Intelligent state monitoring and management system for direct-current resistance tester

Abstract

The invention discloses an intelligent state monitoring and management system for a direct-current resistance tester, and relates to the field of resistance tester monitoring. At present, the implementation state of a direct current resistance tester is inconvenient to monitor; according to the method, the isolated forest model is constructed through the data set constructed by the direct current resistance tester, preliminary anomaly detection and monitoring of the direct current resistance tester are realized, abnormal point data detected by the isolated forest model are further analyzed and recognized by introducing the Gaussian distribution model, detection optimization of the abnormal point data is improved, and the detection accuracy of the direct current resistance tester is improved. Abnormal points can be accurately identified, and accurate judgment of the state of the direct current resistance tester can be realized; according to the method and the system, abnormal characteristics including resistance abnormity, voltage abnormity and the like are classified accurately, if multiple abnormal characteristics exist at the same time, relevance among the multiple characteristics is calculated, fault factors of the current direct-current resistance tester are determined, false alarm and missing alarm of the abnormity are effectively reduced, and accuracy and robustness of state monitoring of the direct-current resistance tester are improved.

Description

Technical Field

[0001] The invention relates to the field of resistance tester monitoring, and in particular to an intelligent monitoring and management system for a DC resistance tester state. Background Art

[0002] As an important measuring tool, DC resistance tester is widely used in many fields such as electronic equipment, automobile manufacturing, power engineering, etc. DC resistance tester evaluates the performance of equipment or circuit by measuring resistance value to ensure that it operates within the normal range.

[0003] During long-term operation, the DC resistance tester may be affected by various factors such as changes in the external environment, equipment aging, improper operation, and operational failures, which may cause performance abnormalities and affect the accuracy of the test results. Therefore, in order to ensure the reliability of the equipment, test accuracy, and normal operation of the system, how to build a status monitoring solution for a pair of DC resistance testers to achieve rapid monitoring and judgment of the status of the DC resistance tester, which is conducive to the subsequent execution of corresponding management and disposal, is a problem that needs to be solved. Summary of the invention

[0004] In view of the above-mentioned shortcomings of the prior art, the present invention provides an intelligent monitoring and management system for the status of a DC resistance tester, which can effectively solve the problem in the prior art that the DC resistance tester may be affected by various situations during operation, making it inconvenient to implement accurate status monitoring of the DC resistance tester.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: The present invention provides a DC resistance tester state intelligent monitoring and management system, which at least comprises: The data acquisition module is used to obtain the key operating data, environmental parameters, and equipment status data of the DC resistance tester to build a data set; The anomaly detection and analysis module trains the isolation forest model through the data set to pre-monitor the anomaly points of the DC resistance tester to obtain an anomaly point set, and also includes: The introduced Gaussian distribution model is trained based on the set of outliers. By calculating the probability density of outliers, the outliers of the DC resistance tester are re-determined to optimize the detection results of the isolation forest, where: Determine the abnormal type characteristics of the abnormal point based on the detection results; And, a multi-type abnormal analysis module is used to calculate the abnormal amplitude of each abnormal type feature according to the sliding window method to obtain the comprehensive abnormal amplitude. If there are still multiple abnormal type features, a correlation analysis is performed to determine the associated abnormal type features by calculating the Pearson correlation coefficient and define the abnormal factors of the DC resistance tester; The risk equipment prediction module is used to determine the associated DC resistance testers with associated risk states based on the abnormal state of the DC resistance tester when it is determined that there is an abnormal DC resistance tester, so as to form pre-monitoring for them.

[0006] Furthermore, the key operating data include resistance and voltage; the environmental parameters include ambient temperature, ambient humidity and ambient air pressure; and the equipment status data include equipment usage time and equipment remaining life.

[0007] Furthermore, outliers are calculated based on the isolation forest model. The abnormality score of is:

[0008] In the formula, is the path length, is the standardized coefficient used to adjust the score, is the number of samples, is the characteristic dimension; based on Determine abnormal points , and construct an outlier set , is the first abnormal threshold.

[0009] Furthermore, the standardized coefficient Dynamically adjust the initial normalization coefficient by introducing feature dimension, data distribution, and anomaly type Get, where:

[0010] Where: is the number of features, For the The anomaly weight of the feature, For the The abnormality score of each feature, is the standardized coefficient after combining the data distribution;

[0011] Where: is the standardized coefficient after the feature dimension is introduced, is the regulating factor, is the coefficient of variation, obtained from the ratio of the standard deviation to the mean of all features;

[0012] Where: is the initial standardized coefficient, 1 to The harmonic number of is the correction factor; Initial standardized coefficients , is the number of samples.

[0013] Furthermore, the Gaussian distribution model introduced by the outlier set is trained, and the method for calculating the probability density of the outlier points is: Through the collection of outliers Train the Gaussian distribution model and assume that the data is Gaussian distribution; For each outlier Calculate the probability density:

[0014] In the formula, For abnormal points The probability density under the Gaussian distribution model is, For the Gaussian distribution model, The probability density function of is the number of Gaussian distributions in the Gaussian distribution model, For the The mixing coefficient of the Gaussian distribution, For the The mean vector of a Gaussian distribution, For the The covariance matrix of a Gaussian distribution, For the The probability density function of a Gaussian distribution is is the function symbol of Gaussian distribution; Combination Re-determine outliers .

[0015] Furthermore, by introducing the weight fusion method to optimize the judgment results of the isolation forest model and the Gaussian distribution model, we have: Construct the weight fusion algorithm expression, then we have:

[0016] In the formula, is the comprehensive evaluation value, is the weight coefficient, which is used to determine the contribution of the isolation forest and Gaussian model; based on Re-evaluate outliers .

[0017] Furthermore, the abnormal point When they exist, the normal ranges of resistance, ambient temperature, humidity, voltage, pressure, usage time and remaining life are defined respectively. When determining the normal range of remaining life, historical data is used for prediction, as follows: The LSTM model is introduced for prediction. Assuming that each time window has w time steps, N time series samples are obtained after the window slides. For each time window, the corresponding actual remaining life data of voltage, resistance, temperature, humidity, and air pressure are extracted from the data set as input; Construct the LSTM model equation as follows: Forget Gate :

[0018] is the output of the forget gate, is the activation function, is the weight matrix of the forget gate, is the hidden state of the previous time step, is the input feature of the current time step, is the biased top of the forget gate; Input Gate and candidate values :

[0019] is the weight matrix of the input gate, is the bias top of the input gate, is the weight matrix of candidate values, is the biased top of the candidate value; Output Gate and the final hidden state :

[0020] is the cell state at the current time step; Fully connected layer:

[0021] is the predicted value of the remaining life corresponding to time step t, is the weight matrix of the output gate, is the bias top of the output gate; Train the LSTM model to minimize the actual remaining lifetime and predicted values The error between them determines the loss function :

[0022] is the number of samples in the dataset, is the predicted remaining life at time step t, is the actual remaining life at time step t; Predict the normal range of remaining life expectancy based on confidence interval and standard deviation.

[0023] Furthermore, the method for determining the abnormal amplitude is: At every point in time In the above example, the abnormal amplitude within the last w time points is calculated according to the sliding window method. The window size w is a hyperparameter. The specific steps are as follows: Set the sliding window size w and include the time point ; In each sliding window, calculate the anomaly type feature The abnormal amplitude , then:

[0024] In the formula, For the current The characteristic value at time, is the historical mean of the anomaly type feature, is the standard deviation of the anomaly type feature; The abnormal amplitude of the comprehensive abnormal type feature is obtained by weighted summing of the abnormal amplitudes of multiple abnormal type features.

[0025] Furthermore, the algorithm expression of the Pearson correlation coefficient is:

[0026] In the formula, is the Pearson correlation coefficient, and They are the abnormal type characteristics, Features and The covariance between and They are and The standard deviation of The Pearson correlation coefficient is used to determine whether there is a correlation between the abnormal type characteristics and to determine the cause of the abnormality.

[0027] Furthermore, the method for determining the associated DC resistance tester is: When the DC resistance tester is found to be in an abnormal state, the current time is determined to be recorded as the abnormal time; Determine the time when the DC resistance tester was tested and was normal last time as the safe time, and construct the abnormal time zone with the abnormal time and the safe time; Obtain the device status value of the DC resistance tester under safe time and the device status value under abnormal time, which are recorded as , get The difference between them is recorded as the risk change value; Obtain the device status values ​​of other associated DC resistance testers that are not monitored at the safe time and the current time, and determine the difference to be recorded as other risk change values; Based on the comparison between other risk change values ​​and risk change values, as well as obtaining the correlation time between the normal time and the current time for detection by the associated DC resistance tester and comparing it with the abnormal time zone, it is determined whether the associated DC resistance tester is in a risky state.

[0028] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects: By constructing an isolation forest model for the data set constructed by the DC resistance tester, the initial anomaly detection and monitoring of the DC resistance tester is realized, and the abnormal point data detected by the isolation forest model is further analyzed and identified by introducing the Gaussian distribution model, so as to improve the detection optimization of the abnormal point data, accurately identify the abnormal points, and realize the accurate judgment of the state of the DC resistance tester; And according to the nature of the anomaly, including resistance anomaly, voltage anomaly, etc., it is accurately classified. At the same time, if there are multiple abnormal features at the same time, the correlation between the multiple features is calculated to further clarify the fault factors of the current DC resistance tester, effectively reduce the false alarms and missed alarms of the anomaly, and improve the accuracy and robustness of the DC resistance tester status monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0030] Figure 1 It is the overall module block diagram of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] During long-term operation, the DC resistance tester may be affected by various factors such as changes in the external environment, equipment aging, improper operation, and operating failures, which may cause abnormal performance and affect the accuracy of the test results. Therefore, in order to ensure the reliability of the equipment, test accuracy, and normal operation of the system, how to build a state monitoring solution for a pair of DC resistance testers and realize rapid monitoring and judgment of the state of the DC resistance tester is a problem that needs to be solved.

[0033] The present invention will be further described below in conjunction with the embodiments.

[0034] Example 1 (see Figure 1 ): A DC resistance tester state intelligent monitoring and management system, comprising at least: The data acquisition module is used to set the acquisition time (can be 3 days, and the frequency is every hour, no limit) to collect the key operating data (voltage, resistance) of the DC resistance tester, and combine the environmental parameters (ambient temperature, ambient humidity and ambient air pressure) and equipment status data (equipment usage time, equipment remaining life) in the external environmental factors to form a data set containing multiple samples. , is the total number of samples in the dataset, For the The feature vector of each sample (time point) contains multiple feature values, and then the operation data of the DC resistance tester, external environmental parameters and equipment status data are combined for subsequent abnormality detection, where:

[0035] In the formula, For the The resistance of the samples, For the The ambient temperature of the sample, For the The ambient humidity of each sample, For the The voltage of each sample, For the The ambient pressure of the sample, For the The device usage time of each sample, For the The remaining life of the equipment of each sample.

[0036] Then the dataset Preprocessing, including data set Cleaning is used to remove missing values ​​and outliers, and then standardization is performed to eliminate dimensional differences, thereby obtaining a new data set:

[0037] In the formula, is the preprocessed data set, After standardization The feature vector of samples is used for subsequent calculation. Here, for the data set The cleaning and standardization are well-known technical methods and will not be described in detail in this embodiment.

[0038] Therefore, through the constructed data set In order to conduct further analysis on the DC resistance tester and clarify the actual status of the DC resistance tester.

[0039] The anomaly detection and analysis module monitors the anomaly points of the DC resistance tester in advance based on the isolation forest model. The method for constructing the isolation forest model is as follows: Set hyperparameters, including the number of trees , used to improve the predictive performance of the model; and the sample subset size , used to reduce the computational burden while retaining enough information to capture data characteristics; After preprocessing the dataset Train an Isolation Forest model pair to build multiple isolated trees, specifically: In the dataset Randomly select samples, and obtain a randomly selected sample set , randomly select a feature To divide the data, , while determining the selected features The minimum and maximum values ​​of , so as to select the partition points later; In the selected features A split point is randomly selected between the minimum and maximum values ​​of , so that the sample set is divided into two subsets, including ,in are the left subset and the right subset respectively, and the left subset :All sample features are in Upper less than or equal to the partition point ; Right subset :All samples in the feature The upper is greater than the dividing point; right Repeat the above steps to build recursively until one of the following conditions is met: The size of the subset is less than or equal to 1; The preset maximum tree depth is reached; Thus, the generated tree is defined as an isolated tree :

[0040] In the formula, and are the left subtree and the right subtree respectively, and are recursively defined as the same structure. is the index of the tree; Therefore, for each outlier Calculate the abnormality score, then:

[0041] In the formula, is the path length, is the optimized standardization coefficient used to adjust the score, is the number of samples, is the characteristic dimension; Set the first abnormal threshold ,like , then Determine outliers and build an outlier set , thereby, achieving the preliminary determination of the abnormal point of the DC resistance tester, that is, the preliminary determination of the abnormal state.

[0042] Optimized standardized coefficient By introducing feature dimension (the increase in feature dimension will affect the sensitivity of the isolation forest model to anomalies. If the proportion of each anomaly component in the high-dimensional feature space is small, it may lead to a low anomaly score and not be correctly identified as an anomaly. Therefore, by introducing feature dimension in the normalization coefficient, the sparse distribution of anomalies caused by the increase in dimension is balanced, and the stability of detection in multiple dimensions is improved), data distribution (considering that the distribution of anomalies in the sample set is sparse, or the ambient temperature and ambient humidity have a great impact, optimization can more accurately identify relatively rare anomaly types. For example, for features with periodic changes such as ambient temperature and humidity, the coefficient can be adjusted to make it more sensitive to deviations from the normal distribution, thereby identifying high-impact anomalies), anomaly type (such as voltage, resistance, remaining life, etc., because different anomaly types have different effects on the health status of the equipment, therefore, corresponding coefficients are set for each anomaly type, so that the model can be more targeted, improve the robustness of the model in practical applications, and make the model more sensitive to potential failure modes) to dynamically adjust the initial normalization coefficient. is obtained, where:

[0043] Where: is the number of features, For the The anomaly weight of the feature, For the The abnormality score of each feature, is the standardized coefficient after combining the data distribution;

[0044] Where: is the standardized coefficient after the feature dimension is introduced, is the regulating factor, is the coefficient of variation, obtained from the ratio of the standard deviation to the mean of all features;

[0045] Where: is the initial standardized coefficient, 1 to The harmonic number of In order to introduce the correction factor of the characteristic dimension d, a logarithmic function is used to control the impact of the growth of the dimension on the path length, then:

[0046] Initial standardized coefficients

[0047] is the number of samples.

[0048] In this embodiment, in order to improve the detection results of the optimized isolation forest, identify samples with high abnormality, and further clarify the abnormal state of the DC resistance tester, a Gaussian distribution model is introduced to optimize the detection results of the isolation forest, as follows: The outlier point set constructed by the isolation forest model Train the Gaussian distribution model and assume that the data is Gaussian distribution; For each outlier Calculate the probability density:

[0049] In the formula, For abnormal points The probability density under the Gaussian distribution model is, For the Gaussian distribution model, The probability density function of is the number of Gaussian distributions in the Gaussian distribution model, For the The mixing coefficient of the Gaussian distribution, For the The mean vector of a Gaussian distribution, For the The covariance matrix of a Gaussian distribution, For the The probability density function of a Gaussian distribution is is the function symbol of Gaussian distribution; Set the second abnormal threshold ,like , then determine is an outlier point; Furthermore, in this example, after using the isolation forest and Gaussian distribution models, the abnormal points of the DC resistance tester can be comprehensively determined, as follows: If exists , then It is determined to be the final outlier point. Therefore, the possible outliers in the data are pre-identified through the isolation forest model, and the Gaussian distribution model is introduced to further analyze the outliers identified by the isolation forest model, so as to more accurately identify the outliers and reduce false positives and negatives. It is worth mentioning that by combining multiple features such as resistance, temperature, humidity, and power supply voltage, the isolation forest and Gaussian distribution models can more comprehensively understand the complex relationships in the data, thereby more effectively detecting anomalies, thereby effectively improving the detection and recognition accuracy.

[0050] Anomaly detection optimization module, further, in this embodiment, in order to further adjust and allocate the judgment results of the isolation forest model and the introduced Gaussian distribution model, so as to optimize the judgment accuracy of the abnormal points of the DC resistance tester by introducing weight fusion, therefore, the specific method is as follows: Obtaining anomaly scores for the isolation forest model , and the Gaussian distribution model for the sample vector The probability density of Construct the weight fusion algorithm expression, then we have:

[0051] In the formula, is the comprehensive evaluation value, is the weight coefficient, which is used to determine the contribution of the isolation forest and Gaussian models. Therefore, by setting the third anomaly threshold, By comparing with the third abnormal threshold, the judgment result of the existence of abnormal points in the DC resistance tester based on the combination of the isolation forest model and the Gaussian distribution model can be determined, so as to further realize accurate abnormal monitoring and judgment of the DC resistance tester, so as to carry out subsequent control and disposal of the DC resistance tester and facilitate the subsequent application of the DC resistance tester.

[0052] Therefore, the abnormal point in the DC resistance tester can be re-determined through the above, and a new abnormal point can be obtained. , and further, that is:

[0053] Therefore, the normal range of resistance, normal range of ambient temperature, normal range of ambient humidity, normal range of voltage, normal range of ambient air pressure, normal range of usage time and normal range of remaining life are defined for each sensor data (which can be set through historical data without limitation). Perform one-by-one matching detection to determine whether it exceeds the preset normal range, that is, to determine the abnormal type characteristics of the DC resistance tester.

[0054] It is worth noting that the normal range of remaining life usually takes into account the influence of environmental parameters such as voltage, resistance, temperature, humidity, air pressure, and the actual remaining life data of the DC resistance tester. Therefore, in order to further accurately define the normal range of remaining life, this embodiment also includes the following steps: Obtain historical data of the DC resistance tester, including voltage, resistance, temperature, humidity, air pressure, and actual remaining life data, and perform normalization processing; The LSTM model is introduced for prediction. Therefore, the historical data is converted into the data format of time steps in advance. Assuming that each time window has w time steps, N time series samples are obtained after the window slides. For each time window, the corresponding voltage, resistance, temperature, humidity, air pressure and other characteristic values ​​are extracted from the data set as input; The corresponding target variable (output) is the remaining life; Construct the LSTM model equation as follows: Forget Gate :

[0055] is the output of the forget gate, is the activation function, is the weight matrix of the forget gate, is the hidden state of the previous time step, is the input feature of the current time step, including resistance, voltage, etc., is the biased top of the forget gate; Input Gate and candidate values :

[0056] is the weight matrix of the input gate, is the bias top of the input gate, is the weight matrix of candidate values, is the biased top of the candidate value; Cell status update :

[0057] is the cell state at the previous time step; Output Gate and the final hidden state :

[0058] is the weight matrix of the output gate, is the bias top of the output gate; Fully connected layer:

[0059] is the predicted value of the remaining life corresponding to time step t; Train the LSTM model to minimize the actual remaining lifetime and predicted values The error between them determines the loss function :

[0060] is the number of samples in the dataset, is the predicted remaining life at time step t, is the actual remaining life at time step t; Based on the confidence interval and standard deviation Determine the predicted remaining life range, then:

[0061] is the lower limit of the remaining life range, is the upper limit of the remaining life range, The confidence level is calculated, which allows the predicted remaining life range to be clearly determined in order to be compared with the abnormal points in the DC resistance tester.

[0062] Further, it should be noted that, assuming that in the process of determining the abnormal type of the DC resistance tester, there are multiple abnormal type characteristics, for example, there are abnormalities in resistance, voltage, etc. at the same time, which makes it impossible to determine the specific abnormal type characteristics of the DC resistance tester. Therefore, in order to further determine the abnormal type characteristics of the DC resistance tester in this case, so as to implement control measures for the DC resistance tester later, this embodiment also includes: The multi-type anomaly analysis module is used to re-determine the specific anomaly cause by calculating the anomaly magnitude of the anomaly type characteristics, as follows: Assume there are multiple exception type features , such as resistance, voltage, etc. At every point in time In the above example, the sliding window method is used to calculate the abnormal amplitude within the last w time points. The window size w is a hyperparameter, which is used to adjust according to the characteristics of the data and the historical records of abnormalities. The specific steps are: Set the sliding window size w and include the time point ; In each sliding window, calculate the anomaly type feature Abnormal amplitude (resistance, voltage, etc.) , then:

[0063] In the formula, For the current The characteristic value (resistance, voltage, etc.) at the time, is the historical mean of the anomaly type feature, is the standard deviation of the anomaly type feature, where: ; ; In the formula, is the starting position covered by sliding the window size w steps back from the current time point i, For time point The characteristic value of Thus, the abnormal amplitude of each abnormal type feature is obtained , by combining the abnormal amplitudes of multiple abnormal type characteristics to make judgments, we have:

[0064] In the formula, is the comprehensive abnormal amplitude, They are the characteristics of each abnormal type (including resistance, voltage, ambient temperature, ambient humidity, etc.) at the time point The abnormal amplitude on are the corresponding weight coefficients respectively; Therefore, the fourth threshold is set to Compare with the fourth threshold value to determine, therefore, if there is an abnormality, perform correlation analysis to determine the presence of associated abnormal type features, specifically: Assuming there are two features including resistance and voltage, it can be expressed as and , calculate the Pearson correlation coefficient:

[0065] In the formula, is the Pearson correlation coefficient, Features and The covariance between and They are and The standard deviation of Corresponding, no further description is given here), among which, The calculation formula is:

[0066] Where: is the sample size, and They are and The corresponding mean (same as above correspond), and Respectively represent the i-th sample in the feature and The value on the Pearson correlation coefficient The value of ,Right now: When it is 1, it indicates two characteristic resistors and voltage Perfect positive correlation (if the positive correlation between resistance and voltage suddenly becomes strong, it may mean that there is an abnormality in the resistance measurement, or the voltage control system may be unstable, resulting in an abnormal relationship between voltage and resistance, so it needs special attention); When it is -1, it indicates two characteristic resistors and voltage Completely negative correlation (e.g., when a voltage controller or power module fails, the voltage may drop abnormally, causing the change in resistance to negatively affect the voltage); When it is equal to 0, it indicates two characteristic resistances and voltage No linear correlation (may indicate a system malfunction or measurement problem); In this way, the status monitoring of the DC resistance tester is realized, and the abnormal points of the DC resistance tester and their corresponding abnormal type status can be determined, which is convenient for the management and control personnel to take corresponding control measures on the DC resistance tester in time, which is beneficial to the normal operation of the subsequent DC resistance tester.

[0067] Finally, it also includes a risk equipment prediction module, which is used to determine the current time as the abnormal time when the DC resistance tester is in an abnormal state, determine the last time the DC resistance tester was tested and was normal as the safe time, and construct an abnormal time zone with the abnormal time and the safe time. , and then, obtain the device status value of the DC resistance tester under the safe time and the device status value under the abnormal time (which can be calculated by machine learning model based on the operating time, operating fault frequency, operating temperature, fault time and other characteristics of the DC resistance tester, such as linear regression model, etc., which will not be described here), respectively, and record them as , get The difference between the two is recorded as the risk change value, and the device status value of other associated DC resistance testers that are not monitored at the safe time is obtained. The device status value of the DC resistance tester at the current time is obtained, and the difference between the two is determined as other risk change values. It is determined whether the other risk change value is greater than or equal to the risk change value. At the same time, it is obtained whether the correlation time between the last normal time of the associated DC resistance tester and the current time is consistent with the abnormal time zone. Same, if both are (greater than or equal to the risk change value and the associated time and abnormal time zone The associated DC resistance tester is marked as a risky state for pre-condition monitoring, thereby facilitating the pre-determination of a large number of associated DC resistance testers with suspected abnormalities, and facilitating the management and control personnel to implement corresponding monitoring and disposal.

[0068] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A DC resistance tester state intelligent monitoring and management system, comprising: The data acquisition module is used to obtain the key operating data, environmental parameters, and equipment status data of the DC resistance tester to build a data set; The anomaly detection and analysis module trains an isolation forest model through a data set to pre-monitor the anomaly points of the DC resistance tester to obtain an anomaly point set, and is characterized by further comprising: The introduced Gaussian distribution model is trained based on the set of outliers. By calculating the probability density of outliers, the outliers of the DC resistance tester are re-determined to optimize the detection results of the isolation forest, where: Determine the abnormal type characteristics of the abnormal point based on the detection results; And, a multi-type abnormal analysis module is used to calculate the abnormal amplitude of each abnormal type feature according to the sliding window method to obtain the comprehensive abnormal amplitude. If there are still multiple abnormal type features, a correlation analysis is performed to determine the associated abnormal type features by calculating the Pearson correlation coefficient and define the abnormal factors of the DC resistance tester; The risk equipment prediction module is used to determine the associated DC resistance testers with associated risk states based on the abnormal state of the DC resistance tester when it is determined that there is an abnormal DC resistance tester, so as to form pre-monitoring for them.

2. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 1 is characterized in that: The key operating data include resistance and voltage; the environmental parameters include ambient temperature, ambient humidity and ambient air pressure; the equipment status data include equipment usage time and equipment remaining life.

3. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 1 is characterized in that: Calculate outliers based on the isolation forest model The abnormality score of is: In the formula, is the path length, is the optimized standardization coefficient used to adjust the score, is the number of samples, is the characteristic dimension; based on Determine outliers , and construct a set of outliers , is the first abnormal threshold.

4. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 3 is characterized in that: The standardized coefficient Dynamically adjust the initial normalization coefficient by introducing feature dimension, data distribution, and anomaly type Get, where: Where: is the number of features, For the The anomaly weight of the feature, For the The abnormality score of each feature, is the standardized coefficient after combining the data distribution; Where: is the standardized coefficient after the feature dimension is introduced, is the regulating factor, is the coefficient of variation; Where: is the initial standardized coefficient, 1 to The harmonic number of is the correction factor; Initial standardized coefficients , is the number of samples.

5. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 3 is characterized in that: The Gaussian distribution model introduced by the outlier set is trained, and the method for calculating the probability density of the outlier points is: Through the collection of outliers Train the Gaussian distribution model and assume that the data is Gaussian distribution; For each outlier Calculate the probability density: In the formula, For abnormal points The probability density under the Gaussian distribution model is, For the Gaussian distribution model, The probability density function of is the number of Gaussian distributions in the Gaussian distribution model, For the The mixing coefficient of the Gaussian distribution, For the The mean vector of a Gaussian distribution, For the The covariance matrix of a Gaussian distribution, For the The probability density function of a Gaussian distribution is is the function symbol of Gaussian distribution; Combination Re-determine outliers .

6. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 1 is characterized in that: By introducing the weight fusion method to optimize the judgment results of the isolation forest model and the Gaussian distribution model, we can obtain: Construct the weight fusion algorithm expression, then we have: In the formula, is the comprehensive evaluation value, is the weight coefficient, which is used to determine the contribution of the isolation forest and Gaussian model; based on Re-evaluate outliers .

7. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 1 is characterized in that: The abnormal point When they exist, the normal ranges of resistance, ambient temperature, humidity, voltage, pressure, usage time and remaining life are defined respectively. When determining the normal range of remaining life, historical data is used for prediction, as follows: The LSTM model is introduced for prediction. Assuming that each time window has w time steps, N time series samples are obtained after the window slides. For each time window, the corresponding actual remaining life data of voltage, resistance, temperature, humidity, and air pressure are extracted from the data set as input; Construct the LSTM model equation as follows: Forget Gate : is the output of the forget gate, is the activation function, is the weight matrix of the forget gate, is the hidden state of the previous time step, is the input feature of the current time step, is the biased top of the forget gate; Input Gate and candidate values : is the weight matrix of the input gate, is the bias top of the input gate, is the weight matrix of candidate values, is the biased top of the candidate value; Output Gate and the final hidden state : is the cell state at the current time step; Fully connected layer: is the predicted value of the remaining life corresponding to time step t, is the weight matrix of the output gate, is the bias top of the output gate; Train the LSTM model to minimize the actual remaining lifetime and predicted values The error between them determines the loss function : is the number of samples in the dataset, is the predicted remaining life at time step t, is the actual remaining life at time step t; Predict the normal range of remaining life expectancy based on confidence interval and standard deviation.

8. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 1 is characterized in that: The method for determining the abnormal amplitude is: At every point in time In the above example, the sliding window method is used to calculate the abnormal amplitude within the last w time points. The window size w is a hyperparameter. The specific steps are as follows: Set the sliding window size w and include the time point ; In each sliding window, calculate the anomaly type feature The abnormal amplitude , then: In the formula, For the current The characteristic value at time, is the historical mean of the anomaly type feature, is the standard deviation of the anomaly type feature; The abnormal amplitude of the comprehensive abnormal type feature is obtained by weighted summing of the abnormal amplitudes of multiple abnormal type features.

9. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 1, characterized in that: The algorithm expression of the Pearson correlation coefficient is: In the formula, is the Pearson correlation coefficient, and They are the abnormal type characteristics, Features and The covariance between and They are and The standard deviation of The Pearson correlation coefficient is used to determine whether there is a correlation between the abnormal type characteristics and to determine the cause of the abnormality.

10. The intelligent monitoring and management system for the state of a DC resistance tester according to claim 1, characterized in that: The method for determining the associated DC resistance tester is: When the DC resistance tester is found to be in an abnormal state, the current time is determined to be recorded as the abnormal time; Determine the time when the DC resistance tester was tested and was normal last time as the safe time, and construct the abnormal time zone with the abnormal time and the safe time; Obtain the device status value of the DC resistance tester under safe time and the device status value under abnormal time, which are recorded as , get The difference between them is recorded as the risk change value; Obtain the device status values ​​of other associated DC resistance testers that are not monitored at the safe time and the current time, and determine the difference to be recorded as other risk change values; Based on the comparison between other risk change values ​​and risk change values, as well as obtaining the correlation time between the normal time and the current time for detection by the associated DC resistance tester and comparing it with the abnormal time zone, it is determined whether the associated DC resistance tester is in a risky state.

Images