A micro-grid source network load storage optimal configuration and operation control strategy

Abstract

This invention discloses an optimal configuration and operation control strategy for microgrid power generation, grid, load, and storage, comprising the following analytical steps: S1, constructing a microgrid operation control model, including the control objective function and constraints; S2, establishing a state prediction model for distributed renewable energy output and load demand based on a Long Short-Time Memory (LSTM) network; S3, establishing a Markov decision process for microgrid operation control based on the prediction results of the state prediction model; S4, training and solving the Markov decision process for microgrid operation control using a deep reinforcement learning algorithm to obtain the optimal solution and form the optimal operation control strategy. This invention can effectively reduce the operating cost of microgrids and improve the economy and security of microgrid operation.

Description

Technical Field

[0001] This invention belongs to the field of microgrid operation and control technology, specifically relating to a microgrid source-grid-load-storage optimization configuration and operation control strategy. Background Technology

[0002] Microgrids are small-scale power generation and distribution systems composed of distributed power sources, energy storage devices, and loads. They can promote the large-scale integration of distributed power sources and renewable energy sources. Distributed generation can make full use of idle space and effectively reduce transmission and distribution costs. Therefore, the proportion of distributed power sources in the power system continues to increase.

[0003] Because new energy power generation is characterized by randomness, volatility, and intermittency, it is difficult to predict and has poor stability, requiring a high degree of flexibility from the power system. Therefore, it is necessary to adopt optimized scheduling methods to coordinate the power generation, grid, load, and storage, and build a new type of power system with a high degree of integration of power generation, grid, load, and storage. This will meet the internal power supply needs of the system while saving power supply costs and reducing economic costs.

[0004] Therefore, in order to solve the above problems, it is necessary to develop a microgrid source-grid-load-storage optimization configuration and operation control strategy. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a microgrid source-grid-load-storage optimization configuration and operation control strategy, which can effectively reduce the operating cost of microgrids and improve the economy and security of microgrid operation.

[0006] The objective of this invention is achieved as follows: a microgrid source-grid-load-storage optimization configuration and operation control strategy, comprising the following analysis steps:

[0007] S1. Construct a microgrid operation control model, including the control objective function and constraints;

[0008] S2. Establish a state prediction model for distributed renewable energy output and load demand based on the Long Short-Time Memory (LSTM) network.

[0009] S3. Based on the prediction results of the state prediction model, establish a Markov decision process for microgrid operation control;

[0010] S4. A deep reinforcement learning algorithm is used to train and solve the Markov decision process for microgrid operation control, so as to obtain the optimal solution and form the optimal operation control strategy.

[0011] Furthermore, in step S1, the control objective function is to minimize operating costs and maximize the distributed renewable energy absorption rate: ,in, , For the target weight factor, satisfying , To minimize operating costs, To maximize the absorption rate of distributed renewable energy.

[0012] Furthermore, the constraints in step S1 include power balance constraints, energy storage constraints, standby unit constraints, and new energy power generation constraints.

[0013] Furthermore, in step S2, a joint feature vector is constructed using distributed renewable energy output, load, and energy storage status as the multi-dimensional time series input to the state prediction model to predict the changing trends of distributed renewable energy output and load demand.

[0014] Furthermore, in step S2, a state prediction model is constructed using a Long Short-Term Memory (LSTM) network. The LSTM network specifically includes a forget gate, an input gate, and an output gate, with the three inputs representing the cell state. Hidden state and Time-input vector The two outputs are the cell state. and hidden state ,and At the same time The output at time step 1, the activation function uses an output of 0-1. function And the hyperbolic tangent function with output -1-1 .

[0015] Furthermore, the establishment of the Markov decision process for microgrid operation control in step S3 includes establishing the state space, action space, reward function, and strategy. .

[0016] Furthermore, in step S4, the dual-delay deep deterministic strategy gradient TD3 algorithm is specifically used for training and solving.

[0017] Furthermore, the training and solving process in step S4 specifically includes the following steps: ① Initialize the environment and network parameters, and set the maximum number of iterations. Experience pool capacity Number of time periods within the operating cycle ② Randomly select load demand, distributed renewable energy output, and electricity price data for a single operating cycle from the training dataset; ③ Obtain the current time period state through the established Markov decision process. And select and adjust actions ④ The state transition probabilities of load and distributed renewable energy are predicted by the state prediction model, and the state transition probability of energy storage charge is calculated by the physical model. The output of load and distributed renewable energy, as well as the state of energy storage charge, are then updated to obtain the state for the next time period. ⑤ Calculate instant rewards ⑥ Determine the current training round number Is it less than the experience pool capacity? If yes, proceed to the next step; otherwise, delete the historical samples and proceed to the next step; ⑦ Transfer the samples to... ⑧ Store the data in the experience replay pool; ⑨ Randomly select a certain number of samples from the experience replay pool to update the network parameters; ⑨ Determine the current time period. Is it less than the number of time periods within the operating cycle? If so, then let , Then return to step ③ to continue; otherwise, proceed to the next step; ⑩ Determine if the maximum number of iterations has been reached. If yes, end the iteration; otherwise, increment the iteration count. Then return to step ③ to continue.

[0018] Furthermore, the network parameter update in step S4 is specifically implemented in the following way: ① The actor network updates the network parameters through policy gradients, as follows: In the formula, For actor network parameters, Let the state transition probability distribution be given by a certain strategy. For a certain random strategy, The fitting function for the critic network evaluation. These are the parameters of the critic network. ① The fitting function for the state-action mapping; ② The critic network uses the state-action value function to evaluate the state... Choose the action The evaluation is performed by updating the network parameters by minimizing the loss function, as follows: In the formula, for Value estimation.

[0019] The beneficial effects of this invention, achieved by employing the above technical solutions, are as follows: By constructing a microgrid operation control model, using a Long Short-Time Memory (LSTM) network to establish a state prediction model for distributed renewable energy output and load demand, predicting key variables with high uncertainty in the system based on historical data, and using a deep reinforcement learning algorithm to train and solve the Markov decision process for microgrid operation control, the optimal solution is obtained and the optimal operation control strategy is formed. This effectively improves the system's prediction accuracy of the changing trends of key variables, enhances the system's ability to perceive environmental uncertainties, minimizes the operating cost of the microgrid, maximizes the absorption of distributed energy, and improves the economic efficiency of microgrid operation and the utilization efficiency of renewable energy. Attached Figure Description

[0020] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0022] like Figure 1 As shown, a microgrid source-grid-load-storage optimization configuration and operation control strategy includes the following analysis steps:

[0023] S1. Construct a microgrid operation control model, including the control objective function and constraints.

[0024] Preferably, in step S1, the control objective function is to minimize operating costs and maximize the distributed renewable energy absorption rate: ;in, , For the target weight factor, satisfying Among them, minimizing operating costs Represented as: In the formula, for The cost of electricity transactions between a microgrid and its upstream grid during a given period. for Time-of-use electricity pricing for Power of the grid interconnection line at the point of common connection between the microgrid and the upper-level grid during a given period; for The cost of standby generator generation in a microgrid during a given period. This refers to the current price of diesel fuel. For effective fuel consumption rate, For generator efficiency, This refers to the rated power of a single standby diesel engine. for The number of diesel generators in operation during a given time period; for Depreciation cost of time-of-use energy storage charging and discharging. This is the depreciation cost factor for energy storage. For electrical energy storage capacity, , They are respectively Time and The state of charge at any given time; among which, maximizing the absorption rate of distributed renewable energy. Represented as: , This represents the total number of new energy generating units. , The first Taiwan's distributed renewable energy units during the period The actual active power output and maximum active power output within the system.

[0025] Preferably, the constraints in step S1 include power balance constraints, energy storage constraints, standby unit constraints, and new energy power generation constraints.

[0026] The power balance constraint is expressed as follows: In the formula, , They are respectively Solar and wind power generation capacity at all times for Power of the grid interconnection line at any time for Total standby generating capacity of the generator units at all times. for Energy storage charging and discharging power at all times, for The power consumption at all times; wherein, the energy storage constraints include energy storage charge and discharge power constraints, state of charge constraints, and energy storage reserve power constraints, specifically, the energy storage charge and discharge power constraints are expressed as follows: , , They represent The charging and discharging power of the energy storage system at all times. , These represent the upper limits of the charging and discharging power of the energy storage system, respectively; the state of charge constraint is expressed as: , express The state of charge of the energy storage system at all times. , These represent the lower and upper threshold values ​​of the state of charge of the energy storage system, respectively; the energy storage backup power constraint is expressed as: , This refers to the critical state of charge for energy storage emergency backup; wherein, the constraints of the backup unit are expressed as: , , They are respectively The cumulative power-on and power-off time at any given moment. , These are the minimum cumulative power-on and power-off times, respectively. , They are respectively , The standby generating unit's operating status is constantly monitored, with 1 indicating startup and 0 indicating shutdown; the aforementioned new energy power generation constraint is expressed as follows: , , They are respectively Solar and wind power generation capacity at all times , These are the maximum instantaneous power output limits for photovoltaic and wind power generation allowed by the equipment.

[0027] S2. Establish a state prediction model for distributed renewable energy output and load demand based on Long Short-Term Memory (LSTM) network.

[0028] Preferably, in step S2, a joint feature vector is constructed using distributed renewable energy output, load, and energy storage status as the multi-dimensional time series input of the state prediction model to predict the changing trends of distributed renewable energy output and load demand.

[0029] Preferably, in step S2, a Long Short-Term Memory (LSTM) network is used to construct a state prediction model. The LSTM network specifically includes a forget gate, an input gate, and an output gate, with the three inputs being the cell state. Hidden state and Time-input vector The two outputs are the cell state. and hidden state ,and At the same time The output at time step 1, the activation function uses an output of 0-1. function And the hyperbolic tangent function with output -1-1 Specifically, the Gate of Oblivion First, read the hidden state. and Time-input vector Then through Activation function To each in the cellular state The output should be a number between 0 and 1, where 1 means "keep completely" and 0 means "discard completely". Specifically: In the formula, Forgetting matrix, The constant vector for the forget gate; the input gate. This will determine the addition of new information to the cell state. The quantity in, and its specific process are expressed as follows: In the formula, For the input matrix, This is the constant vector of the input gate. For candidate state variables, The candidate state variable matrix, A constant vector of candidate states; output gate Decide The output at time t is specifically represented as follows: In the formula, for Output coefficients at time 10:00 For the output matrix, To output their constant vectors, This is the final output.

[0030] S3. Based on the prediction results of the state prediction model, establish a Markov decision process for microgrid operation control.

[0031] Preferably, the establishment of the microgrid operation control Markov decision process in step S3 includes establishing the state space, action space, reward function, and strategy. The state space is represented as: , , , , , They represent Real-time load power, photovoltaic power generation power, wind power generation power, energy storage state of charge, and real-time electricity price; the action space is represented as: , For load regulation, , These are the absorption capacities of solar power and wind power, respectively. The energy storage charging and discharging control quantity; the reward function is expressed as: .

[0032] S4. A deep reinforcement learning algorithm is used to train and solve the Markov decision process for microgrid operation control, so as to obtain the optimal solution and form the optimal operation control strategy.

[0033] Preferably, step S4 specifically employs the dual-delay deep deterministic strategy gradient TD3 algorithm for training and solving, which includes the following steps: ① Initialize the environment and network parameters, and set the maximum number of iterations. Experience pool capacity Number of time periods within the operating cycle ② Randomly select load demand, distributed renewable energy output, and electricity price data for a single operating cycle from the training dataset; ③ Obtain the current time period state through the established Markov decision process. And select and adjust actions ④ The state transition probabilities of load and distributed renewable energy are predicted by the state prediction model, and the state transition probability of energy storage charge is calculated by the physical model. The output of load and distributed renewable energy, as well as the state of energy storage charge, are then updated to obtain the state for the next time period. ⑤ Calculate instant rewards ⑥ Determine the current training round number Is it less than the experience pool capacity? If yes, proceed to the next step; otherwise, delete the historical samples and proceed to the next step; ⑦ Transfer the samples to... ⑧ Store the data in the experience replay pool; ⑨ Randomly select a certain number of samples from the experience replay pool to update the network parameters; ⑨ Determine the current time period. Is it less than the number of time periods within the operating cycle? If so, then let , Then return to step ③ to continue; otherwise, proceed to the next step; ⑩ Determine if the maximum number of iterations has been reached. If yes, then end the iteration; otherwise, set the iteration count to zero. Then return to step ③ to continue.

[0034] Preferably, the network parameter update in step S4 is specifically implemented in the following way: ① The actor network updates the network parameters through policy gradients, as follows: In the formula, For actor network parameters, Let the state transition probability distribution be given by a certain strategy. For a certain random strategy, The fitting function for the critic network evaluation. These are the parameters of the critic network. ① The fitting function for the state-action mapping; ② The critic network uses the state-action value function to evaluate the state... Choose the action The evaluation is performed by updating the network parameters by minimizing the loss function, as follows: In the formula, for Value estimation.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A microgrid source-grid-load-storage optimization configuration and operation control strategy, characterized in that: The analysis includes the following steps: S1. Construct a microgrid operation control model, including the control objective function and constraints; S2. Establish a state prediction model for distributed renewable energy output and load demand based on the Long Short-Time Memory (LSTM) network. S3. Based on the prediction results of the state prediction model, establish a Markov decision process for microgrid operation control; S4. A deep reinforcement learning algorithm is used to train and solve the Markov decision process for microgrid operation control, so as to obtain the optimal solution and form the optimal operation control strategy.

2. The microgrid source-grid-load-storage optimization configuration and operation control strategy according to claim 1, characterized in that: In step S1, the control objective function is to minimize operating costs and maximize the distributed renewable energy absorption rate. ,in, , For the target weight factor, satisfying , To minimize operating costs, To maximize the absorption rate of distributed renewable energy.

3. The microgrid source-grid-load-storage optimization configuration and operation control strategy according to claim 1, characterized in that: The constraints in step S1 include power balance constraints, energy storage constraints, standby unit constraints, and new energy power generation constraints.

4. The microgrid source-grid-load-storage optimization configuration and operation control strategy according to claim 1, characterized in that: In step S2, a joint feature vector is constructed using distributed renewable energy output, load, and energy storage status as the multidimensional time series input to the state prediction model to predict the changing trends of distributed renewable energy output and load demand.

5. The microgrid source-grid-load-storage optimization configuration and operation control strategy according to claim 1, characterized in that: In step S2, a state prediction model is constructed using a Long Short-Term Memory (LSTM) network. The LSTM network specifically includes a forget gate, an input gate, and an output gate, with the three inputs representing the cell state. Hidden state and Time-input vector The two outputs are the cell state. and hidden state ,and At the same time The output at time step 1, the activation function uses an output of 0-1. function And the hyperbolic tangent function with output -1-1 .

6. The microgrid source-grid-load-storage optimization configuration and operation control strategy according to claim 1, characterized in that: The establishment of the Markov decision process for microgrid operation control in step S3 includes establishing the state space, action space, reward function, and strategy. .

7. The microgrid source-grid-load-storage optimization configuration and operation control strategy according to claim 1, characterized in that: In step S4, the dual-delay deep deterministic strategy gradient TD3 algorithm is specifically used for training and solving.

8. The microgrid source-grid-load-storage optimization configuration and operation control strategy according to claim 7, characterized in that: The training and solving process in step S4 specifically includes the following steps: ① Initialize the environment and network parameters, and set the maximum number of iterations. Experience pool capacity Number of time periods within the operating cycle ② Randomly select load demand, distributed renewable energy output, and electricity price data for a single operating cycle from the training dataset; ③ Obtain the current time period state through the established Markov decision process. And select and adjust actions ④ The state transition probabilities of load and distributed renewable energy are predicted by the state prediction model, and the state transition probability of energy storage charge is calculated by the physical model. The output of load and distributed renewable energy, as well as the state of energy storage charge, are then updated to obtain the state for the next time period. ⑤ Calculate instant rewards ⑥ Determine the current training round number Is it less than the experience pool capacity? If yes, proceed to the next step; otherwise, delete the historical samples and proceed to the next step; ⑦ Transfer the samples to... ⑧ Store the data in the experience replay pool; ⑨ Randomly select a certain number of samples from the experience replay pool to update the network parameters; ⑨ Determine the current time period. Is it less than the number of time periods within the operating cycle? If so, then let , Then return to step ③ to continue; otherwise, proceed to the next step; ⑩ Determine if the maximum number of iterations has been reached. If yes, then end the iteration; otherwise, set the iteration count to zero. Then return to step ③ to continue.

9. The microgrid source-grid-load-storage optimization configuration and operation control strategy according to claim 8, characterized in that: The network parameter update in step S4 is specifically implemented in the following way: ① The actor network updates the network parameters through policy gradients, as follows: In the formula, For actor network parameters, Let the state transition probability distribution be given by a certain strategy. For a certain random strategy, The fitting function for the critic network evaluation. These are the parameters of the critic network. ① The fitting function for the state-action mapping; ② The critic network uses the state-action value function to evaluate the state... Choose the action The evaluation is performed by updating the network parameters by minimizing the loss function, as follows: In the formula, for Value estimation.

Images