scheduling optimization model based on the prediction data of S1 and calculating the day-ahead scheduling plan is as follows: The objective function is defined as minimizing the grid operation cost, and the formula is: , in, is the scheduling optimization objective function, is the day-ahead scheduling period, is the output cost of the source-side unit, is the energy storage system cost, is the demand response cost, The output of the source side unit, is the energy storage charging and discharging power, Adjust power for load, Define constraints, including: 1) Grid power balance constraints: , in, is the number of grid nodes, For Node The power at 2) Energy storage charging and discharging constraints: , , in, In the energy storage state, For charging efficiency, is the charging power, is the discharge efficiency, is the discharge power, are the upper and lower limits of the system, 3) Demand response constraints: , , in, is the load regulation factor, is the reference load; Reinforcement learning optimizes scheduling strategies, , in, is the optimal scheduling strategy, is the current grid status, is an optional scheduling action, is the Q-value function for reinforcement learning.

[0012] As a preferred solution of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, in step S3, When wind power output is excessive, control the storage-side equipment to absorb excess power and reduce the output of conventional units; When wind power is insufficient, the storage-side equipment is controlled to release stored energy.

[0013] As a preferred solution of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, the steps of combining the day-ahead dispatch plan of S2 and the real-time forecast data of S1 and using model predictive control MPC for dynamic optimization are as follows: Model predictive control (MPC) is used to perform dynamic optimization by combining the real-time forecast data of S1 and the day-ahead scheduling plan of S2, and the objective function is defined as: , in, To dynamically optimize the objective function, is the MPC prediction time domain length, is the output cost of the source-side unit, is the energy storage cost, is the demand response cost, is the penalty term for the day-ahead scheduling deviation, is the deviation between real-time scheduling and day-ahead scheduling, is the real-time power demand predicted in step S1, is the day-ahead scheduling plan calculated in step S2, The state space model is: , , in, is a state variable, including the output of new energy , flow distribution , load demand , energy storage state , is the system state transfer matrix, is the control input matrix, The control variables include wind power regulation, energy storage charging and discharging power, and load adjustment. is the system noise, is the observed variable, is the observation matrix, To measure noise.

[0014] As a preferred solution of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, the step of using model predictive control MPC for dynamic optimization also includes: Define constraints, including: 1) Grid power balance constraints: , in, is the number of grid nodes, For Node The power at The output of the source side unit, is the charging and discharging power of the energy storage system, Adjust power for load, 2) Energy storage dynamic control constraints: , , in, is the remaining power of the energy storage system, For charging efficiency, is the charging power, is the discharge efficiency, is the discharge power, and are the minimum and maximum capacities of the energy storage system, 3) Wind power consumption strategy: like ,but , like ,but , in, Indicates the power that the energy storage system needs to absorb when there is excess wind power. Indicates the power that the energy storage system needs to supplement when wind power is insufficient. To contribute to wind power, is the load demand, MPC rolling optimization solution, the formula is: , , in, To predict the control time domain length, is the control input set, Indicates that in the rolling optimization time domain Time index within Indicates time The objective function value of is the optimal control input at the current moment, Update the value for the next state.

[0015] As a preferred solution of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, the load-side dynamic demand response includes: When electricity prices are low, guide loads to increase electricity consumption; When electricity prices are at their peak, reduce the power generation of conventional units and give priority to the use of wind power.

[0016] As a preferred solution of the power grid operation control method for realizing source-grid-load-storage coordination described in the present invention, the step of using reinforcement learning to dynamically respond to the load demand on the load side based on the optimization result of S3 is: Define the objective function, the goal is to maximize the long-term benefits of the system, the optimal demand response strategy To maximize the cumulative reward, the formula is: , in, represents the expected value, is the discount factor, For the moment Instant rewards, is the demand response cycle, Reinforcement learning uses the Qlearning method to update the Q value function. The update formula is: , in, is the optimal Q value at the next moment, is the state action value, is the learning rate, is the discount factor. The beneficial effects of the present invention are as follows: the present invention uses LSTM, ARIMA, VAE and MDP models to perform ultra-short-term prediction of power grid operation data, covering new energy output, flow changes, load demand and energy storage status, and optimizes the prediction results by fusing day-ahead prediction with real-time prediction data using a weighted method; in terms of scheduling optimization, a scheduling optimization model is established based on the prediction data, and a day-ahead scheduling plan is calculated to minimize the power grid operation cost, while optimizing the source-side wind and solar output, grid-side flow regulation, load-side demand response and storage-side charging and discharging strategies; scheduling optimization is introduced to reduce the output of conventional units during high wind power output periods and improve the new energy consumption rate; in the real-time scheduling stage, MPC is used for dynamic optimization in combination with the day-ahead scheduling plan and real-time prediction data; when wind power output is in excess, the energy storage equipment is controlled to absorb excess electricity and reduce the output of conventional units; when wind power is insufficient, the energy storage equipment is controlled to release electricity and reduce dependence on thermal power units.

[0017] In the present invention, MPC is based on state space modeling, combined with a forecast rolling window for real-time adjustment, and uses reinforcement learning to dynamically respond to load demand on the load side, so that user loads can increase electricity consumption when electricity prices are low, reduce conventional unit power generation when electricity prices are peak, and give priority to wind power supply, thereby maximizing long-term benefits and minimizing grid operation costs.

[0018] In summary, the present invention effectively alleviates the limitation of the traditional cogeneration system's "heat-based electricity" mode on wind power consumption, reduces the peak-shaving pressure on the power grid, and optimizes the utilization efficiency of the energy storage system, so that energy storage no longer relies solely on fixed charging and discharging rules, but can be optimally scheduled according to real-time demand; in addition, the present invention combines multi-level optimization to achieve deep coordination of source, grid, load and storage, effectively improving the new energy consumption capacity. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0020] Figure 1 The present invention is a flow chart of a method for controlling power grid operation to achieve source-grid-load-storage coordination. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0024] Example 1, reference Figure 1 This embodiment provides a power grid operation control method for realizing source-grid-load-storage coordination, comprising the following steps: Step S1, collecting grid operation data, and using a prediction model to make ultra-short-term predictions on the output of new energy on the source side, the change of the power flow on the grid side, the load demand on the load side, and the energy storage status on the storage side, and generating day-ahead prediction data and real-time prediction data; The grid operation data includes the renewable energy output data on the source side, the grid power flow data on the grid side, the load demand data on the load side, the energy storage status data on the storage side, as well as the meteorological data and electricity price data; The steps for making ultra-short-term forecasts and generating day-ahead forecast data and real-time forecast data are as follows: Based on LSTM model, ARIMA model, variational autoencoder VAE and Markov decision process MDP, ultra-short-term forecasting is performed to generate day-ahead forecast data and real-time forecast data. Long short-term memory network LSTM is used for nonlinear time series prediction to obtain the wind power and photovoltaic output forecast values: and ,in, For the moment Forecasted wind power output, For the moment Forecasted PV output, The autoregressive moving average model ARIMA is used to predict the power flow of the power grid. The prediction formula is: , in, For the moment The flow data, is the order of the autoregressive term, is the autoregression coefficient, for Current trend data, is the moving average term order, is the moving average coefficient, for The random error of time, is the random error at the current moment, Using variational autoencoder VAE for load demand modeling: , , in, is a latent variable, is a nonlinear activation function, is the encoder weight matrix, is the historical load data, is the encoder bias, Reconstruct data for the decoder, is the decoder weight matrix, is the decoder bias, The Markov decision process MDP is used for modeling: , , in, For the moment The energy storage state, For the moment The control actions taken, is the state transfer function, The energy storage state for the next moment, For instant rewards, is the reward function, Combined with day-ahead forecast With real-time prediction , the fusion formula is: , in, is the fused prediction data, is the weight coefficient, is the day-ahead forecast data, To predict data in real time;

[0025] Specifically, in step S1, LSTM, ARIMA, VAE and MDP are used for ultra-short-term forecasting, covering new energy output, power flow changes, load demand and energy storage status, to ensure the accuracy and real-time performance of the forecast data; LSTM is used for nonlinear modeling of new energy output to improve the forecast accuracy of wind and solar output, and the ARIMA model is used for time series modeling of power flow data to improve the predictability of power grid power flow changes. VAE is used for load demand forecasting to optimize load forecasting by learning historical load patterns, and MDP is used for energy storage status forecasting to establish a decision-making model for energy storage charging and discharging; in addition, by fusing the day-ahead forecast with the real-time forecast data, a weighted method is used to optimize the forecast results to make them more in line with the power grid operation requirements; Step S2, based on the forecast data of S1, establish a scheduling optimization model and calculate the day-ahead scheduling plan; The dispatch optimization model integrates the coordinated constraints of the source side, grid side, load side, and storage side, optimizes the wind and solar power output on the source side, the power flow control on the grid side, the demand response on the load side, and the charging and discharging strategy on the storage side, and uses reinforcement learning to calculate the day-ahead dispatch plan to reduce the output of conventional units on the source side during periods of high wind power output.

[0026] Based on the forecast data of S1, the scheduling optimization model is established and the steps to calculate the day-ahead scheduling plan are as follows: The objective function is defined as minimizing the grid operation cost, and the formula is: , in, is the scheduling optimization objective function, is the day-ahead scheduling period, is the output cost of the source-side unit, is the energy storage system cost, is the demand response cost, The output of the source side unit, is the energy storage charging and discharging power, Adjust power for load, Define constraints, including: 1) Grid power balance constraints: , in, is the number of grid nodes, For Node The power at 2) Energy storage charging and discharging constraints: , , in, In the energy storage state, For charging efficiency, is the charging power, is the discharge efficiency, is the discharge power, are the upper and lower limits of the system, 3) Demand response constraints: , , in, is the load regulation factor, is the reference load; Reinforcement learning optimizes scheduling strategies, , in, is the optimal scheduling strategy, is the current grid status, is an optional scheduling action, Q-value function for reinforcement learning;

[0027] Specifically, a power grid dispatch optimization model is established based on S1 forecast data to calculate the day-ahead dispatch plan and minimize the system operating cost. The optimization model is also subject to the grid flow balance, energy storage charging and discharging constraints, and load demand response constraints to ensure the feasibility and stability of the dispatch plan. In terms of optimization methods, reinforcement learning is introduced to calculate the day-ahead dispatch plan, so that the output of conventional units can be reduced during the high-output period of wind power, thereby increasing the consumption rate of new energy. Step S3, combining the day-ahead dispatch plan of S2 and the real-time forecast data of S1, adopts model predictive control MPC to perform dynamic optimization, including: new energy output on the source side, power flow regulation on the grid side, load adjustment on the load side, charging and discharging on the storage side, and giving priority to wind power for heating; In step S3, When wind power output is excessive, control the storage-side equipment to absorb excess power and reduce the output of conventional units; When wind power is insufficient, the storage-side equipment is controlled to release stored energy.

[0028] Combining the day-ahead dispatch plan of S2 and the real-time forecast data of S1, the steps of dynamic optimization using model predictive control MPC are as follows: Model predictive control (MPC) is used to perform dynamic optimization by combining the real-time forecast data of S1 and the day-ahead scheduling plan of S2, and the objective function is defined as: , in, To dynamically optimize the objective function, is the MPC prediction time domain length, is the output cost of the source-side unit, is the energy storage cost, is the demand response cost, is the penalty term for the day-ahead scheduling deviation, is the deviation between real-time scheduling and day-ahead scheduling, is the real-time power demand predicted in step S1, is the day-ahead scheduling plan calculated in step S2, The state space model is: , , in, is a state variable, including the output of new energy , flow distribution , load demand , energy storage state , is the system state transfer matrix, is the control input matrix, The control variables include wind power regulation, energy storage charging and discharging power, and load adjustment. is the system noise, is the observed variable, is the observation matrix, To measure noise.

[0029] The steps of dynamic optimization using model predictive control MPC also include: Define constraints, including: 1) Grid power balance constraints: , in, is the number of grid nodes, For Node The power at The output of the source side unit, is the charging and discharging power of the energy storage system, Adjust power for load, 2) Energy storage dynamic control constraints: , , in, is the remaining power of the energy storage system, For charging efficiency, is the charging power, is the discharge efficiency, is the discharge power, and are the minimum and maximum capacities of the energy storage system, 3) Wind power consumption strategy: like ,but , like ,but , in, Indicates the power that the energy storage system needs to absorb when there is excess wind power. Indicates the power that the energy storage system needs to supplement when wind power is insufficient. To contribute to wind power, is the load demand, MPC rolling optimization solution, the formula is: , , in, To predict the control time domain length, is the control input set, Indicates that in the rolling optimization time domain Time index within Indicates time The objective function value of is the optimal control input at the current moment, Update the value for the next state; Specifically, the S2 day-ahead dispatch plan and the S1 real-time forecast data are combined here, and the model predictive control MPC is used for dynamic optimization. MPC continuously adjusts the new energy output, flow regulation, load adjustment and energy storage charging and discharging strategies through rolling optimization when the real-time forecast data changes. The goal is to minimize the day-ahead dispatch deviation and improve the utilization rate of wind power. The optimization model is affected by the power balance constraints of the power grid, the dynamic control constraints of energy storage and the wind power consumption strategy constraints, which improves the stability, economy and flexibility of the dispatch plan. This step adopts state space modeling and uses the prediction rolling window to achieve dynamic adjustment; Step S4, based on the optimization result of S3, using reinforcement learning to dynamically respond to the load demand on the load side; Load-side dynamic demand response includes: When electricity prices are low, guide loads to increase electricity consumption; When electricity prices are at their peak, reduce the generation of conventional units and give priority to wind power supply; Based on the optimization results of S3, the steps of using reinforcement learning to dynamically respond to load demand on the load side are as follows: Define the objective function, the goal is to maximize the long-term benefits of the system, the optimal demand response strategy To maximize the cumulative reward, the formula is: , in, represents the expected value, is the discount factor, For the moment Instant rewards, is the demand response cycle, Reinforcement learning uses the Qlearning method to update the Q value function. The update formula is: , in, is the optimal Q value at the next moment, is the state action value, is the learning rate, is the discount factor, Specifically, in step S4, based on the dynamic optimization results of step S3, reinforcement learning is used to perform dynamic demand response on the load side, so that user loads can increase electricity consumption when electricity prices are low, reduce the output of conventional units when electricity prices are peak, give priority to wind power supply, maximize long-term benefits, and minimize grid operation costs.

[0030] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A power grid operation control method for realizing source-grid-load-storage coordination, characterized in that: include, Step S1, collecting grid operation data, and using a prediction model to make ultra-short-term predictions on the output of new energy on the source side, the change of the power flow on the grid side, the load demand on the load side, and the energy storage status on the storage side, and generating day-ahead prediction data and real-time prediction data; Step S2, based on the forecast data of S1, establish a scheduling optimization model and calculate the day-ahead scheduling plan; Step S3, combining the day-ahead dispatch plan of S2 and the real-time forecast data of S1, adopts model predictive control MPC to perform dynamic optimization, including: new energy output on the source side, power flow regulation on the grid side, load adjustment on the load side, charging and discharging on the storage side, and giving priority to wind power for heating; Step S4, based on the optimization result of S3, reinforcement learning is used to dynamically respond to the load demand on the load side.

2. A method for controlling power grid operation to achieve source-grid-load-storage coordination as claimed in claim 1, characterized in that: The power grid operation data includes source-side renewable energy output data, grid-side power flow data, load-side load demand data, storage-side energy storage status data, as well as meteorological data and electricity price data.

3. A method for controlling power grid operation to achieve source-grid-load-storage coordination as claimed in claim 2, characterized in that: The steps of performing ultra-short-term forecasting and generating day-ahead forecast data and real-time forecast data are as follows: Based on LSTM model, ARIMA model, variational autoencoder VAE and Markov decision process MDP, ultra-short-term forecasting is performed to generate day-ahead forecast data and real-time forecast data. Long short-term memory network LSTM is used for nonlinear time series prediction to obtain the wind power and photovoltaic output forecast values: and ,in, For the moment Forecasted wind power output, For the moment Forecasted PV output, The autoregressive moving average model ARIMA is used to predict the power flow of the power grid. The prediction formula is: ， in, For the moment The flow data, is the order of the autoregressive term, is the autoregression coefficient, for Current trend data, is the moving average term order, is the moving average coefficient, for The random error of time, is the random error at the current moment, Using variational autoencoder VAE for load demand modeling: ， ， in, is a latent variable, is a nonlinear activation function, is the encoder weight matrix, is the historical load data, is the encoder bias, Reconstruct data for the decoder, is the decoder weight matrix, is the decoder bias, The Markov decision process MDP is used for modeling: ， ， in, For the moment The energy storage state, For the moment The control actions taken, is the state transfer function, The energy storage state for the next moment, For instant rewards, is the reward function, Combined with day-ahead forecast With real-time prediction , the fusion formula is: ， in, is the fused prediction data, is the weight coefficient, is the day-ahead forecast data, For real-time prediction data.

4. A method for controlling power grid operation to achieve source-grid-load-storage coordination as claimed in claim 3, characterized in that: The dispatch optimization model integrates the coordinated constraints of the source side, grid side, load side, and storage side, optimizes the wind and solar power output on the source side, the power flow regulation on the grid side, the demand response on the load side, and the charging and discharging strategy on the storage side, and uses reinforcement learning to calculate the day-ahead dispatch plan to reduce the output of conventional units on the source side during periods of high wind power output.

5. A method for controlling power grid operation to achieve source-grid-load-storage coordination as claimed in claim 4, characterized in that: The steps of establishing a scheduling optimization model based on the forecast data of S1 and calculating the day-ahead scheduling plan are as follows: The objective function is defined as minimizing the grid operation cost, and the formula is: ， in, is the scheduling optimization objective function, is the day-ahead scheduling period, is the output cost of the source-side unit, is the energy storage system cost, is the demand response cost, The output of the source side unit, is the energy storage charging and discharging power, Adjust power for load, Define constraints, including: 1) Grid power balance constraints: ， in, is the number of grid nodes, For Node The power at 2) Energy storage charging and discharging constraints: ， ， in, In the energy storage state, For charging efficiency, is the charging power, is the discharge efficiency, is the discharge power, are the upper and lower limits of the system, 3) Demand response constraints: ， ， in, is the load regulation factor, is the reference load; Reinforcement learning optimizes scheduling strategies, ， in, is the optimal scheduling strategy, is the current grid status, is an optional scheduling action, is the Q-value function for reinforcement learning.

6. A method for controlling power grid operation to achieve source-grid-load-storage coordination as claimed in claim 5, characterized in that: In step S3, When wind power output is excessive, control the storage-side equipment to absorb excess power and reduce the output of conventional units; When wind power is insufficient, the storage-side equipment is controlled to release stored energy.

7. A method for controlling power grid operation to achieve source-grid-load-storage coordination as claimed in claim 6, characterized in that: The steps of combining the day-ahead dispatch plan of S2 and the real-time forecast data of S1 and using the model predictive control MPC for dynamic optimization are as follows: Model predictive control (MPC) is used to perform dynamic optimization by combining the real-time forecast data of S1 and the day-ahead scheduling plan of S2, and the objective function is defined as: ， in, To dynamically optimize the objective function, is the MPC prediction time domain length, is the output cost of the source-side unit, is the energy storage cost, is the demand response cost, is the penalty term for the day-ahead scheduling deviation, is the deviation between real-time scheduling and day-ahead scheduling, is the real-time power demand predicted in step S1, is the day-ahead scheduling plan calculated in step S2, The state space model is: ， ， in, is a state variable, including the output of new energy , flow distribution , load demand , energy storage state , is the system state transfer matrix, is the control input matrix, The control variables include wind power regulation, energy storage charging and discharging power, and load adjustment. is the system noise, is the observed variable, is the observation matrix, To measure noise.

8. A method for controlling power grid operation to achieve source-grid-load-storage coordination as claimed in claim 7, characterized in that: The step of using model predictive control MPC for dynamic optimization also includes: Define constraints, including: 1) Grid power balance constraints: ， in, is the number of grid nodes, For Node The power at The output of the source side unit, is the charging and discharging power of the energy storage system, Adjust power for load, 2) Energy storage dynamic control constraints: ， ， in, is the remaining power of the energy storage system, For charging efficiency, is the charging power, is the discharge efficiency, is the discharge power, and are the minimum and maximum capacities of the energy storage system, 3) Wind power consumption strategy: like ,but , like ,but , in, Indicates the power that the energy storage system needs to absorb when there is excess wind power. Indicates the power that the energy storage system needs to supplement when wind power is insufficient. To contribute to wind power, is the load demand, MPC rolling optimization solution, the formula is: ， ， in, To predict the control time domain length, is the control input set, Indicates that in the rolling optimization time domain Time index within Indicates time The objective function value of is the optimal control input at the current moment, Update the value for the next state.

9. A method for controlling power grid operation to achieve source-grid-load-storage coordination as claimed in claim 8, characterized in that: The load-side dynamic demand response includes: When electricity prices are low, guide loads to increase electricity consumption; When electricity prices are at their peak, reduce the power generation of conventional units and give priority to the use of wind power.

10. A method for controlling power grid operation to achieve source-grid-load-storage coordination according to claim 9, characterized in that: The steps of using reinforcement learning to dynamically respond to the load demand on the load side based on the optimization result of S3 are as follows: Define the objective function, the goal is to maximize the long-term benefits of the system, the optimal demand response strategy To maximize the cumulative reward, the formula is: ， in, represents the expected value, is the discount factor, For the moment Instant rewards, is the demand response cycle, Reinforcement learning uses the Qlearning method to update the Q value function. The update formula is: ， in, is the optimal Q value at the next moment, is the state action value, is the learning rate, is the discount factor.

Images