A topology identification method considering the uncertainty of renewable energy output

Abstract

The present invention discloses a topology identification method taking into account the uncertainty of new energy output, including: S1, collecting historical multi-source measurement data in the distribution network, and processing it into a node voltage probability distribution curve; S2, constructing a topology identification data set according to the node voltage probability distribution curve and its corresponding tie switch state encoding; S3, using the topology identification data set to train a probabilistic graph convolutional neural network as a topology identification model; S4, real-time collection of node voltages in the distribution network, using the established distributed new energy output model to perform flow calculations on it, and obtaining the corresponding node voltage probability distribution curve; S5, inputting the node voltage probability distribution curve into the trained topology identification model for processing, outputting the corresponding tie switch state, and completing topology identification. The present invention effectively improves the accuracy and applicability of distribution network topology identification, and provides effective topology information for subsequent state evaluation and safe and stable control of the distribution network.

Description

Technical Field

[0001] The present invention belongs to the technical field of distributed renewable energy power generation in power systems, and in particular relates to a topology identification method taking into account the uncertainty of renewable energy output. Background Art

[0002] Topology identification is the basis for power flow calculation, pseudo measurement, and state evaluation of distribution networks. It is also the premise for economic planning and operation, safe and stable control. However, with the access of large-scale distributed renewable energy, the dual uncertainty of "source-load" leads to an increase in the frequency of power flow fluctuations and over-limits, which further causes the distribution network feeder switch to operate more frequently, resulting in changes in the network structure, and the operation characteristics tend to be diversified and complex, which makes the safe and stable control of the distribution network put forward higher requirements for the accuracy and timeliness of topology identification. At the same time, the randomness and uncertainty of distributed renewable energy output bring about voltage over-limit, reverse power flow and other problems, which bring new challenges to the topology identification method based on power flow, voltage and other data. How to distinguish whether the change of system electrical parameters is caused by topological changes or fluctuations in renewable energy output is an urgent problem to be solved. Therefore, it is of great significance to study how to effectively use a large amount of user data collected by multiple types of monitoring devices in the distribution network to mine topological change information and consider the uncertainty of renewable energy output in the topological identification process to further improve the level of real-time state estimation of the distribution network.

[0003] At present, domestic and foreign literature has carried out some research on the application of intelligent algorithms in topology identification. On the one hand, in view of the uncertainty of new energy output, some literature models its output as a probability density curve that conforms to the normal distribution or a single Gaussian model. Some literature also points out that it is more practical to model the distributed new energy output using a mixed Gaussian model. The probability flow calculated based on this is also more advantageous when applied to topology identification. On the other hand, for the monitoring devices in the distribution network, some literature uses generative adversarial networks to fill in electrical data to solve the problem that intelligent algorithms require a large data scale; some literature considers the problem of high cost and insufficient coverage of PMU or μPMU, and proposes a data fusion method based on Kalman filtering or linear extrapolation. Based on the acquisition frequency and time scale of PMU, the data collected by SM, SCADA, etc. are aligned and filled to expand the input features of the topology identification model.

[0004] The above strategy collects user data through SCADA systems, smart meters, and PMUs, and based on intelligent algorithms, it has achieved real-time identification of distribution network topology to a certain extent. However, the correlation between different units has not been fully considered in the process of new energy output modeling, resulting in deviations in the flow calculation results. On the other hand, the existing solutions do not fully utilize various monitoring devices such as PQM in the distribution network. Therefore, how to fuse the collected data with different sampling frequencies is one of the difficulties in improving the accuracy of existing topology identification. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the topology identification method taking into account the uncertainty of renewable energy output provided by the present invention solves the problem that traditional topology identification methods often only consider the data fusion method of some measurement devices, but do not fully consider the sampling frequency, sampling accuracy and time scale mismatch of multi-source measurement data in the distribution network.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a topology identification method taking into account the uncertainty of new energy output, comprising the following steps:

[0007] S1, collect historical multi-source measurement data in the distribution network and process it into node voltage probability distribution curve;

[0008] S2. Construct a topology recognition data set based on the node voltage probability distribution curve and its corresponding tie switch state code;

[0009] S3, using the topology recognition dataset to train a probabilistic graph convolutional neural network as a topology recognition model;

[0010] S4, real-time acquisition of node voltages in the distribution network, and use of the established distributed renewable energy output model to calculate the power flow and obtain the corresponding node voltage probability distribution curve;

[0011] S5. Input the node voltage probability distribution curve into the trained topology identification model for processing, output the corresponding tie switch state, and complete the topology identification.

[0012] Furthermore, the step S1 is specifically as follows:

[0013] The historical multi-source measurement data are classified to obtain the voltage history data set under each distribution network topology. For each voltage history data set, the voltage history data collected under each working condition is used as the data basis for fitting the probability density function, and the parameters are estimated. Then, the historical voltage data of each monitoring source under each working condition is represented by a node voltage probability distribution curve.

[0014] Furthermore, the step S2 is specifically as follows:

[0015] S21, importing the topological structure of the distribution network, the historical voltage probability distribution curve of each node of the distribution network and the corresponding state of each interconnection switch of the distribution network;

[0016] S22, partitioning the distribution network topology according to the tie switches;

[0017] S23, abstracting the distribution network topology into a line graph topology according to the topology in each partition;

[0018] S24, discretizing the node voltage probability distribution curve into a fixed number of intervals according to the line graph topology, and then converting it into a node feature vector;

[0019] S25, normalizing the feature vectors of each node, subtracting the normalized feature vectors of adjacent nodes, obtaining node features of corresponding nodes in the line graph topology, and then constructing a topology recognition data set;

[0020] The topology identification data set includes node features in different topology line diagrams in the distribution network and their corresponding tie switch state codes, and the node features correspond to voltage probability distribution curves.

[0021] Furthermore, the step S24 is specifically as follows:

[0022] The voltage data in the node voltage probability distribution curve V =[ v 1,…, v M ] is divided into N intervals, the interval boundaries are [ b 1,…, b N ], the frequency of each interval is calculated by the following formula, and then converted into the corresponding node feature vector;

[0023]

[0024] In the formula, Indicates i Node features, Indicates j Voltage data, Represents the node feature vector, subscript i Indicates interval index, subscript j Indicates the voltage data index.

[0025] Furthermore, in step S3, when training the probabilistic graph convolutional neural network, the process of processing the input data is as follows:

[0026] S31, obtaining an adjacency matrix A and a node feature matrix X of the line graph topology according to the topological atlas set, the node set and the features of each node of the line graph topology;

[0027] S32, performing information aggregation and standardization processing of node features through multiple graph convolution layers and batch normalization operations in a probabilistic graph convolutional neural network, and extracting features between node voltage differences;

[0028] S33, classifying the features between the node voltage differences through the SoftMax layer in the probabilistic graph convolutional neural network, and outputting the classified state vector;

[0029] Each component of the state vector represents the switching state of a corresponding circuit breaker in the power distribution network.

[0030] Furthermore, the loss function for training the probabilistic graph convolutional neural network is:

[0031]

[0032] In the formula, L c Represents the total loss function value of all nodes in the graph; y m Indicates m The true value of each node; Indicates m The observed value of each node.

[0033] Furthermore, in step S4, the established distributed renewable energy output model includes describing the uncertainty of the power system in the distribution network by constructing a mixed Gaussian model to simulate an irregular output curve, and describing the correlation between the outputs of different units by constructing a multidimensional Copula function based on the correlation coefficients between the outputs of different units.

[0034] Furthermore, in the mixed Gaussian model, the probability density function of the distributed generation output in the distribution network is modeled by multiple normal distribution functions, and its expression is:

[0035]

[0036] In the formula, represents the number of Gaussian components, , and Respectively represent m The weights, means and variances of the components, represents the normal distribution function, represents the output of new energy; wherein the Gaussian component satisfies .

[0037] Furthermore, the method of constructing a multidimensional Copula function is as follows:

[0038] S41. Set different output sequences of new energy sourcesX and Y No. i The marginal distribution functions of the elements are F 1( x i )and F 2( y i ), and set F 1( x k )and F 2( y k ) is when X = x k hour, Y = y k The marginal distribution of x k and y k is the element in the new energy output sequence, and then determines the weight parameter oh and conditional parameters i The functional relationship is:

[0039]

[0040]

[0041] In the formula, C n express X and Y Experience Copula, oh 1, oh 2, oh 3 and i 1, i 2, i 3 are oh and i The elements in C T (·), C G (·)and C C (·) represent T-Copula, Gumbel-Copula and Clayton-Copula functions, respectively. α is a hyperparameter;

[0042] S42. Determine hyperparameters α The optimal solution of , and then get the binary Copula function;

[0043] S43, according to the binary Copula function, the joint probability density function of the multidimensional variables is calculated by using the D-Vine theory, and then the multidimensional Copula function is constructed by using the D-Vine structure;

[0044] The joint probability density function of the multidimensional variables The expression is:

[0045]

[0046] In the formula, X for d dimensional vector, X =[ x 1, x 2,…, x d ]; f k ( x k ) is a random variable x k The probability density function of j ( e )and k ( e ) is at the edge e Two conditional nodes connected; e = j ( e ), k ( e )| D ( e ), represents the edge set E i An edge of D ( e ) is the condition set; C M,j(e),k(e)|D(e) is the corresponding conditional Copula probability density function; F ( x j(e) | x D(e) ) is the corresponding conditional Copula probability density function, where x j(e) , x D(e) , x k(e) Represents the edge e Connected edges.

[0047] Furthermore, in step S4, the method for performing power flow calculation on the voltage of the acquisition node in the real-time distribution network using the established distributed new energy output model is specifically as follows:

[0048] S41. Based on the correlation between the outputs of different units described by the multidimensional Copula function, sample and obtain the distributed generation N Group sample value { u s1 ,…, u sN}; u s1 ,…, u sN Indicates the value of the 1st to Nth sampling points of the Copula function;

[0049] S42. According to the inverse function of the marginal cumulative distribution function representing the output of distributed generation, we obtain N Group node injection power W The sample value of { w s1 ,…, w sN}; w s1 ,…, w sN Represents the node injection power of 1-N sampling points;

[0050] S43, will N Group node injection power W Substitute the sampled value into the node injection equation g And branch power flow calculation equation k , collect the node voltage in the distribution network in real time N The sub-power flow is deterministically calculated, and we get N Group system node voltage;

[0051] S44, repeat step S43 to obtain the boundary node voltage of each partition in the distribution network, and perform statistics on it to obtain the corresponding node voltage probability distribution curve.

[0052] The beneficial effects of the present invention are:

[0053] The topology identification method proposed in the invention takes into account the uncertainty of renewable energy output, which makes up for the problem of imperfect modeling of renewable energy output in the existing distribution network topology identification model, and can use monitoring data with different sampling frequencies for model training, effectively improving the accuracy and applicability of distribution network topology identification, and providing effective topology information for subsequent state evaluation and safe and stable control of distribution networks. Specifically:

[0054] (1) The uncertainty of distributed renewable energy output and the correlation between the outputs of different renewable energy units are fully considered. The mixed Gaussian model with the introduction of Copula function is used to accurately characterize the output characteristics of distributed renewable energy.

[0055] (2) The probability density distribution curve is used to characterize the node characteristics under the uncertainty of distributed renewable energy output, so as to solve the problem of different sampling frequencies of different measurement sources, resulting in inconsistent dimensions of the training sample set, and reduce the dependence on expensive high-precision measurement equipment such as PMU;

[0056] (3) A probabilistic graph convolutional neural network is used for topology identification, and the node voltage probability distribution curve is used as the node input feature, which can effectively identify the state changes of the tie switch. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a flow chart of a topology identification method taking into account the uncertainty of new energy output provided by the present invention.

[0058] Figure 2 This is a schematic diagram of the probability density function provided by the present invention.

[0059] Figure 3 A schematic diagram of topological partitioning using IEEE33 nodes as an example provided by the present invention.

[0060] Figure 4 A schematic diagram of the abstract line graph topology process provided by the present invention.

[0061] Figure 5 This is a schematic diagram of the line graph topology corresponding to the IEEE33 node system provided by the present invention. DETAILED DESCRIPTION

[0062] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0063] The embodiment of the present invention provides a topology identification method taking into account the uncertainty of new energy output, such as Figure 1 As shown, the following steps are included:

[0064] S1, collect historical multi-source measurement data in the distribution network and process it into node voltage probability distribution curve;

[0065] S2. Construct a topology recognition data set based on the node voltage probability distribution curve and its corresponding tie switch state code;

[0066] S3, using the topology recognition dataset to train a probabilistic graph convolutional neural network as a topology recognition model;

[0067] S4, real-time acquisition of node voltages in the distribution network, and use of the established distributed renewable energy output model to calculate the power flow and obtain the corresponding node voltage probability distribution curve;

[0068] S5. Input the node voltage probability distribution curve into the trained topology identification model for processing, output the corresponding tie switch state, and complete the topology identification.

[0069] Step S1 of the embodiment of the present invention is specifically as follows:

[0070] The historical multi-source measurement data are classified to obtain the voltage history data set under each distribution network topology. For each voltage history data set, the voltage history data collected under each working condition is used as the data basis for fitting the probability density function, and the parameters are estimated. Then, the historical voltage data of each monitoring source under each working condition is represented by a node voltage probability distribution curve.

[0071] For example, the measurement accuracy, time scale and sampling frequency of multi-source measurement data in the distribution network are different, and they cannot be directly used to fully and accurately reflect the real situation of the distribution network. In this embodiment, a probability distribution curve is used to solve the problem of different sampling frequencies of measurement data, where the expression of the probability density function is:

[0072]

[0073] In the formula, m Represents the average value of historical voltage data, s Indicates the standard deviation of historical voltage data.

[0074] In this embodiment, the historical data of each monitoring under each working condition is represented by a voltage probability distribution curve through the above method, which solves the problem of inconsistent parameter dimensions caused by different sampling frequencies, and can be obtained. m Topology and n Under various working conditions m * n The voltage probability distribution curve is used for subsequent topology identification.

[0075] For example, this embodiment provides the processing effect of the above method on different sampling frequencies. The same set of data is selected, 1000 data points and 100 data points are sampled on average respectively, and the probability density function is calculated as follows: Figure 2 As shown, by calculating the KL divergence, it can be seen that the average divergence of the probability distribution curves constructed under different sampling frequencies is 0.0107. It can be considered that the difference between the two probability distributions can be ignored.

[0076] Step S2 of the embodiment of the present invention is specifically as follows:

[0077] S21, importing the topological structure of the distribution network, the historical voltage probability distribution curve of each node of the distribution network and the corresponding state of each interconnection switch of the distribution network;

[0078] S22, partitioning the distribution network topology according to the tie switches;

[0079] S23, abstracting the distribution network topology into a line graph topology according to the topology in each partition;

[0080] S24, discretizing the node voltage probability distribution curve into a fixed number of intervals according to the line graph topology, and then converting it into a node feature vector;

[0081] S25, normalizing the feature vectors of each node, subtracting the normalized feature vectors of adjacent nodes, obtaining node features of corresponding nodes in the line graph topology, and then constructing a topology recognition data set;

[0082] The topology identification data set includes node features in different topology line diagrams in the distribution network and their corresponding tie switch state codes, and the node features correspond to voltage probability distribution curves.

[0083] In step S24 of this embodiment, in order to avoid the problem that the loss function of the probabilistic graph convolutional neural network is difficult to converge or has low precision, this embodiment uses normalization processing to map each component of the node feature to the interval [0, 1], specifically:

[0084] The voltage data in the node voltage probability distribution curve V =[ v 1,…, v M ] is divided into N intervals, the interval boundaries are [ b 1,…, b N ], the frequency of each interval is calculated by the following formula, and then converted into the corresponding node feature vector;

[0085]

[0086] In the formula, Indicates i Node features, Indicates j Voltage data, Represents the node feature vector, subscript i Indicates interval index, subscript j Indicates the voltage data index.

[0087] For example, in this embodiment, the IEEE 33 node system is taken as an example, and a topology recognition data set is constructed according to the above method. The topology partitioning result corresponding to the IEEE 33 node system is as follows: Figure 3 As shown in the figure, the dotted line represents the tie switch. The imported distribution network topology diagram is as follows Figure 4 The method shown in the figure is abstracted into a line graph, where each node represents a partition. Finally, the partitioned IEEE33 node system can be transformed into Figure 5 Line diagram topology shown.

[0088] In step S3 of the embodiment of the present invention, the network structure and parameters of the constructed probabilistic graph convolutional neural network are shown in Table 1:

[0089] Table 1:

[0090]

[0091] Based on the probabilistic graph convolutional neural network shown in Table 1, when training the probabilistic graph convolutional neural network in this embodiment, the process of processing the input data is as follows:

[0092] S31, obtaining an adjacency matrix A and a node feature matrix X of the line graph topology according to the topological atlas set, the node set and the features of each node of the line graph topology;

[0093] S32, performing information aggregation and standardization processing of node features through multiple graph convolution layers and batch normalization operations in a probabilistic graph convolutional neural network, and extracting features between node voltage differences;

[0094] S33, classifying the features between the node voltage differences through the SoftMax layer in the probabilistic graph convolutional neural network, and outputting the classified state vector;

[0095] Each component of the state vector represents the switching state of a corresponding circuit breaker in the power distribution network.

[0096] In the above processing, for the output state vector, the total loss function value of all nodes is calculated using the observed value and the observed value and the true value through the cross entropy loss function of the following formula. After multiple iterations, the termination condition is met (the number of iterations is reached or the loss function value is less than a certain threshold), indicating that GCN has learned the mapping law between data, and the probabilistic graph convolutional neural network model at this time is saved.

[0097] In this embodiment, the loss function for training the probabilistic graph convolutional neural network is:

[0098]

[0099] In the formula, L cRepresents the total loss function value of all nodes in the graph; y m Indicates m The true value of each node; Indicates m The observed value of each node.

[0100] Exemplarily, for a trained probabilistic graph convolutional neural network (GCN), in this embodiment, the node voltage features of the test set are input into the trained GCN model for real-time circuit breaker switch state observation to obtain the topology of the distribution network. The performance of the evaluation model usually needs to be based on the calculation of the confusion matrix. In this application, the precision, recall, F1 score and other indicators of each label are considered. The following are the definitions and calculation methods of each indicator:

[0101] ①Accuracy P It intuitively represents the ability of the classifier to make correct decisions in all classification tasks, and its expression is:

[0102]

[0103] ②Recall rate R It measures the classifier's ability to identify positive samples, and its expression is:

[0104]

[0105] ③ F 1 score provides a comprehensive quantitative method for the model's precision and recall, and its expression is:

[0106]

[0107] Where: TP n and FP n Respectively represent n The number of positive labels and negative labels classified as positive in samples; FN n Indicates n The number of positive labels classified as negative in samples.

[0108] In step S4 of the embodiment of the present invention, the established distributed renewable energy output model includes describing the uncertainty of the power system in the distribution network by constructing a mixed Gaussian model to simulate an irregular output curve, and describing the correlation between the outputs of different units by constructing a multidimensional Copula function based on the correlation coefficients between the outputs of different units.

[0109] In this embodiment, as the penetration rate of distributed generation (DGs) in the distribution network increases, the uncertainty and randomness of the power system also increase significantly. These uncertainties usually do not conform to the normal distribution and often appear in a non-Gaussian form. In this embodiment, a Gaussian mixture model (GMM) is used to simulate irregular output curves, and these uncertainties are described by combining multiple normal distribution functions. In GMM, the probability density function (PDF) of DG output can be modeled by multiple normal distribution functions, and its expression is:

[0110]

[0111] In the formula, represents the number of Gaussian components, , and Respectively represent m The weights, means and variances of the components, represents the normal distribution function, represents the output of new energy; wherein the Gaussian component satisfies .

[0112] The mean and variance of the mixed distribution are:

[0113]

[0114]

[0115] In this embodiment, the mixed Gaussian model can simulate the probability density function of any distribution, thereby describing non-Gaussian uncertainty. The more components the mixed Gaussian model has, the closer the simulated distribution is to the original data distribution.

[0116] In this embodiment, the method for constructing a multidimensional Copula function is specifically as follows:

[0117] S41. Set different output sequences of new energy sources X and Y No. i The marginal distribution functions of the elements are F 1( x i )and F 2( y i ), and set F 1( x k )and F 2( y k ) is when X = x k hour, Y =y k The marginal distribution of x k and y k is the element in the new energy output sequence, and then determines the weight parameter oh and conditional parameters i The functional relationship is:

[0118]

[0119]

[0120] In the formula, C n express X and Y Experience Copula, oh 1, oh 2, oh 3 and i 1, i 2, i 3 are oh and i The elements in C T (·), C G (·)and C C (·) represent T-Copula, Gumbel-Copula and Clayton-Copula functions, respectively. α is a hyperparameter;

[0121] S42. Determine hyperparameters α The optimal solution of , and then get the binary Copula function;

[0122] S43, according to the binary Copula function, the joint probability density function of the multidimensional variables is calculated by using the D-Vine theory, and then the multidimensional Copula function is constructed by using the D-Vine structure;

[0123] The joint probability density function of the multidimensional variables The expression is:

[0124]

[0125] In the formula, X for d dimensional vector, X =[ x 1, x 2,…, x d ]; fk ( x k ) is a random variable x k The probability density function of j ( e )and k ( e ) is at the edge e Two conditional nodes connected; e = j ( e ), k ( e )| D ( e ), represents the edge set E i An edge of D ( e ) is the condition set; C M,j(e),k(e)|D(e) is the corresponding conditional Copula probability density function; F ( x j(e) | x D(e) ) is the corresponding conditional Copula probability density function, where x j(e) , x D(e) , x k(e) Represents the edge e Connected edges.

[0126] Exemplarily, in step S42 of this embodiment, the maximum expectation algorithm is used to find the optimal solution, and the prior distribution is defined as:

[0127]

[0128] Where: N for oh The number of elements in M for oh The dimension of E ( oh )for oh The expectation is as follows:

[0129]

[0130] make:

[0131]

[0132] In the formula, F (·) is the basis function.

[0133] Take ln p ( T | x i , α , β ) is expected to be:

[0134]

[0135] In the formula, F n for F No. n Column basis function.

[0136] Let the above formula be about α The derivative of is 0, so:

[0137]

[0138] On this basis, α The specific solution steps are as follows:

[0139] ② β When the inverse of the sample variance is selected, an initial value can be obtained α (0) .

[0140] ② Substitution , find oh and i The initial value of .

[0141] ③Substitute the initial value into the formula Repeat the above steps to get the convergence accuracy and calculate the weight coefficient oh and conditional parameters i .

[0142] In step S43 of this embodiment, taking the four-dimensional variable joint density function as an example, its decomposition form is as follows:

[0143]

[0144] Where: c 12 (·) for C 12 The corresponding probability density function is, c 23 (·), c 34 (·), c 13|2 (·), c 24|3 (·), c 14|23 (·) Similarly; F1|2 (·), F 1|2 (·), F 1|2 (·), F 1|2 (·), F 1|2 (·), F 1|2 (·) is the conditional cumulative distribution function.

[0145] In step S4 of the embodiment of the present invention, the method for performing power flow calculation on the voltage of the acquisition node in the real-time acquisition distribution network by using the established distributed new energy output model is specifically as follows:

[0146] S41. Based on the correlation between the outputs of different units described by the multidimensional Copula function, sample and obtain the distributed generation N Group sample value { u s1 ,…, u sN}; u s1 ,…, u sN Indicates the value of the 1st to Nth sampling points of the Copula function;

[0147] S42. According to the inverse function of the marginal cumulative distribution function representing the output of distributed generation, we obtain N Group node injection power W The sample value of { w s1 ,…, w sN}; w s1 ,…, w sN Represents the node injection power of 1-N sampling points;

[0148] S43, will N Group node injection power W Substitute the sampled value into the node injection equation g And branch power flow calculation equation k , collect the node voltage in the distribution network in real time N The sub-power flow is calculated deterministically, and we get N Group system node voltage;

[0149] S44, repeat step S43 to obtain the boundary node voltage of each partition in the distribution network, and perform statistics on it to obtain the corresponding node voltage probability distribution curve.

[0150] In step S5 of the embodiment of the present invention, the node voltage probability distribution curve of the boundary voltage is input as a feature into the established topology identification model to output the state of each tie switch, thereby realizing online identification of the topology.

[0151] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0152] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A topology identification method taking into account the uncertainty of renewable energy output, characterized in that: The following steps are involved: S1, collect historical multi-source measurement data in the distribution network and process it into node voltage probability distribution curve; S2. Construct a topology recognition data set based on the node voltage probability distribution curve and its corresponding tie switch state code; S3, using the topology recognition dataset to train a probabilistic graph convolutional neural network as a topology recognition model; S4, real-time acquisition of node voltages in the distribution network, and use of the established distributed renewable energy output model to calculate the power flow and obtain the corresponding node voltage probability distribution curve; S5, input the node voltage probability distribution curve into the trained topology identification model for processing, output the corresponding tie switch state, and complete the topology identification; The step S1 is specifically as follows: The historical multi-source measurement data are classified to obtain the voltage history data set under each distribution network topology. For each voltage history data set, the voltage history data collected under each working condition is used as the data basis for fitting the probability density function, and the parameters are estimated. Then, the historical voltage data of each monitoring source under each working condition is represented by the node voltage probability distribution curve. In step S4, the distributed renewable energy output model established includes describing the uncertainty of the power system in the distribution network by constructing a mixed Gaussian model to simulate an irregular output curve, and describing the correlation between the outputs of different units by constructing a multidimensional Copula function based on the correlation coefficients between the outputs of different units; In the mixed Gaussian model, the probability density function of the distributed generation output in the distribution network is modeled by multiple normal distribution functions, and its expression is: In the formula, represents the number of Gaussian components, , and Respectively represent m The weights, means and variances of the components, represents the normal distribution function, represents the output of new energy; wherein the Gaussian component satisfies ; The specific method of constructing a multidimensional Copula function is: S41. Set different output sequences of new energy sources X and Y No. i The marginal distribution functions of the elements are F 1( x i )and F 2( y i ), and set F 1( x k )and F 2( y k ) is when X = x k hour, Y = y k The marginal distribution of x k and y k is the element in the new energy output sequence, and then determines the weight parameter ω and conditional parameters θ The functional relationship is: In the formula, C n express X and Y Experience Copula, ω 1, ω 2, ω 3 and θ 1, θ 2, θ 3 are ω and θ The elements in C T (·), C G (·)and C C (·) represent T-Copula, Gumbel-Copula and Clayton-Copula functions, respectively. α is a hyperparameter; S42. Determine hyperparameters α The optimal solution of , and then get the binary Copula function; S43, according to the binary Copula function, the joint probability density function of the multidimensional variables is calculated using the D-Vine theory, and then the multidimensional Copula function is constructed using the D-Vine structure; The joint probability density function of the multidimensional variables The expression is: In the formula, X for d dimensional vector, X =[ x 1, x 2,…, x d ]; f k ( x k ) is a random variable x k The probability density function of j ( e )and k ( e ) is at the edge e Two conditional nodes connected; e = j ( e ), k ( e )| D ( e ), represents the edge set E i An edge of D ( e ) is the condition set; C M,j(e),k(e)|D(e) is the corresponding conditional Copula probability density function; F ( x j(e) | x D(e) ) is the corresponding conditional Copula probability density function, where x j(e) , x D(e) , x k(e) Represents the edge e connected edges; In step S4, the method for calculating the power flow of the voltage of the acquisition node in the real-time distribution network by using the established distributed new energy output model is specifically as follows: S41. Based on the correlation between the outputs of different units described by the multidimensional Copula function, sample and obtain the distributed generation N Group sample value { u s1 ,…, u sN }; u s1 ,…, u sN Indicates the value of the 1st to Nth sampling points of the Copula function; S42. According to the inverse function of the marginal cumulative distribution function representing the output of distributed generation, we obtain N Group node injection power W The sample value of { w s1 ,…, w sN }; w s1 ,…, w sN Indicates the node injection power of 1-N sampling points; S43, will N Group node injection power W Substitute the sampled value into the node injection equation g And branch power flow calculation equation k , the node voltage in the distribution network collected in real time is N The sub-power flow is deterministically calculated, and we get N Group system node voltage; S44, repeat step S43 to obtain the boundary node voltage of each partition in the distribution network, and perform statistics on it to obtain the corresponding node voltage probability distribution curve.

2. The topology identification method taking into account the uncertainty of new energy output according to claim 1, characterized in that: The step S2 is specifically as follows: S21, importing the topological structure of the distribution network, the historical voltage probability distribution curve of each node of the distribution network and the corresponding state of each interconnection switch of the distribution network; S22, partitioning the distribution network topology according to the tie switches; S23, abstracting the distribution network topology into a line graph topology according to the topology in each partition; S24, discretizing the node voltage probability distribution curve into a fixed number of intervals according to the line graph topology, and then converting it into a node feature vector; S25, normalizing the feature vectors of each node, subtracting the normalized feature vectors of adjacent nodes, obtaining node features of corresponding nodes in the line graph topology, and then constructing a topology recognition data set; The topology identification data set includes node features in different topology line diagrams in the distribution network and their corresponding tie switch state codes, and the node features correspond to voltage probability distribution curves.

3. The topology identification method taking into account the uncertainty of new energy output according to claim 2, characterized in that: The step S24 is specifically as follows: The voltage data in the node voltage probability distribution curve V =[ v 1,…, v M ] is divided into N intervals, the interval boundaries are [ b 1,…, b N ], the frequency of each interval is calculated by the following formula, and then converted into the corresponding node feature vector; In the formula, Indicates i Node features, Indicates j Voltage data, Represents the node feature vector, subscript i Indicates interval index, subscript j Indicates the voltage data index.

4. The topology identification method taking into account the uncertainty of new energy output according to claim 1, characterized in that: In step S3, when training the probabilistic graph convolutional neural network, the process of processing the input data is as follows: S31, obtaining an adjacency matrix A and a node feature matrix X of the line graph topology according to the topological atlas set, the node set and the features of each node of the line graph topology; S32, performing information aggregation and standardization processing of node features through multiple graph convolution layers and batch normalization operations in a probabilistic graph convolutional neural network, and extracting features between node voltage differences; S33, classifying the features between the node voltage differences through the SoftMax layer in the probabilistic graph convolutional neural network, and outputting the classified state vector; Each component of the state vector represents the switching state of a corresponding circuit breaker in the power distribution network.

5. The topology identification method taking into account the uncertainty of new energy output according to claim 4, characterized in that: The loss function for training the probabilistic graph convolutional neural network is: In the formula, L c Represents the total loss function value of all nodes in the graph; y m Indicates m The true value of each node; Indicates m The observed value of each node.

Images