Equipment state repair prediction method and system based on support vector machine

Abstract

The invention discloses an equipment state repair prediction method and system based on a support vector machine, and the method comprises the steps: analyzing an equipment operation log of a comprehensive monitoring system through a regular expression, and extracting and calculating the statistical information of the action time of equipment; performing feature extraction on the equipment action time to obtain an equipment action time feature vector set, extracting feature vectors, and performing training by using a support vector machine model; obtaining operation and maintenance data of target equipment, performing feature extraction to obtain an equipment target feature vector set, generating a model verification report, and generating a maintenance suggestion early warning according to a prediction result; the system comprises a data acquisition module, a statistics module, a feature module and a prediction module. According to the method, the equipment state condition is intelligently predicted through the equipment action time feature vector, the prediction accuracy is improved, and the operation and maintenance pressure and the influence of the equipment on the service are reduced.

Description

Technical Field

[0001] The present invention relates to the field of rail transit monitoring systems, and in particular to a method and system for predicting equipment status based on a support vector machine. Background Art

[0002] The scale of urban rail transit operation continues to expand, and the number of automated equipment is also increasing day by day. The demand for intelligent predictive maintenance of key equipment in subway operations is becoming increasingly urgent. Especially in the comprehensive monitoring system, the number of devices that need to be monitored is huge and diverse, and the number of devices monitored in real time reaches hundreds of thousands or even more, which puts a lot of pressure on equipment operation and maintenance personnel. Therefore, we need to use machine learning to automatically and intelligently predict the status of equipment and improve the role of intelligent assistance. In the past, equipment operation and maintenance mainly focused on real-time cross-sectional information such as the operating time and number of operations of the equipment. This type of information reflects the current operation of the equipment, but cannot truly reflect the aging degree and sensitivity of the current equipment in multiple dimensions. There is insufficient research on the operation process of the equipment and the linkage process of related equipment. In order to meet the needs of intelligent operation and maintenance of equipment, a method for predicting the status of equipment based on support vector machines is urgently needed. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a method and system for predicting equipment status based on support vector machines.

[0004] Technical solution: The device status prediction method based on support vector machine of the present invention comprises the following steps: (1) Constructing the equipment action time data set: using regular expressions to parse the equipment operation logs of the integrated monitoring system, extracting the equipment selection time, remote control time, and remote signaling time, and calculating the statistical information of the equipment action time; (2) Constructing a support vector machine model for equipment action time: Extracting features of equipment action time to obtain a set of equipment action time feature vectors, extracting equipment maintenance records, equipment type, point type, equipment operation time, number of operations, equipment manufacturer and model feature vectors, dividing the feature vectors into training set, validation set and test set, and training them using a support vector machine model; (3) Predict equipment status maintenance information: Obtain the operation and maintenance data of the target equipment, perform feature extraction to obtain the target feature vector set of the equipment, import it into the trained support vector machine model, generate a model verification report, and generate maintenance suggestion warnings based on the prediction results.

[0005] Furthermore, the step (1) includes using Python's pandas data analysis library to load the front-end log of the integrated monitoring system, using regular expressions to parse the text into two-dimensional table records, and then cleaning, extracting and converting the data into uplink and downlink data of the equipment operation that can be used, mainly including equipment selection time, equipment selection telesignaling time, equipment remote control time and equipment telesignaling time.

[0006] Furthermore, the device action time in step (1) includes the device selection action time, the device control action time and the associated device time difference; The device selection action time is the device selection signal sending time tx1 minus the device selection confirmation signal time tx2, that is, tx2-tx1; The device control action time is the difference between the device control signal sending time tk1 and the device remote signal confirmation time tk2, that is, tk2-tk1; The associated device time difference is a set of devices with a tree-like association relationship in the integrated monitoring system. When a remote signal of any device is generated, one or more associated device remote signal signals are subsequently generated, and the time difference is the remote signal time difference value of the two device signals.

[0007] Furthermore, the device action time feature vector set in step (2) includes feature vectors of abnormal points, time domain and signal quantity of the device action time; The abnormal point feature vector extraction of the device action time is to extract the feature vector of the device action time as an abnormal point by using a normal distribution model; The time domain feature extraction of the device action time is to extract the device action time series of different periods of the device operation time, calculate the peak value, valley value, mean value, variance and standard deviation of the time series, and form the device time time domain feature vector; The signal feature extraction of the device action time is to extract the current signal value, the action signal value and the current signal value of the device when the associated device is generated during device selection and action operation.

[0008] Furthermore, the step (2) includes cleaning the extracted feature vector data set to generate a data set that can be used by the model, including data discretization, digitization, data replacement, gap filling and normalization operations; Automatically select the optimal parameters of the support vector model and initialize the parameters through the GridSearchCV library of sklearn.model_selection; The feature vector set is divided into training set, validation set and test set. The core function and regularization parameters are changed based on the initialization parameters, and the support vector machine model with the best parameters is selected.

[0009] Furthermore, the core function is selected as a Gaussian function, and the regularization parameter is selected as 100.

[0010] Furthermore, the feature vector set is divided by a cross-validation method.

[0011] Furthermore, the step (3) of acquiring the operation and maintenance data of the target device is to extract the device target feature vector through the system's data acquisition module, statistical module and feature module.

[0012] The equipment status maintenance prediction system based on support vector machine described in the present invention includes a data acquisition module, a statistical module, a feature module and a prediction module; The acquisition module is used to periodically analyze the device front-end log, generate device action time data, and generate records to store in the database; The statistical module is used to count the operation data of the equipment and improve the statistical information of the equipment action time, operation time and operation times in the equipment records; The feature module is used to read device data records and model class data to generate a set of device target feature vectors; The prediction module is used to import the trained support vector machine model, generate a model verification report, and generate a maintenance suggestion warning based on the prediction result.

[0013] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: the present invention adopts support vector machines, which can utilize shorter training time for small sample prediction problems, achieve stronger generalization ability, and has strong practicality; the device status is intelligently predicted through the device action time feature vector, which improves the prediction accuracy and reduces the operation and maintenance pressure and the impact of the equipment on the business. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The flowchart of the equipment status prediction method based on support vector machine is shown in FIG. DETAILED DESCRIPTION

[0015] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0016] like Figure 1 As shown, the device status prediction method based on support vector machine described in the present invention includes the following steps: (1) Constructing a data set of equipment action time Using regular expressions to parse the equipment operation logs of the integrated monitoring system and extract the equipment selection time, remote control time and telesignaling time means using Python's pandas data analysis library to load the front-end logs of the integrated monitoring system, using regular expressions to parse the text into two-dimensional table records, and then cleaning, extracting and converting the data into uplink and downlink data that can be used for equipment operation, mainly including equipment selection time, equipment selection telesignaling time, equipment remote control time and equipment telesignaling time; Analyze the complete process of equipment control for comprehensive monitoring in the equipment front-end log, that is, first remote control selection, remote control selection confirmation, and then remote control and remote signal confirmation; The device action time includes the device selection action time, device control action time and associated device time; The device selection action time refers to the device selection signal sending time tx1 minus the device selection confirmation signal time tx2, that is, tx2-tx1; The device control action time refers to the difference between the device control signal sending time tk1 and the device telesignaling signal confirmation time tk2, that is, tk2-tk1.

[0017] Time difference analysis of associated devices means that there are a large number of devices with tree-like association relationships in the integrated monitoring system. When a telesignaling signal of any device is generated, one or more associated device telesignaling signals are subsequently generated, and the time difference is the telesignaling time difference of the two device signals.

[0018] Calculate statistical information to calculate the variance, sum, difference and median of equipment operation time, number of operations and equipment action time, and use them to generate basic reports for equipment data analysis.

[0019] (2) Construct a device action time support vector machine model.

[0020] Extracting the characteristics of the device action time to obtain a device action time characteristic vector set refers to extracting the characteristic vectors of the abnormal points, time domain and signal quantity of the device action time; Extracting abnormal point features from device action time refers to using a normal distribution model to extract a feature vector that indicates whether the device action time is an abnormal point.

[0021] The time domain features of the equipment action time are extracted, and the equipment action time series of different periods of equipment operation time are extracted. The peak value, valley value, mean, variance and standard deviation of the time series are calculated to form the equipment time domain feature vector set.

[0022] The semaphore feature extraction is performed on the device action time, and the current semaphore value and action semaphore value during device selection and action operation, as well as the current semaphore value of the device when the associated device is generated, are extracted.

[0023] Extracting equipment maintenance records, equipment type, point type, equipment operation time, number of operations, equipment manufacturer and model feature vectors refers to extracting equipment maintenance records, equipment type, point type, equipment manufacturer and model feature vectors from the equipment operation and maintenance system; Divide into training set, validation set and test set and use support vector machine model for training, mainly including: Clean the data set to generate a data set that can be used by the model, including operations such as data discretization, digitization, data replacement, gap filling and normalization; The optimal parameters of the support vector model are automatically selected through the GridSearchCV library of sklearn.model_selection, and the initialization parameters, that is, the core function is selected as the Gaussian function and the regularization parameter is selected as 100.

[0024] Through cross-validation, the feature vector set is divided into training set, validation set and test set. The core function and regularization parameters are changed based on the initialization parameters to select the support vector machine model with the best parameters. (3) Predict equipment status and repair information Obtain the operation and maintenance data of the target device, and extract the target feature vector of the device through the system's data acquisition module, statistics module and feature module; The acquisition module periodically analyzes the equipment front-end log, generates equipment action time data, and stores the generated records in the database; Statistics module, which counts the operation data of the equipment and improves the statistical information of equipment action time, operation time, number of operations, etc. in the equipment records; The feature module reads the device data records and model class data and generates a set of device target feature vectors; The equipment target feature vector set is imported into the trained support vector machine model to predict the maintenance results, generate a model verification report, and generate maintenance suggestion warnings based on the prediction results.

[0025] The equipment status maintenance prediction system based on support vector machine described in the present invention includes a data acquisition module, a statistical module, a feature module and a prediction module; The acquisition module is used to periodically analyze the device front-end log, generate device action time data, and generate records to store in the database; The statistical module is used to count the operation data of the equipment and improve the statistical information of the equipment action time, operation time and operation times in the equipment records; The feature module is used to read device data records and model class data to generate a set of device target feature vectors; The prediction module is used to import the trained support vector machine model, generate a model verification report, and generate a maintenance suggestion warning based on the prediction result.

Claims

1. A device status prediction method based on support vector machine, characterized in that: The steps include: (1) Constructing the equipment action time data set: using regular expressions to parse the equipment operation logs of the integrated monitoring system, extracting the equipment selection time, remote control time, and remote signaling time, and calculating the statistical information of the equipment action time; (2) Constructing a support vector machine model for equipment action time: Extracting features of equipment action time to obtain a set of equipment action time feature vectors, extracting equipment maintenance records, equipment type, point type, equipment operation time, number of operations, equipment manufacturer and model feature vectors, dividing the feature vectors into training set, validation set and test set, and training them using a support vector machine model; (3) Predict equipment status maintenance information: Obtain the operation and maintenance data of the target equipment, perform feature extraction to obtain the target feature vector set of the equipment, import it into the trained support vector machine model, generate a model verification report, and generate maintenance suggestion warnings based on the prediction results.

2. The device status prediction method based on support vector machine according to claim 1 is characterized in that: The step (1) includes using the pandas data analysis library of Python to load the front-end log of the comprehensive monitoring system, using regular expressions to parse the text into two-dimensional table records, and then cleaning, extracting and converting the data into uplink and downlink data of the equipment operation that can be used, mainly including equipment selection time, equipment selection telesignaling time, equipment remote control time and equipment telesignaling time.

3. The device status prediction method based on support vector machine according to claim 1 is characterized in that: The device action time in step (1) includes the device selection action time, the device control action time and the associated device time difference; the device selection action time is the device selection signal sending time tx1 minus the device selection confirmation signal time tx2, that is, tx2-tx1; the device control action time is the device control signal sending time tk1 minus the device telesignaling signal confirmation time tk2, that is, tk2-tk1; the associated device time difference is a set of devices with a tree-like association relationship in the integrated monitoring system. When any one of the device telesignaling signals is generated, one or more associated device telesignaling signals are subsequently generated, and the time difference is the telesignaling time difference of the two device signals.

4. The device status prediction method based on support vector machine according to claim 1 is characterized in that: The device action time feature vector set of step (2) includes feature vectors of abnormal points, time domain and signal quantity of the device action time; the feature vector extraction of abnormal points of the device action time is to extract feature vectors of abnormal points of the device action time using a normal distribution model; the time domain feature extraction of the device action time is to extract the device action time series of different periods of the device operation time, calculate the peak value, valley value, mean value, variance and standard deviation of the time series, and form the device time time domain feature vector; the signal quantity feature extraction of the device action time is to extract the current signal quantity value, action signal quantity value and the current signal quantity value of the device when the associated device is generated during device selection and action operation.

5. The device status prediction method based on support vector machine according to claim 1 is characterized in that: The step (2) includes cleaning the extracted feature vector data set to generate a data set that can be used by the model, including data discretization, digitization, data replacement, gap filling and normalization operations; automatically selecting the optimal parameters of the support vector model through the GridSearchCV library of sklearn.model_selection, and initializing the parameters; dividing the feature vector set into a training set, a validation set and a test set, changing the two parameters of the core function and the regularization parameter based on the initialization parameters, and selecting the support vector machine model with the best parameters.

6. The device status prediction method based on support vector machine according to claim 5 is characterized in that: The core function is selected as a Gaussian function, and the regularization parameter is selected as 100.

7. The device status prediction method based on support vector machine according to claim 5 is characterized in that: The feature vector set is divided by cross-validation.

8. The device status prediction method based on support vector machine according to claim 1 is characterized in that: The step (3) of acquiring the operation and maintenance data of the target device is to extract the device target feature vector through the system's data acquisition module, statistical module and feature module.

9. A device status prediction system based on support vector machine, characterized in that: Including data collection module, statistics module, feature module and prediction module; The acquisition module is used to periodically analyze the device front-end log, generate device action time data, and generate records to store in the database; The statistical module is used to count the operation data of the equipment and improve the statistical information of the equipment action time, operation time and operation times in the equipment records; The feature module is used to read device data records and model class data to generate a set of device target feature vectors; The prediction module is used to import the trained support vector machine model, generate a model verification report, and generate a maintenance suggestion warning based on the prediction result.

Images