A method and system for tracing the source of harmonics in a power distribution network

Abstract

This invention discloses a method and system for tracing the source of harmonics in a distribution network, belonging to the field of distribution network monitoring technology. The method includes: S1, constructing an initial distribution network harmonic causal graph; S2, sensing real-time changes in the distribution network topology and dynamically updating the initial harmonic causal graph to obtain an updated harmonic causal graph; S3, decomposing the updated harmonic causal graph into multiple single-frequency causal subgraphs, performing causal intervention on each subgraph to separate background harmonics from user-side harmonics, calculating the contribution of each user-side harmonic source, and outputting the tracing results. This invention uses a causal graph neural network to replace traditional statistical correlation analysis, thereby effectively distinguishing between causal relationships and spurious correlations in harmonic propagation. By decomposing the harmonic causal graph into multiple independent single-frequency causal subgraphs, accurate decoupling of broadband coupled harmonics is achieved, avoiding error accumulation caused by frequency domain decomposition.

Description

Technical Field

[0001] This invention belongs to the field of power distribution network monitoring technology, specifically, it relates to a method and system for tracing the source of harmonics in power distribution networks. Background Technology

[0002] With the rapid development of new power systems, distributed power sources such as distributed photovoltaic and wind power, as well as power electronic equipment such as electric vehicle charging piles, variable frequency air conditioners, and industrial frequency converters, are being connected to the distribution network in large numbers, leading to increasingly serious harmonic pollution problems in the distribution network. Harmonics not only increase grid line losses and shorten the service life of electrical equipment, but also interfere with the normal operation of relay protection devices and communication systems, and may even cause power system safety accidents. Therefore, accurately identifying the location of harmonic sources and quantifying their contribution is of great significance for achieving precise harmonic control and responsibility allocation. Currently, the methods for tracing the source of harmonics in the distribution network are mainly divided into the following three categories:

[0003] Methods based on harmonic power flow and impedance: These methods calculate the contribution of each harmonic source by establishing a harmonic power flow model of the distribution network or measuring harmonic impedance. However, they suffer from drawbacks: relying on a pre-built static topology model and fixed impedance parameters, they cannot adapt to minute-level topology changes caused by frequent switching of distributed power sources and microgrid connection / disconnection. When the topology changes, the harmonic propagation path calculation is incorrect, and the source tracing accuracy decreases by more than 40%. Furthermore, these methods are based on the linear superposition assumption and are only applicable to scenarios with a single dominant harmonic source. When multiple types of power electronic devices interact to generate broadband coupled harmonics, the harmonic source contributions intertwine, making it impossible to distinguish nonlinear coupling components. When multiple sources are superimposed, the contribution quantification error generally exceeds 30%.

[0004] Signal processing-based methods: These methods separate harmonic signals using signal processing techniques such as independent component analysis and wavelet transform, thereby identifying harmonic sources. Their shortcomings include: relying on statistical correlation analysis, they cannot distinguish between causal relationships and spurious correlations in harmonic propagation; when background harmonic fluctuations in the upstream power grid exceed 2% of the present invention's value, the background harmonics completely mask the harmonic signals on the user side, leading to a false positive rate as high as 60%; furthermore, methods such as independent component analysis require the assumption that each harmonic source signal is independent, while in actual distribution networks, multiple harmonic sources are strongly coupled, failing to meet this assumption and further reducing the accuracy of source tracing.

[0005] Machine learning-based methods: These methods establish a mapping relationship between harmonic characteristics and harmonic source locations by training machine learning models such as neural networks and support vector machines. Their shortcomings include: reliance on large amounts of labeled historical data, poor model generalization ability, and the need to re-collect data and retrain the model when the power grid's operating state or topology changes, resulting in poor real-time performance. Furthermore, machine learning models are black-box models, lacking interpretability and failing to clearly define the physical path of harmonic propagation, making them unsuitable as a basis for harmonic liability allocation. Summary of the Invention

[0006] To address the aforementioned problems and technical deficiencies, this invention employs the following technical solution: a method for tracing the source of harmonics in a power distribution network, comprising the following steps:

[0007] S1. Construct an initial harmonic causal graph of the distribution network. The nodes of the causal graph include the bus, distributed power source, load and upper-level grid connection point in the distribution network. The edges represent the causal relationship of harmonic propagation between nodes.

[0008] S2. Real-time sensing of distribution network topology changes, dynamic incremental updating of the initial distribution network harmonic causality map, to obtain the updated harmonic causality map;

[0009] S3. Decompose the updated harmonic causal spectrum into multiple single-frequency causal subgraphs, perform causal intervention on each single-frequency causal subgraph to separate background harmonics from user-side harmonics, calculate the contribution of each user-side harmonic source and output the source tracing results.

[0010] Preferably, step S1 specifically includes:

[0011] The instantaneous values ​​of three-phase voltage and current at each monitoring node of the distribution network are collected, and noise reduction and normalization preprocessing are performed to extract the amplitude, phase and power characteristics of the 2nd to 25th harmonics.

[0012] An improved PC algorithm is used for causal structure learning, and the physical prior knowledge of the distribution network is introduced to constrain the direction and existence of causal edges. The physical prior knowledge includes power flow direction constraints and harmonic propagation attenuation constraints.

[0013] The initial harmonic causal graph of the distribution network is constructed based on the learned causal structure.

[0014] Furthermore, step S1 also includes:

[0015] Based on the initial harmonic causal graph of the distribution network, a graph attention network model is constructed, and the initial harmonic propagation model is obtained by training with the harmonic characteristics of each node as input.

[0016] Furthermore, step S2, which involves real-time sensing of distribution network topology changes, specifically includes:

[0017] The residual between the measured values ​​of harmonic characteristics of each node and the predicted values ​​of the initial harmonic propagation model is calculated in real time.

[0018] When the residual of a certain region exceeds a preset threshold within three consecutive sampling periods, it is determined that a topological change has occurred in that region.

[0019] Furthermore, the dynamic incremental update of the initial distribution network harmonic causality map in step S2 specifically includes:

[0020] Based on the spatial distribution characteristics of residuals, the region affected by topological changes can be located.

[0021] A sliding window mechanism is adopted, which uses the latest measurement data to relearn only the nodes and edges in the affected area, while preserving the causal relationships in the unchanged areas;

[0022] The parameters of the graph attention network model are adjusted using an incremental learning method to obtain the updated harmonic propagation model and the updated harmonic causality graph.

[0023] Preferably, step S3, which decomposes the updated harmonic causality graph into multiple single-frequency causality subgraphs, specifically includes:

[0024] According to the frequency dimension of the 2nd to 25th harmonics, the updated harmonic causal spectrum is decomposed into single-frequency causal subgraphs corresponding to each harmonic. Each single-frequency causal subgraph contains only the harmonic propagation causal relationship at the corresponding frequency.

[0025] Furthermore, step S3, prior to causal intervention on each single-frequency causal subgraph, also includes:

[0026] Mark the connection point of the upstream power grid as a potential background harmonic node;

[0027] Calculate the causal strength of the potential background harmonic nodes and all downstream nodes in the single-frequency causal subgraph;

[0028] When the causal strength exceeds a preset threshold, the upstream power grid connection point is determined to be a background harmonic source at the current frequency.

[0029] Furthermore, step S3, which involves causal intervention on each single-frequency causal subgraph to separate background harmonics from user-side harmonics, specifically includes:

[0030] For each single-frequency causal subgraph, a backdoor adjustment algorithm is used to perform causal intervention on nodes identified as background harmonic sources;

[0031] By blocking the backdoor path from the background harmonic source to the downstream node, the confusion effect of background harmonics on the downstream node is eliminated, and the node harmonic characteristics caused only by the user-side harmonic source are obtained.

[0032] Furthermore, step S3, calculating the contribution of each user-side harmonic source, specifically includes:

[0033] On each single-frequency causal subgraph after causal intervention, calculate the single-frequency causal contribution of each user-side harmonic source to each monitoring node;

[0034] The single-frequency causal contribution of each user-side harmonic source on all single-frequency causal subgraphs is summed to obtain the total contribution of each user-side harmonic source across the entire frequency band.

[0035] A power distribution network harmonic source tracing system includes:

[0036] The causal graph construction module is used to construct the initial causal graph of the distribution network harmonics and the initial harmonic propagation model;

[0037] The dynamic update module is used to sense changes in the distribution network topology in real time and to dynamically and incrementally update the causal graph and model.

[0038] The decoupling and separation module is used to decompose the updated causal graph into single-frequency causal subgraphs and perform causal intervention to separate background harmonics;

[0039] The contribution calculation module is used to calculate the contribution of each user-side harmonic source and output the source tracing results.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] (1) This invention uses a causal graph neural network to replace the traditional statistical correlation analysis, thereby effectively distinguishing the causal relationship and spurious correlation of harmonic propagation; by decomposing the harmonic causal graph into multiple independent single-frequency causal subgraphs, it achieves accurate decoupling of broadband coupled harmonics and avoids the error accumulation caused by frequency domain decomposition.

[0042] (2) This invention proposes a dynamic incremental update mechanism for causal graphs with anomaly detection, region localization and local update. It uses weighted residuals to perceive topological changes in real time, and only updates the causal structure and model parameters of the changed regions, while keeping the information of the unchanged regions unchanged. Compared with retraining the entire model, the amount of computation is effectively reduced and the topological change response time is significantly reduced.

[0043] (3) This invention applies the backdoor adjustment algorithm in causal intervention to background harmonic separation. By performing causal intervention on the identified background harmonic source nodes, all backdoor paths from them to downstream nodes are blocked, directly eliminating the confusion effect of background harmonics. There is no need to assume that background harmonics and user harmonics are independent of each other, thereby effectively reducing the misjudgment rate.

[0044] (4) The present invention adopts an edge-end collaborative computing architecture, which pushes lightweight tasks such as data preprocessing down to the edge gateway, while the cloud is only responsible for core computing tasks. The single-round source tracing calculation time is only 42ms, which can meet the needs of real-time harmonic governance. At the same time, the causal modeling method of the present invention introduces the physical prior knowledge of the distribution network, which has strong robustness to measurement noise and data missing. Even when 10% of the data of the present invention is missing, the source tracing accuracy can still remain stable. Attached Figure Description

[0045] In the attached diagram:

[0046] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. Generally, the components of the embodiments of the present invention described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0048] Example 1:

[0049] like Figure 1 As shown, a power distribution network harmonic source tracing system includes:

[0050] Data acquisition module: Composed of 12 intelligent power quality monitoring devices, deployed at the bus outlet, distributed power source grid connection point and main load access point of the distribution network, with sampling frequency set at 10kHz and synchronization error ≤1ms;

[0051] Edge computing gateway: Deployed in a 10kV substation, responsible for collecting raw data from various monitoring devices and completing preprocessing tasks such as noise reduction, normalization, and harmonic feature extraction;

[0052] Cloud computing platform: Deployed in the power company's data center, running causal graph construction module, dynamic update module, decoupling and separation module and contribution calculation module, using Intel Xeon Platinum 8375C processor and NVIDIA A10 graphics card to provide computing support;

[0053] Visualization terminal: Deployed in the power company's dispatch center to display harmonic source tracing results, propagation path diagrams, and responsibility allocation reports.

[0054] A method for tracing the source of harmonics in a power distribution network, in its specific implementation, includes the following steps:

[0055] S1. Initial Harmonic Cause-Effect Graph Construction of Distribution Network:

[0056] S101. Data Acquisition and Preprocessing: Instantaneous three-phase voltage and current data from 12 monitoring nodes of the distribution network were collected for 24 consecutive hours. Wavelet threshold denoising was used to remove measurement noise, and the data were normalized to the [0,1] interval using min-max normalization. The amplitudes of the 2nd to 25th harmonics were extracted using Fast Fourier Transform (FFT). Phase and active power Features, among which For harmonic order, Number the nodes;

[0057] S102. Improved PC Algorithm for Causal Structure Learning: An improved PC algorithm incorporating prior physical knowledge of the distribution network is used for causal structure learning. The core improvement lies in adding a physical prior constraint factor to the independence test score of the traditional PC algorithm. The formula is as follows:

[0058]

[0059] in: For nodes To the node The comprehensive score of causal relationships is used, with higher scores indicating stronger causal relationships. For a given set of conditions hour and Mutual information, used to measure statistical correlation; The physical prior weights, with values ​​ranging from [0,1], are determined by the following two constraints: power flow direction constraint: if the node The fundamental active power flows to the node ,but =1, otherwise =0; Harmonic propagation attenuation constraint: ,in This is the harmonic attenuation coefficient (0.02 / km in this embodiment). For nodes and Electrical distance between; overall weight = . ;when >0.5 and > When determining the existence of a slave node To the node The improved algorithm effectively eliminates false causal relationships and improves the accuracy of causal structure learning.

[0060] S103. Initial Causal Graph and Harmonic Propagation Model Construction: Based on the learned causal structure, an initial harmonic causal graph was constructed, comprising 12 monitoring nodes, 4 distributed power generation nodes, 9 load nodes, and 1 upstream grid connection point. A graph attention network model was then constructed based on this graph. The model consists of two graph attention layers, each with 8 attention heads, and uses ReLU as the activation function. Using the 2nd to 25th harmonic features of each node as input and the predicted harmonic voltage of the node as output, the initial harmonic propagation model was trained using the Adam optimizer with a batch size of 32, a learning rate of 0.001, and 100 training epochs.

[0061] S2. Topological change perception and dynamic incremental update of causal graph:

[0062] S201. Topology Change Anomaly Detection: Real-time calculation of the weighted residuals of harmonic characteristics of each node, the formula of which is:

[0063]

[0064] in: For nodes At any moment The weighted residuals; and They are nodes At any moment No. Measured values ​​and model predictions of subharmonic voltage; For harmonic order weights, = / 25, higher harmonics have higher weights because they are more sensitive to topology changes; set a residual threshold. =0.05pu, when the node residuals in a certain region exceed 50% within three consecutive sampling periods. > When this occurs, it is determined that a topological change has taken place in the region.

[0065] S202. Change Region Localization and Incremental Update of Causal Graph: Based on the spatial distribution characteristics of the residuals, the affected region of topological change is located as the subgraph composed of the node with the largest residual and its adjacent nodes. A sliding window mechanism is adopted, using measurement data from the most recent 100 sampling periods, and the improved PC algorithm is re-executed only on the nodes and edges within the affected region to learn the causal structure, preserving the causal relationships of the unchanged regions.

[0066] The parameters of the graph attention network model are adjusted using an incremental learning method, which updates only the model parameters corresponding to nodes in the affected region and freezes the parameters in the unchanged region. The learning rate of incremental learning is set to 0.0001 and the training epochs are 20, resulting in an updated harmonic propagation model and an updated harmonic causal graph. Compared with retraining the entire model, this method reduces the computational cost by more than 85% and the topology change response time is less than 100ms.

[0067] S3. Multi-frequency decoupling, background harmonic separation, and contribution calculation:

[0068] S301. Single-frequency causal subgraph decomposition: According to the frequency dimension of the 2nd to 25th harmonics, the updated harmonic causal spectrum is decomposed into 24 independent single-frequency causal subgraphs. ( =2,3,...,25). Each single-frequency causal subgraph It only includes the causal relationship of harmonic propagation at the corresponding frequency h, and its edge weight is the comprehensive score of causal edges at that frequency. ;

[0069] S302, Background Harmonic Source Identification:

[0070] Mark the upstream power grid connection point as a potential background harmonic node. Calculate its relationship with the single-frequency causal subgraph. The core formula for the harmonic propagation causality intensity of all downstream nodes is:

[0071]

[0072] in: For the upper-level power grid node In frequency Next pair of nodes The causal strength; From node To the node The set of all causal paths; This represents the number of paths.

[0073] Set a causality strength threshold =0.7, when > At that time, the connection point of the upper-level power grid is determined to be the current frequency. Background harmonic sources.

[0074] S303, Causal Intervention Background Harmonic Separation: For each single-frequency causal subgraph determined to contain a background harmonic source. A backdoor adjustment algorithm is used to adjust background harmonic nodes. Causal intervention is performed to eliminate its confusion effect on downstream nodes; the nodes after intervention... The characteristic formula for harmonic voltage is:

[0075]

[0076] in: For causal intervention node In frequency Harmonic voltages below; Background harmonic nodes Values The probability distribution; For nodes Perform intervention operations After that, node The expected value of harmonic voltage.

[0077] This formula can block all backdoor paths from the background harmonic source to the downstream node, and obtain the node harmonic characteristics caused only by the user-side harmonic source, without assuming that the background harmonics and user harmonics are independent of each other.

[0078] S304, Multi-Harmonic Source Contribution Measurement: Single-Frequency Causal Subgraph after Causal Intervention Above, calculate the harmonic sources on each user side. For each monitoring node Single-frequency causal contribution:

[0079]

[0080] in This represents the total number of harmonic sources on the user side.

[0081] The weighted summation of the single-frequency causal contributions of each user-side harmonic source across all single-frequency causal subgraphs yields the total contribution across the entire frequency band:

[0082]

[0083] in The total number of monitoring nodes, The harmonic order weights.

[0084] Source tracing results output: The system outputs the location, type, contribution of each harmonic, and total contribution of each harmonic source on the user side, and generates a harmonic propagation path diagram and a harmonic responsibility division report; when the total contribution of a certain harmonic source exceeds 30%, the system automatically issues an early warning message to prompt the power supply company to carry out targeted management.

[0085] The above embodiments only illustrate preferred embodiments of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A method for tracing the source of harmonics in a power distribution network, characterized in that, Includes the following steps: S1. Construct an initial harmonic causal graph of the distribution network. The nodes of the causal graph include the bus, distributed power source, load and upper-level grid connection point in the distribution network. The edges represent the causal relationship of harmonic propagation between nodes. S2. Real-time sensing of distribution network topology changes, dynamic incremental updating of the initial distribution network harmonic causality map, to obtain the updated harmonic causality map; S3. Decompose the updated harmonic causal spectrum into multiple single-frequency causal subgraphs, perform causal intervention on each single-frequency causal subgraph to separate background harmonics from user-side harmonics, calculate the contribution of each user-side harmonic source and output the source tracing results.

2. A method for tracing the source of harmonics in a distribution network according to claim 1, characterized in that, Step S1 specifically includes: The instantaneous values ​​of three-phase voltage and current at each monitoring node of the distribution network are collected, and noise reduction and normalization preprocessing are performed to extract the amplitude, phase and power characteristics of the 2nd to 25th harmonics. An improved PC algorithm is used for causal structure learning, and the physical prior knowledge of the distribution network is introduced to constrain the direction and existence of causal edges. The physical prior knowledge includes power flow direction constraints and harmonic propagation attenuation constraints. The initial harmonic causal graph of the distribution network is constructed based on the learned causal structure.

3. A method for tracing the source of harmonics in a distribution network according to claim 2, characterized in that, Step S1 also includes: Based on the initial harmonic causal graph of the distribution network, a graph attention network model is constructed, and the initial harmonic propagation model is obtained by training with the harmonic characteristics of each node as input.

4. A method for tracing the source of harmonics in a distribution network according to claim 3, characterized in that, Step S2, which involves real-time sensing of distribution network topology changes, specifically includes: The residual between the measured values ​​of harmonic characteristics of each node and the predicted values ​​of the initial harmonic propagation model is calculated in real time. When the residual of a certain region exceeds a preset threshold within three consecutive sampling periods, it is determined that a topological change has occurred in that region.

5. A method for tracing the source of harmonics in a distribution network according to claim 4, characterized in that, The dynamic incremental update of the initial distribution network harmonic causality map in step S2 specifically includes: Based on the spatial distribution characteristics of residuals, the region affected by topological changes can be located. A sliding window mechanism is adopted, which uses the latest measurement data to relearn only the nodes and edges in the affected area, while preserving the causal relationships in the unchanged areas; The parameters of the graph attention network model are adjusted using an incremental learning method to obtain the updated harmonic propagation model and the updated harmonic causality graph.

6. A method for tracing the source of harmonics in a distribution network according to claim 1, characterized in that, Step S3, which decomposes the updated harmonic causality graph into multiple single-frequency causality subgraphs, specifically includes: According to the frequency dimension of the 2nd to 25th harmonics, the updated harmonic causal spectrum is decomposed into single-frequency causal subgraphs corresponding to each harmonic. Each single-frequency causal subgraph contains only the harmonic propagation causal relationship at the corresponding frequency.

7. A method for tracing the source of harmonics in a distribution network according to claim 6, characterized in that, Step S3, before performing causal intervention on each single-frequency causal subgraph, also includes: Mark the connection point of the upstream power grid as a potential background harmonic node; Calculate the causal strength of the potential background harmonic nodes and all downstream nodes in the single-frequency causal subgraph; When the causal strength exceeds a preset threshold, the upstream power grid connection point is determined to be a background harmonic source at the current frequency.

8. A method for tracing the source of harmonics in a distribution network according to claim 7, characterized in that, Step S3 involves causal intervention on each single-frequency causal subgraph to separate background harmonics from user-side harmonics, specifically including: For each single-frequency causal subgraph, a backdoor adjustment algorithm is used to perform causal intervention on nodes identified as background harmonic sources; By blocking the backdoor path from the background harmonic source to the downstream node, the confusion effect of background harmonics on the downstream node is eliminated, and the node harmonic characteristics caused only by the user-side harmonic source are obtained.

9. A method for tracing the source of harmonics in a power distribution network according to claim 8, characterized in that, Step S3, calculating the contribution of each user-side harmonic source, specifically includes: On each single-frequency causal subgraph after causal intervention, calculate the single-frequency causal contribution of each user-side harmonic source to each monitoring node; The single-frequency causal contribution of each user-side harmonic source on all single-frequency causal subgraphs is summed to obtain the total contribution of each user-side harmonic source across the entire frequency band.

10. A harmonic source tracing system for a power distribution network, characterized in that, include: The causal graph construction module is used to construct the initial causal graph of the distribution network harmonics and the initial harmonic propagation model; The dynamic update module is used to sense changes in the distribution network topology in real time and to dynamically and incrementally update the causal graph and model. The decoupling and separation module is used to decompose the updated causal graph into single-frequency causal subgraphs and perform causal intervention to separate background harmonics; The contribution calculation module is used to calculate the contribution of each user-side harmonic source and output the source tracing results.

Images