An artificial intelligence-based low-voltage power distribution network impedance parameter identification method and system

Abstract

The present application relates to the technical field of power system state estimation, and particularly relates to a low-voltage distribution network impedance parameter identification method and system based on artificial intelligence, which comprises the following steps: constructing an abstract topological structure of a low-voltage distribution network according to the tree-like hierarchical structure characteristics of the low-voltage distribution network; setting a branch parameter calculation pool based on the abstract topological structure of the low-voltage distribution network according to the conductor type and length; determining a low-voltage distribution network electrical quantity distribution data pool under a plurality of types of combinations through power flow calculation; constructing a generative adversarial neural network model, training the model using topological parameters and the low-voltage distribution network electrical quantity distribution data pool, forming an input-output relationship between node parameters and branch parameters, and obtaining impedance parameter identification results. Through the present application, a method for obtaining distribution network branch impedance parameters based on a limited amount of measurement data of low-voltage distribution network nodes is provided, which is conducive to line loss analysis and fault positioning, and can effectively evaluate and correct the impedance parameters of the low-voltage distribution network.

Description

Technical Field

[0001] This invention relates to the field of power system state estimation technology, and in particular to a method and system for identifying impedance parameters of low-voltage distribution networks based on artificial intelligence. Background Technology

[0002] Accurate impedance parameters of the distribution network are fundamental for network power flow calculation, fault location, and state estimation, and are also prerequisites for power grid companies to carry out planning, operation and maintenance, and dispatching. With the large-scale integration of distributed photovoltaic, energy storage, and charging piles into the low-voltage distribution network, accurate impedance parameters play a crucial role in ensuring the full absorption of renewable energy and the integration of multiple energy sources.

[0003] Currently, my country's low-voltage distribution network uses transformer substations as the basic management unit. Line parameters in the low-voltage distribution network are not only the foundation for substation planning and operation analysis, but also play a crucial role in the power grid company's substation line loss analysis and fault location. Furthermore, due to age and maintenance, the actual impedance parameters of lines in the low-voltage distribution network deviate from theoretical values. Failure to update these line parameters in a timely manner will significantly impact the normal operation and maintenance management of the distribution network. Therefore, obtaining accurate line impedance parameters is of great importance for the planning, scheduling, and operation and maintenance management of low-voltage distribution networks. Summary of the Invention

[0004] This invention provides a method and system for identifying impedance parameters in low-voltage distribution networks based on artificial intelligence, which can effectively solve the problems in the background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] An artificial intelligence-based method for identifying impedance parameters in low-voltage distribution networks, the method comprising:

[0007] Based on the tree-like hierarchical structure of the low-voltage distribution network, an abstract topology of the low-voltage distribution network is constructed.

[0008] Based on the abstract topology of the low-voltage distribution network, a branch parameter calculation pool is set according to the conductor type and length;

[0009] Based on the branch parameter calculation pool, a data pool for the distribution of electrical quantities in the low-voltage distribution network under several combinations is determined through power flow calculation.

[0010] A generative adversarial neural network model is constructed and trained using the topological parameters of the abstract topology and the electrical quantity distribution data pool of the low-voltage distribution network to form the input-output relationship between node parameters and branch parameters, thereby obtaining the impedance parameter identification result.

[0011] Furthermore, based on the tree-like hierarchical structure characteristics of the low-voltage distribution network, an abstract topology of the low-voltage distribution network is constructed, including: constructing an abstract topology of the actual physical distribution network according to the tree-like hierarchical structure characteristics of the low-voltage distribution network, using an improved single-tree branch cut-set matrix Z∈C. (n-1)×(n-1) Describe C (n-1)×(n-1) Let Z be a complex space of dimension (n-1)×(n-1), where n is the number of nodes in the low-voltage distribution network; let z be the element in the i-th row and j-th column of the cut set matrix Z. ij , z ij The value is taken as the corresponding branch parameter, i.e., the branch impedance value z. ij =r ij +jx ij , where r ij Let x be the resistive parameter of the branch between nodes i and j. ij Let be the reactance parameter of the branch between nodes i and j.

[0012] Furthermore, it also includes: merging branch parameters using a linearization method, and the impedance value r ij +jx ij Use r ij +jkr ij This indicates that the probability density function of the impedance parameters of low-voltage distribution network branches is established using the kernel density estimation method:

[0013]

[0014] Where n is the number of sample points, K(·) is the Epanechnikov kernel function, and r is the impedance parameter value of the low-voltage distribution network branch. i Let h be the impedance parameter value of the i-th branch, and h>0 is the smoothing coefficient.

[0015] Furthermore, power flow calculations are used to determine several combinations of low-voltage distribution network electrical quantity distribution data pools, including:

[0016] Using the single-tree branch cut set matrix as the topological description, the probability density function of the low-voltage distribution network branch impedance parameters is superimposed;

[0017] Adaptive probability sample extraction is performed based on the probability density function;

[0018] Node data s is determined through power flow calculation. i =p i +jq i With branch data z ij =r ij +jkr ij The mapping relationship between them is used to construct a data pool for the distribution of electrical quantities in the low-voltage distribution network.

[0019] Furthermore, the low-voltage distribution network electrical quantity distribution data pool is implemented using a power flow simulation method, and the optimized bandwidth h of the kernel function in a single simulation is... opt as follows:

[0020]

[0021] Where σ is the standard deviation and n is the total number of samples in the extrapolated dataset.

[0022] Furthermore, a generative adversarial neural network model is constructed, and trained using the topological parameters of the abstract topology structure and the low-voltage distribution network electrical quantity distribution data pool to form the input-output relationship between node parameters and branch parameters, thereby obtaining impedance parameter identification results, including:

[0023] Historical measurement data of injected active power and injected reactive power at the low-voltage distribution network nodes were collected.

[0024] A generator model and a WGAN-GP discriminator model were constructed based on a spatiotemporal graph neural network.

[0025] By combining the generator model and the WGAN-GP discriminator model, a generative adversarial neural network model is formed.

[0026] The generative adversarial neural network model is trained using the historical measurement data and the low-voltage distribution network electrical quantity distribution data pool.

[0027] The actual measurement data of the low-voltage distribution network is obtained, and the actual measurement data is input into the trained generative adversarial neural network model to obtain the impedance parameter identification result.

[0028] Furthermore, a generator model and a discriminator model are constructed based on a spatiotemporal graph neural network, including: constructing a spatiotemporal graph learning module with a graph attention neural network as the core, and the attention layer adopting a multi-head attention mechanism.

[0029] Furthermore, obtaining actual measurement data of the low-voltage distribution network includes:

[0030] By setting corresponding sampling modes on short, medium, and long time scales, active power injection and reactive power injection of the low-voltage distribution network nodes are sampled to obtain active power time series and reactive power time series.

[0031] The active time series and reactive time series from the three modes are synthesized and used as input to the generator model.

[0032] Furthermore, it also includes: using a well-trained discriminator to conduct early warning assessments of potential impedance changes in low-voltage distribution network branches experiencing partial discharge.

[0033] An artificial intelligence-based impedance parameter identification system for low-voltage distribution networks, the system comprising:

[0034] Topology construction module: Based on the tree-like hierarchical structure characteristics of the low-voltage distribution network, construct the abstract topology of the low-voltage distribution network;

[0035] Branch parameter calculation pool setting module: Based on the abstract topology of the low-voltage distribution network, the branch parameter calculation pool is set according to the conductor type and length;

[0036] Power grid electrical quantity distribution data pool determination module: Based on the branch parameter calculation pool, the low-voltage distribution network electrical quantity distribution data pool under several combinations is determined through power flow calculation;

[0037] Impedance parameter identification result acquisition module: Construct a generative adversarial neural network model, train it using the topology parameters of the abstract topology structure and the low-voltage distribution network electrical quantity distribution data pool, form the input-output relationship between node parameters and branch parameters, and obtain the impedance parameter identification result.

[0038] The technical solution of this invention can achieve the following technical effects:

[0039] This invention provides a method for obtaining the impedance parameters of distribution network branches based on limited measurement data of low-voltage distribution network nodes. This method is beneficial for line loss analysis and fault location, and can effectively evaluate and correct the impedance parameters of low-voltage distribution networks. It helps to solve the current problem of difficulty in obtaining the branch parameters of low-voltage distribution networks, and is also conducive to the full absorption of renewable energy and the integration of multiple energy sources.

[0040] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating an artificial intelligence-based method for identifying impedance parameters in low-voltage distribution networks.

[0043] Figure 2 This is a schematic diagram of a graph attention neural network;

[0044] Figure 3 A flowchart illustrating the process for obtaining impedance parameter identification results;

[0045] Figure 4 A flowchart illustrating the process of obtaining actual measurement data for low-voltage distribution networks. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0048] Example 1

[0049] like Figure 1 As shown, an artificial intelligence-based method for identifying impedance parameters in low-voltage distribution networks includes:

[0050] S1: Based on the tree-like hierarchical structure of low-voltage distribution networks, construct an abstract topology for low-voltage distribution networks;

[0051] Specifically, considering the tree-like hierarchical structure of low-voltage distribution networks, an abstract topology of the actual physical distribution network is constructed. The topology edge set is the set of branches in the low-voltage distribution network, and the topology point set includes transformer low-voltage outlet points, low-voltage branch connection points, low-voltage terminal load points, low-voltage distributed power source access points (such as distributed photovoltaics), and low-voltage bidirectional multi-element equipment access points (such as distributed energy storage and electric vehicles).

[0052] S2: Based on the abstract topology of low-voltage distribution network, branch parameter calculation pools are set according to conductor type and length;

[0053] S3: Based on the branch parameter calculation pool, the distribution data pool of electrical quantities of low-voltage distribution network under several combinations is determined by power flow calculation.

[0054] S4: Construct a generative adversarial neural network model, train it using the topological parameters of the abstract topology structure and the data pool of electrical quantity distribution of the low-voltage distribution network, form the input-output relationship between node parameters and branch parameters, and obtain the impedance parameter identification results.

[0055] In this embodiment, various node parameters can be used as attribute characteristics of the low-voltage distribution network, represented as follows: That is, the k-th feature of the i-th node at the t-th time segment, where R n×m×k It is a real space of dimension n×m×k, where n is the number of nodes in the low-voltage distribution network, m is the time section number, and k is the number of feature matrices. k can be determined based on actual measurements in the power grid. Node measurements can include voltage, current, and power values; data can be instantaneous or effective values. The diversity of measurement data can reduce the search space for parameter identification and improve identification efficiency. In specific implementation, considering the availability of existing measurement data in the low-voltage distribution network, the effective values ​​of active and reactive power injected into the nodes are used as attribute features. Since each node contains two feature matrices: an active power matrix and a reactive power matrix, k = 2. When k = 2, the attribute features of the low-voltage distribution network can be represented as s. i,t ∈C n×m , where s i,t =p i,t +jq i,t p is the injected complex power at node i at time t. i,t and q i,t These are the injected active power and injected reactive power of node i at time t, respectively, S t =(s 1,t ,s 2,t ,...,s n,t ) T S represents the characteristic values ​​of all attribute nodes of the distribution network at time segment t, where S = {S1, S2, ..., S...} t ,...,S M} represents all attribute characteristic values ​​of all distribution network nodes across all time segments.

[0056] This invention can identify the branch impedance of a distribution network based on massive historical sampling values ​​from distribution network nodes, thereby obtaining the branch impedance of a low-voltage distribution network, i.e., deriving branch parameters from node data. A mapping function is formed between the node dataset and the branch parameter set as follows: Z = f(S) = f(S1,S2,...,S...). t ,...,S M ), where Z = {z} ij |1≤i,j≤n,i≠j} represents the set of impedance parameters for all branches, and function f represents the fitting function formed by the neural network through the neural network architecture proposed in this invention.

[0057] The present invention provides a method for obtaining the impedance parameters of distribution network branches based on limited measurement data of low-voltage distribution network nodes. This method is beneficial for line loss analysis and fault location, and can effectively evaluate and correct the impedance parameters of low-voltage distribution networks. It helps to solve the current problem of difficulty in obtaining the branch parameters of low-voltage distribution networks, and is also conducive to the full absorption of renewable energy and the integration of multiple energy sources.

[0058] Furthermore, based on the tree-like hierarchical structure of low-voltage distribution networks, an abstract topology of the low-voltage distribution network is constructed, including: constructing an abstract topology of the actual physical distribution network according to the tree-like hierarchical structure of the low-voltage distribution network, and using an improved single-tree branch cut-set matrix Z∈C. (n-1)×(n-1) Describe C (n-1)×(n-1) Let Z be a complex space of dimension (n-1)×(n-1), where n is the number of nodes in the low-voltage distribution network; let z be the element in the i-th row and j-th column of the cut set matrix Z. ij , z ij The value is taken as the corresponding branch parameter, i.e., the branch impedance value z. ij =r ij +jx ij , where r ij Let x be the resistive parameter of the branch between nodes i and j. ij Let be the reactance parameter of the branch between nodes i and j.

[0059] In this embodiment, since the low-voltage distribution network itself has a tree structure, each row of the cut-set matrix represents a cut set formed by a branch, therefore each row has exactly one non-zero element. When the low-voltage node v0 of the transformer is taken as the root node, the branch direction is a hierarchical radial direction, and the branch reference direction starts from the node v0 of the previous layer. i Flow to the next level node v j The elements of the Z matrix are always positive. The cut-set matrix can represent both the topology of the distribution network and the impedance parameter values ​​on the tree branches. In practical applications, it can be appropriately simplified based on the specific scenario. Since the conductor types are basically the same in incremental low-voltage distribution networks, there is a proportional relationship between the electrical parameters. Therefore, the impedance values ​​r of the elements in the Z matrix are... ij +jx ij A single value r can be used ij +jkr ij This indicates that k is a proportionality coefficient, and this simplification can improve the efficiency of subsequent parameter identification search and judgment.

[0060] Furthermore, it also includes: merging branch parameters using a linearization method, and the impedance value r ij +jx ij Use r ij +jkr ijThis indicates that the probability density function of the impedance parameters of low-voltage distribution network branches is established using the kernel density estimation method:

[0061]

[0062] Where n is the number of sample points, K(·) is the Epanechnikov kernel function, and r is the impedance parameter value of the low-voltage distribution network branch. i Let h be the impedance parameter value of the i-th branch, and h>0 is the smoothing coefficient.

[0063] Furthermore, power flow calculations are used to determine several combinations of low-voltage distribution network electrical quantity distribution data pools, including:

[0064] S31: Using a single-tree branch cut set matrix as the topological description, superimposed with the probability density function of the low-voltage distribution network branch impedance parameters;

[0065] S32: Adaptive probability sample extraction based on the probability density function;

[0066] S33: Determine node data s through power flow calculation i =p i +jq i With branch data z ij =r ij +jkr ij The mapping relationship between them is used to construct a data pool for the distribution of electrical quantities in the low-voltage distribution network.

[0067] Furthermore, the low-voltage distribution network electrical quantity distribution data pool is implemented using power flow simulation, and the optimized bandwidth h of the kernel function in a single simulation is... opt as follows:

[0068]

[0069] Where σ is the standard deviation and n is the total number of samples in the extrapolated dataset.

[0070] Furthermore, such as Figure 3 As shown, a generative adversarial neural network model is constructed and trained using topological parameters of an abstract topology structure and a low-voltage distribution network electrical quantity distribution data pool. This establishes the input-output relationship between node parameters and branch parameters, obtaining impedance parameter identification results, including:

[0071] S41: Collect historical measurement data of injected active power and injected reactive power at low-voltage distribution network nodes;

[0072] S42: Constructing a generator model and a WGAN-GP discriminator model based on a spatiotemporal graph neural network;

[0073] S43: The generator model and the WGAN-GP discriminator model are combined to form a generative adversarial neural network model;

[0074] Specifically, the artificial intelligence learning framework proposed in this invention is mainly based on the STGNN spatiotemporal graph neural network model and the GAN generative adversarial neural network model, such as... Figure 1 As shown, a Generative Adversarial Network (GAN) is a neural network that combines generative and adversarial functions. It includes a generator and a discriminator. The generator is a generative network that can generate data similar to real sample data; the discriminator is a discriminative network that can accurately judge and classify the input information. In this invention, both the generator and the discriminator are built based on spatiotemporal graph neural networks. Since there is a strong correlation between node measurement data in low-voltage distribution networks, the core module of STGNN adopts an attention network.

[0075] This invention uses the STGNN model to obtain the topological relationships between nodes in a low-voltage distribution network and the implicit electrical quantity distribution relationships. Traditionally, when using neural networks for distribution network parameter identification, it is difficult to consider the network's intrinsic structure, relying solely on the spatiotemporal distribution of various electrical quantities for judgment, resulting in distorted output results. Furthermore, the model's sensitivity is too high; even small changes in input data can cause jumps in output results. The STGNN model, by incorporating low-voltage distribution network topology and spatiotemporal distribution data of electrical quantities, can effectively uncover the implicit relationships between node and branch data, increasing the ability to identify parameters. Considering the importance of neighboring nodes in spatial dependency learning, the core module of STGNN adopts the GAT graph attention mechanism. Compared to graph neural network models, both GAT and GCN aggregate features from neighboring vertices to the central vertex, utilizing the local stability of the graph structure to learn new vertex feature representations. The difference lies in that GCN utilizes the Laplacian matrix, while GAT utilizes attention coefficients. Therefore, in principle, GAT is more suitable because the relationships between node features are better integrated into the model, making the attention mechanism highly effective in many sequence-based tasks.

[0076] S44: Train the generative adversarial neural network model using historical measurement data and a low-voltage distribution network electrical quantity distribution data pool;

[0077] For the generator, the STGNN model is used to obtain the data correlation and heterogeneity of the low-voltage distribution network at various time scales. Specifically, historical measurement data of low-voltage distribution network nodes is used as input data for the STGNN to generate evaluation values ​​of low-voltage distribution network branch parameters. For the discriminator, a distribution network forward-backward algorithm is first used to construct an accurate data pool of electrical quantity distributions under various real-world scenarios. This pool is then superimposed with the branch parameter evaluation data output by the generator, serving as input data for the discriminator. The STGNN model then distinguishes between real and generated data. The core of the overall GAN ​​framework is based on Nash equilibrium and game theory. The generator and adversary are continuously trained iteratively. Through the game process, the generator and discriminator can continuously promote each other's improvement. Ultimately, the generator can generate data that is very close to the real situation, to the point that the discriminator cannot distinguish between real and generated values. The trained generator can then be used as an impedance parameter identification tool to identify the impedance parameters of distribution network branches.

[0078] As a preferred embodiment of the above, GAN training can be divided into two stages: first, the generator parameters are fixed, and the discriminator is trained; then, the discriminator parameters are fixed, and the generator is trained. During the training process, both models continuously update their parameters to minimize their respective loss functions and output errors. The GAN loss function is as follows:

[0079]

[0080] Where V represents the output of the loss function, This represents the loss function that minimizes the generator. Let E(·) represent the loss function that maximizes the discriminator, and let E(·) represent the expected value of the distribution function; where The term represents the probability that the discriminator will interpret the real data as true when it is input into the discriminator. P represents the probability that the generated data is true when input into the discriminator, where x is the actual data and P is the probability of true. data (x) represents the probability distribution of x, z represents the random admission input, and P z (z) is the probability distribution of z, G(z) represents the generated data of the generator, D(x) represents the discrimination result of the discriminator on the real data, and D(G(z)) represents the discrimination result of the discriminator on the generated data.

[0081] In practical applications, due to the complexity and instability of the training process and the vanishing gradient problem in generative adversarial networks (GANs), improved GAN models, such as WGAN-GP, can be used. WGAN-GP improves the discriminator loss function in WGAN as follows:

[0082]

[0083] Where L(D) is the loss function of the discriminator. It is the Wasserstein distance. This is to penalize the gradient term.

[0084] WGAN-GP adds a gradient penalty term to the loss function of the WGAN discriminator. This gradient penalty is calculated by taking the gradient norm at sampling points between the discriminator's output and the real data, and then using the square of the difference between this norm and 1 as the gradient penalty term, which is added to the loss function. This helps maintain the continuity of the discriminator across the entire input space and reduces the problems of gradient explosion or vanishing. It retains the advantages of WGAN in overcoming gradient vanishing and mode collapse while ensuring that the discriminator does not encounter training difficulties due to excessive differences between the two distributions when updating gradients.

[0085] S45: Obtain actual measurement data of the low-voltage distribution network, input the actual measurement data into the trained generative adversarial neural network model, and obtain impedance parameter identification results.

[0086] Specifically, this invention maps the spatiotemporal hidden representation to the downstream task space, and predicts and evaluates the information of branches (edges between nodes), as illustrated in the diagram below. Figure 2 As shown, the parameter evaluation method often uses a multi-head attention mechanism, constructing K(2) trainable weight matrices W. (1) ,...,W (K) The specific K dimensions can be constructed in the following way:

[0087]

[0088]

[0089]

[0090] In this invention, the parameter sensing evaluation assesses the resistance and reactance parameters on the branch (side).

[0091] Furthermore, a generator model and a discriminator model are constructed based on a spatiotemporal graph neural network, including: constructing a spatiotemporal graph learning module with a graph attention neural network as the core, and the attention layer adopting a multi-head attention mechanism.

[0092] In this embodiment, the attention coefficients in the graph attention model are calculated as follows:

[0093] For node i, calculate its adjacent nodes (j∈N) one by one. i The similarity coefficient between them:

[0094] e ij =a T ([Wh i||Wh j ]),j∈N i

[0095] Where h i h represents the node attribute characteristics of node i. i ∈R F F is the feature dimension of the node. Shared parameters W∈R F′×F The node attribute features are augmented using a linear mapping to generate a new node feature set (which may have different feature dimensions F′). The transformed features of nodes i and j are concatenated using ([·||·]), where a∈R 2F′ It is the weight vector of a single-layer feedforward neural network.

[0096] Calculate the attention coefficient α of node i to j. ij for:

[0097]

[0098] The LeakyReLU (α = 0.2) activation function is used.

[0099] The output characteristics of a node can be obtained through the following equation:

[0100]

[0101] In this invention, the attention layer employs multi-head attention to make the network learning process more stable. It uses K independent attentions to calculate the state, and then averages the calculated K features to obtain the final output representation.

[0102]

[0103] in, These are normalized attention coefficients, obtained from the k-th attention mechanism. This attention mechanism structure can be parallelized during computation, thus offering high efficiency.

[0104] Furthermore, such as Figure 4 As shown, the actual measurement data of the low-voltage distribution network is obtained, including:

[0105] S451: Set corresponding sampling modes on short, medium and long time scales to sample the active and reactive power injection of low-voltage distribution network nodes to obtain active time series and reactive time series.

[0106] Specifically, since the measurement data of each node in the low-voltage distribution network has certain time-series characteristics, corresponding fitting modes can be set for different time periods, such as day and week. The current time can be set as t, the sampling time of node characteristic parameters can be set as Δt, for example, the sampling interval is 5 minutes, the number of samples in a day is 288, the number of samples in a week is 2016, and the time period is ΔT.

[0107] S452: Combine the active and reactive time series from the three modes and use them as input to the generator model.

[0108] For example, in the day mode, P day =(P t-288Δt-ΔT+1 ,P t-288Δt-ΔT+2 ,...,P t-288Δt Q(t) represents the historical time series of active power injected into nodes at the same time the previous day. For low-voltage distribution network power flow, it exhibits a certain trend within a day, reflecting the intraday fluctuations in active power injected into nodes. Similarly, Q(t)... day =(Q t-288Δt-ΔT+1 Q t-288Δt-ΔT+2 ,...,Q t-288Δt ) is the historical time series of reactive power injected by nodes at the same time the previous day. In weekly mode, P week =(P t-2016Δt-ΔT+1 ,P t-2016Δt-ΔT+2 ,...,P t-2016Δt Q is the historical time series of active power injected into nodes at the same time in the previous week. Based on the characteristics of low-voltage distribution networks, active power injection exhibits certain fluctuations within a week, reflecting the periodicity of user electricity consumption behavior. Similarly, the historical time series of reactive power injected into nodes at the same time in the previous week is Q. week =(Q t-2016Δt-ΔT+1 Q t-2016Δt-ΔT+2 ,...,Q t-2016Δt These two modes have the same power grid structure and electrical parameters; therefore, the outputs of both modes are similar. They can be combined as input to the generator model.

[0109] Furthermore, it also includes: using a well-trained discriminator to conduct early warning assessments of potential impedance changes in low-voltage distribution network branches experiencing partial discharge.

[0110] Example 2:

[0111] An artificial intelligence-based impedance parameter identification system for low-voltage distribution networks, the system comprising:

[0112] Topology Construction Module: Based on the tree-like hierarchical structure of low-voltage distribution networks, an abstract topology of the low-voltage distribution network is constructed.

[0113] Branch parameter calculation pool setting module: Based on the abstract topology of low-voltage distribution network, the branch parameter calculation pool is set according to the conductor type and length;

[0114] Power Grid Electrical Quantity Distribution Data Pool Determination Module: Based on the branch parameter calculation pool, the module determines the low-voltage distribution network electrical quantity distribution data pool under several combinations through power flow calculation.

[0115] Impedance parameter identification result acquisition module: Construct a generative adversarial neural network model, use the topological parameters of the abstract topology structure and the low-voltage distribution network electrical quantity distribution data pool for training, form the input-output relationship between node parameters and branch parameters, and obtain the impedance parameter identification result.

[0116] The adjustment system described above in this invention can effectively realize the method for identifying impedance parameters of low-voltage distribution networks based on artificial intelligence. The technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0117] Furthermore, the impedance parameter identification result acquisition module includes:

[0118] Historical measurement data acquisition unit: Collects historical measurement data of injected active power and injected reactive power at low-voltage distribution network nodes;

[0119] Generator-discriminator model building unit: Constructing generator model and WGAN-GP discriminator model based on spatiotemporal graph neural network;

[0120] Generative Adversarial Neural Network (GAN) Model Building Unit: A joint generator model and a WGAN-GP discriminator model form a GAN model;

[0121] Generative Adversarial Neural Network Model Training Unit: The generative adversarial neural network model is trained using historical measurement data and a low-voltage distribution network electrical quantity distribution data pool;

[0122] Impedance parameter identification result acquisition unit: acquires actual measurement data of low-voltage distribution network, inputs the actual measurement data into trained generative adversarial neural network model, and obtains impedance parameter identification results.

[0123] Similarly, the above-mentioned optimization schemes for the system can also achieve the optimization effects corresponding to the methods in Embodiment 1, which will not be repeated here.

[0124] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for identifying impedance parameters in low-voltage distribution networks based on artificial intelligence, characterized in that, The method includes: Based on the tree-like hierarchical structure of the low-voltage distribution network, an abstract topology of the low-voltage distribution network is constructed. Based on the abstract topology of the low-voltage distribution network, a branch parameter calculation pool is set according to the conductor type and length; Based on the branch parameter calculation pool, a data pool for the distribution of electrical quantities in the low-voltage distribution network under several combinations is determined through power flow calculation. A generative adversarial neural network model is constructed, and the topological parameters of the abstract topology structure are used to train the model with the electrical quantity distribution data pool of the low-voltage distribution network to form the input-output relationship between node parameters and branch parameters, and to obtain impedance parameter identification results. Based on the tree-like hierarchical structure of low-voltage distribution networks, an abstract topology of the low-voltage distribution network is constructed, including: constructing an abstract topology of the actual physical distribution network according to the tree-like hierarchical structure of the low-voltage distribution network, and using an improved single-tree branch set matrix. Describe, in which The dimension is complex space, Let be the number of nodes in the low-voltage distribution network; and define the cut set matrix. The Line 1 The elements of the column are , The value is taken as the corresponding branch parameter, i.e., the branch impedance value. ,in For nodes and The resistive parameters of the branches between them, For nodes and The reactive parameters of the branches between them; This also includes: merging branch parameters using a linearization method, and impedance values. use This indicates that the probability density function of the impedance parameters of low-voltage distribution network branches is established using the kernel density estimation method: ； Where n is the number of sample points. For the Epanechnikov kernel function, These are the impedance parameter values ​​for low-voltage distribution network branches. Here is the impedance parameter value of the i-th branch, and h>0 is the smoothing coefficient; A generative adversarial neural network model is constructed and trained using the topology parameters of the abstract topology structure and the electrical quantity distribution data pool of the low-voltage distribution network. This establishes the input-output relationship between node parameters and branch parameters, yielding impedance parameter identification results, including: Historical measurement data of injected active power and injected reactive power at the low-voltage distribution network nodes were collected. A generator model and a WGAN-GP discriminator model were constructed based on a spatiotemporal graph neural network. By combining the generator model and the WGAN-GP discriminator model, a generative adversarial neural network model is formed. The generative adversarial neural network model is trained using the historical measurement data and the low-voltage distribution network electrical quantity distribution data pool. The actual measurement data of the low-voltage distribution network is obtained, and the actual measurement data is input into the trained generative adversarial neural network model to obtain the impedance parameter identification result.

2. The method for identifying impedance parameters of low-voltage distribution networks based on artificial intelligence according to claim 1, characterized in that, The data pool for the distribution of electrical quantities in low-voltage distribution networks under several combinations is determined through power flow calculation, including: Using the single-tree branch cut set matrix as the topological description, the probability density function of the low-voltage distribution network branch impedance parameters is superimposed; Adaptive probability sample extraction is performed based on the probability density function; Node data is determined through power flow calculation. Branch data The mapping relationship between them is used to construct a data pool for the distribution of electrical quantities in the low-voltage distribution network.

3. The method for identifying impedance parameters of low-voltage distribution networks based on artificial intelligence according to claim 1, characterized in that, The low-voltage distribution network electrical quantity distribution data pool is implemented using power flow simulation, with optimized bandwidth of the kernel function in a single simulation. as follows: ； in, denoted as the standard deviation, and n is the total number of samples in the extrapolated dataset.

4. The method for identifying impedance parameters of low-voltage distribution networks based on artificial intelligence according to claim 1, characterized in that, The generator model and the discriminator model are constructed based on the spatiotemporal graph neural network, including: constructing a spatiotemporal graph learning module with graph attention neural network as the core, and the attention layer adopts a multi-head attention mechanism.

5. The method for identifying impedance parameters of low-voltage distribution networks based on artificial intelligence according to claim 1, characterized in that, Obtaining actual measurement data of the low-voltage distribution network includes: By setting corresponding sampling modes on short, medium, and long time scales, active power injection and reactive power injection of the low-voltage distribution network nodes are sampled to obtain active power time series and reactive power time series. The active time series and reactive time series from the three modes are synthesized and used as input to the generator model.

6. The method for identifying impedance parameters of a low-voltage distribution network based on artificial intelligence according to claim 1, characterized in that, Also includes: For potential impedance changes in low-voltage distribution network branches experiencing partial discharge, a well-trained discriminator is used for early warning assessment.

7. A low-voltage distribution network impedance parameter identification system based on artificial intelligence, characterized in that, The system employs the low-voltage distribution network impedance parameter identification method based on artificial intelligence as described in claim 1, wherein the system comprises: Topology construction module: Based on the tree-like hierarchical structure characteristics of the low-voltage distribution network, construct the abstract topology of the low-voltage distribution network; Branch parameter calculation pool setting module: Based on the abstract topology of the low-voltage distribution network, the branch parameter calculation pool is set according to the conductor type and length; Power grid electrical quantity distribution data pool determination module: Based on the branch parameter calculation pool, the low-voltage distribution network electrical quantity distribution data pool under several combinations is determined through power flow calculation; Impedance parameter identification result acquisition module: Construct a generative adversarial neural network model, train it using the topology parameters of the abstract topology structure and the low-voltage distribution network electrical quantity distribution data pool, form the input-output relationship between node parameters and branch parameters, and obtain the impedance parameter identification result.

Images