Primary and distribution cooperative robust state estimation method, device and system and storage medium

Abstract

The invention belongs to the technical field of power systems, and discloses a main distribution cooperative robust state estimation method, device and system and a storage medium, and the method comprises the steps: obtaining historical measurement data, and carrying out the preprocessing of the historical measurement data; and training a pre-constructed convolutional neural network model by using the preprocessed historical measurement data, calculating a measurement residual error by using a tidal current power equation, using a weighted minimum residual error quadratic sum as a loss function, performing back propagation according to the loss function to update parameters of the convolutional neural network model, and performing continuous training to obtain a neural network model. Obtaining a trained convolutional neural network model; and carrying out online estimation by adopting the trained convolutional neural network model, and determining a main-distribution cooperative robust state estimation result. According to the method, training of the convolutional neural network model is converted from a supervised learning problem to a weak supervised learning problem, a large amount of label data is not needed to serve as the premise of model training, and the method has good estimation precision and robust capability.

Description

Technical Field

[0001] The invention relates to a method, device, system and storage medium for estimating a main-distribution coordinated anti-difference state, and belongs to the technical field of power systems. Background Art

[0002] The state estimation (SE) problem of power grids refers to the use of noisy measurements to calculate the current operating state of the power grid. SE is an important inference tool that can provide advanced control and automation functions for power grids. For example, voltage control during normal operation and reconfiguration of feeders during emergency restoration. In the fully observable case, that is, when the number of unknown variables is less than the number of available measurements, the weighted least squares (WLS) method has been well developed and widely used in power grids. When the number of unknown variables is greater than the number of measurements, increasing the number of measurements through pseudo-measurement modeling is a typical practice in distribution networks.

[0003] Traditional methods based on WLS are widely used in power system state estimation. However, in the actual main and distribution integrated network, there are a lot of non-Gaussian noise and gross errors, which will make the weighted least squares method difficult to converge, the accuracy will decrease, and it will be difficult to cope with the complex operating state and random fluctuations caused by the widespread access of new energy. Based on this, a robust state estimation method that can handle various noises, interferences and uncertainties has been proposed. Although robust state estimation can compensate and correct errors and has good anti-error capabilities, it requires the design of suitable algorithms for different systems and environments, so the design is relatively difficult. In addition, robust state estimation usually requires the introduction of non-quadratic estimation criteria, which requires a lot of computing resources and time, and is difficult to adapt to the actual requirements of the main and distribution integrated network. In response to the above problems, machine learning (ML) and signal processing tools have been used for distribution system state estimation and have achieved extensive development, but most of the existing ML-based methods require a large number of labels as training samples, which is often unrealistic in power systems. Summary of the invention

[0004] The purpose of the present invention is to provide a main-distribution coordinated robustness state estimation method, device, system and storage medium. By using the power flow equation to calculate the measurement residual and using the weighted minimum residual square sum as the loss function, the training of the convolutional neural network model is transformed from a supervised learning problem to a weakly supervised learning problem. A large amount of labeled data is no longer required as a prerequisite for model training, and the model has good estimation accuracy and robustness.

[0005] In order to achieve the above objectives / solve the above technical problems, the present invention is implemented by adopting the following technical solutions.

[0006] In one aspect, the present invention provides a method for estimating a main-distribution coordinated robustness state, comprising: Obtain historical measurement data in the main distribution coordination network and pre-process the historical measurement data; The pre-processed historical measurement data is used to train the pre-built convolutional neural network model. The measurement residual is calculated using the power flow equation. The weighted minimum residual square sum is used as the loss function. The parameters of the convolutional neural network model are updated by back propagation according to the loss function. The training is continued to obtain the trained convolutional neural network model. The trained convolutional neural network model is used for online estimation to determine the estimation results of the main and distribution coordinated robustness state.

[0007] Optionally, the historical measurement data includes: active power output by the generator, reactive power output by the generator, active load power of the node, reactive load power of the node and voltage amplitude of the node; The preprocessing of the historical measurement data includes: calculating the active injection power of the node and the reactive injection power of the node through the active power output by the generator, the reactive power output by the generator, the active load power of the node and the reactive load power of the node, and the formula is as follows: ; In the formula, P i For the i The active injection power of each node is P iG For the i The active power output of the generator of each node, P iL For the i Active load power of each node; Q i For the i The reactive injection power of each node is Q iG For the i The reactive power output of the generator of each node, Q iL For the i The reactive load power of each node.

[0008] Optionally, the network structure of the convolutional neural network model includes a convolutional layer, a pooling layer and a fully connected layer; The calculation formula of the convolutional layer is as follows: ; In the formula, For the l The output of the layer kFeatures, For the l Layer k The weight vector corresponding to the features, X l-1 For the l -1 layer output, For the l Layer k The learnable bias corresponding to each feature, f (·) is the nonlinear activation function of the convolutional layer; The calculation formula of the pooling layer is as follows: ; In the formula, For the q The value after pooling of the pooling area, is the first p unpooled values, D q For the q Pooling area; The calculation formula of the fully connected layer is as follows: ; In the formula, For the l Tier k The output of the convolutional layer after activation corresponding to the feature, g (·) is the activation function of the fully connected layer.

[0009] Optionally, the parameters of the convolutional neural network model include fixed parameters and variable parameters. The fixed parameters include the number of neurons in the convolutional layer and the fully connected layer, the sample size used for each batch of model training, and the number of training passes for all samples. The variable parameters include the weight vectors corresponding to the features of each layer and the learnable bias.

[0010] Optionally, the method for training the convolutional neural network model includes: The voltage amplitude of the node, the active injection power of the node and the reactive injection power of the node in the preprocessed historical measurement data are used as the input of the pre-built convolutional neural network model, and the estimated value of the voltage amplitude of the node and the estimated value of the voltage phase angle of the node are used as the output of the convolutional neural network model. At the same time, the power flow equation is used to calculate the measurement residual, and the weighted minimum residual sum of squares is used as the loss function. The variable parameters of the convolutional neural network model are updated by back propagation according to the loss function. Training is continued until the number of training passes for all samples is equal to the preset value, and a trained convolutional neural network model is obtained.

[0011] Optionally, the method of calculating the measurement residual using the power flow equation and using the weighted minimum residual square sum as the loss function includes: For having n nodes, m The loss function of the convolutional neural network model is As shown below: ; In the formula, for The estimated value of the state of the dimension, Contains voltage magnitude estimates and voltage phase angle estimate , z for dimensional measurement vector, h (·) for contact and z A nonlinear function of R is the covariance matrix of the measurement error, R is a diagonal matrix with the first u The element is v The measurement variance corresponding to the measurement value ,in, u = v ; z The expression is: ; In the formula, U , P , Q are the voltage amplitude of the node, the active injection power of the node and the reactive injection power of the node in the preprocessed historical measurement data, respectively. T Represents the transpose of a matrix; exist In the above equation, the active power injected by the node and the reactive power injected by the node are The nonlinear function relationship, that is, the expression of the power flow equation is as follows: ; ; In the formula, and Respectively i Nodes and j The estimated voltage amplitude of each node, For the i Nodes and j The difference in the estimated voltage phase angle of the nodes is For the i The self-conductance of the node, For the i Nodes and j The mutual conductance between nodes is For the i The self-susceptance of each node, For the i Nodes and j The mutual admittance between nodes; The measurement residual is , so the loss function The expression is transformed as follows: ; In the formula, z v For the v A measurement value, For z v The corresponding value calculated by the nonlinear function.

[0012] Optionally, the method of performing online estimation using the trained convolutional neural network model includes: Calculate the active injection power and reactive injection power of the node according to the real-time measured active power output by the generator, the reactive power output by the generator, the active load power of the node, and the reactive load power of the node; The voltage amplitude of the node, the active injection power of the node and the reactive injection power of the node in real-time measurement are used as the input of the trained convolutional neural network model to obtain the estimated value of the voltage amplitude of the node and the estimated value of the voltage phase angle of the node.

[0013] In a second aspect, the present invention provides a main-distribution coordinated robustness state estimation device, comprising: The data acquisition and preprocessing module is configured to acquire historical measurement data in the main distribution coordination network and preprocess the historical measurement data; The model training module is configured to train the pre-built convolutional neural network model using the pre-processed historical measurement data, and calculate the measurement residual using the power flow equation, use the weighted minimum residual square sum as the loss function, and update the parameters of the convolutional neural network model by back propagation according to the loss function, and continuously train to obtain the trained convolutional neural network model; The state estimation module is configured to use the trained convolutional neural network model to perform online estimation to determine the main and distribution coordinated robustness state estimation results.

[0014] In a third aspect, the present invention provides a computer system, comprising: Memory for storing computer programs; A processor is used to execute the computer program to implement the steps of the main and distribution coordinated anti-difference state estimation method described in the first aspect.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the main-distribution coordinated anti-difference state estimation method described in the first aspect.

[0016] Compared with the prior art, the present invention has the following beneficial effects: The present invention trains a pre-constructed convolutional neural network model by using pre-processed historical measurement data, and calculates the measurement residual using the power flow equation, uses the weighted minimum residual square sum as the loss function, and updates the parameters of the convolutional neural network model by back propagation according to the loss function, continuously trains to obtain a trained convolutional neural network model, and then uses the trained convolutional neural network model for online estimation to determine the main distribution coordinated robustness state estimation result; the main distribution coordinated robustness state estimation method provided by the present invention is a physics-based machine learning method, which can use the physical structure of the problem to improve the performance of the standard machine learning algorithm; wherein, the measurement residual is calculated using the power flow equation and the weighted minimum residual square sum is used as the loss function, so that the training of the convolutional neural network model is transformed from a supervised learning problem to a weakly supervised learning problem, and a large amount of labeled data is no longer required as a prerequisite for model training, which is more in line with the actual conditions of the power grid; at the same time, the modification of the loss function reduces the size of the convolutional neural network search space, which in turn helps to better converge during the training process, and has good estimation accuracy and robustness; the method proposed in the present invention can be regarded as a physics-based machine learning method, which uses the physical structure of the problem to improve the performance of the standard machine learning algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 The figure is a schematic diagram of the flow of the main and distribution coordinated robustness state estimation method in an embodiment of the present invention; Figure 2 This is a schematic diagram of a main-distribution integrated network constructed based on the IEEE30 standard transmission network test network and the IEEE33 standard distribution network test network of the present invention; Figure 3 This is a diagram showing the voltage amplitude state estimation result under white noise conditions of the present invention; Figure 4 This is a diagram showing the voltage phase angle state estimation result under white noise conditions of the present invention; Figure 5 This is a diagram showing the voltage amplitude state estimation result under bad data conditions of the present invention; Figure 6 This is a diagram of the voltage phase angle state estimation result under bad data conditions of the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.

[0019] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship.

[0020] Example 1

[0021] like Figure 1 As shown, this embodiment introduces a method for estimating the main and distribution coordinated robustness state, and the steps are as follows: Step S1: Acquire historical measurement data in the main distribution coordination network and pre-process the historical measurement data.

[0022] The historical measurement data include: active power output by the generator, reactive power output by the generator, active load power of the node, reactive load power of the node and voltage amplitude of the node.

[0023] In this embodiment, historical measurement data of 8760 time sections are obtained, the sampling interval is 1 hour, and the 8760 time sections are exactly one year of historical measurement data.

[0024] The preprocessing of the historical measurement data includes: calculating the active injection power of the node and the reactive injection power of the node through the active power output by the generator, the reactive power output by the generator, the active load power of the node and the reactive load power of the node, and the formula is as follows: ; In the formula, P i For the i The active injection power of each node is P iG For the i The active power output of the generator of each node, P iL For the i Active load power of each node; Q i For the i The reactive injection power of each node is Q iG For the iThe reactive power output of the generator of each node, Q iL For the i The reactive load power of each node.

[0025] The preprocessed historical measurement data is divided into a training set and a test set according to a certain ratio. In this embodiment, the preprocessed historical measurement data is divided into a training set and a test set according to a ratio of 15:1.

[0026] Step S2: Use the preprocessed historical measurement data to train the pre-constructed convolutional neural network model, and use the power flow equation to calculate the measurement residual, use the weighted minimum residual square sum as the loss function, and update the parameters of the convolutional neural network model by back propagation according to the loss function, and continue training to obtain a trained convolutional neural network model.

[0027] Step S2.1, using the node voltage amplitude, the node active injection power and the node reactive injection power in the preprocessed historical measurement data as the input of the pre-constructed convolutional neural network model, and using the node voltage amplitude estimation value and the node voltage phase angle estimation value as the output of the convolutional neural network model.

[0028] The network structure of the Convolutional Neural Networks (CNN) model includes a convolutional layer, a pooling layer, and a fully connected layer. The parameters of the convolutional neural network model include fixed parameters and variable parameters. The fixed parameters include the number of neurons in the convolutional layer and the fully connected layer, the sample size used for each batch of model training, and the number of training passes for all samples. The variable parameters include the weight vectors corresponding to the features of each layer and the learnable bias.

[0029] The convolution layer uses the convolution kernel to extract local features from the input measurement and maps the features using a nonlinear activation function. That is, the convolution layer is used to extract effective features of the measurement data. The calculation formula is as follows: ; In the formula, For the l The output of the layer k Features, For the l Layer k The weight vector corresponding to the features, X l-1 For the l -1 layer output, For the l Layer k The learnable bias corresponding to each feature, f (·) is the nonlinear activation function of the convolutional layer.

[0030] The pooling layer is used to reduce the feature dimension while retaining important feature information. This embodiment uses the maximum pooling method to process the pooling layer, and the calculation formula of the pooling layer is as follows: ; In the formula, For the q The value after pooling of the pooling area, is the first p unpooled values, p Represents the first p values, D q For the q pooling area.

[0031] The fully connected layer is used to integrate the extracted features, and its calculation formula is as follows: ; In the formula, For the l Tier k The output of the convolutional layer after activation corresponding to the feature, g (·) is the activation function of the fully connected layer, and the Sigmoid function is selected in this embodiment.

[0032] The fully connected layer passes the calculation results of the multi-level neural network to the output layer and obtains the final power system state estimation value.

[0033] In this embodiment, the sample size used for each batch of training in the convolutional neural network model parameters is set to 32, and the number of training passes for all samples is set to 45. The settings of other parameters of each layer of the convolutional neural network model are shown in Table 1 below: Table 1. Parameter settings of each layer of the convolutional neural network model

[0034] Step S2.2: Calculate the measurement residual using the power flow equation, and use the weighted minimum residual square sum as the loss function.

[0035] For having n nodes, m The loss function of the convolutional neural network model is As shown below: ; In the formula, for The estimated value of the state of the dimension, Contains voltage magnitude estimates and voltage phase angle estimate , z for dimensional measurement vector, h (·) for contact and z A nonlinear function of R is the covariance matrix of the measurement error, R is a diagonal matrix with the first u The element is v The measurement variance corresponding to the measurement value ,in, u = v .

[0036] z The expression is: ; In the formula, U , P , Q are the voltage amplitude of the node, the active injection power of the node and the reactive injection power of the node in the preprocessed historical measurement data, respectively. T Represents the transpose of a matrix.

[0037] exist In the above equation, the active power injected by the node and the reactive power injected by the node are The nonlinear function relationship, that is, the expression of the power flow equation is as follows: ; ; In the formula, and Respectively i Nodes and j The estimated voltage amplitude of each node, For the i Nodes and j The difference in the estimated voltage phase angle of the nodes is For the i The self-conductance of the node, For the i Nodes and j The mutual conductance between nodes is For the i The self-susceptance of each node, For the i Nodes and j The mutual admittance between nodes.

[0038] The measurement residual is , then v The specific expression of the measurement residual corresponding to the measurement value is: , so the loss function The expression is transformed as follows: ; In the formula, z v For the v A measurement value, For z v The corresponding value calculated by the nonlinear function.

[0039] Step S2.3: Update the variable parameters of the convolutional neural network model according to the back propagation of the loss function, and continue training until the number of training passes for all samples is equal to the preset value, so as to obtain a trained convolutional neural network model.

[0040] Step S3: Use the trained convolutional neural network model to perform online estimation to determine the main and distribution coordinated robustness state estimation result. Specifically: Calculate the active injection power and reactive injection power of the node according to the real-time measured active power output by the generator, the reactive power output by the generator, the active load power of the node, and the reactive load power of the node; The voltage amplitude of the node, the active injection power of the node and the reactive injection power of the node in real-time measurement are used as the input of the trained convolutional neural network model to obtain the estimated value of the voltage amplitude of the node and the estimated value of the voltage phase angle of the node.

[0041] Example 2

[0042] In order to verify the superiority of the main and distribution coordinated robust state estimation method provided in Example 1, a test was conducted on a main and distribution integrated network constructed by the IEEE30 standard transmission network test network and the IEEE33 standard distribution network test network.

[0043] like Figure 2 As shown in the figure, the IEEE 30-node system is a transmission network system. The 24th, 26th, and 29th nodes of the transmission network are selected as the boundary nodes connected to the distribution network, and are respectively connected to the IEEE 33-node distribution network. The root node of the distribution network coincides with the load node of the transmission network. In order not to change the original power flow of the transmission network as much as possible, the power transmitted to the distribution network needs to be close to the original load of the transmission network. Therefore, the overall load of the distribution network needs to be multiplied by a factor K At the same time, the load of the connected node is removed, and the load of the node in the transmission network is regarded as the power transmitted to the distribution network.

[0044] K Value selection: The load power at the grid connection point is defined as S L = PL + Jb L , where P L is the active load power at the grid connection point, Q L is the reactive load power at the transmission grid connection point; the root node generator power in the IEEE33-node distribution network power flow result is defined as S G = P G + Jb G , where P G is the active power of the root node generator in the distribution network power flow result, Q G It is the reactive power of the root node generator in the power flow result of the distribution network. At this time, the active power and reactive power are defined as a multiple respectively. K p and K q , K p = P L / P G , K q = Q L / Q G ; Consider mainly to meet the active power matching, K = 0.7 K p +0.3 K p .

[0045] In order to make the comparison between the method of the present invention and the state estimation result based on WLS more intuitive, this embodiment adopts the mean absolute evaluation error e r, mean and the maximum absolute evaluation error e r, max Specifically, the following formulas are used to calculate these indicators: ; ; In the formula, For the t Section r The estimated value of the state quantity, x r ( t ) is the t Sectionr The truth value of a state variable, M is the number of test set sections.

[0046] (1) Performance test when the system measurement noise is Gaussian white noise.

[0047] Figure 3 and Figure 4 The absolute error obtained by comparing the estimated results of the voltage amplitude and voltage phase angle at the node 16 with the true value of the power flow using the method proposed in the present invention and the method based on WLS when the system measurement noise is Gaussian white noise is shown.

[0048] In order to quantitatively analyze the estimation effects of different algorithms in the main distribution integrated network, this embodiment uses the mean absolute estimation error e r, mean and the maximum absolute evaluation error e r, max Compare various algorithms. See Table 2 for details: Table 2 State estimation result indicators under different algorithms

[0049] Depend on Figure 3 , Figure 4 It can be seen from Table 2 that when the system measurement noise is Gaussian white noise, the node voltage amplitude and node voltage phase angle estimation accuracy of the method of the present invention is significantly higher than that of the WLS estimation. Taking node 16 as an example, the average absolute error of the node voltage amplitude and node voltage phase angle of the estimation method proposed by the present invention is reduced by 78.01% and 66.01% respectively compared with WLS; the maximum absolute error of the node voltage amplitude and node voltage phase angle is reduced by 88.68% and 84.57% respectively compared with WLS. Therefore, it can be concluded that when the system measurement noise is Gaussian white noise, the main and distribution coordinated anti-differential state estimation method proposed by the present invention has a higher estimation accuracy than the traditional WLS.

[0050] (2) Performance testing when the system measurement contains bad data.

[0051] Since bad data is inevitable in power system measurements, it is of great practical significance to test the estimation performance of the state estimation algorithm when bad data is included in the system measurements. The robustness test of this embodiment is achieved by adding 5 bad data to the partial power injection measurement. The power injection measurements with bad data added are: active injection of nodes 15, 20, 42, 64, and 91, and reactive injection of nodes 37 and 107.

[0052] Figure 5 and Figure 6The present invention shows the main distribution coordinated anti-difference state estimation method and WLS method when the system measurement contains bad data. The absolute error is obtained by comparing the estimated results of the voltage amplitude and voltage phase angle at the node 16 with the true value of the power flow.

[0053] The average absolute estimation error and maximum absolute estimation error of each algorithm are shown in Table 3: Table 3 State estimation result indicators under different algorithms

[0054] Depend on Figure 5 , Figure 6 It can be seen from Table 3 that when the system measurement contains bad data, the node voltage amplitude and node voltage phase angle estimation accuracy of the method of the present invention is significantly higher than that of the WLS estimation. Taking node 16 as an example, the average absolute error of the node voltage amplitude and node voltage phase angle using the estimation method proposed in the present invention is reduced by 56.23% and 73.71% respectively compared with WLS, and the maximum absolute error of the node voltage amplitude and node voltage phase angle is reduced by 78.04% and 88.21% respectively compared with WLS. Therefore, it can be concluded that when the system measurement contains bad data, the main and distribution coordinated anti-difference state estimation method proposed in the present invention has better anti-difference capability than the traditional WLS.

[0055] Example 3

[0056] Based on the same inventive concept as that of Example 1, this embodiment introduces a main and distribution coordinated robustness state estimation device, including: The data acquisition and preprocessing module is configured to acquire historical measurement data in the main distribution coordination network and preprocess the historical measurement data; The model training module is configured to train the pre-built convolutional neural network model using the pre-processed historical measurement data, and calculate the measurement residual using the power flow equation, use the weighted minimum residual square sum as the loss function, and update the parameters of the convolutional neural network model by back propagation according to the loss function, and continuously train to obtain the trained convolutional neural network model; The state estimation module is configured to use the trained convolutional neural network model to perform online estimation to determine the main and distribution coordinated robustness state estimation results.

[0057] The specific functions of the above modules are implemented by referring to the relevant contents of the method in Example 1, which will not be repeated here. Example 4

[0058] Based on the same inventive concept as Embodiment 1, this embodiment introduces a computer system, including: Memory for storing computer programs; A processor is used to execute the computer program to implement the steps of the main and distribution coordinated anti-difference state estimation method described in Example 1.

[0059] Example 5

[0060] Based on the same inventive concept as Example 1, this embodiment introduces a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the main and distribution coordinated anti-difference state estimation method described in Example 1 are implemented.

[0061] In summary, the present invention uses pre-processed historical measurement data to train a pre-constructed convolutional neural network model, and uses the power flow equation to calculate the measurement residual, uses the weighted minimum residual square sum as the loss function, and updates the parameters of the convolutional neural network model by back propagation according to the loss function, and continuously trains to obtain a trained convolutional neural network model, and then uses the trained convolutional neural network model for online estimation to determine the main distribution coordinated robustness state estimation result; the main distribution coordinated robustness state estimation method provided by the present invention is a physics-based machine learning method, which can use the physical structure of the problem to improve the performance of the standard machine learning algorithm; wherein, the power flow equation is used to calculate the measurement residual and the weighted minimum residual square sum is used as the loss function, so that the training of the convolutional neural network model is transformed from a supervised learning problem to a weakly supervised learning problem, and a large amount of labeled data is no longer required as a prerequisite for model training, which is more in line with the actual conditions of the power grid; at the same time, the modification of the loss function reduces the size of the convolutional neural network search space, which in turn helps to better converge during the training process, and has good estimation accuracy and robustness; the method proposed in the present invention can be regarded as a physics-based machine learning method, which uses the physical structure of the problem to improve the performance of the standard machine learning algorithm.

[0062] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0064] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0065] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0066] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. A method for estimating the state of main and distribution coordinated robustness, characterized in that: include: Obtain historical measurement data in the main distribution coordination network and pre-process the historical measurement data; The pre-processed historical measurement data is used to train the pre-built convolutional neural network model. The measurement residual is calculated using the power flow equation. The weighted minimum residual square sum is used as the loss function. The parameters of the convolutional neural network model are updated by back propagation according to the loss function. The training is continued to obtain the trained convolutional neural network model. The trained convolutional neural network model is used for online estimation to determine the estimation results of the main and distribution coordinated robustness state.

2. The main and distribution coordinated robustness state estimation method according to claim 1 is characterized in that: The historical measurement data includes: active power output by the generator, reactive power output by the generator, active load power of the node, reactive load power of the node and voltage amplitude of the node; The preprocessing of the historical measurement data includes: calculating the active injection power of the node and the reactive injection power of the node through the active power output by the generator, the reactive power output by the generator, the active load power of the node and the reactive load power of the node, and the formula is as follows: ； In the formula, P i For the i The active injection power of each node is P iG For the i The active power output of the generator of each node, P iL For the i Active load power of each node; Q i For the i The reactive injection power of each node is Q iG For the i The reactive power output of the generator of each node, Q iL For the i The reactive load power of each node.

3. The main and distribution coordinated robustness state estimation method according to claim 1 is characterized in that: The network structure of the convolutional neural network model includes a convolutional layer, a pooling layer and a fully connected layer; The calculation formula of the convolutional layer is as follows: ； In the formula, For the l The output of the layer k Features, For the l Layer k The weight vector corresponding to the features, X l-1 For the l -1 layer output, For the l Layer k The learnable bias corresponding to each feature, f (·) is the nonlinear activation function of the convolutional layer; The calculation formula of the pooling layer is as follows: ； In the formula, For the q The value after pooling of the pooling area, is the first p unpooled values, D q For the q Pooling area; The calculation formula of the fully connected layer is as follows: ； In the formula, For the l Tier k The output of the convolutional layer after activation corresponding to the feature, g (·) is the activation function of the fully connected layer.

4. The main and distribution coordinated robustness state estimation method according to claim 1 is characterized in that: The parameters of the convolutional neural network model include fixed parameters and variable parameters. The fixed parameters include the number of neurons in the convolutional layer and the fully connected layer, the sample size used for each batch of model training, and the number of training passes for all samples. The variable parameters include the weight vectors corresponding to the features of each layer and the learnable bias.

5. The main and distribution coordinated robustness state estimation method according to claim 4 is characterized in that: The method for training the convolutional neural network model includes: The voltage amplitude of the node, the active injection power of the node and the reactive injection power of the node in the preprocessed historical measurement data are used as the input of the pre-built convolutional neural network model, and the estimated value of the voltage amplitude of the node and the estimated value of the voltage phase angle of the node are used as the output of the convolutional neural network model. At the same time, the power flow equation is used to calculate the measurement residual, and the weighted minimum residual sum of squares is used as the loss function. The variable parameters of the convolutional neural network model are updated by back propagation according to the loss function. Training is continued until the number of training passes for all samples is equal to the preset value, and a trained convolutional neural network model is obtained.

6. The main and distribution coordinated robustness state estimation method according to claim 2 is characterized in that: The method of calculating the measurement residual using the power flow equation and using the weighted minimum residual square sum as the loss function includes: For having n nodes, m The loss function of the convolutional neural network model is As shown below: ； In the formula, for dimensional state estimate, including voltage amplitude estimate and voltage phase angle estimate , z for dimensional measurement vector, h (·) for contact and z A nonlinear function of R is the covariance matrix of the measurement error, R is a diagonal matrix with the first u The element is v The measurement variance corresponding to the measurement value ,in, u = v ; z The expression is: ； In the formula, U , P , Q are the voltage amplitude of the node, the active injection power of the node and the reactive injection power of the node in the preprocessed historical measurement data, respectively. T Represents the transpose of a matrix; exist In the above equation, the active power injected by the node and the reactive power injected by the node are The nonlinear function relationship of , that is, the expression of the power flow equation is as follows: ； ； In the formula, and Respectively i Nodes and j The estimated voltage amplitude of each node, For the i Nodes and j The difference in the estimated voltage phase angle of the nodes is For the i The self-conductance of the node, For the i Nodes and j The mutual conductance between nodes is For the i The self-susceptance of each node, For the i Nodes and j The mutual admittance between nodes; The measurement residual is , so the loss function The expression is transformed as follows: ； In the formula, z v For the v A measurement value, For z v The corresponding value calculated by the nonlinear function.

7. The method for estimating the main and distribution coordinated robustness state according to claim 1, characterized in that: The method for online estimation using the trained convolutional neural network model includes: Calculate the active injection power and reactive injection power of the node according to the real-time measured active power output by the generator, the reactive power output by the generator, the active load power of the node, and the reactive load power of the node; The voltage amplitude of the node, the active injection power of the node and the reactive injection power of the node in real-time measurement are used as the input of the trained convolutional neural network model to obtain the estimated value of the voltage amplitude of the node and the estimated value of the voltage phase angle of the node.

8. A main and distribution coordinated anti-difference state estimation device, characterized in that: include: The data acquisition and preprocessing module is configured to acquire historical measurement data in the main distribution coordination network and preprocess the historical measurement data; The model training module is configured to train the pre-built convolutional neural network model using the pre-processed historical measurement data, and calculate the measurement residual using the power flow equation, use the weighted minimum residual square sum as the loss function, and update the parameters of the convolutional neural network model by back propagation according to the loss function, and continuously train to obtain the trained convolutional neural network model; The state estimation module is configured to use the trained convolutional neural network model to perform online estimation to determine the main and distribution coordinated robustness state estimation results.

9. A computer system, characterized in that: include: Memory for storing computer programs; A processor is used to execute the computer program to implement the steps of the main and distribution coordinated anti-difference state estimation method described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the main and distribution coordinated robustness state estimation method described in any one of claims 1-7 are implemented.

Images