Medium and low voltage line abnormity identification method based on deep neural network

Abstract

The invention discloses a medium and low voltage line abnormity identification method based on a deep neural network, and the method achieves the comprehensive coverage of key nodes of a power distribution network through the four-stage deployment of station-line-transformer-household, completely captures the abnormal signals at the tail end of a branch line, provides sufficient data support for deep feature learning, solves a problem of a conventional monitoring blind area, and improves the recognition precision of the power distribution network. A convolutional neural network (CNN) and long short-term memory (LSTM) network mixed model automatically extracts time-frequency domain joint features, manual feature design is not needed, subtle changes of transient waveforms are effectively captured, the anomaly identification precision is improved, the limitation that a traditional method depends on artificial experience is broken through, potential fault hidden dangers can be found in advance, sufficient processing time is won for operation and maintenance personnel, and the operation and maintenance efficiency is improved. Through rapid positioning, the fault processing period is shortened, the power failure time is reduced, the power supply reliability is improved, fuzzy C-means clustering is combined with topology analysis, accurate positioning of abnormal sections is achieved, the manual inspection workload is reduced, the fault processing efficiency is improved, and the operation and maintenance cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of smart distribution network automation technology, specifically a method for identifying anomalies in medium and low voltage lines based on deep neural networks. Background Technology

[0002] As a crucial component of the power grid, the power distribution network plays a vital role in ensuring and improving people's livelihoods. Currently, domestic power distribution lines are characterized by their large scale, complex structure, numerous and varied equipment types, and complex operating environments. With economic and social development, the requirements for power supply reliability are increasingly stringent. Medium and low-voltage distribution networks, as the final link in the power system, are characterized by complex topologies (multi-branch, radial), diverse equipment types (cable / overhead hybrid), and harsh operating environments (frequent lightning strikes and tree-related faults). Statistics show that faults in the domestic 10kV distribution network account for over 80% of all power grid faults, with 70% of these faults developing from early anomalies (such as insulation degradation and poor contact).

[0003] Existing anomaly identification methods have many problems, such as: insufficient data acquisition coverage: existing monitoring systems are mainly concentrated at the outgoing end of substations, and the monitoring of key nodes such as branch boxes and distribution transformers is lacking, resulting in incomplete capture of fault features; limited feature extraction capabilities: traditional methods rely on manually designed features, which makes it difficult to capture the joint time-frequency domain features of transient waveforms, and the ability to identify complex faults is weak; poor model adaptability: rule-based discrimination models cannot adapt to dynamic changes in distribution network topology (such as line switching and distributed power source access), and the anomaly identification results are not stable enough.

[0004] Therefore, it is necessary to invent a method for identifying anomalies in medium and low voltage lines based on deep neural networks to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of existing technologies, this invention provides a method for identifying anomalies in medium and low voltage lines based on deep neural networks. By collecting data across the entire domain, learning deep features, and fusing multi-source information, it enables early warning and accurate location of anomalies in medium and low voltage lines, thereby improving the efficiency of power distribution network operation and maintenance and the reliability of power supply.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying anomalies in medium and low voltage lines based on deep neural networks, comprising the following steps:

[0007] Step 1: Deploy transient recording devices at the outgoing terminals, branch boxes, distribution transformers, and important user terminals of medium and low voltage lines to build a wide-area data acquisition network covering "station-line-transformer-user". The deployment density and sampling rate of each node meet the needs of full-area monitoring to ensure complete acquisition of transient data.

[0008] Step 2: Perform time synchronization correction, noise suppression, and standardization on the collected transient data to generate an input sequence suitable for deep neural networks, ensuring data quality and consistency;

[0009] Step 3: Construct a hybrid model that includes a convolutional neural network (CNN) and a long short-term memory network (LSTM). After inputting a standardized sequence, the convolutional layer extracts waveform spatial features, the LSTM layer captures temporal dependencies, and the outputs the probability and severity of the anomaly type, thus achieving deep feature learning and classification.

[0010] Step 4: Combine Bayesian network fusion device ledger, environmental data and historical fault records to analyze the causes of anomalies and trigger graded early warnings, providing multi-dimensional decision support;

[0011] Step 5: Using the fuzzy C-means clustering algorithm, abnormal segments are located based on the differences in transient features between nodes and topological correlation. Topological verification is used to improve the positioning accuracy and achieve precise fault location.

[0012] As a further description of the above technical solution, the transient waveform recording device in step one supports a triggered acquisition mode, which can record transient waveform segments before and after a fault, and realize real-time data transmission through a communication network.

[0013] As a further description of the above technical solution, the convolutional layer of the hybrid network of Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) in step three contains multiple convolutional operations, extracting multi-scale waveform features through convolutional kernels of different sizes, and the LSTM layer contains an appropriate number of memory units to capture temporal features.

[0014] As a further description of the above technical solution, the Bayesian network described in step four integrates multi-source information and outputs the probability distribution of abnormal causes, providing a basis for operation and maintenance decisions.

[0015] As a further description of the above technical solution, the fuzzy C-means clustering algorithm described in step five performs clustering analysis based on multi-dimensional feature vectors, combines topological structure to exclude non-critical segments, and accurately locates abnormal areas.

[0016] As a further description of the above technical solution, the anomaly identification results are visualized through a geographic information system, marking the location of the anomaly segment, the membership degree of the node anomaly, and the topological importance, and supporting real-time viewing on mobile terminals.

[0017] As a further description of the above technical solution, the deep neural network model optimizes parameters through continuous incremental learning, adapts to new fault modes, improves identification performance, and ensures the system's adaptability and advancement.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. This invention achieves comprehensive coverage of key nodes in the distribution network through a four-level deployment of "station-line-transformer-customer," fully capturing abnormal signals at the end of branch lines, providing sufficient data support for deep feature learning, solving the problem of blind spots in traditional monitoring, and using a hybrid CNN and LSTM model to automatically extract joint features in the time and frequency domains without the need for manual feature design. This effectively captures subtle changes in transient waveforms, improves the accuracy of anomaly identification, and breaks through the limitations of traditional methods that rely on human experience. It can detect potential fault hazards in advance, giving maintenance personnel sufficient time to handle them. By quickly locating faults, it shortens the fault handling cycle, reduces power outage time, and improves power supply reliability. Fuzzy C-means clustering combined with topology analysis enables precise location of abnormal sections, reduces the workload of manual inspections, improves fault handling efficiency, and lowers maintenance costs.

[0020] 2. The deep neural network model of this invention can accurately identify various anomaly types, including insulation degradation, poor contact, and high-resistance grounding, providing a basis for targeted operation and maintenance, achieving fine-grained differentiation of fault types. By fusing multi-source information through Bayesian networks, it deeply analyzes the causes of anomalies, and a hierarchical early warning mechanism ensures the rational allocation of operation and maintenance resources, improving response efficiency and optimizing the operation and maintenance decision-making process. This method is compatible with different voltage levels, topologies, and line types, and can seamlessly integrate with existing distribution automation systems, possessing broad engineering application prospects. Edge computing deployment capabilities further enhance real-time performance, suitable for the intelligent upgrade needs of urban and rural distribution networks. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0022] Figure 1 The flowchart shows the method for identifying anomalies in medium and low voltage lines based on deep neural networks provided by this invention. Detailed Implementation

[0023] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] I. Wide-area transient data acquisition and preprocessing

[0026] Level 4 Node Deployment Strategy

[0027] Station-level nodes (outgoing line terminals of substations): Equipped with high-precision waveform recording devices to collect full-frequency voltage / current waveforms at a sampling rate that meets the requirements for capturing transient characteristics, covering all outgoing circuits, ensuring the integrity of transient data of the main line. The transient waveform recording devices support trigger-type acquisition mode, which can record transient waveform segments before and after a fault, and realize real-time data transmission through the communication network to meet the requirements for rapid response.

[0028] Line-level nodes (branch boxes / ring network cabinets): Deploy intelligent transient monitoring terminals to adapt to the sampling rate triggering acquisition of transient segments before and after faults in branch line monitoring. On average, one terminal is set up at each reasonable distance of the line to capture abnormal signals of the branch line.

[0029] Transformer node (distribution transformer): Integrates transient waveform recording function in intelligent terminal to meet the sampling rate monitoring needs of low voltage side monitoring waveform, covering most distribution transformers, and focusing on the impact of low voltage side anomalies on the power grid.

[0030] Household-level nodes (important users): Intermittently collect user-side current waveforms through smart meters to help identify the impact of user equipment anomalies on the power grid.

[0031] Data preprocessing workflow

[0032] Time synchronization correction: Based on the BeiDou satellite time synchronization system, ensure that the synchronization accuracy of data across all nodes meets the requirements of time series analysis, and achieve accurate alignment of waveform data across multiple nodes.

[0033] Noise suppression processing: Wavelet denoising technology is used to filter out power frequency interference and high-frequency noise through multi-layer decomposition and threshold processing, while retaining the effective characteristic frequency bands of transient waveforms to ensure the integrity of characteristic information.

[0034] Standardization processing: Normalize the collected data to eliminate differences in units, provide standardized input for deep neural networks, and improve the stability of model training.

[0035] II. Construction of Deep Neural Network Anomaly Detection Model

[0036] Hybrid CNN and LSTM network architecture

[0037] Input layer: Receives standardized waveform sequences, covering the characteristic window of typical transient processes, including steady-state characteristics before the fault and transient characteristic changes after the fault.

[0038] Convolutional Neural Network (CNN) Layers: Extract spatial features of waveforms through multiple convolutional operations. The first layer uses a suitable number of large-sized convolutional kernels to capture the macroscopic features of the waveform; the second layer uses a larger number of smaller-sized convolutional kernels to extract the detailed features of the waveform, gradually compressing the feature dimension and enhancing the feature abstraction capability.

[0039] Long Short-Term Memory (LSTM) layer: Contains an appropriate number of memory units to capture the time dependence of waveform sequences, identify the evolution pattern of features over time, and effectively model the temporal characteristics of transient processes.

[0040] Output layer: Through a fully connected layer and a classification function, it outputs the probability and severity of the anomaly type, supports the classification of various anomaly states such as insulation degradation, poor contact, and high-resistance grounding, and provides a basis for subsequent decision-making.

[0041] Model training and optimization

[0042] Dataset construction: Collect historical transient waveform data, including normal and various abnormal samples, and divide them into training set, validation set and test set according to a reasonable ratio to ensure data diversity and representativeness.

[0043] Optimization strategy: Adaptive optimization algorithm is adopted to dynamically adjust the learning rate, combined with regularization technology to prevent overfitting, so as to improve the generalization ability of the model and ensure the stability of the model in different scenarios. The deep neural network model optimizes parameters through continuous incremental learning, adapts to new fault modes, improves identification performance, and ensures the adaptability and advancement of the system.

[0044] III. Multi-source information fusion and abnormal decision-making

[0045] Feature vector construction

[0046] Time-domain features: Extract parameters such as waveform peak value, duration, and decay characteristics to reflect the time-dimensional changes of transient processes.

[0047] Frequency domain characteristics: Analyze frequency domain characteristics such as harmonic content and dominant frequency shift to capture the frequency distribution characteristics of transient signals.

[0048] Topological characteristics: Combining the distribution network topology, calculate the node importance index to assess the node's status and influence in the network.

[0049] Environmental characteristics: Integrates real-time environmental data such as temperature, humidity, and pollution levels, and considers the impact of environmental factors on equipment status.

[0050] Multi-level anomaly detection rules

[0051] Level 1 Decision: When the model outputs a high-confidence result, the anomaly type is directly determined and an emergency response is triggered, initiating a rapid processing procedure.

[0052] Secondary judgment: When the confidence level is moderate, combine Bayesian network to fuse multi-source information, conduct anomaly cause analysis, and provide a more comprehensive basis for decision-making.

[0053] Level 3 Judgment: When the confidence level is low, suspicious samples are marked and a manual review process is triggered to ensure the accuracy of the results. The manual review process integrates on-site detection data interface, which can access multiple detection results, correct anomaly identification results, and update model training data.

[0054] IV. Rapid Location Technology for Abnormal Sections

[0055] Fuzzy C-means clustering localization algorithm

[0056] Feature difference calculation: Calculate the transient feature differences between nodes, construct a multi-dimensional feature vector, and comprehensively consider time domain, frequency domain and topological features.

[0057] Fuzzy clustering analysis: The clustering algorithm divides the nodes into faulty segment groups and non-faulty segment groups, and determines the membership degree of each node to the faulty segment.

[0058] Topology correlation verification: By combining the distribution network topology, non-critical sections are excluded, abnormal areas are accurately located, and the accuracy of location is improved.

[0059] Location result visualization: Anomaly identification results are visualized through a geographic information system, marking the location of abnormal segments, the membership degree of abnormal nodes, and topological importance. It supports real-time viewing on mobile terminals, enabling intuitive display and convenient access.

[0060] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for anomaly identification in medium and low voltage lines based on deep neural networks, characterized in that, Includes the following steps: Step 1: Deploy transient recording devices at the outgoing terminals, branch boxes, distribution transformers, and important user terminals of medium and low voltage lines to build a wide-area data acquisition network covering station-level nodes, line-level nodes, transformer-level nodes, and user-level nodes. The deployment density and sampling rate of each node should meet the needs of full-area monitoring to ensure complete acquisition of transient data. Step 2: Perform time synchronization correction, noise suppression, and standardization on the collected transient data to generate an input sequence suitable for deep neural networks, ensuring data quality and consistency; Step 3: Construct a hybrid model that includes a convolutional neural network (CNN) and a long short-term memory network (LSTM). After inputting a standardized sequence, the CNN layer extracts waveform spatial features, and the LSTM layer captures temporal dependencies, outputting the probability and severity of anomaly types to achieve deep feature learning and classification. Step 4: Combine Bayesian network fusion device ledger, environmental data and historical fault records to analyze the causes of anomalies and trigger graded early warnings, providing multi-dimensional decision support; Step 5: Using the fuzzy C-means clustering algorithm, abnormal segments are located based on the differences in transient features between nodes and topological correlation. Topological verification is used to improve the positioning accuracy and achieve precise fault location.

2. The method for anomaly identification of medium and low voltage lines based on deep neural networks according to claim 1, characterized in that: The transient waveform recording device described in step one supports a triggered acquisition mode, which can record transient waveform segments before and after a fault and transmit data in real time through a communication network.

3. The method for anomaly identification of medium and low voltage lines based on deep neural networks according to claim 2, characterized in that: The hybrid network of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) described in step three contains multiple layers of convolution operations. It extracts multi-scale waveform features through convolution kernels of different sizes, and the LSTM layer contains an appropriate number of memory units to capture temporal features.

4. The method for anomaly identification of medium and low voltage lines based on deep neural networks according to claim 3, characterized in that: The Bayesian network described in step four integrates multi-source information and outputs the probability distribution of abnormal causes, providing a basis for operation and maintenance decisions.

5. The method for identifying anomalies in medium and low voltage lines based on deep neural networks according to claim 4, characterized in that: The fuzzy C-means clustering algorithm described in step five performs clustering analysis based on multidimensional feature vectors and combines topological structure to exclude non-critical segments and locate abnormal areas.

6. The method for anomaly identification of medium and low voltage lines based on deep neural networks according to claim 5, characterized in that: The anomaly identification results are visualized through a geographic information system, marking the location of the anomaly segment, the membership degree of the node anomaly, and its topological importance, and supporting real-time viewing on mobile terminals.

7. The method for anomaly identification of medium and low voltage lines based on deep neural networks according to claim 6, characterized in that: Deep neural network models optimize parameters through continuous incremental learning, adapt to new fault modes, improve identification performance, and ensure the system's adaptability and advanced nature.

Images