Power Equipment Health Assessment and Fault Prediction System

Abstract

This invention discloses a power equipment health assessment and fault prediction system, relating to the field of power equipment monitoring and maintenance technology. The system includes: a data acquisition module, a data processing and storage module, a short-term fault prediction model, a medium-term fault prediction model, a long-term fault prediction model, a model fusion and decision-making module, and a user interface. By integrating short-term, medium-term, and long-term fault prediction models, this invention achieves a comprehensive assessment of the health status of power equipment and fault prediction. The short-term model can quickly capture abnormal changes in the current operating status of the equipment and provide immediate fault warnings. The medium-term model, through a combination of time series analysis and mechanistic models, predicts the performance change trend of the equipment over a future period, providing decision support for planned maintenance work. The long-term model focuses on predicting the remaining lifespan of the equipment, providing an important reference for long-term operation and maintenance planning.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring and maintenance technology, specifically a power equipment health assessment and fault prediction system. Background Technology

[0002] With the rapid development of the power industry, the health status and fault prediction of power equipment are crucial to ensuring the stable operation of the power system. During operation, power equipment is affected by various internal and external factors, which can lead to a gradual decline in its performance and eventually failure. Therefore, conducting health assessments and fault predictions for power equipment in order to identify potential problems in a timely manner and take corresponding measures has become an important research direction in the power industry.

[0003] Traditional power equipment health assessments and fault predictions primarily rely on manual inspections and periodic checks. This approach is not only time-consuming and labor-intensive, but also struggles to achieve real-time monitoring and accurate prediction of equipment status. Furthermore, traditional methods are often based on experience-based judgments, lacking scientific data support and intelligent analysis tools, resulting in inaccurate assessment results and insufficient fault prediction capabilities. Specifically, traditional technologies suffer from the following shortcomings: First, limited data acquisition and processing capabilities prevent the acquisition of comprehensive equipment operating status information in real time; second, the fault prediction model is simplistic and cannot adapt to the diversity and complexity of equipment faults; and third, the operation and maintenance strategies lack specificity, failing to develop personalized operation and maintenance plans based on equipment health status and fault risk levels.

[0004] Therefore, developing a power equipment health assessment and fault prediction system will greatly improve the accuracy and reliability of power equipment fault prediction, providing a strong guarantee for the stable operation of the power system. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a power equipment health assessment and fault prediction system. This system integrates data acquisition, processing, storage, and short-term, medium-term, and long-term fault prediction models. Through model fusion and decision-making modules, it achieves accurate assessment of equipment health status and fault early warning. Compared with traditional technologies, this system can detect potential faults earlier, improve prediction accuracy, reduce unplanned downtime, and optimize operation and maintenance strategies. Through the user interface, operation and maintenance personnel can intuitively obtain equipment health information and operation and maintenance suggestions, thereby improving work efficiency.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a power equipment health assessment and fault prediction system, which includes: a data acquisition module, a data processing and storage module, a short-term fault prediction model, a medium-term fault prediction model, a long-term fault prediction model, a model fusion and decision-making module, and a user interface.

[0007] The data acquisition module: by connecting to sensors on the power equipment, it collects parameter data including electrical, mechanical and environmental aspects, and transmits the data to the data processing and storage module in real time through a wireless communication link;

[0008] The data processing and storage module receives data transmitted from the data acquisition module, cleans, filters, and normalizes the received data to remove noise and outliers, extracts feature vectors from power equipment data using the power data similarity feature extraction formula, classifies and stores the processed data in short-term, medium-term, and long-term data storage areas according to the time scale, and simultaneously builds a search index and management system.

[0009] The short-term fault prediction model is constructed based on the changes in the operating status of power equipment over the past few minutes to several hours. Let the operating state vector of the power equipment at time t be X(t) = [x1(t), x2(t), ..., x...]. n (t)] T , where x i (t) represents the i-th running parameter, and the formula is: in, It is parameter x i In the past ΔT s The mean over the time period, σ i (t-ΔT s ) is the corresponding standard deviation, w i,s It is the weight of the i-th parameter in the short-term model, α s It is an index that regulates abnormal sensitivity; when P s (t) > Short-term model fault judgment threshold T s,th Triggering a short-term fault warning immediately outputs detailed fault warning information, including the fault type, location of the fault, and information on related components, and sends the abnormal feature vector V. s (t)=[v s,1 (t),v s,2 (t),…,v s,n (t)] T Passed to the intermediate model, where:

[0010] The mid-term fault prediction model: This model involves in-depth mining and analysis of historical data in the mid-term data storage area, combined with the abnormal feature vector V passed from the short-term model. s (t) Construct a mid-term failure prediction model, the formula is: Among them, z j (t) is the j-th intermediate feature parameter in the intermediate data storage area, w i,m and θ j,m These are the corresponding weights, βm It is a coefficient related to time decay. Real-time data is input into the model for prediction, and the output is a detailed performance change curve of the device in the next 3 days to 2 weeks. It predicts the expected time point when the performance will degrade to the mid-term model failure judgment threshold, sends it to the operation and maintenance personnel to arrange planned maintenance work, and generates a mid-term key performance indicator vector K. m (t+ΔT m )=[k m,1 (t+ΔT m ),k m,2 (t+ΔT m ),…,k m,q (t+ΔT m )] T Passed to the long-term model, where k m,q (t+ΔT m ) represents the predicted value of the q-th interim key performance indicator;

[0011] The long-term failure prediction model is constructed using a combination of theoretical and data-driven methods, and the formula is as follows: Among them, y r (t) is the r-th long-term characteristic parameter in the long-term data storage area, ω q,l and φ r,l These are the corresponding weights. The model predicts the remaining lifespan of power equipment, and the results are fed back into the short-term and medium-term models to adjust their fault judgment thresholds and accuracy requirements.

[0012] The model fusion and decision-making module: Based on the prediction results of the short-term fault prediction model, the medium-term fault prediction model and the long-term fault prediction model, it assigns corresponding dynamic weights to them, uses a weighted summation strategy to obtain the overall health assessment index of the equipment, classifies equipment faults into multiple risk levels according to the health assessment index, formulates personalized operation and maintenance strategies for different levels, receives new data and feedback in real time, and optimizes the parameters and structure of each model according to the new data and operation and maintenance feedback.

[0013] The user interface displays equipment health assessments, fault prediction information, and maintenance suggestions using intuitive charts and text. It also features a historical data query function and supports access from multiple terminal devices.

[0014] Furthermore, the sensors used in the data acquisition module:

[0015] Electrical sensors: voltage transformers, current transformers, and power sensors;

[0016] Mechanical sensors: accelerometers, displacement sensors, and strain gauges;

[0017] Environmental sensors: temperature sensor, humidity sensor, air pressure sensor.

[0018] Furthermore, the data processing and storage module extracts feature vectors from power equipment data using a power data similarity feature extraction formula. Let the original data sequence be D = {d1, d2, ..., d...}. N The extracted feature vector F = [f1, f2, ..., f M The calculation formula is: Where, k(d) i ,d j ) is data point d i With d j The similarity kernel function between them is given by the formula: σ is the kernel function bandwidth parameter, g(d i -d j () is a weighted function for the data differences, and the formula is: α is the adjustment parameter, and λ is the attenuation coefficient. It is the mean of the data sequence D.

[0019] Furthermore, the short-term model fault judgment threshold T in the short-term fault prediction model... s,th The determination of A is obtained by calculating normal operating data. s mean of (t) and standard deviation The formula for calculating the fault judgment threshold in the short-term model is: Where n is a multiple determined based on the tolerance for failure risk.

[0020] Furthermore, the intermediate-term model fault judgment threshold T in the intermediate-term fault prediction model m,th,i The setting is as follows: let the key performance indicators of the equipment be KPIs. For a specific key performance indicator KPI... i The initial design value of this indicator is KPI. i,init The performance degradation limit is α. i Regarding this key performance indicator (KPI) i Mid-term model fault judgment threshold T m,th,i The calculation formula is: T m,th,i =KPI i,init ×(1-α i ).

[0021] Furthermore, in the long-term fault prediction model, the fault judgment thresholds of the short-term and medium-term models are adjusted based on the long-term model:

[0022] For adjusting the fault judgment threshold of the short-term model, let R be the proportion of the equipment's remaining life as assessed by the long-term model. L (t), 0≤R L (t)≤1, R L(t) = 1 indicates that the equipment is brand new, R L (t) = 0 indicates that the equipment has reached the end of its lifespan, T s,th The original short-term model fault judgment threshold is α, and β are adjustment parameters. The new short-term model fault judgment threshold formula is: T s,new,th =T s,th ×(1-α×R L (t))+β×(1-R L (t));

[0023] Regarding the adjustment of the fault judgment threshold for the mid-term model, let the fault judgment threshold for the mid-term model be T. m,th,i The long-term model estimates the rate of equipment performance degradation as D. L (t), where λ and μ are adjustment parameters, the new formula for calculating the fault judgment threshold of the medium-term short-term model is: T m,new,th =T m,th,i ×(1-λ×D L (t))+μ×D L (t).

[0024] Furthermore, the model fusion and decision-making module uses dynamic weights to calculate the reliability index of each model, assuming the short-term model reliability index is R. s Calculate the prediction accuracy over a recent time period, where N is the number of correct predictions. sc The total number of predictions is N. st The timeliness weight of recent predictions w t And w t =e -μ·t Where t is the time interval from the current moment, and μ is the attenuation coefficient, calculated using the following formula: Mid-term model reliability index R m Let the medium-term predicted performance change curve be C. m (t), the actual performance change curve is C ma (t), within the time interval [t1,t2], the similarity S between the two curves is calculated. m And combined with data integrity index I m The formula is: R m =S m ·I m Long-term model reliability index R l Let the predicted long-term remaining life be L. p The actual remaining lifespan is L a The confidence interval width is W l The formula is: Weights w are assigned to the short-term, medium-term, and long-term models based on reliability metrics. s w m w l They are respectively: Where β s β m β l To adjust the index.

[0025] Furthermore, in the model fusion and decision-making module, the overall health assessment index of the equipment condition is set as P, where the prediction result of the short-term failure prediction model is P. s The prediction result of the mid-term failure prediction model is P. m The prediction result of the long-term failure prediction model is P. l The dynamic pairing weights for each model, with the short-term model weights being w. s The intermediate model weights are w m The long-term model weights are w l The overall health assessment index H of the equipment is calculated using the following formula: Where, k s k m k l This is the adjustment coefficient.

[0026] Furthermore, in the model fusion and decision-making module, the risk level classification and personalized operation and maintenance strategies for different levels are defined, assuming the overall equipment health assessment index is H, and a low-risk threshold T is defined. low Medium risk threshold T mid And 0 < T low <T mid <1:

[0027] When H>T low At this time, the equipment is in a low-risk state, and the monitoring frequency f is [not specified]. low Set as the basic monitoring frequency f base Data is collected every hour and a simple inspection is conducted every quarter. Resource allocation mainly focuses on ensuring the availability of daily maintenance resources.

[0028] When T mid <H≤T low At that time, the equipment was in a medium-risk state, and the monitoring frequency was f. mid Increase f base ·k mid (k mid >1), the maintenance plan is formulated for future T pm A detailed inspection will be conducted within a short period of time. pm It can be determined based on the mid-term model prediction results, and the calculation formula is as follows: ΔT m,th It is the time difference between the mid-term performance degradation and the threshold, P m (t) is the current mid-term performance prediction, ΔP mIt refers to the rate of change in medium-term performance, and the proportion of emergency maintenance resources (r) increased by resource allocation. mid ;

[0029] When H≤T mid At that time, the equipment was in a high-risk state, and the monitoring frequency was f. high Significantly increased to f base ·k high k high >k mid Immediately initiate the shutdown and maintenance plan, with maintenance time T. repair The formula is calculated based on the fault type and available maintenance resources: Where t ri r is the time required for the i-th repair step. i It represents the resource sufficiency rate for executing the i-th maintenance step, with resource allocation fully guaranteeing the manpower, material resources, and technical support required for maintenance.

[0030] Furthermore, the low-risk threshold H in the model fusion and decision-making module low And medium risk threshold H mid To determine this, let the mean H value of health devices in historical data be μ. H The standard deviation is σ H Low risk threshold H low Set to: H low =μ H -k1σ H Where k1 is a coefficient determined based on the tolerance for risk, and the medium-risk threshold H mid Set to: H mid =μ H -k2σ H , where k2>k1.

[0031] Compared with existing technologies, this power equipment health assessment and fault prediction system has the following advantages:

[0032] I. This invention integrates short-term, medium-term, and long-term fault prediction models to achieve a comprehensive assessment of the health status of power equipment and fault prediction. The short-term model can quickly capture abnormal changes in the current operating status of the equipment and provide immediate fault warnings. The medium-term model combines time series analysis and mechanistic models to predict the performance change trend of the equipment in the future, providing decision support for planned maintenance work. The long-term model focuses on predicting the remaining life of the equipment, providing an important reference for the long-term operation and maintenance planning of the equipment. This multi-timescale fault prediction and health assessment method improves the accuracy and practicality of prediction.

[0033] Second, this invention introduces dynamic weight allocation and personalized operation and maintenance strategy formulation functions into the model fusion and decision-making module. By calculating the reliability indicators of each model, the system can dynamically adjust the weights of short-term, medium-term, and long-term models in health assessment, thereby ensuring the objectivity and accuracy of the assessment results. At the same time, the system can also formulate personalized operation and maintenance strategies based on the overall health assessment indicators and risk level classification of the equipment, including adjusting the monitoring frequency, formulating maintenance plans, and optimizing resource allocation. This intelligent operation and maintenance decision support function not only improves operation and maintenance efficiency but also reduces operation and maintenance costs.

[0034] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0036] Figure 1 A flowchart for a power equipment health assessment and fault prediction system;

[0037] Figure 2 This is an architecture diagram of a power equipment health assessment and fault prediction system. Detailed Implementation

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1:

[0040] Health assessment and fault prediction of substation transformer equipment

[0041] Data acquisition and processing: In substations, data acquisition modules are deployed for transformer equipment. Electrical sensors (voltage transformers, current transformers, power sensors) collect electrical parameters, mechanical sensors (accelerometers, displacement sensors, strain gauges) monitor mechanical conditions, and environmental sensors (temperature, humidity, and air pressure sensors) acquire environmental information. The acquired data is transmitted wirelessly in real-time to the processing and storage module for cleaning, filtering, and normalization to remove noise and outliers. Feature vectors are then extracted from the power equipment data using a power data similarity feature extraction formula. Let the original data sequence be D = {d1, d2, ..., d...}. N The extracted feature vector F = [f1, f2, ..., f M The calculation formula is: Where, k(d) i ,d j ) is data point d i With d j The similarity kernel function between them is calculated using the following formula: σ is the kernel function bandwidth parameter, g(d i -d j () is a weighted function for the data differences, and its calculation formula is: α is the adjustment parameter, and λ is the attenuation coefficient. It is the mean of the data sequence D, and then it is stored according to the time scale.

[0042] Model prediction and analysis:

[0043] Short-term fault prediction: A short-term fault prediction model is constructed based on the transformer's current operating status from a few minutes to a few hours. Let the transformer's operating state vector at time t be X(t)=[x1(t),x2(t),…,x n (t)] T , where x i (t) represents the i-th operating parameter. The model is constructed based on the short-term anomaly index calculation formula, which is: in, It is parameter x i In the past ΔT s The mean over the time period, σ i (t-ΔT s ) is the corresponding standard deviation, w i,s It is the weight of the i-th parameter in the short-term model, α s It is an index that regulates abnormal sensitivity; when P s (t)>T s,th When the short-term fault threshold is reached, a short-term fault warning is triggered, fault warning information is output, and the abnormal feature vector V is displayed. s (t)=[v s,1 (t),vs,2 (t),…,v s,n (t)] T Passed to the intermediate model, where

[0044] Mid-term fault prediction: A mid-term fault prediction model is constructed by combining time series analysis with a variable instrument operation mechanism model. In-depth mining and analysis of historical data in the mid-term data storage area are conducted, combined with the abnormal feature vector V transmitted from the short-term model. s (t) Construct a mid-term failure prediction model, the formula is: Among them, z j (t) is the j-th intermediate feature parameter in the intermediate data storage area, w i,m and θ j,m These are the corresponding weights, β m It is a coefficient related to time decay. By optimizing the model using historical data and evaluating the model using independent data, real-time data is input for prediction, and the output is the performance change curve of the device in the next 3 days to 2 weeks and the predicted time point when the performance will decline to the warning threshold, generating a mid-term key performance indicator vector K. m (t+ΔT m )=[k m,1 (t+ΔT m ),k m,2 (t+ΔT m ),…,k m,q (t+ΔT m )] T Passed to the long-term model;

[0045] Long-term failure prediction: A long-term failure prediction model is constructed using a combination of theoretical and data-driven methods. The formula is as follows: Among them, y r (t) is the r-th long-term characteristic parameter in the long-term data storage area, ω q,l and φ r,l These are the corresponding weights, which are used to predict the remaining life of the transformer through the model, and then feed back into the short-term and medium-term models to adjust the fault judgment thresholds and accuracy requirements.

[0046] Model fusion and decision-making: Based on the prediction results of short-term, medium-term, and long-term failure prediction models, the model fusion and decision-making module assigns dynamic weights and calculates the reliability index of each model. Let the reliability index of the short-term model be R. s Calculate the prediction accuracy over a recent time period, with N being the number of correct predictions. sc The total number of predictions is N. st The timeliness weight of recent predictions w t =e -μ·tt is the time interval from the current time, μ is the attenuation coefficient, and the formula is: Mid-term model reliability index R m Let the medium-term predicted performance change curve be C. m (t), the actual performance change curve is C ma (t), within the time interval [t1,t2], the similarity S between the two curves is calculated. m And combined with data integrity index I m The calculation formula is: R m =S m ·I m Long-term model reliability index R l Let the predicted long-term remaining life be L. p The actual remaining lifespan is L a The confidence interval width is W l The calculation formula is: Weights w are assigned to the short-term, medium-term, and long-term models based on reliability metrics. s w m w l They are respectively: Where β s β m β l To adjust the index, calculate the overall equipment health assessment index H, and let the prediction result of the short-term failure prediction model be P. s The prediction result of the mid-term failure prediction model is P. m The prediction result of the long-term failure prediction model is P. l The calculation formula is: Where k s k m k l Let T be the adjustment coefficient and the low-risk threshold. low The medium-risk threshold is T. mid (0 <T low <T mid <1), when H>T low At that time, the equipment is in a low-risk state, and the monitoring frequency is set to the basic monitoring frequency f. base When T mid <H≤T low When H ≤ T, the equipment is in a medium-risk state, and the monitoring frequency is increased; when H ≤ T mid When the equipment is in a high-risk state, the monitoring frequency is increased significantly and a shutdown maintenance plan is initiated. At the same time, personalized operation and maintenance strategies and resource allocation plans are formulated according to different risk levels.

[0047] In summary, this embodiment takes substation transformer equipment as the core, constructs a comprehensive data acquisition system, collects electrical, mechanical, and environmental data through multiple types of sensors, processes and stores the data in a standardized manner, and performs short-term fault prediction based on operating parameter vectors and anomaly index formulas to capture sudden faults in real time; the medium-term model integrates multivariate data and mechanistic knowledge to predict medium-term performance trends; the long-term model estimates the remaining lifespan based on characteristic parameters and feeds back to optimize the preceding model; the model fusion module evaluates equipment health based on the prediction results, accurately classifies risk levels, sets monitoring frequencies, plans maintenance and inspection schedules, and allocates resources according to the risk level, forming a closed-loop intelligent operation and maintenance architecture from data to decision-making. This comprehensively ensures the reliable operation of transformers, reduces operation and maintenance costs, and improves power grid stability, providing an efficient and accurate example for substation operation and maintenance management, and has significant engineering application and promotion value.

[0048] Example 2:

[0049] Health assessment and failure prediction of steam turbine generator units in thermal power plants

[0050] In the turbine generator area of ​​the thermal power plant, data acquisition and processing modules are meticulously deployed. Electrical sensors (voltage transformers and current transformers accurately measure generator voltage and current parameters, while power sensors monitor the unit's power output stability and power quality in real time) comprehensively capture electrical operating parameters. Mechanical sensors (accelerometers are closely attached to the turbine journals and bearing housings to sensitively detect vibration; displacement sensors accurately monitor the axial and radial displacement of turbine blades to strictly control blade trajectory; strain gauges are firmly attached to key parts of the turbine cylinder and pipelines to accurately measure stress and strain changes, providing crucial data for mechanical structure health assessment) meticulously monitor mechanical performance. Environmental sensors (temperature sensors are densely distributed around the generator room and cooling system; humidity sensors sense ambient humidity in real time; and air pressure sensors stably monitor the air pressure in the generator room, laying a solid foundation for analyzing the impact of environmental factors on the unit)... The system comprehensively collects environmental information, and the massive amounts of data collected are transmitted in real time to the data processing and storage module via high-speed wireless communication links. The data undergoes rigorous cleaning (removing abnormal data points caused by electromagnetic interference and instantaneous pulses), efficient filtering (using a combination of low-pass, high-pass, and band-pass filters to remove electrical noise and mechanical vibration clutter), and normalization (unifying various parameters to a standard numerical range and eliminating dimensional differences). The data is then intelligently categorized and stored according to time scales. The short-term data storage area accurately retains high-frequency sampling data per minute over the past two hours; the medium-term storage area systematically archives the average data per 10 minutes over the past week; and the long-term storage area integrates the statistical values ​​and extreme values ​​of key parameters for each month over the past year. Simultaneously, an intelligent search index and efficient management system are constructed to ensure that data retrieval and access are fast, accurate, secure, and reliable.

[0051] Model prediction and analysis:

[0052] Short-term fault prediction: A short-term fault prediction model is rigorously constructed based on the current 30-minute operating state vector of the steam turbine generator unit. Let the operating state vector at time t be X(t) = [x1(t), x2(t), ..., x n (t)] T x i (t) represents key operating parameters (such as turbine speed deviation, lubricating oil temperature change rate, and bearing vibration amplitude). A model is constructed based on the short-term anomaly index formula, which is: in, It is parameter x i In the past ΔT s Mean, σ i (t-ΔT s ) represents the standard deviation, w i,s It is the parameter weight, α s Adjusting the sensitivity, such as the vibration amplitude x of the turbine bearing at a certain moment. j (t) mutation, P calculated s (t)>T s,th (The threshold is scientifically set based on statistical analysis of normal operation data and risk tolerance), immediately triggering a sensitive fault warning, outputting detailed fault type (such as bearing wear fault), approximate location (a certain bearing number), and component information, and sending the abnormal feature vector V s (t) Efficient transmission to mid-term models facilitates in-depth analysis;

[0053] Mid-term Fault Prediction: A mid-term fault prediction model is constructed by deeply integrating time series analysis methods with a complex operating mechanism model of steam turbine generator sets. This model deeply mines the rich historical data in the mid-term data storage area and combines it with short-term anomaly feature vectors V. s (t), can be accurately predicted using the formula: Accurate prediction, among which, z j (t) is a mid-term characteristic parameter, w i,m and θ j,m For the weight, β m By correlating time decay, such as predicting the unit efficiency decline curve over the next 10 days, the system accurately determines the time point when key performance indicators (such as the KPI corresponding to heat rate) reach the warning threshold. Based on this, the maintenance team scientifically arranges a deep overhaul plan 5 days later (checking scale buildup in the turbine flow path, adjusting shaft alignment, and optimizing the lubrication system). At the same time, it generates a mid-term key performance indicator vector K. m (t+ΔT m Accurately delivered to the long-term model;

[0054] Long-term failure prediction: A long-term failure prediction model is constructed using a combination of theory and data-driven methods. The formula is as follows: Among them, y r(t) represents long-term characteristic parameters (such as cumulative unit operating time, number of major maintenance operations, and high-temperature creep damage assessment value), ω q,l φ r,l As a weight, the remaining lifespan of the unit is accurately predicted. For example, a unit that has been in operation for 15 years, has undergone 3 major overhauls, and whose long-term characteristic parameters have changed to specific values ​​is predicted to have a remaining lifespan of 3 years. Based on this key result, the short-term and medium-term models are accurately fed back, the short-term failure threshold is dynamically optimized, the sensitivity is intelligently adjusted according to the proportion of remaining lifespan, and the medium-term performance threshold and prediction accuracy are finely corrected. This ensures that the operation and maintenance strategy is closely adapted to the characteristics of each stage of the unit aging process, effectively reducing the risk of failure, extending the service life of equipment, and improving the operating efficiency of thermal power plants.

[0055] Model fusion and decision-making: This module efficiently integrates the results of short-term, medium-term, and long-term fault prediction models, accurately assigns dynamic weights, and scientifically calculates the reliability indicators of each model. The short-term model is evaluated based on its recent prediction accuracy and the number of correct predictions (N). sc With the total number of times N st Ratio combined with timeliness weight w t =e -μ·t μ is the attenuation coefficient, which is determined by the formula: The intermediate-term model is based on the similarity S between the predicted performance curve and the actual performance. m and data integrity I m The calculation formula is: R m =S m ·I m The long-term model determines R using a formula based on the remaining lifetime prediction bias and the confidence interval width. l The formula is: Based on reliability metrics, weights are precisely allocated according to a formula, which is: β s β m β l To adjust the index, the overall health assessment index of the equipment is calculated using the following formula: k s k m k l As an adjustment coefficient, risk levels are precisely classified accordingly, with a low-risk threshold T set. low Medium risk threshold T mid 0 <T low <T mid <1, when H>T low When the risk is low, the monitoring frequency is set to a normal value, and resource allocation focuses on daily maintenance and material supply. mid <H≤T lowAt the medium-risk level, monitoring frequency will be increased (data collected every 30 minutes, in-depth inspections weekly), and maintenance resources will be prepared within 7 days based on the medium-term forecast plan; when H≤T mid When the situation is high-risk, the monitoring frequency increases dramatically (data is collected every 10 minutes, and the equipment is immediately shut down for maintenance). All available manpower, spare parts, and technical experts are mobilized to quickly repair the equipment according to the emergency repair plan, ensuring stable power generation from the thermal power plant and improving the reliability and economic efficiency of power supply.

[0056] In summary, this embodiment focuses on steam turbine generator units in thermal power plants, constructing a comprehensive, multi-layered data-driven health assessment and fault prediction system. Through the comprehensive deployment of sensors to collect electrical, mechanical, and environmental data, which, after rigorous processing and categorized storage, provide a solid data foundation for models at each stage. The short-term model keenly captures potential sudden faults based on real-time parameters; the medium-term model integrates historical trends and short-term anomalies to accurately outline the performance evolution trajectory; and the long-term model accurately quantifies the remaining lifespan based on the unit's full life-cycle characteristics. Furthermore, these three models work together and are dynamically optimized. The model fusion module, through a scientific weight allocation mechanism, refines comprehensive health indicators and accurately defines risk levels, thereby customizing operation and maintenance plans and allocating resources based on risk characteristics. This system forms an efficient closed-loop operation and maintenance strategy, effectively ensuring stable and efficient unit operation, significantly reducing the risk of unplanned shutdowns, deeply optimizing operation and maintenance resource allocation, and continuously improving the economic benefits and power supply reliability of thermal power plants. It sets an innovative example for the operation and maintenance management of key equipment in the energy industry, possessing significant industry promotion value and far-reaching practical significance.

[0057] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A power equipment health assessment and failure prediction system, characterized by, The system comprises a data acquisition module, a data processing and storage module, a short-term fault prediction model, a medium-term fault prediction model, a long-term fault prediction model, a model fusion and decision module, and a user interaction interface. The data acquisition module is connected to the sensors of the power equipment, acquires parameter data including electrical, mechanical, and environmental aspects, and transmits the data to the data processing and storage module in real time through a wireless communication link. The data processing and storage module receives the data transmitted by the data acquisition module, cleans, filters, and normalizes the received data, removes noise and outliers, and extracts feature vectors from the power equipment data through a power data similarity feature extraction method. The method achieves effective feature extraction by mining data similarity and weighted processing data differences. The processed data is classified and stored in short-term, medium-term, and long-term data storage areas according to time scales, and a search index and management system are constructed. The data processing and storage module extracts feature vectors in power equipment data through power data similarity feature extraction formula, sets original data sequence as D={d1, d2, …, d N}, extracts feature vectors F=f1, f2, …, f M ], and the calculation formula is as follows: Wherein, k(d i ,d j ) is a similarity kernel function between data points d i and d j , and the formula is as follows: σ is a kernel function bandwidth parameter, g(d i -d j ) is a weighted function of data difference, and the formula is as follows: α is an adjustment parameter, λ is a decay coefficient, is the mean of data sequence D. The short-term fault prediction model: based on the current several minutes to several hours of running state change data of power equipment to construct short-term fault prediction model, set the running state vector of power equipment at time t as X(t)=[x1(t),x2(t),…,xN(t)], N is the number of running parameters, and the formula is as follows: n (t)] T Wherein x i (t) represents the i th running parameter, is the mean value of the parameter x i in the past ΔT s period, σ i (t-ΔT s ) is the corresponding standard deviation, w i,s is the weight of the i th parameter in the short-term model, and α s is an index for adjusting the sensitivity of the anomaly, when P s (t) > short-term model fault judgment threshold T s,th , the short-term fault warning is triggered, and the detailed fault warning information is immediately output, including the fault type, the location of the fault and the related component information, and the abnormal feature vector V s (t)=[v s,1 (t),v s,2 (t),…,v s,n (t)] T is transmitted to the medium-term model, wherein: ​ The medium-term failure prediction model: through deep mining and analysis of historical data in the medium-term data storage area, combining the abnormal feature vector V delivered by the short-term model s (t) is a medium-term failure prediction model, and the formula is: Wherein, z j (t) is the jth medium-term feature parameter in the medium-term data storage area, w i,m and θ j,m are the corresponding weights, β m is a coefficient related to time decay, real-time data is input into the model for prediction, the output device outputs the detailed performance change curve of the device in the future 3 days to 2 weeks, predicts the time point when the performance decreases to the medium-term model failure judgment threshold, sends it to the operation and maintenance personnel to arrange planned maintenance work, and generates a medium-term key performance indicator vector K m (t+ΔT m )=[k m,1 (t+ΔT m ),k m,2 (t+ΔT m ),…,k m,q (t+ΔT m )] T to the long-term model, wherein k m,q (t+ΔT m ) represents the qth medium-term key performance indicator prediction value; The long-term fault prediction model: a long-term fault prediction model is constructed by using a theory combined with a data-driven method, and a formula is as follows: Wherein, y r (t is the rth long-term characteristic parameter in the long-term data storage area, ω q,l and φ r,l are corresponding weights, the remaining life of the power equipment is predicted by the model, and the result is fed back to the short-term and medium-term models to adjust the fault judgment threshold and the accuracy requirement. The model fusion and decision module assigns dynamic weights to the prediction results of the short-term, medium-term, and long-term fault prediction models, uses a weighted summation strategy to obtain a health assessment index of the overall condition of the equipment, divides the equipment fault into multiple risk levels based on the health assessment index, and develops individualized operation and maintenance strategies for different levels. The system receives new data and feedback in real time and optimizes the model parameters and structure based on the new data and operation and maintenance feedback. The user interaction interface displays the equipment health assessment, fault prediction information, and operation and maintenance suggestions in intuitive charts and text, has a historical data query function, and supports multi-terminal device access.

2. The power equipment health assessment and failure prediction system of claim 1, wherein, The sensors used in the data acquisition module include: Electrical sensors: voltage transformers, current transformers, and power sensors; Mechanical sensors: accelerometers, displacement sensors, and strain gauges; Environmental sensors: temperature sensors, humidity sensors, and barometric pressure sensors.

3. The power equipment health assessment and failure prediction system of claim 1, wherein, The short-term fault prediction model includes a short-term model fault judgment threshold T s,th A s (t) is calculated based on normal operation data and standard deviation The short-term model fault judgment threshold is calculated according to the following formula: Wherein n is a multiple determined according to the fault risk tolerance.

4. The power equipment health assessment and failure prediction system of claim 1, wherein, The intermediate-term fault judgment threshold T in the intermediate-term fault prediction model m,th,i Let the key performance indicators (KPIs) of the equipment be defined as follows: For any key performance indicator (KPI) of the equipment... i The initial design value of this indicator is KPI. i,init The performance degradation limit is α. i Regarding this key performance indicator (KPI) i Mid-term model fault judgment threshold T m,th,i The calculation formula is: T m,th,i =KPI i,init ×(1-α i ).

5. The power equipment health assessment and failure prediction system of claim 1, wherein, In the long-term fault prediction model, the fault judgment thresholds of the short-term and medium-term models are adjusted according to the long-term model. For the adjustment of short-term model fault judgment threshold, let the residual life ratio of the equipment evaluated by the long-term model be R L (t), 0≤R L (t)≤1, R L (t)=1 represents that the equipment is brand new, and R L (t)=0 represents that the equipment is exhausted, and T s,th is the original short-term model fault judgment threshold, and α and β are adjustment parameters, and the new short-term model fault judgment threshold formula is T s,new,th =T s,th ×(1-α×R L (t))+β×(1-R L (t)); For adjustment of the intermediate model fault judgment threshold, let the intermediate model fault judgment threshold be T m,th,i , the device performance decline rate evaluated by the long-term model be D L (t), and λ and μ be adjustment parameters. A new intermediate model fault judgment threshold calculation formula is T m,new,th =T m,th,i ×(1-λ×D L (t))+μ×D L (t).

6. The power equipment health assessment and failure prediction system of claim 1, wherein, The model fusion and decision-making module uses dynamic weights to calculate the reliability index of each model. Let the short-term model reliability index be R. s Calculate the prediction accuracy over a recent time period, where N is the number of correct predictions. sc The total number of predictions is N. st The timeliness weight of recent predictions w t And w t =e -μ·t Where t is the time interval from the current moment, and μ is the attenuation coefficient, calculated using the following formula: Mid-term model reliability index R m Let the medium-term predicted performance change curve be C. m (t), the actual performance change curve is C ma (t), within the time interval [t1,t2], the similarity S between the two curves is calculated. m And combined with data integrity index I m The formula is: R m =S m ·I m Long-term model reliability index R l Let the predicted long-term remaining life be L. p The actual remaining lifespan is L a The confidence interval width is W l The formula is: Weights w are assigned to the short-term, medium-term, and long-term models based on reliability metrics. s w m w l They are respectively: Where β s β m β l To adjust the index.

7. The power equipment health assessment and failure prediction system of claim 1, wherein, The model fuses the health evaluation index of the overall condition of the equipment in the decision module, the prediction result of the short-term fault prediction model is P s , the prediction result of the medium-term fault prediction model is P m , the prediction result of the long-term fault prediction model is P l , the dynamic matching weight corresponding to each model, the short-term model weight is w s , the medium-term model weight is w m , and the long-term model weight is w l , and the overall health evaluation index H of the equipment is calculated, and the formula is: Wherein, k s , k m , and k l are adjustment coefficients.

8. The power equipment health assessment and failure prediction system of claim 1, wherein, The model fuses the division of risk levels in the model fusion and decision module and formulates individual operation and maintenance strategies for different levels. Let the overall health evaluation index of the equipment be H, define a low-risk threshold T low , a medium-risk threshold T mid , and 0 < T low < T mid < 1. When H > T low , the device is in a low-risk state, at which time the monitoring frequency f low is set to a basic monitoring frequency f base , data is collected once an hour, a simple inspection is performed once a quarter, and resource allocation is mainly focused on ensuring daily maintenance resources. When T mid <H≤T low , the device is in a medium-risk state, the monitoring frequency f mid is increased base ·k mid , k mid >1, and the maintenance plan is made to perform detailed inspection in the future T pm , T pm can be determined according to the medium-term model prediction result, and the calculation formula is as follows: ΔT m,th is the time difference of the medium-term performance falling to the threshold value, P m (t) is the current medium-term performance prediction value, ΔP m is the medium-term performance change rate, and resource allocation increases the reserve proportion r mid of emergency maintenance resources; When H ≤ T mid , the device is in a high-risk state, and the monitoring frequency f high is greatly increased to f base · k high , k high > 1 mid , and the maintenance plan is immediately started, with maintenance time T repair . According to the fault type and maintenance resource conditions, the formula is: where t ri is the time required for the i th maintenance step, and r i is the resource sufficiency rate for performing the i th maintenance step, with resources being fully allocated to ensure manpower, material resources, and technical support required for maintenance.

9. The power equipment health assessment and failure prediction system of claim 8, wherein, said model fuses the determination of the low risk threshold H low and the medium risk threshold H mid , assuming that the mean value of the H values of the healthy devices in the historical data is μ H and the standard deviation is σ H , the low risk threshold H low is set as: H low = μ H - k1σ H where k1 is a coefficient determined according to the tolerance to risk, the medium risk threshold H mid is set as: H mid = μ H - k2σ H where k2 > k1.

Images