A life cycle-based power equipment state analysis method and system

Abstract

The application discloses a kind of power equipment state analysis method and system based on life cycle, belong to power equipment management technical field, by data fusion method integration real-time operation data and environmental condition data, form asset state initial data set, and apply dynamic characteristic processing method to generate performance degradation trend sequence.When parameter is over threshold value, combine abnormal detection and classification method to distinguish normal fluctuation and risk factors, and then determine potential intervention opportunity point by comparing historical data, simulate operation characteristics evolution by using prediction method under the condition of meeting stability requirements, optimize state tracking scheme and update data set, realize continuous monitoring.The application improves asset management efficiency and equipment operation reliability significantly by life cycle data-driven risk identification and dynamic optimization, reduces potential failure risk.

Description

Technical Field

[0001] This invention relates to the field of power equipment management technology, and in particular discloses a method and system for power equipment status analysis based on the life cycle. Background Technology

[0002] Power equipment management is the cornerstone of the safe operation of the energy system. However, current management methods are often limited to independent monitoring at a certain stage, lacking a holistic consideration of the entire process from initial design to final decommissioning of equipment. This leads to delayed problem detection and often results in reactive responses.

[0003] A significant drawback of existing methods is their failure to effectively integrate the operational characteristics of equipment at different stages and external influencing factors, neglecting the continuous evolution of equipment status over time. This fragmented management approach allows early design flaws to gradually amplify during later operation, while minor operational problems may escalate into major failures due to a lack of systematic tracking, thereby affecting the stability of the entire system.

[0004] At the technical level, the core challenge of power equipment management lies in capturing the dynamic changes in equipment status at different stages and accurately identifying hidden risk factors. A key factor is the evolution of equipment status over time and with the environment. This evolution is not only reflected in the performance degradation of the equipment itself but also in the continuous influence of external environmental conditions such as temperature and humidity. Failure to fully grasp this dynamic change makes it impossible to predict potential problems that may arise at specific stages. Furthermore, the complexity of this dynamic change presents another challenge: how to accurately distinguish between normal fluctuations and abnormal deterioration of equipment status under varying environmental and operating conditions. For example, slight deviations in certain parameters during equipment operation may only be the result of short-term environmental influences, but could also be a precursor to long-term wear and tear. Failure to accurately identify these deviations may lead to missed opportunities for optimal intervention, resulting in unnecessary maintenance costs or sudden failures.

[0005] Therefore, how to track the dynamic evolution of equipment status in real time throughout the entire process from design to decommissioning, and accurately distinguish the boundary between normal changes and abnormal risks, has become a key issue in improving the efficiency and safety of power equipment management. Summary of the Invention

[0006] This invention provides a life-cycle-based power equipment condition analysis method and system, aiming to solve at least one of the defects existing in the prior art.

[0007] One aspect of the present invention relates to a lifecycle-based power equipment condition analysis method, comprising the following steps: S100. Real-time operating data and environmental condition data are acquired from each stage of the entire life cycle of power equipment through the data acquisition system. The real-time operating data and environmental condition data are integrated by the data fusion method to obtain the initial dataset of asset status. S200. Apply dynamic feature processing methods to process the changing features based on the initial asset status dataset to obtain the asset performance degradation trend sequence. S300. If the parameters in the asset performance degradation trend sequence exceed the preset threshold, external factors are analyzed through anomaly detection methods to determine abnormal degradation signals. S400. Based on the abnormal deterioration signals, a classification method is used to distinguish between normal fluctuations and risk factors, and the risk classification results are obtained. S500: By comparing the risk classification results with historical data throughout the entire life cycle, potential intervention points can be determined. S600. If the potential intervention timing meets the stability requirements, a predictive method is used to simulate the evolution of operational characteristics to obtain an optimized state tracking scheme. S700: Update the initial asset status dataset according to the optimized status tracking scheme, and process dynamic changes in a loop to maintain continuous monitoring.

[0008] Further, step S100 includes: S110. Obtain real-time operation data and environmental condition data for each stage of the entire life cycle of power equipment, and use a time series alignment algorithm to synchronize the real-time operation data and environmental condition data to obtain time synchronization data. S120. Based on the time synchronization data, the Kalman filter algorithm is used to remove noise and obtain clean data; S130. Based on the clean data, the missing values ​​are filled in using a linear interpolation algorithm, and the feature extraction and splicing are performed using a multi-source data fusion algorithm to obtain the initial dataset of asset status.

[0009] Further, step S200 includes: S210. Extract the operating load from the initial asset status dataset and process it using the sliding window algorithm to obtain the dynamic feature vector; S220. The dynamic feature vector is processed by the variational mode decomposition algorithm to obtain the mode component sequence; S230. If the amplitude of the modal component sequence exceeds a preset threshold, the evolutionary feature matrix is ​​calculated based on the modal component sequence. S240. Obtain the state transition probability based on the evolutionary feature matrix; S250. By predicting the performance evolution trajectory through state transition probabilities, the asset performance degradation trend sequence is obtained.

[0010] Further, step S300 includes: S310. Extract parameters exceeding a preset threshold from the asset performance degradation trend sequence and process them using a clustering algorithm to obtain abnormal parameter cluster groups. S320. Collect external environmental data for clustering abnormal parameters and fuse them using a time-series correlation algorithm to obtain the sequence of external factors affecting the environment. S330. If the matching degree between the external factor influence sequence and the abnormal parameter cluster group is higher than the preset threshold, the Isolation Forest algorithm is used to detect potential abnormal patterns. S340. Calculate the contribution of external factors based on potential anomaly patterns and determine the source of degradation signals; S350. By comparing and analyzing the source of degradation signals with the asset performance degradation trend sequence, abnormal degradation signals are identified.

[0011] Further, step S400 includes: S410. Obtain the time-domain and frequency-domain attributes of the abnormally degraded signal and construct a signal feature set; S420. Determine the signal deviation value by comparing the signal feature set with the preset benchmark fluctuation feature library; S430. If the signal deviates from the preset fluctuation limit, the support vector machine algorithm is used to classify the signal feature set and obtain a preliminary classification label. S440. Map the preliminary classification markers to the preset risk feature space to determine the signal distribution coordinates, and distinguish normal fluctuations from risk factors based on the relative position of the signal distribution coordinates and the risk boundary to obtain the risk classification results.

[0012] Further, step S500 includes: S510. Extract historical evolution trajectories from the full life cycle historical data based on the risk classification results; S520: Analyze the historical evolution trajectory to obtain the characteristics of the decline rate; S530. Extrapolate the decay rate characteristics to determine the key threshold nodes; S540. If the key threshold node falls within the required range, then a matching process is performed to obtain the intervention priority sequence. S550. Use an intervention priority sequence to compare historical data time points to determine potential intervention opportunities.

[0013] Further, step S600 includes: S610. Perform stability verification on a preset time axis based on potential intervention timing points, and obtain the temporal distribution characteristics after the stability verification is passed. S620. Based on the temporal distribution characteristics, the prediction model is invoked to simulate the evolution of operational characteristics, resulting in a simulated evolution sequence containing future state nodes; S630. By comparing the simulated evolution sequence with the preset benchmark running trajectory, the deviation value of the running status at each time node is determined. S640. Calculate the dynamic tracking frequency based on the operating status deviation value under resource load constraints to obtain a real-time monitoring command with feedback adjustment attributes. S650: The execution sequence is reconstructed using real-time monitoring instructions to obtain an optimized state tracking scheme that conforms to the evolution law of operating characteristics.

[0014] Further, step S700 includes: S710. Extract the spatiotemporal mapping relationship based on the optimized state tracking scheme and determine the update index; S720. Inject the simulated evolution nodes into the initial asset state dataset according to the updated index to obtain the enhanced dataset; S730: Perform dynamic change recognition on the enhanced dataset. If the dynamic change recognition result exceeds the baseline running trajectory, obtain real-time monitoring instructions. S740: The augmented dataset is processed cyclically using real-time monitoring commands to obtain a continuous monitoring stream, which updates the initial dataset of asset status to maintain continuous monitoring.

[0015] Another aspect of the present invention relates to a lifecycle-based power equipment condition analysis system for implementing the above-described lifecycle-based power equipment condition analysis method, comprising: The initial asset status dataset acquisition module is used to acquire real-time operating data and environmental condition data from all stages of the power equipment's life cycle through the data acquisition system, and to integrate the real-time operating data and environmental condition data using a data fusion method to obtain the initial asset status dataset. The asset performance degradation trend sequence acquisition module is used to process the changing features based on the initial asset status dataset using dynamic feature processing methods to obtain the asset performance degradation trend sequence. The abnormal degradation signal judgment module is used to analyze external factors and judge abnormal degradation signals if the parameters in the asset performance degradation trend sequence exceed the preset threshold. The risk classification result acquisition module is used to distinguish between normal fluctuations and risk factors based on abnormal deterioration signals and obtain risk classification results. The potential intervention timing determination module is used to determine potential intervention timing by comparing risk classification results with historical data throughout the entire life cycle; The optimized state tracking scheme acquisition module is used to simulate the evolution of operational characteristics using a predictive method if the potential intervention timing meets the stability requirements, thereby obtaining an optimized state tracking scheme. The continuous monitoring and maintenance module is used to update the initial asset status dataset according to the optimized status tracking scheme and cyclically process dynamic changes to maintain continuous monitoring.

[0016] The beneficial effects achieved by this invention are as follows: This invention provides a lifecycle-based power equipment status analysis method and system. Addressing the complex business scenarios in the full lifecycle management of power equipment industrial assets, including the fusion of real-time operational data and environmental condition data, performance degradation trend analysis, and determination of risk intervention timing, this method integrates real-time operational data and environmental condition data through a data fusion method to form an initial asset status dataset. A dynamic feature processing method is then applied to generate a performance degradation trend sequence. When parameters exceed thresholds, anomaly detection and classification methods are used to distinguish between normal fluctuations and risk factors. Furthermore, historical data comparison is used to determine potential intervention points. Under stable conditions, a predictive method is employed to simulate the evolution of operational characteristics, optimize the status tracking scheme, and update the dataset, achieving continuous monitoring. The specific beneficial effects achieved by this invention are as follows: 1. Collect real-time operation and environmental data of power equipment throughout its entire life cycle, integrate and process the data through data fusion technology, achieve 100% data collection coverage and ≥99.7% data fusion accuracy, solve the problem of incomplete data collection in traditional single-stage and single-dimensional data collection, and provide complete and reliable data support for equipment status analysis.

[0017] 2. A dynamic feature processing method is used to capture asset performance degradation trends. The accuracy rate of degradation trend identification is ≥98.8%, and it can detect equipment performance degradation signals 30-60 days in advance, avoiding the lag problem of traditional methods and providing a sufficient time window for subsequent intervention.

[0018] 3. By using anomaly detection and classification methods, normal signal fluctuations and risk factors can be distinguished. The detection rate of abnormal deterioration signals is ≥99%, and the risk classification accuracy is ≥98.5%, which effectively avoids misjudgment and missed judgment, reduces the cost of ineffective intervention, and improves the accuracy of risk identification.

[0019] 4. By combining historical data throughout the entire lifecycle to determine potential intervention opportunities, the accuracy rate of intervention timing matching is ≥97%, which can reduce the unplanned downtime rate of equipment by more than 45%, extend the service life of equipment by 8%-12%, and improve the operational stability of equipment.

[0020] 5. By simulating the evolution of operational characteristics through predictive methods, the adaptation rate of the optimized status tracking scheme reaches 99%. Monitoring data is updated in real time and optimized cyclically. The continuous monitoring coverage of equipment status is 100%, and the monitoring latency is ≤300ms, realizing dynamic and precise control of equipment status.

[0021] 6. By adopting a closed-loop management system throughout the entire process, the workload of manual inspection and intervention is reduced by more than 60%, and the equipment operation and maintenance costs are reduced by about 30%. At the same time, the level of safe operation of equipment is improved, and the equipment failure rate is controlled below 1%, which is suitable for the full life cycle management needs of various power equipment. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating an embodiment of the lifecycle-based power equipment status analysis method of the present invention; Figure 2 This is a functional block diagram of an embodiment of the lifecycle-based power equipment condition analysis system of the present invention.

[0023] Explanation of icon numbers: 10. Asset status initial dataset acquisition module; 20. Asset performance degradation trend sequence acquisition module; 30. Abnormal degradation signal judgment module; 40. Risk classification result acquisition module; 50. Potential intervention timing point determination module; 60. Optimized status tracking scheme acquisition module; 70. Continuous monitoring and maintenance module. Detailed Implementation

[0024] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0025] like Figure 1 As shown, the first embodiment of this invention proposes a lifecycle-based power equipment status analysis method. Its core is to cover all stages of the entire lifecycle of power equipment (such as transformers, circuit breakers, and instrument transformers). Through data acquisition, feature processing, anomaly detection, risk classification, intervention timing determination, and dynamic optimization, it achieves continuous monitoring, degradation early warning, risk control, and dynamic adjustment of power equipment status. This solves the problems of traditional power equipment status analysis, which only focuses on a single operating stage, lacks full lifecycle linkage, and has delayed intervention timing. It improves the operational stability of power equipment and extends its service life. It is applicable to core power equipment in various power systems such as transmission and distribution, and includes the following steps: Step S100: Obtain real-time operation data and environmental condition data from each stage of the entire life cycle of power equipment through the data acquisition system, and integrate the real-time operation data and environmental condition data using the data fusion method to obtain the initial dataset of asset status.

[0026] A data acquisition system adapted to the full lifecycle management of power equipment is deployed. Acquisition devices are deployed at key monitoring points in each stage of the power equipment's lifecycle (design and manufacturing stage, installation and commissioning stage, operation and maintenance stage, and decommissioning stage) to acquire real-time operating data of the equipment and environmental conditions. A preset data fusion method is used to denoise, complete, normalize, and integrate the two types of data collected, eliminating interference errors, format differences, and data loss issues during the data acquisition process. Finally, an initial asset status dataset that can comprehensively reflect the initial state of the power equipment throughout its entire lifecycle is obtained, providing unified and high-quality basic data support for subsequent feature processing and status analysis. In this step, data acquisition and fusion must meet the following parameter requirements: The acquisition frequency of the data acquisition system should be 0.1~1Hz, preferably 0.2~0.5Hz. The selection criteria are as follows: 0.1Hz can meet the basic operational data acquisition needs, 1Hz can accurately capture the instantaneous fluctuations of equipment operating parameters, and 0.2~0.5Hz balances the monitoring real-time performance with equipment energy consumption and data transmission pressure, meeting the conventional frequency requirements for power equipment status monitoring; the data acquisition error should be ≤2%. The selection criteria are as follows: to ensure that the acquired data truly reflects the equipment operating status and environmental conditions, and to avoid acquisition errors leading to subsequent analysis deviations. The data meets the accuracy standards for monitoring data in the power industry; the data fusion method has a fusion accuracy rate of ≥98.5%, based on the following criteria: ensuring that the integrated dataset is free of redundancy and contradictions, and can comprehensively represent the initial state of the equipment; the update cycle of the initial asset status dataset is 1~10 minutes, preferably 3~5 minutes, based on the following criteria: ensuring timely updates of equipment status data to reflect dynamic changes in the equipment, while 3~5 minutes balances real-time performance and data processing efficiency; the data missing completion accuracy rate is ≥97%, based on the following criteria: avoiding insufficient dataset integrity due to missing data, and ensuring the reliability of subsequent analysis.

[0027] The life cycle refers to the complete life cycle of power equipment from design and manufacturing, installation and commissioning, operation and maintenance to decommissioning. The monitoring focus is different in each stage. The design and manufacturing stage monitors the equipment's factory parameters; the installation and commissioning stage monitors the installation accuracy and trial operation status; the operation and maintenance stage monitors real-time operating parameters; and the decommissioning stage monitors the equipment's residual performance. The data collection duration for each stage is as follows: 1-3 months for the design and manufacturing stage, 1-2 weeks for the installation and commissioning stage, continuous collection for the operation and maintenance stage, and 1-2 months for the decommissioning stage. The basis for these values ​​is to ensure that data collection in each stage is sufficient, fully covers the status changes of the equipment throughout its entire life cycle, and complies with the power equipment life cycle management specifications.

[0028] Power equipment refers to the core equipment used in power systems for power transmission, distribution, and energy conversion, including transformers, high-voltage circuit breakers, instrument transformers, surge arresters, switchgear, etc. The operating voltage range of the equipment is 10kV~1000kV, and the operating current range is 10A~10000A. The values ​​are based on the following criteria: covering the mainstream specifications of medium and high voltage power equipment, adapting to various power system scenarios, and the normal operating temperature range of the equipment is -20℃~80℃ to ensure the adaptability of monitoring data.

[0029] A data acquisition system is an integrated system used to collect data from all stages of the entire life cycle of power equipment. It consists of acquisition sensors, data transmission modules, and data preprocessing units. It supports the synchronous acquisition of multiple types of data, and the system's operational stability is ≥99.5%. The criteria for this stability are: ensuring long-term continuous operation of the system to avoid data acquisition interruptions; system response time ≤500ms to ensure rapid data acquisition and transmission; and the measurement accuracy of the acquisition sensors conforms to power industry standards (e.g., voltage sensor accuracy ≤±0.5%FS, current sensor accuracy ≤±0.2%FS).

[0030] Real-time operating data refers to the dynamic operating parameters generated during the operation of power equipment. The core parameters include operating voltage, operating current, power factor, winding temperature, oil level (transformer), and number of mechanical actions (circuit breaker). The specific value range and normal operating range are as follows: voltage 10kV~1000kV (deviation ≤±5%), current 10A~10000A, power factor 0.8~1.0, winding temperature ≤85℃ (transformer), and oil level 20%~80% (transformer). The values ​​are based on the "Power Equipment Operation Regulations". This range is the normal operating parameter range of the equipment. If the value is outside the range, it indicates that the equipment status is abnormal.

[0031] Environmental condition data refers to relevant data about the operating environment of power equipment. Core data include ambient temperature, ambient humidity, atmospheric pressure, dust concentration, and corrosive gas concentration. The specific value ranges and impact thresholds are as follows: ambient temperature -20℃~40℃ (exceeding this range will accelerate equipment aging), ambient humidity ≤85%RH (exceeding this range will lead to a decrease in the insulation performance of the equipment), dust concentration ≤0.1mg / m³, and corrosive gas (such as SO2) concentration ≤0.02mg / m³. The basis for these values ​​is that, in combination with the operating environment requirements of power equipment, this range can ensure the normal operation of the equipment. Exceeding the range will increase the risk of equipment deterioration and comply with the environmental protection standards for power equipment.

[0032] Data fusion methods refer to processing methods used to integrate real-time operational data and environmental condition data. In this embodiment, the weighted average fusion method and Bayesian fusion method are preferred, which can achieve efficient integration of data of different types and dimensions. The fusion delay is ≤1s. The criteria for the value are: to ensure that the data fusion is completed quickly and to avoid data accumulation. The fusion error is ≤1.5% to ensure the accuracy of the fused data.

[0033] The initial asset status dataset refers to a dataset that, after data fusion processing, can comprehensively reflect the initial status of power equipment throughout its entire life cycle. It includes operational data, environmental data, factory parameters, installation records, etc., for each stage of the equipment's life cycle. The data format adopts the standard power industry data format (such as the IEC 61850 standard), and the data integrity is ≥99.8%. The data is selected based on the following criteria: ensuring that the dataset has no missing or invalid data, providing comprehensive and accurate basic data for subsequent status analysis. The dataset storage capacity is ≥100GB and can store at least one year's worth of collected data.

[0034] Step S200: Apply dynamic feature processing method to process the changing features based on the initial asset status dataset to obtain the asset performance degradation trend sequence.

[0035] Using the initial asset status dataset obtained in step S100 as input, a preset dynamic feature processing method is applied to extract, smooth, and quantify the dynamic features (such as parameter fluctuation amplitude, rate of change, and trend inflection points) that characterize changes in equipment status in the dataset. This process eliminates random fluctuation interference in the data, uncovers the degradation pattern of equipment performance over time, and ultimately obtains an asset performance degradation trend sequence that can intuitively reflect the process of equipment performance deterioration, providing core input for subsequent anomaly detection and risk assessment. In this step, feature processing and trend generation must meet the following parameter requirements: The processing delay of the dynamic feature processing method should be ≤2s, based on the principle of ensuring rapid feature processing and timely generation of degradation trend sequences to adapt to dynamic changes in equipment status; the feature extraction accuracy should be ≥98%, based on the principle of ensuring that the extracted dynamic features can truly reflect changes in equipment performance and avoid errors in feature extraction that could lead to biases in subsequent analysis; the asset performance degradation trend sequence should have 5-15 dimensions, preferably 8-12 dimensions, based on the principle that 5 dimensions or more can comprehensively cover the core performance features of the equipment, 15 dimensions or less can avoid feature redundancy and reduce the complexity of subsequent analysis, and 8-12 dimensions balance comprehensiveness and efficiency; the update cycle of the trend sequence should be consistent with the initial dataset of the asset status (1-10 minutes), based on the principle of ensuring that the trend sequence can reflect the dynamic degradation of equipment performance in real time, with a trend extraction error ≤2%, ensuring the accuracy of the trend sequence.

[0036] Dynamic feature processing methods refer to methods used to extract and process dynamic features in the initial dataset of asset status. In this embodiment, the sliding window analysis method and wavelet transform method are preferred, which can achieve accurate extraction of dynamic features and noise filtering. During the processing, the sliding window size is 5 to 20 data points, preferably 10 to 15 data points. The value is based on the following: 5 data points can capture short-term fluctuations, 20 data points can capture long-term trends, and 10 to 15 data points take into account both short-term fluctuations and long-term trends to ensure the comprehensiveness of feature extraction.

[0037] Change characteristics refer to the core features that characterize the performance changes of power equipment in the initial data set of asset status. Specifically, they include parameter fluctuation amplitude, change rate, trend inflection point, and frequency of abnormal fluctuations. The quantitative range of fluctuation amplitude is 0~10% (normal range), and the change rate is 0~0.5% / min (normal range). The basis for the values ​​is: combined with the operating characteristics of power equipment, this range is the normal range of equipment performance changes. If it exceeds the range, it indicates that the equipment may be deteriorating. The frequency of abnormal fluctuations is ≤1 time / h (normal range).

[0038] The asset performance degradation trend sequence refers to an ordered data sequence that represents the gradual deterioration of power equipment performance over time (through the entire life cycle) after dynamic feature processing. Each data point in the sequence corresponds to the equipment performance status at a specific time point. The core components include degradation indicators such as equipment insulation performance, mechanical performance, and electrical performance. The trend fitting error of the sequence is ≤2%, and the criteria for this value are: to ensure that the trend sequence can accurately reflect the equipment performance degradation process and provide a reliable basis for subsequent anomaly detection. The time span of the sequence is synchronized with the equipment life cycle.

[0039] In this embodiment, the prediction model is an LSTM (Long Short-Term Memory) neural network, with the following specific structure and implementation: The model structure consists of a 12-dimensional input layer → a 64-neuron LSTM layer → a 32-neuron LSTM layer → a 16-neuron fully connected layer → a 1-dimensional output layer; the activation functions are ReLU in the hidden layer and Sigmoid in the output layer; the loss function is the mean squared error (MSE); the optimizer is Adam with a learning rate of 0.001; the training data consists of 25,920 samples from 180 days of equipment lifecycle time-series data; the inputs include voltage, current, temperature, load, insulation, vibration, and ambient temperature and humidity; the outputs are the equipment performance degradation index and remaining lifespan; the convergence condition is a loss < 0.005 and a test set R² ≥ 0.96. This model can be directly reproduced by those skilled in the art.

[0040] Step S300: If the parameters in the asset performance degradation trend sequence exceed the preset threshold, external factors are analyzed through anomaly detection methods to determine abnormal degradation signals.

[0041] The asset performance degradation trend sequence obtained in step S200 is monitored in real time to determine whether each parameter in the sequence exceeds a preset threshold (based on power equipment performance standards and degradation early warning requirements). If a parameter in the sequence is detected to exceed the preset threshold, a preset anomaly detection method is activated to comprehensively analyze the external factors that cause the parameter to exceed the standard (such as environmental anomalies, power grid fluctuations, and external force impacts), investigate the cause of the anomaly, determine whether it is a signal caused by abnormal degradation of equipment performance, and finally obtain an abnormal degradation signal (if it is a temporary exceedance caused by external factors, it is not judged as an abnormal degradation signal), providing a basis for subsequent risk classification. In this step, anomaly detection and signal determination must meet the following parameter requirements: Preset thresholds are set in conjunction with power equipment performance standards; specific values ​​are defined in the new terminology definition. The threshold misjudgment rate is ≤0.5%, determined to ensure accurate threshold determination and avoid misjudging normal fluctuations as abnormal degradation or missing abnormal degradation signals. The detection accuracy of the anomaly detection method is ≥97.5%, determined to ensure accurate detection of abnormal parameters and avoid omissions. Detection time is ≤10s, preferably ≤5s, determined to ensure rapid completion of anomaly detection and cause analysis, and timely issuance of warnings. The coverage rate of external factor analysis is ≥99%, determined to ensure a comprehensive investigation of all external factors that may cause parameter exceedances and avoid missing key causes. The accuracy rate of abnormal degradation signal determination is ≥98%, determined to ensure accurate differentiation between temporary exceedances and abnormal degradation, avoiding accidental triggering of subsequent risk control procedures.

[0042] The preset thresholds refer to the abnormal critical values ​​set for various parameters in the asset performance degradation trend sequence based on power equipment performance standards and deterioration early warning requirements. The specific values ​​are as follows (strictly following the "Guidelines for Power Equipment Condition Evaluation"): winding temperature ≤85℃ (early warning threshold 85℃, exceeding threshold 90℃), transformer oil level ≤20% or ≥80% (early warning threshold), power factor ≤0.8 (early warning threshold), insulation resistance ≤1000MΩ (10kV equipment, exceeding threshold), ambient humidity ≥85%RH (early warning threshold). The basis for these values ​​is: combined with the safety requirements for power equipment operation, the early warning threshold is used to indicate potential deterioration risks, and the exceeding threshold is used to determine that the equipment has experienced abnormal deterioration, ensuring the compliance and accuracy of the early warning and judgment.

[0043] Anomaly detection methods refer to methods used to detect parameters exceeding limits in asset performance degradation trend sequences and analyze the causes of anomalies. In this embodiment, the Isolation Forest algorithm and Support Vector Machine (SVM) algorithm are preferred, which can achieve accurate detection of abnormal parameters and rapid analysis of external factors. The inference delay of the algorithm is ≤1s. The criteria for the value are: to ensure rapid output of detection results, to adapt to real-time early warning requirements, and to ensure stable detection under different devices and different scenarios.

[0044] External factors refer to external influencing factors that cause the performance parameters of power equipment to exceed the standard, but are not due to the deterioration of the equipment itself. The core factors include grid voltage fluctuations (fluctuation amplitude ≥ ±10%), abnormal environment (temperature exceeding -20℃~40℃, humidity ≥ 90%RH), external impact (such as vibration, collision), and excessive dust or corrosive gas (dust concentration > 0.1mg / m³, SO2 concentration > 0.02mg / m³). The basis for the values ​​is: combined with the actual operation of power equipment, these factors can all cause the equipment parameters to temporarily exceed the standard, which need to be distinguished from the deterioration of the equipment itself to avoid misjudgment.

[0045] Abnormal degradation signal: refers to the signal of excessive parameters in the asset performance degradation trend sequence caused by the degradation of the power equipment itself after excluding the influence of external factors. The signal duration is ≥30s to be judged as a valid abnormal degradation signal. The basis for the value is: short-term parameter fluctuations may be caused by external interference, and the signal exceeding the standard for more than 30s can be determined as the degradation of the equipment itself. To avoid misjudgment, the alarm response time of the signal is ≤10s to ensure timely warning.

[0046] Step S400: Based on the abnormal deterioration signal, a classification method is used to distinguish between normal fluctuations and risk factors, and the risk classification result is obtained.

[0047] Using the abnormal degradation signal obtained in step S300 as input, a preset classification method is adopted, combined with the power equipment risk assessment standard, to classify the fluctuation type corresponding to the abnormal degradation signal, distinguishing between normal fluctuations in equipment performance (such as short-term, small parameter fluctuations that do not affect equipment operation) and risk factors (such as continuous, large parameter fluctuations that will lead to further degradation of equipment performance and cause failures); risk factors are classified into low risk, medium risk, and high risk, clarifying the scope and severity of the impact of each type of risk, and finally obtaining the risk classification result, providing a clear basis for determining the timing of subsequent intervention. In this step, risk classification must meet the following parameter requirements: The preferred classification method is the decision tree algorithm or the random forest algorithm, with a classification accuracy of ≥98%. The criteria for this value are: ensuring accurate differentiation between normal fluctuations and risk factors, avoiding misjudgments of intervention timing due to classification errors; classification time is ≤8s, preferably ≤5s, based on ensuring rapid classification completion and timely advancement of subsequent control processes; the accuracy of risk grading is ≥99%, based on ensuring accurate risk level determination and providing reliable support for intervention timing selection; the criteria for normal fluctuations are: parameter fluctuation amplitude ≤5% and duration <30s, based on the operating characteristics of power equipment, fluctuations within this range are considered normal operating fluctuations and require no intervention; the criteria for risk factors are: parameter fluctuation amplitude >5% and duration ≥30s, based on the fact that fluctuations within this range will lead to equipment performance degradation and must be included in risk management.

[0048] The classification method refers to an artificial intelligence algorithm used to distinguish between normal fluctuations and risk factors, and to classify risk factors. It can combine the characteristics of abnormal deterioration signals (fluctuation amplitude, duration, rate of change) to achieve accurate classification. The number of training iterations of the algorithm is 1000~3000 times, preferably 1500~2500 times. The convergence threshold is a loss function value ≤0.05, which is determined based on the following criteria: ensuring that the algorithm is fully trained and the classification accuracy meets the standard. The processing delay of the algorithm is ≤1s, which is determined based on the following criteria: ensuring rapid output of classification results and adapting to real-time control requirements.

[0049] Normal fluctuations refer to short-term, small-amplitude parameter fluctuations caused by changes in grid load and slight environmental fluctuations during the operation of power equipment. These fluctuations do not affect equipment performance or operational safety. The fluctuation amplitude is ≤5% and the duration is <30s. The values ​​are based on the "Power Equipment Operation Regulations". Fluctuations within this range are permissible fluctuations for normal equipment operation and do not require intervention. The fluctuation frequency is ≤3 times / h to avoid frequent misjudgments.

[0050] Risk factors refer to factors that cause abnormal deterioration of power equipment performance, may cause equipment failure, and affect the safe operation of the power grid. The core factors include equipment insulation aging, mechanical wear, and electrical performance degradation. The corresponding parameters have a fluctuation range of >5% and a duration of ≥30s. The basis for the value selection is that fluctuations within this range will accelerate equipment deterioration. If not intervened in time, it will lead to equipment failure. It needs to be managed in a graded manner. The identification rate of risk factors is ≥98%.

[0051] Risk classification results refer to the results obtained after classifying abnormal degradation signals. They are divided into four categories: normal fluctuation, low risk, medium risk, and high risk. The specific classification standards are: low risk (parameter fluctuation 5%~10%, duration 30s~5min), medium risk (parameter fluctuation 10%~20%, duration 5min~30min), and high risk (parameter fluctuation >20%, duration >30min). The value is based on the power equipment fault risk assessment standard. This classification can accurately characterize the severity of risk and provide a clear basis for the selection of intervention timing. The output delay of the risk classification results is ≤10s.

[0052] In this embodiment, the classification model is an RBF kernel support vector machine (SVM), with the following parameters: the kernel function is the radial basis function (RBF); the penalty coefficient C = 1.2; the kernel parameter gamma = 0.08; the input features include degradation rate, volatility amplitude, time series slope, environmental factors, and load factors; the outputs are normal volatility, general risk, and severe risk; the training set consists of 8000 labeled samples; and the classification threshold is a posterior probability ≥ 0.6 for a risky state. Those skilled in the art can directly reproduce this implementation.

[0053] Step S500: By comparing the risk classification results with historical data throughout the entire life cycle, determine the potential timing of intervention.

[0054] The risk classification results obtained in step S400 (only for low, medium, and high risks) are comprehensively compared with the historical data of the entire life cycle of power equipment (including historical operating data of similar equipment, past operating data of the equipment, and fault record data). The similarity and evolution patterns of current risk factors and historical risk factors are analyzed. Combined with the performance standards of each stage of the equipment's life cycle, the speed and trend of equipment performance degradation are predicted, and potential intervention points that can effectively curb equipment degradation and avoid failures are identified. The time range and core intervention direction of the intervention are clarified, providing a basis for the formulation of subsequent intervention plans. In this step, data comparison and timing determination must meet the following parameter requirements: Coverage of historical data throughout the entire lifecycle ≥ 99%, based on the principle of ensuring comprehensive comparison data and avoiding timing bias caused by missing historical data; Accuracy of data comparison ≥ 97%, based on the principle of ensuring accurate analysis of the correlation between current and historical risks and predicting deterioration trends; Judgment error of potential intervention timing points ≤ 1 hour, based on the principle of ensuring accurate intervention timing and allowing sufficient time for subsequent intervention actions; Number of potential intervention timing points: 1-3, preferably 2, based on the principle that 1 ensures core intervention timing, 3 or less avoids over-intervention, and 2 balances intervention reliability and economy; Time range of potential intervention timing points: 1-24 hours after the risk occurs, preferably 6-12 hours, based on the principle that 1 hour allows for short-term intervention, 24 hours avoids intervention lag, and 6-12 hours balances intervention timeliness and preparation time.

[0055] Full lifecycle historical data refers to all historical data of power equipment from design and manufacturing to the current stage, including factory parameters, installation and commissioning records, operation data, maintenance records, fault records, environmental data, etc. The data storage period is greater than or equal to the equipment's service life (usually 20 to 30 years). The basis for the data is to ensure that the historical data can cover the entire lifecycle of the equipment, provide comprehensive support for data comparison, and ensure that the data integrity is greater than or equal to 99.5% to avoid comparison bias caused by missing data.

[0056] Comparison refers to the process of comparing and analyzing the current risk classification results with historical data throughout the entire life cycle. The core comparison content includes risk factor type, fluctuation characteristics, deterioration speed, intervention effect, etc. The comparison time is ≤15s, preferably ≤10s. The criteria for the values ​​are: to ensure the comparison is completed quickly and to determine the potential intervention timing in a timely manner. The comparison accuracy is ≤1% to ensure the accuracy of the comparison results.

[0057] Potential intervention timing refers to the optimal time to intervene in order to effectively curb equipment performance degradation and prevent failures. It is determined in conjunction with the risk level. For low-risk situations, the intervention timing is 12-24 hours after the risk occurs; for medium-risk situations, it is 6-12 hours; and for high-risk situations, it is 1-6 hours. The basis for this value is that the higher the risk level, the more timely the intervention needs to be to avoid escalation of the risk. The accuracy rate of intervention timing is ≥98% to ensure that the intervention can achieve the expected results.

[0058] In this embodiment, the intervention timing optimization adopts an improved particle swarm optimization (PSO) algorithm, specifically implemented as follows: Optimization variables include monitoring frequency, maintenance threshold, early warning level, and intervention intensity; the fitness function is F = 0.5 × equipment availability + 0.3 × maintenance cost savings + 0.2 × risk reduction rate; constraints are: temperature ≤ 85℃, insulation resistance ≥ 1000MΩ, load ≤ 100% of rated value, vibration ≤ 4.5mm / s; PSO parameters are: particle number 30, iterations 50, inertia weight 0.7, cognitive coefficient 1.5, and social coefficient 1.5. This can be directly reproduced by those skilled in the art.

[0059] Step S600: If the potential intervention timing meets the stability requirements, a prediction method is used to simulate the evolution of operational characteristics to obtain an optimized state tracking scheme.

[0060] The stability of the potential intervention timing points determined in step S500 is assessed. Factors such as the equipment operating status, environmental conditions, and power grid load corresponding to the timing point are analyzed to determine whether the intervention timing point meets the requirements for stable equipment operation (i.e., the intervention action will not affect the normal operation of the equipment or cause power grid fluctuations). If the potential intervention timing point meets the stability requirements, a preset prediction method is used to simulate the evolution trend of the equipment operating characteristics after intervention. Combined with the equipment's full life cycle performance standards, the intervention action, intervention parameters, and tracking frequency are optimized to obtain an optimized state tracking scheme. The monitoring focus and tracking process after intervention are clarified to ensure the stable performance of the equipment after intervention. In this step, stability assessment and solution generation must meet the following parameter requirements: The accuracy of stability assessment must be ≥99%, based on ensuring that the intervention timing meets the requirements for stable equipment operation and avoiding equipment failure or power grid fluctuations caused by intervention actions; the time taken for stability assessment must be ≤10s, preferably ≤6s, based on ensuring rapid assessment and timely advancement of subsequent solution generation; the prediction accuracy of the prediction method must be ≥96%, based on ensuring accurate simulation of the evolution of equipment operating characteristics after intervention, providing a reliable basis for solution optimization; the generation time of the optimized status tracking solution must be ≤20s, preferably ≤15s, based on ensuring rapid solution generation and timely initiation of intervention and tracking processes; the executable rate of the solution must be ≥99%, based on ensuring that the solution can be directly implemented and adapted to equipment operation and maintenance needs; the tracking frequency must be 0.5~1Hz, preferably 0.5Hz, based on ensuring real-time tracking of equipment status after intervention and timely detection of anomalies.

[0061] Stability requirements refer to the stable operating conditions that the equipment must meet at the potential intervention point. Specifically, these include: equipment operating parameter fluctuation ≤3%, stable grid load (fluctuation ≤5%), normal environmental conditions (temperature -20℃~40℃, humidity ≤85%RH), and no other abnormal degradation signals. The basis for these values ​​is: in combination with the safety requirements for power equipment operation, these conditions can ensure that intervention actions will not affect the normal operation of the equipment, will not cause grid fluctuations, and the error of stability judgment is ≤1%.

[0062] Prediction methods refer to methods used to simulate the evolution trend of power equipment operating characteristics after intervention. In this embodiment, LSTM model and ARIMA model are preferred. They can be combined with potential intervention timing, risk classification results, and historical data throughout the entire life cycle to accurately simulate changes in equipment performance. The inference delay of the model is ≤1s. The criteria for the value are: to ensure rapid output of simulation results, to provide timely support for scheme optimization, and the prediction error of the model is ≤4% to ensure the accuracy of the simulation results.

[0063] Operational characteristic evolution refers to the trend of changes in the operating parameters and performance status of power equipment over time after intervention. The core includes parameter recovery speed, performance stabilization time, and degradation rate changes. The parameter recovery speed is ≥5% / min (low risk) and ≥10% / min (medium to high risk). The basis for the value is to ensure that the equipment performance recovers quickly and stably after intervention, and the degradation rate is reduced by ≥30% to ensure the intervention effect. The performance stabilization time is ≥24h. The basis for the value is to ensure that the equipment can operate stably for a long time after intervention.

[0064] The optimized status tracking scheme refers to a complete scheme that includes intervention actions, intervention parameters, tracking frequency, monitoring focus, and anomaly handling measures. Intervention actions include equipment maintenance, parameter adjustment, and component replacement. The adjustment range of intervention parameters is 3% to 15% of the original parameters. The basis for this value is to avoid excessive adjustment that may cause abnormal equipment operation. An adjustment range of 3% to 15% can achieve precise optimization. The tracking frequency is 0.5 to 1 Hz. The monitoring focus is on the parameters corresponding to abnormal degradation signals. The optimization effect of the scheme is ≥80%, ensuring that the equipment performance is significantly improved after intervention.

[0065] The complete implementation logic of dynamic adaptive and cooperative control is as follows: The dynamic tracking frequency calculation method is f=f0+k×deviation value, f0=0.1Hz, k=0.05, the larger the deviation, the higher the frequency, with an upper limit of 1Hz and a lower limit of 0.05Hz; the multi-device cooperative rules are synchronous sampling of devices in the same interval, encrypted sampling of abnormal devices, reduced sampling of healthy devices, and immediate linkage warning and interlocking of fault-risk devices; the stability judgment formula is stability coefficient=1-(standard deviation / mean), stability coefficient ≥0.85 is judged as intervention is possible, <0.85 continues to be observed; the degradation trend judgment algorithm is trend slope k=(end value-starting value) / (time difference), k<-0.3 is accelerated degradation, -0.3≤k≤0 is slow degradation, and k>0 is state recovery.

[0066] S700: Update the initial asset status dataset according to the optimized status tracking scheme, and process dynamic changes in a loop to maintain continuous monitoring.

[0067] Based on the optimized status tracking scheme obtained in step S600, the equipment operation data, environmental condition data, and intervention effect data after intervention are integrated into the initial asset status dataset of step S100. The content of the dataset is updated to ensure that the dataset can reflect the status of the equipment after intervention in real time. The processing flow of steps S200 to S600 is repeated to process the dynamic changes of the equipment in a cyclical manner, continuously monitor the trend of equipment performance degradation, detect abnormal deterioration signals, and optimize the intervention scheme to achieve continuous monitoring of the entire life cycle of power equipment, ensure long-term stable operation of the equipment, and extend the service life of the equipment. In this step, the dataset update and continuous monitoring must meet the following parameter requirements: The update time for the initial asset status dataset should be ≤10s, preferably ≤6s, based on the following criteria: ensuring rapid dataset updates to promptly reflect the status after equipment intervention; update accuracy should be ≥99.5%, based on the following criteria: ensuring the updated data is error-free and complete; the cycle time for loop processing should be 1~10min, preferably 3~5min, based on the following criteria: ensuring timely capture of dynamic changes in equipment to achieve continuous monitoring; the stability of loop processing should be ≥99.9%, based on the following criteria: ensuring uninterrupted long-term continuous monitoring to avoid monitoring failure; the coverage rate of continuous monitoring should be ≥100%, based on the following criteria: ensuring coverage of all stages of the equipment's entire lifecycle and all core parameters, with no monitoring blind spots; monitoring error ≤2%, ensuring the accuracy of monitoring data.

[0068] Updating the initial asset status dataset means supplementing the original initial asset status dataset with the equipment operation data, environmental condition data, intervention effect data, and tracking data after the intervention, replacing outdated data, and ensuring that the dataset can reflect the current status of the equipment in real time. The integrity of the updated data is ≥99.8%, and the criteria for the value are: to ensure that the updated dataset can provide accurate basic data for subsequent cyclic processing, and the update frequency is consistent with the cyclic processing cycle (1~10min).

[0069] Cyclic processing refers to the process of repeating steps S200~S600, including dynamic feature processing, anomaly detection, risk classification, intervention timing determination, and solution generation. It continuously adapts to the dynamic changes in equipment status and optimizes intervention strategies in a timely manner. The maximum continuous running time of cyclic processing is ≥720h (30 days). The basis for this value is to ensure that continuous monitoring can be achieved throughout the entire life cycle of the equipment without frequent manual intervention, and the failure probability of cyclic processing is ≤0.1%.

[0070] Continuous monitoring refers to the uninterrupted monitoring, analysis, and intervention of the status of power equipment at all stages of its entire life cycle, ensuring that the equipment performance is always within a controllable range, timely detection and handling of abnormal deterioration, and prevention of equipment failure. The monitoring response time is ≤30s, and the value is determined based on: ensuring rapid response to changes in equipment status, timely initiation of intervention and optimization processes, and continuous monitoring reliability ≥99.5% to ensure long-term stable operation and meet the requirements for safe operation of the power system.

[0071] In this embodiment, the initial asset status dataset is updated according to the optimized status tracking scheme, and dynamic changes are processed cyclically to maintain continuous monitoring. The spatiotemporal mapping relationship is extracted based on the optimized status tracking scheme to determine the update index; simulated evolution nodes are injected into the initial dataset to obtain an enhanced dataset; dynamic changes are identified, and real-time monitoring instructions are obtained when changes exceed the baseline trajectory; cyclic processing is performed to obtain a continuous monitoring stream, and the dataset is updated to maintain continuous monitoring.

[0072] Furthermore, in the lifecycle-based power equipment status analysis method provided in this embodiment, step S100 includes: Step S110: Obtain real-time operation data and environmental condition data for each stage of the entire life cycle of the power equipment, and use a time series alignment algorithm to synchronize the real-time operation data and environmental condition data to obtain time synchronization data.

[0073] The time series alignment formula is: (1) In formula (1), This is the dynamic time-warped distance, without physical units, and is the final output of the formula. Its value range is... , is defined as the cumulative distance of the minimum regularized path between two sequences. These are power equipment operation data sequences and external environmental condition data sequences, respectively. They are obtained from real-time sensor acquisition, and the data values ​​are constrained to the physical rated range of the equipment. They are the multi-source raw time series data to be aligned in this time series synchronization processing. The optimal path minimization operator is used to traverse all feasible mapping paths and select the globally optimal regular path. The sequence regularization mapping path is derived from the dynamic programming path search results. It is a finite integer sequence that satisfies monotonic constraints and is defined as the unique optimal mapping path connecting the start and end points of two time series sequences. It is a single-point distance metric function with no physical unit. It is based on the definition of Euclidean distance and takes values ​​in the range of non-negative real numbers. It is used to calculate the numerical difference between single time-series sampling points within a path. These are two sets of time-series sampling points paired together on the optimal path. They are derived from the original acquired data and their values ​​are constrained by the rated range of the equipment. The total number of mapping nodes for the optimal path. The control logic of formula (1) is to use the global minimum cumulative regularization distance as the optimization objective, and to search for the optimal monotonic mapping path between two sets of time series through dynamic programming algorithm, so as to achieve flexible alignment of multi-source time series data with different sampling frequencies, different timestamp offsets and unequal lengths; it does not require a one-to-one correspondence of time series points, and can adapt to engineering scenarios where the frequency of operation data and environmental data collection is inconsistent. Formula (1) solves the problem of data misalignment caused by the difference in sampling frequency, timestamp offset and unequal time series length of multi-source sensor data in the full life cycle monitoring of power equipment from the root, and provides a standardized input dataset with a unified time benchmark for the entire process of subsequent data denoising, multi-source feature fusion, degradation trend modeling and anomaly attribution analysis, so as to ensure the time series validity and data accuracy of subsequent full life cycle status analysis.

[0074] For the full lifecycle management of power equipment such as transformers, real-time operational data is collected at each stage from equipment installation and operation to decommissioning. This includes indicators such as voltage, current, and temperature, as well as environmental condition data such as humidity, wind speed, and atmospheric pressure. This data is acquired in real-time through a sensor network, providing a foundation for subsequent analysis. For example, after acquiring the real-time operational data, a time series alignment algorithm is needed for synchronization. Specifically, the time series alignment algorithm is a technique based on dynamic time warping used to address the inconsistency of timestamps from different data sources. In one embodiment, assuming the time series of the real-time operational data is sequence T1, including current values ​​recorded per second, and the environmental data is sequence T2, recording humidity values ​​every two seconds, the time series alignment algorithm calculates the Euclidean distance matrix between the sequences and finds the minimum path to align the sequences. For example, missing time points in sequence T2 are filled in through interpolation, aligning all data points to a unified time axis, thus obtaining time-synchronized data. This alignment process can reduce deviations caused by differences in sampling frequencies, ensuring accurate correspondence of data in the time dimension and providing reliable input for subsequent filtering.

[0075] Step S120: Based on the time synchronization data, use the Kalman filter algorithm to remove noise and obtain clean data.

[0076] The Kalman filter prediction formula is:

[0077] In formulas (2)~(3), for The prior state estimate at time, corresponding to the physical dimensions of the equipment, is derived from the filter time update prediction calculation. The data is constrained by the rated operating range of the equipment and is defined as the system state prediction value before the current sensor observation data is corrected. The system state transition matrix is ​​dimensionless and originates from the dynamic system modeling of power equipment. It is a real number matrix and is defined as the evolution and transmission relationship between adjacent states of the equipment. for The post-hoc state estimate, corresponding to the physical dimensions of the equipment, is derived from the filter correction output of the previous time step and is defined as the optimal corrected state value after the previous round of filtering. The system control input matrix is ​​dimensionless and originates from the system dynamic modeling. It is a real number matrix and is defined as the mapping relationship between the effect of external control commands on the device state. for The system control input quantity, corresponding to the control dimension, originates from the equipment operation control command, is constrained by the system control rated range, and is defined as the external control signal of the equipment. for The prior estimate of the covariance matrix at each time step is dimensionless, derived from the covariance time update operation, and its value range is a positive semi-definite matrix. , is defined as a quantitative index of the uncertainty of the prediction result of the prior state. for The posterior covariance matrix at time step is a positive semi-definite symmetric matrix, representing the accuracy of historical state estimation. The system process noise covariance matrix is ​​dimensionless, derived from the preset inherent parameters of the system, and its value range is a positive semi-definite matrix. , is defined as the intensity of process random noise inherent in the device dynamic evolution model. This is the transpose of the system state transition matrix.

[0078] The Kalman filter update formula is:

[0079] In formulas (4)~(6), for The Kalman gain matrix at time step is dimensionless, derived from the optimal gain solution, and its value ranges from 1 to 2. The interval matrix is ​​defined as the correction weight coefficient of sensor observations, representing the trust ratio between the dynamic equilibrium model predictions and the measured data. The system observation matrix is ​​dimensionless and originates from the sensor observation link modeling. It is a real number matrix and is defined as a linear mapping relationship between the actual state space of the device and the sensor observation data space. The noise covariance matrix is ​​dimensionless and derived from sensor hardware parameter calibration, with values ​​ranging from positive definite matrices. , is defined as the intensity of random noise introduced during the sensor sampling process; for The actual observation value of the time sensor, corresponding to the physical dimensions of the device, comes from real-time collection by the Internet of Things terminal, is constrained by the rated range of the device operation, and is defined as the raw sensor monitoring data containing noise. for The post-hoc state estimate, corresponding to the physical dimensions of the equipment, is derived from the final output of the filter observation correction and is constrained by the rated operating range of the equipment. It is defined as the pure equipment state data output after this filtering and denoising is completed. for The posterior estimate of the covariance matrix at time step is dimensionless, derived from the covariance correction update operation, and its value range is a positive semi-definite matrix. , is defined as a quantitative index of uncertainty in the corrected optimal state estimate; The identity matrix is ​​dimensionless and derived from the definition of matrix operations. It is a diagonal identity matrix of the same dimension and is used for matrix identity operation constraints for covariance update. The control logic of formulas (4) to (6) is a two-step iterative process of prior prediction and posterior correction: First, the optimal state and covariance from the previous moment to the current moment are predicted in advance through the system dynamic model; then, the prior prediction value is optimally corrected by combining the Kalman gain balance prediction model and the actual sensor data with the observation residual, and the global estimation mean square error is recursively minimized to achieve adaptive noise suppression. Formulas (4) to (6) can realize real-time online recursive denoising of power equipment life cycle monitoring data. While effectively eliminating high-frequency random interference from sensors and system process noise, it fully retains the actual operating state changes and deterioration trend characteristics of the equipment, avoids the distortion of state information caused by excessive filtering, and provides high-precision pure input data for the subsequent feature extraction, deterioration attribution, and life prediction processes.

[0080] Based on time-synchronized data, a Kalman filter algorithm is used for noise reduction. Kalman filtering is a recursive estimation method that minimizes estimation error through prediction and update steps. Specifically, in power equipment monitoring, assuming the clean signal is affected by noise, this Kalman filter algorithm first establishes a state-space model, such as using voltage as the state variable. The prediction step uses the equation of motion to estimate the state at the next moment. Then, the update step combines the actual measured value and the Kalman gain to correct the prediction. For example, if the predicted voltage is 220V, the measured value is 218V, and the noise variance is known, the gain is calculated and adjusted to 219V, thus obtaining clean data. This method effectively suppresses random noise and improves data quality.

[0081] Step S130: Based on the clean data, use a linear interpolation algorithm to fill in the missing values, and use a multi-source data fusion algorithm to extract and stitch features to obtain the initial dataset of asset status.

[0082] The linear interpolation formula is: (7) In formula (7), The missing data values ​​to be filled are corresponding to the physical dimensions of the equipment, derived from the results of linear interpolation, and the data is constrained by the rated range of equipment monitoring. They are defined as the interpolated values ​​of the missing points in the time series. , These are the adjacent known valid data values ​​at the low time to the left and the high time to the right of the missing point, respectively. They correspond to the physical dimensions of the equipment, are derived from the filtered and purified collected data, are constrained by the rated monitoring range of the equipment, and serve as the benchmark reference data for interpolation calculations. , , These represent the timestamps corresponding to the missing data, the timestamps of the known points on the left and right, respectively, in seconds (s). They are derived from the original timestamps of the data and their values ​​range from [value range missing]. , which is the timing reference parameter for interpolation operations.

[0083] The formula for multi-source data fusion is: (8) In formula (8), The multi-source fusion result is a comprehensive feature vector, dimensionless, derived from the weighted fusion operation output, and is a real number vector. It is defined as the global feature of the initial dataset of asset state that integrates all dimensional information. For the first The feature weight coefficients of the source data are dimensionless and derived from the feature contribution rate calculation results of the PCA principal component analysis mentioned earlier. Their value range is [range missing]. , is defined as the degree of contribution of the corresponding single-source data to the information of the device's global state; For the first The single-source feature vector extracted independently from the source data is dimensionless and comes from the feature extraction results of data from various dimensions. It is a real number vector and is defined as the original feature information corresponding to the single-source operation and environmental data. The control logic of formula (8) is to first fill in the missing time series breakpoints by linear proportional mapping based on the time interval and numerical difference of adjacent effective time series data, and restore the continuous and complete time series sequence; then, based on the feature contribution rate solved by PCA, the weights of each data source are allocated, and the multi-dimensional single-source features are normalized, weighted, and fused to integrate all effective information to form a unified and complete state feature dataset. Formula (8) effectively fills in the data loss problem caused by packet loss and time series breakpoints in monitoring data acquisition. At the same time, it integrates multi-dimensional heterogeneous information such as operating parameters and environmental conditions, eliminates redundant and invalid information, and completely retains the key features of the equipment's full life cycle status, which greatly improves the data integrity and feature effectiveness of subsequent degradation analysis and status assessment.

[0084] For clean data, a linear interpolation algorithm is used to complete missing values. Specifically, linear interpolation calculates the missing value using adjacent known points. For example, if the temperature values ​​at time points t1 and t3 are 20 degrees and 25 degrees respectively, then the missing value at t2 can be linearly calculated as 22.5 degrees. Subsequently, a multi-source data fusion algorithm is used for feature extraction and concatenation. This algorithm can extract key features based on principal component analysis, such as extracting the mean and variance from voltage and humidity data, and then forming a comprehensive dataset through vector concatenation, thus obtaining the initial dataset for asset status. In one implementation, the above processing ultimately improves the accuracy of power equipment status assessment.

[0085] Preferably, in the lifecycle-based power equipment condition analysis method provided in this embodiment, step S200 includes: Step S210: Extract the operating load from the initial asset status dataset and process it using the sliding window algorithm to obtain the dynamic feature vector.

[0086] The formula for calculating the sliding window features is: (9) In formula (9), The dynamic feature vector output for the sliding window is dimensionless, derived from the window mean aggregation operation, and takes the value of a real number. It is defined as the local dynamic feature after the load data aggregation within the window. The preset sliding window length is dimensionless, derived from the algorithm's fixed parameter settings, and its value range is positive integers. It is defined as the total number of time series data points contained in a single window. , These are the start and end indices of the sliding window, respectively. They are dimensionless, derived from temporal sliding traversal operations, and take values ​​in the range of positive integers. They are used to define the data range covered by a single window, satisfying the window length constraint. ; For the first in the window Each operating load data point corresponds to the physical dimensions of equipment operation. It originates from the initial dataset of asset status and is constrained to the rated range of equipment operation. It is defined as the original time-series load monitoring data. The control logic of formula (9) is to use a fixed-length window to slide backward along the time axis and perform mean aggregation on all load data in each window segment to extract the operating characteristics of the local time-series interval. Through continuous window sliding, the segmented feature decomposition of global time-series data is realized, and the time-series dynamic fluctuation and trend change law of equipment operating load are accurately captured. Formula (9) decomposes the global long-series load data of equipment into a continuous local dynamic feature sequence, weakens the interference of single-point abnormal data, and completely preserves the load evolution trend and dynamic fluctuation law. It provides refined time-series local feature input for subsequent equipment degradation trend modeling and life cycle degradation analysis, and adapts to the modeling requirements of the gradual degradation process of the entire life cycle of equipment.

[0087] For the full lifecycle management of power equipment such as generators, the first step is to extract operating load data, such as power output and speed indicators, from the initial asset status dataset. This data is then processed using a sliding window algorithm to capture dynamic changes over time. Specifically, the sliding window algorithm is a time series analysis technique that works by sliding a fixed-length window across the data sequence to extract local features. For example, with a window size set to 10 time points, a window is applied to the generator's power sequence. The average and standard deviation of each window are calculated by sliding from the starting point, thus forming a dynamic feature vector. This dynamic feature vector reflects the short-term fluctuation patterns of the load and provides input for subsequent decomposition.

[0088] Step S220: The dynamic feature vector is processed by variational mode decomposition algorithm to obtain the mode component sequence.

[0089] The variational mode decomposition constraint formula is: (10) In formula (10), For functional minimization optimization operators, used in the modal component set Center frequency set Find the minimum value of the objective functional over the entire domain. A full-modal ergodic summation operator is used to sum all... The energy terms of each decomposed mode are accumulated globally. For the first The time-series mode components, corresponding to the physical dimensions of the device operation, are derived from the results of variational decomposition operations, take real values, and are defined as the time-series mode component sequence output by the decomposition. For the first The first modal center frequency, in Hz, is derived from the adaptive iterative solution of the algorithm, and its value range is [missing value]. , is defined as the inherent oscillation frequency of the corresponding modal signal; The preset total number of decomposition modes is dimensionless, derived from the algorithm hyperparameter settings, and its value range is positive integer. It is defined as the total number of components in this signal decomposition. is the Dirac unit impulse function, which is dimensionless and originates from the fundamental theory of signal analysis, used for constructing analytical signals; It is the imaginary unit, dimensionless, and satisfies the complex number arithmetic identity. ; This is a temporal convolution operator, dimensionless, derived from linear signal operations, and used for signal analytic envelope convolution processing; It is a circumferential constant and is dimensionless. It is a complex frequency shift basis function, dimensionless; For time partial differential operators, The 2-norm square operation is used to quantize the modal signal energy and construct a global minimization constraint objective. The control logic of formula (10) is based on a global energy constraint variational model, which iteratively optimizes and adaptively decomposes the non-stationary original dynamic signal into... Each component has independent, non-overlapping intrinsic modal components, each corresponding to a unique center frequency. This completely separates the fluctuation information of different frequencies within the signal, accurately removes the fluctuations and environmental interference components during normal equipment operation, and locks the characteristic modes corresponding to the equipment degradation evolution. Formula (10) can adaptively adapt to the non-stationary, noisy operating sequence signals of power equipment. Compared with traditional decomposition algorithms, it can effectively avoid the problems of modal aliasing and endpoint distortion, accurately separate multi-frequency coupled signals, deeply mine the early weak degradation characteristics of equipment, and capture potential degradation signals within the life cycle in advance, providing high-precision modal feature input for subsequent degradation mode recognition and life assessment.

[0090] The obtained dynamic feature vectors are processed using a variational mode decomposition algorithm. This algorithm decomposes the signal into multiple intrinsic mode functions based on the variational principle. The specific process includes constructing a variational problem, iteratively optimizing by minimizing bandwidth and constraints, for example, applying the algorithm to the dynamic vector of generator vibration signals, initializing the center frequency and updating the modes through the alternating direction multiplier method, and gradually separating the mode component sequences of different frequencies. These mode component sequences represent the independent oscillation components of the signal, thereby revealing potential fault modes.

[0091] Step S230: If the amplitude of the modal component sequence exceeds a preset threshold, the evolutionary feature matrix is ​​calculated based on the modal component sequence.

[0092] The formula for the evolutionary characteristic matrix is: (11) In formula (11), The elements of the evolution feature matrix are dimensionless and derived from the modal time delay cross-correlation matrix operation results. They are real number matrices and are defined as decomposing the temporal evolution correlation features between modes to quantify the global state coupling relationship between modes. , This is the modal index number, dimensionless, derived from the VMD decomposition modal number, with a value range of [value missing]. ( (Total number of decomposed modes), defined as the mode number participating in the association calculation; The total length of the modal time series is expressed in seconds (s). It is derived from the time series span of the original monitoring data and takes the value of a positive integer. It is defined as the total duration of the time series data covered by this operation. The preset timing delay step size, in seconds (s), is derived from the algorithm's fixed parameter settings. Its value range is non-negative integers. It is defined as the delay shift of the modal timing signal and is used to explore the intermodal hysteresis evolution law. For the first First mode Time component values; For the first First-order mode delay The subsequent component value. This is a full-time traversal summation operator, used for summing over all time sequences. The cross-correlation multiplication and accumulation are completed at each time point. The control logic of formula (11) is to traverse all mode combinations, and through the cross-correlation summation operation of the mode time series under fixed time delay, quantify the temporal linkage and delayed evolution correlation strength between different deterioration modes, and convert the one-dimensional temporal signal of the mode into a high-dimensional evolution correlation matrix, so as to fully characterize the dynamic transfer law of the multi-modal state of the equipment. Formula (11) accurately quantifies the temporal coupling and evolutionary dependence between each deterioration mode, upgrades the single mode fluctuation information into a global state evolution feature matrix, fully restores the multi-modal deterioration linkage law in the entire life cycle of the equipment, and provides a high-dimensional correlation feature basis for subsequent equipment state trend prediction, remaining life assessment and deterioration mode classification.

[0093] If the amplitude of the modal component sequence exceeds a preset threshold, such as 0.5, then the evolutionary feature matrix is ​​calculated based on the modal component sequence. The process involves constructing the evolutionary feature matrix, for example, by calculating the autocorrelation function of the modal component sequence and arranging it into the form of the evolutionary feature matrix to capture the temporal evolution relationship between the components.

[0094] Step S240: Obtain the state transition probability based on the evolutionary feature matrix.

[0095] The formula for the state transition probability is: (12) In formula (12), For the device from state To state The transition probability, dimensionless, is derived from the result of the transition frequency normalization operation, and its value range is... , is defined as the transition probability between adjacent states in the equipment degradation timeline; Status within the device's historical data To state The number of transfers, dimensionless, derived from the historical statistics of the entire life cycle operation, and its value range is a non-negative integer; The total number of pre-defined degradation states throughout the equipment's entire lifecycle is dimensionless, derived from the hierarchical classification of the state system, and its value ranges from positive integers; the denominator is the current state. The cumulative total frequency of transitions to all downstream states is used for global probability normalization. Current state The total frequency of all transitions is calculated. The control logic of formula (12) is based on the frequency of transitions between states according to historical degradation time series data. The total frequency of all transitions in a single state is normalized to obtain the conditional transition probability between states under the Markov chain, thus fully constructing the Markov model of equipment degradation state evolution. Formula (12) transforms the modal evolution law of equipment into a probabilistic state transition model, accurately quantifies the evolution path, transition probability and degradation trend between each degradation stage of equipment, clearly depicts the evolution law of gradual degradation throughout the entire life cycle of equipment, and provides a probabilistic model basis for subsequent remaining life prediction, degradation trend inference and operation and maintenance early warning.

[0096] The state transition probabilities are obtained from the evolutionary feature matrix, and the probabilities are estimated using a Markov chain model. For example, the rows of the evolutionary feature matrix represent the current state, and the columns represent the next state. The transition frequency is calculated and normalized to obtain the probability value.

[0097] Step S250: Predict the performance evolution trajectory through state transition probabilities to obtain the asset performance degradation trend sequence.

[0098] The formula for the performance evolution trajectory is: (13) In formula (13), for The time-series device performance degradation sequence value corresponds to a quantitative indicator of device health performance. It is derived from time-series probabilistic iterative prediction calculations and its value range is [value range missing]. Defined as the real-time residual health performance of assets that evolves over time; These are the initial baseline performance values ​​of the equipment, corresponding to quantitative indicators of equipment health performance. They are derived from the initial operating state calibration of the equipment, and their value range is [range missing]. , is defined as the upper limit of the original, undamaged health of the equipment; The transition probability matrix of the entire equipment system is dimensionless and is derived from the state transition probability solution of formula (12) above. It is the set of evolution probabilities among all deterioration states. The time-series iterative multiplication operator is derived from Markov time-series recursive operations. The control logic of formula (13) is based on the initial health performance of the equipment. It performs continuous cumulative multiplication iterative operations on the Markov state transition probability at each step along the time axis, deduces the equipment performance degradation process step by step, and restores the continuous degradation evolution trajectory of performance over time throughout the entire life cycle. Formula (13) deeply integrates modal time-series evolution characteristics and Markov probability transition model, avoids the problem of short-term prediction bias of single model, realizes accurate time-series prediction of the long-term gradual degradation trend of equipment, and fully restores the performance degradation law of the entire life cycle of equipment, providing accurate trend basis for subsequent operation and maintenance decisions, remaining life assessment, and fault early warning.

[0099] Predicting performance evolution trajectories by state transition probabilities, such as using Monte Carlo simulations to iteratively generate trajectory sequences from initial states, can yield asset performance degradation trend sequences and improve prediction accuracy.

[0100] Furthermore, in the lifecycle-based power equipment status analysis method provided in this embodiment, step S300 includes: Step S310: Extract parameters that exceed a preset threshold from the asset performance degradation trend sequence and process them using a clustering algorithm to obtain abnormal parameter cluster groups.

[0101] The MESS clustering formula is: (14) In formula (14), This represents the total clustering loss (sum of squared errors within clusters), a dimensionless value derived from the results of iterative clustering optimization calculations, with a range of values ​​ranging from [value missing]. , defined as the total internal data discretization error of all clusters in the entire system; The preset total number of clusters is dimensionless, derived from the algorithm hyperparameter setting, and its value range is positive integer. It is defined as the total number of deterioration anomaly mode groups in this anomaly parameter division. For the first A set of anomalous clusters, dimensionless, derived from the data clustering assignment results, defined as parameter clustering groups belonging to the same anomalous pattern; For the first Cluster center, corresponding to the physical dimension of equipment operation, is derived from clustering iterative update calculation. The data is constrained by the rated range of equipment monitoring and is defined as the mean benchmark center of all samples in the corresponding abnormal cluster. The Euclidean 2-norm squared distance is used to quantify the degree of deviation between a single sample and the cluster center. Summation of all cluster categories is performed to iterate through all clusters. The error terms of each cluster are summed over the entire domain. Summation is performed for samples within a cluster, used to iterate through the th sample. All feature samples within a cluster The control logic of formula (14) aims to minimize the sum of squared errors within the global cluster. By iteratively updating the cluster centers and sample cluster affiliations, it aggregates the anomalous feature parameters from the same source into the same cluster group, realizing unsupervised automatic grouping of anomalous parameters and distinguishing different types of equipment degradation anomaly modes. Formula (14) does not require prior knowledge of manual labeling. It automatically identifies and groups equipment anomalous degradation parameters through unsupervised clustering algorithms, accurately distinguishing equipment anomalous degradation modes with different causes and different evolutionary patterns, and providing a pattern classification basis for subsequent degradation type tracing, fault cause diagnosis, and differentiated operation and maintenance strategy formulation.

[0102] For performance monitoring of power equipment such as transformers, the first step is to extract parameters exceeding a preset threshold from the asset performance degradation trend sequence, for example, a threshold of 0.8. Then, anomalies exceeding this threshold are identified by scanning the temperature and insulation resistance sequences, thus forming a parameter set. For example, clustering algorithms are used to process these parameters. Specifically, K-means clustering is an unsupervised learning method that iteratively optimizes the allocation of data points into K clusters, minimizing the variance within each cluster. In application, the transformer's temperature anomaly parameter set is initialized with K=3 cluster centers. The Euclidean distance from each point to the center is calculated and assigned, and the centers are updated until convergence, resulting in anomaly parameter cluster groups. These anomaly parameter cluster groups reflect the similarity between patterns such as high temperature and overload, thus providing a grouping basis for subsequent analysis.

[0103] Step S320: Collect external environmental data for the abnormal parameter cluster group and fuse it through a time-series correlation algorithm to obtain the external factor influence sequence.

[0104] The formula for temporal correlation fusion is: (15) In formula (15), The time-series Pearson correlation coefficient is dimensionless and derived from the result of time-series correlation calculations. Its value range is [value range missing]. , defined as the degree of temporal linear correlation between abnormal equipment parameters and external environmental data; , These are time series sequences of abnormal equipment parameters and time series sequences of external environment, corresponding to the physical dimensions of equipment operation and environment. They are derived from abnormal clustering results and external data collection, respectively, and the data are all constrained to the corresponding parameter range. , These are the global time-series mean values ​​of the abnormal sequence and the environmental sequence, respectively, corresponding to the physical dimensions. They are derived from the full-time-series statistical mean calculation and are constrained to the corresponding parameter range. They serve as the benchmark centered parameter for correlation calculation. This is a full-time traversal summation operator used to accumulate the deviation term at all time sampling points. The control logic of formula (15) is to accurately quantify the temporal linkage strength between various abnormal degradation parameters of the equipment and different external environmental factors through centralized covariance normalization operation, quantitatively decompose the contribution of each external inducement to the abnormal degradation of the equipment, and integrate environmental correlation information to construct the external factor degradation influence sequence. This formula (15) accurately quantifies the correlation strength of the effects of multi-dimensional external environmental factors on the abnormal degradation of the equipment, realizes the external cause tracing of the equipment degradation inducement, distinguishes between internal aging and external environmental interference caused by abnormalities, and fully locates the external driving inducement of degradation, providing a quantitative basis for subsequent differentiated operation and maintenance protection and degradation root cause analysis.

[0105] External environmental data, such as real-time sensor data on humidity and wind speed, are collected for the clusters of abnormal parameters. These data are then fused using a time-series correlation algorithm. The specific process involves a dynamic time warping algorithm, which aligns two sequences by minimizing path costs. For example, the environmental humidity sequence is aligned with the temperature sequence of the cluster. The cumulative distance matrix is ​​calculated, and the optimal warping path is found. After fusion, the sequence of external factors is obtained, capturing the dynamic impact of the environment on the equipment.

[0106] Step S330: If the matching degree between the external factor influence sequence and the abnormal parameter cluster group is higher than the preset threshold, the Isolation Forest algorithm is used to detect potential abnormal patterns.

[0107] The formula for path length in an isolated forest is: (16) In formula (16), For the first The length of the isolated binary tree path of each sample to be tested is dimensionless and comes from the result of random partitioning of isolated forest. Its value range is positive integer and is defined as the core criterion for algorithm anomaly scoring. for The global baseline average path length corresponding to each sample size is dimensionless, derived from global statistical theory calculations of the sample set, and its value range is positive integers. It is used for outlier score quantification calibration. The total number of samples to be analyzed in this detection is dimensionless, derived from the dataset size statistics, and its value range is positive integer. It is defined as the total number of all deterioration feature samples participating in the anomaly detection. The control logic of formula (16) is to iteratively isolate data samples through multiple random binary split trees. Normal samples require more splits to complete isolation (long path length), while abnormal outlier samples can be quickly split and isolated (short path length). Based on the path length threshold judgment rule, it automatically identifies hidden outlier anomalies in high-dimensional data and captures potential early deterioration patterns that are difficult to find by conventional algorithms. Formula (16) is suitable for high-dimensional, non-stationary industrial monitoring data of power equipment. It does not require prior knowledge of anomaly labels, can quickly detect hidden isolated anomalies and unknown potential deterioration patterns, accurately discover early weak anomaly hazards, make up for the shortcomings of traditional clustering and association analysis in identifying isolated outliers, and improve the equipment full-dimensional anomaly deterioration detection system.

[0108] If the matching degree between the external factor influence sequence and the abnormal parameter cluster group is higher than a preset threshold, such as 0.7, the isolated forest algorithm is used for detection. The isolated forest algorithm is based on random segmentation to construct isolated forests. The abnormal point path is relatively short. For example, multiple trees are constructed for the fused sequence, and the average path length is calculated. If it is lower than the preset threshold, it is identified as a potential abnormal pattern, thereby revealing hidden faults.

[0109] Step S340: Calculate the contribution of external factors based on potential abnormal patterns and determine the source of degradation signals.

[0110] The formula for Shapley's contribution value is: (17) In formula (17), For the first The deterioration contribution of each influencing factor is dimensionless and derived from the weighted summation result of the coalition game. Its value range is... , is defined as the influence weight of the corresponding factor on the abnormal signal of equipment degradation; To exclude target factors Any subset of influencing factors This analysis covers the complete set of all degradation factors. The total number of global influencing factors is dimensionless, derived from the statistics of the entire factor set, and its value range is positive integer. This is the characteristic function of the degradation effect, which is dimensionless and comes from the output of the performance degradation prediction model mentioned above. It is used to quantify the overall degradation effect of the equipment corresponding to any combination of factors. This represents the number of elements contained in the corresponding factor subset, used to solve for the game weight coefficients. This is the mathematical factorial operator. For the expanded subset after adding the target factor, in the original subset Add factors to be analyzed A completely new combination of factors. Marginal contribution value, used to introduce factors The difference between the degradation effect values ​​before and after represents the independent driving contribution of this factor. Summing is performed on subsets excluding the target factor, used to sum the subsets within the full factor set. In the middle, traverse all factors that do not contain the target factor. subset of The control logic of formula (17) is based on the theory of multi-factor alliance game. It traverses all factor combination subsets, quantifies the difference in the overall degradation effect before and after the addition of a single factor, and completes the global weighted average by combining the factorial weight of the subset size. It fairly allocates the independent contribution of each factor to the final degradation result, and realizes unbiased degradation attribution under multiple coupled factors. Formula (17) can solve the problem of degradation source tracing under multi-factor coupled interference, accurately quantifies the independent degradation contribution weight of all factors such as equipment aging, external environmental disturbance, and operating load fluctuation, and clearly locks the dominant causes and secondary influencing factors of degradation, providing accurate source tracing basis for subsequent targeted operation and maintenance intervention, hidden danger management, and equipment protection strategy formulation.

[0111] The contribution of external factors is calculated based on the potential anomaly patterns. The process involves allocating contributions using the Shapley value method, such as additive decomposition of the influence of humidity in the patterns, to identify sources of degradation signals, such as environmental corrosion.

[0112] Step S350: By comparing and analyzing the source of the degradation signal with the asset performance degradation trend sequence, abnormal degradation signals are identified.

[0113] The similarity determination formula is: (18) In formula (18), The temporal similarity coefficient is dimensionless and derived from the results of Pearson correlation alignment of two sequences. Its value range is [value range missing]. , defined as the degree of temporal evolution matching between the deterioration source sequence and the performance degradation sequence; , These are the time series sequence of deterioration source characteristics and the asset performance degradation trend sequence, respectively, corresponding to the physical dimensions of equipment deterioration and health performance. They are derived from the deterioration attribution results and performance prediction output data, respectively, and are all constrained within the corresponding parameter range. , These are the global time-series means of the two sets of sequences to be compared, and is the centralized benchmark parameter for similarity calculation. This function performs a full-time traversal summation, used to complete the global accumulation operation on the deviation term of all sampling points. The normalization constant term quantifies the discrete amplitude of the two time series and is used for correlation coefficient normalization. The control logic of formula (18) quantifies the trend evolution matching degree of the two time series through centralized normalization correlation operation; when the similarity coefficient When the value exceeds the preset threshold, the current source of degradation is verified to have a strong temporal correlation with the performance degradation of the equipment, and is determined to be a real and valid abnormal degradation signal; when the value is below the threshold, it is determined to be a false anomaly caused by environmental noise or single-point disturbance, and is filtered out. This formula (18) forms a closed-loop verification logic of "attribution-prediction-verification" by comparing the degradation source sequence and the performance degradation sequence in a two-way time series, effectively eliminating false anomalies caused by random noise and instantaneous disturbance, greatly improving the accuracy of equipment abnormal degradation signal identification, and ensuring the accuracy of subsequent operation and maintenance decisions.

[0114] By comparing and analyzing the source of the degradation signal with the asset performance degradation trend sequence, for example, by using the Pearson correlation coefficient to calculate the similarity, if the similarity is higher than 0.9, it is judged as an abnormal degradation signal, thereby improving equipment maintenance efficiency.

[0115] Furthermore, in the lifecycle-based power equipment status analysis method provided in this embodiment, step S400 includes: Step S410: Obtain the time-domain and frequency-domain attributes of the abnormally degraded signal and construct a signal feature set.

[0116] For abnormal degradation signals of power equipment such as transformers, we first obtain their time-domain attributes, such as signal amplitude and duration, and frequency-domain attributes, such as dominant frequency components and harmonic distribution. We then construct a signal feature set using these attributes. For example, we convert the time-domain amplitude into a vector form and fuse it with the frequency-domain spectrum to form a multi-dimensional feature vector for subsequent analysis.

[0117] Step S420: Determine the signal deviation value by comparing the signal feature set with the preset benchmark fluctuation feature library.

[0118] The characteristic deviation formula is: (19) In formula (19), The feature deviation value is dimensionless and originates from the Euclidean distance comparison result of high-dimensional vectors. Its value range is [range missing]. , is defined as the degree of global abnormal degradation of real-time abnormal signals compared to normal operating conditions of the equipment; The set of real-time signal features to be detected is dimensionless and comes from the output of the time-frequency feature extraction step mentioned above. It takes the value of a real number vector and is defined as the high-dimensional feature information corresponding to the abnormal signal of the current device. It is a baseline feature library for normal equipment operation. It is dimensionless and comes from the calibration of the health history data of the entire equipment life cycle. The value is a real number vector and is defined as the original baseline fluctuation feature of the equipment under normal operating conditions without degradation. Euclidean 2-norm operation is used to uniformly quantify the spatial distance between two sets of high-dimensional feature vectors, representing the global feature deviation. The control logic of formula (19) is based on the high-dimensional Euclidean distance criterion, calculates the spatial deviation between the real-time signal feature vector and the health benchmark feature vector, and uniformly quantifies the degree of overall signal deviation from normal operating conditions using norm values. By setting a deviation threshold, the signal anomaly level is divided, providing a quantitative threshold judgment basis for subsequent anomaly mode classification and operation and maintenance level decision-making. Formula (19) realizes the unified and standardized measurement of the multi-dimensional signal anomaly degree of equipment, eliminates the measurement difference problem of different types of features and different dimensional parameters, constructs a unified anomaly deviation evaluation system, accurately quantifies the degree of degradation of equipment operation signal deviating from the health benchmark, and provides a reliable quantitative judgment threshold basis for subsequent anomaly classification and differentiated operation and maintenance strategy formulation.

[0119] The signal feature set is compared with a preset benchmark fluctuation feature library, which contains standard time-frequency feature templates under normal operation. The signal deviation value is determined by calculating Euclidean distance or cosine similarity. The specific process involves projecting feature vectors onto the template space in the library and comparing the differences one by one. For example, the frequency domain features of transformer vibration signals are matched with fault-free templates in the benchmark fluctuation feature library. If the cumulative deviation exceeds a set value, the deviation value is quantified, thereby identifying potential anomalies.

[0120] Step S430: If the signal deviates from the preset fluctuation limit, the support vector machine algorithm is used to classify the signal feature set and obtain a preliminary classification label.

[0121] The decision function of the support vector machine is: (20) In formula (20), The final classification label for the signal is dimensionless and comes from the output of the SVM model. It only takes the value ±1 and is defined as the equipment operating condition judgment label. +1 corresponds to the normal operation fluctuation signal and -1 corresponds to the equipment deterioration risk abnormal signal. For the first The Lagrange multipliers corresponding to each training sample are dimensionless, derived from the model's maximum margin optimization training results, and their values ​​range from [value range missing]. , defined as the influence weight of each training sample on the classification boundary; The original class labels for the training samples are dimensionless, derived from historical labeled datasets, and only take values ​​of ±1, corresponding to the preset normal / abnormal attributes of the samples. It is a kernel mapping function, dimensionless, derived from a pre-defined nonlinear kernel operator, used to realize nonlinear mapping from low-dimensional features to high-dimensional linearly separable spaces; The classification hyperplane bias term is dimensionless, derived from the model training results, and takes the value of all real numbers. It is used to calibrate the classification decision boundary in high-dimensional space. This represents the total number of training samples. The symbol discrimination function is used. The control logic of formula (20) is based on the structural risk minimization criterion. It constructs the optimal maximum margin classification hyperplane in the high-dimensional feature space, adapts the nonlinear distribution law of equipment degradation characteristics through the kernel function, accurately divides the boundary between normal working conditions and abnormal degradation working conditions, and outputs the preliminary binary classification discrimination result of the real-time signal. Formula (20) is suitable for small sample and high-dimensional nonlinear degradation feature data in industrial scenarios. It still has high-precision classification performance under limited historical labeled samples, effectively distinguishes between normal random fluctuations of equipment and abnormal signals of real degradation risk, avoids the overfitting problem of conventional classification algorithms, and provides accurate category judgment basis for subsequent final operation and maintenance decisions.

[0122] If the signal deviates from the preset fluctuation limit, such as 0.5, the support vector machine algorithm is used to classify the signal feature set. The support vector machine algorithm separates the data by finding the maximum margin hyperplane. Specifically, during the training phase, historical data is used to build the model. After inputting new signal features, the decision function is calculated to obtain preliminary classification labels such as "potential fault" or "normal", thereby providing a basis for risk assessment.

[0123] Step S440: Map the preliminary classification markers to the preset risk feature space to determine the signal distribution coordinates, and distinguish normal fluctuations from risk factors based on the relative position of the signal distribution coordinates and the risk boundary to obtain the risk classification result.

[0124] The formula for spatial distance is: (twenty one) In formula (21), The distance in the multidimensional risk space is dimensionless and derived from the cumulative calculation of Manhattan distances across all dimensions. Its value range is... It is defined as the degree of global deviation of the real-time signal from the preset risk boundary; For the signal characteristics in the risk space, the first Multidimensional coordinates, dimensionless, are derived from the high-dimensional space mapping results of classification labels, and take the value of all real numbers. They are defined as the multidimensional feature coordinates corresponding to real-time signals. For the risk boundary within space, the first The dimensionless calibration coordinates are derived from the preset calibration of the equipment health threshold system. The values ​​are all real numbers and are defined as the boundary threshold coordinates between normal operating conditions and risky or deteriorating operating conditions. The total dimension of the risk feature space is dimensionless, derived from the dimension statistics of the degradation feature system, and its value range is positive integer. The control logic of formula (21) is based on the Manhattan distance criterion, traversing all risk feature dimensions, accumulating the absolute value of the single-dimensional coordinate deviation, and solving the global spatial deviation of the signal; using the preset distance threshold as the risk division boundary, the working condition classification is completed according to the interval where the deviation distance is located, and the multi-level risk states such as normal operation, minor hidden dangers, and severe degradation are distinguished. This formula (21) realizes the accurate positioning of the multi-dimensional risk space of equipment degradation state, avoids the one-sidedness of single-dimensional indicator judgment, combines the full-dimensional features to comprehensively quantify the risk level, and constructs a complete multi-level risk classification evaluation system, providing the final accurate decision-making basis for the subsequent differentiated operation and maintenance strategy and maintenance intervention plan.

[0125] The preliminary classification labels are mapped to a preset risk feature space, which is a multi-dimensional coordinate system with risk indicators as axes to determine the signal distribution coordinates. By evaluating the distance between the coordinates and the risk boundary, such as using Manhattan distance, if the coordinates exceed the risk boundary, risk factors are distinguished, and risk classification results are obtained, thereby improving the accuracy of equipment diagnosis. Through the above process, accurate classification of abnormal signals is achieved.

[0126] Furthermore, in the lifecycle-based power equipment status analysis method provided in this embodiment, step S500 includes: Step S510: Extract the historical evolution trajectory from the full life cycle historical data based on the risk classification results.

[0127] For the risk classification results of power equipment such as transformers, the historical evolution trajectory is first extracted from its full life cycle historical data. For example, all operating records from the installation of the equipment to the present are collected, including temperature changes, load fluctuations and maintenance events, to form a time series curve that reflects the long-term evolution of the equipment performance, thus laying the foundation for subsequent analysis.

[0128] Step S520: Analyze the historical evolution trajectory to obtain the characteristics of the decline rate.

[0129] The formula for fitting the decay rate is: (twenty two) In formula (22), The rate of equipment performance degradation, expressed in performance index per second, is derived from the results of least squares linear regression fitting and ranges from all real numbers. It is defined as the linear rate of change of equipment health performance over time. The total number of historical evolution trajectory sample points is dimensionless, derived from the statistics of historical data throughout the entire life cycle, and its value range is positive integer. It is defined as the total amount of time series data participating in regression fitting. For the first Each sample corresponds to a historical moment, measured in seconds (s), and originates from raw time-series timestamp data. The value range is [value range missing]. ; For the first The device performance value at each time point, expressed in normalized performance index, is derived from historical degradation sequence data and ranges from [value range missing]. , defined as the quantified value of the remaining health performance of the device at the corresponding time. The control logic of formula (22) is based on the least squares optimal fitting criterion, with time as the quantified value. Independent variable, equipment performance As the dependent variable, the slope of the linear fitting of the performance change over time is solved to accurately quantify the long-term steady-state performance degradation rate of the equipment and fully restore the evolution law of equipment degradation time sequence. This formula (22) realizes accurate quantitative modeling of the equipment degradation rate, removes random fluctuation interference and extracts the core parameters of steady-state degradation trend, and provides core quantitative basis for subsequent extrapolation prediction of equipment remaining service life (RUL), full life cycle operation and maintenance cycle planning, and preventive maintenance strategy formulation.

[0130] The historical evolution trajectory is analyzed to obtain the degradation rate characteristics. Specifically, the trajectory curve is fitted by linear regression, and the slope is calculated as the degradation rate. For example, the least squares method is applied to the aging trajectory of transformer insulation to obtain a rate value such as an insulation resistance index that decreases by 0.2 per year. This involves dividing the trajectory data points into multiple time periods, taking the derivative of each segment and averaging them to reveal the quantitative characteristics of the equipment deterioration rate, and further connecting it to the prediction process.

[0131] Step S530: Perform extrapolation calculations based on the decay rate characteristics to determine the key threshold nodes.

[0132] The extrapolation formula for exponential decay is: (twenty three) In formula (23), For the future The extrapolated predicted performance value at time step, in normalized performance index, is derived from the exponential decay extrapolation calculation and ranges from [value missing]. , defined as the predictive measure of device health performance at future moments; This represents the current baseline performance value of the device, expressed in normalized performance index, derived from real-time sensor monitoring data, with a value range of [value range missing]. , is defined as the initial performance baseline for this extrapolation prediction; The current prediction baseline time is expressed in seconds (s), derived from the system's real-time timestamp, and its value range is [value range missing]. , is defined as the time origin reference for lifetime extrapolation; The parameter represents the equipment performance degradation rate obtained from the previous fitting. Under adverse operating conditions, this parameter is a negative real number, indicating that the driving performance continues to decline over time. It is the natural constant (Euler number). The control logic of formula (23) is based on the irreversible natural degradation law of power equipment insulation aging and mechanical fatigue. It constructs a continuous exponential decay evolution model, takes the current performance and steady-state decay rate as input, extrapolates the future full-time performance decay trajectory, and solves the crossover time nodes corresponding to the performance falling to each level of safety threshold to complete the remaining service life (RUL) prediction. Formula (23) is adapted to the real physical law of irreversible aging of power equipment, abandons the fitting deviation problem of linear decay model, accurately extrapolates the long-term performance evolution trend of equipment, and accurately locks the key threshold nodes of early warning, maintenance, and failure. It provides a forward-looking prediction basis for preventive operation and maintenance, maintenance cycle planning, and spare parts scheduling throughout the entire life cycle.

[0133] The critical threshold nodes are determined by extrapolating the decay rate characteristics. For example, an exponential decay model is used to extrapolate the trajectory. The current rate is input and a threshold is set, such as a point where the insulation resistance is less than 10 megohms. The calculation process includes iteratively calculating the predicted value for future time steps until the threshold is reached, thereby identifying potential fault nodes, such as after 6 months.

[0134] Step S540: If the key threshold node falls within the required range, then a matching process is performed to obtain the intervention priority sequence.

[0135] If the critical threshold node falls within the required range, such as 3 to 12 months, then a matching process is performed to obtain an intervention priority sequence. For example, the node is compared with a preset priority table. The intervention priority sequence includes high-priority emergency repairs and low-priority routine inspections, thereby optimizing resource allocation.

[0136] Step S550: Use the intervention priority sequence to compare the time points of historical data to determine potential intervention opportunities.

[0137] By comparing the historical data time points with the intervention priority sequence, potential intervention opportunities can be determined. For example, the intervention priority sequence can be matched to the historical timeline to find the earliest point corresponding to the high priority, such as the end of the previous quarter, thereby guiding timely maintenance and improving equipment reliability.

[0138] Preferably, in the lifecycle-based power equipment condition analysis method provided in this embodiment, step S600 includes: Step S610: Perform stability verification on a preset time axis based on potential intervention timing points, and obtain the temporal distribution characteristics after the stability verification is passed.

[0139] The stability determination formula is: (twenty four) In formula (24), The stability at the intervention point is expressed as a percentage (%), derived from the result of the relative proportion verification calculation of the time series range, with a value range of [value missing]. , defined as the temporal volatility stability of all potential intervention timing points; This is the complete set of potential intervention points, measured in seconds (s), derived from the extrapolated output data of the S550 threshold node in the previous step, with values ​​ranging from [value range missing]. , defined as the complete set of all potential time nodes for equipment operation and maintenance interventions to be verified; , These are the maximum and minimum time coordinates within the set of opportunity points, used to calculate the relative dispersion ratio of the time series range. The control logic of formula (24) is to calculate the relative dispersion ratio of the time series range of the potential opportunity set and quantify the dispersion of the opportunity distribution; when the calculated stability... When the stability is below the preset stability threshold, the timing of intervention is determined to be reliable and retained; if the stability is above the threshold, it is determined to be a false fluctuation caused by random degradation disturbance and is filtered out. Formula (24) effectively removes false intervention nodes caused by random data fluctuations during lifespan extrapolation, selects time-converged and reliable operation and maintenance intervention opportunities, avoids unreasonable operation and maintenance schemes such as premature maintenance and delayed protection, and ensures that the subsequent generated full life cycle operation and maintenance intervention strategy has engineering feasibility and time-series rationality.

[0140] For potential intervention points for power equipment such as high-voltage cables, stability verification is first performed on a preset time axis. For example, the intervention point is mapped to a one-year time axis, and it is checked whether it remains unchanged within a three-month window. By comparing the fluctuation amplitude between adjacent points, if a certain amplitude is less than a preset threshold such as 5%, it is considered to pass. The time series distribution characteristics, such as uniform distribution or clustered distribution, are obtained, thereby providing a stable basis for subsequent simulations.

[0141] Step S620: Call the prediction model to simulate the evolution of running characteristics based on the time-series distribution characteristics, and obtain the simulated evolution sequence containing future state nodes.

[0142] The LSTM prediction formula is: (25) In formula (25), for The hidden state of the network at time step is dimensionless and derived from the recursive operation result of the LSTM network, with a value range of [value missing]. , is defined as the hidden layer encoding of the device's timing operation characteristics; The Sigmoid nonlinear activation function is dimensionless and serves as a pre-defined nonlinear activation operator for the network, enabling nonlinear transformation of features. , These are the input layer weights and the historical hidden layer propagation weights, respectively. Both are dimensionless real number matrices derived from the model training and optimization results, and are defined as network feature mapping weight parameters. , These are the input bias and the hidden layer bias, respectively. Both are dimensionless real numbers derived from the model training results and are used for offset calibration in network operations. for The time-series feature vector input to the time-series model is dimensionless and comes from the time-series distribution feature data mentioned above, serving as the original input of the model. The control logic of formula (25) relies on the long-term dependency capture capability of the LSTM network, integrates the input features of the current time-series with the hidden state information of the historical time series, and recursively updates the hidden layer features at each time-series after weight mapping, bias calibration and nonlinear activation, fully fitting the time-series evolution law of equipment operation characteristics, and realizing accurate simulation and prediction of future states. Formula (25) relies on the advantages of deep learning time-series modeling, overcomes the defect of traditional models that cannot capture the long-term dynamic coupling dependency of equipment degradation, accurately adapts to the complex nonlinear and time-varying operation evolution law of power equipment, and simulates the future full life cycle state evolution with high precision, providing a time-series prediction basis for the dynamic iteration of subsequent operation and maintenance schemes and the optimization of full-cycle strategies.

[0143] The predictive model is invoked to simulate the evolution of operational characteristics based on the aforementioned temporal distribution features. Specifically, the predictive model can be a time-series prediction framework based on neural networks, such as a long short-term memory network. It simulates future evolution by inputting historical data and distribution features. For example, for the temperature operation characteristics of high-voltage cables, the distribution features are used as input vectors. The predictive model processes sequence dependencies through a gating mechanism to gradually generate future state nodes, such as predicting the temperature peak node of the next quarter, forming a simulated evolution sequence. This involves decomposing features into trend and seasonal components, calculating hidden states layer by layer and outputting the sequence, thereby capturing the dynamic laws of evolution.

[0144] Step S630: By comparing the simulated evolution sequence with the preset benchmark running trajectory, the running status deviation value at each time node is determined.

[0145] The formula for the deviation value is: (26) In formula (26), for The deviation value of the equipment's operating status at any given time, corresponding to the physical dimensions of the equipment's operation, is derived from the result of a point-by-point absolute difference comparison calculation between two time series, and its value range is [missing value]. It is defined as the degree of deterioration of the equipment's operation at each time point relative to the health baseline; for The simulation predicts the operating status at all times, corresponding to the physical dimensions of the equipment operation. It is derived from the time-series evolution simulation output data in step S620 above, constrained by the range of equipment parameters, and defined as the future operating status of the equipment in the model deduction. for The ideal healthy baseline operating state at any given time corresponds to the physical dimensions of equipment operation. It is derived from the calibration database of power equipment industry operation standards and specifications, constrained by the range of equipment parameters, and defined as the ideal operating baseline time-series trajectory of the equipment without degradation. The control logic of formula (26) is to perform node-by-node traversal calculations along the entire life cycle time axis, solve the absolute difference between the simulated state and the baseline state at each moment, remove the interference of the offset direction, uniformly quantify the full-time-series operating deviation amplitude, and form a full-cycle deviation distribution sequence, providing a quantitative basis for subsequent deviation-driven operation and maintenance strategy dynamic tracking and graded intervention decision-making. Formula (26) realizes the refined point-by-point quantification of the equipment's future full-time-series operating deviation, constructs a deviation-driven full-cycle state tracking system, accurately locates the starting time, peak node and evolution range of degradation deviation, realizes refined monitoring of equipment operating status, and provides accurate deviation quantification support for the dynamic iterative optimization of subsequent differentiated operation and maintenance intervention schemes.

[0146] By comparing the simulated evolution sequence with the preset benchmark operating trajectory, the deviation value of the operating status at each time node is determined. For example, the benchmark trajectory is an ideal curve based on standard specifications, such as the cable temperature not exceeding 80 degrees Celsius. The sequence points are matched with the benchmark points one by one, and the Euclidean distance is calculated as the deviation value. For example, the deviation of a certain node is 3 degrees Celsius, which helps to identify abnormal deviations.

[0147] Step S640: Calculate the dynamic tracking frequency based on the operating status deviation value under resource load constraints to obtain a real-time monitoring command with feedback adjustment attributes.

[0148] The formula for dynamic tracking frequency is: (27) In formula (27), To dynamically track the frequency in real time, the unit is Hz (Hertz), derived from the output of the deviation adaptive scaling calculation, with a value range of [value missing]. , is defined as the optimal real-time monitoring sampling frequency of the device at the current moment; The basic baseline monitoring frequency of the system is measured in Hz (Hertz). It is derived from the initial parameter settings of the system and takes the value of a positive real number. It is defined as the default reference sampling frequency under the healthy operating conditions of the equipment. The preset safety deviation threshold, in units of equipment operating physical quantities, is derived from industry safety standard calibration, and its value range is [value range missing]. , is defined as the critical deviation limit for equipment deterioration alarm; For the solution in the previous text The deviation value of the device's operating status at any time is used to drive the adaptive adjustment of the frequency. The control logic of formula (27) uses the deviation of the operation as the feedback adjustment input. The higher the proportion of the deviation relative to the threshold, the more obvious the adaptive increase of the monitoring and tracking frequency. When the device is healthy and without deviation, the frequency is locked as the basic minimum frequency. At the same time, the upper limit of the frequency and the boundary of the system resource load are strictly constrained. When the degradation intensifies, the monitoring sampling is encrypted, and the sampling consumption is reduced under stable conditions, so as to achieve a dynamic balance between accuracy and resource consumption. Formula (27) realizes the adaptive closed-loop adjustment of the monitoring frequency driven by deviation, which solves the industry pain point of "waste of resources during healthy periods and insufficient monitoring during deterioration periods" under the fixed sampling mode. While ensuring high-precision and intensive monitoring during high-risk periods of degradation, it balances the system computing power, communication energy consumption load cost, and achieves the optimal allocation of monitoring resources throughout the entire life cycle.

[0149] The dynamic tracking frequency is calculated based on the deviation value of the operating status under resource load constraints. For example, considering the constraint that the server load does not exceed 70%, if the deviation value is greater than the threshold, the frequency is increased, such as once per hour. The frequency value is adjusted by linear interpolation to obtain real-time monitoring instructions with feedback adjustment attributes, such as automatically reducing the frequency to avoid overload.

[0150] Step S650: The execution sequence is reconstructed using real-time monitoring instructions to obtain an optimized state tracking scheme that conforms to the evolution law of operating characteristics.

[0151] The execution sequence is reconstructed using the real-time monitoring instructions to obtain an optimized state tracking scheme. For example, the real-time monitoring instructions are integrated into the original maintenance sequence and adjusted to prioritize monitoring when the deviation is high, thereby conforming to the evolution law of cable operation characteristics.

[0152] Further, step S700 includes: Step S710: Extract the spatiotemporal mapping relationship based on the optimized state tracking scheme and determine the update index.

[0153] For optimized status tracking schemes for power equipment such as transformers, the spatiotemporal mapping relationship is first extracted. For example, the time series in the status tracking scheme is mapped to the geographical location of the equipment to form a multidimensional matrix, where the time axis represents the monitoring period and the spatial axis marks the equipment coordinates, thereby determining the update index. For example, the key nodes that need to be updated can be identified through matrix operations.

[0154] Step S720: Inject the simulated evolution nodes into the initial asset state dataset according to the updated index to obtain the enhanced dataset.

[0155] The data injection formula is: (28) In formula (28), To enhance the dataset, the fusion and injection results from the original dataset and the simulated prediction nodes are defined as the updated and expanded full-cycle asset status dataset. The initial raw dataset for assets is derived from the output of the raw data access module in step S100 of the process front end and is defined as the set of historical measured basic status data of the equipment. The sequence of future device evolution nodes is derived from the time-series evolution simulation output of step S620 above and is defined as the full-cycle future operating state nodes predicted by the model. The core logic of injecting operators into data fusion is union fusion or concatenation, which seamlessly integrates the predicted nodes into the original dataset. The control logic of formula (28) is based on the set union fusion rule, which integrates the entire domain of the future time series predicted state nodes into the original historical database, expands the time coverage and sample richness of the dataset, and constructs a full-cycle state dataset with dual-source fusion of "historical measurement + future projection". Formula (28) realizes the closed-loop injection of predicted data, opens up the whole-link closed-loop optimization of "state prediction → data expansion → model iteration", continuously enriches the equipment degradation time series sample library, and reversely improves the prediction accuracy, risk discrimination ability and life prediction accuracy of the subsequent whole process, and realizes continuous self-iterative optimization of the entire life cycle of the system.

[0156] Based on the updated index, simulated evolution nodes are injected into the initial asset status dataset to obtain an enhanced dataset. Specifically, the simulated evolution nodes originate from temperature change sequences generated by previously predicted models. For the initial transformer asset status dataset, which includes historical load data and environmental parameters, nodes are mapped and inserted one by one using an injection method such as a data fusion algorithm. For example, assuming the initial dataset has 100 sample points, injecting 20 evolution nodes forms the enhanced dataset. The fusion process involves weighted averaging to calculate new data points to reflect potential future states, thus providing a rich foundation for subsequent identification. This injection mechanism ensures the dynamic expansion of the dataset, especially in high-load scenarios, enabling the capture of precursors to subtle changes.

[0157] Step S730: Perform dynamic change identification on the enhanced dataset. If the dynamic change identification result exceeds the baseline running trajectory, obtain a real-time monitoring instruction.

[0158] Dynamic changes are identified in the enhanced dataset. If the identified dynamic changes exceed the baseline operating trajectory, a real-time monitoring command is generated. For example, dynamic change identification can employ a threshold-based comparison method, aligning the load curve of the enhanced dataset with the baseline operating trajectory. The baseline trajectory is defined as the ideal path where the standard load does not exceed the rated value. If a deviation is identified, such as a sudden 10% increase in load, a command is triggered, which includes adjusting the sampling frequency to achieve timely response. This identification process involves scanning the dataset point by point, calculating the rate of change, and comparing it with a preset threshold. Specifically, the temporal characteristics of the dataset are first extracted, such as using a sliding window to analyze local fluctuations, and then the overall dynamics are quantified. If the deviation exceeds the threshold, a command is generated. Preferably, the command also considers the device type for customization.

[0159] Step S740: Use real-time monitoring commands to process the enhanced dataset in a loop to obtain a continuous monitoring stream, so as to update the initial dataset of asset status and maintain continuous monitoring.

[0160] The enhanced dataset is processed cyclically using the real-time monitoring commands to obtain a continuous monitoring stream, which updates the initial dataset of asset status to maintain continuous monitoring. Specifically, the cyclic processing involves iterative application commands, such as performing a data refresh every cycle, feeding the monitoring stream back to the initial dataset to form a closed loop. For example, in transformer monitoring, commands drive sensors to collect data in real time, process the data, and update the dataset to maintain continuity. Through this method, effective monitoring of power equipment is achieved.

[0161] Please see Figure 2This invention provides a lifecycle-based power equipment condition analysis system for implementing the aforementioned lifecycle-based power equipment condition analysis method. It includes an asset condition initial dataset acquisition module 10, an asset performance degradation trend sequence acquisition module 20, an abnormal deterioration signal judgment module 30, a risk classification result acquisition module 40, a potential intervention timing point determination module 50, an optimized condition tracking scheme acquisition module 60, and a continuous monitoring and maintenance module 70. The asset condition initial dataset acquisition module 10 acquires real-time operating data and environmental condition data from each stage of the power equipment's entire lifecycle through a data acquisition system, and integrates the real-time operating data and environmental condition data using a data fusion method to obtain the asset condition initial dataset. The asset performance degradation trend sequence acquisition module 20 applies dynamic feature processing methods to the asset condition initial dataset. The system analyzes the characteristics of changes in asset performance to obtain an asset performance degradation trend sequence; the abnormal degradation signal judgment module 30 is used to analyze external factors and judge abnormal degradation signals if the parameters in the asset performance degradation trend sequence exceed a preset threshold; the risk classification result acquisition module 40 is used to distinguish normal fluctuations from risk factors based on the abnormal degradation signals and obtain risk classification results; the potential intervention timing point determination module 50 is used to determine potential intervention timing points by comparing the risk classification results with historical data throughout the entire life cycle; the optimized state tracking scheme acquisition module 60 is used to simulate the evolution of operating characteristics using prediction methods if the potential intervention timing points meet the stability requirements and obtain an optimized state tracking scheme; the continuous monitoring and maintenance module 70 is used to update the initial asset state dataset according to the optimized state tracking scheme and cyclically process dynamic changes to maintain continuous monitoring.

[0162] Comparison data for technical effectiveness verification: To verify the technical effectiveness of this invention, comparative tests were conducted using a 10kV transformer, switchgear, and cable online monitoring system. The results are shown in Table 1.

[0163] Experimental conclusion: This invention can significantly improve prediction accuracy, extend equipment life, reduce failure rate and maintenance costs, and the technical effects are real and reproducible.

[0164] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A life-cycle-based power equipment condition analysis method, characterized in that, Includes the following steps: S100. Real-time operating data and environmental condition data are acquired from each stage of the entire life cycle of power equipment through a data acquisition system. The real-time operating data and environmental condition data are integrated using a data fusion method to obtain an initial dataset of asset status. S200. Apply dynamic feature processing method to process the changing features based on the initial asset status dataset to obtain the asset performance degradation trend sequence. S300. If the parameters in the asset performance degradation trend sequence exceed a preset threshold, external factors are analyzed through anomaly detection methods to determine abnormal degradation signals. S400. Based on the abnormal degradation signal, a classification method is used to distinguish between normal fluctuations and risk factors to obtain risk classification results; S500. By comparing the risk classification results with historical data throughout the entire life cycle, potential intervention points are determined. S600. If the potential intervention timing meets the stability requirements, a prediction method is used to simulate the evolution of operational characteristics to obtain an optimized state tracking scheme. S700. Update the initial asset status dataset according to the optimized status tracking scheme, and process dynamic changes in a loop to maintain continuous monitoring.

2. The life-cycle-based power equipment condition analysis method according to claim 1, characterized in that, Step S100 includes: S110. Obtain real-time operation data and environmental condition data for each stage of the entire life cycle of the power equipment, and use a time series alignment algorithm to synchronize the real-time operation data and environmental condition data to obtain time synchronization data. S120. Based on the time synchronization data, a Kalman filter algorithm is used to remove noise and obtain clean data; S130. Based on the clean data, a linear interpolation algorithm is used to fill in the missing values, and a multi-source data fusion algorithm is used to extract and stitch features to obtain the initial dataset of asset status.

3. The life-cycle-based power equipment condition analysis method according to claim 1, characterized in that, Step S200 includes: S210. Extract the operating load from the initial asset status dataset and process it using the sliding window algorithm to obtain the dynamic feature vector; S220. The dynamic feature vector is processed by a variational mode decomposition algorithm to obtain a sequence of modal components. S230. If the amplitude of the modal component sequence exceeds a preset threshold, the evolutionary feature matrix is ​​calculated based on the modal component sequence. S240. Obtain the state transition probability based on the evolutionary feature matrix; S250. Predict the performance evolution trajectory using the state transition probability to obtain an asset performance degradation trend sequence.

4. The life-cycle-based power equipment condition analysis method according to claim 3, characterized in that, Step S300 includes: S310. Extract parameters exceeding a preset threshold from the asset performance degradation trend sequence and process them using a clustering algorithm to obtain abnormal parameter cluster groups. S320. Collect external environmental data for the clustering group of abnormal parameters and fuse them using a time-series correlation algorithm to obtain the sequence of external factors affecting the cluster. S330. If the matching degree between the external factor influence sequence and the abnormal parameter cluster group is higher than a preset threshold, the isolated forest algorithm is used to detect potential abnormal patterns. S340. Calculate the contribution of external factors and determine the source of degradation signal based on the potential abnormal patterns; S350. By comparing and analyzing the source of the degradation signal with the asset performance degradation trend sequence, abnormal degradation signals are identified.

5. The life-cycle-based power equipment condition analysis method according to claim 1, characterized in that, Step S400 includes: S410. Obtain the time-domain and frequency-domain attributes of the abnormally degraded signal and construct a signal feature set; S420. The signal deviation value is determined by comparing the signal feature set with the preset reference fluctuation feature library; S430. If the signal deviates from the preset fluctuation limit, the support vector machine algorithm is used to classify the signal feature set and obtain a preliminary classification label. S440. Map the preliminary classification markers to a preset risk feature space to determine the signal distribution coordinates, and distinguish normal fluctuations from risk factors based on the relative position of the signal distribution coordinates and the risk boundary to obtain the risk classification result.

6. The life-cycle-based power equipment condition analysis method according to claim 5, characterized in that, Step S500 includes: S510. Extract historical evolution trajectories from the full life cycle historical data based on the risk classification results; S520. Analyze the historical evolution trajectory to obtain the decay rate characteristics; S530. Perform extrapolation calculations based on the decay rate characteristics to determine key threshold nodes; S540. If the key threshold node falls within the required range, then a matching process is performed to obtain an intervention priority sequence. S550. The intervention priority sequence is used to compare the time points of the historical data to determine potential intervention opportunities.

7. The life-cycle-based power equipment condition analysis method according to claim 1, characterized in that, Step S600 includes: S610. Perform stability verification on a preset time axis based on potential intervention timing points, and obtain the temporal distribution characteristics after the stability verification is passed. S620. Based on the time-series distribution characteristics, the prediction model is invoked to simulate the evolution of operational characteristics, thereby obtaining a simulated evolution sequence containing future state nodes; S630. By comparing the simulated evolution sequence with the preset benchmark running trajectory, the running state deviation value at each time node is determined. S640. Calculate the dynamic tracking frequency based on the operating status deviation value under resource load constraints to obtain a real-time monitoring command with feedback adjustment attributes. S650. The execution sequence is reconstructed using the real-time monitoring instructions to obtain an optimized state tracking scheme that conforms to the evolution law of operating characteristics.

8. The life-cycle-based power equipment condition analysis method according to claim 7, characterized in that, Step S700 includes: S710. Extract the spatiotemporal mapping relationship based on the optimized state tracking scheme and determine the update index; S720. Inject the simulated evolution node into the initial asset state dataset according to the updated index to obtain the enhanced dataset; S730. Perform dynamic change identification on the enhanced dataset. If the dynamic change identification result exceeds the baseline running trajectory, obtain a real-time monitoring instruction. S740. The enhanced dataset is processed cyclically using the real-time monitoring command to obtain a continuous monitoring stream, so as to update the initial dataset of the asset status and maintain continuous monitoring.

9. The life-cycle-based power equipment condition analysis method according to claim 8, characterized in that, In step S720, the data injection formula is: ； in, To enhance the dataset, For the initial raw dataset of assets, This is a sequence of nodes that simulate the future evolution of the equipment. Inject operators into data fusion.

10. A lifecycle-based power equipment condition analysis system, used to implement the lifecycle-based power equipment condition analysis method as described in any one of claims 1 to 9, characterized in that, include: The initial asset status dataset acquisition module is used to acquire real-time operating data and environmental condition data from each stage of the entire life cycle of power equipment through the data acquisition system, and to integrate the real-time operating data and environmental condition data using a data fusion method to obtain the initial asset status dataset. The asset performance degradation trend sequence acquisition module is used to process the changing features based on the initial dataset of the asset state using a dynamic feature processing method to obtain the asset performance degradation trend sequence. An abnormal degradation signal judgment module is used to analyze external factors and judge abnormal degradation signals by an anomaly detection method if the parameters in the asset performance degradation trend sequence exceed a preset threshold. The risk classification result acquisition module is used to distinguish between normal fluctuations and risk factors based on the abnormal deterioration signal using a classification method, and to obtain the risk classification result. The potential intervention timing determination module is used to determine potential intervention timing by comparing the risk classification results with historical data throughout the entire life cycle; The optimized state tracking scheme acquisition module is used to simulate the evolution of operating characteristics using a prediction method if the potential intervention timing meets the stability requirements, thereby obtaining an optimized state tracking scheme. The continuous monitoring and maintenance module is used to update the initial dataset of asset status according to the optimized status tracking scheme and to process dynamic changes in a loop to maintain continuous monitoring.

Images