Power grid transmission section power flow adjustment unit selection method based on deep learning

Abstract

The application discloses a power grid transmission section power flow adjustment unit selection method based on deep learning, and comprises the following steps: step one, constructing a power grid transmission section power flow adjustment unit sample database; step two, constructing a power grid transmission section power flow adjustment unit selection model based on deep learning; and step three, generating a power flow adjustment unit online according to a real-time operation state of the power grid. The power flow sensitivity of each unit to the transmission section is calculated repeatedly in the traditional method, and the power flow is calculated repeatedly, which is changed into offline work, so that the algorithm for generating the power flow adjustment unit online according to the real-time operation state of the power grid is realized, a large number of power flow calculations are avoided, the power flow adjustment unit of the transmission section is quickly selected, and the power grid power flow adjustment work efficiency is greatly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid operation safety analysis, in particular to a power grid transmission section power flow adjustment unit selection method based on deep learning in a data-driven mode. BACKGROUND

[0002] With the development of social economy, the demand for electricity is increasing, and the high proportion of renewable energy connected to the grid and the increasing load level make the structure and operation state of the power grid complex and changeable, and the power grid transmission section power blockage occurs more frequently. When the power grid transmission section is blocked, the power flow needs to be adjusted in time to avoid further power grid disaster. The traditional power grid transmission section blocking adjustment needs to select the unit with high sensitivity to the key section line power flow, and adjust the output of the unit, so that the power flow of the transmission section can be transferred to other lines of the power grid. However, calculating the power flow sensitivity of each unit to the transmission section requires repeated calculation of the power flow, which is time-consuming and difficult to ensure the timeliness of power flow adjustment. Under this background, it is increasingly important to use deep learning to train the massive power grid transmission section power flow adjustment unit offline, and to select the power system power flow adjustment unit online quickly.

[0003] Deep learning can learn the internal laws and representation levels of sample data, has the advantages of automatically extracting target features, can learn the features of data directly from data, and represents the complex nonlinear mapping relationship between data. The deep learning model can be trained offline, and the online calculation is fast. Deep learning has the advantages of efficient learning of sample features and fast online calculation, and has been widely used in power systems. The convolution kernel parameter sharing and interlayer connection sparsity in the convolution neural network hidden layer enable the convolution neural network to learn the grid features with small calculation amount, have stable effect, and have no additional feature engineering requirements for data. It can be seen that the convolution neural network can learn the mapping relationship between the power grid operation state and the key section and the power flow adjustment unit.

[0004] The structure of the power system is becoming increasingly complex with the improvement of economic level, and repeated calculation of the power flow is time-consuming and difficult to meet the timeliness requirements of power flow adjustment. Therefore, how to use deep learning algorithm to train the massive power grid transmission section power flow adjustment unit offline, to excavate the mapping relationship between the power grid operation state and the key section and the power flow adjustment unit, to construct the power grid transmission section power flow adjustment unit selection deep learning model, and to realize the online fast and accurate selection of the transmission section and the power flow adjustment unit. SUMMARY

[0005] The present application provides a power grid transmission section power flow adjustment unit selection method based on deep learning.

[0006] The present application is realized by the following technical solutions:

[0007] The power grid transmission section power flow adjustment unit selection method based on deep learning comprises the following steps:

[0008] Step one: build a power grid transmission section power flow adjustment unit sample database.

[0009] Step two: build a deep learning-based power grid transmission section power flow adjustment unit selection model.

[0010] Step three: generate power flow adjustment units online in real-time operation state of power grid.

[0011] Further: in the step one, the specific method for building the power grid transmission section power flow adjustment unit sample database is as follows:

[0012] Provide random model power grid operation data, including randomly generated power grid topology and load, adjust the generator data, and use the optimal power flow to minimize the power grid loss, thereby generating generator operation data.

[0013] According to the determined power grid operation data state, calculate the load rate of the power grid transmission section, and according to the load rate, classify and extract the larger load transmission section as the key section line of the region.

[0014] By adjusting the output power of each generator, the change of the key section line power flow is calculated, and the ratio of the amplitude value before and after the adjustment of each generator is used as the sensitivity coefficient of the generator and the key section line, and the generator set is sorted in descending order of sensitivity, and the generator set sorting result is saved.

[0015] At this point, the sample data is obtained, which is the power grid operation data, including the power grid topology state, the active power P, the reactive power Q, the voltage amplitude U, the voltage phase angle θ, the key section line, and the generator set order.

[0016] According to the above steps, multiple samples can be generated to form the power grid transmission section power flow adjustment unit sample database.

[0017] Further: in the step two, the specific method for building the deep learning-based power grid transmission section power flow adjustment unit selection model is as follows:

[0018] In order to more accurately map the relationship between the power grid operation state and the key section and the power flow adjustment unit, the input and output data format of the deep learning model is determined.

[0019] The power grid operation data is taken as the input data of the convolutional neural network model, including the power grid topology state, the active power P, the reactive power Q, the voltage amplitude U and the voltage phase angle θ of each power grid node, the key section line and the generator set sequence are taken as the output data of the model, so as to train the convolutional neural network model, and the specific operation is as follows.

[0020] A power grid topology including N power grid nodes and M generator set nodes is provided, a power grid transmission section power flow adjustment unit selection model based on deep learning is constructed, and the power grid transmission section and the power flow adjustment unit sequence are calculated.

[0021] One: preliminary work, the power grid topology is represented by the power grid association matrix H, H is an N×N matrix, when the nodes i and j are directly connected, then in the association matrix H ij =1, when the nodes i and j are not directly connected, then in the association matrix H ij =0.

[0022] The active power P, the reactive power Q, the voltage amplitude U and the voltage phase angle θ of each power grid node form a node feature vector, and then N nodes form an N×4 node feature matrix A.

[0023] The power grid association matrix H and the node feature matrix A are spliced by column to obtain an N×(N+4) input matrix X.

[0024] The maximum number of power grid topology lines is a, the maximum number of power grid generator sets is b, the maximum number of key section lines is c, each key section line can be composed of different lines, corresponding to a group of generator power flow adjustment sequences, a key section line can be composed of an a-dimensional line vector L, when the i-th line is contained in the key section line, then L i =1, otherwise L i =0.

[0025] The unit power flow adjustment sequence corresponding to a key section line can be represented by a b-dimensional unit power flow adjustment sequence vector G, when the j-th unit sequence is 3, then G j =3.

[0026] The line vector L and the unit power flow adjustment sequence vector G are spliced to obtain an a+b-dimensional vector, all key section lines and their corresponding unit power flow adjustment sequences can form a c×(a+b)-dimensional output matrix Y.

[0027] Two: build a power grid transmission section power flow adjustment unit selection model based on deep learning, adopt the convolutional neural network model in deep learning to map the relationship between power grid operation state and key section and power flow adjustment unit, the convolutional neural network includes input layer, convolution layer, pooling layer, full connection layer and output layer, the input layer is input matrix X, the input data is first extracted features by the convolution kernel in the convolution layer, the corresponding feature matrix is obtained, the jth convolution feature matrix h i,j Of the i th convolution layer contains convolution kernel W and bias matrix B and activation function f, then the feature matrix h i,j Is:

[0028] h i,j =f(X*W+B)

[0029] The pooling layer is the next layer of the convolution layer, which extracts data feature information again to further reduce the size of the feature matrix, the pooling matrix of the pooling layer is composed of multiple feature matrices, the data is transmitted to the output layer after the full connection layer, and the output matrix is the output matrix Y.

[0030] Three: offline training of power grid transmission section power flow adjustment unit selection model based on deep learning, through the generated power grid transmission section power flow adjustment unit sample database, according to the construction of power grid transmission section power flow adjustment unit selection model, the weight matrix in the model is continuously optimized by using the stochastic gradient descent algorithm, so as to improve the selection accuracy of power grid transmission section power flow adjustment unit, until the iteration number is reached or the model accuracy requirement is met, the optimization model is terminated, and the offline training of the model is completed.

[0031] Further: in the step three, the specific method that the power grid real-time operation state on-line generates the power flow adjustment unit is as follows: the power grid real-time operation data is directly obtained from the power grid online real-time operation, including power grid topology state, active power P, reactive power Q, voltage amplitude U and voltage phase angle θ of each power grid node, so as to form the input matrix of the power grid transmission section power flow adjustment unit selection model based on deep learning according to the first step in step two, input the obtained input matrix into the trained model in step two, and the output matrix of the model can be quickly obtained, and the power grid transmission section and the power flow adjustment unit sequence can be directly obtained according to the output matrix.

[0032] The beneficial effects of the present application are:

[0033] Compared with the prior art, the power grid power transmission section power flow adjustment unit selection method based on deep learning can guarantee the timeliness of power flow adjustment more. The power grid power transmission section power flow adjustment unit selection model can be trained offline to mine the mapping relationship between the power grid operation state and the key section and the power flow adjustment unit, realize offline training of the massive power grid power transmission section power flow adjustment unit, and the power grid real-time operation state can online quickly generate the power transmission section and the power flow adjustment unit. The power grid real-time safety and stability analysis is provided. The power grid power transmission section power flow adjustment unit selection method based on deep learning has the rapid effectiveness of online calculation, changes the traditional calculation of the power flow sensitivity of each unit to the power transmission section into offline work, thereby realizing the algorithm of generating the power flow adjustment unit online according to the power grid real-time operation state, avoiding a large amount of power flow calculation, quickly selecting the power transmission section power flow adjustment unit, and greatly improving the power grid power flow adjustment work efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The method flowchart of the present application is shown in the following figure.

[0035] Figure 2 The power grid power transmission section power flow adjustment unit selection model based on deep learning is constructed. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0037] Please refer to Figure 1 and Figure 2 wherein Figure 1 The method flowchart of the present application is shown in the following figure. Figure 2 The power grid power transmission section power flow adjustment unit selection model based on deep learning is constructed.

[0038] The present application is realized by the following technical scheme:

[0039] The power grid power transmission section power flow adjustment unit selection method based on deep learning comprises the following steps:

[0040] Step 1: Construct a power grid power transmission section power flow adjustment unit sample database.

[0041] The specific method is as follows:

[0042] The random model power grid operation data is provided, including the random generation of power grid topology and load, the generator data is adjusted, the power grid loss is minimized by using the optimal flow, and the generator operation data is generated.

[0043] The load rate of the power transmission section is calculated according to the determined power grid operation data state, and according to the load rate, the load is extracted as the key section line of the region.

[0044] By adjusting the output power of each generator, the change of the key section line power flow is calculated, and the ratio of the amplitude value before and after the adjustment of each generator is used as the sensitivity coefficient of the generator and the key section line, and the generator set is sorted in descending order of sensitivity, and the generator set sorting result is saved.

[0045] At this point, the sample data is obtained, which is the power grid operation data, including the power grid topology state, the active power P, the reactive power Q, the voltage amplitude U, the voltage phase angle θ, the key section line, and the generator set sequence.

[0046] According to the above steps, multiple samples can be generated to form a power transmission section power flow adjustment unit sample database.

[0047] Step two: build a power transmission section power flow adjustment unit selection model based on deep learning.

[0048] The specific method is as follows (see Figure 2 shown):

[0049] In order to more accurately map the relationship between the power grid operation state and the key section and the power flow adjustment unit, the input and output data format of the deep learning model is determined.

[0050] The power grid operation data is used as the input data of the convolutional neural network model, including the power grid topology state, the active power P, the reactive power Q, the voltage amplitude U, and the voltage phase angle θ of each power grid node, and the key section line and the generator set sequence are used as the output data of the model, so as to train the convolutional neural network model, and the specific operation is as follows.

[0051] A power grid topology including N power grid nodes and M generator set nodes is provided, a power transmission section power flow adjustment unit selection model based on deep learning is built, and the power transmission section and the power flow adjustment unit sequence are calculated.

[0052] One: preliminary work, the power grid topology is represented by the power grid association matrix H, H is an N*N matrix, when the nodes i and j are directly connected, H ij =1, when the nodes i and j are not directly connected, H ij =0.

[0053] The active power P, the reactive power Q, the voltage amplitude U, and the voltage phase angle θ of each power grid node form a node feature vector, and the N nodes form an N×4-dimensional node feature matrix A.

[0054] The power grid association matrix H and the node feature matrix A are spliced by column to obtain an N×(N+4)-dimensional input matrix X.

[0055] The maximum number of power grid topology lines is a, the maximum number of power grid generator sets is b, and the maximum number of key section lines is c. Each key section line can be composed of different lines, corresponding to a group of generator power flow adjustment sequences. A key section line can be composed of an a-dimensional line vector L. When the i-th line is contained in the key section line, L i = 1, and L i = 0 when it is not contained.

[0056] The generator power flow adjustment sequence corresponding to a key section line can be represented by a b-dimensional generator power flow adjustment sequence vector G. When the j-th generator sequence is 3, G j = 3.

[0057] Splicing the line vector L and the generator power flow adjustment sequence vector G can obtain an a+b-dimensional vector. All key section lines and their corresponding generator power flow adjustment sequences can form a c×(a+b)-dimensional output matrix Y.

[0058] Two: Construct a power grid transmission section power flow adjustment generator selection model based on deep learning. A convolutional neural network model in deep learning is used to map the relationship between the power grid operating state and the key section and the power flow adjustment generator. The convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The input layer is the input matrix X. The input data is first extracted by the convolutional kernel in the convolutional layer to obtain the corresponding feature matrix. The j-th convolutional feature matrix h i,j of the i-th convolutional layer contains the convolutional kernel W, the bias matrix B, and the activation function f. Therefore, the feature matrix h i,j is:

[0059] h i,j = f(X*W+B)

[0060] The pooling layer is the next layer of the convolutional layer, which further extracts the data feature information and further reduces the size of the feature matrix. The pooling matrix of the pooling layer is composed of multiple feature matrices. After the data passes through the fully connected layer, it is transmitted to the output layer. The output matrix is the output matrix Y.

[0061] Three: offline training of power grid transmission section power flow adjustment unit selection model based on deep learning, through the generated power grid transmission section power flow adjustment unit sample database, according to the construction of power grid transmission section power flow adjustment unit selection model, the weight matrix in the model is continuously optimized by using the stochastic gradient descent algorithm, so as to improve the accuracy of power grid transmission section power flow adjustment unit selection, until the iteration number is reached or the model accuracy requirement is met, the optimization model is terminated, and the offline training of the model is completed.

[0062] Step three: online generation of power flow adjustment unit in real-time operation state of power grid.

[0063] The specific method is as follows: the real-time operation data of power grid is directly obtained from the online real-time operation of power grid, including the topological state of power grid, the active power P, the reactive power Q, the voltage amplitude U and the voltage phase angle θ of each power grid node, so as to form the input matrix of power grid transmission section power flow adjustment unit selection model based on deep learning according to the first step in step two, input the obtained input matrix into the trained model in step two, and the output matrix of the model can be quickly obtained according to the output matrix, and the power grid transmission section and the power flow adjustment unit sequence can be directly obtained.

[0064] The beneficial effects of the present application are:

[0065] Compared with the prior art, the power grid transmission section power flow adjustment unit selection method based on deep learning can better guarantee the timeliness of power flow adjustment. The power grid transmission section power flow adjustment unit selection model can be trained offline to mine the mapping relationship between power grid operation state and key section and power flow adjustment unit, realize offline training of massive power grid transmission section power flow adjustment unit, online fast generation of transmission section and power flow adjustment unit in real-time operation state of power grid, and quick real-time adjustment of generator unit output, so as to avoid further power grid disaster, thereby providing a basis for real-time safety and stability analysis of power grid. The power grid transmission section power flow adjustment unit selection method based on deep learning has the advantages of fast and effective online calculation, and the traditional calculation of the power flow sensitivity of each unit to the transmission section needs to be repeatedly calculated, which is changed to offline work, so as to realize the algorithm of online generation of power flow adjustment unit according to the real-time operation state of power grid, avoid a large number of power flow calculations, quickly select the power flow adjustment unit of the transmission section, and greatly improve the work efficiency of power grid power flow adjustment.

[0066] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only preferred examples of the present application and are not intended to limit the present application. Various changes and improvements can be made to the present application without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1.A method for power grid transmission section power flow adjustment unit selection based on deep learning, characterized in that, The method comprises the following steps: Step 1: Constructing a power grid transmission section power flow adjustment unit sample database; Random model power grid operation data is provided, including randomly generated power grid topology and load, the power grid loss is minimized by adjusting the generator data and using optimal power flow, thereby generating generator operation data; According to the determined power grid operation data state, the load rate of the power grid transmission section is calculated, and according to the load rate, the larger load transmission section is extracted as the key section line of the region; By adjusting the output power of each generator, the change of the key section line power flow is calculated, and the ratio of the amplitude value before and after the adjustment of each generator is used as the sensitivity coefficient of the generator and the key section line, and the generator set is sorted in descending order of sensitivity, and the generator set sorting result is saved; At this point, the sample data is obtained, which is the power grid operation data, including the power grid topology state, the active power P, the reactive power Q, the voltage amplitude U, the voltage phase angle θ, the key section line, and the generator set order; According to the above steps, multiple samples can be generated to form a power grid transmission section power flow adjustment unit sample database; Step 2: Constructing a power grid transmission section power flow adjustment unit selection model based on deep learning; In order to more accurately map the relationship between the power grid operation state and the key section and the power flow adjustment unit, the input and output data formats of the deep learning model are determined; The power grid operation data is used as the input data of the convolutional neural network model, including the power grid topology state, the active power P, the reactive power Q, the voltage amplitude U, and the voltage phase angle θ of each power grid node, and the key section line and the generator set order are used as the output data of the model, so as to train the convolutional neural network model, and the specific operation is as follows: A power grid topology including N power grid nodes and M generator set nodes is provided, and a power grid transmission section power flow adjustment unit selection model based on deep learning is constructed to obtain the power grid transmission section and the power flow adjustment unit order: One: preliminary work, power grid topology is represented by power grid incidence matrix H, H is N x N matrix, when the power grid topology node i, j is directly connected, then the incidence matrix H ij = 1, when the node i, j is not directly connected, then the incidence matrix H ij = 0; The active power P, the reactive power Q, the voltage amplitude U, and the voltage phase angle θ of each power grid node form a node feature vector, and N nodes form an N×4-dimensional node feature matrix A; The power grid association matrix H and the node feature matrix A are spliced by column to obtain an N×(N+4)-dimensional input matrix X; The maximum number of power grid topology lines is a, the maximum number of power grid generator sets is b, the maximum number of key section lines is c, each key section line can be composed of different lines, corresponding to a group of generator power flow adjustment sequences, a key section line can be composed of an a-dimensional line vector L, when the i-th line is contained in the key section line, L i = 1, and L i = 0 when the i-th line is not contained. The unit power flow adjustment sequence corresponding to a key section line can be represented by a b-dimensional unit power flow adjustment sequence vector G. When the jth unit sequence is 3, G j = 3; The line vector L and the unit power flow adjustment order vector G are spliced to obtain an a+b-dimensional vector, and all key section lines and their corresponding unit power flow adjustment orders can form a c×(a+b)-dimensional output matrix Y; Two: Construct the power grid transmission section power flow adjustment unit selection model based on deep learning. The convolutional neural network model in deep learning is used to map the relationship between the power grid operation state and the key section and the power flow adjustment unit. The convolutional neural network includes an input layer, a convolution layer, a pooling layer, a full connection layer and an output layer. The input layer is an input matrix X. The input data is first extracted by the convolution kernel in the convolution layer to obtain the corresponding feature matrix. The jth convolution feature matrix h i,j of the ith convolution layer contains the convolution kernel W, the bias matrix B and the activation function f. The feature matrix h i,j is: h i,j = f(X*W + B) The pooling layer is the next layer of the convolutional layer, which further extracts the data feature information and further reduces the size of the feature matrix. The pooling matrix of the pooling layer is composed of multiple feature matrices. After the data passes through the fully connected layer, it is transmitted to the output layer, and the output matrix is the output matrix Y. Three: offline training of power grid transmission section power flow adjustment unit selection model based on deep learning, through the generated power grid transmission section power flow adjustment unit sample database, according to the construction of power grid transmission section power flow adjustment unit selection model, using stochastic gradient descent algorithm to optimize the weight matrix in the model, so as to improve the selection accuracy of power grid transmission section power flow adjustment unit, until the iteration number is reached or the model accuracy requirement is met, terminate the optimization model, complete the offline training of the model; Step three: online generation of power flow adjustment unit in real-time operation state of power grid; The real-time operation state of power grid directly obtains the real-time operation data of power grid, including the topological state of power grid, the active power P, the reactive power Q, the voltage amplitude U and the voltage phase angle θ of each power grid node, so as to form the input matrix of the power grid transmission section power flow adjustment unit selection model based on deep learning according to the first step in step two, input the obtained input matrix into the trained model in step two, and calculate the output matrix of the model, so as to directly obtain the power grid transmission section and the sequence of power flow adjustment unit according to the output matrix.

Images