Voltage sensitivity calculation method, device, equipment and medium based on residual network

Abstract

The present invention discloses a voltage sensitivity calculation method based on a residual network, which relates to the technical field of voltage detection, and is used to solve the problems of high computational complexity and large deviation of calculation results in existing voltage sensitivity detection. The method includes the following steps: obtaining the node voltage amplitude and the historical data of the system state parameters of the node, and constructing a training set; performing residual dendritic neural network training, and constructing an explicit expression of the node voltage amplitude with respect to the system state parameters; taking a single system state parameter as the principal element, calculating the partial derivative, and obtaining a sensitivity expression; obtaining the real-time data of the system state parameters and substituting them into the sensitivity expression, and calculating the node voltage sensitivity under the current grid topology. The present invention also discloses a voltage sensitivity calculation device, an electronic device, and a computer storage medium based on a residual network. The present invention performs residual dendritic neural network training through historical data, and then obtains a sensitivity expression, with low computational complexity and high precision.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network node voltage detection, and in particular to a node voltage sensitivity calculation method, device, equipment and medium based on a residual dendritic neural network. Background Art

[0002] With the continuous penetration of renewable energy represented by distributed photovoltaics in the power system, the problem of voltage fluctuation at the distribution network nodes has become more and more serious. The sensitivity of different state parameters of the system is usually used as the key parameter of the node voltage regulation and management link. For example, real-time data of state parameters such as active power and reactive power are injected. The traditional node voltage sensitivity calculation method mainly adopts the inversion of the power flow Jacobian matrix, and its accuracy is seriously dependent on the accuracy of the grid topology and line parameters. Due to the current distribution network’s common problems such as unclear medium-voltage line-to-transformer relationship, unclear low-voltage topology, and unknown line parameters, the traditional node voltage sensitivity calculation method cannot be applied to actual projects, and the calculation deviation is large; in addition, the inversion operation of the power flow Jacobian matrix has a high computational complexity and there is a risk of dimensionality explosion. Therefore, a more accurate sensitivity calculation method is urgently needed. Summary of the invention

[0003] In order to overcome the shortcomings of the prior art, one of the objects of the present invention is to provide a voltage sensitivity calculation method based on a residual network, which obtains the node voltage amplitude and the historical data of the system state parameters of the node, and then obtains the explicit expression of the voltage amplitude with respect to all state parameters of the system, and obtains the sensitivity expression by taking partial derivatives.

[0004] One of the purposes of the present invention is achieved by the following technical solution:

[0005] A voltage sensitivity calculation method based on a residual network comprises the following steps:

[0006] Obtain the node voltage amplitude and the historical data of the system state parameters of the node, and build a training set based on the acquired historical data;

[0007] The training set is used as an input value to train a residual dendritic neural network, and an explicit expression of the node voltage amplitude with respect to all state parameters of the system is constructed according to the trained residual dendritic neural network model;

[0008] Taking a single system state parameter of a node as a principal element, calculating the partial derivative of the explicit expression, and obtaining a sensitivity expression of the node voltage amplitude with respect to the single system state parameter;

[0009] The real-time data of system state parameters are obtained and substituted into the sensitivity expression to calculate the node voltage sensitivity under the current grid topology.

[0010] Furthermore, the system status parameters include active power and reactive power.

[0011] In order to ensure the accuracy of model training, avoid the small difference between adjacent measurement data, which causes the neural network to be unable to accurately mine data features, and prevent the data difference from being too small, which causes the model training to fail due to matrix singularity, it also includes: screening the system state parameter historical data, and constructing the training set according to the screened data. The data screening and training set construction process includes the following steps:

[0012] Step 1.1: Randomly select The system state parameters of each node at the time are taken as the reference operating point;

[0013] Step 1.2: Get the first System status parameter historical data;

[0014] Step 1.3: Calculate the The historical data of system status parameters and Operating point spacing ;

[0015] Step 1.4: Compare the distance with the operating status determination threshold:

[0016] If the distance is less than or equal to the running state determination threshold, , and return to step 1.2 to continue data screening;

[0017] Step 1.5: If the distance is greater than the running state judgment threshold, The historical data of system status parameters is taken as the running points as the data in the training set, complete a data screening, and let , , return to step 1.2 to continue building the training set.

[0018] Furthermore, the calculation of the data screening satisfies:

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] in, represents the set of all nodes in the network, ; , Node No. Historical injected active power and reactive power measurement data; , They are the nodes under the reference operating point The injected active power and reactive power measurement data; For the The historical measurement data and The distance between the reference operating points; is the rated capacity of the system; is the running status judgment threshold; when When , it is determined to be valid measurement data, used as the data in the training set, and used as the A reference operating point; on the contrary, when , it is judged as invalid measurement data and removed from the training set.

[0024] Further, performing residual dendritic neural network training according to the training set includes:

[0025] Replacing the residual dendritic neural network input layer parameter vector with the system state parameter historical data;

[0026] Replacing the residual dendritic neural network output layer vector with the node voltage amplitude;

[0027] Model training is performed according to the training set. During training, the linear change weight matrix and the residual weight matrix of the model are updated by back propagation. The back propagation calculation satisfies:

[0028] ,

[0029] ,

[0030] ,

[0031] ,

[0032] ,

[0033] ,

[0034] ,

[0035] in, For loss, is the learning rate, Single-layer module for the network The linear transfer part, and Residual dendritic neural network single layer module The input and output, is the original input vector of the residual dendritic neural network, It is Modules to The linear variation weight matrix of each module, It is Modules to The residual weight matrix of the modules, Denotes the Hadamard product.

[0036] Furthermore, the established explicit expression of node voltage is directly fitted by the residual dendritic neural network obtained by the last weight update, and the explicit expression of node voltage amplitude with respect to all state parameters of the system is constructed, including:

[0037] According to the training result of the residual dendritic neural network, the residual dendritic neural network obtained by the last weight update is fitted to obtain the explicit expression, which satisfies:

[0038] ,

[0039] in, is the voltage amplitude of the selected node; , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training, For All possible combinations of any two elements in For For all possible combinations of three elements in the node and nodes For Node Any two nodes outside.

[0040] Furthermore, the system state parameter of interest is selected as the principal element and the partial derivative is calculated to obtain an explicit expression of the node voltage sensitivity. The single system state parameter of the node is used as the principal element, and the partial derivative of the explicit expression is calculated to obtain a sensitivity expression of the node voltage amplitude with respect to the single system state parameter, including:

[0041] Select the required single system state parameter, substitute it into the explicit expression to find the partial derivative, and obtain the sensitivity expression of the single system state parameter, which satisfies:

[0042] ,

[0043] in, is the voltage amplitude of the selected node, is the selected single system state parameter, For Node About this status parameter The sensitivity of , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training, For All possible combinations of any two elements in For For all possible combinations of three elements in the node and nodes For Node Any two nodes outside.

[0044] The second purpose of the present invention is to provide a voltage sensitivity calculation device based on a residual network, which constructs a residual dendritic neural network model through historical data of system state parameters, and constructs a sensitivity expression of the voltage amplitude with respect to a single system state parameter based on the model, thereby obtaining the node voltage sensitivity.

[0045] The second object of the present invention is achieved by adopting the following technical solution:

[0046] A voltage sensitivity calculation device based on a residual network, comprising:

[0047] A construction module is used to obtain the node voltage amplitude and the historical data of the system state parameters of the node, and to construct a training set based on the obtained historical data;

[0048] A training module, used for training the residual dendritic neural network using the training set as an input value, and constructing an explicit expression of the node voltage amplitude with respect to all state parameters of the system according to the trained residual dendritic neural network model;

[0049] An expression generation module is used to calculate the partial derivative of the explicit expression with a single system state parameter of the node as the principal element to obtain a sensitivity expression of the node voltage amplitude with respect to the single system state parameter;

[0050] The sensitivity calculation module is used to obtain the real-time data of the system state parameters and substitute them into the sensitivity expression to calculate the node voltage sensitivity under the current grid topology.

[0051] The third object of the present invention is to provide an electronic device for performing one of the objects of the invention, which includes a processor, a storage medium and a computer program, wherein the computer program is stored in the storage medium, and when the computer program is executed by the processor, the above-mentioned voltage sensitivity calculation method based on the residual network is implemented.

[0052] A fourth object of the present invention is to provide a computer-readable storage medium storing one of the objects of the invention, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned voltage sensitivity calculation method based on the residual network is implemented.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The voltage sensitivity calculation method based on the residual network of the present invention adopts a data-driven approach as a whole, and only uses the historical measurement data of the system to realize the real-time calculation of the node voltage sensitivity, breaking through the limitation that the traditional sensitivity solution method needs to be accurately calculated under the conditions of clear distribution network topology and clear line impedance parameters; by collecting the historical operation data of the node voltage amplitude and the system state parameters, a training data set is constructed to carry out the training of the residual dendritic neural network, and the accuracy of the training data and the trained model can be guaranteed without obtaining the real-time system state parameter data; the residual dendritic neural network is used to complete the polynomial fitting of the power flow calculation, and the explicit expression of the node voltage amplitude with respect to the system state parameter is obtained, and then by selecting a single system state parameter as the principal element and calculating the partial derivative thereof, the explicit expression of the sensitivity of the node voltage with respect to each system state parameter is obtained, and the real-time measurement value of the system is substituted into the expression to obtain the real-time value of the sensitivity of the node voltage with respect to the selected state parameter. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 is a flow chart of a voltage sensitivity calculation method based on a residual network according to Embodiment 1;

[0056] Figure 2 is a schematic diagram of the residual dendritic neural network model structure of Example 1;

[0057] Figure 3 is a data screening flow chart of Embodiment 2;

[0058] Figure 4 is a grid topology diagram of the IEEE33 node calculation example of the third embodiment;

[0059] Figure 5 It is a summary diagram of the initial operating point parameters of the IEEE33 node calculation example of the third embodiment;

[0060] Figure 6 This is a theoretical value matrix diagram of voltage sensitivity of some nodes injected with active power in Example 3;

[0061] Figure 7 1. It is a matrix diagram of calculated values ​​of active power voltage sensitivity injected by some nodes in Embodiment 3;

[0062] Figure 8 1. It is a theoretical value matrix diagram of reactive power voltage sensitivity injected by some nodes of the third embodiment;

[0063] Fig. 9 1. It is a theoretical value matrix diagram of reactive power voltage sensitivity injected by some nodes of the third embodiment;

[0064] Fig.10 is a structural block diagram of a voltage sensitivity calculation device based on a residual network according to a fourth embodiment;

[0065] Fig.11 It is a structural block diagram of an electronic device of Embodiment 5. DETAILED DESCRIPTION

[0066] The present invention will be described in more detail below in conjunction with the accompanying drawings. It should be noted that the following description of the present invention with reference to the accompanying drawings is only illustrative and not restrictive. The various embodiments can be combined with each other to form other embodiments not shown in the following description.

[0067] Embodiment 1

[0068] Embodiment 1 provides a voltage sensitivity calculation method based on a residual network, which aims to utilize a large amount of historical operation data of system state parameters, construct a training data set through data screening, carry out residual dendritic neural network training, and construct a nonlinear mapping relationship between system state parameters and node voltage amplitudes, thereby obtaining an explicit expression of node voltage amplitude with respect to different system state parameters; then, the system state parameter of interest is selected as the principal element, the partial derivative of the node voltage expression is calculated, and an explicit expression of voltage sensitivity is obtained; finally, the system state parameters obtained by real-time measurement are substituted to realize rapid perception of node voltage sensitivity.

[0069] The node voltage sensitivity calculation of this embodiment needs to be considered based on unknown network topology and line parameters, and is only driven by historical measurement data to obtain real-time information.

[0070] This embodiment is an improvement based on the residual dendritic neural network. Compared with the traditional residual neural network, the biggest feature of the dendritic neural network is that the activation function between the traditional network layers is replaced by the Hadamard product. Specifically, the dendritic neural network (Dendrite Net, DD) involved in this embodiment is a new network architecture different from traditional cell neural networks such as CNN, RNN, and LSTM. The dendritic neural network focuses on simulating the information processing process of the neural dendrites. It uses the Hadamard product to replace the nonlinear activation function in the cell network. It has the advantages of controllable accuracy, low computational complexity, and not easy to overfit. It has achieved good results in the application verification of classification and regression problems. In the field of regression, the dendritic neural network is equivalent to using Taylor expansion to approximate the original function, but the high-order power obtained by the Hadamard product avoids the problem that Taylor expansion is difficult to obtain the derivative components of each order of the original function. Therefore, the use of dendritic neural networks for nonlinear regression is a hot topic in the field of engineering applications.

[0071] This embodiment adopts a data-driven approach and only uses historical system measurement data to achieve real-time calculation of node voltage sensitivity, breaking through the limitation that traditional sensitivity solution methods require clear distribution network topology and clear line impedance parameters to achieve accurate calculation.

[0072] Please refer to Figure 1 As shown, a voltage sensitivity calculation method based on a residual network is characterized by comprising the following steps:

[0073] S1, obtaining the node voltage amplitude and the historical data of the system state parameters of the node, and constructing a training set based on the obtained historical data;

[0074] The above-mentioned system state parameters can be selected according to the needs and specific system conditions. In this embodiment, the system state parameters are based on the two key parameters of the voltage control and management link, namely, active power and reactive power.

[0075] S2, using the training set as input value to perform residual dendritic neural network training, and constructing an explicit expression of node voltage amplitude with respect to all state parameters of the system according to the trained residual dendritic neural network model;

[0076] S2 trains the residual dendritic neural network based on the training set constructed by S1, fits the unified mapping between all state parameters of the system and the node voltage, and then establishes an explicit expression of the node voltage amplitude with respect to all state parameters of the system.

[0077] Please refer to Figure 2 As shown, the above residual dendritic neural network model structure is as follows:

[0078] ,

[0079] ,

[0080] ,

[0081] ,

[0082] in, is the original input vector of the residual dendritic neural network; Y is the final output vector of the residual dendritic neural network; and It is a single-layer network module Input and output; Single-layer module for the network The linear transfer part; It is Modules to The linear variation weight matrix of each module; It is Modules to The residual weight matrix of each module; Denotes the Hadamard product.

[0083] The training process of S2 includes:

[0084] Replacing the residual dendritic neural network input layer parameter vector with the system state parameter historical data;

[0085] Replacing the residual dendritic neural network output layer vector with the node voltage amplitude;

[0086] Model training is performed according to the training set. During training, the linear change weight matrix and the residual weight matrix of the model are updated by back propagation. The back propagation calculation satisfies:

[0087] ,

[0088] ,

[0089] ,

[0090] ,

[0091] ,

[0092] ,

[0093] ,

[0094] in, For loss, is the learning rate, Single-layer module for the network The linear transfer part, and Residual dendritic neural network single layer module The input and output, is the original input vector of the residual dendritic neural network, It is Modules to The linear variation weight matrix of each module, It is Modules to The residual weight matrix of the modules, Denotes the Hadamard product.

[0095] In S2, an explicit expression of the node voltage amplitude with respect to all state parameters of the system is constructed, including:

[0096] According to the training result of the residual dendritic neural network, the residual dendritic neural network obtained by the last weight update is fitted to obtain the explicit expression, which satisfies:

[0097] ,

[0098] in, is the voltage amplitude of the selected node; , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training. The elements in the linear variation weight matrix and the residual weight matrix corresponding to each layer module are obtained only through linear combination of addition and multiplication. For All possible combinations of any two elements in For For all possible combinations of three elements in the node and nodes For Node Any two nodes outside.

[0099] In this embodiment, a 4-layer residual dendritic neural network is used as an example to illustrate the polynomial coefficients The combined relationship between the linear change weight matrix and the residual weight matrix is ​​as follows:

[0100]

[0101]

[0102]

[0103]

[0104] in, For the The linear change weight matrix of the layer module Row, No. The element value of the column; For the The residual weight matrix of the layer module Row, No. The element value of the column.

[0105] When the system state parameter is selected to inject active and reactive power into each node, that is, The explicit expression will directly reflect the direct relationship between the node voltage and the injected power of each node. The above expression is only illustrated by taking a 4-layer residual dendritic neural network as an example. In practice, the number of network layers can be adjusted according to the accuracy requirements to obtain a higher-order polynomial approximation structure.

[0106] S3, taking the single system state parameter of the node as the principal element, calculating the partial derivative of the explicit expression, and obtaining the sensitivity expression of the node voltage amplitude with respect to the single system state parameter;

[0107] The above-mentioned single system state parameter can be selected according to the calculation requirements, and each system state parameter corresponds to a sensitivity expression, for example, active power and reactive power each correspond to a sensitivity expression.

[0108] S3 Specifically, this embodiment takes a 4-layer residual dendritic neural network as an example, takes a single system state parameter of a node as the principal element, calculates the partial derivative of the explicit expression, and obtains a sensitivity expression of the node voltage amplitude with respect to the single system state parameter, including:

[0109] Select the required single system state parameter, substitute it into the explicit expression to find the partial derivative, and obtain the sensitivity expression of the single system state parameter, which satisfies:

[0110] ,

[0111] in, is the voltage amplitude of the selected node, is the selected single system state parameter, For Node About this status parameter The sensitivity of , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training, For All possible combinations of any two elements in For All possible combinations of any three elements in .

[0112] The above description is based on a 4-layer residual dendritic neural network as an example. The number of network layers can actually be adjusted according to the accuracy requirements. This embodiment will not repeat the calculation of various numbers of layers.

[0113] S4. Obtain real-time data of system status parameters and substitute them into the sensitivity expression to calculate node voltage sensitivity under the current grid topology.

[0114] In summary, the method of this embodiment can train the residual dendritic neural network only through historical operation data without the need for accurate network topology and line parameters, and obtain an explicit expression of the sensitivity of the node voltage amplitude to the system state parameters in a data-driven manner. Taking advantage of the fact that the dendritic neural network calculation only includes Hadamard products and matrix multiplications, the computational complexity is significantly lower than that of traditional cell neural networks containing nonlinear functions, and can achieve fast calculation of node voltage sensitivity in microseconds, which can provide efficient guidance for voltage management in actual engineering.

[0115] Embodiment 2

[0116] The second embodiment is carried out on the basis of the first embodiment, and aims to increase the accuracy of the model through data screening.

[0117] Since the system operating status does not change much between adjacent measurement times, if the historical data is not screened, the neural network may not be able to accurately mine data features due to the small difference in the adjacent measurement data, or even fail to train the model due to matrix singularity when the data difference is too small. Therefore, it is necessary to set the operating status judgment threshold to preliminarily screen the data to ensure smooth model training.

[0118] Please refer to Figure 3 As shown, the historical data of the system state parameters are screened, and the training set is constructed based on the screened data. The data screening and training set construction process includes the following steps:

[0119] Step 1.1: Randomly select The system state parameters of each node at the time are taken as the reference operating point;

[0120] In step 1.1, in this embodiment, the active power and reactive power As a reference operating point;

[0121] Step 1.2: Get the first System status parameter historical data: , ;

[0122] Step 1.3: Calculate the The historical data of system status parameters and Operating point spacing ;

[0123] It should be noted that in step 1.1, , .

[0124] Step 1.4: Compare the distance with the operating status determination threshold:

[0125] If the distance is less than or equal to the running state determination threshold, , and return to step 1.2 to continue data screening;

[0126] Step 1.5: If the distance is greater than the running state determination threshold , the first The historical data of system status parameters is taken as the running points as the data in the training set, complete a data screening, and let , , return to step 1.2 to continue building the training set.

[0127] The calculation of the above data screening satisfies:

[0128] ,

[0129] ,

[0130] ,

[0131] ,

[0132] in, represents the set of all nodes in the network, ; , Node No. Historical injected active power and reactive power measurement data; , They are the nodes under the reference operating point The injected active power and reactive power measurement data; For the The historical measurement data and The distance between the reference operating points; is the rated capacity of the system; is the running state judgment threshold, which can be selected based on experience and actual conditions. For example, the threshold can be selected as 0.01; , it means that the historical measurement point is out of the reference operating point, so it is determined to be valid measurement data, used as the data in the training set, and used as the first A reference operating point; on the contrary, when , it indicates that the historical measurement point is still in The reference operating points are judged as invalid measurement data and removed from the training set.

[0133] Embodiment 3

[0134] Embodiment 3 is an experimental simulation performed according to the methods described in Embodiment 1 and Embodiment 2.

[0135] In this embodiment, in order to fully reflect the actual effect of the voltage sensitivity calculation method based on the residual network, the IEEE 33-node distribution network standard calculation example is used as a basis.

[0136] In this embodiment, the IEEE 33-node distribution network topology includes 33 nodes, 32 lines and 5 tie switches, the system voltage level is 12.66 kV, and the system base capacity is 1 MVA. Figure 4 The distribution network model shown in the figure has all five tie switches disconnected, node 1 is selected as the balancing node, and the remaining 32 nodes are PQ nodes. The initial active power and reactive power injected into the PQ node are as follows: Figure 5 After running the model, the voltage amplitude of each node in the system at the initial reference operating point can be obtained through power flow calculation simulation. To simulate historical operating data, the node is generated in the following way No. Active power and reactive power at historical measurement moments:

[0137] ,

[0138] ,

[0139] in, , Node Initial injected active power and reactive power; , Node No. Active power and reactive power at each historical measurement moment; , They are respectively normal distribution random numbers with a mean of 0 and a standard deviation of 0.1, which are used to simulate the normal fluctuation of node injection power at different historical measurement moments; , They are respectively normal distributed random numbers with a mean of 0 and a standard deviation of 0.025%, which are used to simulate the measurement error of the instrument.

[0140] The above method can generate a large amount of node injection power data, and the corresponding node voltage amplitude data can be obtained after power flow calculation. In this example, the precise grid topology and line parameters are only required to obtain simulation data. In actual engineering, measured data can be used directly, so precise grid topology and line parameters are not required.

[0141] The generated active power and reactive power injected into each node are used as system state variables. Through the method proposed in the present invention, the sensitivity of the node voltage amplitude to the active power and reactive power injected into each node can be calculated. Figure 6 Theoretical values ​​of active power voltage sensitivity matrix injected into some nodes, Figure 7 Inject the calculated values ​​of active power voltage sensitivity matrix into some nodes; Figure 8 Theoretical value of reactive power voltage sensitivity matrix injected into some nodes, Fig. 9 The calculated value of the reactive power voltage sensitivity matrix is ​​injected into some nodes. OK The values ​​in the columns represent The voltage of the node corresponding to the row The columns correspond to the sensitivity of the nodes to inject active / reactive power. It can be seen that the node voltage sensitivity calculated by the method described in the present invention is 0.94% at the selected time section, and the average relative error is 0.49%; the maximum relative error of the reactive power voltage sensitivity matrix is ​​0.46%, and the average relative error is 0.27%. It can be seen that the method described in the present invention can accurately perceive the node voltage sensitivity.

[0142] Embodiment 4

[0143] Embodiment 4 discloses a device corresponding to the voltage sensitivity calculation method based on the residual network of the above embodiment, which is a virtual device structure of the above embodiment. Please refer to Fig.10 As shown, including:

[0144] A construction module 210 is used to obtain the node voltage amplitude and the historical data of the system state parameters of the node, and to construct a training set based on the obtained historical data;

[0145] A training module 220 is used to train the residual dendritic neural network using the training set as an input value, and to construct an explicit expression of the node voltage amplitude with respect to all state parameters of the system according to the trained residual dendritic neural network model;

[0146] An expression generation module 230 is used to calculate the partial derivative of the explicit expression with a single system state parameter of the node as the principal element to obtain a sensitivity expression of the node voltage amplitude with respect to the single system state parameter;

[0147] The sensitivity calculation module 240 is used to obtain the real-time data of the system state parameters and substitute it into the sensitivity expression to calculate the node voltage sensitivity under the current grid topology.

[0148] Preferably, the system status parameters include active power and reactive power.

[0149] Preferably, the system state parameter historical data is screened, and the training set is constructed based on the screened data. The data screening and training set construction process includes the following steps:

[0150] Step 1.1: Randomly select The system state parameters of each node at the time are taken as the reference operating point;

[0151] Step 1.2: Get the first System status parameter historical data;

[0152] Step 1.3: Calculate the The historical data of system status parameters and Operating point spacing ;

[0153] Step 1.4: Compare the distance with the operating status determination threshold:

[0154] If the distance is less than or equal to the running state determination threshold, , and return to step 1.2 to continue data screening;

[0155] Step 1.5: If the distance is greater than the running state judgment threshold, The historical data of system status parameters is taken as the running points as the data in the training set, complete a data screening, and let , , return to step 1.2 to continue building the training set.

[0156] Preferably, the calculation of data screening satisfies:

[0157] ,

[0158] ,

[0159] ,

[0160] ,

[0161] in, represents the set of all nodes in the network, ; , Node No. Historical injected active power and reactive power measurement data; , They are the nodes under the reference operating point The injected active power and reactive power measurement data; For the The historical measurement data and The distance between the reference operating points; is the rated capacity of the system; is the running status judgment threshold; when When , it is determined to be valid measurement data, used as the data in the training set, and used as the A reference operating point; on the contrary, when , it is judged as invalid measurement data and removed from the training set.

[0162] Preferably, performing residual dendritic neural network training according to the training set includes:

[0163] Replacing the residual dendritic neural network input layer parameter vector with the system state parameter historical data;

[0164] Replacing the residual dendritic neural network output layer vector with the node voltage amplitude;

[0165] Model training is performed according to the training set. During training, the linear change weight matrix and the residual weight matrix of the model are updated by back propagation. The back propagation calculation satisfies:

[0166] ,

[0167] ,

[0168] ,

[0169] ,

[0170] ,

[0171] ,

[0172] ,

[0173] in, For loss, is the learning rate, Single-layer module for the network The linear transfer part, and Residual dendritic neural network single layer module The input and output, is the original input vector of the residual dendritic neural network, It is Modules to The linear variation weight matrix of each module, It is Modules to The residual weight matrix of the modules, Denotes the Hadamard product.

[0174] Preferably, an explicit expression of the node voltage amplitude with respect to all state parameters of the system is constructed, including:

[0175] According to the training result of the residual dendritic neural network, the residual dendritic neural network obtained by the last weight update is fitted to obtain the explicit expression, which satisfies:

[0176] ,

[0177] in, is the voltage amplitude of the selected node; , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training, For All possible combinations of any two elements in For All possible combinations of any three elements in .

[0178] Preferably, a single system state parameter of the node is used as the principal element, and the partial derivative of the explicit expression is calculated to obtain a sensitivity expression of the node voltage amplitude with respect to the single system state parameter, including:

[0179] Select the required single system state parameter, substitute it into the explicit expression to find the partial derivative, and obtain the sensitivity expression of the single system state parameter, which satisfies:

[0180] ,

[0181] in, is the voltage amplitude of the selected node, is the selected single system state parameter, For Node About this status parameter The sensitivity of , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training, For All possible combinations of any two elements in For All possible combinations of any three elements in .

[0182] Embodiment 5

[0183] Fig.11 A schematic diagram of the structure of an electronic device provided in Embodiment 5 of the present invention is shown in FIG. Fig.11 As shown, the electronic device includes a processor 310, a memory 320, an input device 330 and an output device 340; the number of processors 310 in the computer device can be one or more. Fig.11 A processor 310 is taken as an example; the processor 310, the memory 320, the input device 330 and the output device 340 in the electronic device can be connected via a bus or other means. Fig.11 The example of connecting through bus is taken in the following.

[0184] The memory 320, as a computer-readable storage medium, can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the voltage sensitivity calculation method based on the residual network in the embodiment of the present invention (for example, the construction module 210, the training module 220, the expression generation module 230 and the sensitivity calculation module 240 in the voltage sensitivity calculation device based on the residual network). The processor 310 executes various functional applications and data processing of the electronic device by running the software programs, instructions and modules stored in the memory 320, that is, implements the voltage sensitivity calculation method based on the residual network in the above-mentioned embodiments 1 to 3.

[0185] The memory 320 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 320 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 320 may further include a memory remotely arranged relative to the processor 310, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0186] The input device 330 may be used to receive input user identity information, system status parameter historical data, etc. The output device 340 may include a display device such as a display screen.

[0187] Embodiment 6

[0188] Embodiment 7 of the present invention further provides a storage medium containing computer executable instructions, which can be used for a computer to execute a voltage sensitivity calculation method based on a residual network, the method comprising:

[0189] Obtain the node voltage amplitude and the historical data of the system state parameters of the node, and build a training set based on the acquired historical data;

[0190] The training set is used as an input value to train a residual dendritic neural network, and an explicit expression of the node voltage amplitude with respect to all state parameters of the system is constructed according to the trained residual dendritic neural network model;

[0191] Taking a single system state parameter of a node as a principal element, calculating the partial derivative of the explicit expression, and obtaining a sensitivity expression of the node voltage amplitude with respect to the single system state parameter;

[0192] The real-time data of system state parameters are obtained and substituted into the sensitivity expression to calculate the node voltage sensitivity under the current grid topology.

[0193] Of course, the computer executable instructions of a storage medium including computer executable instructions provided in an embodiment of the present invention are not limited to the method operations described above, and can also execute related operations in the voltage sensitivity calculation method based on the residual network provided in any embodiment of the present invention.

[0194] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for an electronic device (which can be a mobile phone, a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0195] It is worth noting that in the above-mentioned embodiment of the voltage sensitivity calculation device based on the residual network, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0196] For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of the present invention.

Claims

1. A voltage sensitivity calculation method based on residual network, characterized in that: The following steps are involved: Obtain the node voltage amplitude and the historical data of the system state parameters of the node, and build a training set based on the acquired historical data; The training set is used as an input value to train the residual dendritic neural network, and an explicit expression of the node voltage amplitude with respect to all state parameters of the system is constructed according to the trained residual dendritic neural network model; wherein the residual dendritic neural network uses the Hadamard product to replace the activation function between the residual neural network layers; and an explicit expression of the node voltage amplitude with respect to all state parameters of the system is constructed, including: According to the training result of the residual dendritic neural network, the residual dendritic neural network obtained by the last weight update is fitted to obtain an explicit expression of the node voltage amplitude with respect to all state parameters of the system, and the explicit expression satisfies: ， in, is the voltage amplitude of the selected node; , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training, For All possible combinations of any two elements in For For all possible combinations of three elements in the node and nodes For Node Any two nodes outside Taking a single system state parameter of a node as a principal element, calculating the partial derivative of the explicit expression, and obtaining a sensitivity expression of the node voltage amplitude with respect to the single system state parameter; The real-time data of system state parameters are obtained and substituted into the sensitivity expression to calculate the node voltage sensitivity under the current grid topology.

2. The voltage sensitivity calculation method based on residual network according to claim 1, characterized in that: The system status parameters include active power and reactive power.

3. The voltage sensitivity calculation method based on the residual network according to claim 1 or 2, characterized in that: The system state parameter historical data is screened, and the training set is constructed based on the screened data. The data screening and training set construction process includes the following steps: Step 1.1: Randomly select The system state parameters of each node at the time are taken as the reference operating point; Step 1.2: Get the first System status parameter historical data; Step 1.3: Calculate the The historical data of system status parameters and Operating point spacing ; Step 1.4: Compare the distance with the operating status determination threshold: If the distance is less than or equal to the running state determination threshold, , and return to step 1.2 to continue data screening; Step 1.5: If the distance is greater than the running state judgment threshold, The historical data of system status parameters is taken as the running points as the data in the training set, complete a data screening, and let , , return to step 1.2 to continue building the training set.

4. The voltage sensitivity calculation method based on residual network according to claim 3, characterized in that: The calculation of the data screening satisfies: ， ， ， ， in, represents the set of all nodes in the network, ; , Node No. Historical injected active power and reactive power measurement data; , They are the nodes under the reference operating point The injected active power and reactive power measurement data; For the The historical measurement data and The distance between the reference operating points; is the rated capacity of the system; is the running status judgment threshold; when When , it is determined to be valid measurement data, used as the data in the training set, and used as the A reference operating point; on the contrary, when , it is judged as invalid measurement data and removed from the training set.

5. The voltage sensitivity calculation method based on residual network according to claim 1, characterized in that: Performing residual dendritic neural network training according to the training set includes: Replacing the residual dendritic neural network input layer parameter vector with the system state parameter historical data; Replacing the residual dendritic neural network output layer vector with the node voltage amplitude; Model training is performed according to the training set. During training, the linear change weight matrix and the residual weight matrix of the model are updated by back propagation. The back propagation calculation satisfies: ， ， ， ， ， ， ， in, For loss, is the learning rate, Single-layer module for the network The linear transfer part, and Residual dendritic neural network single layer module The input and output, is the original input vector of the residual dendritic neural network, It is Modules to The linearly changing weight matrix of each module, It is Modules to The residual weight matrix of each module, Denotes the Hadamard product.

6. The voltage sensitivity calculation method based on residual network according to claim 1, characterized in that: Taking the single system state parameter of the node as the principal element, the partial derivative of the explicit expression is calculated to obtain the sensitivity expression of the node voltage amplitude with respect to the single system state parameter, including: Select the required single system state parameter, substitute it into the explicit expression to find the partial derivative, and obtain the sensitivity expression of the single system state parameter, which satisfies: ， in, is the voltage amplitude of the selected node, is the selected single system state parameter, For Node About this status parameter The sensitivity of , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training, For All possible combinations of any two elements in For For all possible combinations of three elements in the node and nodes For Node Any two nodes outside.

7. A voltage sensitivity calculation device based on a residual network, characterized in that: It includes: A construction module is used to obtain the node voltage amplitude and the historical data of the system state parameters of the node, and to construct a training set based on the obtained historical data; A training module is used to train the residual dendritic neural network using the training set as an input value, and to construct an explicit expression of the node voltage amplitude with respect to all state parameters of the system according to the trained residual dendritic neural network model; wherein the residual dendritic neural network uses Hadamard product to replace the activation function between the residual neural network layers; constructing an explicit expression of the node voltage amplitude with respect to all state parameters of the system, including: According to the training result of the residual dendritic neural network, the residual dendritic neural network obtained by the last weight update is fitted to obtain an explicit expression of the node voltage amplitude with respect to all state parameters of the system, and the explicit expression satisfies: ， in, is the voltage amplitude of the selected node; , is the initial bias parameter of the network and system status parameters A collection of where the network initial bias parameters The value is 1; are the polynomial coefficients obtained through residual dendritic neural network training, For All possible combinations of any two elements in For For all possible combinations of three elements in the node and nodes For Node Any two nodes outside An expression generation module is used to calculate the partial derivative of the explicit expression with a single system state parameter of the node as the principal element to obtain a sensitivity expression of the node voltage amplitude with respect to the single system state parameter; The sensitivity calculation module is used to obtain the real-time data of the system state parameters and substitute them into the sensitivity expression to calculate the node voltage sensitivity under the current grid topology.

8. An electronic device comprising a processor, a storage medium and a computer program, wherein the computer program is stored in the storage medium, wherein: When the computer program is executed by a processor, the voltage sensitivity calculation method based on the residual network according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the voltage sensitivity calculation method based on the residual network according to any one of claims 1 to 6 is implemented.

Images