Low-voltage motor insulation resistance on-line monitoring and early warning method, system, equipment and medium

Abstract

The invention relates to a low-voltage motor insulation resistance online monitoring and early warning method, system and device and a medium. The method comprises the following steps: firstly, acquiring multi-dimensional operation data through a data acquisition link to form an original monitoring data set; preprocessing the original monitoring data set, and focusing core information through dimension reduction processing to obtain a key parameter set; and then inputting the key parameter set into a dynamic correlation model constructed based on a long short-term memory network, and outputting a prediction result of the variation trend of the insulation resistance value along with time and the deterioration degree quantification. And finally, carrying out early warning analysis based on a prediction result, and generating early warning fault features in combination with a real degradation judgment process. According to the method, the insulation resistance trend and the degradation degree are accurately predicted, the defect of insufficient real-time performance of off-line detection is overcome, the accuracy and the pertinence of fault early warning are improved, and equipment shutdown and safety risks caused by insulation faults are effectively reduced.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electrical engineering, and particularly relates to a low-voltage motor insulation resistance online monitoring and early warning method, system, device and medium. BACKGROUND

[0002] In industrial production and civil facilities, low-voltage motors are widely used as core power equipment, and their operation reliability is directly related to production efficiency and system safety. Insulation performance is a key indicator for measuring the health status of a low-voltage motor, and the degradation or abnormal decrease of insulation resistance can easily cause motor short circuit, grounding and other faults, which may lead to equipment downtime for maintenance, or even cause safety accidents and significant economic losses. Traditional insulation resistance monitoring relies on offline oscillometer measurement or periodic shutdown detection, which has the defects of long monitoring interval and poor real-time performance, and cannot capture the dynamic degradation trend and early abnormal signals of insulation performance. At the same time, due to environmental interference, complex working conditions and other factors, the existing online monitoring technology often has large noise in monitoring data, inaccurate feature extraction, high false alarm rate, and cannot effectively support the fine evaluation and early fault warning of the insulation state of a low-voltage motor. SUMMARY

[0003] Therefore, it is necessary to provide a low-voltage motor insulation resistance online monitoring and early warning method, system, device and medium which can accurately predict the insulation resistance trend and degradation degree, improve the accuracy and pertinence of fault warning, and effectively reduce the equipment downtime and safety risks caused by insulation faults.

[0004] In a first aspect, the application provides a low-voltage motor insulation resistance online monitoring and early warning method, which comprises the following steps:

[0005] Obtaining multi-dimensional operation data including insulation resistance, environmental parameters and working condition characteristics to obtain an original monitoring data set; the environmental parameters include temperature and humidity data; the working condition characteristics include load mutation characteristics and voltage stress distribution data.

[0006] Noise filtering and associated feature extraction are performed on the original monitoring data set, and a key parameter set is obtained through dimension reduction processing; the associated features include response coefficients of insulation resistance and environmental temperature, sensitive coefficients of insulation resistance and environmental humidity, and coupling correlation degrees of insulation resistance and load mutation.

[0007] The key parameter set is input into a trained dynamic correlation model to obtain a prediction result of the change trend of insulation resistance value over time and the degradation degree quantization; the dynamic correlation model uses a long short-term memory network to analyze the mapping relationship between parameter dynamic correlation and insulation degradation degree.

[0008] Based on the prediction result, the parameters exceeding the preset threshold are subjected to mutation feature detection and real degradation judgment to obtain early warning fault features, which include mutation point time sequence information, degradation mode code, fault position coordinates and parameter abnormal value.

[0009] In one of the embodiments, noise filtering and correlation feature extraction are performed on the original monitoring data set, and key parameter set is obtained through dimension reduction processing, including:

[0010] The time sequence data signal in the original monitoring data set is obtained, including time sequence record data of insulation resistance value, temperature, humidity, load mutation and voltage stress distribution.

[0011] The time sequence data signal is subjected to noise filtering processing by wavelet transform to obtain denoised multi-dimensional time sequence signal.

[0012] The correlation features of insulation resistance and environmental parameters and working condition characteristics are extracted from the denoised multi-dimensional time sequence signal to generate a feature matrix.

[0013] The feature matrix is subjected to dimension reduction processing by principal component analysis to obtain a key parameter set focusing on the core characteristics of insulation state.

[0014] In one of the embodiments, the key parameter set is input into the trained dynamic correlation model to obtain the prediction result of the change trend of insulation resistance value with time and the degradation degree quantization, including:

[0015] The insulation resistance related feature data and environmental humidity data are extracted from the key parameter set; the feature data includes voltage, current and temperature parameters; the environmental humidity data has high correlation with the insulation resistance.

[0016] The feature data is input into the trained dynamic correlation model to obtain the trend prediction value of the insulation resistance with time; the dynamic correlation model is constructed by using long short-term memory network.

[0017] The feature importance analysis is performed on the environmental humidity data and the feature data by using random forest algorithm to obtain the influence weight of humidity on insulation resistance degradation.

[0018] The parameters of the dynamic correlation model are adjusted according to the influence weight to obtain the optimized trend prediction value, and the insulation degradation index is calculated by using the formula; the insulation degradation index reflects the aging state of the insulation material.

[0019] The abnormal fluctuation feature is extracted from the optimized trend prediction value, and the insulation degradation index is updated by using the abnormal fluctuation feature.

[0020] The optimized trend prediction value and the updated insulation degradation index are integrated to obtain the prediction result of the change and degradation of the insulation resistance value.

[0021] In one embodiment, the insulation deterioration index is calculated by the following formula:

[0022]

[0023] wherein D represents the insulation deterioration index, R pred (t) represents the t-time insulation resistance optimization prediction value output by the dynamic correlation model, R base represents the insulation resistance reference value under the normal state of the device, calibrated by historical health data, W humid represents the weight of the influence of humidity on insulation deterioration calculated by the random forest algorithm, with a value range of [0, 1], ΔH(t) represents the over-standard amount of environmental humidity H(t) relative to the normal humidity threshold H thresh , ΔH(t) = H(t) - H thresh , F fluct (t) represents the abnormal fluctuation feature quantization value extracted from the optimization trend prediction value, obtained by normalizing the deviation variance of the prediction value and the smooth trend line, with a value range of [0, 1], α, β, γ respectively represent the weight coefficients of insulation resistance trend, humidity influence, and abnormal fluctuation, α + β + γ = 1, determined by historical fault data training optimization.

[0024] In one embodiment, based on the prediction results, the parameters exceeding the preset threshold are detected for mutation feature and real deterioration judgment, to obtain early warning fault features, including:

[0025] Extracting early warning parameters exceeding the preset early warning threshold from the prediction results; the early warning parameters include insulation state parameter abnormal value, environmental correlation parameter over-standard value, and working condition feature abnormal value.

[0026] Using the Isolation Forest algorithm to extract the mutation points of the early warning parameters to obtain a mutation feature set containing mutation time, mutation amplitude, and correlation parameter fluctuation features.

[0027] Based on the mutation feature set, querying the pre-established deterioration mode library to obtain real deterioration early warning results including deterioration type and confidence through feature matching degree calculation; the deterioration mode library stores the mapping relationship between mutation features and insulation deterioration types in historical fault cases.

[0028] If the deterioration confidence in the real deterioration early warning result exceeds the judgment threshold, the fault positioning result including the fault component and the deterioration degree level is determined based on the deterioration mode library combined with the device structure parameters, and the early warning fault features are generated.

[0029] In one embodiment, the feature matching degree is calculated by the following formula:

[0030]

[0031] wherein, M represents the feature matching degree, x i represents the i-th feature parameter in the current set of mutation features, y i represents the i-th corresponding feature parameter of the historical failure mode in the degradation mode library, σ i represents the historical fluctuation standard deviation of the i-th feature parameter, ω i represents the importance weight of the i-th feature parameter, which is obtained by training the historical data through a random forest algorithm, and the value range is [0, 1], and ∑ω i > 0.

[0032] In one of the embodiments, after the pre-warning failure feature is generated, the method further comprises:

[0033] Taking the pre-warning failure feature as a training sample, a failure classification model is constructed through a random forest algorithm, and the failure type and prediction probability are output; the failure classification model adopts multi-decision tree ensemble learning, and the feature importance is calculated through feature split gain.

[0034] The real-time operation log of the equipment is matched by using the failure classification model, potential failure modes are extracted, and an extended failure feature set containing new feature dimensions and association rules is generated; the potential failure modes include feature combinations that have not triggered the pre-warning threshold but have a degradation trend.

[0035] Based on the degradation sensitivity weight of each feature in the extended failure feature set, the collection frequency of the equipment monitoring parameters and the pre-warning threshold setting are optimized to obtain an adaptive dynamic pre-warning and monitoring strategy.

[0036] In a second aspect, the application also provides a low-voltage motor insulation resistance online monitoring and pre-warning system, which comprises:

[0037] A data acquisition module is configured to obtain multi-dimensional operation data including insulation resistance, environmental parameters and working condition features to obtain an original monitoring data set; the environmental parameters include temperature and humidity data; the working condition features include load mutation features and voltage stress distribution data;

[0038] A data processing module is configured to filter out noise and extract associated features from the original monitoring data set, and obtain a key parameter set through dimension reduction processing; the associated features include the response coefficient of insulation resistance and environmental temperature, the sensitivity coefficient of insulation resistance and environmental humidity, and the coupling correlation degree of insulation resistance and load mutation;

[0039] A prediction analysis module is configured to input the key parameter set into a trained dynamic correlation model to obtain a prediction result of the change trend of the insulation resistance value over time and the degradation degree quantization; the dynamic correlation model uses a long short-term memory network to analyze the mapping relationship between parameter dynamic correlation and insulation degradation degree;

[0040] The early warning diagnosis module is configured for detecting mutation characteristics and judging real deterioration of parameters exceeding the preset threshold based on the prediction result, to obtain early warning fault characteristics, which include mutation point time sequence information, deterioration mode coding, fault position coordinates and parameter abnormal values.

[0041] In a third aspect, the present application also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the foregoing method when executing the computer program.

[0042] In a fourth aspect, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the foregoing method.

[0043] The low-voltage motor insulation resistance online monitoring and early warning method, system, computer device and storage medium described above first acquire multi-dimensional operation data including insulation resistance, environmental parameters (temperature and humidity data) and working condition characteristics (load mutation characteristics and voltage stress distribution data) through a data acquisition link to form an original monitoring data set. Then, the original monitoring data set is preprocessed: noise filtering technology is used to reduce data interference, response coefficients of insulation resistance and environmental temperature, sensitive coefficients of insulation resistance and environmental humidity, coupling correlation degrees of insulation resistance and load mutation and other correlation characteristics are extracted, and then core information is focused through dimension reduction processing to obtain a key parameter set. Subsequently, the key parameter set is input into a dynamic correlation model constructed based on a long short-term memory network, the model learns the mapping relationship between parameter dynamic correlation and insulation deterioration degree, and outputs the prediction result of the change trend of the insulation resistance value over time and the quantification of the deterioration degree. Finally, early warning analysis is carried out based on the prediction result: mutation characteristic detection is carried out on parameters exceeding the preset threshold, and real deterioration judgment process is combined to generate early warning fault characteristics including mutation point time sequence information, deterioration mode coding, fault position coordinates and parameter abnormal values. This method integrates multi-dimensional data of environment and working conditions, optimizes data quality through noise filtering and feature extraction, solves the problems of large data interference and unclear characteristics in traditional monitoring; the dynamic correlation model constructed by using the long short-term memory network realizes accurate prediction of the insulation resistance trend and the deterioration degree, and overcomes the defect of insufficient real-time performance of offline detection; structured early warning fault characteristics are generated through mutation characteristic detection and deterioration judgment, which improves the accuracy and pertinence of fault early warning, provides reliable technical support for real-time monitoring, early fault diagnosis and operation and maintenance decision of low-voltage motor insulation state, and effectively reduces equipment downtime and safety risks caused by insulation faults. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the embodiments or the related art description will be briefly introduced. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor based on these drawings.

[0045] Figure 1 The flow chart of the low-voltage motor insulation resistance online monitoring and early warning method provided by the embodiment of the present application is shown in the figure.

[0046] Figure 2 The structural block diagram of the low-voltage motor insulation resistance online monitoring and early warning system provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the present application more clear and explicit, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0048] In one of the embodiments, as shown in the figure, the present application provides a low-voltage motor insulation resistance online monitoring and early warning method, which can include the following steps: Figure 1

[0049] Step S101, acquiring multi-dimensional operation data including insulation resistance, environmental parameters and working condition characteristics to obtain an original monitoring data set; the environmental parameters include temperature and humidity data; the working condition characteristics include load mutation characteristics and voltage stress distribution data.

[0050] Specifically, through the deployment of sensing devices in the low-voltage motor and the operating environment, multi-dimensional operation data is collected in real time, including the insulation resistance value of the motor body, environmental parameters (such as temperature data and humidity data around the motor) and working condition characteristic data (such as current / voltage fluctuation characteristics when the load mutates, distribution data of voltage stress in the motor winding).

[0051] Step S102, noise filtering and associated feature extraction are performed on the original monitoring data set, and key parameter set is obtained through dimension reduction processing; the associated features include the response coefficient of insulation resistance and environmental temperature, the sensitive coefficient of insulation resistance and environmental humidity, and the coupling correlation degree of insulation resistance and load mutation.

[0052] ​First, the noise filtering algorithm (such as wavelet transform) is used to eliminate high-frequency interference signals in the data and improve the data quality; then the correlation features are extracted from the denoised multi-dimensional data, including the response coefficient of the insulation resistance to the change of the environmental temperature (reflecting the influence degree of temperature on the insulation resistance), the sensitive coefficient of the insulation resistance to the environmental humidity (quantifying the correlation strength of humidity and insulation performance), and the coupling correlation degree of the insulation resistance and load mutation (characterizing the effect of load fluctuation on the insulation state), and these features are integrated into a feature matrix; finally, through dimension reduction methods such as principal component analysis, redundant information is removed, and the key parameter set focusing on the core characteristics is obtained.

[0053] In step S103, the key parameter set is input into the trained dynamic correlation model to obtain the prediction results of the change trend of the insulation resistance value over time and the degradation degree quantification; the dynamic correlation model uses a long short-term memory network to analyze the mapping relationship between parameter dynamic correlation and insulation degradation degree.

[0054] The key parameter set obtained by preprocessing is input into the trained dynamic correlation model, which is constructed based on a long short-term memory network (LSTM) and can capture the dynamic correlation relationship between parameters and time sequence characteristics. By learning the mapping rules of parameter changes and insulation degradation in historical data, the model outputs the prediction results of the change trend of the insulation resistance value over time. At the same time, the model quantitatively calculates the insulation degradation degree based on the dynamic change characteristics of the parameters, and finally forms the prediction results containing the resistance trend and the degradation degree quantification index.

[0055] In step S104, based on the prediction results, the parameters exceeding the preset threshold are detected for mutation characteristics and real degradation, and the early warning fault features are obtained; the early warning fault features include mutation point time sequence information, degradation mode coding, fault position coordinates, and parameter abnormal values.

[0056] Based on the prediction results, the parameters (such as insulation resistance abnormal decrease value, degradation degree exceeding limit value, etc.) exceeding the preset threshold are analyzed: the isolated forest and other mutation feature detection algorithms are used to extract the parameter mutation points, and the specific time of mutation occurrence, mutation amplitude, and fluctuation characteristics of the associated parameters are determined; the real degradation is judged by combining the pre-established degradation mode library, and the degradation type and confidence are confirmed by feature matching; if the degradation confidence exceeds the judgment threshold, the fault components and degradation level are located by combining the device structure parameters, and finally the early warning fault features containing the mutation point time sequence information, degradation mode coding, fault position coordinates, and parameter abnormal values are generated, providing structured information for fault warning and operation and maintenance decision-making.

[0057] The low-voltage motor insulation resistance online monitoring and early warning method first acquires multi-dimensional operation data including insulation resistance, environmental parameters (temperature and humidity data) and working condition characteristics (load mutation characteristics and voltage stress distribution data) through a data acquisition link to form an original monitoring data set. Then, the original monitoring data set is preprocessed: noise filtering technology is used to reduce data interference, response coefficients of insulation resistance and environmental temperature, sensitive coefficients of insulation resistance and environmental humidity, coupling correlation degrees of insulation resistance and load mutation are extracted, and core information is focused through dimension reduction processing to obtain a key parameter set. Subsequently, the key parameter set is input into a dynamic correlation model constructed based on a long short-term memory network. The model learns the mapping relationship between parameter dynamic correlation and insulation deterioration degree, and outputs the change trend of insulation resistance value over time and the prediction result of the deterioration degree quantization. Finally, based on the prediction result, early warning analysis is carried out: parameters exceeding the preset threshold are subjected to mutation feature detection, and real deterioration judgment process is combined to generate early warning fault features including mutation point time sequence information, deterioration mode code, fault position coordinates and parameter abnormal value. This method integrates multi-dimensional data of environment and working conditions, optimizes data quality through noise filtering and feature extraction, solves the problems of large data interference and unclear features in traditional monitoring; the dynamic correlation model constructed by using the long short-term memory network realizes accurate prediction of the insulation resistance trend and the deterioration degree, and overcomes the defect of insufficient real-time performance of offline detection; the structured early warning fault features are generated through mutation feature detection and deterioration judgment, which improves the accuracy and pertinence of fault early warning, provides reliable technical support for real-time monitoring, early fault diagnosis and operation and maintenance decision of low-voltage motor insulation state, and effectively reduces equipment downtime and safety risks caused by insulation faults.

[0058] In one of the embodiments, the noise filtering and correlation feature extraction are performed on the original monitoring data set, and the key parameter set is obtained through dimension reduction processing, which can include the following steps:

[0059] Step S201, time sequence data signals in the original monitoring data set are acquired; the time sequence data signals include time sequence record data of insulation resistance value, temperature, humidity, load mutation and voltage stress distribution.

[0060] Step S202, wavelet transform is used to perform noise filtering processing on the time sequence data signals to obtain denoised multi-dimensional time sequence signals.

[0061] Preferably, the time sequence signals are decomposed into wavelet components of different frequencies, high-frequency noise components (such as sensor interference, electromagnetic noise, etc.) and low-frequency effective signals (real change trend of parameters) are separated, and the denoised multi-dimensional time sequence signals are obtained by reconstructing the low-frequency components.

[0062] Step S203, correlation features of insulation resistance respectively associated with environmental parameters and working condition characteristics are extracted from the denoised multi-dimensional time sequence signals to generate a feature matrix.

[0063] For environmental parameters (temperature, humidity), the response coefficient of insulation resistance change with temperature (ratio of resistance change to temperature change) and the sensitive coefficient of insulation resistance change with humidity (ratio of resistance change rate to humidity change rate) are calculated; for working condition characteristics (load mutation, voltage stress distribution), the coupling correlation degree of insulation resistance and load mutation (quantifying correlation strength through Pearson correlation coefficient or mutual information) and the correlation characteristics with voltage stress distribution are calculated. These quantified correlation characteristics are integrated by dimension to form a structured feature matrix, which fully retains the key influencing factors of insulation state.

[0064] Step S204, using principal component analysis to reduce the dimension of the feature matrix, and obtaining a key parameter set focusing on the core characteristics of the insulation state.

[0065] By calculating the covariance matrix of the feature matrix, the principal component component with the largest variance contribution (i.e. the feature combination that can explain the largest variation degree of data) is extracted, and the redundant components with low contribution degree are removed, and finally the dimension-reduced key parameter set is obtained. These key parameters focus on the core characteristics of the insulation state (such as temperature sensitivity, humidity coupling, load correlation, etc.), while retaining the core information, reducing the data dimension, reducing the computational complexity of the subsequent model, and improving the analysis efficiency.

[0066] First, the time series data signals are extracted from the original monitoring data set, covering the time series record data of insulation resistance value, temperature, humidity, load mutation and voltage stress distribution, ensuring to contain the dynamic change information related to the insulation state. Then, the extracted time series data signals are processed by wavelet transform for noise filtering, by decomposing the high-frequency noise components of the signals and reconstructing the effective signals, to obtain the denoised multi-dimensional time series signals, and improve the signal-to-noise ratio of the data. Based on the denoised multi-dimensional time series signals, further extract the correlation characteristics of insulation resistance respectively with environmental parameters (temperature, humidity), working condition characteristics (load mutation, voltage stress distribution), including response coefficient, sensitive coefficient, coupling correlation degree, etc., integrate these features to form a feature matrix. Finally, the feature matrix is processed by principal component analysis to retain the principal component components reflecting the core correlation law of the insulation state and remove redundant information, and obtain a key parameter set focusing on the core characteristics of the insulation state.

[0067] The embodiment effectively improves data quality and analysis efficiency through targeted data processing steps: the noise filtering processing of wavelet transform eliminates the interference signals in the original data, providing a reliable foundation for subsequent feature extraction; the extraction of correlation features quantifies the internal relationship between insulation resistance and environmental and working condition factors, avoiding the omission of key influencing factors; the dimension reduction processing of principal component analysis simplifies the data dimension, reduces the subsequent model calculation complexity, and retains the core information of insulation state evaluation. The accurate transformation from the original monitoring data to the core feature parameters is realized, ensuring the accuracy and reliability of the monitoring and early warning results.

[0068] In one embodiment, the key parameter set is input into the trained dynamic correlation model to obtain the prediction results of the change trend of insulation resistance value over time and the quantification of the degradation degree, which can include the following steps:

[0069] Step S301, extracting insulation resistance related feature data and environmental humidity data from the key parameter set; the feature data includes voltage, current and temperature parameters; the environmental humidity data has a high correlation with the insulation resistance.

[0070] Focusing on the core input data extraction in the key parameter set, two types of data are explicitly selected: one is insulation resistance related feature data, specifically including voltage parameters (such as working voltage, voltage fluctuation value), current parameters (such as load current, current harmonic component) and temperature parameters (such as winding temperature, ambient temperature) in motor operation, which directly reflect the electrical operating state of the motor; the other is environmental humidity data, which is extracted separately as a key correlation parameter because humidity has a significant correlation with the physical properties of insulation resistance (such as high humidity leading to moisture degradation of insulation materials).

[0071] Step S302, inputting the feature data into the trained dynamic correlation model to obtain the trend prediction value of the insulation resistance change over time; the dynamic correlation model is constructed using a long short-term memory network.

[0072] The trained dynamic correlation model is constructed based on a long short-term memory network (LSTM), which has the core advantage of effectively capturing long-term dependencies in time series data. The model learns the dynamic correlation rules of voltage, current, temperature and other feature parameters and insulation resistance changes in historical data, and outputs the trend prediction value of the insulation resistance change over time, i.e. the numerical change curve of the insulation resistance in the future period of time, providing a basic trend basis for subsequent degradation analysis.

[0073] Step S303, performing feature importance analysis on the environmental humidity data and the feature data using a random forest algorithm to obtain the influence weight of humidity on insulation resistance degradation.

[0074] The algorithm calculates the contribution of the humidity parameter in the insulation resistance degradation prediction by constructing multiple decision trees, and finally outputs the influence weight of humidity on the insulation resistance degradation (the value range is [0, 1]). The weight quantifies the influence intensity of the humidity factor in the insulation degradation process, and the higher the weight, the more significant the influence of humidity on the current insulation state, providing a quantitative basis for model optimization.

[0075] In step S304, the parameters of the dynamic correlation model are adjusted according to the influence weight to obtain an optimized trend prediction value, and an insulation degradation index is calculated using a formula; the insulation degradation index reflects the aging state of the insulation material.

[0076] Preferably, the parameters of the dynamic correlation model (such as the hidden layer weight of LSTM and the length of the time window) are adjusted so that the model focuses more on the learning of high-influence-weight parameters, thereby generating an optimized trend prediction value. At the same time, an insulation degradation index is calculated using a preset formula (fusing factors such as insulation resistance trend deviation and humidity influence weight), which quantifies the aging degree of the insulation material in numerical form (0-1 range), and the higher the value, the more serious the aging state, directly reflecting the degradation level of insulation performance.

[0077] In step S305, abnormal fluctuation features are extracted from the optimized trend prediction value, and the insulation degradation index is updated using the abnormal fluctuation features.

[0078] The abnormal fluctuation features specifically include the instantaneous drop amplitude of the insulation resistance prediction value, the continuous fluctuation frequency, and the deviation amount from the smoothed trend line.

[0079] In step S306, the optimized trend prediction value and the updated insulation degradation index are integrated to obtain the prediction result of the insulation resistance value change and degradation.

[0080] Specifically, first, two types of core data are extracted from the key parameter set: one is the insulation resistance related feature data, including voltage, current and temperature parameters; the other is the environmental humidity data which has a high correlation with the insulation resistance. The extracted feature data is input into a dynamic correlation model based on a long short-term memory network, and through the learning of the model on the dynamic correlation relationship of the parameters, the trend prediction value of the insulation resistance change over time is output. At the same time, the random forest algorithm is used for feature importance analysis on the environmental humidity data and the feature data, and the influence weight of humidity on the insulation resistance degradation is quantitatively obtained; the parameters of the dynamic correlation model are adjusted according to the influence weight to generate an optimized trend prediction value, and an insulation degradation index reflecting the aging state of the insulation material is calculated based on a preset formula. Subsequently, abnormal fluctuation features (such as instantaneous deviation and fluctuation frequency) are extracted from the optimized trend prediction value, and are integrated into the update calculation of the insulation degradation index; finally, the optimized trend prediction value and the updated insulation degradation index are integrated to form a prediction output containing the insulation resistance value change rule and the quantitative result of the degradation degree.

[0081] The embodiment improves the prediction accuracy and the reliability of the degradation evaluation through multi-link optimization design: the targeted extraction of feature data and high correlation humidity data ensures the effectiveness of the model input; the long short-term memory network captures the time series characteristics combined with the feature importance analysis of the random forest, which realizes the dynamic optimization of the model parameters, so that the trend prediction value is more consistent with the actual operating state; the calculation of the insulation degradation index combines the humidity influence weight and the abnormal fluctuation characteristics, and comprehensively quantifies the insulation aging state, avoiding the limitations of single parameter evaluation.

[0082] In one embodiment, the insulation degradation index can be calculated by the following formula:

[0083]

[0084] Wherein, D represents the insulation degradation index, R pred (t) represents the t time insulation resistance optimized prediction value output by the dynamic correlation model, R base represents the insulation resistance reference value under normal state of the equipment, which is calibrated by historical health data, W humid represents the influence weight of humidity on insulation degradation calculated by the random forest algorithm, the value range is [0, 1], ΔH(t) represents the over-standard quantity of environment humidity H(t) relative to normal humidity threshold H thresh , ΔH(t) = H(t) - H thresh , F fluct (t) represents the abnormal fluctuation characteristic quantization value extracted from the optimized trend prediction value, which is obtained by normalizing the deviation variance of the prediction value and the smooth trend line, the value range is [0, 1], α, β, γ respectively represent the weight coefficients of insulation resistance trend, humidity influence and abnormal fluctuation, α+β+γ=1, which are determined by historical fault data training optimization.

[0085] The calculation formula of the embodiment realizes the accurate quantitative evaluation of the insulation state through the scientific fusion of multi-dimensional factors: the core reflects the degradation trend of the resistance itself through the deviation proportion of the insulation resistance optimized prediction value and the reference value; the product term of humidity influence weight and humidity over-standard quantity quantifies the acceleration effect of high humidity environment on insulation degradation, which reflects the targeted influence of environmental factors; the abnormal fluctuation characteristic quantization value is included to capture the unstable change of insulation resistance and enhance the sensitivity to potential sudden degradation; and the weight coefficients determined by historical fault data training optimization ensure the reasonable allocation of the contribution of each factor. The overall formula comprehensively integrates the insulation itself characteristics, environmental correlation influence and dynamic fluctuation characteristics, so that the calculation result can objectively reflect the real aging state of the insulation material, and provides a reliable quantitative basis for subsequent degradation judgment and early warning decision, effectively improving the accuracy and scientificity of the insulation state evaluation.

[0086] In one embodiment, the parameter exceeding the preset threshold is subjected to mutation feature detection and real degradation judgment based on the prediction result to obtain a pre-warning failure feature, which can include the following steps:

[0087] Step S401, extracting a pre-warning parameter exceeding a preset pre-warning threshold from the prediction result; the pre-warning parameter includes an insulation state parameter abnormal value, an environment-related parameter over-limit value, and a working condition feature abnormal value.

[0088] Preferably, the insulation state parameter abnormal value includes insulation resistance value abnormal drop data, insulation degradation index over-limit value, and other indicators directly reflecting insulation performance; the environment-related parameter over-limit value covers insulation resistance and temperature response coefficient over-limit value, humidity sensitivity coefficient abnormal value, and corresponding environment temperature / humidity measured over-limit data; and the working condition feature abnormal value includes coupling correlation abnormal value and load mutation peak over-limit data.

[0089] Step S402, using an isolation forest algorithm to extract a mutation point of the pre-warning parameter to obtain a mutation feature set containing a mutation time, a mutation amplitude, and a correlation parameter fluctuation feature.

[0090] Specifically, the algorithm identifies the time when the parameter value suddenly deviates from the normal trend (mutation time) according to the time sequence change feature of the pre-warning parameter, calculates the deviation amount (mutation amplitude) of the parameter value at this time from the historical mean value, and synchronously records the fluctuation features (such as fluctuation amplitude and frequency change) of the correlation parameters (such as humidity and load) before and after the mutation. The final output mutation feature set integrates these information in a structured form, accurately locates the key nodes of the abnormal change of the insulation state, and provides time and numerical dimension basis for degradation reason analysis.

[0091] Step S403, querying a pre-established degradation mode library based on the mutation feature set to obtain a real degradation pre-warning result including a degradation type and a confidence level through feature matching degree calculation; the degradation mode library stores the mapping relationship between the mutation features and the insulation degradation types in historical failure cases.

[0092] The mode library stores the mapping relationship between various types of mutation features and insulation degradation types in historical failure cases (such as “humidity sensitivity coefficient sudden rise + insulation resistance slow drop” corresponding to insulation moisture degradation). Through a feature matching degree calculation formula (combining feature deviation, historical fluctuation standard deviation, and feature importance weight), the similarity between the current mutation feature and the historical mode is quantified, and the degradation type (such as partial discharge, insulation aging, etc.) with the highest matching degree and the corresponding confidence level (value range 0-1) are output. The higher the confidence level, the more reliable the matching between the current feature and the historical degradation mode, realizing accurate mapping from data features to degradation types.

[0093] Step S404, if the degradation confidence in the real degradation early warning result exceeds the determination threshold, the fault positioning result including the fault component and the degradation degree level is determined based on the degradation mode library combined with the equipment structure parameters, and the early warning fault feature is generated.

[0094] Combined with the fault component records of the historical cases in the degradation mode library, the typical fault location corresponding to the current degradation type is matched, and the fault component (such as the stator winding and the lead insulation) is further accurately positioned combined with the equipment structure parameters (such as the winding distribution and the insulation material layout); at the same time, the degradation degree level (such as slight degradation, moderate degradation and severe degradation) is determined according to the mutation amplitude and the degradation index. The finally generated early warning fault feature integrates the mutation point timing information, the degradation mode code (corresponding to the specific degradation type), the fault location coordinates and the parameter abnormal value, forms a structured fault description, and provides clear fault positioning and severity information for operation and maintenance decision.

[0095] First, the early warning parameters exceeding the preset early warning threshold are extracted from the prediction result, including the insulation state parameter abnormal value (such as abnormal decrease of insulation resistance, degradation index exceeding limit), the environment related parameter exceeding value (such as abnormal temperature and humidity sensitive coefficient, environment parameter measured exceeding standard) and the working condition feature abnormal value (such as abnormal load coupling correlation degree, load mutation peak exceeding standard). For these early warning parameters, the isolated forest algorithm is used for mutation point extraction, the specific time of parameter mutation, the mutation amplitude and the fluctuation characteristics of the related parameters (such as temperature and humidity, load) are determined through the isolated analysis of abnormal data by the algorithm, and a structured mutation feature set is formed. Based on the set, the degradation mode library (storing the mapping relationship between the mutation features and the degradation types in the historical faults) is queried, the corresponding degradation type and confidence are output through feature matching degree calculation, and the real degradation early warning result is obtained; if the confidence exceeds the determination threshold, the fault component and the degradation degree level are located combined with the degradation mode library and the equipment structure parameters (such as winding layout, insulation material distribution), and the early warning fault feature including the mutation point timing information, the degradation mode code, the fault location coordinates and the parameter abnormal value is finally generated.

[0096] The present embodiment improves the accuracy and pertinence of fault early warning through multi-link precise analysis: the classification extraction of early warning parameters ensures the comprehensive coverage of abnormal signals, avoiding the omission of key degradation factors; the mutation point extraction of isolated forest algorithm realizes the accurate positioning of abnormal change nodes, providing time and numerical basis for degradation reason tracing; the feature matching of degradation mode library combined with confidence judgment effectively distinguishes the real degradation and interference signals, reducing the false positive rate; the fault positioning based on the equipment structure parameters realizes the accurate mapping from the degradation type to the specific component.

[0097] In one of the embodiments, the feature matching degree can be calculated by the following formula:

[0098]

[0099] wherein, M represents the feature matching degree, x i represents the i-th feature parameter in the current set of mutation features, y i represents the i-th corresponding feature parameter of the historical failure mode in the degradation mode library, σ i represents the historical fluctuation standard deviation of the i-th feature parameter, ω i represents the importance weight of the i-th feature parameter, which is obtained by training the historical data through a random forest algorithm, and the value range is [0, 1], and ∑ω i > 0.

[0100] In this embodiment, the deviation of the current feature parameter and the corresponding parameter of the historical mode is calculated, and the historical fluctuation standard deviation of the parameter is normalized, which effectively eliminates the influence of the dimension difference of different features. The feature importance weight obtained by the random forest algorithm is introduced, so that the key features (such as insulation resistance mutation amplitude and humidity sensitivity coefficient) have higher weight in matching, which improves the pertinence of matching. Through weighted summation and weight normalization processing, the result is constrained in a reasonable range, and it is ensured that the matching degree value can intuitively reflect the feature similarity. The feature deviation, historical fluctuation characteristics and feature importance are comprehensively integrated, which significantly improves the accuracy of feature matching, provides reliable quantitative basis for real degradation type recognition and confidence evaluation, and effectively supports the accurate early warning and diagnosis of insulation failure.

[0101] In one of the embodiments, after generating the early warning failure feature, the following steps can be further included:

[0102] Step S501, taking the early warning failure feature as a training sample, a failure classification model is constructed through a random forest algorithm, and the failure type and prediction probability are output; the failure classification model adopts multi-decision tree ensemble learning, and the feature importance is calculated through feature split gain.

[0103] Preferably, the learning results of multiple independent decision trees are integrated to improve the classification stability, and each decision tree calculates the contribution of different features to failure classification based on feature split gain (such as Gini index or information gain) to obtain the importance weight of each feature (such as higher weight of key features such as insulation resistance mutation amplitude and humidity sensitivity coefficient). After the model training is completed, the corresponding failure type (such as insulation damp, partial discharge, etc.) and prediction probability (value range 0-1) can be output for the input feature data, and the higher the prediction probability, the stronger the credibility of the classification result.

[0104] Step S502, using the fault classification model to perform feature matching on the equipment real-time operation log, extracting potential fault patterns, and generating an extended fault feature set containing new feature dimensions and association rules; the potential fault patterns include feature combinations that have not triggered the warning threshold but have a degradation trend.

[0105] Through the feature matching function of the model, the similarity of log data and known fault features is compared, and feature combinations that have not triggered the warning threshold but have a degradation trend (such as a 2% monthly decline in insulation resistance and a humidity consistently above 60% RH) are identified. Such combinations constitute potential fault patterns. Based on these patterns, new feature dimensions (such as the timing correlation strength between features and multi-parameter coupling coefficients) are further extracted, and association rules (such as "when the load fluctuation frequency > 5 times / hour and the winding temperature > 70°C, the insulation degradation risk increases by 30%") are summarized. Finally, the extended fault feature set containing new dimensions and rules is integrated, breaking through the limitations of traditional threshold warnings.

[0106] Step S503, based on the degradation sensitivity weights of each feature in the extended fault feature set, optimize the collection frequency of equipment monitoring parameters and the setting of warning thresholds, and obtain the adaptive dynamic warning and monitoring strategy.

[0107] Optimizing the monitoring strategy based on the degradation sensitivity weights of each feature in the extended fault feature set: for high sensitivity features (such as insulation resistance and humidity), increase the collection frequency (such as from 1 time per hour to 1 time every 10 minutes) to ensure that the dynamic changes of key parameters are accurately captured; for low sensitivity features (such as ambient pressure), maintain regular collection frequency to reduce data redundancy. At the same time, according to the distribution rules of new features in the extended feature set (such as the normal fluctuation range of humidity and load coupling features), dynamically adjust the upper and lower limits of the warning threshold (such as the humidity association threshold is automatically corrected according to the seasonal baseline value), and finally form a dynamic warning and monitoring strategy that adapts to the real-time running state of the equipment, improving the accuracy of the warning and the efficiency of resource utilization.

[0108] First, the early warning fault features are taken as training samples, and a fault classification model is constructed by using a random forest algorithm: through ensemble learning of multiple decision trees, the importance of each feature is calculated by using feature split gain, the model outputs the fault type and the corresponding prediction probability. Subsequently, the fault classification model is used to match the features of the real-time operation log of the equipment, and the feature combination with a deterioration trend (such as the coupled features of a slow decline in insulation resistance and a continuously high humidity) that does not trigger the early warning threshold is identified, the potential fault mode is extracted, and then an extended fault feature set containing new feature dimensions (such as the collaborative change rate between features) and association rules (such as “when the coupling coefficient of load fluctuation and temperature rise exceeds 0.6, the deterioration accelerates”) is generated. Finally, based on the deterioration sensitivity weight (determined by feature importance analysis) of each feature in the extended fault feature set, the collection frequency of the equipment monitoring parameters (high sensitivity features improve the collection density) and the early warning threshold setting (dynamically adjusted according to the distribution of new features) are optimized to form a dynamic early warning and monitoring strategy that adapts to the running state of the equipment.

[0109] The fault classification model based on random forest of the embodiment uses the advantage of ensemble learning to improve the accuracy of fault type identification, and the feature importance analysis provides a basis for key parameter focusing; the extraction of potential fault mode breaks through the limitation of traditional threshold early warning, can capture early deterioration trend, and expands the time window of fault early warning; the optimization of dynamic monitoring strategy makes the resource allocation more accurate, the high-frequency collection of high-sensitivity features ensures the capture of key information, and the adaptive adjustment of threshold reduces the risk of false alarm and missed alarm. The intelligent level and early warning timeliness of low-voltage motor insulation monitoring are significantly improved, which provides technical support for efficient operation and maintenance of the whole life cycle of the equipment.

[0110] In one of the embodiments, as shown in Figure 2 The application also provides a low-voltage motor insulation resistance online monitoring and early warning system, which can include:

[0111] The data acquisition module 601 is used to acquire multi-dimensional operation data including insulation resistance, environmental parameters and working condition features, and obtain an original monitoring data set; the environmental parameters include temperature and humidity data; the working condition features include load mutation features and voltage stress distribution data;

[0112] The data processing module 602 is used to filter noise and extract associated features from the original monitoring data set, and obtain a key parameter set through dimension reduction processing; the associated features include the response coefficient of insulation resistance and environmental temperature, the sensitivity coefficient of insulation resistance and environmental humidity, and the coupling correlation degree of insulation resistance and load mutation;

[0113] The prediction analysis module 603 is configured to input the key parameter set into the trained dynamic correlation model to obtain a prediction result of a change trend of the insulation resistance value over time and a quantified degree of deterioration; and the dynamic correlation model adopts a long short-term memory network to analyze a mapping relationship between parameter dynamic correlation and the degree of insulation deterioration.

[0114] The early warning diagnosis module 604 is configured to perform mutation feature detection and real deterioration judgment on parameters exceeding a preset threshold based on the prediction result to obtain early warning fault features; and the early warning fault features include mutation point time sequence information, deterioration mode coding, fault position coordinates and parameter abnormal values.

[0115] The above low-voltage motor insulation resistance online monitoring and early warning system realizes insulation state monitoring and early warning through the cooperation of the four core modules: the data acquisition module acquires multi-dimensional operating data, including motor insulation resistance values, environmental parameters (temperature, humidity) and working condition characteristics (load mutation features, voltage stress distribution data), and aggregates to form an original monitoring data set; the data processing module pre-processes the original data set, eliminates interference signals through noise filtering, extracts correlation features such as response coefficients of insulation resistance and temperature, sensitive coefficients of humidity, and coupling correlation degrees of load mutation, and then processes through dimension reduction to obtain a key parameter set focusing on insulation core characteristics; the prediction analysis module inputs the key parameter set into a dynamic correlation model based on a long short-term memory network, and outputs a change trend of the insulation resistance over time and a quantified degree of deterioration; the model realizes accurate prediction by learning parameter dynamic correlation rules; and the early warning diagnosis module performs mutation feature detection (such as mutation time and amplitude) on parameters exceeding the limit based on the prediction result, completes real deterioration judgment in combination with a deterioration mode library, and finally generates early warning fault features including mutation point time sequence, deterioration mode coding, fault position coordinates and parameter abnormal values.

[0116] The data acquisition module of the device ensures comprehensive coverage of multi-dimensional key parameters, providing complete input for analysis; the data processing module optimizes data quality through noise filtering and feature extraction, laying a reliable foundation for model prediction; the prediction analysis module accurately captures the insulation resistance trend and the degree of deterioration by using the time sequence analysis advantage of the long short-term memory network, breaking through the real-time limitations of traditional offline detection; and the early warning diagnosis module realizes accurate fault positioning through mutation detection and pattern matching, and outputs structured early warning features providing clear basis for operation and maintenance decisions. The real-time, accuracy and intelligent level of low-voltage motor insulation state monitoring are improved, the equipment downtime risk caused by insulation failure is effectively reduced, and technical support is provided for safe and stable operation of equipment.

[0117] It should be understood that although the steps in the flowcharts related to the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other sequences. Moreover, at least some of the steps in the flowcharts related to the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of the steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or stages.

[0118] In an embodiment, a computer device is provided, comprising a memory and a processor, the memory stores a computer program, and the processor implements the steps of the low-voltage motor insulation resistance online monitoring and early warning method, system, device and medium as described above when executing the computer program.

[0119] In an embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the steps of the above method embodiments.

[0120] For the device embodiment, since it basically corresponds to the method embodiment, the relevant part can be referred to the part of the method embodiment. The device embodiments described above are only schematic, and the components described as separate components can or can not be physically separate, and the components displayed as a unit can or can not be a physical unit, i.e. they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present disclosure according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0121] The above-described embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the application. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application.

Claims

1. A method for online monitoring and early warning of insulation resistance of low-voltage motors, characterized in that, The method includes: The raw monitoring dataset is obtained by acquiring multidimensional operational data including insulation resistance, environmental parameters, and operating condition characteristics; the environmental parameters include temperature and humidity data; the operating condition characteristics include load change characteristics and voltage stress distribution data. The original monitoring dataset is subjected to noise filtering and correlation feature extraction, and a key parameter set is obtained through dimensionality reduction. The correlation features include the response coefficient of the insulation resistance to ambient temperature, the sensitivity coefficient to ambient humidity, and the coupling correlation degree with load mutation. The key parameter set is input into the trained dynamic correlation model to obtain the prediction results of the insulation resistance value changing over time and the degree of degradation; the dynamic correlation model uses a long short-term memory network to analyze the mapping relationship between the dynamic correlation of parameters and the degree of insulation degradation. Based on the prediction results, mutation feature detection and actual degradation judgment are performed on parameters exceeding the preset threshold to obtain early warning fault features; the early warning fault features include mutation point time sequence information, degradation mode encoding, fault location coordinates and parameter anomaly values.

2. The method according to claim 1, characterized in that, The process involves noise filtering and correlation feature extraction of the original monitoring dataset, followed by dimensionality reduction to obtain a set of key parameters, including: Acquire time-series data signals from the original monitoring dataset; the time-series data signals include time-series records of insulation resistance values, temperature, humidity, load abrupt changes, and voltage stress distribution; The time-series data signal is subjected to wavelet transform for noise filtering to obtain a denoised multidimensional time-series signal; The insulation resistance is associated with environmental parameters and operating conditions respectively from the denoised multidimensional time-series signal to generate a feature matrix; Principal component analysis was used to reduce the dimensionality of the feature matrix to obtain a set of key parameters focusing on the core characteristics of the insulation state.

3. The method according to claim 1, characterized in that, The step of inputting the key parameter set into the trained dynamic correlation model to obtain a quantitative prediction result of the insulation resistance value changing over time and the degree of degradation includes: Insulation resistance-related feature data and environmental humidity data are extracted from the set of key parameters; the feature data includes voltage, current and temperature parameters; the environmental humidity data has a high correlation with the insulation resistance. The feature data is input into the trained dynamic correlation model to obtain the predicted value of the insulation resistance changing over time; the dynamic correlation model is constructed using a long short-term memory network. The random forest algorithm was used to analyze the feature importance of the environmental humidity data and feature data to obtain the influence weight of humidity on the degradation of insulation resistance. The parameters of the dynamic correlation model are adjusted according to the influence weights to obtain the optimized trend prediction value, and the insulation degradation index is calculated using the formula; the insulation degradation index reflects the aging state of the insulation material. Extract abnormal fluctuation features from the optimized trend prediction values, and update the insulation degradation index using the abnormal fluctuation features; By integrating the optimized trend prediction value and the updated insulation degradation index, the prediction results of insulation resistance value change and degradation are obtained.

4. The method according to claim 3, characterized in that, The insulation degradation index is calculated using the following formula: Where D represents the insulation degradation index, R pred (t) represents the optimized predicted value of insulation resistance at time t output by the dynamic correlation model, R base This represents the reference value of insulation resistance under normal equipment conditions, calibrated using historical health data. (W) humid The value represents the weight of humidity on insulation degradation calculated by the random forest algorithm, ranging from [0,1]. ΔH(t) represents the ambient humidity H(t) at time t relative to the normal humidity threshold H. thresh The overscalar quantity, ΔH(t)=H(t)-H thresh F fluct (t) represents the quantified value of abnormal fluctuation characteristics extracted from the optimized trend prediction value. It is obtained by normalizing the variance of the deviation between the prediction value and the smooth trend line and takes a value range of [0,1]. α, β, and γ represent the weight coefficients of insulation resistance trend, humidity influence, and abnormal fluctuation, respectively. α+β+γ=1 and are determined by training and optimization using historical fault data.

5. The method according to claim 1, characterized in that, The step of detecting abrupt changes and determining actual degradation of parameters exceeding a preset threshold based on the prediction results to obtain early warning fault characteristics includes: The prediction results are used to extract warning parameters that exceed the preset warning threshold; the warning parameters include abnormal values ​​of insulation status parameters, excessive values ​​of environmental related parameters, and abnormal values ​​of operating condition characteristics. The isolated forest algorithm is used to extract mutation points from the warning parameters, resulting in a mutation feature set that includes the mutation time, mutation magnitude, and fluctuation characteristics of related parameters. Based on the set of mutation features, a pre-established degradation pattern library is queried, and a real degradation warning result including degradation type and confidence level is obtained through feature matching degree calculation; the degradation pattern library stores the mapping relationship between mutation features and insulation degradation types in historical fault cases; If the degradation confidence level in the actual degradation warning result exceeds the judgment threshold, then based on the degradation pattern library and the equipment structural parameters, the fault location result including the faulty component and the degradation level is determined, and a warning fault feature is generated.

6. The method according to claim 5, characterized in that, The feature matching degree is calculated using the following formula: Where M represents the feature matching degree, x i y represents the i-th feature parameter in the current mutation feature set. i σ represents the i-th corresponding feature parameter of the historical failure mode in the degradation mode library. i ω represents the historical standard deviation of the i-th feature parameter. i Let represent the importance weight of the i-th feature parameter, obtained by training on historical data using the random forest algorithm, with a value range of [0,1], and ∑ω i >0.

7. The method according to claim 5, characterized in that, After generating the early warning fault characteristics, the method further includes: Using the aforementioned early warning fault features as training samples, a fault classification model is constructed using the random forest algorithm to output the fault type and prediction probability; the fault classification model employs multi-decision tree ensemble learning, and calculates feature importance through feature split gain; The fault classification model is used to perform feature matching on the real-time operation logs of the equipment, extract potential fault patterns, and generate an extended fault feature set containing new feature dimensions and association rules; the potential fault patterns include feature combinations that have not triggered the warning threshold but show a deterioration trend. Based on the degradation sensitivity weights of each feature in the extended fault feature set, the collection frequency of equipment monitoring parameters and the setting of early warning thresholds are optimized to obtain an appropriate dynamic early warning and monitoring strategy.

8. A low-voltage motor insulation resistance online monitoring and early warning system, characterized in that, The system includes: The data acquisition module is used to acquire multi-dimensional operational data, including insulation resistance, environmental parameters, and operating condition characteristics, to obtain the raw monitoring dataset; the environmental parameters include temperature and humidity data; the operating condition characteristics include load change characteristics and voltage stress distribution data. The data processing module is used to perform noise filtering and correlation feature extraction on the original monitoring dataset, and to obtain a set of key parameters through dimensionality reduction; the correlation features include the response coefficient of the insulation resistance to ambient temperature, the sensitivity coefficient to ambient humidity, and the coupling correlation degree with load mutation. The predictive analysis module is used to input the key parameter set into the trained dynamic correlation model to obtain the prediction results of the insulation resistance value changing over time and the degree of degradation; the dynamic correlation model uses a long short-term memory network to analyze the mapping relationship between the dynamic correlation of parameters and the degree of insulation degradation. The early warning and diagnosis module is used to detect mutation features and determine actual degradation of parameters that exceed a preset threshold based on the prediction results, and obtain early warning fault features; the early warning fault features include mutation point time sequence information, degradation mode encoding, fault location coordinates and parameter anomaly values.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images