Transformer area identification optimization algorithm based on KCL and power frequency distortion characteristics

Abstract

The invention discloses a transformer area identification optimization algorithm based on KCL and power frequency distortion characteristics, and the algorithm carries out the precise analysis of the current of each node in a power system through employing the KCL, employs the feature extraction technologies of PCA, wavelet transform and the like, and employs the deep learning algorithms of a CNN, a long and short time memory network (LSTM) and the like. Characteristic vector extraction, model training and predictive analysis are carried out on power frequency (50Hz / 60Hz) distortion signals sent by transformer area identification equipment in a power grid, effective characteristic current information is extracted, and the accuracy and efficiency of transformer area identification are improved. According to the method, different transformer areas can be effectively distinguished in a complex power network, and the method is particularly suitable for an intelligent power distribution network with nonlinear loads and power electronic equipment. Through experimental verification, the algorithm shows excellent recognition performance in various operation states, and the monitoring and fault diagnosis capabilities of the power distribution system are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of power system signal processing and substation identification, and in particular to a substation identification optimization algorithm based on KCL and power frequency distortion characteristics. Background Art

[0002] With the continued development of smart grids and distribution automation systems, power systems are placing higher demands on the accuracy of structural identification and dynamic perception of substations (the customer areas served by distribution transformers). Substation identification is fundamental to applications such as distribution network operation and maintenance, fault location, and power quality monitoring. Its results directly impact the operating efficiency and dispatch strategies of distribution systems.

[0003] Currently, common methods for identifying substations rely primarily on static topological information of the distribution network and on-site construction records. However, due to the complexity of the grid structure and dynamic changes in actual operation, such as user migration, temporary cable adjustments, and switch operations, these static methods often fail to accurately reflect real-time electrical connection relationships. Furthermore, traditional methods rely heavily on user-side response signals and are susceptible to factors such as communication interference and signal attenuation, resulting in insufficient robustness in identification.

[0004] Furthermore, with the increasing integration of nonlinear loads and power electronic devices, the harmonic components and distortion characteristics of power frequency signals are becoming increasingly complex. Traditional substation identification methods based on steady-state electrical data (such as voltage and current averages) struggle to effectively exploit the inter-node electrical relationships inherent in these distorted power frequency signals, and therefore are unable to meet the recognition accuracy and real-time requirements of the new distribution network environment.

[0005] Some existing technologies have attempted to improve substation identification by introducing signal analysis or neural network methods, but the following problems still exist: First, Kirchhoff's current law (KCL) is not used as a constraint in the identification process, and there is a lack of consistent judgment on the logic of current flow; second, the power frequency pulse signal response and node distortion characteristics are not effectively combined for fusion analysis, and the stability and accuracy of the identification process need to be improved. Summary of the Invention

[0006] In order to solve the above problems, the present invention provides a substation identification optimization algorithm based on KCL and power frequency distortion characteristics. By combining the KCL current balance analysis and the power frequency distortion signal feature extraction method, a substation identification optimization algorithm that can adapt to the actual dynamic operation state of the power grid is constructed to improve the accuracy, robustness and real-time response capability of identification.

[0007] The present invention is achieved through the following technical solution: a substation identification optimization algorithm based on KCL and power frequency distortion characteristics, the algorithm comprising the following steps: Step 1: Data collection: Collect voltage, current and power frequency distortion signal data of multiple nodes in the power distribution system through smart meters or power quality analysis devices; Step 2: Pulse signal injection: 2 milliseconds before the zero crossing point of the first half cycle of the AC voltage, a high-amplitude, narrow-pulse-width power-frequency pulse signal with an amplitude of 6 amperes and a pulse width of 2 milliseconds is injected into the power line through each node to stimulate and form a response feature; Step 3: Initial topology construction: Constructing an initial electrical topology model based on the node response characteristics, electrical parameter data, and the existing distribution network structure; Step 4: Current balance analysis: Apply Kirchhoff's current law to analyze the balance of current flowing into and out of each node, which is used to identify the electrical connection relationship between nodes and potential topological anomalies. Step 5: Feature extraction: Performing principal component analysis or wavelet transform processing on the power frequency distortion signal to extract frequency domain or time-frequency domain feature vectors containing harmonic components, distortion factors, harmonic phases, etc.; Step 6: Area identification optimization: The current balance analysis results and the feature vector are input into an optimized recognition model, which is based on a convolutional neural network or a long short-term memory network structure to divide and identify the substation area; Step 7: Dynamic verification and adjustment: Combining historical operation data with real-time feedback, the substation division results are verified and evaluated, and the model parameters are dynamically updated or the structure is adjusted according to the system operation status to achieve adaptive optimization of the substation division results.

[0008] As a preferred technical solution, the acquisition period of the power frequency distortion signal is less than 1 second to ensure the response capability to fast dynamic changes in load.

[0009] As a preferred technical solution, the injection of the pulse signal is performed by the master node uniformly issuing an identification start command, and each sub-node synchronously performs the injection operation after receiving the command, and only one round of pulse signals is injected during each round of identification.

[0010] As a preferred technical solution, during the initial topology construction process, a graph structure is used to store the electrical connection relationship between nodes and lines, and the electrical parameters between nodes and edges are modeled in the form of a weighted graph.

[0011] As a preferred technical solution, in the current balance analysis step, a verification analysis of ΣI_in = ΣI_out is performed on each node based on the KCL principle. If the unbalance value exceeds a preset threshold, it is determined that the node has an abnormality or a topological mismatch connection.

[0012] As a preferred technical solution, the wavelet transform used in the feature extraction uses the Daubechies wavelet function to perform multi-scale decomposition on the power frequency signal of each node and extract energy spectrum features.

[0013] As a preferred technical solution, the dimension of the feature vector extracted by the principal component analysis PCA in step five is determined by the principal component with a cumulative contribution rate greater than 95%, ensuring a balance between feature representativeness and compression efficiency.

[0014] As a preferred technical solution, the optimized recognition model in step six is constructed based on the historical substation sample data set during the training phase. The model training adopts a cross-validation strategy and the particle swarm optimization algorithm is used to optimize the parameters.

[0015] As a preferred technical solution, in the dynamic verification and adjustment step, a rolling window method is used to correct the model recognition threshold or network layer structure in real time according to the size of the recognition error.

[0016] The beneficial effects of the present invention are as follows: by injecting a specific power frequency pulse signal before the voltage crosses zero, and combining it with KCL current balance analysis, the present invention can accurately restore the actual electrical connection relationship between each node, avoiding the identification deviation caused by the traditional method relying on static topology information; The present invention collects voltage, current and power frequency distortion signals from each node, introduces feature extraction methods such as principal component analysis (PCA) and wavelet transform, and combines deep learning models to fuse and identify multi-dimensional features, significantly improving the system's recognition robustness under complex working conditions such as nonlinear loads and interference from power electronic equipment. The present invention adopts a dynamic model adjustment mechanism, which can automatically update the optimization model parameters and substation division results based on real-time operation data and historical feedback results, realize continuous learning and adaptive optimization of substation identification, and adapt to the rapid changes in power grid operation status; By accurately identifying the structure and boundary relationship of substations, the present invention can provide reliable basic data support for power companies in applications such as load management, fault location, and distributed energy access control, further improving the efficiency and intelligence level of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 It is a principle block diagram of the present invention; Figure 2 It is the algorithm flow chart of the present invention; Figure 3 Schematic diagram of current balance analysis based on KCL of the present invention; Figure 4 This is a waveform diagram of the power frequency distortion signal feature extraction of the present invention; Figure 5 This is a schematic diagram of the station area identification optimization results of the present invention. DETAILED DESCRIPTION

[0019] All features disclosed in this specification, or all steps in the disclosed methods or processes, except mutually exclusive features and / or steps, can be combined in any manner.

[0020] Any feature disclosed in this specification (including any appended claims, abstract, and drawings), unless otherwise stated, may be replaced by other equivalent or similar features. In other words, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

[0021] like Figure 1-Figure 5 As shown, the present invention is an optimization algorithm for substation identification based on KCL and distortion characteristics, which includes collecting electrical data, injecting identification signals, building initial topology, performing current balance analysis, extracting node features, identifying substations based on optimization models and dynamically adjusting results, etc., forming a complete distributed substation intelligent identification technology chain.

[0022] The specific method is as follows: First, during the data collection phase, smart meters or power quality analyzers are deployed at each distribution node to collect key electrical parameters during grid operation in real time. Collected data includes three-phase AC voltage and current waveforms, as well as the node's power frequency distortion signal (including harmonic components, total harmonic distortion (THD), distortion factor, and harmonic phase angle). To enhance dynamic response, the power frequency signal sampling period is preferably set to less than 1 second, with a sampling frequency of at least 1 kHz to ensure frequency domain fidelity.

[0023] After data collection initialization is complete, the master node sends a unified zone identification start command to all sub-nodes. Using a high-precision clock synchronization mechanism, each sub-node injects a standardized identification signal into the power line through its pulse injection circuit 2 milliseconds before the zero crossing of the first half-cycle of the power frequency voltage. The signal is preferably a narrow pulse current with an amplitude of 6 amps and a duration of 2 milliseconds.

[0024] This pulse signal serves as an identification excitation signal and can quickly stimulate the electrical response differences between the nodes in the system, thereby assisting in establishing accurate electrical connection relationships.

[0025] The system then compares and analyzes the response current waveforms of each node before and after the pulse injection. By simultaneously measuring the corresponding voltage and current waveforms' mutation points, distortion trends, and amplitude response strength, a graph of the coupling relationships between the nodes can be obtained.

[0026] Combining the node's own voltage and current data and ledger information, an initial topology model is constructed using graph structure modeling, where nodes represent electricity users, edges represent line connection relationships, and edge weights reflect line resistance, reactance, conductance and other characteristics.

[0027] The system then performs a current balance analysis on this current data. This analysis, based on Kirchhoff's Current Law (KCL), imposes the constraint that the sum of the input currents must equal the sum of the output currents at each node, specifically ΣI_in = ΣI_out.

[0028] By calculating the residual ΔI of this formula and comparing it with the error tolerance set by the system (such as 5% of rated current), if the tolerance is exceeded, the node is marked as a potential abnormal node. This process can be used to automatically detect line misconnections, loop interference or missing branch problems.

[0029] After completing topology and connection accuracy analysis, the system extracts features from the power frequency distortion signals of the nodes. This feature extraction involves principal component analysis (PCA) and wavelet transform (WT). When using the wavelet transform, the Daubechies (db4) wavelet is used to perform a 3-5 level multi-scale decomposition of the power frequency signal, obtaining its energy spectrum and instantaneous amplitude changes in different frequency bands. When using PCA, the signal matrix is normalized and covariance analysis is performed, and the first several principal components with a cumulative contribution rate greater than 95% are extracted for subsequent modeling.

[0030] The KCL current balance features are combined with the distortion signal feature vector to form the input for the recognition model. The recognition model is a fusion deep learning model consisting of a convolutional neural network (CNN) and a long short-term memory (LSTM) network.

[0031] CNN is used to extract the spatial topological relationship and amplitude change pattern between nodes, and LSTM is used to model the load fluctuation trend and response inertia in multiple time periods.

[0032] The model training is based on the historical station area division dataset and adopts a cross-validation mechanism. During training, the particle swarm optimization algorithm is used to adjust hyperparameters such as network depth, convolution kernel size, and learning rate, ultimately achieving a balance between recognition accuracy and generalization ability.

[0033] After the model is trained and deployed, the system automatically divides the substations in each round of recognition. The recognition results are output as a label matrix and mapped into the power grid GIS system, allowing power grid dispatchers to visually view the substation structure and node affiliation relationships.

[0034] To improve the model's stability and long-term usability in complex power distribution environments, the system also features a dynamic verification and adaptive optimization module. This module analyzes the error trends in recognition results over rolling time windows (e.g., the last hour or 12 hours) and determines whether to trigger fine-tuning of the model structure based on indicators such as decreased recognition accuracy and sudden changes in node attribution.

[0035] Fine-tuning may include strategies such as increasing the LSTM memory step, adjusting the number of convolution channels, and changing the feature fusion layer structure to ensure that the model always maintains a good fit with the real power grid.

[0036] In addition, this implementation is applicable to complex grid environments with a large number of nonlinear loads, power electronic equipment, or distributed energy access; This algorithm can be deployed in a centralized master server for batch analysis, or it can be ported to edge computing terminals such as smart circuit breakers and substation monitoring devices to achieve lightweight local substation identification.

[0037] Through the algorithm structure and data flow described in the above implementation mode, the present invention can stably, quickly and accurately identify the boundaries of distribution system substations under different operating conditions, especially in scenarios with severe harmonic interference and intense load dynamics, showing excellent recognition ability and robustness, and has good engineering practicality and promotion value.

[0038] The proposed algorithm was applied to a city's smart distribution network. By installing smart meters and power quality analyzers, current, voltage, and harmonic data were collected at each node in real time. The algorithm first performed a KCL-based current distribution balance analysis, combined with power frequency harmonic feature extraction, and then utilized a particle swarm optimization algorithm to segment and optimize the substations. Experimental results showed that compared to traditional topology-based recognition methods, the proposed algorithm improved substation recognition accuracy by 20%, and significantly enhanced recognition robustness in harmonic interference environments. Furthermore, a dynamic adjustment function enabled the system to respond to changes in the grid topology in real time, ensuring the real-time and accurate recognition results.

[0039] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that are not conceived through creative work should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection defined in the claims.

Claims

1. A substation identification optimization algorithm based on KCL and power frequency distortion characteristics, characterized by: The algorithm comprises the following steps: Step 1: Data collection: Collect voltage, current and power frequency distortion signal data of multiple nodes in the power distribution system through smart meters or power quality analysis devices; Step 2: Pulse signal injection: 2 milliseconds before the zero crossing point of the first half cycle of the AC voltage, a high-amplitude, narrow-pulse-width power-frequency pulse signal with an amplitude of 6 amperes and a pulse width of 2 milliseconds is injected into the power line through each monitoring node to stimulate and form a response feature; Step 3: Initial topology construction: Constructing an initial electrical topology model based on the node response characteristics, electrical parameter data, and the existing distribution network structure; Step 4: Current balance analysis: Apply Kirchhoff's current law to analyze the balance of current flowing into and out of each node, which is used to identify the electrical connection relationship between nodes and potential topological anomalies. Step 5: Feature extraction: Performing principal component analysis or wavelet transform processing on the power frequency distortion signal to extract frequency domain or time-frequency domain feature vectors containing harmonic components, distortion factors, harmonic phases, etc.; Step 6: Area identification optimization: The current balance analysis results and the feature vector are input into an optimized recognition model, which is based on a convolutional neural network or a long short-term memory network structure to divide and identify the substation area; Step 7: Dynamic verification and adjustment: Combining historical operation data with real-time feedback, the substation division results are verified and evaluated, and the model parameters are dynamically updated or the structure is adjusted according to the system operation status to achieve adaptive optimization of the substation division results.

2. The substation area identification optimization algorithm based on KCL and power frequency distortion characteristics according to claim 1 is characterized by: The acquisition period of the power frequency distortion signal is less than 1 second to ensure the response capability to fast dynamic load changes.

3. The substation area identification optimization algorithm based on KCL and power frequency distortion characteristics according to claim 1 is characterized by: The injection of the pulse signal is carried out by the master node sending a unified identification start command, and each sub-node synchronously performs the injection operation after receiving the command, and only one round of pulse signals is injected during each round of identification.

4. The substation area identification optimization algorithm based on KCL and power frequency distortion characteristics according to claim 1 is characterized by: During the initial topology construction process, a graph structure is used to store the electrical connection relationship between nodes and lines, and the electrical parameters between nodes and edges are modeled in the form of a weighted graph.

5. The substation area identification optimization algorithm based on KCL and power frequency distortion characteristics according to claim 1 is characterized in that: In the current balance analysis step, a check analysis of ΣI_in = ΣI_out is performed on each node based on the KCL principle. If the unbalance value exceeds a preset threshold, it is determined that the node is abnormal or has a topological mismatch connection.

6. The substation area identification optimization algorithm based on KCL and power frequency distortion characteristics according to claim 1 is characterized by: The wavelet transform used in the feature extraction uses the Daubechies wavelet function to perform multi-scale decomposition on the power frequency signal of each node and extract the energy spectrum features.

7. The substation identification optimization algorithm based on KCL and power frequency distortion characteristics according to claim 1 is characterized by: The dimension of the feature vector extracted by the principal component analysis (PCA) in step five is determined by the principal component with a cumulative contribution rate greater than 95%, thereby ensuring a balance between feature representativeness and compression efficiency.

8. The substation area identification optimization algorithm based on KCL and power frequency distortion characteristics according to claim 1 is characterized by: The optimization recognition model in step six is constructed based on the historical substation sample data set during the training phase. The model training adopts a cross-validation strategy and the particle swarm optimization algorithm is used to optimize the parameters.

9. The substation area identification optimization algorithm based on KCL and power frequency distortion characteristics according to claim 1 is characterized in that: In the dynamic verification and adjustment step, the topology number of each node in the model, the identification time scale and the network layer structure are corrected in real time using a rolling time window method according to the size of the identification error.

Images