A novel method, system, device and medium for real-time positioning of harmonics in power system

Abstract

The present invention provides a new type of power system harmonic real-time positioning method, system, equipment and medium, the method is to continuously obtain the fundamental waveform data and voltage waveform monitoring data set of the new power system at multiple sampling moments, generate a standard harmonic waveform data set according to the fundamental waveform data at each sampling moment, and then input the voltage waveform monitoring data set and the standard harmonic waveform data set into a pre-constructed harmonic characteristic analysis network to perform harmonic characteristic analysis to obtain a harmonic characteristic data set, and then generate a corresponding harmonic characteristic graph according to the harmonic characteristic data of each harmonic collection point in the harmonic characteristic data set at each sampling moment and the network topology of the new power system, and input the harmonic characteristic graphs at all sampling moments into a pre-constructed graph neural network to perform harmonic positioning analysis to obtain the corresponding system harmonic positioning result. The present invention can perform real-time harmonic positioning analysis on the new power system, and effectively improve the harmonic positioning efficiency and positioning accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a novel method, system, equipment and medium for real-time positioning of harmonics in power systems. Background Art

[0002] With the increasing application of power electronic equipment in new power systems, harmonic pollution in the electrical environment has also increased, which has a great impact on the power supply quality and operation safety of the power system. Therefore, reliable tracing and locating of power system harmonics can provide reliable guarantee for the safe and stable operation of new power systems.

[0003] The existing harmonic source tracing and positioning methods for new power systems mainly include two categories: time-frequency domain analysis methods and deep learning methods. Among them, the time-frequency domain analysis method uses frequency domain analysis, time domain analysis or complex frequency domain analysis to decompose the waveform to obtain the phase, frequency, amplitude and other characteristics of different harmonics, and then uses intervention or non-intervention methods to analyze the harmonic type and locate the source of the harmonic based on the harmonic characteristics. This time-frequency domain analysis method based on fixed rules involves a mutual constraint relationship between the computational burden and the function complexity, and it is difficult to effectively balance the computational efficiency and computational accuracy; at the same time, the existing deep learning method directly inputs the collected power system monitoring data into the network model for harmonic status evaluation, which is very easy to overfit and makes it difficult to cope with the high-complexity power grid harmonics in actual deployment. Therefore, it is urgent to provide a new power system harmonic source tracing and positioning method that can take into account both harmonic positioning efficiency and positioning accuracy. Summary of the invention

[0004] The purpose of the present invention is to provide a novel real-time harmonic positioning method for a power system. The harmonic characteristic diagram of the power system is constructed by the harmonic characteristic analysis results of a harmonic characteristic analysis network including multiple parallel twin harmonic characteristic analysis subnetworks, and then a graph neural network model is used to perform harmonic positioning analysis based on the harmonic characteristic diagram. The powerful linear / nonlinear fitting ability of the neural network can be effectively utilized to perform real-time harmonic positioning analysis on the new power system, thereby effectively improving the harmonic positioning efficiency and positioning accuracy.

[0005] In order to achieve the above objectives, it is necessary to provide a new method, system, equipment and medium for real-time positioning of harmonics in power system in response to the above technical problems.

[0006] In a first aspect, an embodiment of the present invention provides a novel method for real-time positioning of harmonics in a power system, the method comprising the following steps:

[0007] Continuously acquiring fundamental waveform data and voltage waveform monitoring data sets of the new power system at multiple sampling moments; the voltage waveform monitoring data sets include voltage waveform data of each harmonic collection point;

[0008] Generate a corresponding standard harmonic waveform data set according to the fundamental waveform data at each sampling moment; the standard harmonic waveform data set includes a number of fundamental wave integer multiple standard harmonic waveform data;

[0009] Inputting the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-built harmonic characteristic analysis network for harmonic characteristic analysis to obtain a corresponding harmonic characteristic data set; the harmonic characteristic analysis network includes a plurality of parallel twin harmonic characteristic analysis sub-networks;

[0010] Generate a corresponding harmonic characteristic diagram according to the harmonic characteristic data of all harmonic collection points in the harmonic characteristic data set at each sampling moment and the network topology of the novel power system;

[0011] The harmonic feature graphs of all sampling moments are input into the pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results.

[0012] Furthermore, the twin harmonic characteristic analysis subnetwork includes two multilayer perceptron networks with shared weights and one fully connected neural network; the number of the twin harmonic characteristic analysis subnetworks is consistent with the number of the fundamental integer multiple standard harmonic waveform data; the method for constructing the harmonic characteristic analysis network includes:

[0013] Acquire historical monitoring data of the novel power system; the historical monitoring data includes historical fundamental waveform data and corresponding historical voltage waveform monitoring data set;

[0014] Performing multi-modal noise superposition processing on the voltage waveform data of each harmonic collection point in the historical voltage waveform monitoring data set to obtain a corresponding pre-processed historical voltage waveform data set;

[0015] According to the historical fundamental waveform data, a historical standard harmonic waveform data set is generated, and according to the historical standard harmonic waveform data set and the preprocessed historical voltage waveform data set, each twin harmonic characteristic analysis sub-network is trained based on a random sampling method to obtain the harmonic characteristic analysis network.

[0016] Furthermore, the output result of the twin harmonic characteristic analysis subnetwork is a harmonic characteristic triplet; the harmonic characteristic triplet includes confidence, amplitude ratio and phase difference between the fundamental integral multiple standard harmonic waveform data contained in the voltage waveform data and the fundamental waveform data;

[0017] The step of training each twin harmonic characteristic analysis sub-network based on the random sampling method to obtain the harmonic characteristic analysis network includes:

[0018] Based on the cross entropy loss function, the confidence in each harmonic characteristic triple is comprehensively evaluated to obtain the corresponding confidence estimation deviation;

[0019] Based on the two-dimensional Gaussian distribution loss function, the amplitude ratio and phase difference between the fundamental integral multiple standard harmonic waveform data and the fundamental waveform number data contained in the voltage waveform data in each harmonic characteristic triplet are comprehensively evaluated to obtain the corresponding harmonic component estimation deviation;

[0020] According to the confidence estimation deviation and the harmonic component estimation deviation, each twin harmonic characteristic analysis sub-network is back-propagated and updated for optimization until a preset training termination condition is reached, thereby obtaining the harmonic characteristic analysis network.

[0021] Furthermore, the two-dimensional Gaussian distribution loss function is expressed as:

[0022]

[0023] In the formula, represents the harmonic component estimation bias; and Respectively represent the amplitude ratio and phase difference between the fundamental wave integer multiple standard harmonic waveform data and the fundamental wave waveform number data contained in the voltage waveform data in the i-th harmonic characteristic triplet of each harmonic acquisition point; and They represent the mean and standard deviation of the amplitude ratio in the two-dimensional Gaussian distribution respectively; and They represent the mean and standard deviation of the phase difference in the two-dimensional Gaussian distribution respectively; represents the correlation coefficient between the amplitude ratio and the phase difference in the two-dimensional Gaussian distribution; k represents the number of harmonic characteristic triplets.

[0024] Furthermore, the step of inputting the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-built harmonic characteristic analysis network for harmonic characteristic analysis to obtain the corresponding harmonic characteristic data set includes:

[0025] The voltage waveform data of each harmonic collection point in each voltage waveform monitoring data set are combined with each fundamental wave integer multiple standard harmonic waveform data in the corresponding standard harmonic waveform data set and input into the corresponding twin harmonic characteristic analysis subnetwork for harmonic characteristic analysis to obtain the harmonic characteristic data of each harmonic collection point at the corresponding sampling time; the harmonic characteristic data includes a plurality of harmonic characteristic triplets;

[0026] According to the harmonic characteristic data of all harmonic collection points at each sampling time, a corresponding harmonic characteristic data set is generated.

[0027] Furthermore, the graph neural network includes an encoding network, a preset time attention network and a decoding network; the encoding network and the decoding network both include a multi-layer fully connected neural network; the loss function of the graph neural network is expressed as:

[0028]

[0029] In the formula,

[0030]

[0031] in, It represents the comprehensive loss of harmonic positioning; represents the harmonic localization loss of each graph node; Represents all nodes and node relationships in the entire topology domain; Indicates the weight of each node data after taking the upper boundary weight; and is a constant value; and Represents the distance between the two nodes closest to the outlier; Indicates the network weight size before the boundary weight constraint; Represents the plane distance of the harmonic generation point output by the graph neural network from the node; Indicates the true plane distance of the harmonic generation point from the node.

[0032] Furthermore, the step of inputting the harmonic feature graphs at all sampling moments into a pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results includes:

[0033] Inputting the harmonic feature graphs at all sampling moments into the encoding network for feature encoding to obtain corresponding harmonic feature vectors;

[0034] Inputting the harmonic feature vector into the preset time attention network for feature fusion to obtain the corresponding fused harmonic feature;

[0035] The fused harmonic feature is input into the decoding network for decoding processing to obtain the system harmonic positioning result.

[0036] In a second aspect, an embodiment of the present invention provides a novel power system harmonic real-time positioning system, the system comprising:

[0037] A voltage waveform acquisition module is used to continuously acquire fundamental waveform data and voltage waveform monitoring data sets of the new power system at multiple sampling moments; the voltage waveform monitoring data set includes voltage waveform data of each harmonic acquisition point;

[0038] A standard harmonic generation module is used to generate a corresponding standard harmonic waveform data set according to the fundamental waveform data at each sampling moment; the standard harmonic waveform data set includes a number of fundamental integer multiple standard harmonic waveform data;

[0039] A harmonic characteristic analysis module is used to input the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-built harmonic characteristic analysis network for harmonic characteristic analysis to obtain a corresponding harmonic characteristic data set; the harmonic characteristic analysis network includes a plurality of parallel twin harmonic characteristic analysis sub-networks;

[0040] A topology diagram construction module, used to generate a corresponding harmonic characteristic diagram according to the harmonic characteristic data of all harmonic collection points in the harmonic characteristic data set at each sampling moment and the network topology of the novel power system;

[0041] The harmonic location analysis module is used to input the harmonic feature graphs of all sampling moments into the pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results.

[0042] In a third aspect, an embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0043] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0044] The above-mentioned application provides a new type of real-time harmonic positioning method, system, computer equipment and storage medium for power system harmonics. The method realizes the continuous acquisition of fundamental waveform data of the new power system at multiple sampling moments and a voltage waveform monitoring data set including voltage waveform data of each harmonic collection point. After generating a standard harmonic waveform data set including several fundamental integer multiple standard harmonic waveform data according to the fundamental waveform data at each sampling moment, the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set are input into a pre-constructed harmonic characteristic analysis network including multiple parallel twin harmonic characteristic analysis subnetworks for harmonic characteristic analysis to obtain the corresponding harmonic characteristic data set, and then the corresponding harmonic characteristic graph is generated according to the harmonic characteristic data of each harmonic collection point in the harmonic characteristic data set at each sampling moment and the network topology of the new power system, and the harmonic characteristic graphs at all sampling moments are input into a pre-constructed graph neural network for harmonic positioning analysis to obtain the corresponding system harmonic positioning result. Technical solution. Compared with the existing technology, this new method for real-time harmonic positioning in power systems constructs a harmonic characteristic diagram of the power system based on the harmonic characteristic analysis results of a harmonic characteristic analysis network including multiple parallel twin harmonic characteristic analysis subnetworks, and then uses a graph neural network model to perform harmonic positioning analysis based on the harmonic characteristic diagram. It can effectively utilize the powerful linear / nonlinear fitting ability of neural networks to perform real-time harmonic positioning analysis on new power systems, effectively improving the harmonic positioning efficiency and positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 1 is a flow chart of a novel method for real-time harmonic location in a power system according to an embodiment of the present invention;

[0046] Figure 2 is a schematic diagram of the structure of a harmonic characteristic analysis network in an embodiment of the present invention;

[0047] Figure 3 It is a schematic diagram of the structure of the neural network in the embodiment of the present invention;

[0048] Figure 4 2 is a schematic diagram of the structure of a novel power system harmonic real-time positioning system according to an embodiment of the present invention;

[0049] Figure 5 It is a diagram of the internal structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical scheme and beneficial effects of the present application clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the embodiments described below are part of the embodiments of the present invention and are only used to illustrate the present invention, but are not used to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] The novel power system harmonic real-time positioning method provided by the present invention can be understood as a novel power system harmonic source tracing positioning method that takes into account both harmonic positioning efficiency and positioning accuracy, based on the existing deep learning method, which is very prone to overfitting when evaluating the power system harmonic state, making it difficult to cope with the application status of the highly complex power grid harmonics in actual deployment. The following embodiments will explain the novel power system harmonic real-time positioning method of the present invention in detail.

[0052] In one embodiment, Figure 1 As shown, a novel power system harmonic real-time positioning method is provided, comprising the following steps:

[0053] S11. Continuously obtain fundamental waveform data and voltage waveform monitoring data sets of the new power system at multiple sampling moments; wherein, the fundamental waveform data can be obtained by using an existing power system fundamental waveform extraction algorithm, which is not specifically limited here; the voltage waveform monitoring data set can be understood as the voltage waveform monitoring data of each harmonic collection point collected according to a fixed duration and frequency by all harmonic collectors pre-deployed on the power system, that is, the voltage waveform monitoring data set includes the voltage waveform data of each harmonic collection point; it should be noted that all harmonic collection points in the power system can be selected according to actual application requirements, which is not specifically limited here, and the number of continuously obtained sampling moments can be determined according to actual application requirements. In order to ensure the timeliness of harmonic positioning, this embodiment preferably selects 5 sampling moments.

[0054] S12. Generate a corresponding standard harmonic waveform data set according to the fundamental waveform data at each sampling time; the standard harmonic waveform data set includes a number of fundamental integer multiple standard harmonic waveform data; wherein, the number of fundamental integer multiple standard harmonic waveform data can be set according to actual application requirements, and the standard harmonic waveform data set constructed in this embodiment preferably includes all standard waveforms of 1 to 10 times the fundamental waveform. The generation of the corresponding standard harmonic waveform data set can use PyCharm software and Python language to write a standard waveform generator for generating standard sinusoidal waveform data of fundamental integer multiple harmonics as comparison benchmark data used for comparative analysis of various voltage waveform data in the harmonic characteristic analysis network.

[0055] S13, input the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-constructed harmonic characteristic analysis network for harmonic characteristic analysis to obtain a corresponding harmonic characteristic data set; wherein the harmonic characteristic analysis network can be understood as a neural network model for performing harmonic characteristic analysis on the voltage waveform data of each harmonic collection point to determine whether there are harmonics in the voltage waveform data, as well as information such as the proportion of various harmonic amplitudes and phase differences; in order to improve the accuracy and efficiency of harmonic detection, this embodiment preferably adopts Figure 2 The harmonic characteristic analysis network shown includes a plurality of parallel twin harmonic characteristic analysis subnetworks, and the number of the twin harmonic characteristic analysis subnetworks is consistent with the number of the fundamental wave integer multiple standard harmonic waveform data.

[0056] Specifically, the twin harmonic characteristic analysis subnetwork includes two multilayer perceptron networks with shared weights and one fully connected neural network, wherein the multilayer perceptron network with shared weights is used to extract harmonic features from voltage waveform data and one fundamental integer multiple standard harmonic waveform data, respectively, and the generated harmonic features are input into the fully connected neural network for comparison and analysis to obtain the corresponding harmonic characteristic triplet, that is, the output result of the twin harmonic characteristic analysis subnetwork is the harmonic characteristic triplet , which can be expressed as:

[0057]

[0058] in, It is the confidence level, indicating whether there is a common frequency part between the voltage waveform data and the compared fundamental wave integer multiple standard harmonic waveform data. Its value range is (0,1). If it is greater than 0.5, it means there is a common frequency part, and if it is less than 0.5, it means there is no common frequency part. is the result of the harmonic component estimation subtask of the twin harmonic characteristic analysis subnetwork, is the similarity, which indicates the amplitude ratio of the fundamental integral multiple standard harmonic waveform data contained in the voltage waveform data to the fundamental waveform data. is the offset, which indicates the phase difference between the fundamental wave integer multiple standard harmonic waveform data and the fundamental wave waveform data contained in the voltage waveform data; Function represents the activation function, that is, the ReLu function; and It is a twin network, which represents two multi-layer perceptron networks with shared weights, that is, when extracting voltage waveform data And fundamental wave integer multiple standard harmonic waveform data When the features are used, the same network weights are used, and the two networks average their loss values ​​before propagation during back propagation.

[0059] In order to ensure the anti-interference ability and robustness of the harmonic characteristic analysis network, this embodiment preferably adopts adding multimodal noise to the training data during the model training process. Specifically, the method for constructing the harmonic characteristic analysis network includes:

[0060] Obtain historical monitoring data of the new power system; the historical monitoring data includes historical fundamental waveform data and corresponding historical voltage waveform monitoring data sets; wherein, the method for obtaining the historical fundamental waveform data and the historical voltage waveform monitoring data set can refer to the relevant description of the above-mentioned fundamental waveform data and voltage waveform monitoring data set; the waveform length of the historical voltage waveform data of each harmonic collection point in the specific historical voltage waveform monitoring data set can be determined according to the sampling frequency and fundamental frequency actually used for the sensor. For example, considering the fluctuation of the sensor data, two cycles of waveform data can be selected, and 100 points are evenly sampled in each cycle, that is, the voltage waveform data with a length of 200.

[0061] The voltage waveform data of each harmonic collection point in the historical voltage waveform monitoring data set is subjected to multimodal noise superposition processing to obtain a corresponding pre-processed historical voltage waveform data set; wherein the multimodal noise superposition processing can be understood as adding Gaussian distribution noise to each voltage waveform data, and the corresponding pre-processed historical voltage waveform data can be expressed as:

[0062]

[0063] in, Represents the voltage waveform data in the historical voltage waveform monitoring data set; For voltage waveform data Superposition of multimodal noise The voltage waveform data afterwards; represents Gaussian distributed noise, is the mean of the Gaussian distribution, is the variance of Gaussian distribution; is Poisson distributed noise, is the Poisson distribution parameter.

[0064] According to the historical fundamental waveform data, a historical standard harmonic waveform data set is generated, and according to the historical standard harmonic waveform data set and the preprocessed historical voltage waveform data set, each twin harmonic characteristic analysis sub-network is trained based on a random sampling method to obtain the harmonic characteristic analysis network; wherein, the method for obtaining the historical standard harmonic waveform data set as comparison data can also refer to the generation description of the above-mentioned standard harmonic waveform data set, which will not be repeated here; it should be noted that the data length of each standard waveform data in the historical standard harmonic waveform data set needs to be consistent with the length and frequency of the above-mentioned historical voltage waveform data.

[0065] In order to ensure the reliability of the training of each twin harmonic characteristic analysis subnetwork, this embodiment preferably uses random sampling combined with a loss function based on cross entropy and two-dimensional Gaussian distribution to update and train each twin harmonic characteristic analysis subnetwork on the basis of randomly superimposing a type of noise on each voltage waveform data to construct a model training set, so as to ensure the accuracy of the harmonic characteristic triplet output by the twin harmonic characteristic analysis subnetwork. Specifically, the steps of training each twin harmonic characteristic analysis subnetwork based on the random sampling method to obtain the harmonic characteristic analysis network include:

[0066] Based on the cross entropy loss function, the confidence in each harmonic characteristic triple is comprehensively evaluated to obtain the corresponding confidence estimation deviation; wherein the cross entropy loss function is expressed as:

[0067]

[0068] Wherein, k represents the number of harmonic characteristic triplets, that is, the length of the waveform data; Indicates the confidence estimation deviation, ensuring more accurate judgment of whether a certain frequency band contains harmonics; is the confidence value in the i-th harmonic characteristic triplet of each harmonic acquisition point output by the twin harmonic characteristic analysis subnetwork, express The corresponding true value.

[0069] Based on the two-dimensional Gaussian distribution loss function, the amplitude ratio and phase difference between the fundamental integer multiple standard harmonic waveform data and the fundamental waveform number data contained in the voltage waveform data in each harmonic characteristic triplet are comprehensively evaluated to obtain the corresponding harmonic component estimation deviation; wherein, the two-dimensional Gaussian distribution loss function is expressed as:

[0070]

[0071] In the formula, represents the harmonic component estimation bias; and They respectively represent the amplitude ratio and phase difference between the fundamental wave integer multiple standard harmonic waveform data and the fundamental wave waveform number data contained in the voltage waveform data of the i-th harmonic characteristic triplet of each harmonic acquisition point output by the twin harmonic characteristic analysis subnetwork; and They represent the mean and standard deviation of the amplitude ratio in the two-dimensional Gaussian distribution respectively; and They represent the mean and standard deviation of the phase difference in the two-dimensional Gaussian distribution respectively; It represents the correlation coefficient between the amplitude ratio and the phase difference in the two-dimensional Gaussian distribution; k represents the number of harmonic characteristic triplets, that is, the length of the waveform data.

[0072] According to the confidence estimation deviation and the harmonic component estimation deviation, each twin harmonic characteristic analysis subnetwork is back-propagated and updated to optimize until the preset training termination condition is reached, thereby obtaining the harmonic characteristic analysis network; wherein, the process of back-propagating and updating each twin harmonic characteristic analysis subnetwork can be understood as successively transmitting the confidence estimation deviation and the harmonic component estimation deviation obtained by each round of model training calculation of each twin harmonic characteristic analysis subnetwork to the twin harmonic characteristic analysis subnetwork for back-propagation update of the parameters of each neuron in the neural network, so that the neural network can simultaneously determine whether the corresponding harmonic exists The ratio of the amplitude of the harmonic to the fundamental With phase It should be noted that the preset training termination condition can be selected according to actual application requirements, such as reaching a preset number of iterative training times or reaching a preset analysis accuracy, etc., and is not specifically limited here.

[0073] This embodiment constructs a model training set by randomly superimposing a type of noise on each voltage waveform data, and adopts random sampling to sample a fixed number of data from massive historical voltage waveform data as training samples, and combines the loss function design based on cross entropy and two-dimensional Gaussian distribution to update the training of the twin harmonic characteristic analysis sub-network. This can ensure the correct direction of the neural network gradient descent and the reliability of the twin harmonic characteristic analysis sub-network training, as well as the robustness of the harmonic characteristic analysis network and the accuracy of the harmonic characteristic analysis, thereby providing a reliable analysis basis for the subsequent harmonic source tracing and positioning based on harmonic characteristics.

[0074] After constructing a harmonic characteristic analysis network capable of performing real-time analysis on the voltage waveform data of each harmonic collection point through the above method steps, the voltage waveform monitoring data set and the corresponding standard harmonic waveform data set used as a comparison benchmark can be simultaneously input into the harmonic characteristic analysis network for real-time harmonic characteristic analysis to obtain a harmonic characteristic data set including the harmonic characteristic data of each harmonic collection point. Specifically, the step of inputting the voltage waveform monitoring data set and the corresponding standard harmonic waveform data set at each sampling moment into the pre-constructed harmonic characteristic analysis network for harmonic characteristic analysis to obtain the corresponding harmonic characteristic data set includes:

[0075] The voltage waveform data of each harmonic collection point in each voltage waveform monitoring data set are combined with each fundamental integer multiple standard harmonic waveform data in the corresponding standard harmonic waveform data set and input into the corresponding twin harmonic characteristic analysis sub-network for harmonic characteristic analysis to obtain the harmonic characteristic data of each harmonic collection point at the corresponding sampling time; the harmonic characteristic data include multiple harmonic characteristic triplets; assuming that the number of twin harmonic characteristic analysis sub-networks is 10, that is, the standard harmonic waveform data set includes 1 to 10 times the fundamental integer multiple standard harmonic waveform data, then the corresponding harmonic characteristic data acquisition process of each harmonic collection point can be understood as: the voltage waveform data of each harmonic collection point and 1 times the fundamental integer multiple standard harmonic waveform data are input into the first twin harmonic characteristic analysis sub-network for harmonic characteristic analysis to obtain the voltage waveform data of each harmonic collection point and The comparison analysis results (harmonic characteristic triplets) of 1 times the fundamental integer multiple standard harmonic waveform data (fundamental waveform data); the voltage waveform data of each harmonic collection point and the 2 times the fundamental integer multiple standard harmonic waveform data are input into the second twin harmonic characteristic analysis sub-network for harmonic characteristic analysis, and the comparison analysis results (harmonic characteristic triplets) of the voltage waveform data of each harmonic collection point and the 2 times the fundamental integer multiple standard harmonic waveform data are obtained; and by analogy, the comparison results of the voltage waveform data of each harmonic collection point and the different fundamental integer multiple standard harmonic waveform data can be given by the 10 twin harmonic characteristic analysis sub-networks in the harmonic characteristic analysis network, that is, 10 harmonic characteristic triplets can be obtained for each harmonic collection point. Assuming that the waveform data of each harmonic collection point is 200 in length, the output of the harmonic characteristic analysis network is data of 30 dimensions. It should be noted that the analysis process of the twin harmonic characteristic analysis subnetwork in acquiring the harmonic characteristic data of each harmonic collection point at each sampling moment mentioned above can be found in the relevant description involved in the aforementioned twin harmonic characteristic analysis subnetwork structure, which will not be repeated here; in addition, the number of the above-mentioned twin harmonic characteristic analysis subnetworks is only an exemplary description and can be adjusted according to actual needs in actual applications.

[0076] According to the harmonic characteristic data of all harmonic collection points at each sampling time, a corresponding harmonic characteristic data set is generated.

[0077] This embodiment introduces distribution contrast monitoring technology and uses contrast learning methods to set up multiple groups of parallel twin neural networks to decompose the harmonic detection problem into the detection of specific harmonic components. It also introduces multimodal noise to build a model training data set and designs a loss function based on two-dimensional Gaussian distribution and cross entropy. It comprehensively detects the presence of harmonics in the signal waveform and the amplitude, phase and other information of the harmonics to train and construct a harmonic characteristic analysis network, which can effectively improve the anti-interference ability and robustness of the harmonic characteristic analysis network while effectively improving the accuracy of the harmonic component detection results.

[0078] S14. Generate a corresponding harmonic characteristic diagram based on the harmonic characteristic data of all harmonic collection points in the harmonic characteristic data set at each sampling moment and the network topology of the new power system; wherein the harmonic characteristic diagram at each sampling moment can be understood as a virtual network mapping relationship diagram of the power system that integrates the harmonic characteristic data of each harmonic collection point at that moment; the harmonic characteristic diagram can be understood as using the harmonic collection point as the graph node in the harmonic characteristic diagram, and based on the circuit connection relationship and spatial position relationship of each harmonic collection point in the network topology of the new power system as the node edge, starting from the right direction of the main control device, from near to far, counterclockwise, each harmonic collection point is numbered in sequence from 1 according to positive integers, and the edge relationship of the graph node is constructed based on the connection relationship between each number. It should be noted that the node information data of each graph node is the multiple harmonic characteristic triplets extracted by the harmonic characteristic analysis network mentioned above. The harmonic characteristic diagram constructed in the above way takes into account the harmonic characteristic data and spatial position data of each harmonic collection point, which can provide strong support for the graph neural network to perform harmonic identification and effective source tracing and positioning.

[0079] S15. Input the harmonic characteristic graphs of all sampling moments into the pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results; wherein, the graph neural network can select any network model that can analyze the harmonic characteristic graph, in order to ensure that on the basis of comprehensive extraction and analysis of the spatial topological graph relationship of the overall power system, the location information of the abnormal nodes generating harmonics can be accurately generated on the original topological graph, such as Figure 3 As shown, the present embodiment preferably sets the graph neural network to include an encoding network (Encoder), a preset time attention network and a decoding network (Decoder), and the encoding network and the decoding network both include multi-layer fully connected neural networks, wherein the encoding network includes 3 layers of fully connected layer neural networks (the number of neurons in each network is 100, 50, and 20, respectively), that is, the feature length obtained after processing by the encoding network is 20; the decoding network is composed of 3 layers of fully connected networks (the number of neurons in each network is 20, 50, and 100, respectively); correspondingly, the preset time attention network can be understood as a time attention mechanism introduced to enhance the neural network's ability to extract and analyze time dimension information in consideration of problems such as time asynchrony and electromagnetic disturbance in the power grid signal transmission process. The preset time attention network can be implemented with reference to the existing time attention mechanism and is not specifically limited here.

[0080] In this embodiment, a graph neural network consisting of an encoding network, a preset time attention network and a decoding network is used to significantly reduce the impact of time asynchrony in the power system on harmonic source tracing through feature extraction and fusion methods, facilitating more accurate harmonic source tracing. The output of the graph neural network includes one or more triple results , where n represents the graph node number in the harmonic feature graph, and Represents the plane distance of the harmonic generation point from the nth graph node.

[0081] The above-mentioned pre-constructed graph neural network can be understood as a network model with optimal parameters obtained based on the harmonic characteristic data generated by the aforementioned harmonic characteristic analysis network training combined with the corresponding power system network topology structure training; in order to ensure the accuracy of the graph neural network model in predicting the harmonic anomaly location, this embodiment preferably designs a loss function with boundary weights as the loss function of the graph neural network, which is specifically expressed as:

[0082]

[0083] In the formula,

[0084]

[0085] in, It represents the comprehensive loss of harmonic positioning; represents the harmonic localization loss of each graph node; Represents all nodes and node relationships in the entire topology domain; Indicates the weight of each node data after taking the upper boundary weight; and is a constant value; and Represents the distance between the two nodes closest to the outlier; Indicates the network weight size before the boundary weight constraint; Represents the plane distance of the harmonic generation point output by the graph neural network from the node; Indicates the true plane distance of the harmonic generation point from the node.

[0086] Specifically, the step of inputting the harmonic feature graphs at all sampling moments into a pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results includes:

[0087] The harmonic feature graphs at all sampling moments are input into the encoding network for feature encoding to obtain corresponding harmonic feature vectors; wherein, the process of obtaining the harmonic feature vectors can refer to the relevant technical implementation of feature encoding using a multi-layer fully connected network.

[0088] The harmonic feature vector is input into the preset time attention network for feature fusion to obtain the corresponding fused harmonic feature; wherein, the process of obtaining the fused harmonic feature can refer to the relevant technical implementation of the specific selected time attention mechanism.

[0089] The fused harmonic feature is input into the decoding network for decoding processing to obtain the system harmonic positioning result.

[0090] In this embodiment, feature extraction is performed based on a coding network including a multi-layer fully connected network, which can effectively weaken the influence of data fluctuations, data noise and data transmission speed on the waveform; at the same time, by introducing a time attention network to perform feature matching and fusion based on the time dimension, the perception ability of the graph neural network for time dimension data is effectively improved, and the time positioning of key harmonics and the time synchronization of the overall system are achieved, thereby reducing the impact of time asynchrony on data analysis results and providing reliable guarantee for the accuracy of harmonic source tracing and positioning results in the power system.

[0091] The embodiment of the present application continuously obtains the fundamental waveform data of the new power system at multiple sampling moments and a voltage waveform monitoring data set including voltage waveform data of each harmonic collection point, generates a standard harmonic waveform data set including several standard harmonic waveform data of integer multiples of the fundamental waveform according to the fundamental waveform data at each sampling moment, inputs the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-constructed harmonic characteristic analysis network including multiple parallel twin harmonic characteristic analysis subnetworks to perform harmonic characteristic analysis to obtain the corresponding harmonic characteristic data set, then generates the corresponding harmonic characteristic graph according to the harmonic characteristic data of each harmonic collection point in the harmonic characteristic data set at each sampling moment and the network topology of the new power system, and The harmonic characteristic diagrams of all sampling moments are input into a pre-constructed graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results. This solution effectively solves the application defect of existing harmonic location of new power systems that it is difficult to balance positioning efficiency and positioning accuracy. The harmonic characteristic diagram of the power system is constructed by the harmonic characteristic analysis results of a harmonic characteristic analysis network based on multiple parallel twin harmonic characteristic analysis subnetworks, and then a graph neural network model is used to perform harmonic location analysis based on the harmonic characteristic diagram. This can ensure the high efficiency and generalization of the harmonic characteristic analysis network and the graph neural network, and use the powerful linear / nonlinear fitting ability of the neural network to perform real-time harmonic location analysis on the new power system, effectively improving the harmonic location efficiency and positioning accuracy.

[0092] It should be noted that although the steps in the above flowchart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0093] In one embodiment, Figure 4 As shown, a novel power system harmonic real-time positioning system is provided, the system comprising:

[0094] The voltage waveform acquisition module 1 is used to continuously acquire fundamental waveform data and voltage waveform monitoring data sets of the new power system at multiple sampling moments; the voltage waveform monitoring data sets include voltage waveform data of each harmonic acquisition point;

[0095] The standard harmonic generation module 2 is used to generate a corresponding standard harmonic waveform data set according to the fundamental waveform data at each sampling moment; the standard harmonic waveform data set includes a number of fundamental integer multiple standard harmonic waveform data;

[0096] The harmonic characteristic analysis module 3 is used to input the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-built harmonic characteristic analysis network for harmonic characteristic analysis to obtain the corresponding harmonic characteristic data set; the harmonic characteristic analysis network includes a plurality of parallel twin harmonic characteristic analysis sub-networks;

[0097] A topology map construction module 4 is used to generate a corresponding harmonic characteristic map according to the harmonic characteristic data of all harmonic collection points in the harmonic characteristic data set at each sampling moment and the network topology of the novel power system;

[0098] The harmonic location analysis module 5 is used to input the harmonic feature graphs of all sampling moments into a pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results.

[0099] For the specific limitations of the new power system harmonic real-time positioning system, please refer to the limitations of the new power system harmonic real-time positioning method mentioned above, and the corresponding technical effects can also be obtained equivalently, which will not be repeated here. Each module in the above-mentioned new power system harmonic real-time positioning system can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0100] Figure 5 FIG. 1 shows an internal structure diagram of a computer device in an embodiment, and the computer device may specifically be a terminal or a server. Figure 5As shown, the computer device includes a processor, a memory, a network interface, a display, a camera and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a new type of power system harmonic real-time positioning method is realized. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0101] It can be understood by those skilled in the art that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the present application scheme, and does not constitute a limitation on the computer device to which the present application scheme is applied. The specific computing device may include more or fewer components than shown in the figure, or combine certain components, or have the same component arrangement.

[0102] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0103] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0104] In summary, the embodiment of the present invention provides a novel power system harmonic real-time positioning method and system, wherein the novel power system harmonic real-time positioning method realizes continuous acquisition of fundamental waveform data of the novel power system at multiple sampling moments and a voltage waveform monitoring data set including voltage waveform data of each harmonic acquisition point, after generating a standard harmonic waveform data set including several fundamental integer multiple standard harmonic waveform data according to the fundamental waveform data at each sampling moment, the voltage waveform monitoring data set and the corresponding standard harmonic waveform data set at each sampling moment are input into a pre-constructed harmonic characteristic analysis network including multiple parallel twin harmonic characteristic analysis subnetworks for harmonic characteristic analysis to obtain the corresponding harmonic characteristic data set, and then the harmonic characteristic data set of each harmonic is concentrated in the harmonic characteristic data set at each sampling moment. A technical solution is proposed that the harmonic characteristic data of the wave collection points and the network topology of the new power system are used to generate the corresponding harmonic characteristic diagram, and the harmonic characteristic diagrams at all sampling moments are input into a pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results. This method constructs the harmonic characteristic diagram of the power system based on the harmonic characteristic analysis results of a harmonic characteristic analysis network including multiple parallel twin harmonic characteristic analysis subnetworks, and then uses a graph neural network model to perform harmonic location analysis based on the harmonic characteristic diagram. On the basis of ensuring the high efficiency and generalization of the harmonic characteristic analysis network and the graph neural network, the powerful linear / nonlinear fitting ability of the neural network can be used to perform real-time harmonic location analysis on the new power system, effectively improving the harmonic location efficiency and positioning accuracy.

[0105] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0106] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A novel method for real-time harmonic location in power system, characterized in that: The method comprises the following steps: Continuously acquiring fundamental waveform data and voltage waveform monitoring data sets of the new power system at multiple sampling moments; the voltage waveform monitoring data sets include voltage waveform data of each harmonic collection point; Generate a corresponding standard harmonic waveform data set according to the fundamental waveform data at each sampling moment; the standard harmonic waveform data set includes a number of fundamental wave integer multiple standard harmonic waveform data; Inputting the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-built harmonic characteristic analysis network for harmonic characteristic analysis to obtain a corresponding harmonic characteristic data set; the harmonic characteristic analysis network includes a plurality of parallel twin harmonic characteristic analysis sub-networks; Generate a corresponding harmonic characteristic diagram according to the harmonic characteristic data of all harmonic collection points in the harmonic characteristic data set at each sampling moment and the network topology of the novel power system; The harmonic feature graphs at all sampling moments are input into the pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results; The twin harmonic characteristic analysis subnetwork includes two multilayer perceptron networks with shared weights and one fully connected neural network; the number of the twin harmonic characteristic analysis subnetworks is consistent with the number of the fundamental integer multiple standard harmonic waveform data; the construction method of the harmonic characteristic analysis network includes: Acquire historical monitoring data of the novel power system; the historical monitoring data includes historical fundamental waveform data and corresponding historical voltage waveform monitoring data set; Performing multi-modal noise superposition processing on the voltage waveform data of each harmonic collection point in the historical voltage waveform monitoring data set to obtain a corresponding pre-processed historical voltage waveform data set; Generate a historical standard harmonic waveform data set according to the historical fundamental waveform data, and train each twin harmonic characteristic analysis subnetwork based on a random sampling method according to the historical standard harmonic waveform data set and the preprocessed historical voltage waveform data set to obtain the harmonic characteristic analysis network; The output result of the twin harmonic characteristic analysis subnetwork is a harmonic characteristic triplet; the harmonic characteristic triplet includes confidence, amplitude ratio and phase difference between the fundamental integral multiple standard harmonic waveform data contained in the voltage waveform data and the fundamental waveform data; The step of training each twin harmonic characteristic analysis sub-network based on the random sampling method to obtain the harmonic characteristic analysis network includes: Based on the cross entropy loss function, the confidence in each harmonic characteristic triple is comprehensively evaluated to obtain the corresponding confidence estimation deviation; Based on the two-dimensional Gaussian distribution loss function, the amplitude ratio and phase difference between the fundamental integral multiple standard harmonic waveform data and the fundamental waveform number data contained in the voltage waveform data in each harmonic characteristic triplet are comprehensively evaluated to obtain the corresponding harmonic component estimation deviation; According to the confidence estimation deviation and the harmonic component estimation deviation, each twin harmonic characteristic analysis subnetwork is back-propagated and updated for optimization until a preset training termination condition is reached, thereby obtaining the harmonic characteristic analysis network; The two-dimensional Gaussian distribution loss function is expressed as: In the formula, represents the harmonic component estimation bias; and Respectively represent the amplitude ratio and phase difference between the fundamental wave integer multiple standard harmonic waveform data and the fundamental wave waveform number data contained in the voltage waveform data in the i-th harmonic characteristic triplet of each harmonic acquisition point; and They represent the mean and standard deviation of the amplitude ratio in the two-dimensional Gaussian distribution respectively; and They represent the mean and standard deviation of the phase difference in the two-dimensional Gaussian distribution respectively; represents the correlation coefficient between the amplitude ratio and the phase difference in the two-dimensional Gaussian distribution; k represents the number of harmonic characteristic triplets.

2. The novel power system harmonic real-time positioning method according to claim 1, characterized in that: The step of inputting the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-built harmonic characteristic analysis network for harmonic characteristic analysis to obtain the corresponding harmonic characteristic data set comprises: The voltage waveform data of each harmonic collection point in each voltage waveform monitoring data set are combined with each fundamental wave integer multiple standard harmonic waveform data in the corresponding standard harmonic waveform data set and input into the corresponding twin harmonic characteristic analysis subnetwork for harmonic characteristic analysis to obtain the harmonic characteristic data of each harmonic collection point at the corresponding sampling time; the harmonic characteristic data includes a plurality of harmonic characteristic triplets; According to the harmonic characteristic data of all harmonic collection points at each sampling time, a corresponding harmonic characteristic data set is generated.

3. The novel power system harmonic real-time positioning method according to claim 1, characterized in that: The graph neural network includes an encoding network, a preset time attention network and a decoding network; the encoding network and the decoding network both include a multi-layer fully connected neural network; the loss function of the graph neural network is expressed as: In the formula, in, It represents the comprehensive loss of harmonic positioning; represents the harmonic localization loss of each graph node; Represents all nodes and node relationships in the entire topology domain; Indicates the weight of each node data after taking the upper boundary weight; and is a constant value; and Represents the distance between the two nodes closest to the outlier; Indicates the network weight size before the boundary weight constraint; Represents the plane distance of the harmonic generation point output by the graph neural network from the node; Indicates the true plane distance of the harmonic generation point from the node.

4. The novel power system harmonic real-time positioning method as claimed in claim 3 is characterized in that: The step of inputting the harmonic feature graphs at all sampling moments into a pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results includes: Inputting the harmonic feature graphs at all sampling moments into the encoding network for feature encoding to obtain corresponding harmonic feature vectors; Inputting the harmonic feature vector into the preset time attention network for feature fusion to obtain the corresponding fused harmonic feature; The fused harmonic feature is input into the decoding network for decoding processing to obtain the system harmonic positioning result.

5. A new type of power system harmonic real-time positioning system, characterized in that: The novel power system harmonic real-time positioning method as claimed in any one of claims 1 to 4 is applied, wherein the system comprises: A voltage waveform acquisition module is used to continuously acquire fundamental waveform data and voltage waveform monitoring data sets of the new power system at multiple sampling moments; the voltage waveform monitoring data set includes voltage waveform data of each harmonic acquisition point; A standard harmonic generation module is used to generate a corresponding standard harmonic waveform data set according to the fundamental waveform data at each sampling moment; the standard harmonic waveform data set includes a number of fundamental integer multiple standard harmonic waveform data; A harmonic characteristic analysis module is used to input the voltage waveform monitoring data set at each sampling moment and the corresponding standard harmonic waveform data set into a pre-built harmonic characteristic analysis network for harmonic characteristic analysis to obtain a corresponding harmonic characteristic data set; the harmonic characteristic analysis network includes a plurality of parallel twin harmonic characteristic analysis sub-networks; A topology diagram construction module, used to generate a corresponding harmonic characteristic diagram according to the harmonic characteristic data of all harmonic collection points in the harmonic characteristic data set at each sampling moment and the network topology of the novel power system; The harmonic location analysis module is used to input the harmonic feature graphs of all sampling moments into the pre-built graph neural network for harmonic location analysis to obtain the corresponding system harmonic location results.

6. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images