Power equipment asset health management and predictive maintenance service system

Abstract

The invention relates to the technical field of power equipment operation and maintenance management, in particular to a power equipment asset health management and predictive maintenance service system which comprises a data acquisition and integration module, a feature engineering module, a health assessment and prediction engine maintenance decision and early warning module and a service interface module. The data acquisition and integration module acquires equipment operation parameters through multiple types of sensors, and associates pre-stored equipment asset information to generate an equipment comprehensive data stream; the feature engineering module cleans and standardizes the equipment comprehensive data stream, and constructs a space-time correlation feature matrix; the health assessment and prediction engine comprises a health state assessment unit and a fault prediction unit, the health state assessment unit outputs a health index HI by using a gradient boosting decision tree, and the fault prediction unit outputs a fault probability and a remaining service life RUL in a future preset time period; and the maintenance decision and early warning module generates a grading early warning signal and a maintenance strategy scheme. The intelligent level of operation and maintenance of power equipment is improved, reliable operation of the equipment is guaranteed, and the operation and maintenance cost is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment operation and maintenance management, in particular to a power equipment asset health management and predictive maintenance service system. BACKGROUND

[0002] As the core component of the power system, the health status and stable operation of the power equipment are directly related to the overall power supply reliability and safety. However, the traditional power equipment operation and maintenance management mode has many limitations and cannot meet the needs of the development of intelligent and efficient power systems.

[0003] From the perspective of data management, the traditional operation and maintenance has a single means of collecting power equipment operation parameters, and cannot fully deploy multiple sensors to capture the multi-dimensional operation characteristics of the equipment. The collected multi-source operation data is often stored in a scattered manner, lacks effective integration mechanism, and not only has non-uniform data format and non-synchronized time stamp, but also is not associated with equipment asset information, resulting in prominent data fragmentation problem. At the same time, there are abnormal values and missing fragments in the original data, and there is a lack of systematic data cleaning and standardization process, so the data quality cannot be guaranteed, and a reliable foundation cannot be provided for subsequent equipment health analysis.

[0004] In terms of health assessment and fault prediction, the traditional mode highly depends on the experience judgment of operation and maintenance personnel, cannot quantitatively evaluate the equipment health level, and cannot accurately identify the early existing hidden faults of the equipment. The fault prediction ability is particularly weak, and the strategies of post-maintenance or regular maintenance are mostly adopted: post-maintenance needs to carry out rush repair after the equipment fails, which is easy to cause power interruption and affect the stable operation of the power system; regular maintenance does not combine with the actual health status of the equipment, which may cause over-maintenance of healthy equipment, resulting in waste of human and spare parts resources, and may not timely maintain the sub-healthy equipment, increasing the risk of failure.

[0005] In summary, the traditional power equipment operation and maintenance management has obvious deficiencies in data integration, health assessment, fault prediction, maintenance decision and information transmission, and an equipment asset health management and predictive maintenance service system is needed, which can realize efficient data integration, accurate health assessment, early fault prediction, scientific maintenance decision and safe information transmission, so as to improve the intelligent level and reliability of power equipment operation and maintenance. SUMMARY

[0006] The purpose of the present application is to provide a power equipment asset health management and predictive maintenance service system to solve the problems raised in the background.

[0007] To achieve the above purpose, the present application provides the following technical scheme: A power equipment asset health management and predictive maintenance service system, comprising: The data acquisition and integration module is used to collect time-series operating parameters in real time through various sensors deployed on power equipment, and associate them with pre-stored equipment asset information to generate a comprehensive equipment data stream; The feature engineering module, connected to the data acquisition and integration module, is used to clean and standardize the comprehensive data stream from the device, and performs the following operations to construct a multi-dimensional feature set: Extract time-domain statistical features and frequency-domain features from operating parameters; Based on the topological relationships of devices, the operating parameters of related devices are aggregated to generate spatial association features; By integrating time-domain statistical features, frequency-domain features, spatial correlation features, and static attribute features from equipment asset information, a unified spatiotemporal correlation feature matrix is ​​constructed. A health assessment and prediction engine, connected to the feature engineering module, includes: The health status assessment unit is used to input the spatiotemporal correlation feature matrix into the health assessment model trained with historical data and output a quantitative equipment health index HI. The health assessment model is a machine learning model based on gradient boosting decision tree. The fault prediction unit is used to input the health index HI and the spatiotemporal correlation feature matrix into the time series prediction model trained by historical fault data, and output the probability of the device failing within a preset time period and the predicted remaining service life RUL. The maintenance decision and early warning module, connected to the health assessment and prediction engine, is used for: Based on the probability of failure and the remaining useful life (RUL), combined with the preset warning threshold, a graded warning signal is automatically generated. The warning signal includes at least the warning level and the predicted time of failure. Based on the warning level, the predicted failure time point, and the health index HI, a maintenance strategy plan is generated that includes maintenance timing and resource type. The service interface module connects to the maintenance decision and early warning module. It is used to encapsulate graded early warning signals and maintenance strategy plans into a standard data format and distribute them to designated user terminals through the API interface.

[0008] As a preferred solution, the data acquisition and integration module generates a comprehensive data stream for the device through the following steps: Collect raw data streams from multiple sources, including vibration sensors, temperature sensors, current sensors, and voltage sensors deployed on power equipment, to form an initial heterogeneous time-series dataset; Multimodal data alignment is performed on heterogeneous time-series datasets to unify the timestamps and sampling frequencies of data streams from various sensors, thereby generating synchronized time-series data. Perform data validity verification on synchronized time series data, identify and remove outliers and missing segments, and generate pre-cleaned steady-state operating parameters; The steady-state operating parameters are matched with the static attributes of equipment model, commissioning date, and historical maintenance records pre-stored in the asset database to generate semi-structured equipment data with association identifiers; Semi-structured equipment data with associated identifiers are injected with equipment topology relationships and asset hierarchy information to construct a comprehensive equipment data stream with a unified data model.

[0009] As a preferred approach, the feature engineering module performs data cleaning and standardization on the device's comprehensive data stream and constructs a multi-dimensional feature set. The process specifically includes: Anomaly detection is performed on the input device integrated data stream, and outlier identification methods based on box plots are used to remove outlier data points that deviate significantly from the normal range; Missing values ​​are processed in the data sequence after removing outlier data points. A time series-based linear interpolation method is used to fill in the missing data segments to form a complete sequence of operating parameters. The complete sequence of operating parameters is standardized by using the Z-score normalization method to convert the operating parameters of different dimensions into a standard normal distribution with zero mean and unit variance, thus generating standardized operating parameters. Extract time-domain statistical features from standardized operating parameters. The time-domain statistical features include at least the mean, variance, skewness, kurtosis, and overthreshold rate of the operating parameters within a specified time window. The standardized operating parameters are transformed from the time domain to the frequency domain. The spectral distribution of the operating parameters is calculated by fast Fourier transform, and frequency domain features are extracted from them. The frequency domain features include at least the amplitude of the main frequency components, the centroid frequency of the spectrum, and the spectral variance. Based on the device topology, we locate related devices that are electrically connected or physically adjacent to the current device, aggregate the standardized operating parameters of the related devices at the same time, and generate spatial correlation features through a weighted average algorithm. The extracted time-domain statistical features, frequency-domain features, and spatial correlation features are vertically concatenated with the static attribute features from equipment asset information. The static attribute features include at least the equipment model code, years of operation, and number of historical maintenance. The vertically concatenated hybrid feature set is subjected to dimensionality consistency checks and feature selection. Redundant features are removed and key features are retained, ultimately constructing a unified spatiotemporal correlation feature matrix.

[0010] As a preferred option, the health status assessment unit is used to quantitatively assess the equipment's health index, and the process includes: Receive the spatiotemporal correlation feature matrix from the feature engineering module; The spatiotemporal correlation feature matrix is ​​input into the pre-trained health assessment model. The health assessment model uses the gradient boosting decision tree algorithm and is trained on historical data to learn the mapping relationship between the health status of the device and the features. The health assessment model uses its built-in multiple decision trees to predict input features tree by tree, and combines the prediction results of each tree through an additive model to obtain a preliminary health score. The initial health score is standardized and scaled to convert it to a preset health index range, generating a quantified device health index HI. The output device health index (HI) serves as the result of the device's health status assessment.

[0011] As a preferred embodiment, the fault prediction unit is used to input the health index HI and the spatiotemporal correlation feature matrix into a time-series prediction model trained with historical fault data, and output the probability of the device failing within a preset time period and the predicted remaining useful life RUL, including the following steps: Receive the device health index HI from the health status assessment unit and the spatiotemporal correlation feature matrix from the feature engineering module; The equipment health index HI and the spatiotemporal correlation feature matrix are aligned and spliced ​​in chronological order to generate an enhanced temporal feature sequence. The enhanced temporal feature sequence is input into a pre-trained temporal prediction model, which is trained based on historical fault data and outputs the fault probability value of the device at each time point in a future preset time period, forming a fault probability sequence. Based on the failure probability sequence, the remaining service life (RUL) of the equipment is calculated by identifying the first time point at which the failure probability value exceeds the preset failure threshold. The probability sequence of output device failure within a preset time period and the predicted remaining useful life (RUL).

[0012] As a preferred solution, the maintenance decision and early warning module automatically generates graded early warning signals based on the failure probability and remaining useful life (RUL), combined with preset early warning thresholds. The process includes the following steps: Receive the failure probability sequence and remaining useful life (RUL) from the failure prediction unit; Based on the fault probability sequence, the maximum fault probability value within a preset time period is extracted, and this maximum fault probability value is compared with multiple preset probability thresholds. The preliminary warning level is determined based on the comparison results. Based on the remaining useful life (RUL), calculate the predicted failure time point starting from the current time point; The initial warning level is correlated with the predicted time of failure, and the warning level is corrected by combining the preset time sensitivity rules to generate the final warning level. Based on the final warning level and the predicted failure time, a graded warning signal is generated, which includes at least the warning level and the predicted failure time.

[0013] As a preferred approach, a maintenance strategy plan is generated based on the warning level, the predicted failure time, and the health index HI, including the following steps: Receive the fault probability sequence and remaining useful life (RUL) from the fault prediction unit, and the health index (HI) from the health status assessment unit, and obtain the warning level and predicted fault time point from the generated graded warning signal. Based on the warning level, it is mapped to a preset maintenance urgency level, where the maintenance urgency level is used to indicate the degree of urgency of maintenance actions; Based on the predicted failure time point and combined with the maintenance urgency level, a maintenance time window is calculated, which defines the effective time range for performing maintenance operations. Based on the Health Index (HI), assess the current health status of the equipment, and determine the required maintenance resource types according to the health status and preset resource mapping rules. The maintenance resource types include at least the skill requirements of maintenance personnel, spare parts types, and tools and equipment. By integrating maintenance urgency levels, maintenance time windows, and maintenance resource types, a maintenance timing suggestion is generated, which specifies a particular maintenance execution time point or time period. Based on recommendations on maintenance resource types and timing, a complete maintenance strategy plan is assembled, which includes at least maintenance timing and resource types.

[0014] As a preferred solution, the service interface module is used to encapsulate tiered early warning signals and maintenance strategy plans into a standard data format and distribute them to designated user terminals via an API interface, including: Receive tiered early warning signals and maintenance strategy plans from the maintenance decision and early warning module; The data format of the graded early warning signals and maintenance strategy schemes is converted and encapsulated into a standard data packet conforming to a predefined JSON schema. This standard data packet includes a message header that identifies the data type, a payload containing detailed information about the early warning signals and maintenance strategies, and metadata describing the data source and timestamp. Perform data integrity verification on the encapsulated standard data packets and verify that the data has not been damaged or tampered with during the encapsulation process by calculating the hash value; The verified standard data packets are distributed to the specified user terminals through the RESTful API interface, which is configured to use the HTTPS protocol for secure transmission and employs a token-based authentication mechanism to ensure access authorization. During the data distribution process, the response status of the API interface is monitored in real time, and an automatic retransmission mechanism is executed in case of transmission failure or timeout until the data is successfully delivered to the user terminal or the maximum number of retries is reached. Record the status log of data distribution, including the sending time, the receiving terminal identifier, and the transmission result, for subsequent auditing and service quality analysis.

[0015] As can be seen from the technical solution provided by the present invention above, the beneficial effects of the power equipment asset health management and predictive maintenance service system provided by the present invention are: By collaborating with the data acquisition and integration module and the feature engineering module, the problems of scattered, inconsistent quality, and single-dimensional data in traditional power equipment operation and maintenance are improved. The data acquisition and integration module deploys multiple types of sensors to collect operating parameters, aligns multimodal data to unify timestamps and sampling frequencies, and removes outliers and missing segments by combining data validity verification. Then, it associates static asset information of the equipment and injects topological relationships and hierarchical information to generate a unified equipment comprehensive data stream, avoiding the chaos of asynchronous time and unrelated attributes of multi-source data. The feature engineering module uses box plot method to identify outliers, linear interpolation to fill missing data, and Z-score normalization to eliminate dimensional differences. It extracts spatial correlation and static attribute features in the time and frequency domains and merges them into a spatiotemporal correlation feature matrix. While removing data redundancy, it achieves multi-dimensional feature coverage and provides high-quality and high-availability data support for subsequent intelligent analysis. The health assessment and prediction engine of this invention replaces the traditional experience-based judgment of equipment status with a scientific model, enabling quantifiable health status and predictable fault risks. The health status assessment unit adopts a gradient boosting decision tree model, inputting the spatiotemporal correlation feature matrix into a model trained with historical data. Through tree-by-tree prediction and addition model combination of multiple decision trees, a preliminary health score is output, which is then standardized into a fixed-range health index HI, intuitively reflecting the current health level of the equipment and avoiding the crudeness of traditional binary judgment. The fault prediction unit concatenates HI with the spatiotemporal correlation feature matrix into an enhanced temporal feature sequence, inputs it into a temporal prediction model trained with historical fault data, outputs a fault probability sequence within a preset time period, and calculates the remaining service life RUL by identifying the first time point when the fault probability exceeds the threshold, allowing maintenance personnel to know in advance the possible time and risk level of the fault, thus avoiding the dilemma of passive repair after a sudden fault. The maintenance decision and early warning module of this invention deeply integrates risk warning with maintenance planning, improving the accuracy and efficiency of operation and maintenance decisions. The hierarchical early warning signal generation unit generates early warning levels by comparing the maximum failure probability threshold with time sensitivity and clearly predicts the failure time point, avoiding the ambiguity of traditional early warnings that only report risks without reporting time, allowing operation and maintenance personnel to quickly focus on high-priority tasks. The maintenance strategy solution generation unit maps maintenance urgency based on early warning levels, calculates maintenance time windows based on predicted failure time points, and then matches resource types according to HI, which avoids the hidden danger of delayed maintenance of high-risk equipment and eliminates the waste of investing high-specification resources in low-risk equipment, realizing precise resource allocation for on-demand operation and maintenance, reducing ineffective operation and maintenance costs, and shortening maintenance preparation time. The service interface module of this invention constructs a standardized, encapsulated, secure, verified, and reliable information output channel, solving the problems of non-standardized, insecure, and easily lost information transmission in traditional operations and maintenance. The module encapsulates tiered early warning signals and maintenance strategy schemes into standard data packets conforming to a predefined JSON schema, ensuring that different user terminals can parse them uniformly. Data integrity is verified through hash value calculation to prevent data corruption or tampering. Then, HTTPS protocol encryption and token-based authentication mechanisms are used to prevent information from being eavesdropped on in the network or illegally obtained by unauthorized terminals. The API interface response status is monitored in real time, and automatic retransmission is performed in case of transmission failure or timeout, and transmission logs are recorded to ensure that maintenance instructions are not lost, delayed, or leaked, ensuring that maintenance personnel can obtain accurate early warning information and maintenance plans in a timely manner, and ensuring that maintenance actions are implemented efficiently. This invention promotes the upgrading of power equipment operation and maintenance from routine maintenance to predictive maintenance through intelligent data analysis, decision-making, and transmission throughout the entire process. By identifying fault risks in advance and formulating maintenance plans, it can avoid power outages caused by sudden equipment failures, significantly improving the reliability of power supply. Accurate health assessments and fault predictions can reduce unnecessary downtime maintenance, extend equipment lifespan, optimize the allocation of resources such as manpower and spare parts, reduce overall operation and maintenance costs, and provide strong support for the long-term safe and economical operation of power equipment assets. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall structure of a power equipment asset health management and predictive maintenance service system according to the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] 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 embodiments.

[0019] like Figure 1 As shown, this embodiment of the invention provides a power equipment asset health management and predictive maintenance service system, including: The data acquisition and integration module is used to collect time-series operating parameters in real time through various sensors deployed on power equipment, and associate them with pre-stored equipment asset information to generate a comprehensive equipment data stream; The feature engineering module, connected to the data acquisition and integration module, is used to clean and standardize the comprehensive data stream from the device, and performs the following operations to construct a multi-dimensional feature set: Extract time-domain statistical features and frequency-domain features from operating parameters; Based on the topological relationships of devices, the operating parameters of related devices are aggregated to generate spatial association features; By integrating time-domain statistical features, frequency-domain features, spatial correlation features, and static attribute features from equipment asset information, a unified spatiotemporal correlation feature matrix is ​​constructed. A health assessment and prediction engine, connected to the feature engineering module, includes: The health status assessment unit is used to input the spatiotemporal correlation feature matrix into the health assessment model trained with historical data and output a quantitative equipment health index HI. The health assessment model is a machine learning model based on gradient boosting decision tree. The fault prediction unit is used to input the health index HI and the spatiotemporal correlation feature matrix into the time series prediction model trained by historical fault data, and output the probability of the device failing within a preset time period and the predicted remaining service life RUL. The maintenance decision and early warning module, connected to the health assessment and prediction engine, is used for: Based on the probability of failure and the remaining useful life (RUL), combined with the preset warning threshold, a graded warning signal is automatically generated. The warning signal includes at least the warning level and the predicted time of failure. Based on the warning level, the predicted failure time point, and the health index HI, a maintenance strategy plan is generated that includes maintenance timing and resource type. The service interface module connects to the maintenance decision and early warning module. It is used to encapsulate graded early warning signals and maintenance strategy plans into a standard data format and distribute them to designated user terminals through the API interface.

[0020] In this embodiment, the data acquisition and integration module generates the device's comprehensive data stream through the following steps: Collect raw data streams from multiple sources, including vibration sensors, temperature sensors, current sensors, and voltage sensors deployed on power equipment, to form an initial heterogeneous time-series dataset; Multimodal data alignment is performed on heterogeneous time-series datasets to unify the timestamps and sampling frequencies of data streams from various sensors, thereby generating synchronized time-series data. Perform data validity verification on synchronized time series data, identify and remove outliers and missing segments, and generate pre-cleaned steady-state operating parameters; The steady-state operating parameters are matched with the static attributes of equipment model, commissioning date, and historical maintenance records pre-stored in the asset database to generate semi-structured equipment data with association identifiers; Inject equipment topology relationships and asset hierarchy information into semi-structured equipment data with associated identifiers to construct a comprehensive equipment data stream with a unified data model; Furthermore, the data acquisition and integration module is the "data cornerstone" of the power equipment asset health management and predictive maintenance service system. It captures the real-time operating status of the equipment through various sensors deployed on the power equipment, while simultaneously associating it with pre-stored equipment asset information. This transforms scattered, heterogeneous data into a unified, usable comprehensive equipment data stream, providing high-quality data input for subsequent feature engineering, health assessment, and maintenance decisions, ensuring reliable data support for the entire system from the source. The core function of the data acquisition and integration module is to realize the entire process of "acquisition-alignment-cleaning-association-integration" of power equipment operating data. Specifically, it first... The system collects real-time equipment operating parameters with time-series characteristics using various sensors such as vibration, temperature, current, and voltage, forming an initial heterogeneous time-series dataset. Then, it unifies the time dimension and sampling frequency through multimodal data alignment, and removes outlier and missing data through data validity verification. Subsequently, the processed steady-state operating parameters are associated with static attributes in the asset database, such as equipment model, commissioning date, and historical maintenance records. Finally, it injects equipment topology relationships and asset hierarchy information, ultimately generating a comprehensive equipment data stream with a unified data model, providing a structurally complete and information-rich data foundation for subsequent modules. The data acquisition and integration module consists of the following sub-modules: Multi-source data acquisition unit: The multi-source data acquisition unit is responsible for acquiring raw operational data from various sensors deployed on the power equipment, serving as the module's data input source. This unit targets the core needs of power equipment operation monitoring by selectively acquiring data from four key sensor types: 1) vibration sensor data, used to capture the vibration state of mechanical components and reflect potential problems such as bearing wear and component loosening; 2) temperature sensor data, used to monitor temperature changes in key parts such as windings and joints, providing early warning of overheating faults; 3) current sensor data, recording current fluctuations during equipment operation and reflecting load changes and circuit anomalies; and 4) voltage sensor data, acquiring voltage parameters across the equipment to monitor power supply stability. Through real-time acquisition of these four types of sensor data, an initial heterogeneous time-series dataset containing different physical quantities and acquisition frequencies is formed, comprehensively covering the core state dimensions of equipment operation. Data alignment and preprocessing unit: The data alignment and preprocessing unit is a key link in improving data quality. It mainly completes two steps: "data synchronization - validity verification" to transform heterogeneous raw data into steady-state operating parameters. Multimodal data alignment: Due to differences in sampling frequencies and timestamp recording methods among different types of sensors, direct use can lead to inconsistent data time dimensions. This unit synchronizes heterogeneous time-series datasets by unifying the timestamp format (e.g., using standard UTC time) and adjusting the sampling frequency (interpolating at the highest sampling frequency or downsampling at the lowest sampling frequency). This ensures that all sensor data correspond at the same time point, generating synchronized time-series data and solving the problem of "data not being synchronized and therefore unable to be jointly analyzed". Data validity verification: Synchronized time series data may contain outliers due to sensor failures or signal interference, or missing data segments due to communication interruptions. This unit identifies and removes outlier data points that exceed reasonable ranges through preset validity rules (such as threshold judgment based on the normal operating parameter range of the equipment), while marking missing data segments. Finally, it generates steady-state operating parameters that have been preliminarily cleaned and have no obvious anomalies or missing data, ensuring that the data has the basic conditions for subsequent correlation and analysis. Asset information association unit: The asset information association unit is responsible for supplementing the operational data with equipment asset context information, realizing the fusion of "dynamic operational data - static asset information". This unit first retrieves the equipment static attributes pre-stored in the asset database, including equipment model (reflecting the equipment hardware configuration and performance parameters), commissioning date (the basis for calculating the equipment's years of operation), and historical maintenance records (reflecting the equipment's past faults and repairs). Then, through a unique equipment identifier (such as equipment number), it matches and associates the processed steady-state operating parameters with the corresponding static attributes, so that each piece of operational data is accompanied by the equipment's basic asset information, generating semi-structured equipment data with association identifiers. This association processing makes the operational data no longer isolated values, but "meaningful data" with asset background, laying the foundation for subsequent health assessment analysis combined with the equipment's own characteristics. Topology and hierarchical information injection unit: The topology and hierarchy information injection unit is responsible for integrating the relationships between devices and the asset management hierarchy to achieve "spatial dimension supplementation" of the data. This unit first obtains the topology relationships of devices in the power system, including electrical connections between devices (such as the connection between transformers and switchgear) and physical proximity relationships (such as devices within the same distribution cabinet). At the same time, it retrieves asset hierarchy information (such as the three-level management hierarchy of plant-workshop-equipment). Then, it injects these topology relationships and hierarchy information into semi-structured device data with association identifiers. Through a unified data format (such as defining fixed fields to store topology codes and hierarchy codes), it constructs a comprehensive device data stream with a unified data model. After injecting topology information, the data can reflect the mutual influence between devices, providing a basis for extracting spatial correlation features in subsequent feature engineering. Injecting hierarchy information facilitates subsequent data statistics and analysis according to management dimensions.

[0021] In this embodiment, the feature engineering module is the "core feature processing" of the power equipment asset health management and predictive maintenance service system. It receives the comprehensive equipment data stream output by the data acquisition and integration module, eliminates data noise and format differences through data cleaning and standardization, and then extracts key features from multiple dimensions including time domain, frequency domain, spatial correlation, and static attributes. Finally, it integrates these features to construct a unified spatiotemporal correlation feature matrix, providing accurate and comprehensive feature input for subsequent health assessment and fault prediction. It serves as a crucial bridge connecting raw data and model analysis. The core function of the feature engineering module is to realize the entire process of "cleaning-standardization-feature extraction-fusion optimization" of the comprehensive equipment data. Specifically, it first performs anomaly detection and missing value repair on the comprehensive equipment data stream. Data points deviating from the normal range are removed, missing segments are filled, and Z-score normalization is used to unify the format of parameters with different dimensions. Then, time-domain statistical features (such as mean and variance) and frequency-domain features (such as the amplitude of major frequency components) are extracted from the standardized data. Spatial correlation features are generated based on equipment topology, combined with static attribute features of equipment asset information (such as equipment model and years of operation). Finally, the multi-dimensional features are vertically concatenated, and redundancy is removed through dimensional consistency checks and feature selection to construct a unified spatiotemporal correlation feature matrix, ensuring that the features accurately reflect the equipment health status and meet the needs of subsequent model analysis. The feature engineering module performs data cleaning and standardization on the comprehensive equipment data stream and constructs a multi-dimensional feature set, specifically including: Data cleaning and standardization unit: Data cleaning and standardization are fundamental steps to ensure feature quality. Through a three-step process of "anomaly removal, missing data repair, and format unification," the comprehensive data stream of the equipment is transformed into clean, standardized operating parameters. Anomaly detection: An outlier identification method based on box plot is adopted. First, the quartiles (Q1, Q3) and interquartile range (IQR=Q3-Q1) of each operating parameter in the integrated data stream of the equipment are calculated. Then, the anomaly judgment range is set (below Q1-1.5IQR or above Q3+1.5IQR). Data points are screened one by one, and outliers that deviate significantly from the range (such as voltage surge data caused by instantaneous interference from sensors) are removed to avoid anomalies affecting the accuracy of subsequent feature extraction. Missing value handling: For missing segments that may exist in the data sequence after removing outliers (such as missing data at some time points due to brief communication interruptions), a time series-based linear interpolation method is adopted. Based on the valid data points before and after the missing segment, the value at the missing position is calculated by linear fitting to fill the missing segment and form a complete and continuous sequence of running parameters, ensuring that there are no data breaks during feature extraction. Data standardization: Considering the differences in the units of different operating parameters (such as temperature in °C, current in A, and voltage in kV), the Z-score normalization method is used to convert the complete operating parameter sequence into a standard normal distribution with zero mean and unit variance. The processed data eliminates the influence of units, allowing different types of parameters to be fused and analyzed in the same dimension, ultimately generating standardized operating parameters. Multi-dimensional feature extraction unit: The multi-dimensional feature extraction unit is the core of the module. It extracts key features that reflect the health status of the equipment from four dimensions: time domain, frequency domain, spatial correlation and static attributes, and comprehensively covers the dynamic and static information of equipment operation. Time-domain statistical feature extraction: Based on standardized operating parameters, within a specified time window (such as 5 minutes or 10 minutes, set according to equipment monitoring needs), calculate statistical indicators reflecting the data distribution and changing trends. These indicators include at least the mean (reflecting the average level of the parameters), variance (reflecting the degree of parameter fluctuation), skewness (reflecting the symmetry of the parameter distribution), kurtosis (reflecting the steepness of the parameter distribution), and threshold exceedance rate (the percentage of times the parameter exceeds the safety threshold). This forms a time-domain statistical feature set, capturing the time-domain dynamic patterns of equipment operation. Frequency domain feature extraction: Standardized operating parameters are transformed from the time domain to the frequency domain. The spectral distribution of the parameters is calculated using Fast Fourier Transform (FFT), and the energy proportion of different frequency components is analyzed. Key frequency domain indicators are extracted from this, including at least the amplitude of the main frequency components (the amplitude corresponding to the highest energy frequency in the spectrum, reflecting the main frequency of equipment vibration, electromagnetic and other characteristics), the centroid frequency of the spectrum (the center frequency of the energy distribution in the spectrum, reflecting the central tendency of the frequency distribution), and the spectral variance (reflecting the dispersion of the frequency distribution). This forms a frequency domain feature set, capturing the hidden frequency domain information of equipment operation (such as frequency shifts caused by wear of mechanical parts). Spatial correlation feature extraction: Based on pre-stored device topology relationships, locate related devices that are electrically connected (e.g., transformers and switchgear connected by lines) or physically adjacent (e.g., devices in the same distribution cabinet) to the current device, and collect standardized operating parameters of the related devices at the same time. Using a weighted average algorithm (with weights set according to the degree of influence between the related devices and the current device, e.g., 0.6 for directly connected devices and 0.4 for indirectly connected devices), calculate the weighted average of the parameters of the related devices at the same time, generating a spatial correlation feature set that reflects the mutual influence between devices (e.g., the impact of overload on the current device from a related device). Static attribute feature extraction: Retrieve equipment asset information pre-stored in the asset database, extract static attribute indicators related to equipment health status, including at least equipment model code (reflecting equipment hardware configuration and performance benchmark), years of operation (reflecting equipment aging degree), and number of historical maintenance (reflecting equipment past failure frequency), forming a static attribute feature set to provide static background information for equipment health assessment; Feature fusion and optimization unit: The feature fusion and optimization unit is responsible for integrating multi-dimensional features into a unified and efficient feature matrix. Through a three-step operation of "splicing-checking-filtering", it improves the effectiveness and practicality of features. Vertical feature concatenation: The time-domain statistical features, frequency-domain features, spatial correlation features and static attribute features generated by the multi-dimensional feature extraction unit are vertically concatenated. That is, the various features at the same time are combined into a feature vector in chronological order, so that each feature vector contains the device's time-domain, frequency-domain, spatial correlation and static attribute information, forming a hybrid feature set. Dimensional consistency check: Perform dimensional consistency check on the mixed feature set to check whether the number of dimensions of each feature vector is consistent (e.g., all vectors contain 10 time-domain features, 5 frequency-domain features, 3 spatial correlation features, and 3 static attribute features, for a total of 21 dimensions). If there are feature vectors with missing or redundant dimensions, ensure that all feature vectors have consistent dimensions by filling in the missing dimensions (e.g., by filling with the average of features from similar devices) or deleting redundant dimensions, so as to avoid errors in subsequent model analysis due to inconsistent dimensions. Feature selection: Feature importance assessment methods (such as analysis of variance and correlation analysis) are used to calculate the correlation between each feature in the mixed feature set and the equipment health status (historical health data). Features with low correlation and redundancy (such as redundant features with a correlation greater than 0.9 with other features) are removed, and key features that contribute highly to equipment health assessment and fault prediction are retained. Finally, a unified spatiotemporal correlation feature matrix is ​​constructed to improve the computational efficiency and analysis accuracy of subsequent models.

[0022] In this embodiment, the health assessment and prediction engine is the "intelligent analysis hub" of the power equipment asset health management and predictive maintenance service system. It receives the spatiotemporal correlation feature matrix output by the feature engineering module, quantifies the current health level of the equipment through the health status assessment unit, and predicts future failure risks and remaining service life of the equipment through the fault prediction unit. This provides accurate health diagnosis results and fault prediction data for the maintenance decision-making and early warning module, making it a key core component in realizing "from condition monitoring to predictive maintenance." The core function of the health assessment and prediction engine is to achieve the dual tasks of "quantitative assessment - future prediction" of the equipment's health status. Specifically, it first... The health status assessment unit inputs the spatiotemporal correlation feature matrix into a gradient boosting decision tree model trained on historical data, outputting a quantitative health index (HI) that intuitively reflects the current health level of the equipment. Then, the fault prediction unit fuses the health index HI with the spatiotemporal correlation feature matrix to form an enhanced time-series feature sequence, which is input into a time-series prediction model trained on historical fault data, outputting a fault probability sequence and predicted remaining useful life (RUL) for a preset future time period. Finally, HI, the fault probability sequence, and RUL are synchronously transmitted to the maintenance decision and early warning module, providing data support for tiered early warning and maintenance strategy formulation. The health assessment and prediction engine includes: Health status assessment unit: The core function of the health status assessment unit is to transform abstract equipment characteristics into quantifiable health indicators, accurately assessing the current health status of the equipment. The specific process includes the following steps: Feature reception and preparation: Receives the spatiotemporal correlation feature matrix from the feature engineering module in real time. This matrix contains the device's temporal statistical features, frequency domain features, spatial correlation features, and static attribute features. Ensures that the feature dimensions match the input dimensions of the pre-trained model. If there is a dimension difference, an anomaly warning is triggered (the dimension consistency must be ensured in advance during the model training phase). Pre-trained model call: Load the health assessment model trained on historical data. This model is built based on the gradient boosting decision tree algorithm. During the model training phase, the mapping relationship between "spatiotemporal correlation features and health status" has been learned through a large amount of historical equipment operation data (including normal state, sub-health state, and pre-failure state data). It can output a preliminary score of the corresponding health level based on the input features. Tree-by-tree prediction and score combination: The health assessment model incorporates multiple independent decision trees. Each decision tree makes predictions based on different feature subsets in the spatiotemporal correlation feature matrix, outputting a health score for a single tree. The model uses an additive model to sum the predicted scores of all decision trees to obtain a preliminary health score for the equipment's health status. This score integrates the judgment results of multiple trees, reducing the prediction bias of a single decision tree. Health Index Standardization: Since the initial health score may vary with the model training data, it needs to be standardized and scaled to convert it to a preset health index range (e.g., 0-100, where 100 represents complete health and 0 represents near failure). The standardization process is achieved through linear mapping to ensure that the health indices of different devices and at different times are comparable, ultimately generating a quantified device health index HI. Output results: The equipment health index HI is used as the health status assessment result and transmitted in real time to the fault prediction unit (for subsequent fault prediction) and the maintenance decision and early warning module (for maintenance strategy formulation). At the same time, HI and the corresponding timestamp are stored in the historical health database to facilitate the tracking of equipment health status change trends. Fault prediction unit: The core function of the fault prediction unit is to predict future fault risks and remaining service life based on the current health status and historical trends of the equipment. The specific process includes the following steps: Multi-source data reception and alignment: The device health index HI from the health status assessment unit is received, and the spatiotemporal correlation feature matrix from the feature engineering module is also received. The HI and the spatiotemporal correlation feature matrix are aligned in chronological order according to the timestamp to ensure that the feature matrix at each time point corresponds to a unique HI value, thus avoiding prediction deviations caused by time misalignment. Enhanced temporal feature construction: The aligned HI is concatenated with the spatiotemporal correlation feature matrix to generate an enhanced temporal feature sequence. The addition of HI can supplement the direct indicators of equipment health status and complement the spatiotemporal correlation features (which reflect the details of equipment operation), so that the feature sequence can more comprehensively reflect the potential change trend of equipment from normal to failure and improve prediction accuracy. Time-series prediction model inference: Load the time-series prediction model trained with historical fault data. During the training phase, the model has learned the correlation between the changes in equipment characteristics over time and the occurrence of faults through a large amount of historical data of "feature sequence - fault occurrence time" of equipment. Input the enhanced time-series feature sequence into the model, and the model outputs the fault probability value of the equipment at each time point in the future within a preset time period (such as the next 7 days or 30 days, set according to operation and maintenance requirements), forming a continuous fault probability sequence. Remaining useful life calculation: Based on the output failure probability sequence, a preset failure threshold (e.g., 80%, representing a high-risk level for equipment failure at that time point) is used. By traversing the failure probability sequence, the time point at which the first failure probability value exceeds the preset failure threshold is identified. This time point is the predicted failure occurrence time point. The remaining useful life (RUL) of the equipment is obtained by subtracting the current time point from the predicted failure occurrence time point (units such as days or hours, set according to the precision of the preset time period). Prediction output: The failure probability sequence (including failure probability at each time point) and the calculated remaining useful life (RUL) within the future preset time period are synchronously transmitted to the maintenance decision and early warning module to provide data support for the generation of graded early warning signals and maintenance timing planning. At the same time, the prediction results, along with the corresponding equipment number and timestamp, are stored in the failure prediction database to facilitate subsequent verification of prediction accuracy.

[0023] In this embodiment, the maintenance decision-making and early warning module is the "operation and maintenance command core" of the power equipment asset health management and predictive maintenance service system. It receives the failure probability, remaining useful life (RUL), and health index (HI) output by the health assessment and prediction engine, automatically generates tiered early warning signals through preset rules, clarifies the equipment failure risk level and predicts the failure time, and formulates strategic plans including maintenance timing and resource types based on the health status. This provides operation and maintenance personnel with precise and executable action instructions, serving as a crucial bridge connecting "health prediction" and "actual maintenance." The core function of the maintenance decision-making and early warning module is to achieve integrated processing of "risk early warning - maintenance planning." Specifically, it first generates tiered early warning signals based on the fault probability sequence and remaining useful life (RUL) output by the fault prediction unit, combined with preset early warning thresholds and time sensitivity rules, clearly marking the equipment's early warning level and predicted fault time point. Then, based on the early warning level, predicted fault time point, and health index (HI) output by the health status assessment unit, it generates a complete maintenance strategy plan by mapping maintenance urgency, calculating maintenance time windows, and matching maintenance resource types, clarifying "when to maintain" and "what resources to use for maintenance." Finally, it transmits the tiered early warning signals and maintenance strategy plan to the service interface module, providing direct basis for operation and maintenance execution. The maintenance decision and early warning module includes: Tiered early warning signal generation unit: The core function of the graded early warning signal generation unit is to transform fault prediction data into intuitive risk warning information, clarifying the equipment fault risk level and time point. The specific process includes the following steps: Predictive data reception: Real-time reception of the failure probability sequence and remaining useful life (RUL) from the failure prediction unit, ensuring that the data includes the failure probability value at each time point within the future preset time period (such as the failure probability of each day in the next 30 days) and the accurate RUL value (such as 20 days). If the data is missing or has an incorrect format, a data verification alarm is triggered, and the system waits for the failure prediction unit to retransmit the data. Maximum failure probability extraction and preliminary level determination: Extract the maximum failure probability value within a preset time period from the failure probability sequence, and compare this value with multiple preset probability thresholds (such as a low-risk threshold of 30%, a medium-risk threshold of 60%, and a high-risk threshold of 80%). For example, a maximum failure probability of 25% is determined as a preliminary warning level of "low risk", 55% is "medium risk", and 85% is "high risk", thus forming a preliminary warning level. Predicted failure time point calculation: Based on the remaining useful life (RUL), the predicted failure time point of the equipment is calculated starting from the current time point; for example, if the current time is May 1st and the RUL is 20 days, then the predicted failure time point is May 21st. This time point will serve as key time information for early warning signals to guide the operation and maintenance time planning. Warning Level Correction: The initial warning level is correlated with the predicted failure time point, and the initial level is corrected according to the preset time sensitivity rules. For example, if the initial level is "medium risk" but the predicted failure time point is only 5 days (short-term risk), the level is corrected to "high risk" according to the time sensitivity rules. If the initial level is "medium risk" but the predicted failure time point is 60 days (long-term risk), the "medium risk" level is maintained, and the final accurate warning level is generated. Early warning signal assembly: Based on the final early warning level and the predicted failure time point, a complete hierarchical early warning signal is assembled; the signal should at least include the early warning level (such as low, medium, and high risk) and the predicted failure time point (such as a specific date or number of days) to ensure that maintenance personnel can quickly grasp the core information of equipment failure risk; Maintenance strategy generation unit: The core function of the maintenance strategy generation unit is to formulate a feasible maintenance execution plan based on early warning information and health status, clarifying the maintenance timing and required resources. The specific process includes the following steps: Multi-source information reception: Receives the fault probability sequence and remaining service life (RUL) from the fault prediction unit, the health index (HI) from the health status assessment unit, and the final warning level and predicted fault time point output by the graded warning signal generation unit, ensuring that all information is complete and consistent in time dimension (e.g., all corresponding to the same monitoring cycle of the same equipment). Maintenance urgency level mapping: Based on the final warning level, it is mapped to a preset maintenance urgency level; for example, the warning level "low risk" corresponds to "low urgency" (can be arranged within the regular operation and maintenance cycle), "medium risk" corresponds to "medium urgency" (a plan needs to be formulated within 1 week), and "high risk" corresponds to "high urgency" (preparation needs to be started within 24 hours). This level directly indicates the urgency of the maintenance action; Maintenance time window calculation: Based on the predicted failure time and the maintenance urgency level, the effective time range for performing maintenance operations (i.e., the maintenance time window) is determined. For example, if the warning level is "high risk" and the predicted failure time is May 21, the maintenance time window is set to May 16-May 20 (5 days before the failure); if the warning level is "medium risk" and the predicted failure time is June 20, the time window is set to June 10-June 18 (10-2 days before the failure), ensuring that maintenance is completed before the failure occurs. Maintenance resource type determination: Assess the current health status of the equipment based on the Health Index (HI), and determine the required maintenance resource type by combining the preset health status and resource mapping rules; for example, when the HI is 85 (good health status) and the warning level is "low risk", the resource type is "routine maintenance personnel + basic testing tools"; when the HI is 40 (poor health status) and the warning level is "high risk", the resource type is "experienced maintenance personnel + professional testing equipment + spare parts (such as bearings, sensors)". The resource type should at least include the maintenance personnel skill requirements, spare parts types, and tools and equipment; Maintenance strategy assembly: Integrate maintenance urgency level, maintenance time window and maintenance resource type to generate specific maintenance timing suggestions (such as "arrange senior maintenance personnel with professional testing equipment and spare bearings for maintenance between May 16 and May 20"). Then, based on the maintenance resource type and timing suggestions, assemble a complete maintenance strategy plan. The plan should at least clearly define the two core contents of maintenance timing and resource type. The module's workflow includes: Initialization phase: Rules and threshold configuration: Preset multiple probability thresholds for graded early warning (e.g., low risk 30%, medium risk 60%, high risk 80%), time sensitivity rules (e.g., short-term risk < 7 days, medium-term risk 7-30 days, long-term risk > 30 days), maintenance urgency level mapping rules (e.g., high risk → high urgency, medium risk → medium urgency, low risk → low urgency), and health status and resource mapping rules (e.g., HI < 50 points → senior personnel + professional equipment + spare parts, HI ≥ 50 points → regular personnel + basic tools), ensuring that all rules meet the actual needs of power equipment operation and maintenance; Interface and database configuration: Configure the communication interface with the health assessment and prediction engine to ensure that fault probability sequences, RUL and HI can be received in real time; configure the output interface with the service interface module to ensure that the graded early warning signals and maintenance strategy schemes can be transmitted completely; establish an early warning and strategy database to store historical early warning signals, maintenance strategy schemes and corresponding equipment numbers and timestamps, so as to facilitate subsequent operation and maintenance traceability and rule optimization. Parameter validation: Perform logical validation on preset thresholds and rules, such as checking whether the probability thresholds are sorted in "low < medium < high" and whether the time intervals of the time sensitivity rules do not overlap, to ensure that the rule logic is consistent and avoid contradictions between subsequent warnings and strategy generation; Data reception and verification stage: The graded early warning signal generation unit receives the fault probability sequence and RUL output by the fault prediction unit, and the maintenance strategy scheme generation unit simultaneously receives the fault probability sequence, RUL, HI, and the early warning level and predicted fault time point from the graded early warning signal generation unit. The received data is checked for integrity and consistency. The fault probability sequence is checked to see if it contains the probability values ​​of all time points in the future preset time period, whether RUL is a valid value (non-negative), and whether HI is within the preset range (e.g., 0-100). If there is a problem with the data, an alarm is triggered, requiring the upstream module to retransmit. If the data is normal, proceed to the next step. The generation stage of graded early warning signals: Extract the maximum failure probability value from the failure probability sequence, compare it with the preset probability threshold, and determine the preliminary warning level (e.g., if the maximum probability is 75% > 60% and < 80%, the preliminary level is medium risk). The failure time point is predicted based on RUL (e.g., if the current date is May 1st, with an RUL of 15 days, the predicted failure time is May 16th). The failure time type is determined by combining the time sensitivity rules (e.g., 15 days is considered a medium-term risk). The initial warning level is revised according to the time sensitivity rule (e.g., if the initial medium risk is plus the medium risk, it is maintained as medium risk; if the initial medium risk is plus the short-term risk, it is revised to high risk), and the final warning level is generated. Assemble the final warning level and the predicted failure time point to form a graded warning signal (e.g., "Warning level: medium risk; predicted failure time point: May 16"). Maintenance strategy solution generation phase: The urgency level is mapped to the final warning level (e.g., medium risk → medium urgency). Based on the predicted failure time and maintenance urgency level, calculate the maintenance time window (e.g., for medium urgency + predicted failure on May 16, the time window is set to May 6-May 14). Based on the HI assessment of the equipment health status (e.g., HI 65 points), the maintenance resource type is determined in combination with the resource mapping rules (e.g., HI ≥ 50 points → routine maintenance personnel + basic testing tools). Integrate the maintenance urgency level (medium urgency), maintenance time window (May 6-May 14), and resource type (regular personnel + basic tools) to generate maintenance timing suggestions ("Arrange regular maintenance personnel to carry out maintenance with basic testing tools within May 6-May 14"), and then assemble them into a complete maintenance strategy plan; Result output and storage stage: Transmit the graded early warning signals and maintenance strategy schemes to the service interface module, ensuring that the data format meets the standard requirements of the service interface (such as compatibility with subsequent JSON schema encapsulation). The tiered early warning signals, maintenance strategy plans, corresponding device numbers, and current timestamps are written into the early warning and strategy database to complete data storage, which facilitates subsequent operation and maintenance review and module rule optimization.

[0024] In this embodiment, the service interface module is used to encapsulate the graded early warning signals and maintenance strategy schemes into a standard data format and distribute them to designated user terminals through the API interface, including: Receive tiered early warning signals and maintenance strategy plans from the maintenance decision and early warning module; The data format of the graded early warning signals and maintenance strategy schemes is converted and encapsulated into a standard data packet conforming to a predefined JSON schema. This standard data packet includes a message header that identifies the data type, a payload containing detailed information about the early warning signals and maintenance strategies, and metadata describing the data source and timestamp. Perform data integrity verification on the encapsulated standard data packets and verify that the data has not been damaged or tampered with during the encapsulation process by calculating the hash value; The verified standard data packets are distributed to the specified user terminals through the RESTful API interface, which is configured to use the HTTPS protocol for secure transmission and employs a token-based authentication mechanism to ensure access authorization. During the data distribution process, the response status of the API interface is monitored in real time, and an automatic retransmission mechanism is executed in case of transmission failure or timeout until the data is successfully delivered to the user terminal or the maximum number of retries is reached. Record the status log of data distribution, including the sending time, the identifier of the receiving terminal, and the transmission result, for subsequent auditing and service quality analysis; Furthermore, the service interface module serves as the "information output bridge" of the power equipment asset health management and predictive maintenance service system. It receives tiered early warning signals and maintenance strategy plans output by the maintenance decision and early warning module. Through standardized encapsulation, security verification, and controllable distribution, it accurately transmits key operation and maintenance information to designated user terminals, ensuring that operation and maintenance personnel can obtain equipment early warnings and maintenance instructions in a timely, secure, and complete manner. It is a crucial link connecting the system's core analytical capabilities with terminal applications. The core function of the service interface module is to realize the entire process of information transmission management, from "information encapsulation to security verification, accurate distribution, and status monitoring." Specifically, it first receives tiered early warning signals (including early warning level and predicted fault time) and maintenance strategy plans (including maintenance timing and resource type) generated by the maintenance decision and early warning module. Then, it converts these two types of information into standard data packets conforming to a predefined JSON schema and verifies data integrity through hash value calculation. Subsequently, it uses a RESTful API interface, employing HTTPS protocol and token authentication mechanism, to distribute the data packets to designated user terminals. Simultaneously, it monitors the transmission status in real time, automatically retransmits in case of failure or timeout, and records transmission logs, ultimately ensuring that operation and maintenance information is delivered to user terminals securely, reliably, and promptly. The service interface module includes: Data receiving and format conversion unit: The data receiving and format conversion unit is the "preprocessing center" for information transmission. It is responsible for converting non-standardized operation and maintenance information into a unified format, laying the foundation for subsequent distribution. The specific process includes the following steps: Operation and maintenance information reception: Real-time reception of tiered early warning signals and maintenance strategy plans from the maintenance decision and early warning module, ensuring that both types of information contain complete core content: The tiered early warning signal should include the warning level and the predicted time of failure; The maintenance strategy plan must cover maintenance timing and resource types (including maintenance personnel skill requirements, spare parts types, tools and equipment). If information is missing, a data retransmission request will be triggered. Standard format encapsulation: The received graded early warning signals and maintenance strategy plans are converted into data formats and encapsulated into standard data packets conforming to a predefined JSON schema. This data packet contains three core parts: first, a message header identifying the data type (e.g., the "data_type" field is labeled "early_warning" or "maintenance_plan"); second, a payload containing detailed information about the early warning signal and maintenance strategy (e.g., in early warning signals, the "warning_level" field records the level, and the "predicted_fault_time" field records the time point; in maintenance strategies, the "maintenance_window" field records the timing, and the "resource_type" field records resource information); and third, metadata describing the data source and timestamp (e.g., the "data_source" field is labeled "maintenance_decision_module," and the "timestamp" field records the encapsulation time), ensuring a consistent data packet structure and complete information. Data integrity verification unit: The data integrity verification unit acts as a "security gatekeeper" for information transmission. It verifies whether data has been damaged or tampered with during the encapsulation process by calculating hash values. The specific process includes the following steps: Hash value calculation: The encapsulated standard data packet is subjected to hash value calculation (the document does not specify a specific hash algorithm, but it is inferred to be a mainstream algorithm such as SHA-256 based on common security practices), generating a unique data packet hash value that can accurately reflect the content characteristics of the data packet; if any modification is made to the content of the data packet (such as the warning level being tampered with or the maintenance timing being adjusted), the hash value will change significantly; Integrity verification: The calculated hash value is bound and stored with the standard data packet. Before subsequent transmission, the matching between the data packet content and the hash value is checked again. If they match, the data is determined to be undamaged or tampered with and is allowed to enter the distribution stage. If they do not match, the data is determined to be abnormal and a re-encapsulation process is triggered until a standard data packet with content matching the hash value is generated, thus preventing abnormal data transmission to the user terminal. API Distribution and Security Control Unit: The API distribution and security control unit acts as the "channel manager" for information transmission, responsible for delivering data packets to designated terminals through secure and controllable interfaces. The specific process includes the following steps: API interface configuration: Load the preset RESTful API interface parameters, specify the interface access address, request method (such as POST method to submit data packets), and data transmission rate limit (to avoid excessive bandwidth consumption by a single terminal), and ensure that the interface meets the access requirements of the user terminal; Secure transmission configuration: Enable HTTPS protocol to encrypt the transmission channel to prevent data packets from being eavesdropped or intercepted during network transmission; at the same time, enable token-based authentication mechanism; user terminals must obtain a unique access token assigned by the system in advance, and carry the token when requesting API interfaces. The system will only allow data packets to be received after verifying the validity of the token, so as to prevent unauthorized terminals from obtaining operation and maintenance information. Targeted distribution execution: Based on the system's preset terminal allocation rules (such as sending maintenance information of a certain type of equipment to the corresponding substation maintenance terminal), standard data packets that have passed integrity verification are distributed to the designated user terminals through the configured RESTful API interface to ensure that the information accurately reaches the target audience; Transmission monitoring and retransmission unit: The transmission monitoring and retransmission unit is the "reliability guarantor" of information transmission. It ensures the successful delivery of data packets through real-time monitoring and automatic retransmission. The specific process includes the following steps: Transmission status monitoring: During the data packet distribution process, the response status of the API interface is monitored in real time. By receiving the "data received" confirmation signal from the user terminal, it is determined whether the data packet has been successfully delivered. If no confirmation signal is received or abnormal responses such as "transmission timeout" or "data corruption" are received, the transmission is determined to have failed. Automatic retransmission execution: For data packets that fail to be transmitted, the automatic retransmission mechanism is activated; the data packets are redistributed according to the preset retransmission interval (e.g., 5 seconds after the first failure, 10 seconds after the second failure, increasing to a maximum interval of 30 seconds), and the number of retransmissions is recorded at the same time; if the number of retransmissions reaches the preset maximum number of retries (e.g., 5 times) and still fails, a manual intervention alarm is triggered to notify the system administrator to troubleshoot the transmission fault (e.g., terminal offline, network interruption). Log recording unit: The log recording unit acts as a "process tracer" for information transmission, providing a basis for subsequent auditing and optimization by recording detailed transmission information. The specific process includes the following steps: Log information collection: Throughout the entire data packet distribution process, key status information is collected, including data transmission time (the timestamp field records the distribution time after encapsulation), receiving terminal identifier (such as terminal number and IP address), transmission result (such as "success", "failure (timeout)", "success after 3 retransmissions"), and data packet hash value (used to trace data integrity). Log storage and archiving: Collected log information is stored in the system log database in chronological order, and the logs are archived at a preset period (e.g., monthly) to ensure that the log data can be queried for a long time; read and write permission control is enabled during the log storage process, allowing only administrators to query or export logs to prevent logs from being tampered with; The module's workflow includes: Initialization phase: Configure standardized templates and parameters: Predefine JSON schema templates corresponding to tiered early warning signals and maintenance strategy schemes, and clarify the field structure and requirements of message headers, payloads, and metadata; configure RESTful API interface parameters (access address, request method, transmission rate limit), HTTPS protocol encryption parameters (certificate configuration, encryption algorithm), and token generation rules (token format, validity period, permission association logic); set automatic retransmission parameters (initial retransmission interval of 5 seconds, maximum retransmission interval of 30 seconds, maximum number of retries of 5), log storage path, and archiving cycle (monthly archiving). Authorized terminals and token distribution: Based on the operation and maintenance management requirements, a unique terminal identifier is assigned to each user terminal (such as substation operation and maintenance terminal, regional operation and maintenance center terminal). A corresponding access token is generated based on the terminal permissions. The token is distributed to the terminal and the correspondence between "terminal identifier-token-permission scope" is recorded to ensure that the terminal can access the API interface through token verification. Database and Interface Testing: Initialize the log database and test the log writing and query functions; simulate the hierarchical early warning signals and maintenance strategy schemes output by the maintenance decision and early warning module, and test the entire process of data reception, format conversion, integrity verification, and API distribution to ensure that there are no abnormalities in each link and that the module can run normally; Data reception and encapsulation stage: The data receiving and format conversion unit receives the hierarchical early warning signals and maintenance strategy schemes output by the maintenance decision and early warning module, checks the integrity of the information (such as whether the early warning signal contains the early warning level and whether the maintenance scheme contains the resource type), and if it is complete, it enters the encapsulation stage; if it is missing, it sends a retransmission request to the maintenance decision module. According to the predefined JSON schema template, the graded early warning signals and maintenance strategy schemes are encapsulated into a standard data packet, and the message header (annotating the data type), payload (filling in detailed early warning and maintenance information), and metadata (annotating the data source and encapsulation time) are filled in to generate a structured JSON data packet; Integrity verification and security configuration phase: The data integrity verification unit calculates a hash value (such as SHA-256) for the standard data packet, binds the hash value to the data packet and stores it, and verifies the matching of the data packet content with the hash value again. If they match, the data packet enters the distribution stage; if they do not match, the data packet is repackaged and verified again. The API distribution and security control unit loads the HTTPS protocol configuration and enables encrypted transmission channels; at the same time, it reads the "terminal identifier-token" correspondence to confirm the validity of the token of the terminal to be distributed, in order to prepare for subsequent identity verification. API distribution and status monitoring phase: According to the preset terminal allocation rules, the designated receiving terminal for the current data packet is determined, and the data packet is sent to the terminal through the RESTful API interface. When the terminal requests the interface, it carries an access token. The system verifies the validity and permission scope of the token. After successful verification, the encrypted standard data packet is transmitted. The transmission monitoring and retransmission unit monitors the interface response status in real time: if a "data received" confirmation signal is received from the terminal, the transmission is considered successful; if no confirmation signal is received or a "timeout" or "data corruption" response is received, the automatic retransmission mechanism is activated, and the data packet is retransmitted at a preset interval, while the number of retransmissions is recorded; if the retransmission fails after the maximum number of retransmissions, a manual intervention alarm is triggered. Log recording and archiving phase: The log recording unit collects information throughout the entire transmission process, including data transmission time, receiving terminal identifier, transmission result (success / failure reason / retransmission count), and data packet hash value, and writes the information into the log database; Historical logs in the log database are classified and archived according to a preset archiving period (e.g., monthly) (by terminal, by date), and expired logs are cleaned up (e.g., logs within 1 year are retained) to ensure the efficient operation of the log database and meet long-term traceability requirements. Results feedback phase: If the user terminal supports feedback, the module receives the "information read" confirmation signal returned by the terminal and records the signal to the log; if the terminal reports "data packet parsing error", the data packet re-encapsulation and distribution process is triggered to ensure that the terminal can correctly parse the operation and maintenance information.

[0025] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A power equipment asset health management and predictive maintenance service system, characterized in that: include: The data acquisition and integration module is used to collect time-series operating parameters in real time through various sensors deployed on power equipment, and associate them with pre-stored equipment asset information to generate a comprehensive equipment data stream; The feature engineering module, connected to the data acquisition and integration module, is used to clean and standardize the comprehensive data stream from the device, and performs the following operations to construct a multi-dimensional feature set: Extract time-domain statistical features and frequency-domain features from operating parameters; Based on the topological relationships of devices, the operating parameters of related devices are aggregated to generate spatial association features; By integrating time-domain statistical features, frequency-domain features, spatial correlation features, and static attribute features from equipment asset information, a unified spatiotemporal correlation feature matrix is ​​constructed. A health assessment and prediction engine, connected to the feature engineering module, includes: The health status assessment unit is used to input the spatiotemporal correlation feature matrix into the health assessment model trained with historical data and output a quantitative equipment health index HI. The health assessment model is a machine learning model based on gradient boosting decision tree. The fault prediction unit is used to input the health index HI and the spatiotemporal correlation feature matrix into the time series prediction model trained by historical fault data, and output the probability of the device failing within a preset time period and the predicted remaining service life RUL. The maintenance decision and early warning module, connected to the health assessment and prediction engine, is used for: Based on the probability of failure and the remaining useful life (RUL), combined with the preset warning threshold, a graded warning signal is automatically generated. The warning signal includes at least the warning level and the predicted time of failure. Based on the warning level, the predicted failure time point, and the health index HI, a maintenance strategy plan is generated that includes maintenance timing and resource type. The service interface module connects to the maintenance decision and early warning module. It is used to encapsulate graded early warning signals and maintenance strategy plans into a standard data format and distribute them to designated user terminals through the API interface.

2. The power equipment asset health management and predictive maintenance service system according to claim 1, characterized in that: The data acquisition and integration module generates a comprehensive data stream for the device through the following steps: Collect raw data streams from multiple sources, including vibration sensors, temperature sensors, current sensors, and voltage sensors deployed on power equipment, to form an initial heterogeneous time-series dataset; Multimodal data alignment is performed on heterogeneous time-series datasets to unify the timestamps and sampling frequencies of data streams from various sensors, thereby generating synchronized time-series data. Perform data validity verification on synchronized time series data, identify and remove outliers and missing segments, and generate pre-cleaned steady-state operating parameters; The steady-state operating parameters are matched with the static attributes of equipment model, commissioning date, and historical maintenance records pre-stored in the asset database to generate semi-structured equipment data with association identifiers; Semi-structured equipment data with associated identifiers are injected with equipment topology relationships and asset hierarchy information to construct a comprehensive equipment data stream with a unified data model.

3. The power equipment asset health management and predictive maintenance service system according to claim 1, characterized in that: The feature engineering module performs data cleaning and standardization on the device's comprehensive data stream and constructs a multi-dimensional feature set. The process specifically includes: Anomaly detection is performed on the input device integrated data stream, and outlier identification methods based on box plots are used to remove outlier data points that deviate significantly from the normal range; Missing values ​​are processed in the data sequence after removing outlier data points. A time series-based linear interpolation method is used to fill in the missing data segments to form a complete sequence of operating parameters. The complete sequence of operating parameters is standardized by using the Z-score normalization method to convert the operating parameters of different dimensions into a standard normal distribution with zero mean and unit variance, thereby generating standardized operating parameters. Extract time-domain statistical features from standardized operating parameters. The time-domain statistical features include at least the mean, variance, skewness, kurtosis, and overthreshold rate of the operating parameters within a specified time window. The standardized operating parameters are transformed from the time domain to the frequency domain. The spectral distribution of the operating parameters is calculated by fast Fourier transform, and frequency domain features are extracted from them. The frequency domain features include at least the amplitude of the main frequency components, the centroid frequency of the spectrum, and the spectral variance. Based on the device topology, we locate related devices that are electrically connected or physically adjacent to the current device, aggregate the standardized operating parameters of the related devices at the same time, and generate spatial correlation features through a weighted average algorithm. The extracted time-domain statistical features, frequency-domain features, and spatial correlation features are vertically concatenated with the static attribute features from equipment asset information. The static attribute features include at least the equipment model code, years of operation, and number of historical maintenance. The vertically concatenated hybrid feature set is subjected to dimensionality consistency checks and feature selection. Redundant features are removed and key features are retained, ultimately constructing a unified spatiotemporal correlation feature matrix.

4. The power equipment asset health management and predictive maintenance service system according to claim 1, characterized in that: The health status assessment unit is used to quantitatively assess the equipment health index, and its process includes: Receive the spatiotemporal correlation feature matrix from the feature engineering module; The spatiotemporal correlation feature matrix is ​​input into the pre-trained health assessment model. The health assessment model uses the gradient boosting decision tree algorithm and is trained on historical data to learn the mapping relationship between the health status of the device and the features. The health assessment model makes predictions on the input features tree by tree through its built-in multiple decision trees, and obtains a preliminary health score by combining the prediction results of each tree through an additive model. The initial health score is standardized and scaled to convert it to a preset health index range, generating a quantified device health index HI. The output device health index (HI) serves as the result of the device's health status assessment.

5. The power equipment asset health management and predictive maintenance service system according to claim 1, characterized in that: The fault prediction unit is used to input the health index HI and the spatiotemporal correlation feature matrix into a time-series prediction model trained with historical fault data, and output the probability of the device failing within a preset time period and the predicted remaining useful life RUL, including the following steps: Receive the device health index HI from the health status assessment unit and the spatiotemporal correlation feature matrix from the feature engineering module; The equipment health index HI and the spatiotemporal correlation feature matrix are aligned and spliced ​​in chronological order to generate an enhanced temporal feature sequence. The enhanced temporal feature sequence is input into a pre-trained temporal prediction model, which is trained based on historical fault data and outputs the fault probability value of the device at each time point in a future preset time period, forming a fault probability sequence. Based on the failure probability sequence, the remaining service life (RUL) of the equipment is calculated by identifying the first time point when the failure probability value exceeds the preset failure threshold. The probability sequence of output device failure within a preset time period and the predicted remaining useful life (RUL).

6. The power equipment asset health management and predictive maintenance service system according to claim 1, characterized in that: The maintenance decision and early warning module automatically generates graded early warning signals based on the failure probability and remaining useful life (RUL), combined with preset early warning thresholds. The process includes the following steps: Receive the failure probability sequence and remaining useful life (RUL) from the failure prediction unit; Based on the fault probability sequence, the maximum fault probability value within a preset time period is extracted, and this maximum fault probability value is compared with multiple preset probability thresholds. The preliminary warning level is determined based on the comparison results. Based on the remaining useful life (RUL), calculate the predicted failure time point starting from the current time point; The initial warning level is correlated with the predicted time of failure, and the warning level is corrected by combining the preset time sensitivity rules to generate the final warning level. Based on the final warning level and the predicted failure time, a graded warning signal is generated, which includes at least the warning level and the predicted failure time.

7. The power equipment asset health management and predictive maintenance service system according to claim 1, characterized in that: Based on the warning level, predicted failure time, and health index HI, a maintenance strategy plan is generated that includes maintenance timing and resource type, including the following steps: Receive the fault probability sequence and remaining useful life (RUL) from the fault prediction unit, and the health index (HI) from the health status assessment unit, and obtain the warning level and predicted fault time point from the generated graded warning signal. Based on the warning level, it is mapped to a preset maintenance urgency level, where the maintenance urgency level is used to indicate the degree of urgency of maintenance actions; Based on the predicted failure time point and combined with the maintenance urgency level, a maintenance time window is calculated, which defines the effective time range for performing maintenance operations. Based on the Health Index (HI), assess the current health status of the equipment, and determine the required maintenance resource types according to the health status and preset resource mapping rules. The maintenance resource types include at least the skill requirements of maintenance personnel, spare parts types, and tools and equipment. By integrating maintenance urgency levels, maintenance time windows, and maintenance resource types, a maintenance timing suggestion is generated, which specifies a particular maintenance execution time point or time period. Based on recommendations on maintenance resource types and timing, a complete maintenance strategy plan is assembled, which includes at least maintenance timing and resource types.

8. The power equipment asset health management and predictive maintenance service system according to claim 1, characterized in that: The service interface module is used to encapsulate tiered early warning signals and maintenance strategy schemes into a standard data format and distribute them to designated user terminals via an API interface, including: Receive tiered early warning signals and maintenance strategy plans from the maintenance decision and early warning module; The data format of the graded early warning signals and maintenance strategy schemes is converted and encapsulated into a standard data packet conforming to a predefined JSON schema. This standard data packet includes a message header that identifies the data type, a payload containing detailed information about the early warning signals and maintenance strategies, and metadata describing the data source and timestamp. Perform data integrity verification on the encapsulated standard data packets, and verify that the data has not been damaged or tampered with during the encapsulation process by calculating the hash value; The verified standard data packets are distributed to the specified user terminals through the RESTful API interface, which is configured to use the HTTPS protocol for secure transmission and employs a token-based authentication mechanism to ensure access authorization. During the data distribution process, the response status of the API interface is monitored in real time, and an automatic retransmission mechanism is executed in case of transmission failure or timeout until the data is successfully delivered to the user terminal or the maximum number of retries is reached. Record the status log of data distribution, including the sending time, the identifier of the receiving terminal, and the transmission result, for subsequent auditing and service quality analysis.

Images