CN111680815A - A Hierarchical Optimization Reconstruction Method of Microgrid Based on BP Neural Network - Google Patents

Abstract

The invention discloses a method for hierarchical optimization and reconstruction of a microgrid based on a BP neural network, which includes S1: establishing a microgrid optimization and reconstruction model, and constructing an objective function and constraint conditions of the microgrid optimization and reconstruction; S2: establishing a BP neural network model , and train and test the neural network; S3: Using the idea of hierarchical optimization, in the microgrid reconstruction, it is divided into the first and second level optimization process according to whether the power flow calculation is performed, and the training is used in the second level optimization process. The BP neural network replaces the power flow calculation; S4: Introduce the comprehensive evaluation method, construct the comprehensive evaluation function, comprehensively evaluate the output results of the BP network, and select the optimal reconstruction scheme. The invention has the advantages that the reconstruction process realizes offline training of the BP neural network, online input of switch states, direct output of network losses, balancing of node power and voltage results, saving calculation time and significantly improving microgrid reconstruction efficiency.

Description

A Hierarchical Optimization Reconstruction Method of Microgrid Based on BP Neural Network

technical field

The invention relates to the technical field of microgrid optimization and reconstruction, in particular to a microgrid optimization and reconstruction method based on BP neural network.

Background technique

In recent years, with the development of new energy power generation, microgrid has attracted more and more attention. Microgrid is a small power generation and distribution system that organically integrates distributed power sources, loads, energy storage devices, converters, and monitoring and protection devices. It has two modes of islanding and grid-connected operation. The microgrid may fail during operation. At this time, it is necessary to adjust the switching state of the microgrid. After the microgrid is reconstructed, the load can be balanced, the network loss can be effectively reduced, and the power quality can be improved, thereby improving its reliability and safety. sex. With the development of network reconstruction technology, scholars have begun to study the accuracy and speed of network reconstruction while paying attention to network reconstruction algorithms.

In the existing microgrid reconstruction algorithm, as the number of network nodes increases, the number of variables also increases, which increases the amount of computation and computation time. In addition, since some intelligent algorithms can only find the local optimal solution and cannot find the global optimal solution, it cannot be guaranteed whether the final reconstruction result reaches the optimal state. In the process of microgrid reconstruction, the idea of hierarchical optimization is introduced, and the BP neural network is used to replace the power flow calculation, so as to ensure the reconstruction accuracy, reduce the pressure of power flow calculation during the microgrid reconstruction, and greatly improve the reconstruction efficiency.

SUMMARY OF THE INVENTION

In view of the above-mentioned defects of the prior art, the purpose of the present invention is to provide a microgrid optimization and reconstruction method based on BP neural network. network, and introduces the idea of hierarchical optimization, online calculation of network loss, balance of node power offset and voltage deviation, etc., saving calculation time and significantly improving the efficiency of microgrid reconstruction.

In order to solve the above-mentioned technical problems, the present invention is realized by the following technical solutions: a method for hierarchical optimization and reconstruction of microgrid based on BP neural network, the method comprises the following steps:

S1: Establish a microgrid optimization and reconstruction model, and construct an objective function and constraints for microgrid optimization and reconstruction;

S2: Establish a BP neural network model, and train and test the neural network;

S3: Using the idea of hierarchical optimization, in the reconstruction of the microgrid, the optimization process that does not involve power flow calculation is placed in the first-level processing, and the optimization process involving power flow calculation is treated as the second-level optimization process; only when the first-level optimization process is satisfied The second-level optimization process is performed only for the solution of the optimization conditions, and the trained BP neural network is used to replace the power flow calculation in the second-level optimization process;

S4: Introduce the comprehensive evaluation method, construct a comprehensive evaluation function, comprehensively evaluate the output results of the BP network, and select the optimal reconstruction scheme.

Preferably, under the constraints of step S1, the objective function of the microgrid optimization and reconstruction is:

S11: Microgrid reconstruction needs to meet the goal of minimum load shedding, and its objective function is:

In the formula: i∈Ω, Ω is the set of nodes to remove the load after reconstruction; S _i represents the load corresponding to node i;

The stable operation of the microgrid after reconstruction must meet the following constraints:

1) Balanced node power constraints:

P _tmin ≤P _t ≤P _tmax

In the formula: P _t is the adjustable active power of the balance node t; P _tmax is the upper limit of the adjustable active power of the node t; P _tmin is the lower limit of the adjustable active power of the node t;

2) Branch power constraints:

P _Bj ≤ P _Bjmax

In the formula: P _Gx is the generated power of the micro-power source x in the micro-grid, X is the number of micro-power sources retained after reconstruction; P _Li is the load active power retained by node i after the micro-grid reconstruction, and N is the retained power after reconstruction. number of nodes;

3) Power balance constraints:

In the formula: P _Gx is the generated power of the micro-power source x in the micro-grid, X is the number of micro-power sources retained after reconstruction; P _Li is the load active power retained by node i after the micro-grid reconstruction, and N is the retained power after reconstruction. number of nodes;

4) Micro power generation power constraints:

P _Gmin ≤P _G ≤P _Gmax

In the formula: P _G is the total generated power of the micro-power after the micro-grid reconstruction; P _Gmin is the lower limit of the generated power in the micro-grid; P _Gmax is the upper limit of the generated power in the micro-grid;

5) Node voltage constraints:

U _imin ≤U _i ≤U _imax

In the formula: U _i is the voltage of node i; U _imin is the lower limit of the voltage of node i; U _imax is the upper limit of the voltage of node i.

Preferably, the construction of the BP neural network in the step S2 is as follows:

S21: In order to improve the training accuracy, the neural network model constructed by the present invention contains two hidden layers, each

The number of nodes in a hidden layer is determined according to the following formula:

In the formula, h is the number of hidden layer nodes, z and v are the number of input layer and output layer nodes, respectively, and c is an adjustment constant between 1 and 10.

Preferably, the training and testing process of the BP neural network in the step S2 is as follows:

S22: Use the Matpower toolkit to calculate the power flow of the microgrid corresponding to different switching states, including the balance node power P _t , the network loss L, the maximum node voltage U _max and the minimum voltage U _min ; collect the above data;

S23: In the collected data, arbitrarily select 80% of the data as the training set, and use the neural network toolbox in Matlab for training;

S24: For the BP neural network obtained by training, the remaining 20% are imported into the BP neural network as the test set data, and the output results are compared with the data in the test set to determine that the BP neural network meets the requirements.

Preferably, the process of applying the hierarchical optimization idea to the microgrid reconstruction in the step S3 is as follows:

S31: In the first-level optimization process, deal with the objective function and constraints that do not involve any power flow calculation; consider the balance between the power generation and the load, perform integer programming, and obtain a load switch state combination that satisfies the balance condition, and follow the The switch combination solution set D is obtained by sorting the power supply load shedding amount in ascending order;

S32: In the second-level optimization process, the objective functions and constraints involved in the power flow calculation are processed, and the switch combination solution set D obtained by the first-level optimization is put into the BP neural network one by one for prediction, and the prediction result is obtained;

S33: For the load shedding amount, determine whether the prediction result satisfies the relevant constraints of the stable operation of the microgrid; if only one switch combination satisfies, the switch combination is the optimal solution for the microgrid reconstruction; on the contrary, if there are multiple groups of switches If the output result of the switch combination satisfies the constraints, the optimal solution for microgrid reconstruction is selected according to the comprehensive evaluation method; if all combinations are not satisfied, the next load shedding amount is selected and the above steps are repeated until the optimal solution is found.

Preferably, the comprehensive evaluation method in the step S4 is as follows:

S41: If the output results of multiple switch combinations for the same load shedding amount satisfy the constraints, multiple sets of corresponding network loss, voltage deviation and power deviation outputs will be obtained; set the reference values of network loss, voltage deviation and power deviation, and set the The output results of each group of switches are normalized, and the corresponding weight coefficient k _i is selected according to the preference of the decision maker, and the switch combination with the smallest comprehensive evaluation function is selected as the optimal solution by using the comprehensive evaluation function.

Preferably, the comprehensive evaluation function in the step S4 is constructed as follows:

S42: Microgrid reconstruction, when the load shedding amount is the same and the results meet the constraints, it needs to meet the

The goal of the minimum comprehensive evaluation function value is as follows:

分别为网络损耗、电压偏差和平衡节点功率偏差的归一化处理值；In the formula: k ₁ , k ₂ , k ₃ are the weight coefficients of network loss, voltage deviation and balance node power deviation, and 0<k ₁ , k ₂ , k ₃ <1, k ₁ +k ₂ +k ₃ =1 ;

are the normalized processing values of network loss, voltage deviation and balance node power deviation, respectively;

S43: Before carrying out the comprehensive evaluation, the output result should be normalized first, and its calculation

Methods as below:

In the formula: L is the system network loss value, L ^* is the set network loss reference value; ΔU is the voltage deviation, which refers to the difference between the highest voltage and the lowest voltage of the node in the entire system, U ^* is the system reference voltage; ΔP _t is Power deviation refers to the deviation between the power of the balance node and the set reference power, P _t ^* is the reference power of the balance node refers to the average value of the upper and lower limits of the power of the balance node; among them,

In the formula: P _Bj , Q _Bj , R _Bj are the active power, reactive power and resistance corresponding to branch j, respectively, where Q _Bj is obtained according to the load power factor and active power; U _Bj is the end of branch j Voltage; r _j is the branch state after the corresponding reconstruction, 1 means connected, 0 means disconnected; J is the number of branches contained in the system.

Preferably, in the step S3, the exception handling process is as follows:

S34: If the output results of all switch combinations in the switch state set D do not meet the constraints, it indicates that the fault reconstruction is abnormal, and the exception processing mechanism is entered; in all switch combinations that do not meet the constraints, the node voltage of each switch combination is calculated. Deviation, the balance node power exceeds the limit, and the comprehensive limit Y calculation is performed, as shown below:

In the formula, w ₁ , w ₂ are weight coefficients and 0<w ₁ , w ₂ <1, w ₁ +w ₂ =1; i is the minimum voltage node, dU _imin is the absolute value of the lower limit of the minimum voltage, U _imin is the lower limit value of the node voltage; j is the maximum voltage node, dU _jmax is the absolute value of the upper limit of the maximum voltage, U _jmax is the upper limit value of the node voltage; t is the balance node, dP _tmin is the absolute value of the lower limit power of the balance node, P _tmin is the lower limit of the power of the balance node, dP _tmax is the absolute value of the over-limit power on the balance node, and P _tmax is the upper limit of the power of the balance node; the switch combination with the smallest comprehensive over-limit Y is selected as the optimal reconstruction output solution.

Compared with the prior art, the advantages of the present invention are:

(1) By citing the idea of hierarchical optimization, the load flow calculation pressure during microgrid reconstruction can be reduced.

(2) Using the BP neural network to realize offline training, input the switch state online, and output the result directly, without the need for network reconstruction power flow calculation, saving computing time.

Description of drawings

Fig. 1 is a schematic flowchart of a hierarchical optimization and reconstruction method of microgrid based on BP neural network.

Figure 2 is a schematic diagram of the structure of the improved 33-node microgrid system in the microgrid hierarchical optimization and reconstruction method based on BP neural network.

Figure 3 shows the BP neural network model in the hierarchical optimization and reconstruction method of microgrid based on BP neural network.

Fig. 4 is the training mean square error diagram of the BP neural network model in the hierarchical optimization and reconstruction method of the microgrid based on the BP neural network.

FIG. 5 is a fitting diagram of the training data of the BP neural network model in the hierarchical optimization and reconstruction method of the microgrid based on the BP neural network.

Figure 6 is a graph of the test output error of the BP neural network in the hierarchical optimization and reconstruction method of the microgrid based on the BP neural network.

Detailed ways

The following describes in detail the embodiments of the present invention, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary, and are intended to explain the present invention and should not be construed as limiting the present invention.

As shown in Figure 1 to Figure 6; a BP neural network-based hierarchical optimization and reconstruction method for microgrid, which includes:

S1: Establish a microgrid optimization and reconstruction model, and construct an objective function and constraints for microgrid optimization and reconstruction;

S2: Establish a BP neural network model, and train and test the neural network;

S3: Using the idea of hierarchical optimization, in the reconstruction of the microgrid, the optimization process that does not involve power flow calculation is placed in the first-level process, and the optimization process involving power flow calculation is treated as the second-level optimization process. The second-level optimization process is performed only when the first-level optimization conditions are met, and the trained BP neural network is used to replace the power flow calculation in the second-level optimization process;

S4: Introduce the comprehensive evaluation method, construct a comprehensive evaluation function, comprehensively evaluate the output results of the BP network, and select the optimal reconstruction scheme.

More specifically, step 1) in the optimization and reconstruction of the microgrid, there are:

S11: Microgrid reconstruction needs to meet the goal of minimum load shedding, and its objective function is:

In the formula: i∈Ω, Ω is the set of nodes to remove the load after reconstruction; S _i represents the load corresponding to node i.

The stable operation of the microgrid after reconstruction must meet the following constraints:

1) Balanced node power constraints:

P _tmin ≤P _t ≤P _tmax ; (2)

In the formula: P _t is the adjustable active power of the balance node t; P _tmax is the upper limit of the adjustable active power of the node t; P _tmin is the lower limit of the adjustable active power of the node t.

2) Branch power constraints:

P _{Bj ≤P Bjmax} ; (3)

In the formula: P _Bj is the active power flowing through branch j; P _Bjmax is the upper limit of active power transmission of branch j.

3) Power balance constraints:

In the formula: P _Gx is the generated power of the micro-power source x in the micro-grid, X is the number of micro-power sources retained after reconstruction; P _Li is the load active power retained by node i after the micro-grid reconstruction, and N is the retained power after reconstruction. number of nodes.

4) Micro power generation power constraints:

P _Gmin ≤P _G ≤P _Gmax ; (5)

In the formula: _PG is the total generated power of the micro-power after the micro-grid reconstruction; _PGmin is the lower limit of the generated power in the micro-grid; _PGmax is the upper limit of the generated power in the micro-grid.

5) Node voltage constraints:

U _imin ≤U _i ≤U _imax ; (6)

In the formula: U _i is the voltage of node i; U _imin is the lower limit of the voltage of node i; U _imax is the upper limit of the voltage of node i.

Step 2) In the construction of BP neural network, the training and testing steps are as follows,

S21: In order to improve the training accuracy, the neural network model constructed by the present invention contains two hidden layers, and the number of nodes in each hidden layer is determined according to the following formula:

In the formula, h is the number of hidden layer nodes, z and v are the number of input layer and output layer nodes, respectively, and c is an adjustment constant between 1 and 10.

S22: Use the Matpower toolkit to calculate the power flow of the microgrid corresponding to different switching states, including the balance node power P _t , the network loss L, the maximum node voltage U _max and the minimum voltage U _min and so on. The above data are collected.

S23: Among the collected data, arbitrarily select 80% of the data as the training set, and the remaining 20% as the test set. For the data in the training set, the switching state of the three-level load is used as the input of the BP neural network and imported into the input layer; the balanced node power, network loss, node maximum voltage and minimum voltage calculated by the Matpower toolkit are used as the output. , import the output layer. Use the neural network toolbox in Matlab to select the number of hidden layers and the number of nodes corresponding to each hidden layer, and then set the training target minimum error, maximum training times, learning rate and transfer function, within the maximum training range to train.

S24: For the BP neural network obtained by training, import the switch state in the test set data into the input layer, run the BP neural network, and obtain the output results of balancing node power, network loss, node maximum voltage and minimum voltage, etc., and output the obtained The result is compared with the data in the test set, and the error of each group of results is calculated. If the error of each group of results is within the given error range, the trained BP neural network meets the requirements.

Step 3) After obtaining the BP neural network, the process of applying the hierarchical optimization idea to the microgrid reconstruction is as follows:

S31: In the first-level optimization process, the objective functions and constraints that do not involve any power flow calculation are processed. Considering the balance between power generation and load, integer programming is performed to obtain load switch state combinations that meet the balance conditions, and the switch combination solution set D is obtained according to the ascending order of power supply load shedding.

S32: In the second-level optimization process, the objective functions and constraints involved in the power flow calculation are processed. The switch combination solution set D obtained by the first-level optimization is put into the BP neural network for prediction one by one, and the prediction result is obtained.

S33: For the load shedding amount, determine whether the prediction result satisfies the relevant constraints of the stable operation of the microgrid. If only one set of switch combinations is satisfied, the switch combination is the optimal solution for microgrid reconstruction. On the contrary, if the output results of multiple groups of switch combinations satisfy the constraints, the optimal solution for microgrid reconfiguration is selected according to the comprehensive evaluation method. If all combinations are not satisfied, select the next load shedding amount and repeat the above steps until the optimal solution is found.

S34: If the output results of all switch combinations in the switch state set D do not satisfy the constraint conditions, it indicates that the fault reconstruction is abnormal, and the exception processing mechanism is entered. In all switch combinations that do not meet the constraints, the node voltage deviation of each switch combination and the balance node power over-limit value are calculated comprehensively, as shown below:

In the formula, w ₁ and w ₂ are weight coefficients and 0<w ₁ , w ₂ <1, and w ₁ +w ₂ =1. i is the minimum voltage node, dU _imin is the absolute value of the lower limit of the minimum voltage, U _imin is the lower limit of the node voltage; j is the maximum voltage node, dU _jmax is the absolute value of the upper limit of the maximum voltage, U _jmax is the upper limit of the node voltage value; t is the balance node, dP _tmin is the absolute value of the lower limit power of the balance node, P _tmin is the lower limit value of the balance node power, dP _tmax is the absolute value of the upper limit power of the balance node, and P _tmax is the balance node power upper limit value . The switch combination with the smallest comprehensive threshold Y is selected as the optimal reconstruction output solution.

Step 4) The comprehensive evaluation process is as follows:

S41: If the output results of multiple switch combinations for the same load shedding amount satisfy the constraints, multiple sets of corresponding network loss, voltage deviation and power deviation outputs will be obtained. Set the reference values of network loss, voltage deviation and power deviation, normalize the output results of each group of switches, and select the corresponding weight coefficient k _i according to the preference of the decision maker, and use the comprehensive evaluation function to select The switch combination with the smallest comprehensive evaluation function value is regarded as the optimal solution.

S42: Microgrid reconstruction, when the load shedding amount is the same and the results meet the constraints, the goal of the minimum comprehensive evaluation function value needs to be met, and the objective function is:

分别为网络损耗、电压偏差和平衡节点功率偏差的归一化处理值。In the formula: k ₁ , k ₂ , k ₃ are the weight coefficients of network loss, voltage deviation and balance node power deviation, and 0<k ₁ , k ₂ , k ₃ <1, k ₁ +k ₂ +k ₃ =1 ;

are the normalized processing values of network loss, voltage deviation and balance node power deviation, respectively.

S43: Before the comprehensive evaluation, the output results should be normalized first, and the calculation method is as follows:

In the formula: L is the system network loss value, L ^* is the set network loss reference value; ΔU is the voltage deviation, which refers to the difference between the highest voltage and the lowest voltage of the node in the entire system, U ^* is the system reference voltage; ΔP _t is The power deviation refers to the deviation between the power of the balance node and the set reference power, and P _t ^* is the balance node reference power, which refers to the average value of the upper and lower limits of the balance node power. in,

In the formula: P _Bj , Q _Bj , R _Bj are the active power, reactive power and resistance corresponding to branch j, respectively, where Q _Bj is obtained according to the load power factor and active power; U _Bj is the end of branch j Voltage; r _j is the branch state after the corresponding reconstruction, 1 means connected, 0 means disconnected; J is the number of branches contained in the system.

More specifically, in Figure 2, it is assumed that the branch impedance and load data in the microgrid are shown in Table 1. A micro gas turbine MT with a capacity of 2.5MW is installed at node 1, a photovoltaic power generation PV1 with a capacity of 2MW is installed at node 3, and a wind turbine WG1 with a capacity of 1MW is installed at node 6. Among them, the gas turbine is used as the balance node of the system, and the rest of the generators and loads are in PQ mode

Table 1 Branch impedance and load data

Before the simulation, the active power constraint range of the gas turbine is set to 0-2.1MW; the voltage constraint range of each node is 0.9-1.1; the grid voltage reference value is 12.66kV; Both generator and load are in PQ mode. Assuming that the microgrid needs to be reconstructed, the active load of 0.9-1.1 MW in the network needs to be removed, and the method of the present invention is used for reconstruction. Among them, Table 2 represents the first-level optimization results, Table 3 represents the second-level optimization results, and Table 5 represents the optimal reconstruction results as follows:

Table 2 The first-level optimization results

Table 3 Second-level optimization results

When the cut load is 0.9MW, the active power of the balance nodes corresponding to the three groups of structures are all larger than the capacity of the gas turbine, which does not satisfy the balance between the load and the power generation. Therefore, these three groups of switch combinations are eliminated. When the next load shedding amount is 1.0MW, the active power of the balance node of the corresponding three groups of structures is less than the capacity of the gas turbine, which satisfies the balance relationship between the load and the power generation; and the minimum and maximum voltages are both within the allowable range, so The three groups of structures with a load shedding capacity of 1.0 MW were comprehensively evaluated to determine the optimal solution.

The weights of the comprehensive evaluation function (ie, formula (9)) are selected as k ₁ =0.8, k ₂ =0.1, and k ₃ =0.1. Table 4 shows the comprehensive evaluation results with a load removal amount of 1.0 MW.

Table 4. Comprehensive evaluation results of load shedding of 1MW

Table 5 Optimal reconstruction results of the method of the present invention

In order to show the advantages of the method of the present invention, compared with the hierarchical optimization method that directly calculates the power flow, the optimal solution results can be obtained as shown in Table 6.

Table 6 Optimal reconstruction results of comparison methods

According to the results in Table 5 and Table 6, it can be seen that the final switch combinations (ie reconstruction schemes) obtained by the two methods are consistent, indicating that the two methods have the same accuracy. Comparing the reconstruction time of the method of the present invention and the comparison method, the reconstruction time of the method of the present invention is 0.073s, and the reconstruction time of the comparison method is 6.571s, the time is shortened by about 90 times, and the calculation efficiency is obviously improved.

The above are only specific embodiments of the present invention, but the technical features of the present invention are not limited thereto. Any changes or modifications made by those skilled in the art in the field of the present invention are all covered by the patent scope of the present invention. among.

Claims (8)

1. a microgrid hierarchical optimization and reconstruction method based on BP neural network, is characterized in that, described method comprises the following steps: S1: Establish a microgrid optimization and reconstruction model, and construct an objective function and constraints for microgrid optimization and reconstruction; S2: Establish a BP neural network model, and train and test the neural network; S3: Using the idea of hierarchical optimization, in the reconstruction of the microgrid, the optimization process that does not involve power flow calculation is placed in the first-level processing, and the optimization process involving power flow calculation is treated as the second-level optimization process; only when the first-level optimization process is satisfied The second-level optimization process is performed only for the solution of the optimization conditions, and the trained BP neural network is used to replace the power flow calculation in the second-level optimization process; S4: Introduce the comprehensive evaluation method, construct a comprehensive evaluation function, comprehensively evaluate the output results of the BP network, and select the optimal reconstruction scheme. 2. the microgrid hierarchical optimization and reconstruction method based on BP neural network as claimed in claim 1, is characterized in that, described step S1 is under constraint condition, and the objective function of microgrid optimization reconstruction is: S11: Microgrid reconstruction needs to meet the goal of minimum load shedding, and its objective function is:

In the formula: i∈Ω, Ω is the set of nodes to remove the load after reconstruction; S _i represents the load corresponding to node i; The stable operation of the microgrid after reconstruction must meet the following constraints: 1) Balanced node power constraints: P _tmin ≤P _t ≤P _tmax In the formula: P _t is the adjustable active power of the balance node t; P _tmax is the upper limit of the adjustable active power of the node t; P _tmin is the lower limit of the adjustable active power of the node t; 2) Branch power constraints: P _Bj ≤P _{Bj max} In the formula: P _Gx is the generated power of the micro-power source x in the micro-grid, X is the number of micro-power sources retained after reconstruction; P _Li is the load active power retained by node i after the micro-grid reconstruction, and N is the retained power after reconstruction. number of nodes; 3) Power balance constraints:

In the formula: P _Gx is the generated power of the micro-power source x in the micro-grid, X is the number of micro-power sources retained after reconstruction; P _Li is the load active power retained by node i after the micro-grid reconstruction, and N is the retained power after reconstruction. number of nodes; 4) Micro power generation power constraints: P _{G min} ≤P _G ≤P _{G max} In the formula: P _G is the total generated power of the micro-power after the micro-grid reconstruction; P _{G min} is the lower limit of the generated power in the micro-grid; P _{G max} is the upper limit of the generated power in the micro-grid; 5) Node voltage constraints: U _imin ≤U _i ≤U _imax In the formula: U _i is the voltage of node i; U _imin is the lower limit of the voltage of node i; U _imax is the upper limit of the voltage of node i. 3. the microgrid hierarchical optimization reconstruction method based on BP neural network as claimed in claim 1, is characterized in that, in described step S2, the construction of BP neural network is as follows: S21: In order to improve the training accuracy, the neural network model constructed by the present invention contains two hidden layers, and the number of nodes in each hidden layer is determined according to the following formula:

In the formula, h is the number of hidden layer nodes, z and v are the number of input layer and output layer nodes, respectively, and c is an adjustment constant between 1 and 10. 4. the microgrid grading optimization and reconstruction method based on BP neural network as claimed in claim 1, is characterized in that, in described step S2, the training and testing process of BP neural network is as follows: S22: Use the Matpower toolkit to calculate the power flow of the microgrid corresponding to different switching states, including the balance node power P _t , the network loss L, the maximum node voltage U _max and the minimum voltage U _min ; collect the above data; S23: In the collected data, arbitrarily select 80% of the data as the training set, and use the neural network toolbox in Matlab for training; S24: For the BP neural network obtained by training, the remaining 20% are imported into the BP neural network as the test set data, and the output results are compared with the data in the test set to determine that the BP neural network meets the requirements. 5. the microgrid hierarchical optimization and reconstruction method based on BP neural network as claimed in claim 1, is characterized in that, in described step S3, the hierarchical optimization thought is applied to process in microgrid reconstruction as follows: S31: In the first-level optimization process, deal with the objective function and constraints that do not involve any power flow calculation; consider the balance between the power generation and the load, perform integer programming, and obtain a load switch state combination that satisfies the balance condition, and follow the The switch combination solution set D is obtained by sorting the power supply load shedding amount in ascending order; S32: In the second-level optimization process, the objective functions and constraints involved in the power flow calculation are processed, and the switch combination solution set D obtained by the first-level optimization is put into the BP neural network one by one for prediction, and the prediction result is obtained; S33: For the load shedding amount, determine whether the prediction result satisfies the relevant constraints of the stable operation of the microgrid; if only one switch combination satisfies, the switch combination is the optimal solution for the microgrid reconstruction; on the contrary, if there are multiple groups of switches If the output result of the switch combination satisfies the constraints, the optimal solution for microgrid reconstruction is selected according to the comprehensive evaluation method; if all combinations are not satisfied, the next load shedding amount is selected and the above steps are repeated until the optimal solution is found. 6. the microgrid grading optimization and reconstruction method based on BP neural network as claimed in claim 1, is characterized in that, in described step S4, comprehensive evaluation method is as follows: S41: If the output results of multiple switch combinations for the same load shedding amount satisfy the constraints, multiple sets of corresponding network loss, voltage deviation and power deviation outputs will be obtained; set the reference values of network loss, voltage deviation and power deviation, and set the The output results of each group of switches are normalized, and the corresponding weight coefficient k _i is selected according to the preference of the decision maker, and the switch combination with the smallest comprehensive evaluation function is selected as the optimal solution by using the comprehensive evaluation function. 7. the microgrid hierarchical optimization and reconstruction method based on BP neural network as claimed in claim 1, is characterized in that, in described step S4, comprehensive evaluation function is constructed as follows: S42: Microgrid reconstruction, when the load shedding amount is the same and the results meet the constraints, the goal of the minimum comprehensive evaluation function value needs to be met, and the objective function is:

In the formula: k ₁ , k ₂ , k ₃ are the weight coefficients of network loss, voltage deviation and balance node power deviation, and 0<k ₁ , k ₂ , k ₃ <1, k ₁ +k ₂ +k ₃ =1 ;

are the normalized processing values of network loss, voltage deviation and balance node power deviation, respectively; S43: Before the comprehensive evaluation, the output results should be normalized first, and the calculation method is as follows:

In the formula: L is the system network loss value, L ^* is the set network loss reference value; ΔU is the voltage deviation, which refers to the difference between the highest voltage and the lowest voltage of the node in the entire system, U ^* is the system reference voltage; ΔP _t is Power deviation refers to the deviation between the power of the balance node and the set reference power, P _t ^* is the reference power of the balance node refers to the average value of the upper and lower limits of the power of the balance node; among them,

In the formula: P _Bj , Q _Bj , R _Bj are the active power, reactive power and resistance corresponding to branch j, respectively, where Q _Bj is obtained according to the load power factor and active power; U _Bj is the end of branch j Voltage; r _j is the branch state after the corresponding reconstruction, 1 means connected, 0 means disconnected; J is the number of branches contained in the system. 8. The BP neural network-based microgrid hierarchical optimization and reconstruction method according to claim 1, wherein in the step S3, the abnormality processing process is as follows: S34: If the output results of all switch combinations in the switch state set D do not meet the constraints, it indicates that the fault reconstruction is abnormal, and the exception processing mechanism is entered; in all switch combinations that do not meet the constraints, the node voltage of each switch combination is calculated. Deviation, the balance node power exceeds the limit, and the comprehensive limit Y calculation is performed, as shown below:

In the formula, w ₁ , w ₂ are weight coefficients and 0<w ₁ , w ₂ <1, w ₁ +w ₂ =1; i is the minimum voltage node, dU _imin is the absolute value of the lower limit of the minimum voltage, U _imin is the lower limit value of the node voltage; j is the maximum voltage node, dU _jmax is the absolute value of the upper limit of the maximum voltage, U _jmax is the upper limit value of the node voltage; t is the balance node, dP _tmin is the absolute value of the lower limit power of the balance node, P _tmin is the lower limit of the power of the balance node, dP _tmax is the absolute value of the over-limit power on the balance node, and P _tmax is the upper limit of the power of the balance node; the switch combination with the smallest comprehensive over-limit Y is selected as the optimal reconstruction output solution.

Images