Switch abnormity prompting method based on AI regulation and control

Abstract

The invention discloses a switch abnormity prompting method based on AI regulation and control, relates to the technical field of switch abnormity detection, is used for solving the problems that the equipment failure rate is increased and the maintenance efficiency is reduced, and provides a switch abnormity prompting method by separately marking circuits in a switch, setting acquisition time and acquiring data of marked objects within the acquisition time. The method comprises the following steps: processing data by adopting a decision tree method, setting a plurality of state weights, detecting response time of a marked object, generating an abnormal feature in combination with the state weights, importing the marked object into different levels of abnormal detection mechanisms according to the abnormal feature, monitoring a monitoring signal generated by the abnormal detection mechanism in real time, and obtaining monitoring information. The operation records of all the marked objects are called, the prompt index is calculated after the clustering algorithm is introduced and the monitoring information is integrated, whether abnormal prompt is carried out or not is judged according to the prompt index, the response efficiency of the system to abnormal fluctuation is enhanced, and the equipment failure rate is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of switch abnormality detection, and more specifically, to a switch abnormality prompt method based on AI regulation. Background Art

[0002] In existing automation systems and intelligent monitoring devices, equipment operating status monitoring is of great significance to the safety and stability of the system. Traditional equipment status monitoring methods usually rely on manual regular inspections or single parameter measurements, such as identifying anomalies only through threshold judgments or simple alarm signals.

[0003] The existing technology has the following deficiencies: At present, existing methods are unable to combine multi-dimensional data for comprehensive analysis, lack the ability to deeply identify abnormal states, and are unable to adopt corresponding processing strategies for different types of switch anomalies. It is difficult to detect abnormal phenomena in a timely manner, resulting in increased system operation risks, increased equipment failure rates, reduced system reliability, and reduced maintenance efficiency. Therefore, a switch anomaly prompt method based on AI control is proposed.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a switch abnormality prompt method based on AI control, which solves the problems raised in the above-mentioned background technology by monitoring line status, decision tree data processing, multi-state weight analysis, and hierarchical abnormality detection and clustering mechanism.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a switch abnormality prompt method based on AI control, comprising the following steps: Step S1: Mark each line in the switch, set a collection time, and collect line data of the marked line within the collection time; In step S1, the line data of the marked line are the marked line vibration frequency, the marked line ground current and the marked line noise intensity; Select a period of time as the collection time, and collect line data for each marked line within the collection time as follows: Multiple time points are randomly set during the acquisition time, and the vibration frequency, ground current, and noise intensity of the marked line are detected at each time point to obtain the vibration frequency, ground current, and noise intensity of the marked line at each time point; Step S2: Process the line data using a decision tree method and set multiple state weights, detect the line data response time of the marked line, and generate abnormal features based on the state weights; Step S3: Selectively import the marked lines into line anomaly detection mechanisms at different levels based on the anomaly characteristics, monitor the monitoring signals generated by the line anomaly detection mechanisms in real time, and obtain monitoring information; Step S4: The closing times of each marked line in the switch are retrieved, integrated with the monitoring information, and then fed into a clustering algorithm to calculate the switch prompt index. It is determined whether to generate abnormal prompt information based on the switch prompt index.

[0007] In a preferred embodiment, in step S2, the vibration frequencies of the marked line at adjacent time points of the marked line are subtracted within the acquisition time. The difference between the vibration frequencies of the marked line at the previous time point and the vibration frequencies of the marked line at the next time point is used as the instantaneous vibration frequency change at the next time point. The subtraction process is performed once for every two adjacent time points, and multiple instantaneous vibration frequency changes are generated within the acquisition time. When calculating the instantaneous ground current variation, the difference between the marked line ground current at the latter time point and the marked line ground current at the previous time point in adjacent time points is taken as the instantaneous ground current variation at the previous time point.

[0008] In a preferred embodiment, in step S2, before using the decision tree method, the first time point and the last time point in the acquisition time are removed and integrated into a time point data set, and the noise intensity corresponding to each time point in the time point data set is merged into a noise data set; After removing the instantaneous vibration frequency variation at the last time point in the order of time points during the acquisition time, the instantaneous vibration frequency variation is merged into a vibration frequency data set; After removing the instantaneous ground current variation at the first time point in the order of time points during the acquisition time, the instantaneous ground current variation is merged into a ground current data set; The time point dataset, noise dataset, vibration frequency dataset and ground current dataset are used as input to build a decision tree model to screen the target line data.

[0009] In a preferred embodiment, in step S2, the steps of constructing a decision tree model to screen target line data are as follows: Data preparation: Take the time point dataset, noise dataset, vibration frequency dataset, and ground current dataset as input, and the target line data as output; Set the root node: Use the vibration frequency dataset and the ground current dataset to build two decision trees respectively, and select the maximum value in the vibration frequency dataset and the ground current dataset as the root node of the two decision trees respectively; Construct a decision tree: Use the root node as the parent node, and connect each parent node to its two child nodes on the left and right. After the connection is completed, use the child node as the new parent node and repeat the operation until the number of nodes matches the number of data in the vibration frequency dataset or the ground current dataset. Fill the nodes with data from the vibration frequency dataset or the ground current dataset in descending order. Retrieve and record the time point corresponding to the data in each node and the noise intensity at the corresponding time point from the time point dataset and the noise dataset. Screening time points: Count the number of layers of the two decision trees, screen out all nodes in the middle layer of the two decision trees and mark them, select the time point corresponding to the maximum noise intensity from the record information of the marked nodes in the two decision trees, and compare the time points selected by the two decision trees; Feature selection: If the time point selected by the two decision trees is the same, the instantaneous vibration frequency change and instantaneous ground current change at the corresponding time point are used as the target line data; if the time point selected by the two decision trees is different, all time points between the two time points are selected in the time point data set, and the instantaneous vibration frequency change and instantaneous ground current change at the middle time point are used as the target line data; Target output: Output the filtered target line data as output data.

[0010] In a preferred embodiment, in step S2, the state weight is set according to the target line data. The specific steps are as follows: The instantaneous vibration frequency change and instantaneous ground current change in the target line data are set as input variables, and the input variables are used as logistic regression intermediate variables to generate state weights: , where i is the sequence number of the three input variables, Generate state weight for the i-th input variable, e is the natural base, is the intermediate variable of the i-th logistic regression; When the input variable is the instantaneous vibration frequency change, the generated state weight is set to the vibration frequency weight; when the input variable is the instantaneous ground current change, the generated state weight is set to the ground current weight.

[0011] In a preferred embodiment, in step S2, the line data response time of the marker line includes the response time of the marker line vibration frequency and the response time of the marker line ground current; the time difference between the time point corresponding to the maximum value of the marker line vibration frequency and the time point corresponding to the minimum value of the marker line vibration frequency within the acquisition time is calculated as the response time of the marker line vibration frequency, and the response time of the marker line ground current is calculated similarly; The vibration frequency weight, the ground current weight, the response time of the vibration frequency of the marked line, and the response time of the ground current of the marked line are weighted and summed to generate the abnormal characteristics of the marked line.

[0012] In a preferred embodiment, in step S3, when the abnormal characteristics of the marked line exceed the preset abnormal threshold, the marked line is matched with the preset high-level detection mechanism; otherwise, the marked line is matched with the preset low-level detection mechanism; In each layer of detection, it is determined whether the line information of the marked line meets the detection conditions. If it meets the conditions, it will enter the next layer. If it does not meet the conditions, a monitoring signal will be generated. The monitoring information is the number of monitoring signals generated after the marked line passes the line anomaly detection mechanism.

[0013] In a preferred embodiment, in step S4, the number of closures of each marked line is normalized to generate a normalized result of the number of closures of the marked line; The normalized closing times of the marked lines are multiplied by the number of monitoring signals of the corresponding marked lines to generate the marked line prompt index. The marked line prompt indexes are combined into a prompt index dataset and then passed through the clustering algorithm for screening: Arrange the data in the prompt index data set from small to large, select the median of the data as the core point, set the screening difference, mark the data before or after the core point, and when the absolute value of the difference between the marked point and the core point is lower than the screening difference, filter out the marked point and the core point, and repeat the operation with the marked point as the new core point until the absolute value of the difference between the marked point and the core point exceeds the screening difference, stop the operation, and calculate the average value of the data in the filtered prompt index data set as the switch prompt index; When the switch prompt index exceeds the preset prompt threshold, an abnormal prompt message is generated.

[0014] Technical effects and advantages of the present invention: The present invention marks each line in the switch, sets the collection time, collects data of the marked object within the collection time, processes the data using a decision tree method and sets multiple state weights, detects the response time of the marked object, generates abnormal features in combination with the state weights, imports the marked object into different levels of abnormal detection mechanisms according to the abnormal features, monitors the monitoring signals generated by the abnormal detection mechanism in real time and obtains monitoring information, calls the operation records of each marked object, calculates the prompt index after integrating the clustering algorithm with the monitoring information, and determines whether to issue an abnormal prompt based on the prompt index. This can realize automatic recognition and hierarchical detection of abnormal states, effectively improve the system's fault diagnosis accuracy and response speed, enhance the system's response efficiency to abnormal fluctuations, and reduce equipment failure rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the implementation of a switch abnormality prompt method based on AI control in the present invention.

[0016] Figure 2 This is a schematic diagram of the steps of a switch abnormality prompt method based on AI control of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making creative efforts are within the scope of protection of the present invention.

[0018] Example 1: A switch abnormality prompt method based on AI control, such as Figures 1 to 2 As shown, the following steps are included: Step S1: Mark each line in the switch, set a collection time, and collect line data of the marked line within the collection time; Step S2: Process the line data using a decision tree method and set multiple state weights, detect the line data response time of the marked line, and generate abnormal features based on the state weights; Step S3: Selectively import the marked lines into line anomaly detection mechanisms at different levels based on the anomaly characteristics, monitor the monitoring signals generated by the line anomaly detection mechanisms in real time, and obtain monitoring information; Step S4: The closing times of each marked line in the switch are retrieved, integrated with the monitoring information, and then fed into a clustering algorithm to calculate the switch prompt index. It is determined whether to generate abnormal prompt information based on the switch prompt index.

[0019] The specific implementation is as follows: In step S1, each line in the switch is marked and line data of each marked line is collected; The line data of the marker line include the marker line vibration frequency, the marker line ground current, and the marker line noise intensity.

[0020] It should be noted that the information database is a structured collection used to store, manage and retrieve data. The information database in this example is an information database that stores switch information, which contains the stored current ranges of each line under normal operation.

[0021] By collecting the status data of the marked lines and setting multiple status weights, the marked lines are judged for abnormalities, thereby improving the accuracy of line judgment.

[0022] Select a period of time as the collection time, and collect line data for each marked line within the collection time as follows: When collecting the vibration frequency of the marked line and the ground current of the marked line, multiple time points are randomly set within the collection time, and at each time point, the vibration frequency and ground current of the marked line are detected using a vibration sensor and a ground current monitor to obtain the vibration frequency and ground current of the marked line at each time point; When collecting the noise intensity of the marked line, the noise of the marked line is detected at each time point using a noise meter during the collection time, and the detection result is used as the noise intensity at the corresponding time point.

[0023] It should be noted that the time period of the collection time is not uniquely set. For example, the time period of the collection time can be set to half an hour, etc., which will not be described in detail here.

[0024] In step S2, the vibration frequencies of the marker line at adjacent time points of the marker line are subtracted within the acquisition time. Specifically, the vibration frequency of the marker line at the previous time point and the vibration frequency of the marker line at the next time point are subtracted as the instantaneous vibration frequency change at the next time point. The subtraction process is performed once for every two adjacent time points, and multiple instantaneous vibration frequency changes are generated within the acquisition time. The ground currents of the marked line at adjacent time points of the marked line are subtracted during the acquisition time. Specifically, the ground current of the marked line at the latter time point is subtracted from the ground current of the marked line at the previous time point between two adjacent time points as the instantaneous ground current change at the previous time point. Similarly, multiple instantaneous ground current changes are generated during the acquisition time. It should be explained that the greater the instantaneous vibration frequency change or the instantaneous current change, the more likely it is that the circuit will become abnormal.

[0025] The first and last time points within the acquisition time are removed and integrated into a time point data set, and the noise intensity corresponding to each time point in the time point data set is merged into a noise data set; After removing the instantaneous vibration frequency variation at the last time point in the order of time points during the acquisition time, the instantaneous vibration frequency variation is merged into a vibration frequency data set; After removing the instantaneous ground current variation at the first time point in the order of time points during the acquisition time, the instantaneous ground current variation is merged into a ground current data set; The time point dataset, noise dataset, vibration frequency dataset, and ground current dataset are used as input to build a decision tree model. The specific steps are as follows: Data preparation: The time point dataset, noise dataset, vibration frequency dataset, and ground current dataset are used as input, and the target line data is used as output.

[0026] Set the root node: Use the vibration frequency dataset and the ground current dataset to construct two decision trees respectively. Split the instantaneous vibration frequency changes and instantaneous ground current changes in the two datasets downward in descending order. Use the largest instantaneous vibration frequency change and instantaneous ground current change as the root nodes of the two decision trees respectively. Select the time point corresponding to the instantaneous vibration frequency change from the time point dataset and record it. Select the noise intensity at the recording time point from the noise dataset and record it. Repeat the same operation for the instantaneous ground current change. Construct a decision tree: Take the root node as the first parent node, and connect two child nodes through each parent node. When constructing a decision tree using the vibration frequency dataset, select the two largest instantaneous vibration frequency changes that are not nodes in the vibration frequency dataset, in descending order, as the left and right child nodes of the current parent node. From the time point dataset and the noise dataset, call the time point of the instantaneous vibration frequency change of the corresponding node and the noise intensity at the corresponding time point and record them. Use the left and right child nodes as two new parent nodes, and each data point derived from each dataset is incorporated into the decision tree. The same process is used when constructing a decision tree using the ground current dataset. Screening time points: Count the number of layers of the two decision trees, screen out all nodes in the middle layer of the two decision trees and mark them, call the record information of the marked nodes, select the time point corresponding to the maximum noise intensity from the record information of the marked nodes in the two decision trees, and compare the time points selected by the two decision trees; Feature selection: If the time point selected by the two decision trees is the same, the instantaneous vibration frequency change and instantaneous ground current change at the corresponding time point are used as the target line data; if the time point selected by the two decision trees is different, all time points between the two time points are selected in the time point data set, and the instantaneous vibration frequency change and instantaneous ground current change at the middle time point are used as the target line data; Target output: Output the filtered target line data as output data.

[0027] It should be noted that in the decision tree, the parent node and the connected left and right child nodes are in different layers, and the left and right child nodes are in the same layer. When counting the number of layers of the two decision trees, if the number of layers of the decision tree is an even number, all the nodes in the middle two layers of the decision tree are screened out and marked. If the time points selected by the two decision trees are not the same time point and the number of time points between the two time points in the time point data set is an even number, the instantaneous vibration frequency change and the instantaneous ground current change corresponding to the time point with the largest noise intensity in the middle two time points are taken as the target line data.

[0028] Since the vibration frequency and ground current change in one direction when the line fails, the data detected at the time point by the noise intensity is imported into the decision tree for processing to output the target line data, thereby improving the accuracy of subsequent data analysis. According to the target line data, the state weight is set, and the specific steps are as follows: Set input: set the instantaneous vibration frequency change and the instantaneous ground current change in the target line data as input variables respectively; Build a logistic regression model: generate state weights for the input variables as logistic regression intermediate variables: Where i is the serial number of the three input variables, Generate state weights for the i-th input variable, e is the natural base, The i-th logistic regression intermediate variable; Weight setting: when the input variable is the instantaneous vibration frequency change, the generated state weight is set as the vibration frequency weight; when the input variable is the instantaneous ground current change, the generated state weight is set as the ground current weight; The response time of the line data of the marked line includes the response time of the vibration frequency of the marked line and the response time of the ground current of the marked line; the time difference between the time point corresponding to the maximum value of the vibration frequency of the marked line and the time point corresponding to the minimum value of the vibration frequency of the marked line in the collection time is calculated as the response time of the vibration frequency of the marked line, and the response time of the ground current of the marked line is calculated in the same way.

[0029] The vibration frequency weight, the ground current weight, the response time of the vibration frequency of the marked line, and the response time of the ground current of the marked line are weighted and summed to generate the abnormal feature of the marked line.

[0030] It should be noted that the larger the abnormal feature of the marked line, the more likely the marked line is to appear abnormal, and the more it needs to be detected in depth.

[0031] In step S3, the line anomaly detection mechanism is divided into high-level detection mechanism and low-level detection mechanism. When the abnormal feature of the marked line exceeds the preset abnormal threshold, the marked line is matched with the high-level detection mechanism; otherwise, the marked line is matched with the low-level detection mechanism.

[0032] It should be noted that the high level, the low level and the abnormal threshold can be set according to the actual situation, for example, the double-layer detection is set as the low level, the five-layer detection is set as the high level, and the abnormal threshold is set as 0.6, etc. No more analysis is made here.

[0033] After the marked line is imported into the line anomaly detection mechanism of different levels according to the abnormal feature of the marked line, it is judged in each layer whether the line information of the marked line meets the detection condition. If it meets the condition, it enters the next layer, otherwise a monitoring signal is generated. The monitoring information is the number of monitoring signals generated after the marked line passes through the line anomaly detection mechanism. The more the number of monitoring signals, the more likely the marked line is to be abnormal.

[0034] It should be noted that the detection conditions in the line anomaly detection mechanism are set by experimenters in the field according to actual conditions, and the line information acquisition channel of the marked line is not unique, which is not analyzed here.

[0035] In step S4, the more the number of times the line is closed, the faster the corresponding line insulation layer is corroded, the more likely the corresponding line is to be abnormal, and the more likely the switch as a whole is to be abnormal.

[0036] After calling the number of times the switch is closed, the Max-min standardization method is used for standardization processing: , is the number of times the marked line is closed, is the minimum value of the number of times the marked line in the switch is closed, is the maximum value of the number of times the marked line in the switch is closed, is the result of the corresponding marked line after standardization.

[0037] After multiplying the result of the standardization of the number of times the marked line is closed and the number of monitoring signals corresponding to the marked line, a marked line prompt index is generated, and all the marked line prompt indexes in the switch are combined into a prompt index data set; The prompt index data set is input into the clustering algorithm for screening, and the specific steps are as follows: Set the core point: arrange the data in the prompt index data set from small to large, and select the median of the data as the core point; Core clustering: set a screening difference, start with the core point, expand to both sides according to the screening difference, mark the data before or after the core point, compare the absolute value of the difference between the marked point and the core point with the screening difference, if the absolute value of the difference is lower than the screening difference, then the marked point and the core point are merged into the screened prompt index data set, and the marked point is used as a new core point for repeated operation, until the calculation result of the absolute value of the difference between the marked point and the core point exceeds the screening difference, then stop operation.

[0038] Calculate the switch prompt index: calculate the average value of the data in the screened prompt index data set as the switch prompt index.

[0039] Compare the switch prompt index with the preset prompt threshold value, when the switch prompt index exceeds the preset prompt threshold value, generate an abnormal prompt information and send it to the user end.

[0040] It should be noted that the abnormal prompt information is a text message that indicates that a risk has begun to exist. The settings of the screening difference and the prompt threshold are not unique and can be set according to specific circumstances such as the age of the switch. They will not be elaborated here.

[0041] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0042] Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0043] As used herein, the singular forms "a," "an," and "the" may also include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "include," "comprising," "having," and the like specify the presence of stated features, integers, steps, operations, components, parts, or combinations thereof, but do not preclude the presence or addition of one or more other features, integers, steps, operations, components, parts, or combinations thereof. Furthermore, the term "and / or" as used in this specification includes any and all combinations of the relevant listed items.

[0044] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0045] The above description of the disclosed embodiments will enable those skilled in the art to implement or use various modifications of these embodiments, and it will be apparent to those skilled in the art that the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A switch abnormality prompt method based on AI control, characterized in that: The following steps are involved: Step S1: Mark each line in the switch, set a collection time, and collect line data of the marked line within the collection time; In step S1, the line data of the marked line are the marked line vibration frequency, the marked line ground current and the marked line noise intensity; Select a period of time as the collection time, and collect line data for each marked line within the collection time as follows: Multiple time points are randomly set during the acquisition time, and the vibration frequency, ground current, and noise intensity of the marked line are detected at each time point to obtain the vibration frequency, ground current, and noise intensity of the marked line at each time point; Step S2: Process the line data using a decision tree method and set multiple state weights, detect the line data response time of the marked line, and generate abnormal features based on the state weights; Step S3: Selectively import the marked lines into line anomaly detection mechanisms at different levels based on the anomaly characteristics, monitor the monitoring signals generated by the line anomaly detection mechanisms in real time, and obtain monitoring information; Step S4: The closing times of each marked line in the switch are retrieved, integrated with the monitoring information, and then fed into a clustering algorithm to calculate the switch prompt index. It is determined whether to generate abnormal prompt information based on the switch prompt index.

2. The switch abnormality prompt method based on AI control according to claim 1 is characterized by: In step S2, the vibration frequencies of the marker line at adjacent time points of the marker line are subtracted within the acquisition time. The difference between the vibration frequencies of the marker line at the previous time point and the vibration frequencies of the marker line at the next time point is used as the instantaneous vibration frequency change at the next time point. The subtraction process is performed once for every two adjacent time points, and multiple instantaneous vibration frequency changes are generated within the acquisition time. When calculating the instantaneous ground current variation, the difference between the marked line ground current at the latter time point and the marked line ground current at the previous time point in adjacent time points is taken as the instantaneous ground current variation at the previous time point.

3. The switch abnormality prompt method based on AI control according to claim 2 is characterized in that: In step S2, before using the decision tree method, the first time point and the last time point in the acquisition time are removed and integrated into a time point data set, and the noise intensity corresponding to each time point in the time point data set is merged into a noise data set; After removing the instantaneous vibration frequency variation at the last time point in the order of time points during the acquisition time, the instantaneous vibration frequency variation is merged into a vibration frequency data set; After removing the instantaneous ground current variation at the first time point in the order of time points during the acquisition time, the instantaneous ground current variation is merged into a ground current data set; The time point dataset, noise dataset, vibration frequency dataset and ground current dataset are used as input to build a decision tree model to screen the target line data.

4. The switch abnormality prompt method based on AI control according to claim 3 is characterized by: In step S2, the steps for constructing a decision tree model to screen target line data are as follows: Data preparation: Take the time point dataset, noise dataset, vibration frequency dataset, and ground current dataset as input, and the target line data as output; Set the root node: Use the vibration frequency dataset and the ground current dataset to build two decision trees respectively, and select the maximum value in the vibration frequency dataset and the ground current dataset as the root node of the two decision trees respectively; Construct a decision tree: Use the root node as the parent node, and connect each parent node to its two child nodes on the left and right. After the connection is completed, use the child node as the new parent node and repeat the operation until the number of nodes matches the number of data in the vibration frequency dataset or the ground current dataset. Fill the nodes with data from the vibration frequency dataset or the ground current dataset in descending order. Retrieve and record the time point corresponding to the data in each node and the noise intensity at the corresponding time point from the time point dataset and the noise dataset. Screening time points: Count the number of layers of the two decision trees, screen out all nodes in the middle layer of the two decision trees and mark them, select the time point corresponding to the maximum noise intensity from the record information of the marked nodes in the two decision trees, and compare the time points selected by the two decision trees; Feature selection: If the time point selected by the two decision trees is the same, the instantaneous vibration frequency change and instantaneous ground current change at the corresponding time point are used as the target line data; if the time point selected by the two decision trees is different, all time points between the two time points are selected in the time point data set, and the instantaneous vibration frequency change and instantaneous ground current change at the middle time point are used as the target line data; Target output: Output the filtered target line data as output data.

5. The switch abnormality prompt method based on AI control according to claim 4 is characterized in that: In step S2, the state weight is set according to the target line data. The specific steps are as follows: The instantaneous vibration frequency change and instantaneous ground current change in the target line data are set as input variables, and the input variables are used as logistic regression intermediate variables to generate state weights: , where i is the sequence number of the three input variables, Generate state weight for the i-th input variable, e is the natural base, is the intermediate variable of the i-th logistic regression; When the input variable is the instantaneous vibration frequency change, the generated state weight is set to the vibration frequency weight; when the input variable is the instantaneous ground current change, the generated state weight is set to the ground current weight.

6. The switch abnormality prompt method based on AI control according to claim 4 is characterized in that: In step S2, the line data response time of the marking line includes the response time of the marking line vibration frequency and the response time of the marking line ground current; Calculate the time difference between the time point corresponding to the maximum value of the vibration frequency of the marker line and the time point corresponding to the minimum value of the vibration frequency of the marker line within the acquisition time as the response time of the vibration frequency of the marker line. Similarly, calculate the response time of the ground current of the marker line. The vibration frequency weight, the ground current weight, the response time of the vibration frequency of the marked line, and the response time of the ground current of the marked line are weighted and summed to generate the abnormal characteristics of the marked line.

7. The switch abnormality prompt method based on AI control according to claim 6 is characterized by: In step S3, when the abnormal characteristics of the marked line exceed the preset abnormal threshold, the marked line is matched with the preset high-level detection mechanism; otherwise, the marked line is matched with the preset low-level detection mechanism; In each layer of detection, it is determined whether the line information of the marked line meets the detection conditions. If it meets the conditions, it will enter the next layer. If it does not meet the conditions, a monitoring signal will be generated. The monitoring information is the number of monitoring signals generated after the marked line passes the line anomaly detection mechanism.

8. The switch abnormality prompt method based on AI control according to claim 7 is characterized in that: In step S4, the closing times of each marked line are normalized to generate a normalized result of the closing times of the marked lines; The normalized closing times of the marked lines are multiplied by the number of monitoring signals of the corresponding marked lines to generate the marked line prompt index. The marked line prompt indexes are combined into a prompt index dataset and then passed through the clustering algorithm for screening: Arrange the data in the prompt index data set from small to large, select the median of the data as the core point, set the screening difference, mark the data before or after the core point, and when the absolute value of the difference between the marked point and the core point is lower than the screening difference, filter out the marked point and the core point, and repeat the operation with the marked point as the new core point until the absolute value of the difference between the marked point and the core point exceeds the screening difference, stop the operation, and calculate the average value of the data in the filtered prompt index data set as the switch prompt index; When the switch prompt index exceeds the preset prompt threshold, an abnormal prompt message is generated.

Images