Three-phase distribution network harmonic source locating method considering topology switching and new energy harmonic pseudo measurement

Abstract

The application provides a three-phase power distribution network harmonic source positioning method considering topology switching and new energy harmonic pseudo measurement, a harmonic emission model is constructed based on the harmonic emission characteristics of distributed new energy, the harmonic characteristics of the monitored nodes are migrated to the unmonitored nodes through transfer learning to generate harmonic pseudo measurement data, and the harmonic pseudo measurement data is used to make up for the missing measurement of the unmonitored nodes; the harmonic pseudo measurement data and the actual harmonic measurement data are used to realize harmonic state estimation through a multi-layer perception topology adaptive graph convolution network, the network adapts to topology switching by dynamically fusing the learnable weight of the multi-order adjacency matrix, and embeds the power distribution network harmonic transfer equation as a physical constraint into a loss function, and outputs the harmonic injection current estimation value of each node; the harmonic injection current estimation value is continuously accumulated, the statistical characteristics of the harmonic injection current of each node are analyzed through statistical analysis, and the harmonic source nodes in the power distribution network are comprehensively identified in combination with a preset criterion.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of power distribution network harmonic source positioning, and particularly relates to a three-phase power distribution network harmonic source positioning method considering topology switching and new energy harmonic pseudo-measurement. BACKGROUND

[0002] With the large-scale grid connection of renewable energy and power electronic devices, the number of three-phase power distribution network harmonic sources has increased dramatically, and presents obvious nonlinearity and imbalance. Harmonic pollution is becoming increasingly serious. Harmonics not only cause safety hazards such as overheating of electrical equipment and accelerated insulation aging, but also may cause system risks such as power grid resonance and relay protection misoperation, which seriously threaten power supply reliability and power quality. Therefore, effective positioning of harmonic sources is a key prerequisite for governing harmonic pollution, clarifying pollution responsibility, and ensuring safe and economic operation of power grids.

[0003] Current single-point harmonic source positioning methods measure the harmonic voltage and current at the point of common coupling (PCC), establish a Norton or Thevenin equivalent circuit model of the system, analyze the phase relationship of harmonic impedance to determine the harmonic power flow direction, and thus determine whether the harmonic source is mainly located on the system side or the user side. Multi-point harmonic source positioning methods obtain harmonic voltage and current measurements at multiple key nodes of the power grid, combine known network topology and parameters, and use harmonic state estimation algorithms to invert the location and injection of multiple harmonic sources in the system.

[0004] Harmonic state estimation uses accurate network topology and harmonic impedance parameter information to construct the admittance matrix of each harmonic, uses the harmonic voltage and current of the key nodes as measurement data, and uses weighted least squares, Kalman filter or particle filter methods to solve the harmonic state of the system. However, such methods require that the harmonic measurements be over-determined or critically positive definite, i.e., the dimension of the harmonic measurement points is greater than or equal to the dimension of the unknown harmonic state variables. Due to the cost constraints of harmonic monitoring device deployment, three-phase power distribution networks usually only deploy harmonic monitoring devices at key buses. Single-type harmonic monitoring cannot achieve global coverage of the network, resulting in a large number of nodes in the harmonic monitoring blind area. The above-mentioned incompleteness of power distribution network harmonic measurement information seriously restricts the accuracy and robustness of harmonic state estimation methods, and affects the accurate positioning of harmonic sources in power distribution networks.

[0005] With the development of artificial intelligence technology, data-driven harmonic state estimation methods have gradually become the focus of attention. Such methods use advanced deep learning techniques to directly extract harmonic propagation rules from historical data, learning the nonlinear mapping relationship between harmonic measurement data and state variables, thereby avoiding the dependence on the accurate physical model of the distribution network. However, the data-driven method is essentially a topologically dependent static fitting, and the model training is highly dependent on historical data under a specific network structure. In actual distribution networks, due to the frequent access or exit of devices such as high-penetration distributed energy and electric vehicle charging stations, the system topology and operation mode exhibit multi-modal switching characteristics. When the power grid topology changes, the original data distribution is destroyed, and the model loses its generalization ability and becomes invalid, so data must be collected and trained again; more importantly, the "black box" nature of data-driven methods weakens the interpretability of the results, making it difficult to support the practical application value of harmonic source localization results. SUMMARY

[0006] In view of the defects and deficiencies of the prior art, the present application provides a three-phase distribution network harmonic source positioning method considering topology switching and new energy harmonic pseudo-measurement, to solve the problems of low harmonic source positioning accuracy and poor robustness caused by sparse measurement, dynamic topology change and complex new energy harmonic characteristics.

[0007] The method first constructs a distributed new energy harmonic emission model to quantify the dynamic correlation between new energy output power and harmonic current, and uses federated transfer learning technology to transfer the harmonic characteristics of nodes with installed monitoring devices to unmonitored nodes, generating harmonic pseudo-measurement data to fill in the measurement blind area and improve system measurement observability. Secondly, a multi-layer perception topology adaptive graph convolution network (MPTAGCN) is designed to realize harmonic state estimation. The network adaptively tracks the topology switching of the distribution network by dynamically fusing the learnable weights of multi-order adjacency matrices, captures the complex correlation between nodes by combining edge weight convolution and dynamic attention mechanism, and embeds the distribution network harmonic propagation equation as a physical constraint into the loss function to ensure that the estimation result conforms to the physical characteristics of the power grid, and outputs the harmonic injection current estimation value of each node. Finally, the harmonic injection current estimation value is continuously accumulated, and statistical characteristics such as mean, variance and 95% quantile are calculated, and combined with preset criteria (such as the mean exceeding the threshold set based on the national standard limit and the reference current, the variance being relatively small or the 95% quantile being relatively large, etc.), to comprehensively identify the harmonic source nodes in the distribution network.

[0008] The present application effectively improves the accuracy and robustness of three-phase distribution network harmonic source positioning under dynamic topology through the coordinated design of pseudo-measurement generation, topology adaptive harmonic state estimation and statistical positioning, providing technical support for harmonic pollution control and safe operation of the power grid.

[0009] The technical scheme adopted by the present application to solve the technical problems is as follows:

[0010] A three-phase distribution network harmonic source positioning method considering topology switching and new energy harmonic pseudo measurement: a harmonic emission model is constructed based on the harmonic emission characteristics of distributed new energy, the harmonic characteristics of the monitored nodes are migrated to the unmonitored nodes through transfer learning to generate harmonic pseudo measurement data, which is used to make up for the missing measurement of unmonitored nodes; the harmonic pseudo measurement data and the actual harmonic measurement data are used to realize harmonic state estimation through a multi-layer perception topology adaptive graph convolution network, the network adapts to topology switching by dynamically fusing the learnable weight of the multi-order adjacency matrix, and embeds the distribution network harmonic transfer equation as a physical constraint into the loss function, and outputs the harmonic injection current estimation value of each node; based on the harmonic injection current estimation value, the statistical characteristics of the harmonic injection current of each node are analyzed through statistical analysis, and the harmonic source nodes in the distribution network are identified comprehensively according to the preset criterion.

[0011] Further, the harmonic emission characteristics of the distributed new energy include the dynamic correlation between the new energy output power and the harmonic current, and the hth harmonic current in the harmonic emission model is a function of the new energy output power, the carrier frequency, the grid impedance, the DC side voltage and the grid voltage.

[0012] Further, the transfer learning adopts federal transfer learning, the source domain is the harmonic monitoring data of the distributed new energy grid connection point where the harmonic monitoring device is installed, and the target domain is the distributed new energy without monitoring device; the input of the transfer model includes the output power, node voltage and current of the distributed new energy, and the output is the harmonic current of the node; when there is no training data in the target domain, the last network parameter of the pre-training model of the source domain is added to the disturbance and applied to the target domain.

[0013] Further, the multi-layer perception topology adaptive graph convolution network includes a topology adaptive graph convolution neural network, an edge weight convolution, and a dynamic attention mechanism; wherein the topology adaptive graph convolution neural network obtains the node feature matrix of the next layer by the sum of the product of the multi-order normalized adjacency matrix and the corresponding layer node feature matrix and the learnable weight matrix, and processes it through an activation function; the node feature matrix includes voltage and current harmonic components; the multi-order adjacency matrix discards redundant orders through Dropout optimization.

[0014] Further, the edge weight convolution dynamically generates a convolution kernel weight matrix based on edge features through a multi-layer perception machine to capture the complex correlation between nodes in the graph structure; the edge features include line impedance parameters and topology connection types.

[0015] Further, the dynamic attention mechanism obtains the attention coefficient by processing the original attention score of the adjacent nodes through the LeakyReLU activation function and then normalizing it through softmax, so as to differentially allocate the influence weight of the neighbor node information on the new features of the current node.

[0016] Further, the loss function of the physical constraint comprises a mean square error of the predicted harmonic state and the real harmonic state, and a bias term of the estimated measurement value and the actual measurement value; wherein the bias term is a square of the bias when the absolute value of the bias does not exceed a threshold, and is a product of the threshold and the absolute value of the bias minus half of the square of the threshold when the absolute value of the bias exceeds the threshold.

[0017] Further, the statistical characteristics comprise a mean, a variance and a 95% quantile of the harmonic injection current estimation value of each node; wherein the mean reflects the average contribution level of the node to the harmonic, the variance reflects the fluctuation degree of the estimation value around the mean, and the 95% quantile reflects the extreme level of the harmonic injection current.

[0018] And a three-phase distribution network harmonic source positioning system considering topology switching and new energy harmonic pseudo-measurement, comprising:

[0019] a pseudo-measurement generation module, configured to construct a harmonic emission model based on the harmonic emission characteristics of the distributed new energy, and to migrate the harmonic characteristics of the monitored nodes to the unmonitored nodes by transfer learning to generate harmonic pseudo-measurement data, the harmonic pseudo-measurement data being used to make up for the missing measurements of the unmonitored nodes;

[0020] a harmonic state estimation module, configured to utilize the harmonic pseudo-measurement data and the actual harmonic measurement data, and to realize harmonic state estimation by a multi-layer perception topology adaptive graph convolution network, the network adapting to topology switching by dynamically fusing the learnable weights of multi-order adjacency matrices, and embedding a distribution network harmonic transfer equation as a physical constraint into a loss function, and outputting harmonic injection current estimation values of each node;

[0021] a harmonic source positioning module, configured to continuously accumulate based on the harmonic injection current estimation values, to analyze the statistical characteristics of the harmonic injection current of each node by statistical analysis, and to comprehensively identify the harmonic source nodes in the distribution network in combination with a preset criterion.

[0022] And a computer device, comprising a memory, a processor and a computer program stored on the memory, the processor realizing the method as described above when executing the computer program.

[0023] A non-transitory computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to realize the method as described above.

[0024] Compared with the prior art, the application and the preferred scheme thereof solve the problems of model failure or precision reduction of the traditional method when the topology is changed by designing a multi-layer perception topology adaptive graph convolution network (MPTAGCN) which dynamically fuses the characteristics of multi-order adjacency matrix learnable weights and can adaptively track the dynamic switching of the power distribution network topology; meanwhile, the power distribution network harmonic propagation equation is embedded into the loss function as a physical constraint, which ensures that the harmonic state estimation result conforms to the physical characteristics of the power grid and significantly improves the precision and robustness of the harmonic state estimation under dynamic working conditions.

[0025] Meanwhile, the pseudo-measurement generation method based on harmonic emission characteristic migration migrates the harmonic characteristics of the monitored nodes to the unmonitored nodes through federal transfer learning, effectively fills the measurement blind area of the power distribution network, improves the observability of the system harmonic measurement, and provides more complete data support for subsequent state estimation; and the positioning method based on statistical harmonic data comprehensively determines the harmonic source through long-term accumulation and statistical analysis of the harmonic injection current estimation value combined with multi-dimensional indicators such as mean, variance and extreme value, avoids the misjudgment caused by the influence of measurement errors, model deviations and other factors in single estimation, and improves the reliability and accuracy of harmonic source identification.

[0026] In summary, the application provides a more adaptive solution for three-phase power distribution network harmonic source positioning in the scenario of high penetration of new energy and dynamic changes of topology through the collaborative optimization of multiple technical links, which supports harmonic pollution control and safe and economic operation of the power grid. BRIEF DESCRIPTION OF DRAWINGS

[0027] The application will be further described in detail below in combination with the drawings and specific embodiments:

[0028] Figure 1 The three-phase power distribution network harmonic source positioning method flowchart considering topology switching and new energy harmonic pseudo-measurement for the embodiments of the application. DETAILED DESCRIPTION

[0029] In order to make the features and advantages of the application more obvious and easy to understand, the following embodiments are described in detail as follows:

[0030] It should be pointed out that the following detailed description is exemplary and is intended to provide further description of the present application. Unless otherwise specified, all technical and scientific terms used in the specification have the same meaning as generally understood by those skilled in the art to which the present application belongs.

[0031] It is to be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0032] In view of the complex working conditions such as sparse harmonic measurement and dynamic change of system topology in the current three-phase power distribution network harmonic source positioning process, the present application provides a three-phase power distribution network harmonic source positioning method considering system topology switching and new energy harmonic pseudo measurement.

[0033] It includes the design of the following three main parts:

[0034] 1. Distributed new energy harmonic pseudo measurement generation method based on harmonic emission characteristic migration: This method considers the influence of control delay, dead zone effect and other factors on system harmonics, and constructs a distributed new energy harmonic emission characteristic model; and uses the migration learning method to construct a new energy harmonic pseudo measurement generation method based on the emission characteristic modeling, to improve the observability of system harmonic measurement.

[0035] 2. Harmonic state estimation method based on multi-layer perception topology adaptive graph convolutional neural network (MPTAGCN): In view of the problem that the dynamic switching and real-time topology change of distributed energy access are difficult to track accurately, a multi-layer perception topology adaptive graph convolutional neural network (MPTAGCN) method is proposed, and the system parameters and harmonic transfer equation of the power distribution network are introduced as physical constraints to ensure that the harmonic state meets the physical characteristics.

[0036] 3. Power distribution network harmonic source positioning method based on statistical harmonic data: This method constructs a harmonic source identification criterion, realizes multi-point harmonic source positioning in the power distribution network based on the statistical indicators of harmonic injection current estimation data in a specific period, and avoids the problem that the single positioning result is inaccurate due to measurement error, inaccurate model, background harmonic, load model error and other reasons in the real power distribution network.

[0037] The present application will be further demonstrated and introduced through specific embodiments:

[0038] In the first aspect, for the power distribution network system with high penetration of distributed new energy, due to economic cost and other factors, there are a large number of monitoring blind areas in the power distribution network, and the traditional harmonic state estimation method based on power grid equation has the problem of insufficient observability in the power distribution network, and it is difficult to realize real-time perception and accurate grasp of the harmonic state of the whole network. Therefore, the present application provides a distributed new energy harmonic pseudo measurement generation method based on harmonic emission characteristic migration, and the implementation process includes:

[0039] (1) Harmonic emission characteristics of distributed photovoltaic

[0040] The power generation of distributed photovoltaic has strong uncertainty due to weather changes and other factors, and the harmonic current injected into the distribution network by the inverter presents significant uncertainty. If the full-bridge topology is adopted for the distributed photovoltaic inverter and the inverter is controlled in the bipolar PWM modulation mode, the double Fourier series expansion of the inverter output voltage v inv (t) is as follows:

[0041] (1)

[0042] wherein t is time; V dc is the DC side voltage; the grid voltage ; the inverter output voltage , S(t) is a switching function; the modulation wave is , M is the modulation degree, is the angular frequency of the modulation signal, is the initial phase of the modulation signal; is the carrier angular frequency; the bridge arm switching conduction duty cycle dx=M+0.5; h is the harmonic order; k is the fundamental sideband harmonic order; J k (⋅) is the kth order Bessel function. The fundamental voltage component v inv，1 (t) of the inverter output and its effective value V inv，1 are as follows:

[0043] (2)

[0044] The fundamental current component i(t) is determined by the fundamental voltage difference and the grid fundamental impedance Z1:

[0045] (3)

[0046] wherein Z1=R+jω1L, R is the system resistance, L is the system inductance, and j is the imaginary unit. Assuming unit power factor operation, the effective value I1 of the fundamental current is:

[0047] (4)

[0048] wherein is the power factor (power factor), is the phase angle. The active power P is the product of the grid fundamental voltage V grid and the effective value of the fundamental current I1:

[0049] (5)

[0050] The modulation degree M can be solved from equation (5) as follows:

[0051] (6)

[0052] The hth harmonic voltage is given by the double Fourier series expansion of the inverter output voltage (1) as:

[0053] (7)

[0054] The harmonic current I h is determined by the harmonic voltage V h and the grid harmonic impedance Z h , whose calculation formula is as follows:

[0055] (8)

[0056] Ignoring the internal loss of the converter, the active power on the AC side and the DC side is balanced, and the instantaneous power theory can be obtained , and M is substituted into the above formula to obtain the relationship expression of the hth harmonic current I h emitted by the distributed photovoltaic power P pv :

[0057] (9)

[0058] As can be seen from equation (9), when the distributed photovoltaic power P pv increases, the modulation degree M also increases, and the harmonic voltage is affected by the modulation degree, the DC side voltage and the Bessel function, so when the output power of the inverter changes, the harmonic current injected into the grid will also change. Therefore, according to equation (9), a distributed photovoltaic harmonic emission model considering power electronic characteristics can be established, and its simplified expression is:

[0059] (10).

[0060] (2) Generation of regional distributed photovoltaic harmonic pseudo-measurement based on transfer learning

[0061] Transfer learning (TL) is an advanced technology in the field of artificial intelligence, which transfers knowledge from data-rich source domains to improve the performance of target domain tasks. Suppose {F1, F2,..., F N} is the source domain of N data providers, {D1, D2,..., D N} is the N data sets of the corresponding data providers, {X1, X2,..., X N} is the feature set of the N data sets, and {Y1, Y2,..., Y N} is the label set of the N data sets. If for any i, j {1,..,N}, Xi = X j , Y i = Y j , then it is called homogeneous transfer learning. Model-based homogeneous transfer learning is to directly transfer the model network trained in the source domain to the target domain, reuse its network architecture and training parameters, and then use the target domain input data to realize the inference and prediction of new samples.

[0062] Therefore, the present application adopts federal transfer learning to realize the construction of a regional distributed photovoltaic harmonic emission model. It is worth noting that the source domain data is the harmonic monitoring data of the distributed photovoltaic grid-connected point in the region where the harmonic monitoring device is installed, and the target domain is the distributed photovoltaic in the region where the monitoring device is not installed. According to the distributed photovoltaic harmonic emission model formula (10), the input of the transfer model is the distributed photovoltaic output power, node voltage and node current, and the output is the node harmonic current I h , and the minimum cross-entropy loss function is used. After the source domain training is completed, the parameters are transferred to the target domain, that is, the pre-trained neural network parameters are fixed, and the last layer network parameters are fine-tuned. However, due to the lack of sample data and training data in the target domain, the last layer network parameters of the source domain are disturbed and used in the target domain.

[0063] In the second aspect, with the large-scale grid connection of distributed new energy, the power supply mode of the distribution network changes, the line tie switch is frequently switched, the system topology dynamically changes, and the method of relying on manual recording of topology changes is low in efficiency and difficult to reflect the real network structure in time, which easily causes the estimation model to be mismatched, resulting in the decrease of the estimation accuracy of the traditional method. Therefore, the present application proposes a multilayer perceptron topology adaptive graph convolutional neural network (MPTAGCN) for the three-phase distribution network harmonic state estimation problem, and the algorithm architecture and technical details are described as follows:

[0064] (1) TAGCN

[0065] Topology Adaptive Graph Convolutional Network (TAGCN) is one of the variants of Graph Convolutional Network (GCN). Compared with GCN, which takes K=1 after Chebyshev polynomial approximation of the convolution kernel, TAGCN uses k graph convolution kernels to extract local features of different sizes, and keeps k as a hyperparameter. By introducing a linear combination of multi-order adjacency matrices, the model can adaptively adjust the importance of different order neighbor information. The mathematical derivation process is as follows:

[0066] (11)

[0067] (12)

[0068] (13)

[0069] (14)

[0070] where l is the l-th layer of graph convolution hidden layer; is the f-th input feature of the (l+1)-th hidden layer; is the f-th output feature of the l-th hidden layer; is k graph convolution kernels; is the data of all graph nodes related to the c-th feature input by the l-th hidden layer; is the coefficient of each graph convolution kernel; is the normalized adjacency matrix after adding self-loop; I is the identity matrix; D is the sum of the original adjacency matrix A and the identity matrix I is the degree matrix of D; is the activation function; is the learning bias; is the N l dimensional identity matrix. The core innovation of TAGCN is to introduce a learnable weight matrix , that is, each layer assigns an independent weight matrix to K+1 adjacency matrix powers, realizing dynamic fusion of neighbor information of different orders, so formula (11) can also be written as formula (15).

[0071] (15)

[0072] (16)

[0073] (17)

[0074] where H is the node feature matrix constructed by all node feature information , is the feature vector of node i; F is the feature dimension of the node; is the loss function; is the learnable weight matrix of the l-th layer and the k-th order, which is optimized through backpropagation (formula (16)) during the training process. If the neighbor information of a certain order k contributes more to reducing the loss, the corresponding weight will significantly increase, otherwise it will be suppressed, and the redundant order is discarded through Dropout to automatically select the optimal neighborhood range.

[0075] (2) ECC and MLP

[0076] Edge Conditioned Convolution (ECC) is a convolution operation for graph data, which dynamically generates convolution kernels by explicitly using edge information (such as edge weight, type or feature) in the graph, which can better capture complex relationships in graph structure. For the neighbor node j of node i, the convolution output of ECC is:

[0077] (18)

[0078] (19)

[0079] where N(i) is the neighbor set of node i; is the number of neighbors of set i, which is used as a normalization factor; is a weight matrix dynamically generated by edge features , which is implemented through a Multilayer Perceptron (MLP). The MLP can represent the high-order nonlinear relationship hidden in the measurement data through the synergistic effect of nonlinear activation function and hierarchical structure, so as to capture complex patterns such as multivariate coupling and non-stationary dynamic characteristics. Specifically, it can be expressed as formula (19).

[0080] (3) Dynamic attention

[0081] The TAGCN network model mainly considers the features of nodes themselves and the positional relationship between nodes, and does not consider the features of the edges between nodes, and the attention coefficient of any node is relatively constant. Therefore, the calculation process of the attention coefficient is improved, and the improved attention coefficient calculation formula is:

[0082] (20)

[0083] (21) ​

[0084] (22)

[0085] (23)

[0086] (24)

[0087] (25)

[0088] where, is the output vector of the projected node i; is the learnable weight matrix; is the i-th input vector; is the original attention score between adjacent nodes i and j; a is the learnable weight vector; LeakyReLU is a nonlinear activation function, Generally 0.2; is the attention coefficient after softmax normalization, different important neighbor node information will be given different weights to determine its influence on the new features of node ; is the new feature of each output node i that integrates neighborhood information; is the activation function, and in this example, eLU activation function is used.

[0089] (4) Loss function

[0090] The MPTAGCN algorithm incorporates physical equation constraints in the network architecture, and then proposes a cumulative loss function to enhance the estimation accuracy and robustness of the model. That is, the mean square error between the actual state and the target state is used as the loss function of supervised learning , and the physical constraints are used for self-supervised learning between the actual measurement value and the estimated measurement value, and the specific mathematical expression is:

[0091] (26)

[0092] where, is the predicted harmonic state output by the measurement ; is the real harmonic state of the distribution network; is the real harmonic state of the distribution network; is the measured value calculated according to the node predicted value, i.e. , H is the Jacobian matrix; N is the number of network nodes; is the physical model constraint, which selects δ by multiplying the standard deviation of the measured value σ with the actual measured value z, and takes the value after comparing it with .

[0093] In a third aspect, in actual power systems, due to factors such as measurement errors, model deviations, background harmonic interference and load model uncertainties, the estimated values of all node harmonic currents are usually not zero. Therefore, the determination of the harmonic source location is difficult to directly depend on a single positioning result, and must be identified through long-term statistical analysis. For this purpose, the present application proposes a power distribution network harmonic source positioning method based on statistical harmonic data. The method uses multi-point harmonic monitoring data and pseudo-measurement data of the power distribution network, and applies the MPTAGCN algorithm to estimate the harmonic injection current of each node. By accumulating the harmonic injection current estimation results for a long time, the statistical characteristics of the estimated injection current of each node are analyzed, and finally the harmonic source node in the power distribution network is identified. The specific implementation steps are as follows:

[0094] Step 1: Data acquisition. Based on harmonic measurement data and MPTAGCN algorithm, the harmonic injection current estimation value I h of each node for each harmonic is obtained, and obviously unreasonable results are eliminated. This process can last for several weeks, months or even longer.

[0095] Step 2: Key statistical index calculation. The following key statistics are calculated for each node to represent the long-term statistical characteristics of the node harmonic injection current:

[0096] (1) Mean μ_I h : reflects the average contribution level of the node to the harmonic. The greater the positive value, the greater the average contribution.

[0097] (27)

[0098] (2) Variance σ²_I h and standard deviation σ_I h : reflect the volatility of the node harmonic injection current estimation value around the mean. Large volatility may mean that the node is greatly affected by other sources, or its own harmonic emission is unstable (such as impact load). Small variance and large mean indicate that the node is a stable and significant harmonic source.

[0099] (28)

[0100] (2) Key quantile--95% quantile I h _95%: reflects the extreme level of harmonic injection current. It is important for evaluating peak impact and device capacity.

[0101] (29)

[0102] Step 3: Comprehensive determination of harmonic source node. Based on the long-term statistical results, the following criteria are combined to comprehensively determine whether the node is a harmonic source node.

[0103] (1) Significant positive mean value: its absolute value is much larger than other nodes, or exceeds the set threshold. The threshold of this embodiment is 0.1 The definitions are as follows:

[0104] (30)

[0105] Wherein: I is the national standard h harmonic current limit value, I set I is the reference current; K is the set proportion coefficient, which is 0.5.

[0106] (2) Relatively small standard deviation / variance: compared with the mean value, the relatively small standard deviation / variance indicates that the injection is relatively stable, and is not dominated by measurement noise or model error. For the fluctuation source, the variance will be large, and at this time, the high 95% quantile needs to be combined.

[0107] (3) High 95% quantile: even if the mean value is not large, but the 95% quantile is high, which indicates that there is significant intermittent or impact harmonic emission at the node.

[0108] Based on the design of the above embodiments, the implementation steps of the three-phase power distribution network harmonic source positioning method considering topology switching and new energy harmonic pseudo measurement according to the present application are as shown in Figure 1 The processes include the following:

[0109] 1. Data collection

[0110] Collecting distributed new energy harmonic data in the region;

[0111] Collecting power output data of missing distributed new energy harmonic monitoring nodes in the region.

[0112] 2. Model construction and pseudo measurement generation

[0113] Performing federated transfer learning on the distributed new energy harmonic data in the region;

[0114] Combining the federated transfer learning results and the power output data of the missing monitoring nodes to construct a harmonic emission characteristic model;

[0115] Generating distributed harmonic pseudo measurements using the harmonic emission characteristic model.

[0116] 3. Data input and network processing

[0117] Preparing part of the node harmonic monitoring data;

[0118] Inputting part of the node harmonic monitoring data and the distributed harmonic pseudo measurements into the multi-layer perception topology adaptive graph convolutional neural network (MPTAGCN).

[0119] 4. Harmonic current estimation and long-term analysis

[0120] Output each node harmonic injection current estimation value by MPTA GCN;

[0121] Carry out long-term data analysis (statistical feature calculation) on each node harmonic injection current estimation value.

[0122] 5. Harmonic source positioning

[0123] According to the preset threshold standard, the harmonic source positioning is realized based on the long-term data analysis result.

[0124] Based on the same inventive concept, the application further provides a computer device, which comprises one or more processors and a memory for storing one or more computer programs; the program comprises program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are configured to implement one or more instructions, and are specifically configured to load and execute one or more instructions in the computer storage medium to implement the above method.

[0125] It should be further noted that based on the same inventive concept, the present application also provides a computer storage medium, which stores a computer program, and the computer program is run by a processor to execute the above method. The storage medium can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of the computer readable storage medium include: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus.

[0126] It should be noted that unless otherwise defined, technical or scientific terms used in the present application should be understood as having the common meaning in the field of the present application to those having ordinary skill in the art. The terms "first", "second" and similar terms used in the present application do not denote any order, quantity or importance, but are only used to distinguish different components. The terms "include" or "contain" and similar terms mean that the elements or objects before the terms cover the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connected" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationships, and when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0127] The above is only the preferred embodiment of the present application, and does not limit other forms of the present application. Any skilled person in the art can modify or change the above disclosed technical content into equivalent embodiments with equivalent changes. However, any simple modification, equivalent change and modification of the above embodiments made according to the technical essence of the present application without departing from the technical solution content of the present application still falls within the protection scope of the present application.

[0128] The application is not limited to the above best mode, and anyone can derive other various forms of three-phase power distribution network harmonic source positioning methods considering topology switching and new energy harmonic pseudo measurement under the inspiration of the application. Any equivalent changes and modifications made within the scope of the application should be included in the scope of the application.

Claims

1. A method for locating harmonic sources in a three-phase distribution network, taking into account topology switching and pseudo-measurement of harmonics from new energy sources, characterized in that: A harmonic emission model is constructed based on the harmonic emission characteristics of distributed new energy sources. Through transfer learning, the harmonic characteristics of monitored nodes are transferred to unmonitored nodes to generate harmonic pseudo-measurement data. The harmonic pseudo-measurement data is used to compensate for the measurement gaps of unmonitored nodes. Using the aforementioned harmonic pseudo-measurement data and actual harmonic measurement data, harmonic state estimation is achieved through a multilayer sensing topology adaptive graph convolutional network. The network adapts to topology switching by dynamically fusing learnable weights of multi-order adjacency matrices and embedding the distribution network harmonic transfer equation as a physical constraint into the loss function, outputting estimated values ​​of harmonic injection current for each node. Based on the estimated values ​​of harmonic injection current, the network continuously accumulates data and, through statistical analysis of the statistical characteristics of harmonic injection current for each node, comprehensively identifies harmonic source nodes in the distribution network by combining preset criteria. The multilayer perceptual topological adaptive graph convolutional network includes a topological adaptive graph convolutional neural network, edge weight convolution, and a dynamic attention mechanism. The topological adaptive graph convolutional neural network obtains the next layer's node feature matrix by summing the product of a multi-order normalized adjacency matrix, the corresponding layer's node feature matrix, and the learnable weight matrix, and then processing this sum with an activation function. The node feature matrix contains voltage and current harmonic components. The multi-order adjacency matrix is ​​optimized using Dropout to discard redundant orders. The edge weight convolution dynamically generates the convolution kernel weight matrix based on edge features using a multilayer perceptron to capture the complex relationships between nodes in the graph structure; the edge features include line impedance parameters and topology connection types. The dynamic attention mechanism obtains attention coefficients by processing the original attention scores of neighboring nodes through the LeakyReLU activation function and then normalizing them through softmax, so as to differentiate the influence weight of neighboring node information on the new features of the current node. The loss function of the physical constraint includes the mean square error between the predicted harmonic state and the actual harmonic state, and a deviation term between the estimated measurement value and the actual measurement value; wherein, the deviation term is the square of the deviation when the absolute value of the deviation does not exceed the threshold, and is the product of the threshold and the absolute value of the deviation minus half of the square of the threshold when the absolute value of the deviation exceeds the threshold.

2. The method for locating harmonic sources in a three-phase distribution network, taking into account topology switching and pseudo-measurement of harmonics from new energy sources, as described in claim 1, is characterized in that: The harmonic emission characteristics of the distributed renewable energy source include the dynamic correlation between the renewable energy output power and the harmonic current. In the harmonic emission model... h The subharmonic current is a function of the power output of new energy sources, carrier frequency, grid impedance, DC side voltage, and grid voltage.

3. The method for locating harmonic sources in a three-phase distribution network, considering topology switching and pseudo-measurement of harmonics from new energy sources, as described in claim 1, is characterized in that: The transfer learning adopts federated transfer learning. The source domain is the harmonic monitoring data of the distributed new energy grid-connected points with harmonic monitoring devices installed, and the target domain is the distributed new energy without monitoring devices installed. The input of the transfer model includes the output power, node voltage and current of the distributed new energy, and the output is the harmonic current of the node. When there is no training data in the target domain, the parameters of the last layer of the pre-trained model in the source domain are perturbed and then applied to the target domain.

4. The method for locating harmonic sources in a three-phase distribution network, taking into account topology switching and pseudo-measurement of harmonics from new energy sources, as described in claim 1, is characterized in that: The statistical characteristics include the mean, variance, and 95th percentile of the estimated harmonic injection current at each node; where the mean reflects the average contribution level of the node to the harmonics, the variance reflects the degree of fluctuation of the estimated value around the mean, and the 95th percentile reflects the extreme level of the harmonic injection current.

5. A three-phase distribution network harmonic source location system considering topology switching and pseudo-measurement of new energy harmonics, used to implement the method as described in claim 1, characterized in that, include: The pseudo-measurement generation module is used to construct a harmonic emission model based on the harmonic emission characteristics of distributed new energy sources. Through transfer learning, the harmonic characteristics of monitored nodes are transferred to unmonitored nodes to generate harmonic pseudo-measurement data. The harmonic pseudo-measurement data is used to make up for the measurement gaps of unmonitored nodes. The harmonic state estimation module is used to estimate the harmonic state by using the pseudo-harmonic measurement data and the actual harmonic measurement data through a multi-layer sensing topology adaptive graph convolutional network. The network adapts to topology switching by dynamically fusing the learnable weights of the multi-order adjacency matrix and embedding the distribution network harmonic transfer equation as a physical constraint into the loss function, and outputs the estimated value of the harmonic injection current of each node. The harmonic source location module is used to continuously accumulate the estimated harmonic injection current value, and to identify the harmonic source nodes in the distribution network by statistically analyzing the statistical characteristics of the harmonic injection current of each node and combining the preset criteria.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-4.

Images