Power distribution network operation state evaluation and fault early warning method and system based on data fusion and deep learning
By using multi-source data fusion and deep learning methods, the problems of data silos and insufficient early warning in the distribution network have been solved, realizing intelligent fault early warning and rapid location of the distribution network, and improving power supply reliability.
Patent Information
- Application Number
- CN202511462283.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-14
- Publication Date
- 2026-01-16
AI Technical Summary
The existing power distribution network suffers from severe data silos, lacks an effective data fusion mechanism, has insufficient traditional early warning capabilities, relies on manual experience, has a delayed response, and cannot achieve early warning, thus affecting the reliability of power supply.
We employ multi-source heterogeneous data acquisition and preprocessing, use deep learning models for high-dimensional feature extraction and fusion, utilize deep autoencoders and CNN-LSTM models for risk quantification and fault warning, and combine interpretable AI technology to assist in locating the source of anomalies.
It achieves deep integration of multi-source data, can automatically identify complex potential faults, provide early warnings, shorten fault troubleshooting time, and improve power supply reliability and intelligence.
Smart Images
Figure CN121352104A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of power system monitoring and data analysis technology, specifically to a method and system for evaluating the operating status of distribution networks and providing early warning of faults based on data fusion and deep learning. Background Technology
[0002] With the rapid development of smart grids and the energy internet, a massive number of sensing devices have been deployed in distribution networks, such as smart meters in advanced metering architectures (AMIs), fault indicators on distribution lines, online oil chromatography monitoring devices on transformers, and video surveillance in substations. These devices generate massive amounts of heterogeneous, multi-timescale operational data (voltage, current, power, temperature, images, etc.).
[0003] Currently, the main problems in the monitoring of the operation status and fault handling of power distribution networks are as follows:
[0004] The problem of data silos is serious: data from different sources are analyzed independently, lacking an effective integration mechanism, making it difficult to form a unified and comprehensive understanding of the power grid status.
[0005] Insufficient early warning capability: Traditional threshold alarm methods are too simple and cannot identify complex and progressive potential faults (such as aging of line insulation, overheating of transformers, etc.), resulting in high false alarm and missed alarm rates.
[0006] Reliance on human experience: Status assessment and fault diagnosis rely heavily on the experience of schedulers and operations personnel, which is inefficient and difficult to handle when data volumes surge.
[0007] Delayed response: The fault is usually located and isolated only after it occurs, which makes it impossible to provide early warning and affects the reliability of power supply.
[0008] Therefore, there is an urgent need for an analytical method that can deeply integrate multi-source heterogeneous data, automatically mine the inherent correlations in the data, and achieve intelligent early warning. Summary of the Invention
[0009] The purpose of this invention is to provide a method and system for evaluating the operating status and providing early warning of faults in power distribution networks based on data fusion and deep learning, thereby improving the intelligence level and power supply reliability of power distribution networks.
[0010] To achieve the objectives of this invention, the technical solution provided by this invention is as follows:
[0011] First aspect
[0012] This application provides a method for power distribution network operation status assessment and fault early warning based on data fusion and deep learning, including the following steps:
[0013] Step S1: Multi-source heterogeneous data acquisition and preprocessing to obtain preprocessed data;
[0014] Step S2: Based on the preprocessed data, perform high-dimensional feature extraction and fusion to obtain a fused feature vector;
[0015] Step S3: Based on the fused feature vector, perform operational status assessment and risk quantification to obtain a risk index;
[0016] Step S4: Based on the risk index, perform early warning and location of faults.
[0017] Specifically, the multi-source heterogeneous data acquisition in step S1 involves: acquiring multi-source time-series data of the distribution network from the data platform, including: voltage, current, and power data from the SCADA system; fine-grained electricity consumption data from the user side of the smart meter AMI; zero-sequence current and temperature data from the fault indicator; data from online monitoring devices; and meteorological data.
[0018] The meteorological data includes temperature, humidity, and wind speed data.
[0019] Specifically, the preprocessing in step S1 involves cleaning, aligning, and normalizing the collected data, and using interpolation or aggregation methods to unify the data onto the same time series for different time granularities.
[0020] Specifically, step S2 includes the following:
[0021] Step S2.1: Use neural networks to automatically extract features from different types of data to capture their local patterns and short-term dependencies;
[0022] Step S2.2: Use a neural network to model the extracted feature sequences and the original time series to capture the long-term dependencies and dynamic evolution patterns of the data over time;
[0023] Step S2.3: Concatenate feature vectors from different data sources to form a unified, high-dimensional fusion feature vector, thereby achieving deep fusion at the data level.
[0024] Specifically, step S3 includes the following:
[0025] Step S3.1: Input the fused high-dimensional feature vector into a deep autoencoder and train the autoencoder on historical normal data so that it learns to reconstruct data patterns under normal working conditions;
[0026] Step S3.2: Calculate the reconstruction error of real-time data. The magnitude of the reconstruction error directly quantifies the degree of deviation between the current operating state and the normal state, i.e., the risk index. The larger the error, the more abnormal the system state and the higher the risk.
[0027] Specifically, step S4 includes the following:
[0028] Step S4.1: Set a dynamic risk threshold. When the risk index exceeds the threshold, the system triggers an early warning.
[0029] Step S4.2: Analyze the CNN-LSTM model using gradient-weighted class activation mapping or similar interpretable AI techniques to generate a feature importance heatmap. This heatmap can indicate which data sources and which features at which time points contribute the most to the current high-risk state, thereby helping operations and maintenance personnel to quickly locate the source of the anomaly.
[0030] Second aspect
[0031] This application provides a power distribution network operation status assessment and fault early warning system based on data fusion and deep learning, including the following units: data acquisition and preprocessing unit, high-dimensional feature extraction and fusion unit, operation status assessment and risk quantification unit, and fault early warning and location unit;
[0032] The data acquisition and preprocessing unit is used for multi-source heterogeneous data acquisition and preprocessing to obtain preprocessed data.
[0033] The high-dimensional feature extraction and fusion unit is used to perform high-dimensional feature extraction and fusion based on the preprocessed data to obtain a fused feature vector;
[0034] The operation status assessment and risk quantification unit is used to perform operation status assessment and risk quantification based on the fused feature vector to obtain a risk index;
[0035] The fault early warning and location unit is used to perform fault early warning and location based on the risk index.
[0036] Specifically, the multi-source heterogeneous data acquisition involves collecting multi-source time-series data of the power distribution network from the data platform, including: voltage, current, and power data from the SCADA system; fine-grained electricity consumption data from the user side of the smart meter AMI; zero-sequence current and temperature data from the fault indicator; data from online monitoring devices; and meteorological data, including temperature, humidity, and wind speed data.
[0037] The high-dimensional feature extraction and fusion unit specifically performs the following steps:
[0038] Step S2.1: Use neural networks to automatically extract features from different types of data to capture their local patterns and short-term dependencies;
[0039] Step S2.2: Use a neural network to model the extracted feature sequences and the original time series to capture the long-term dependencies and dynamic evolution patterns of the data over time;
[0040] Step S2.3: Concatenate feature vectors from different data sources to form a unified, high-dimensional fusion feature vector, thereby achieving deep fusion at the data level.
[0041] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0042] (1) This application breaks through the data silo and realizes the effective fusion and collaborative analysis of multi-dimensional data such as electrical, physical, chemical and environmental data through deep learning models, so as to perceive more comprehensively;
[0043] (2) This application utilizes unsupervised or self-supervised learning (autoencoder) to discover unknown and weak fault precursor features, realizing the transformation from "post-event analysis" to "pre-event warning", which greatly advances the warning time window.
[0044] (3) This application combines interpretable AI technology, which not only knows that "there is a problem", but also makes a preliminary judgment on "where the problem may be", which greatly shortens the troubleshooting time;
[0045] (4) The deep learning model of this application can continuously learn new data, continuously optimize and update the benchmark of the “normal” state, and adapt to the power grid structure and the commissioning of new equipment. Attached Figure Description
[0046] Figure 1 This is a schematic diagram of the process for a distribution network operation status assessment and fault early warning method based on data fusion and deep learning, provided in an embodiment of the present invention. Detailed Implementation
[0047] 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 only some embodiments of the present invention, and not all embodiments. 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.
[0048] This invention relates to the field of power system monitoring and data analysis technology, and in particular to a method and system for real-time and accurate assessment of the operating status of the distribution network and early warning of faults by using multi-source data such as smart meters, distribution automation terminals, and online monitoring devices, and performing data fusion analysis through deep learning models.
[0049] like Figure 1 As shown, this invention provides a method for power distribution network operation status assessment and fault early warning based on data fusion and deep learning, including the following steps:
[0050] Step S1: Multi-source heterogeneous data acquisition and preprocessing to obtain preprocessed data;
[0051] The data platform collects multi-source time-series data of the distribution network, including: voltage, current, and power data from the SCADA system; fine-grained electricity consumption data from smart meters (AMIs) on the user side; zero-sequence current and temperature data from fault indicators; data from online monitoring devices (such as DGA oil chromatography data, temperature, and partial discharge); and meteorological data (temperature, humidity, and wind speed).
[0052] The data is cleaned, aligned, and normalized. For different time granularities (seconds, minutes, hours), interpolation or aggregation methods are used to unify the data onto the same time series.
[0053] Step S2: Based on the preprocessed data, perform high-dimensional feature extraction and fusion to obtain a fused feature vector;
[0054] Neural networks (such as 1D-CNN) are used to automatically extract features from different types of data (electrical quantities, temperature, chemical quantities, etc.) to capture their local patterns and short-term dependencies.
[0055] Neural networks are used to model the extracted feature sequences and the original time series in order to capture the long-term dependencies and dynamic evolution of the data over time.
[0056] Feature vectors from different data sources are concatenated to form a unified, high-dimensional fusion feature vector, thereby achieving deep data fusion.
[0057] Step S3: Based on the fused feature vector, perform operational status assessment and risk quantification to obtain a risk index;
[0058] The fused high-dimensional feature vector is input into a deep autoencoder. This autoencoder is trained on historical normal data so that it learns to reconstruct data patterns under normal operating conditions.
[0059] Calculate the reconstruction error (e.g., mean squared error, MSE) of real-time data. The magnitude of the reconstruction error directly quantifies the degree of deviation between the current operating state and the "normal" state, i.e., the risk index. The larger the error, the more abnormal the system state, and the higher the risk.
[0060] Step S4: Based on the risk index, perform early warning and location of faults.
[0061] Set dynamic risk thresholds. When the risk index exceeds the threshold, the system triggers an alert.
[0062] Gradient-weighted class activation mapping (Grad-CAM) or similar interpretable AI techniques are used to analyze CNN-LSTM models and generate feature importance heatmaps. These heatmaps indicate which data sources and which features at which points in time contribute most to the current high-risk state, thus helping maintenance personnel quickly locate the source of the anomaly (e.g., a specific line or transformer substation).
[0063] In addition, this application also provides a power distribution network operation status assessment and fault early warning system based on data fusion and deep learning, corresponding to the method, including the following units: data acquisition and preprocessing unit, high-dimensional feature extraction and fusion unit, operation status assessment and risk quantification unit, and fault early warning and location unit;
[0064] The data acquisition and preprocessing unit is used for multi-source heterogeneous data acquisition and preprocessing to obtain preprocessed data.
[0065] The high-dimensional feature extraction and fusion unit is used to perform high-dimensional feature extraction and fusion based on the preprocessed data to obtain a fused feature vector;
[0066] The operation status assessment and risk quantification unit is used to perform operation status assessment and risk quantification based on the fused feature vector to obtain a risk index;
[0067] The fault early warning and location unit is used to perform fault early warning and location based on the risk index.
[0068] Specifically, the multi-source heterogeneous data acquisition involves collecting multi-source time-series data of the power distribution network from the data platform, including: voltage, current, and power data from the SCADA system; fine-grained electricity consumption data from the user side of the smart meter AMI; zero-sequence current and temperature data from the fault indicator; data from online monitoring devices; and meteorological data, including temperature, humidity, and wind speed data.
[0069] The high-dimensional feature extraction and fusion unit specifically performs the following steps:
[0070] Step S2.1: Use neural networks to automatically extract features from different types of data to capture their local patterns and short-term dependencies;
[0071] Step S2.2: Use a neural network to model the extracted feature sequences and the original time series to capture the long-term dependencies and dynamic evolution patterns of the data over time;
[0072] Step S2.3: Concatenate feature vectors from different data sources to form a unified, high-dimensional fusion feature vector, thereby achieving deep fusion at the data level.
[0073] Among them, the high-dimensional feature extraction and fusion unit has a built-in pre-trained multi-channel CNN-LSTM model and a deep autoencoder, which is the core computing engine.
[0074] In addition, the early warning and location unit is also used to display the overall risk situation of the system, the warning list, the location information (heat map), and the historical data backtracking.
[0075] Finally, it should be noted that the above embodiments are merely illustrative and explanatory of the present invention, and are not intended to limit the present invention to the scope of the described embodiments. Furthermore, those skilled in the art will understand that the present invention is not limited to the above embodiments, and many more variations and modifications can be made based on the teachings of the present invention, all of which fall within the scope of protection claimed by the present invention.
Claims
1. A power distribution network operation state evaluation and fault early warning method based on data fusion and deep learning, characterized in that, The method comprises the following steps: Step S1: multi-source heterogeneous data collection and preprocessing, to obtain preprocessed data; Step S2: based on the preprocessed data, high-dimensional feature extraction and fusion are performed to obtain a fusion feature vector; Step S3: based on the fusion feature vector, running state evaluation and risk quantification are performed to obtain a risk index; Step S4: based on the risk index, early fault warning and positioning are performed.
2. The power distribution network operation state evaluation and fault early warning method based on data fusion and deep learning according to claim 1, characterized in that, The multi-source heterogeneous data collection in step S1 specifically comprises: collecting multi-source time series data of the power distribution network from a data platform, including: voltage, current and power data of a SCADA system; user-side fine-grained power consumption data of an intelligent electric meter AMI; zero sequence current and temperature data of a fault indicator; online monitoring device data; and meteorological data.
3. The power distribution network operation state evaluation and fault early warning method based on data fusion and deep learning according to claim 2, characterized in that, The meteorological data comprises temperature, humidity and wind speed data.
4. The power distribution network operation state evaluation and fault early warning method based on data fusion and deep learning according to claim 1, characterized in that, The preprocessing in step S1 specifically comprises: cleaning, aligning and normalizing the collected data, and using interpolation or aggregation methods to unify the data to the same time series according to different time granularities.
5. The power distribution network operating state assessment and fault early warning method based on data fusion and deep learning according to claim 1, characterized in that, The step S2 specifically comprises the following: Step S2.1: using a neural network to automatically extract features of different types of data respectively, to capture local patterns and short-term dependencies; Step S2.2: using a neural network to model the extracted feature sequences and the original time series, to capture long-term dependencies and dynamic evolution rules of the data over time; Step S2.3: splicing the feature vectors from different data sources to form a unified, high-dimensional fusion feature vector, thereby realizing deep fusion at the data level.
6. The power distribution network operating state assessment and fault early warning method based on data fusion and deep learning according to claim 1, characterized in that, The step S3 specifically comprises the following: Step S3.1: inputting the fused high-dimensional feature vector into a deep autoencoder, and training the autoencoder on historical normal data, so that the autoencoder learns to reconstruct the data pattern under normal conditions; Step S3.2: calculating the reconstruction error of real-time data, and the reconstruction error directly quantifies the deviation of the current running state from the normal state, i.e., the risk index; the greater the error, the more abnormal the system state, and the higher the risk.
7. The power distribution network operating state assessment and fault early warning method based on data fusion and deep learning according to claim 1, characterized in that, The step S4 specifically comprises the following: Step S4.1: setting a dynamic risk threshold, and triggering a warning when the risk index exceeds the threshold; Step S4.2: using gradient-weighted class activation mapping or similar interpretable AI technology to analyze the CNN-LSTM model, to generate a feature importance heat map, which can indicate which data sources, time points and features contribute most to the current high-risk state, thereby assisting operation and maintenance personnel to quickly locate the abnormal source.
8. A power distribution network operation state evaluation and fault early warning system based on data fusion and deep learning, characterized in that, The method comprises the following units: a data collection and preprocessing unit, a high-dimensional feature extraction and fusion unit, a running state evaluation and risk quantification unit, and an early fault warning and positioning unit; The data collection and preprocessing unit is configured to collect and preprocess multi-source heterogeneous data, to obtain preprocessed data; The high-dimensional feature extraction and fusion unit is configured to perform high-dimensional feature extraction and fusion based on the preprocessed data, to obtain a fusion feature vector; The running state evaluation and risk quantification unit is configured to perform running state evaluation and risk quantification based on the fusion feature vector, to obtain a risk index; and The early fault warning and positioning unit is configured to perform early fault warning and positioning based on the risk index. The operation state evaluation and risk quantification unit is configured to perform operation state evaluation and risk quantification based on the fusion feature vector, and obtain a risk index. The fault early warning and positioning unit is configured to perform fault early warning and positioning based on the risk index.
9. The power distribution network operating state assessment and fault early warning system based on data fusion and deep learning according to claim 8, characterized in that, The multi-source heterogeneous data collection specifically comprises: collecting multi-source time series data of the power distribution network from a data platform, including: voltage, current and power data of a SCADA system; user-side fine-grained power consumption data of an intelligent electric meter AMI; zero sequence current and temperature data of a fault indicator; online monitoring device data; and weather data, the weather data including temperature, humidity and wind speed data.
10. The power distribution network operating state assessment and fault early warning system based on data fusion and deep learning according to claim 8, characterized in that, The high-dimensional feature extraction and fusion unit specifically performs the following steps: Step S2.1: using a neural network to automatically extract features of different types of data respectively, and capturing local patterns and short-term dependencies thereof; Step S2.2: using a neural network to model the extracted feature sequence and the original time series, so as to capture long-term dependencies and dynamic evolution rules of the data in time; Step S2.3: splicing feature vectors from different data sources to form a unified, high-dimensional fusion feature vector, thereby realizing deep fusion at the data level.