A substation equipment fault intelligent early warning method and system based on multi-source data fusion

Abstract

This invention discloses an intelligent early warning method and system for substation equipment faults based on multi-source data fusion, belonging to the field of substation equipment operation and maintenance technology. The method includes: collecting multi-source data such as substation equipment operating parameters, environmental perception, image monitoring, and historical operation and maintenance data; eliminating data heterogeneity through preprocessing such as noise reduction and outlier removal; integrating multi-dimensional information using a three-level fusion strategy of data layer, feature layer, and decision layer; constructing an improved deep belief network early warning model, and optimizing parameters using a jellyfish swarm optimization algorithm to achieve accurate fault identification and risk classification; triggering graded early warnings and pushing information, and recording data for model iteration. The system includes six modules, including multi-source data acquisition and preprocessing. This invention solves the problems of single data and low early warning accuracy in traditional methods, improving the timeliness and reliability of fault early warning, and providing technical support for intelligent operation and maintenance of substations.

Description

Technical Field

[0001] This invention belongs to the field of substation equipment operation and maintenance technology, specifically relating to a method and system for intelligent early warning of substation equipment faults based on multi-source data fusion. Background Technology

[0002] The content in this section only provides background information related to this invention and may not constitute prior art.

[0003] As the core hub of the power system, the stability of the substation's equipment operation directly affects the security and reliability of power supply. With the rapid development of the power system towards high voltage, large capacity, and intelligence, the scale of substation equipment is growing exponentially. The operating environment faces severe challenges such as complex electromagnetic environment, extreme weather conditions, and multi-physics field coupling effects. Equipment failures lead to frequent power grid accidents, seriously affecting the safe and economical operation of the power system.

[0004] Current mainstream substation equipment fault early warning technologies have several shortcomings: First, threshold-based monitoring methods can only identify visible faults, lacking sufficient ability to identify early potential faults, resulting in a high false negative rate and difficulty in achieving early fault warnings. Second, traditional fault diagnosis models often rely on single-parameter analysis, lacking systematic integration of multi-dimensional data, resulting in limited information dimensions, low fault identification accuracy, and high false alarm rate, making it difficult to cope with the diverse equipment and complex fault types in substations. Third, the parameters of existing early warning models are mostly statically set, failing to adaptively adjust according to changes in equipment operating conditions, resulting in poor model generalization ability. Fourth, multi-source data suffers from problems such as heterogeneous formats and scale differences, lacking effective integration between data, forming information silos, and failing to fully leverage the synergistic effect of multi-source data.

[0005] Furthermore, traditional manual inspection methods are inefficient, labor-intensive, and susceptible to human error, leading to missed or false detections. They are no longer sufficient to meet the high accuracy and rapid response requirements of modern substations for fault identification. Therefore, there is an urgent need for an intelligent early warning method and system that can integrate multi-source data, improve early warning accuracy, and provide early fault warnings. This would address the shortcomings of existing technologies and drive the development of substation operation and maintenance towards intelligence and automation. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent early warning method and system for substation equipment faults based on multi-source data fusion. This addresses the problems of traditional early warning methods, such as single data sources, low early warning accuracy, high false alarm and missed alarm rates, and poor model generalization ability. The goal is to achieve accurate and timely early warning of substation equipment faults, thereby improving substation operation and maintenance efficiency and intelligence. To achieve the above objectives, the technical solution adopted by this invention is as follows: Firstly, this application provides an intelligent early warning method for substation equipment faults based on multi-source data fusion, comprising the following steps: S1. Multi-source data acquisition: Deploy multiple types of acquisition devices to simultaneously acquire substation equipment operating parameter data, environmental perception data, image monitoring data, and historical operation and maintenance data to construct a raw multi-source dataset. Operating parameter data includes core equipment operating parameters such as voltage, current, power, insulation resistance, partial discharge, and oil chromatography data. Environmental perception data includes environmental parameters affecting equipment operation, such as ambient temperature, humidity, air pressure, dust concentration, and electromagnetic interference intensity. Image monitoring data includes visible light and infrared images of equipment appearance, instrument readings, and component status, used to capture information such as surface defects and temperature anomalies. Historical operation and maintenance data includes equipment fault records, maintenance logs, calibration records, and historical operating parameter data, providing data support for fault diagnosis and model training.

[0007] This step employs a three-pronged acquisition strategy encompassing space, time, and domain. Spatially, it deploys three levels of surveillance cameras—close-up, mid-range, and long-range—to achieve multi-angle, all-around coverage of the equipment. Close-up cameras focus on equipment details, mid-range cameras cover the areas associated with the equipment, and long-range cameras monitor the overall environment of the entire site. In terms of time, it combines timed acquisition with real-time triggering. Under normal conditions, data is collected at preset intervals (e.g., 5 minutes / time). When abnormal signals are detected (e.g., sudden temperature rise or voltage change), the sampling frequency is automatically increased to capture instantaneous changes in the equipment's status, ensuring the comprehensiveness and timeliness of data acquisition.

[0008] S2. Data Preprocessing: The original multi-source dataset is subjected to noise reduction, outlier removal, missing value imputation, and standardization to eliminate data heterogeneity and interference, resulting in a standardized dataset. Since the collected multi-source data comes from different devices, it suffers from inconsistent formats, different units, noise interference, and missing data. Directly using it for model training would severely impact the accuracy of early warning systems; therefore, preprocessing is necessary. S21. Noise Reduction Processing: The moving average method is used to smooth and reduce noise in the operating parameter data. The current value of each data point is averaged with its N neighboring data points to obtain the noise reduction result for that data point, thus eliminating the impact of short-term spike noise on the data. The image monitoring data is optimized using defogging and anti-shake algorithms to reduce the interference of factors such as ambient light and fog, and improve image clarity.

[0009] S22. Outlier Removal: Calculate the standardized value of each dimension of data. When the standardized value is greater than the preset outlier judgment threshold (e.g., 3), it is regarded as an outlier and removed. The standardized value is the difference between the data point and the mean of the dimension divided by the standard deviation of the dimension. This method can effectively remove abnormal data caused by sensor failure and acquisition error.

[0010] S23. Missing value completion: For continuous data such as voltage, current, and temperature, interpolation is performed using the average of the valid data points before and after the data point to complete the data, maintaining the continuity and trend of the data sequence; for discrete data such as fault type and maintenance method, mode completion is used to ensure the integrity of the data.

[0011] S24. Standardization: Map the data of each dimension to the interval [0,1]. The specific calculation formula is: x'=(x-x_min) / (x_max-x_min), where x is the original data, x_min is the minimum value of the data in this dimension, and x_max is the maximum value of the data in this dimension. Standardization eliminates the dimensional differences between different physical quantities, laying the foundation for subsequent data fusion and model training.

[0012] S3. Multi-source data fusion: Employing a three-level fusion strategy of data layer, feature layer, and decision layer, it achieves deep integration of multi-dimensional information to obtain comprehensive equipment status assessment results. Multi-source data fusion is the core of improving early warning accuracy. Through three-level fusion, the synergistic effect of various types of data can be fully utilized, breaking down information silos. S31. Data Layer Fusion: Spatiotemporally align data from the same source, stitch together images from multiple cameras on the same device into a panoramic view to ensure the spatial integrity of image data; synchronize timestamps from different sensors to keep operating parameter data, environmental data, and image data consistent in time, avoiding fusion errors caused by time deviations.

[0013] S32. Feature Layer Fusion: Principal Component Analysis (PCA) is used to reduce the dimensionality of high-dimensional operating parameter data, remove redundant features, and extract key statistical features related to faults, including mean, standard deviation, slope, and range. Convolutional Neural Networks (CNNs) are used to encode features in image monitoring data, extracting feature vectors such as texture and shape of equipment appearance and component status. All types of features are integrated to generate a high-dimensional equipment status feature matrix, achieving effective fusion of multi-source features. This step uses time-series data segmentation and a multi-scale sliding window method to extract features, which can capture dynamic features related to fault evolution and improve feature effectiveness.

[0014] S33. Decision-level fusion: The fuzzy integral algorithm is used to assign adaptive weights to the multi-dimensional feature recognition results output by the feature layer. The weight allocation is dynamically adjusted based on the fault recognition contribution rate of each feature. The integrated operation generates the comprehensive equipment status evaluation result, which effectively solves the problem of insufficient reliability of single feature recognition and improves the accuracy and stability of the fusion result.

[0015] S4. Fault Early Warning Modeling: Based on the fused comprehensive state data, an improved deep belief network early warning model is constructed. The jellyfish swarm optimization algorithm is used to globally optimize the model parameters, and the optimal early warning model is obtained through training. Traditional deep belief networks suffer from random parameter initialization and are prone to getting trapped in local optima. This invention improves the fault identification capability of the model by improving the model structure and optimizing the parameters. S41. Constructing an improved deep belief network model: A multi-layer improved restricted Boltzmann machine (RBM) stacking structure is adopted, and a fault sensitivity coefficient matrix is ​​introduced. This matrix is ​​multiplied element-wise with the high-dimensional equipment state feature matrix to form a fault sensitivity input matrix. The calculation method of the fault sensitivity coefficient matrix is ​​as follows: Based on the safety boundary of each key parameter of the substation equipment, a fault risk amplification coefficient is set. Combined with the deviation between the feature components and the safety threshold, the fault sensitivity coefficient of each feature component is calculated to form a fault sensitivity coefficient matrix. This matrix can enhance the model's sensitivity to fault features and improve the early fault identification capability.

[0016] S42. Model Parameter Optimization: Taking the model's fault identification accuracy, false negative rate, and false positive rate as optimization objectives, the jellyfish swarm optimization algorithm is used to perform global search and dynamic tuning of the weight parameters and bias parameters of the deep belief network. The jellyfish swarm optimization algorithm has the characteristics of fast convergence speed and strong global search capability, which can effectively avoid the model from getting trapped in local optima and obtain the optimal parameter combination.

[0017] S43. Model Training and Validation: The standardized dataset is divided into training, validation, and test sets in a 7:2:1 ratio. The optimized early warning model is trained using the training set. The model hyperparameters (such as the number of network layers and the learning rate) are adjusted using the validation set. The model performance is validated using the test set to ensure that the model's fault identification accuracy is not less than 98%, the false negative rate is not higher than 0.007 times / day, and the false positive rate is not higher than 0.02 times / day, thus meeting the actual needs of substation fault early warning.

[0018] S5. Early Warning Triggering and Feedback: Real-time collected and processed multi-source data is input into the optimal early warning model, outputting the equipment fault risk level. Based on the risk level, the corresponding early warning mechanism is triggered, pushing early warning information to the maintenance terminal, and recording early warning data for model iteration and updates. The fault risk level is divided into three levels: Level 1 warning corresponds to minor anomalies, triggering an audible and visual alarm, pushing a reminder message, and recommending regular inspections by maintenance personnel; Level 2 warning corresponds to moderate anomalies, triggering SMS and APP push notifications, pushing the fault location, abnormal parameters, and preliminary handling suggestions, requiring maintenance personnel to investigate promptly; Level 3 warning corresponds to severe anomalies, triggering an emergency audible and visual alarm and multi-channel push notifications, pushing emergency handling plans, requiring immediate shutdown and repair to prevent the fault from escalating. Simultaneously, data such as early warning time, fault type, and handling results are recorded and used regularly for model iteration and optimization to improve the model's generalization ability and early warning accuracy.

[0019] Secondly, this application also provides an intelligent early warning system for substation equipment faults based on multi-source data fusion, used to implement the above-mentioned early warning method, including: Multi-source data acquisition module: Composed of various types of sensors, monitoring cameras, and data acquisition terminals, it is used to collect operating parameter data, environmental perception data, image monitoring data, and historical operation and maintenance data of substation equipment to construct a raw multi-source dataset. Among them, the sensors include voltage sensors, current sensors, temperature sensors, humidity sensors, partial discharge sensors, SF6 sensors, and vibration sensors, which are deployed in key equipment areas such as transformers, circuit breakers, and disconnect switches in the substation. The monitoring cameras include visible light cameras and infrared thermal imaging cameras to realize real-time monitoring of equipment appearance and temperature anomalies. The data acquisition terminal is responsible for data aggregation and initial transmission.

[0020] Data preprocessing module: Connected to the multi-source data acquisition module, it is used to perform noise reduction, outlier removal, missing value completion, and standardization on the original multi-source dataset, and output a standardized dataset. This module adopts a modular design, which can flexibly adjust the preprocessing algorithm parameters according to the characteristics of different types of data to ensure the preprocessing effect.

[0021] Data fusion module: Connected to the data preprocessing module, it adopts a three-level fusion strategy of data layer, feature layer and decision layer to perform spatiotemporal alignment, feature extraction and encoding, and fuzzy integral fusion on standardized datasets, and output the comprehensive status evaluation results of the equipment; this module has multiple built-in fusion algorithms and can dynamically select the optimal fusion strategy according to the actual application scenario.

[0022] Early warning modeling module: Connected to the data fusion module, it is used to build an improved deep belief network early warning model. It uses the jellyfish swarm optimization algorithm to optimize model parameters, trains the model using the training set, verifies the model performance using the test set, and outputs the optimal early warning model. This module supports online adjustment of model parameters and offline training, which facilitates model maintenance and updates.

[0023] Early warning push module: Connected to the early warning modeling module, it receives the fault risk level output by the optimal early warning model, triggers the corresponding early warning mechanism, and pushes the early warning information, fault location, and emergency handling suggestions to the operation and maintenance terminal; it supports three push methods: SMS, APP push, and audible and visual alarm. The operation and maintenance terminal includes a mobile APP, a monitoring center terminal, and a handheld terminal for operation and maintenance personnel, ensuring that operation and maintenance personnel can obtain early warning information in a timely manner.

[0024] Data storage module: Connected to each of the above modules, it stores raw multi-source data, preprocessed data, fused data, model parameters, early warning records, and historical operation and maintenance data, providing data support for model iteration and fault tracing. It adopts a hybrid storage architecture, with real-time data stored in in-memory databases such as Redis to ensure high throughput and meet real-time early warning requirements; historical data is stored in relational databases such as MySQL and time-series databases such as InfluxDB to support long-term querying and analysis.

[0025] The beneficial effects of this invention are as follows: 1. The technical solution involved in this invention adopts a multi-source data fusion strategy, which integrates multi-dimensional data such as operating parameters, environmental perception, image monitoring and historical operation and maintenance, breaks down information silos, gives full play to the synergistic effect of various types of data, solves the problem of single data in traditional early warning methods, and improves the comprehensiveness and accuracy of fault identification.

[0026] 2. The technical solution involved in this invention designs a three-level fusion architecture, which gradually realizes the deep integration of multi-source data from the data layer, feature layer and decision layer. Combined with the fuzzy integral algorithm, it realizes adaptive weight allocation, which effectively improves the reliability of the fusion result and reduces the limitations of single feature recognition.

[0027] 3. The technical solution involved in this invention constructs an improved deep belief network early warning model, introduces a fault sensitivity coefficient matrix to enhance the model's sensitivity to fault characteristics, and uses a jellyfish swarm optimization algorithm to optimize model parameters. This solves the problems of traditional models being prone to getting trapped in local optima and having poor generalization ability, thereby improving the model's fault identification accuracy and early warning capability. It can effectively identify potential faults and advance the early warning response time by 4-6 maintenance cycles.

[0028] 4. The technical solution involved in this invention realizes full-process automation of data collection, preprocessing, fusion, early warning, and push notification, replacing the traditional manual inspection method, reducing the labor intensity of operation and maintenance personnel, and improving the efficiency of substation operation and maintenance; at the same time, different early warning mechanisms are triggered according to the fault risk level, and targeted emergency handling suggestions are pushed to facilitate the rapid response of operation and maintenance personnel and prevent the fault from escalating.

[0029] 5. The technical solution system involved in this invention adopts a modular design, with a clear structure and strong scalability. It can flexibly adjust the acquisition equipment and model parameters according to the actual scale and equipment type of the substation to adapt to the application requirements of different scenarios. The data storage adopts a hybrid storage architecture, which takes into account both real-time and long-term storage requirements, and provides strong support for model iteration and fault tracing. Attached Figure Description

[0030] Figure 1 This is a flowchart of the early warning method of the present invention; Figure 2This is a structural block diagram of the early warning system of the present invention; Figure 3 This is a schematic diagram of the three-level architecture for multi-source data fusion of the present invention; Figure 4 This is a schematic diagram of the structure of the improved deep belief network of the present invention. Detailed Implementation

[0031] 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 only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0032] Example 1 like Figure 1-4 As shown in the figure, an intelligent early warning method for substation equipment faults based on multi-source data fusion proposed in this embodiment of the invention includes the following steps: S1. Multi-source data acquisition: Within the 220kV substation, voltage sensors, current sensors, partial discharge sensors, SF6 sensors, temperature sensors, and humidity sensors are deployed to simultaneously acquire operating parameter data of transformers and circuit breakers (voltage range 0-252kV, current range 0-1000A, partial discharge quantity 0-1000pC) and environmental sensing data (temperature -20℃-60℃, humidity 0-100%RH). Ten visible light cameras and five infrared thermal imaging cameras are deployed to acquire image data of equipment appearance, instrument readings, and component status. Equipment fault records, maintenance logs, calibration records, and historical operating parameter data from the substation's operation and maintenance system for the past five years are retrieved to construct the original multi-source dataset. The acquisition cycle is set to 5 minutes / time; when a temperature surge exceeding 5℃ / min or a voltage change exceeding 10% is detected, the sampling frequency is increased to 1 minute / time.

[0033] S2. Data Preprocessing: S21. Noise Reduction Processing: The operating parameter data is smoothed and denoised using the moving average method (N=5) to eliminate short-term spike noise; the image data is processed using histogram equalization dehazing algorithm and Gaussian filtering anti-shake algorithm to improve image clarity.

[0034] S22. Outlier Removal: Calculate the standardized values ​​of data in each dimension, set the outlier threshold to 3, and when the standardized value is greater than 3, it is considered an outlier and removed. The removal rate is controlled within 2%.

[0035] S23. Missing value completion: Linear interpolation is used to complete continuous data, and mode is used to complete discrete data. After completion, the data integrity reaches more than 99.5%.

[0036] S24. Standardization: Map the data of each dimension to the [0,1] interval to eliminate the difference in units and obtain a standardized dataset.

[0037] S3. Multi-source data fusion: S31. Data Layer Fusion: Stitching images from three cameras on the same device into a panoramic view, synchronizing the timestamps of all sensors to ensure spatiotemporal consistency of data.

[0038] S32. Feature layer fusion: PCA is used to reduce the dimensionality of the operating parameter data and extract four types of statistical features: mean, standard deviation, slope, and range; CNN is used to encode the image data and extract a 256-dimensional feature vector; after integration, a high-dimensional device status feature matrix (dimension 1000×300) is generated.

[0039] S33. Decision-level fusion: Using a fuzzy integral algorithm, adaptive weights are assigned to the recognition results output by the feature layer to generate a comprehensive equipment status assessment result.

[0040] S4. Fault Early Warning Modeling: S41. Construct an improved deep belief network model: adopt a 3-layer improved RBM stacked structure, introduce a fault sensitivity coefficient matrix, and multiply it with the feature matrix to obtain the fault sensitivity input matrix.

[0041] S42. Model parameter optimization: The jellyfish swarm optimization algorithm was adopted, with the population size set to 50, the number of iterations to 100, and the learning rate to 0.01. The model weights and bias parameters were optimized.

[0042] S43. Model Training and Validation: The standardized dataset was divided into training, validation, and test sets in a ratio of 7:2:1. After training, the model's fault identification accuracy was 98.5%, the false negative rate was 0.005 times / day, and the false positive rate was 0.015 times / day, meeting the early warning requirements.

[0043] S5. Early Warning Triggering and Feedback: Real-time data collection and input into the model, outputting the fault risk level; when an abnormal increase in the partial discharge of the transformer is detected, the model outputs a level-two early warning and pushes the warning information to the maintenance personnel via SMS and APP, including the fault location (#1 transformer), abnormal parameters (partial discharge 800pC), and preliminary handling suggestions (shutdown to check insulation performance); after the maintenance personnel handle the situation, they record the handling results for model iterative optimization.

[0044] Example 2 A substation equipment fault intelligent early warning system based on multi-source data fusion, used to implement the early warning method described in Example 1, includes: Multi-source data acquisition module: including 20 voltage sensors, 20 current sensors, 15 partial discharge sensors, 10 SF6 sensors, 15 temperature sensors, and 15 humidity sensors, deployed in key equipment areas of the substation; 10 visible light cameras (1080P resolution) and 5 infrared thermal imaging cameras (temperature measurement range -20℃-150℃), deployed around the equipment; 5 data acquisition terminals responsible for data aggregation and transmission.

[0045] Data preprocessing module: Employs an industrial-grade processor with built-in noise reduction, outlier removal, missing value completion, and standardization algorithms. It can process collected data in real time, output standardized datasets, and has a processing latency of ≤1s.

[0046] Data fusion module: It adopts FPGA chip to realize a three-level fusion strategy, supports real-time fusion of multi-source data, fusion delay ≤2s, and fusion accuracy ≥99%.

[0047] Early warning modeling module: Deployed on an industrial server, it has an improved deep belief network and jellyfish swarm optimization algorithm built in, supports online model training and parameter adjustment, and the model training time is ≤24h.

[0048] Early warning push module: Supports SMS, APP push, and audible and visual alarms, and can be linked with the mobile APP of operation and maintenance personnel and the monitoring center terminal. The early warning information push delay is ≤3s.

[0049] Data storage module: It uses Redis in-memory database (stores real-time data, capacity 100GB), MySQL relational database (stores historical operation and maintenance data, capacity 500GB) and InfluxDB time series database (stores historical running parameter data, capacity 1TB), and supports data query and export.

[0050] The system was tested at the 220kV substation for 3 months and triggered 12 warnings, including 7 Level 1 warnings, 4 Level 2 warnings, and 1 Level 3 warning. The warning accuracy rate was 98.3%, with no missed or false alarms. It effectively prevented the escalation of 3 equipment failures and improved the substation's operation and maintenance efficiency and safety.

[0051] Example 3 A method for intelligent early warning of substation equipment faults based on multi-source data fusion, applied to a 110kV distributed substation, includes the following steps: S1. Multi-source data acquisition: Eight voltage sensors, eight current sensors, six partial discharge sensors, five SF6 sensors, eight temperature sensors, and eight humidity sensors are deployed to simultaneously collect operating parameter data (voltage range 0-126kV, current range 0-800A, partial discharge quantity 0-800pC) and environmental sensing data (temperature -15℃-55℃, humidity 10%-95%RH) of distribution transformers and disconnectors within the distributed substation. Six visible light cameras and three infrared thermal imaging cameras are deployed to cover each distribution unit, collecting image data of equipment appearance and terminal status. Equipment fault records, maintenance logs, and historical operating parameter data from the substation operation and maintenance system for the past three years are retrieved to construct the original multi-source dataset. The standard acquisition cycle is set to 3 minutes / time. When a sudden change in partial discharge quantity exceeding 20% ​​or a sudden increase in ambient humidity exceeding 10% / h is detected, the sampling frequency is increased to 1 minute / time.

[0052] S2. Data Preprocessing: S21. Noise Reduction Processing: Weighted moving average method (N=4, weight coefficients are 0.4, 0.3, 0.2, 0.1) is used to smooth and reduce noise in the operating parameter data to enhance data stability; adaptive dehazing algorithm and median filtering anti-shake algorithm are used for image data to adapt to the complex outdoor lighting environment of distributed substations.

[0053] S22. Outlier Removal: Calculate the standardized values ​​of data in each dimension, set the outlier threshold to 2.8, and when the standardized value is >2.8, it is considered an outlier and removed. The removal rate is controlled within 1.8%.

[0054] S23. Missing value completion: For continuous data, linear interpolation combined with trend prediction is used for completion; for discrete data, mode combined with nearest neighbor value is used for completion. After completion, the data integrity reaches more than 99.6%.

[0055] S24. Standardization process: Map the data of each dimension to the interval [0,1] and use the formula x'=(x-x_min) / (x_max-x_min) to complete the standardization and eliminate the difference in dimensions.

[0056] S3. Multi-source data fusion: S31. Data Layer Fusion: Stitching images from two cameras in the same power distribution unit into a partial panoramic view, synchronizing all sensor timestamps to ensure data spatiotemporal consistency, adapting to the characteristics of distributed substations with multiple units and dispersed layout.

[0057] S32. Feature layer fusion: PCA is used to reduce the dimensionality of the operating parameter data and extract four types of statistical features: mean, standard deviation, slope, and range; a lightweight CNN is used to encode the image data and extract a 128-dimensional feature vector; after integration, a high-dimensional equipment status feature matrix (dimension 800×200) is generated, which reduces the hardware computing power pressure of distributed substations.

[0058] S33. Decision-making level fusion: The fuzzy integral algorithm is adopted to dynamically adjust the feature weights according to the equipment type of the distributed substation (distribution transformer, disconnector switch) to generate the comprehensive equipment status assessment result.

[0059] S4. Fault Early Warning Modeling: S41. Construct an improved deep belief network model: adopt a 2-layer improved RBM stacked structure, introduce a fault sensitivity coefficient matrix, and adjust the coefficient thresholds in combination with the fault characteristics of distributed substation equipment to form a fault sensitivity input matrix.

[0060] S42. Model parameter optimization: The jellyfish swarm optimization algorithm is adopted, with a population size of 40, an iteration count of 80, and a learning rate of 0.015. The model weights and bias parameters are optimized to adapt to the scenario where the amount of data in distributed substations is relatively small.

[0061] S43. Model Training and Validation: The standardized dataset was divided into training, validation and test sets in a ratio of 6:2:2. After training, the model's fault identification accuracy was 98.2%, the false negative rate was 0.006 times / day, and the false positive rate was 0.018 times / day, meeting the early warning requirements of distributed substations.

[0062] S5. Early Warning Triggering and Feedback: Real-time data collection and input into the model, outputting the fault risk level; when an abnormal increase in temperature of the disconnector terminal is detected (exceeding 80℃), the model outputs a level two early warning, pushing the fault location (#3 disconnector), abnormal parameters, and handling suggestions (check for loose wiring) via APP push and on-site audible and visual alarms; after the maintenance personnel handle the situation, they record the relevant data for model iteration and optimization.

[0063] The method was tested in a 110kV distributed substation for two months, triggering a total of 8 early warnings with an accuracy rate of 98.1%, 0 missed reports and 0 false alarms. It effectively reduced the manual inspection pressure of distributed substations and improved the timeliness of equipment operation and maintenance in remote areas.

[0064] Comparative Example The traditional single-data early warning method is applied to the same 110kV distributed substation as in Example 3, and the specific scheme is as follows: 1. Data Acquisition: Only operating parameter data (voltage, current, partial discharge) of substation equipment is collected. Environmental perception data, image monitoring data and historical operation and maintenance data are not collected. The normal acquisition cycle is 5 minutes / time, and there is no abnormal trigger speed-up mechanism.

[0065] 2. Data preprocessing: Only simple noise reduction (moving average method, N=3) and outlier removal (threshold of 3.5) are performed on the running parameter data. Missing value completion and standardization are not performed. The data is used directly for model training.

[0066] 3. Data fusion: There is no multi-source data fusion step; fault diagnosis is based solely on a single operating parameter.

[0067] 4. Early warning modeling: A common deep belief network model is adopted, without introducing a fault sensitivity coefficient matrix or optimizing the model parameters. The model is trained directly using the default parameters.

[0068] 5. Early warning trigger: The system judges the fault based on a single parameter threshold and outputs a fault prompt. There is no hierarchical early warning mechanism. The information is pushed through on-site audible and visual alarms only.

[0069] This traditional method was tested for two months at the same 110kV distributed substation, triggering a total of 7 warnings, including 3 false alarms and 2 missed alarms. The warning accuracy rate was only 57.1%, the missed alarm rate was 0.033 times / day, and the false alarm rate was 0.05 times / day. It failed to identify potential faults such as abnormal temperature of disconnector terminals and deterioration of insulation performance of distribution transformers in a timely manner, which led to the amplification of a minor equipment fault and affected the stability of power supply.

[0070] Comparative Analysis: Compared with the comparative example, Example 3 integrates multi-source data and multi-dimensional information, combined with an improved early warning model and parameter optimization, resulting in a 41 percentage point increase in early warning accuracy, an 81.8% reduction in false alarm rate, and a 64% reduction in false alarm rate. It also features tiered early warning and multi-channel push functions, enabling early identification of potential faults and preventing their escalation. This is significantly superior to traditional single-data early warning methods, fully demonstrating the technical advantages and practicality of this invention.

[0071] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0072] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent early warning of substation equipment faults based on multi-source data fusion, characterized in that, Includes the following steps: S1. Multi-source data acquisition: Deploy multiple types of acquisition devices to simultaneously acquire substation equipment operating parameter data, environmental sensing data, image monitoring data, and historical operation and maintenance data to construct a raw multi-source dataset. Operating parameter data includes voltage, current, power, insulation resistance, partial discharge, and oil chromatography data; environmental sensing data includes ambient temperature, humidity, air pressure, dust concentration, and electromagnetic interference intensity; image monitoring data includes visible light and infrared images of equipment appearance, instrument readings, and component status; historical operation and maintenance data includes equipment fault records, maintenance logs, calibration records, and historical operating parameter data. S2. Data Preprocessing: Noise reduction, outlier removal, missing value imputation, and standardization are performed on the original multi-source dataset to eliminate data heterogeneity and interference, resulting in a standardized dataset. Specifically, this includes: S21. Noise Reduction Processing: The moving average method is used to smooth and reduce noise in the operating parameter data to eliminate short-term spike noise; the defogging and anti-shake algorithms are used to optimize the image monitoring data to reduce environmental interference. S22. Outlier Removal: Calculate the standardized value of data for each dimension. When the standardized value is greater than the preset outlier threshold, it is considered an outlier and removed. The standardized value is the difference between the data point and the mean of the dimension divided by the standard deviation of that dimension. S23. Missing value imputation: For continuous data, interpolation is performed using the average of the valid data points before and after the missing data point; for discrete data, the mode is used to imput the missing value, thus maintaining the continuity and trend of the data sequence. S24. Standardization: Map the data of each dimension to the [0,1] interval to eliminate the dimensional differences between different physical quantities and obtain a standardized dataset; S3. Multi-source data fusion: Employing a three-tiered fusion strategy involving data layer, feature layer, and decision layer, deep integration of multi-dimensional information is achieved to obtain comprehensive equipment status assessment results; specifically including: S31. Data layer fusion: Spatiotemporal alignment of data from the same source, stitching images from multiple cameras on the same device into a panoramic view, synchronizing timestamps from different sensors, and ensuring spatiotemporal consistency of data; S32. Feature Layer Fusion: Principal Component Analysis (PCA) is used to reduce the dimensionality of high-dimensional operating parameter data and extract key statistical features; Convolutional Neural Network (CNN) is used to encode features of image monitoring data to extract texture and shape features of equipment appearance and component status; and various types of features are integrated to generate a high-dimensional equipment status feature matrix. S33. Decision-level fusion: The fuzzy integral algorithm is used to assign adaptive weights to the multi-dimensional feature recognition results output by the feature layer, and the comprehensive equipment status evaluation results are generated through integral calculation to improve the reliability of feature fusion; S4. Fault Early Warning Modeling: Based on the fused comprehensive state data, an improved deep belief network early warning model is constructed. The jellyfish swarm optimization algorithm is used to globally optimize the model weights and bias parameters, training to obtain the optimal early warning model; specifically including: S41. Construct an improved deep belief network model: adopt a multi-layer improved restricted Boltzmann machine stacking structure, introduce a fault sensitivity coefficient matrix, multiply it with the high-dimensional equipment state feature matrix element by element to form a fault sensitivity input matrix, and enhance the model's sensitivity to fault features; S42. Model parameter optimization: Taking the model fault identification accuracy, false negative rate, and false positive rate as optimization objectives, the jellyfish swarm optimization algorithm is used to perform global search and dynamic tuning of the weight parameters and bias parameters of the deep belief network to obtain the optimal parameter combination. S43. Model Training and Validation: Divide the standardized dataset into training set, validation set and test set. Use the training set to train the optimized early warning model, use the validation set to adjust the model hyperparameters, and use the test set to validate the model performance to ensure that the model meets the early warning accuracy requirements. S5. Early Warning Triggering and Feedback: Input the multi-source data collected and processed in real time into the optimal early warning model, output the equipment fault risk level, trigger the corresponding early warning mechanism according to the risk level, push the early warning information to the operation and maintenance terminal, and record the early warning data for model iteration and update; among them, the fault risk level is divided into three levels: Level 1 early warning corresponds to minor anomaly, Level 2 early warning corresponds to moderate anomaly, and Level 3 early warning corresponds to severe anomaly. Different levels correspond to different early warning response strategies and emergency handling suggestions.

2. The intelligent early warning method for substation equipment faults based on multi-source data fusion according to claim 1, characterized in that, In step S1, data acquisition adopts a three-in-one acquisition strategy of "space-time-domain"; in the spatial dimension, three levels of monitoring cameras, namely close-up, mid-range and long-range, are deployed to achieve multi-angle coverage of the equipment; in the time dimension, a combination of timed acquisition and real-time triggering is adopted. Under normal conditions, data is collected according to a preset cycle, and the sampling frequency is automatically increased when abnormal signals are detected.

3. The intelligent early warning method for substation equipment faults based on multi-source data fusion according to claim 1, characterized in that, In step S32, the extracted statistical features include mean, standard deviation, slope, and range. Key features related to fault evolution are extracted through time series data segmentation and multi-scale sliding window methods.

4. The intelligent early warning method for substation equipment faults based on multi-source data fusion according to claim 1, characterized in that, In step S41, the fault sensitivity coefficient matrix is ​​calculated as follows: based on the safety boundary of each key parameter of the substation equipment, a fault risk amplification coefficient is set, and the fault sensitivity coefficient of each characteristic component is calculated by combining the deviation between the characteristic component and the safety threshold to form a fault sensitivity coefficient matrix.

5. A substation equipment fault intelligent early warning system based on multi-source data fusion, characterized in that, To implement the early warning method according to any one of claims 1-4, comprising: Multi-source data acquisition module: Composed of various types of sensors, monitoring cameras and data acquisition terminals, it is used to collect substation equipment operating parameter data, environmental perception data, image monitoring data and historical operation and maintenance data to build a raw multi-source dataset; Data preprocessing module: Connected to the multi-source data acquisition module, it is used to perform noise reduction, outlier removal, missing value completion and standardization on the original multi-source dataset, and output a standardized dataset; Data fusion module: Connected to the data preprocessing module, it adopts a three-level fusion strategy of data layer, feature layer and decision layer to perform spatiotemporal alignment, feature extraction and encoding, and fuzzy integral fusion on standardized datasets, and outputs the comprehensive status evaluation results of the equipment. Early warning modeling module: Connected to the data fusion module, it is used to build an improved deep belief network early warning model. It uses the jellyfish swarm optimization algorithm to optimize the model parameters, trains the model through the training set, verifies the model performance using the test set, and outputs the optimal early warning model. Early warning push module: Connected to the early warning modeling module, it is used to receive the fault risk level output by the optimal early warning model, trigger the corresponding early warning mechanism, and push the early warning information, fault location and emergency handling suggestions to the operation and maintenance terminal; Data storage module: Connected to each of the above modules, it is used to store raw multi-source data, preprocessed data, fused data, model parameters, early warning records and historical operation and maintenance data, providing data support for model iteration and fault tracing.

6. The intelligent early warning system for substation equipment faults based on multi-source data fusion according to claim 5, characterized in that, The sensors in the multi-source data acquisition module include voltage sensors, current sensors, temperature sensors, humidity sensors, partial discharge sensors, SF6 sensors, and vibration sensors; the monitoring cameras include visible light cameras and infrared thermal imaging cameras, which are deployed in the key equipment areas of the substation.

7. The intelligent early warning system for substation equipment faults based on multi-source data fusion according to claim 5, characterized in that, The data storage module adopts a hybrid storage architecture, with real-time data stored in an in-memory database to ensure high throughput, and historical data stored in a relational database and a time-series database to support long-term querying and analysis.

8. The intelligent early warning system for substation equipment faults based on multi-source data fusion according to claim 5, characterized in that, The early warning push module supports three push methods: SMS, APP push, and audible and visual alarm. The operation and maintenance terminals include mobile APP, monitoring center terminal, and handheld terminals for operation and maintenance personnel.

Images