Metering transformer fault intelligent prediction method and system based on machine learning

Abstract

The invention discloses a metering transformer fault intelligent prediction method and system based on machine learning, and relates to the technical field of electric power metering monitoring, and the method comprises the steps: collecting the electrical parameters, load data and environmental parameters of a metering transformer, and carrying out the multi-modal fusion analysis to extract a key feature sequence; constructing a degradation analysis model based on machine learning in combination with historical operation data and shared data of equipment of the same model; and predicting the degradation state of the core component by using the degradation analysis model, calculating a health state score, performing fault risk assessment, and performing intelligent load distribution optimization according to a fault risk assessment result. The technical problems that in the prior art, fault early warning is difficult to carry out in the early stage, load distribution cannot be dynamically adjusted, and the service life of equipment is affected in a traditional method are solved, early fault early warning of the metering transformer is achieved through degradation analysis and health assessment of the core component, and the service life of the equipment is prolonged. And the service life of the equipment is prolonged through load optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power metering monitoring, in particular to a metering transformer fault intelligent prediction method and system based on machine learning. BACKGROUND

[0002] As a key device in the power system, the metering transformer is widely used in electric energy metering, relay protection and power monitoring, and its operation reliability is directly related to the safety and stability of the power grid. However, due to the long-term high load or even overload operation of the metering transformer, its core components (such as the core and the coil) are easily affected by multiple factors such as electricity, heat, mechanics and environment, leading to gradual degradation, such as core magnetization rate reduction, coil insulation aging, etc.

[0003] These degradation problems are usually difficult to be found in the early stage by conventional monitoring means, and once developed into a sudden failure, it will cause serious economic loss and power grid safety hazard. At the same time, the existing load distribution strategy is mostly based on fixed rules, which cannot dynamically adjust the load of high-risk equipment, easily leading to further aging or failure of the equipment, affecting the service life of the equipment. SUMMARY

[0004] The present application provides a metering transformer fault intelligent prediction method and system based on machine learning, which is used to solve the technical problems that the traditional method in the prior art cannot perform fault early warning in the early stage, and the load distribution cannot be dynamically adjusted, resulting in that the load pressure of high-risk equipment cannot be relieved, affecting the service life of the equipment.

[0005] In a first aspect, the present application provides a metering transformer fault intelligent prediction method based on machine learning, which comprises: collecting operation data of a target metering transformer through a sensor, the operation data comprising electrical parameters, load data and environmental parameters; performing multi-modal fusion analysis on the operation data and extracting a key feature sequence; using a degradation analysis model, taking the key feature sequence as input, to predict the degradation state of the core component of the target metering transformer; according to the degradation state of the core component, combining the operation load history data and the environmental parameters, calculating the health state score of the target metering transformer; based on the health state score, performing fault risk assessment of the target metering transformer, according to the fault risk assessment result, performing intelligent load distribution, and using load optimization strategy to perform load distribution optimization.

[0006] A second aspect of this application provides a machine learning-based intelligent fault prediction system for metering transformers. The system includes: an operational data acquisition module for acquiring operational data of a target metering transformer via sensors, the operational data including electrical parameters, load data, and environmental parameters; a key feature extraction module for performing multimodal fusion analysis on the operational data and extracting key feature sequences; a degradation analysis model construction module and a core component degradation prediction module, the core component degradation prediction module using the degradation analysis model and the key feature sequences as input to predict the degradation state of the core components of the target metering transformer; a health status assessment module for calculating a health status score of the target metering transformer based on the core component degradation state, combined with historical operational load data and the environmental parameters; and a fault risk assessment and optimization module for assessing the fault risk of the target metering transformer based on the health status score, performing intelligent load allocation based on the fault risk assessment results, and optimizing the load allocation using load optimization strategies.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0008] This application provides a machine learning-based intelligent fault prediction method and system for metering transformers, which relates to the field of power metering and monitoring technology. It collects operational data through sensors and performs multimodal fusion analysis. A degradation analysis model is constructed by combining historical and shared data. This model predicts the degradation status of core components and calculates a health status score, performing fault risk assessment. Based on the assessment results, load allocation is dynamically optimized. This solves the technical problems of existing methods, such as the difficulty in early fault warning and the inability to dynamically adjust load allocation, which leads to unrelieved load pressure on high-risk equipment and affects equipment lifespan. The method achieves early fault warning for metering transformers through core component degradation analysis and health assessment, and extends equipment lifespan through load optimization. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic diagram of the intelligent prediction method for metering transformer faults based on machine learning provided in the embodiments of this application;

[0011] Figure 2 A schematic diagram of the structure of a machine learning-based intelligent prediction system for metering transformer faults provided in this application embodiment.

[0012] Figure labeling: 11. Operational data acquisition module; 12. Key feature extraction module; 13. Core component degradation prediction module; 14. Health status assessment module; 15. Fault risk assessment and optimization module. Detailed Implementation

[0013] This application provides a machine learning-based intelligent prediction method and system for metering transformer faults, which solves the technical problems in the prior art where traditional methods are difficult to provide early warning of faults and cannot dynamically adjust load distribution, resulting in the inability to alleviate the load pressure on high-risk equipment and affecting the service life of the equipment.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to such processes, methods, products, or devices.

[0016] Example 1, as Figure 1 As shown, this application provides a machine learning-based intelligent prediction method for metering transformer faults, the method comprising:

[0017] P10: The operating data of the target metering transformer is collected by the sensor. The operating data includes electrical parameters, load data and environmental parameters.

[0018] Specifically, collecting operational data of the target metering transformer through sensors is the fundamental step of the entire intelligent fault prediction method. Its core lies in comprehensively acquiring multi-dimensional data on the operating status of the equipment, namely the dynamic parameters generated by the equipment during operation, including electrical parameters, load data, and environmental parameters. These data together reflect the health status of the metering transformer and the external working environment.

[0019] The acquisition of electrical parameters includes key indicators such as current, voltage, and power factor. Current and voltage are the core input and output quantities of the instrument transformer, reflecting its operating load and the presence of abnormal fluctuations, while the power factor can be used to assess the load's efficiency and stability. The acquisition of these parameters relies on high-precision electromagnetic sensors, employing high-frequency sampling technology to ensure the capture of minute changes.

[0020] The load data collection mainly includes the cumulative time of high-load operation, load fluctuation frequency, and load change rate. The cumulative time of high-load operation refers to the duration for which the instrument transformer operates close to or beyond the design load. Its accumulation may lead to degradation problems such as a decrease in core magnetization or coil loss. The load fluctuation frequency is used to assess the stability of the equipment load. High-frequency load fluctuations may be a precursor to sudden failures. The load change rate can be used to detect the severity of load changes and provide a basis for predicting the load pressure on the equipment.

[0021] The environmental parameters collected include temperature and humidity, vibration spectrum, and pollution level. Temperature and humidity directly affect the insulation performance and material aging rate of the equipment, especially in high humidity environments where degradation of insulation materials may occur. Vibration spectrum is an important indicator of the mechanical stability of the equipment. By collecting vibration signals and analyzing their spectral characteristics, early detection of component loosening or structural problems caused by mechanical vibration can be achieved. Pollution level reflects the impact of the external environment on the equipment; for example, the adhesion of dust and pollutants may increase the risk of equipment discharge.

[0022] To ensure the accuracy of data acquisition, multifunctional sensors with high sensitivity and anti-interference capabilities are required, and time synchronization mechanisms must be used to ensure that various types of data are aligned on the same time scale. Simultaneously, distributed data storage and edge computing technologies can be employed to perform preliminary data preprocessing during the acquisition phase, such as noise reduction, filtering, and outlier detection. Real-time updates provide reliable data support for subsequent multimodal data fusion, degradation analysis, and fault prediction, ensuring the robustness and accuracy of the entire system under complex operating conditions.

[0023] P20: Perform multimodal fusion analysis on the running data and extract key feature sequences.

[0024] Furthermore, step P20 in this embodiment of the application also includes:

[0025] P21: Perform feature filtering and dimensionality reduction on the electrical parameters, load data and environmental parameters to extract a set of key features, which includes different types of key features;

[0026] P22: Use feature interaction analysis to identify the nonlinear correlation between key features; P23: Based on the nonlinear correlation, integrate different types of key features into the key feature sequence through a multimodal fusion model.

[0027] It should be understood that multimodal fusion analysis is performed on the operational data and key feature sequences are extracted. Through feature selection and dimensionality reduction, feature interaction analysis, and multimodal fusion model, electrical parameters, load data, and environmental parameters are integrated to provide refined input data for subsequent prediction models.

[0028] First, in the feature selection and dimensionality reduction stage, a set of key features is extracted from the electrical parameters, load data, and environmental parameters. These parameters each contain a large number of raw data dimensions. For example, electrical parameters such as voltage, current, and power factor may include multiple dimensions of information, such as instantaneous fluctuations and trend changes. Load data may record various variables such as high-load operating time, load change rate, and fluctuation frequency. Environmental parameters include temperature and humidity, vibration spectrum characteristics, and pollution levels. Due to the diversity and redundancy of data dimensions, feature importance analysis methods, such as random forest feature selection or principal component analysis (PCA), can be used to select the features most relevant to fault prediction, forming a set of key features. For example, current fluctuation amplitude and temperature and humidity changes contribute significantly to insulation aging, while high-frequency components of the vibration spectrum may be closely related to the degradation of mechanical components. The set of key features includes different types of key features, such as time series features and environmental impact factors.

[0029] Next, feature interaction analysis is used to further identify the nonlinear correlations between key features. The purpose of feature interaction analysis is to uncover information that a single feature cannot express independently. For example, the co-variation of voltage fluctuations and load fluctuations may reveal the electrical stress of equipment under high loads, while the correlation between vibration characteristics and environmental pollution levels may indicate the accelerated wear of mechanical components by the external environment. To this end, machine learning-based interaction analysis tools (such as XGBoost or SHAP value analysis) can be used to quantify the interaction contributions between different features. For example, SHAP values ​​can explain the weight of the interaction effect between load fluctuation rate and vibration spectrum in fault prediction, thereby revealing latent fault modes.

[0030] Furthermore, based on the aforementioned nonlinear correlation, a multimodal fusion model integrates different types of key features into a unified key feature sequence. This multimodal fusion model can employ a deep learning architecture to simultaneously process temporal data and static environmental features. For example, CNNs can extract spatial pattern features of electrical parameters, LSTMs can capture the temporal dynamic changes of load fluctuations, and Transformers establish global correlations among all features through a self-attention mechanism. During the integration process, a multimodal feature alignment mechanism unifies feature timestamps and standardizes data scales, thereby generating a unified key feature sequence. This key feature sequence integrates electrical parameters, load characteristics, and environmental influences, possessing stronger expressive power and lower redundancy, providing high-quality data support for the subsequent degradation analysis model.

[0031] P30: Using a degradation analysis model, with the key feature sequence as input, predict the degradation status of the core components of the target metering transformer.

[0032] Furthermore, before predicting the degradation state, in order to construct a degradation analysis model, step P30 of this application embodiment also includes:

[0033] P31a: Collect historical operating data of the target metering transformer and extract shared data of the same model of equipment; P32a: Perform data clustering analysis on the historical operating data and shared data of the same model of equipment according to the core component type, and integrate them to generate a cross-device collaborative learning set, which contains multiple data subsets; P33a: Use the cross-device collaborative learning set as training data, and combine machine learning principles to perform independent component degradation trend training and overall degradation trend training respectively, generate core component analysis unit and overall trend analysis unit, and fuse them to generate the degradation analysis model.

[0034] Optionally, acquiring historical operating data of the target metering transformer and shared data of equipment of the same model, and building a degradation analysis model through machine learning, is a key step in realizing the prediction of degradation of the core components and the overall performance of the metering transformer.

[0035] First, during the data acquisition phase, historical operating data of the target metering transformer is collected, while shared data is extracted from the online monitoring system of the same model of equipment. Historical operating data refers to the electrical parameters, load data, and environmental parameters recorded during long-term operation of the equipment, comprehensively reflecting the long-term impact of the equipment's workload and external environment on its performance. Shared data from the same model of equipment, accumulated under other operating conditions, supplements the limitations of data from a single device, expanding the diversity and generalization ability of the training data. This data is shared through a federated learning mechanism, providing a more comprehensive foundation for collaborative learning while ensuring device privacy and data security.

[0036] Next, cluster analysis is performed on historical operating data and shared data from equipment of the same model, based on the type of core components. Core components include iron cores and coils, and their degradation patterns typically exhibit different characteristics. For example, a decrease in the magnetic susceptibility of the iron core may be related to high-load operating time, while coil losses may be significantly affected by temperature and humidity. Using feature similarity-based clustering algorithms (such as K-means or DBSCAN), data from different equipment are aggregated into multiple subsets, each corresponding to a specific operating state or degradation pattern of a core component. The resulting cross-device collaborative learning set contains multiple data subsets, each reflecting not only the operating characteristics of individual equipment but also common patterns among equipment of the same model.

[0037] Furthermore, using the cross-device collaborative learning ensemble as training data and incorporating machine learning principles, separate training is conducted for the independent degradation trend of core components and the overall degradation trend. Independent degradation trend training targets a single core component, extracting its core component training dataset from the cross-device collaborative learning ensemble. Time series models (such as LSTM or GRU) are used to predict the component's degradation state over time, generating core component analysis units for each core component. For example, a coil aging analysis model can capture the aging rate of a coil under specific environmental parameters. Overall degradation trend training, through ensemble learning methods, extracts the overall training dataset from the cross-device collaborative learning ensemble—a dataset containing aging data from multiple core components. Machine learning principles are used to train and predict the comprehensive degradation trend among multiple core components, such as how the coordinated degradation of the core and coil affects the overall health of the device, generating an overall trend analysis unit.

[0038] Finally, the core component analysis unit and the overall trend analysis unit are combined using model fusion techniques, such as weighted ensemble or joint training of deep learning models, to generate a complete degradation analysis model. This model can accurately predict the global trend of the overall health status of the equipment, from the fine-grained degradation of a single core component, providing hierarchical degradation analysis capabilities for equipment operation.

[0039] Furthermore, using a degradation analysis model with key feature sequences as input, the degradation trend and degradation level of core components (such as iron cores and coils) can be accurately predicted.

[0040] First, the key feature sequence generated in the previous steps is input into the degradation analysis model. The key feature sequence is a comprehensive set of features extracted from electrical parameters, load data, and environmental parameters. It has been time-aligned and nonlinearly correlated through multimodal fusion analysis, possessing high expressive power and temporal continuity. The feature sequence includes not only static parameters (such as high-load operating time and average load level) but also dynamic characteristics (such as load fluctuation frequency and temperature rise rate), providing the model with detailed operational status information.

[0041] The core of the degradation analysis model comprises a core component analysis unit and an overall trend analysis unit, respectively handling the degradation state of individual components and the overall degradation trend of the equipment. In the core component analysis, the degradation analysis model models key components such as the core and coils separately, for example, predicting the gradual decrease in core magnetic susceptibility using time series models (such as LSTM or GRU). Core magnetic susceptibility is a key indicator reflecting core performance, and its decline is usually directly related to long-term high-load operation, while coil losses can be significantly affected by changes in temperature and humidity. The model constructs an accurate prediction function by learning the relationship between these characteristics and degradation results from historical data.

[0042] Next, the overall trend analysis unit uses ensemble learning methods (such as XGBoost or Random Forest) to synthesize the degradation states of core components into a global degradation trend for the device. For example, when both the core and coil degrade simultaneously, the model can quantify their synergistic effect on the overall health of the device. This process fully leverages the potential correlations between multiple components, making the predictions more comprehensive and accurate.

[0043] In practical implementation, the degradation analysis model can employ an adaptive weight allocation mechanism to adjust the weights of key features. Through feature importance analysis, it dynamically focuses on the features that have the greatest impact on degradation prediction. For example, in a high-humidity environment, the weight of temperature and humidity features may be higher than that of load fluctuation features, thereby improving the model's adaptability.

[0044] Ultimately, the model outputs the degradation status of core components, including numerical results (range 0-100) of core susceptibility and coil loss. These results can be further classified into normal, slight degradation, and severe degradation, providing clear reference for equipment operation and maintenance. The prediction results can also be presented through a visualization platform, allowing maintenance personnel to understand the equipment's health status in real time. Through this step, the degradation analysis model transforms complex operational data into intuitive degradation status indicators, laying the technical foundation for intelligent management of equipment operation. Simultaneously, the collaborative analysis of core components and overall trends further improves the accuracy and reliability of predictions.

[0045] P40: Calculate the health status score of the target metering transformer based on the degradation status of the core components, combined with historical operating load data and the environmental parameters.

[0046] Furthermore, step P40 in this embodiment of the application also includes:

[0047] P41: Construct a multi-level health status scoring system, which includes a core component scoring unit, an operating load scoring unit, and an environmental parameter scoring unit; P52: Based on the core component scoring unit, receive the degradation status of the core components, calculate the degradation scores of the core and coil respectively, and generate a core component score; P43: Using the operating load scoring unit, calculate the cumulative high load time, load fluctuation frequency, and load change rate scores based on the historical operating load data, and generate an operating load score; P44: Through the environmental parameter scoring unit, calculate the temperature and humidity, vibration spectrum, and pollution level scores based on the environmental parameters, and generate an environmental parameter score; P45: According to the multi-level scoring formula, fuse the core component score, operating load score, and environmental parameter score to obtain the health status score of the target metering transformer.

[0048] The multi-level scoring formula is as follows:

[0049] ;in, Score the overall health status of the target metering transformer. Scoring of core components Operating load score and Environmental parameter scores are weighted coefficients for each of the multi-level scores.

[0050] It should be understood that, based on the degradation status of the core components, combined with historical operating load data and environmental parameters, a multi-level health status scoring system is used to comprehensively assess the impact of the equipment's core components, operating load, and external environmental factors, generating an overall health score.

[0051] First, a multi-level health status scoring system is constructed, comprising a core component scoring unit, an operational load scoring unit, and an environmental parameter scoring unit. Each unit addresses health influencing factors from different dimensions, making the scoring process hierarchical and refined. For example, the core component scoring unit focuses on the health status of the core and coils, while the operational load scoring unit focuses on the impact of load fluctuations and high-load operating time, and the environmental parameter scoring unit assesses the effects of external conditions such as temperature, humidity, and vibration on equipment health.

[0052] Next, the core component scoring unit calculates degradation scores for both the core and coil based on their degradation status. The core degradation score is primarily based on the decrease in the core's magnetic susceptibility, while the coil degradation score reflects the degree of coil loss, with a scoring range of 0-100. The core component score is calculated by weighting these two scores, ensuring that the health status of each component is accurately reflected.

[0053] Then, in the operating load scoring unit, load-related scores are calculated based on historical operating load data. Specifically, these include a cumulative high-load time score, assessing the cumulative duration the equipment operates near or above its design load; a load fluctuation frequency score, reflecting load instability; and a load change rate score, quantifying the degree of drastic load changes. The load scoring is designed to capture the operating stress of the equipment under different load conditions.

[0054] Next, using the environmental parameter scoring unit, temperature and humidity scores, vibration spectrum scores, and pollution level scores are calculated based on the environmental parameter data. The temperature and humidity score assesses potential changes in equipment insulation performance; the vibration spectrum score reflects the potential damage to the equipment caused by mechanical vibration; and the pollution level score assesses the long-term effects of external pollutants on equipment health. These scores comprehensively capture the impact of environmental conditions on equipment operation.

[0055] Finally, by integrating the core component scores, operating load scores, and environmental parameter scores using a multi-level scoring formula, an overall health status score for the target metering transformer is generated. The specific formula is as follows:

[0056] ;in, Score the overall health status of the target metering transformer. Scoring of core components Operating load score and The environmental parameter scores are weighted coefficients for multi-level scores. These weighted coefficients are determined through training with historical data to ensure that the influence of each scoring dimension matches the actual situation.

[0057] Through the aforementioned scoring system, the health status of the target metering transformer can be quantified into intuitive numerical results, and classified into categories such as healthy, average, or poor, providing a clear basis for equipment maintenance and operation management. This process relies on multi-dimensional feature fusion and weighted calculation, fully demonstrating the accuracy and scientific nature of intelligent health assessment.

[0058] P50: Based on the health status score, conduct a fault risk assessment of the target metering transformer, perform intelligent load allocation according to the fault risk assessment results, and optimize the load allocation using a load optimization strategy.

[0059] Furthermore, step P50 in this embodiment of the application also includes:

[0060] P51: Receive the health status score and its changing trend; P52: Extract the operating load characteristics and environmental parameters of the target metering transformer, and fuse them with the health status score to form a multi-dimensional input feature set; P53: Use a risk assessment model, combined with the multi-dimensional input feature set, to perform a fault risk assessment and obtain the fault risk assessment result.

[0061] In one possible embodiment of this application, by receiving health status scores and their changing trends, extracting multidimensional feature data, and constructing a risk assessment model, the failure risk can be accurately predicted and the load allocation strategy optimized.

[0062] First, during the health status score receiving phase, the system receives the latest health status score and its changing trend of the target metering transformer. The health status score is an indicator used to quantify the overall health status of the equipment. Combined with the changing trend (such as the rate of score decline), it can reflect the dynamic characteristics of equipment degradation. For example, if the score continues to decline and the rate of decline exceeds a preset threshold, it may indicate that the equipment has entered a rapid degradation phase, requiring close monitoring.

[0063] Next, the feature fusion stage begins. Real-time operating load characteristics and environmental parameters are extracted from the target metering transformers, including load fluctuation frequency, cumulative high load time, load change rate, as well as temperature and humidity, vibration spectrum, and pollution level. These features reflect the operating pressure on the equipment and the potential impact of the external environment on equipment performance. By fusing with the health status score, a multi-dimensional input feature set is formed, which includes both the current health status of the equipment and the combined effects of load and environmental factors. The feature set undergoes feature alignment and standardization to ensure temporal consistency and data scale uniformity across different feature dimensions.

[0064] Subsequently, a risk assessment model is used to analyze the failure risk of the target equipment. The risk assessment model can employ machine learning-based classification algorithms, such as random forests or deep neural networks, and is trained using historical failure data. The model takes a multi-dimensional feature set as input, combines health status scores and real-time operational data, and outputs the probability of occurrence for each potential failure type (e.g., 60% probability of insulation aging and 40% probability of coil loss). Simultaneously, based on the overall risk level, the model classifies equipment risk into low, medium, and high risk categories, providing a clear basis for intelligent load allocation.

[0065] After completing the fault risk assessment, intelligent load allocation is performed based on the assessment results. For high-risk equipment, the load pressure is reduced first to minimize the possibility of further degradation; for low-risk equipment, the load is appropriately increased to balance system operating efficiency. Based on the risk assessment results and current load status, the optimal load adjustment scheme is dynamically output. The load optimization strategy is verified through simulation to ensure the positive impact of the adjustment scheme on overall equipment health and system efficiency.

[0066] Furthermore, step P50 in this embodiment of the application also includes:

[0067] P54: Based on the fault risk assessment results, extract the risk level, fault occurrence probability, and health status score of the target metering transformer; P55: Collect the real-time operating status of the target metering transformer; P56: Use the real-time operating status, risk level, fault occurrence probability, and health status score as status inputs, define the load adjustment strategy as the action output, and construct a reinforcement learning model; P57: Based on the reinforcement learning model, dynamically adjust the load allocation strategy through the reward function to generate the load optimization strategy.

[0068] The reward function formula is as follows: ;in, For the cost of degradation, To the extent of risk mitigation, To improve system efficiency, , , These are the weighting coefficients for degradation cost, risk mitigation level, and system efficiency improvement, respectively, which are dynamically adjusted through training.

[0069] Optionally, further optimization can be achieved to generate intelligent load allocation and load optimization strategies based on the results of fault risk assessment.

[0070] First, the risk level, failure probability, and health status score of the target metering transformer are extracted based on the failure risk assessment results. The risk level is a classification of the potential failure probability of the equipment, such as low, medium, and high risk, used to determine the urgency of equipment operation; the failure probability is the likelihood of a specific failure type (such as insulation aging or coil loss) calculated based on historical data and real-time operating characteristics; the health status score provides a quantitative assessment of the overall operating status of the equipment. These data together constitute a description of the current health status of the equipment.

[0071] Next, the real-time operating status of the target metering transformer is collected, including operating characteristics such as current load, load fluctuation frequency, and cumulative high load time. This real-time status data reflects the current working pressure of the equipment and serves as an important reference for load adjustment strategies.

[0072] Then, the above data is input into a reinforcement learning model for modeling. Specifically, real-time operating status, risk level, failure probability, and health status score are used as state inputs, and load adjustment strategies are used as action outputs to construct a dynamic optimization model based on reinforcement learning. The reinforcement learning model learns the optimal load adjustment scheme autonomously by continuously interacting with the environment. For example, when the load pressure is high and the equipment is at a high risk level, the model can output an action to reduce the load; while when the equipment health is good and the load fluctuation is small, the load allocation can be appropriately increased to balance the equipment operating pressure and system efficiency.

[0073] The core of the reinforcement learning model lies in the design of the reward function. This reward function measures the quality of load adjustment actions, taking into account the following factors: degradation cost (increasing load may exacerbate equipment degradation, resulting in a negative reward); risk mitigation level (reducing the load on high-risk equipment can decrease the probability of failure, resulting in a positive reward); and system efficiency improvement (reasonable load allocation improves the overall system operating efficiency, leading to additional rewards).

[0074] Based on the aforementioned reward function, the reinforcement learning model can dynamically adjust the load allocation strategy and generate a load optimization strategy, ensuring that the optimized load allocation balances equipment health and system performance. For example, for high-risk equipment, the model prioritizes allocating lower loads to prevent further equipment degradation; while for low-risk equipment, the load can be appropriately increased to balance the overall load pressure. Furthermore, based on fault risk assessment, intelligent and dynamic optimization of load allocation for metering transformers is achieved, reducing equipment fault risk and improving system operating efficiency, thus providing technical assurance for the safe and efficient operation of metering transformers.

[0075] In summary, the embodiments of this application have at least the following technical effects:

[0076] This application collects operational data through sensors and performs multimodal fusion analysis. It then constructs a degradation analysis model by combining historical data and shared data. The degradation analysis model is used to predict the degradation status of core components and calculate a health status score to conduct a fault risk assessment. Based on the assessment results, the load allocation is dynamically optimized to achieve accurate fault prediction and improved operational efficiency.

[0077] The technology achieves the goal of early fault warning for metering transformers through degradation analysis and health assessment of core components, and extends the service life of the equipment through load optimization.

[0078] Example 2, based on the same inventive concept as the machine learning-based intelligent prediction method for metering transformer faults in the foregoing examples, such as... Figure 2As shown, this application provides a machine learning-based intelligent fault prediction system for metering transformers. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0079] The operation data acquisition module 11 is used to acquire the operation data of the target metering transformer through sensors. The operation data includes electrical parameters, load data and environmental parameters.

[0080] The key feature extraction module 12 is used to perform multimodal fusion analysis on the running data and extract key feature sequences.

[0081] The core component degradation prediction module 13 is used to predict the degradation state of the core components of the target metering transformer by using the degradation analysis model and taking the key feature sequence as input.

[0082] The health status assessment module 14 is used to calculate the health status score of the target metering transformer based on the degradation status of the core components, combined with historical operating load data and environmental parameters.

[0083] The fault risk assessment and optimization module 15 is used to assess the fault risk of the target metering transformer based on the health status score, perform intelligent load allocation according to the fault risk assessment results, and optimize the load allocation using a load optimization strategy.

[0084] Furthermore, the key feature extraction module 12 is also used to perform the following steps:

[0085] The electrical parameters, load data, and environmental parameters are subjected to feature filtering and dimensionality reduction to extract a set of key features, which includes different types of key features. Feature interaction analysis is used to identify the nonlinear correlation between the key features. Based on the nonlinear correlation, the different types of key features are integrated into the key feature sequence through a multimodal fusion model.

[0086] Furthermore, the core component degradation prediction module 13 is also used to perform the following steps:

[0087] Historical operating data of the target metering transformer is collected, and shared data of the same model of equipment is extracted. Based on the historical operating data and the shared data of the same model of equipment, data clustering analysis is performed according to the core component type, and a cross-device collaborative learning set is generated. The cross-device collaborative learning set contains multiple data subsets. Using the cross-device collaborative learning set as training data, and in combination with machine learning principles, independent component degradation trend training and overall degradation trend training are performed respectively to generate core component analysis units and overall trend analysis units, and these are then merged to generate the degradation analysis model.

[0088] Furthermore, the health status assessment module 14 is also used to perform the following steps:

[0089] A multi-level health status scoring system is constructed, comprising a core component scoring unit, an operating load scoring unit, and an environmental parameter scoring unit. Based on the core component scoring unit, the degradation status of the core components is received, and degradation scores for the core and coil are calculated separately to generate a core component score. Using the operating load scoring unit, based on historical operating load data, cumulative high load time, load fluctuation frequency, and load change rate scores are calculated to generate an operating load score. Through the environmental parameter scoring unit, based on environmental parameters, temperature and humidity, vibration spectrum, and pollution level scores are calculated to generate an environmental parameter score. According to the multi-level scoring formula, the core component score, operating load score, and environmental parameter score are fused and calculated to obtain the health status score of the target metering transformer.

[0090] Furthermore, the health status assessment module 14 is also used to perform the following steps:

[0091] The multi-level scoring formula is as follows: ;in, Score the overall health status of the target metering transformer. Scoring of core components Operating load score and Environmental parameter scores are weighted coefficients for each of the multi-level scores.

[0092] Furthermore, the fault risk assessment and optimization module 15 is also used to perform the following steps:

[0093] The system receives the health status score and its changing trend; extracts the operating load characteristics and environmental parameters of the target metering transformer, and integrates them with the health status score to form a multi-dimensional input feature set; uses a risk assessment model, combined with the multi-dimensional input feature set, to perform a fault risk assessment and obtain the fault risk assessment result.

[0094] Furthermore, the fault risk assessment and optimization module 15 is also used to perform the following steps:

[0095] Based on the fault risk assessment results, the risk level, fault probability, and health status score of the target metering transformer are extracted; the real-time operating status of the target metering transformer is collected; the real-time operating status, risk level, fault probability, and health status score are used as status inputs, and a load adjustment strategy is defined as the action output to construct a reinforcement learning model; based on the reinforcement learning model, the load allocation strategy is dynamically adjusted through a reward function to generate the load optimization strategy. The reward function formula is: ;in, For the cost of degradation, To the extent of risk mitigation, To improve system efficiency, , , These are the weighting coefficients for degradation cost, risk mitigation level, and system efficiency improvement, respectively, which are dynamically adjusted through training.

[0096] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0097] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0098] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A machine learning-based intelligent fault prediction method for metering transformers, characterized in that, The method includes: The operating data of the target metering transformer is collected by sensors, and the operating data includes electrical parameters, load data and environmental parameters; Multimodal fusion analysis was performed on the operational data, and key feature sequences were extracted; Using a degradation analysis model, with the key feature sequence as input, the degradation status of the core components of the target metering transformer is predicted. Based on the degradation status of the core components, combined with historical operating load data and environmental parameters, the health status score of the target metering transformer is calculated. Based on the health status score, a fault risk assessment is performed on the target metering transformer. Intelligent load allocation is then performed based on the fault risk assessment results, and a load optimization strategy is used to optimize the load allocation.

2. The intelligent fault prediction method for metering transformers based on machine learning as described in claim 1, characterized in that, The operational data is subjected to multimodal fusion analysis, and key feature sequences are extracted, including: The electrical parameters, load data, and environmental parameters are subjected to feature filtering and dimensionality reduction to extract a set of key features, which includes different types of key features. Feature interaction analysis is used to identify nonlinear correlations between key features; Based on the aforementioned nonlinear correlation, a multimodal fusion model is used to integrate different types of key features into the key feature sequence.

3. The intelligent fault prediction method for metering transformers based on machine learning as described in claim 1, characterized in that, The construction of the degradation analysis model includes: Collect historical operating data of the target metering transformer and extract shared data of the same model of equipment; Based on the historical operating data and shared data of the same model of equipment, data clustering analysis is performed according to the core component type, and a cross-device collaborative learning set is generated. The cross-device collaborative learning set contains multiple data subsets. Using the cross-device collaborative learning set as training data, and combining machine learning principles, the degradation analysis model is trained and generated.

4. The intelligent fault prediction method for metering transformers based on machine learning as described in claim 3, characterized in that, Using the cross-device collaborative learning dataset as training data, and combining machine learning principles, the degradation analysis model is trained and generated, including: Using the cross-device collaborative learning set as training data, the core component training dataset and the overall training dataset are extracted respectively. By combining machine learning principles, we conduct independent component degradation trend training and overall degradation trend training respectively, and generate core component analysis units and overall trend analysis units. The degradation analysis model is generated by integrating the core component analysis unit and the overall trend analysis unit.

5. The intelligent fault prediction method for metering transformers based on machine learning as described in claim 1, characterized in that, Based on the degradation status of the core components, combined with historical operating load data and environmental parameters, a health status score for the target metering transformer is calculated, including: A multi-level health status scoring system is constructed, which includes a core component scoring unit, an operational load scoring unit, and an environmental parameter scoring unit. Based on the core component scoring unit, the degradation status of the core component is received, and the degradation scores of the iron core and the coil are calculated respectively to generate the core component score; Using the aforementioned operating load scoring unit, based on the historical operating load data, the cumulative high load time, load fluctuation frequency, and load change rate scores are calculated to generate an operating load score; The environmental parameter scoring unit calculates temperature and humidity, vibration spectrum, and pollution level scores based on the environmental parameters to generate an environmental parameter score. Based on the multi-level scoring formula, the core component score, operating load score, and environmental parameter score are fused and calculated to obtain the health status score of the target metering transformer.

6. The intelligent fault prediction method for metering transformers based on machine learning as described in claim 5, characterized in that, The multi-level scoring formula is as follows: ； in, Score the overall health status of the target metering transformer. Scoring of core components Operating load score and Environmental parameter scores are weighted coefficients for each of the multi-level scores.

7. The intelligent fault prediction method for metering transformers based on machine learning as described in claim 1, characterized in that, Based on the health status score, a fault risk assessment of the target metering transformer is performed, including: Receive the health status score and its changing trend; Extract the operating load characteristics and environmental parameters of the target metering transformer, and fuse them with the health status score to form a multi-dimensional input feature set; Using a risk assessment model, combined with the multidimensional input feature set, a fault risk assessment is performed to obtain the fault risk assessment result.

8. The intelligent fault prediction method for metering transformers based on machine learning as described in claim 7, characterized in that, Intelligent load allocation is performed based on the fault risk assessment results, including: Based on the fault risk assessment results, the risk level, fault occurrence probability, and health status score of the target metering transformer are extracted. Collect the real-time operating status of the target metering transformer; Using the real-time operating status, risk level, failure probability, and health status score as status inputs, and defining the load adjustment strategy as the action output, a reinforcement learning model is constructed. Based on the reinforcement learning model, the load allocation strategy is dynamically adjusted through the reward function to generate the load optimization strategy.

9. The intelligent fault prediction method for metering transformers based on machine learning as described in claim 8, characterized in that, The reward function formula is as follows: ； in, For the cost of degradation, To the extent of risk mitigation, To improve system efficiency, , , These are the weighting coefficients for degradation cost, risk mitigation level, and system efficiency improvement, respectively, which are dynamically adjusted through training.

10. A machine learning-based intelligent fault prediction system for metering transformers, characterized in that, The system includes: The system includes an operation data acquisition module, which is used to acquire the operation data of the target metering transformer through sensors. The operation data includes electrical parameters, load data, and environmental parameters. A key feature extraction module is used to perform multimodal fusion analysis on the running data and extract key feature sequences; A core component degradation prediction module is used to predict the degradation state of the core components of the target metering transformer by using the degradation analysis model and taking the key feature sequence as input. A health status assessment module is used to calculate the health status score of the target metering transformer based on the degradation status of the core components, combined with historical operating load data and environmental parameters. The fault risk assessment and optimization module is used to assess the fault risk of the target metering transformer based on the health status score, perform intelligent load allocation according to the fault risk assessment results, and optimize the load allocation using a load optimization strategy.

Images