A data integration collection method and system for a device management platform

Abstract

The present application relates to the technical field of data processing, in particular to a kind of data integration collection method and system for equipment management platform, comprising: according to the data change in each type operation data time sequence, obtain reference duration, obtain the time of each historical fault repair with the same fault type as the current fault repair as the termination time, the time sequence of each type operation data history with the reference duration as duration, determine the final extension demand degree of the operation data time sequence corresponding to the current fault repair, to obtain target duration, obtain the time sequence of all types of operation data update with the time of the current fault repair as the termination time, duration as target duration, as the time domain matching data of the current fault repair, complete data integration.The present application can obtain higher data integration effect of adaptation by combining the abnormal performance of equipment operation data in actual scene, the influence of multi-source data correlation and data training, improve the data integration efficiency.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a data integration and acquisition method and system for device management platforms. Background Technology

[0002] The equipment service management platform collects equipment operation data and historical work order data, and uses machine learning and data mining technologies to build multi-dimensional profiles of equipment and users. Based on these profiles, the platform can predict demand and optimize resources, and schedule maintenance personnel and spare parts information in advance, which helps reduce resource waste and improve service efficiency.

[0003] The platform regularly imports equipment maintenance record data into the platform's local dataset for data training to increase the prediction accuracy of the equipment record model. When training and importing multi-source collected data, data integration is required first. Specifically, discrete maintenance data is matched with continuous operating data of a certain time period. For example, for a single equipment failure record, all relevant sensor readings (such as temperature, vibration, etc.) collected within 16 hours before its occurrence are matched to achieve time-series alignment and integration of data, thereby forming a wide table to prepare for subsequent data training.

[0004] Existing problem: When matching discrete maintenance record data of equipment with continuous running data of corresponding duration in the traditional way, a fixed duration of running data segment is often selected before the maintenance record. However, in the actual equipment management platform scenario of time-series alignment and integration of multi-source data, if the continuous running data matched with discrete maintenance data is too short, it will be difficult for the running data to reflect the complete abnormal performance of the equipment. If the matched continuous running data is too long, it will increase the redundant computing resources consumed during data training, reduce data processing efficiency, and further lead to a low data integration fit of discrete maintenance data using the traditional method of time-series alignment of running data segments. Summary of the Invention

[0005] This invention provides a data integration and acquisition method and system for device management platforms to solve existing problems.

[0006] The present invention provides a data integration and acquisition method and system for an equipment management platform, which adopts the following technical solution:

[0007] One embodiment of the present invention provides a data integration and acquisition method for a device management platform, the method comprising the following steps:

[0008] Obtain the time series sequence of each type of operation data with the current fault reporting time as the end time and the duration as a preset period; based on the data changes in the time series sequence of each type of operation data, obtain the time of sudden change in the operation data of each type;

[0009] Based on the time interval between the time changes in all types of operational data, obtain the reference duration; obtain the historical time series sequence of each type of operational data, with the time of each historical fault report that is the same as the fault type of the current fault report as the end time and the duration as the reference duration;

[0010] Based on the difference between the time series sequence of each type of operational data and the historical time series sequence of each type of operational data, determine the final extended demand degree of the operational data time series corresponding to the current fault report;

[0011] Based on the final extended demand degree of the operational data time series corresponding to the current fault repair, obtain the target duration; obtain the update time series of all types of operational data with the current fault repair time as the end time and the duration as the target duration, and use it as the time domain matching data for the current fault repair to complete the data integration.

[0012] Furthermore, the specific steps for obtaining the runtime mutation time for each type of runtime data are as follows:

[0013] In each type of running data time series, the mean of the data values ​​at all times is obtained and denoted as the first mean. The inverse normalized value of the absolute value of the difference between the data value at each time and the first mean is obtained and denoted as the device durability performance at each time.

[0014] With a preset threshold N, the local time period consisting of the N closest times to the x-th time period and the x-th time period is obtained. The standard deviation of the data values ​​of all times within the local time period is recorded as the volatility of the x-th time period.

[0015] The product of the inverse proportional value of the device durability performance at each moment and the volatility at each moment is recorded as the operational anomaly coefficient at each moment.

[0016] Based on the operational anomaly coefficients at all times, determine the timing of operational mutations for each type of operational data.

[0017] Furthermore, the specific steps for determining the time of operational mutation for each type of operational data based on the operational anomaly coefficients at all times are as follows:

[0018] In each type of runtime data time series, with time as the horizontal axis and runtime anomaly coefficient as the vertical axis, the least squares method is used to perform curve fitting to obtain the runtime anomaly change curve;

[0019] On the abnormal change curve, obtain the normalized value of the slope of the data point corresponding to the x-th time point, which is denoted as the trend value. Obtain the mean of the data values ​​of all times before the x-th time point, which is denoted as the second mean. Obtain the mean of the data values ​​of all times after the x-th time point, which is denoted as the third mean. Obtain the product of the absolute value of the difference between the second mean and the third mean and the trend value, which is denoted as the probability of abnormal change at the x-th time point.

[0020] The moment when the probability of abnormal mutation is greatest is recorded as the mutation moment for each type of running data.

[0021] Furthermore, the specific steps for obtaining the reference duration are as follows:

[0022] For all types of runtime data, the time interval between any two runtime mutation times is used as the clustering distance. The DBSCAN density clustering algorithm is used to perform clustering operations to obtain several clusters and the cluster center time of each cluster.

[0023] The cluster with the most runtime mutation times is identified and denoted as the target cluster.

[0024] In the target cluster, the inversely proportional normalized value of the time interval between each run mutation time and the cluster center time is obtained and recorded as the first interval value. The product of the first interval value and the number of run mutation times is recorded as the outlier tendency factor for each run mutation time in the target cluster.

[0025] Based on the magnitude of the out-of-state tendency factor at each runtime mutation moment in the target cluster, high-probability out-of-state moments are determined.

[0026] The time interval between the high-probability anomaly moment and the current moment is recorded as the reference duration.

[0027] Furthermore, the specific steps involved in determining the high-probability abnormal situation are as follows:

[0028] In the target cluster, the moment of the operational mutation with the largest outlier tendency factor is denoted as the high-probability outlier moment.

[0029] Furthermore, the specific steps for determining the final extended demand degree of the runtime data timing corresponding to the current fault report are as follows:

[0030] In the historical time series of operational data of type y corresponding to all historical fault reports of the same type as the current fault report, the average value of the data at the same time is obtained in turn to form the historical time series of the average value of operational data of type y.

[0031] In the time sequence of the y-th type of operational data corresponding to the current fault report, obtain the time sequence segment of the y-th type of operational data between the time of the high-probability abnormal situation and the current time.

[0032] The value of the time series analysis of the operational data corresponding to the current fault report is determined based on the difference between the historical time series of the average value of each type of operational data and the time series segments of each type of operational data.

[0033] Obtain the intermediate time in the time series of each type of operation data corresponding to the current fault report, and denote it as the target time. Obtain the normalized value of the difference between the target time and the high-probability abnormal time, and denote it as the expansion coefficient corresponding to the current fault report.

[0034] The product of the time series analysis value of the operational data corresponding to the current fault repair and the expansion coefficient is denoted as the time series expansion demand of the operational data corresponding to the current fault repair.

[0035] Based on the time series expansion requirement of the current fault repair, determine the final time series expansion requirement of the current fault repair.

[0036] Furthermore, the specific steps for determining the time-series analysis value of the operational data corresponding to the current fault report are as follows:

[0037] Obtain the inverse proportional normalized value of the Pearson correlation coefficient between the historical time series of the mean of the y-th type of running data and the time series segment of the y-th type of running data, and denote it as the first uncorrelated value;

[0038] In the historical time series sequence of the mean of the data of type y and the time series sequence of the data of type y, the sum of the absolute values ​​of the differences between two elements with the same ordinal value is recorded as the second uncorrelated value;

[0039] The product of the first irrelevant value and the second irrelevant value is denoted as the current and historical irrelevantness of the y-th type of running data;

[0040] The sum of the uncorrelation values ​​between the current and historical data of all types of operational data is recorded as the operational data time series analysis value corresponding to the current fault report.

[0041] Furthermore, the specific steps for determining the final extended demand degree of the operational data time series corresponding to the current fault repair based on the extended demand degree of the operational data time series corresponding to the current fault repair are as follows:

[0042] The historical time series of all types of operation data corresponding to all historical fault reports with the same fault type as the current fault report is used as input to the pre-trained model to obtain the data packet ratio and gradient norm during the model training and evaluation process.

[0043] The normalized value of the ratio of data packet percentage to gradient norm is denoted as the first ratio. The normalized value of the product of the first ratio and the time series extension demand degree of the current fault repair is denoted as the final time series extension demand degree of the current fault repair.

[0044] Furthermore, the specific steps for obtaining the target duration are as follows:

[0045] The product of the preset basic cycle and the final extended demand degree of the operation data time sequence corresponding to the current fault repair is obtained and denoted as the first product;

[0046] When the target time is greater than or equal to the high-probability abnormal time, the sum of the preset base period and the first product is recorded as the target duration.

[0047] When the target time is less than the high-probability abnormal time, the difference between the preset base period and the first product is recorded as the target duration.

[0048] The present invention also proposes a data integration and acquisition system for a device management platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program stored in the memory to implement the steps of the aforementioned data integration and acquisition method for a device management platform.

[0049] The beneficial effects of the technical solution of the present invention are:

[0050] In this embodiment of the invention, a time series sequence of each type of operational data is obtained, with the current fault reporting time as the end time and a duration of a preset period. A reference duration is obtained based on the data changes in each type of operational data time series. A historical time series sequence of each type of operational data, with the time of each historical fault reporting of the same fault type as the current fault reporting time as the end time and a duration of the reference duration, is also obtained. Based on the difference between each type of operational data time series and its historical counterpart, the final extended demand degree of the operational data time series corresponding to the current fault reporting is determined. Thus, based on the equipment durability and local fluctuation performance analysis, the operational anomaly coefficient at a single moment is obtained, and an operational anomaly change curve is constructed. Operational mutation analysis points corresponding to each type of operational data are obtained from the anomaly change curve. Then, high-probability anomaly points are screened by combining the correlation performance of multiple types of operational data, and collaborative analysis is performed with other types of operational data under the same historical equipment maintenance conditions to obtain the time series extended demand degree. Finally, the time series extended demand degree is corrected through the training effect of historical data, ensuring the accuracy of the selection of time-domain matching data for subsequent current fault reporting. The system obtains the target duration and retrieves the update time sequence of all types of operational data with the current fault reporting time as the end time and the target duration as the time domain matching data for the current fault reporting, thus completing data integration. In this way, the invention can achieve a more suitable data integration effect by combining abnormal behavior of equipment operational data in actual scenarios, the correlation of multi-source data, and the impact on data training, thereby improving data integration efficiency. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating the steps of a data integration and acquisition method for a device management platform according to the present invention.

[0053] Figure 2 A schematic diagram showing the selection of analysis time for fault repair;

[0054] Figure 3 A schematic diagram of abrupt change analysis points on the abnormal change curve of single-type equipment operation data;

[0055] Figure 4 This is a schematic diagram showing the variation curves of the historical time series of the average value of the same type of running data and the time series segments of the running data. Detailed Implementation

[0056] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a data integration and acquisition method and system for a device management platform proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0058] The following description, in conjunction with the accompanying drawings, details a specific solution for a data integration and acquisition method and system for an equipment management platform provided by this invention.

[0059] Please see Figure 1 The diagram illustrates a flowchart of a data integration and acquisition method for a device management platform according to an embodiment of the present invention. The method includes the following steps:

[0060] Step S001: Obtain the time sequence of each type of operation data with the current fault reporting time as the end time and the duration as a preset period; based on the data changes in the time sequence of each type of operation data, obtain the time of sudden change in the operation data of each type.

[0061] This embodiment primarily involves time-series alignment of discrete maintenance data and continuous operational data from equipment within the platform, achieving data integration from multiple sources. Since power equipment is a critical infrastructure in today's rapidly developing industrial environment, the equipment management platform for power equipment integrates several cutting-edge technologies: IoT technology for real-time data acquisition, big data technology for in-depth mining of the potential value of equipment operational data, and artificial intelligence technology for demand forecasting and resource optimization. When processing multi-source data, the platform first performs time-series alignment and integration of discrete maintenance data and continuous operational data to improve the accuracy of subsequent data training. Traditionally, continuous time-domain data with a fixed operating period preceding the discrete maintenance data is selected as the alignment data; for example, matching the operational data readings collected by all relevant sensors within 16 hours prior to a single equipment failure record. However, the traditional time-series alignment method using a preset fixed operational data period has low matching accuracy. Therefore, this embodiment combines the actual equipment management platform scenario to obtain the time-series alignment results between discrete maintenance record data and continuous multi-type operational data to improve data integration efficiency.

[0062] First, preliminary preparations and data collection are required. Specifically, this involves: reading auxiliary analysis data from various modules within the equipment management platform, including the multi-sensor storage module, work order data storage module, and training parameter storage module. The work order data storage module reads real-time maintenance record data awaiting time-series alignment, including fault reporting time and fault detection type information. The multi-sensor storage module reads real-time operational data awaiting time-series alignment, including but not limited to vibration, pressure, temperature, and current data. The training parameter storage module reads the pre-trained model trained based on the historical time-series aligned maintenance record data and operational data. This model is an LSTM (Long Short-Term Memory) prediction model, a predictive model based on a long short-term memory network. It can handle and learn complex patterns and long-term dependencies in time-series data. By integrating data from different sources and systems, the LSTM model can obtain more comprehensive historical information, thereby improving the accuracy of predicting future events. This is a well-known technology. The model's input is a historical single equipment fault record matched with all relevant sensor readings collected within the 16 hours prior to its occurrence. The read data is uploaded to the data acquisition system for subsequent analysis and processing.

[0063] This allows for the real-time acquisition of various types of operational data (such as vibration, pressure, temperature, and current) during equipment operation, as well as the time and type of each fault report, and pre-trained models.

[0064] The data for each type of operation was collected once per second for the duration of the most recent month. The min-max normalization method was used to normalize each type of operation data to unify the units of measurement; this is a well-known technique, and the specific method will not be described here.

[0065] It should be noted that the purpose of this embodiment is to perform time-series alignment and integration of discrete maintenance record data and continuous operation data. Specifically, for a single discrete fault reporting timestamp to be integrated, a continuous time period is selected within the total selectable duration of its corresponding operation data to perform time-series alignment with the discrete reporting data. Specifically, the abnormal behavior of the operation data at each moment within the total selectable duration is first analyzed. Based on the abnormal behavior, the correlation between multiple types of operation data is analyzed to obtain high-probability anomalies within the total selectable duration. Then, the time-series expansion requirement degree is obtained by combining historical data under the same fault detection type and the training impact analysis of historical data. The time-series expansion requirement degree is used to realize the time-series alignment and integration process of discrete and continuous data.

[0066] Further explanation is needed: First, the operational mutation analysis point in the corresponding operational data of the equipment maintenance record needs to be determined. Considering that the operational period aligned with the discrete maintenance data should include the complete abnormal condition performance of the equipment as much as possible, within the total selectable duration of the time axis, the durability performance and local fluctuations of the operational data at a single moment are analyzed to obtain anomaly coefficients. Then, an operational anomaly change curve is constructed, and the steepness of the change curve is used to determine the operational mutation analysis point corresponding to a single type of operational data in the time domain. This operational mutation analysis point may be the time point at which the equipment experienced an abnormal condition. Since traditionally, a preset fixed 16-hour duration of operational data is selected as the time-series alignment data for fault record data, directly selecting 16-hour data for alignment may result in poor fit. Therefore, in this embodiment, twice the preset fixed value is used as the total selectable duration range on the time axis. The duration of the operational data for time-series alignment is determined within the total selectable duration range through the analysis in this embodiment to improve fit. A schematic diagram of the fault maintenance analysis duration selection is shown below. Figure 2 As shown, Figure 2 The time axis is from left to right. Starting from the discrete maintenance record point, the traditional preset selection duration (16 hours) and the total selectable duration range (32 hours) are obtained in reverse time order.

[0067] Preferably, in one embodiment of the present invention, the method for obtaining the runtime mutation time of each type of runtime data includes:

[0068] The preset period (total selectable duration range) is 32 hours, and this will be used as an example for explanation.

[0069] Obtain the time sequence of each type of running data, with the current fault reporting time as the end time and the duration as a preset period.

[0070] For each type of running data time series, obtain the mean of the data values ​​at all times, denoted as the first mean, and obtain the absolute value of the difference between the data value at each time point and the first mean. The inversely proportional normalized value is denoted as the device durability performance at each moment.

[0071] It should be noted that in this embodiment, the following is used: As The inverse proportional normalized value, This is a linear normalization function used to normalize data values ​​to a range of 0 to 1. For each type of operational data, the smaller the difference between the data value at each moment and the data value at other moments, the better the equipment's durability. For example, for equipment temperature parameters, under normal conditions, the equipment temperature should have a small difference from historical temperatures.

[0072] The preset quantity threshold N is 6, and this will be used as an example for explanation.

[0073] In each type of running data time series, obtain the N closest time periods to the x-th time period and the local time period consisting of the x-th time period. The standard deviation of the data values ​​of all times within the local time period is denoted as the volatility of the x-th time period.

[0074] It should be noted that when equipment malfunctions, its various operating parameters will exhibit unstable fluctuations, such as current fluctuations in electrical equipment before overload. Therefore, for each moment, the larger the standard deviation of the current type of operating data within a local time period, the more severe the fluctuations. At the same time, the lower the equipment durability performance at each moment, the greater the abnormal performance of the equipment.

[0075] In each type of operational data time series, the product of the inverse proportional value of the device durability performance at each moment and the volatility at each moment is obtained and recorded as the operational anomaly coefficient at each moment.

[0076] The difference in device durability performance at each moment is calculated by subtracting the difference from 1, which is used as the inverse proportional value of device durability performance at each moment.

[0077] In each type of runtime data time series, with time as the horizontal axis and runtime anomaly coefficient as the vertical axis, curve fitting is performed using the least squares method to obtain the runtime anomaly change curve.

[0078] The least squares method is a well-known technique, and its specific method will not be introduced here.

[0079] On the abnormal change curve, obtain the slope of the data point corresponding to the x-th time point. The normalized value is denoted as the trend value. The mean of the data values ​​at all times before time x (including time x) is denoted as the second mean. The mean of the data values ​​at all times after time x (excluding time x) is denoted as the third mean. The product of the absolute value of the difference between the second mean and the third mean and the trend value is denoted as the probability of abrupt change at time x.

[0080] It should be noted that: with As The normalized value. If the slope of the data point change on the abnormal change curve corresponding to the current type of operating data at each time point is higher, it indicates that the time point is more likely to be the time point of equipment abnormality. At the same time, the greater the difference in the operating abnormality coefficient before and after the time point, the higher the probability that the time point is the time point of abnormality.

[0081] On the abnormal change curve, the moment with the highest probability of abnormal change is recorded as the time of change for each type of running data, and the data point at the time of change is the point of change analysis.

[0082] It should be noted that if multiple moments with the highest probability of abrupt changes exist on the abnormal operation curve, the first one in chronological order will be selected for subsequent analysis. A schematic diagram of the analysis points for abrupt changes on the abnormal operation curve of equipment single-type operation data is shown below. Figure 3 As shown, Figure 3 The horizontal axis of the abnormal change curve represents time (t), the vertical axis represents the abnormal coefficient, and the dots marked on the abnormal change curve represent the points of sudden change analysis.

[0083] Step S002: Obtain the reference duration based on the time interval between the sudden change times of all types of running data; obtain the historical time sequence of each type of running data with the time of each historical fault report (which is the same as the fault type of the current fault report) as the end time and the duration as the reference duration.

[0084] It should be noted that: further, it is necessary to collaboratively analyze the correlation of multi-source data and its impact on data training to obtain the time-series expansion requirement. Considering that when equipment malfunctions, the operational data of different types should be correlated, that is, the differences in the operational mutation analysis points corresponding to the different types of operational data in the time domain should be small. Therefore, based on this rule, the high-probability malfunction points that can most accurately express the time of malfunction occurrence are selected. Furthermore, for the same fault detection type, the changes reflected by the operational data may be similar. If the operational data in the current fault record does not meet this similarity feature, it indicates that the analytical value of the operational data is higher. In this case, the selection range of time-domain operational data can be appropriately expanded.

[0085] It should be further noted that if the selected data period is too long, it will affect the accuracy and efficiency of subsequent data training. Therefore, the time-series alignment selection period for the above analysis has been adjusted based on historical data training performance. To further determine the time of anomaly occurrence, the correlation analysis of multiple types of operational data is combined. When an anomaly occurs in the equipment, it should have a correlation impact on all types of operational data. Therefore, the operational change analysis points for each type of operational data should have small differences on the time axis.

[0086] Preferably, in one embodiment of the present invention, the method for obtaining the reference duration includes:

[0087] For all types of runtime data, the time interval between any two runtime mutation times is used as the clustering distance. The DBSCAN density clustering algorithm is used to perform clustering operations to obtain several clusters and the cluster center time of each cluster.

[0088] DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm, a well-known technique, and its specific method will not be described here. The cluster center time for each cluster is the mean of all runtime mutation times within that cluster.

[0089] Count the number of runtime mutation times in each cluster, and select the cluster with the most runtime mutation times as the target cluster.

[0090] If there are multiple clusters with the largest number of runtime mutation times, then one of them will be randomly selected for subsequent analysis.

[0091] Within the target cluster, obtain the time interval between each mutation time and the cluster center time. The inverse proportional normalized value is denoted as the first interval value. The product of the first interval value and the number of runtime mutation times in the target cluster is denoted as the outlier tendency factor for each runtime mutation time in the target cluster.

[0092] It should be noted that: with As The inverse proportional normalized value indicates that the more runtime mutation moments there are in the target cluster, the more they conform to the clustering pattern of runtime mutation moments of various types of runtime data, and the greater the possibility that they are anomaly moments. At the same time, the closer the runtime mutation moment is to the cluster center moment, the greater the probability that the runtime mutation moment is an anomaly moment.

[0093] Within the target cluster, the moment of operational mutation with the largest anomaly tendency factor is denoted as the high-probability anomaly moment. That is, the operational mutation analysis point corresponding to the high-probability anomaly moment is the high-probability anomaly point.

[0094] The time interval between the high-probability anomaly moment and the current moment is recorded as the reference duration.

[0095] Obtain the historical time sequence of each type of running data, with the time of each historical fault report (which is the same as the fault type reported in the current fault report) as the end time and the duration as the reference duration.

[0096] Step S003: Based on the difference between the time series sequence of each type of operational data and the historical time series sequence of each type of operational data, determine the final extended demand degree of the operational data time series corresponding to the current fault report.

[0097] Preferably, in one embodiment of the present invention, the method for obtaining the final extended demand degree of the operating data timing corresponding to the current fault report includes:

[0098] Taking the y-th type of operational data as an example, in the historical time series of all historical fault reports with the same fault type as the current fault report, the average value of the data at the same time is obtained in turn to form the historical time series of the average value of the y-th type of operational data.

[0099] In the time sequence of the y-th type of operating data corresponding to the current fault report, obtain the time sequence segment of the y-th type of operating data between the time of the high probability of abnormal situation and the current time.

[0100] It should be noted that the historical time series sequence of the mean of the data of type y is of the same length as the time series sequence segment of the data of type y. A schematic diagram illustrating the change curves of the historical time series sequence of the mean of the data of the same type and the time series sequence segment of the data is shown below. Figure 4 As shown. Figure 4 In the diagram, the horizontal axis represents time (t), and the vertical axis represents the operational data values. The current maintenance record's operational data change curve and the historical operational average change curve are in different time periods, but their durations are both the time interval between a high-probability anomaly point (a high-probability anomaly moment) and a discrete maintenance time point (the current moment). Figure 4 The two change curves are displayed aligned.

[0101] Obtain the Pearson correlation coefficient between the historical time series of the mean of the data of type y and the time series segment of the data of type y. The inversely proportional normalized value is denoted as the first uncorrelated value.

[0102] Among them, with As The inverse proportional normalized value, the Pearson correlation coefficient is a well-known technique, and the specific method will not be introduced here. The smaller the Pearson correlation coefficient, the less similar the two data sequences are.

[0103] In the historical time series sequence of the mean of the data of type y and the time series sequence segment of the data of type y, obtain the absolute value of the difference between two elements with the same ordinal value, and record the sum of the absolute values ​​of the differences between all two elements with the same ordinal value as the second uncorrelated value.

[0104] The product of the first irrelevant value and the second irrelevant value is denoted as the current and historical irrelevantness of the data of the y-th type.

[0105] The sum of the uncorrelation values ​​between the current and historical data of all types of operational data is recorded as the operational data time series analysis value corresponding to the current fault report.

[0106] It should be noted that: the larger the first irrelevance value, the greater the difference between the current analysis of various types of operational data and the historical data change patterns under the same fault. At the same time, the larger the second irrelevance value, that is, the greater the difference in operational data values, the less the operational data corresponding to the maintenance record data to be aligned in the current analysis conforms to the general fault characteristics. Therefore, the more important it is to analyze its operational data, that is, the higher the scalable value of the operational data.

[0107] Obtain the midpoint of the time series of operational data for each type corresponding to the current fault report, and denote it as the target time. Then, obtain the difference between the target time and the high-probability abnormal time. The normalized value is denoted as the expansion coefficient corresponding to the current fault report.

[0108] The product of the time series analysis value of the operational data corresponding to the current fault repair and the expansion coefficient is denoted as the time series expansion demand of the operational data corresponding to the current fault repair.

[0109] It should be noted that: with As The normalized value. Since the total selectable duration is twice the traditional preset selection duration, if the high probability of an anomaly time is shorter than the target time, the traditional preset selection duration needs to be extended; conversely, if the high probability of an anomaly time is longer than the target time, the traditional preset selection duration needs to be reduced.

[0110] The historical time series of all types of operational data corresponding to all historical fault reports with the same fault type as the current fault report is used as input to the pre-trained model to obtain the data packet ratio and gradient norm during the model training and evaluation process.

[0111] Among them, the proportion of data packets and the gradient norm can be calculated and analyzed during the training and evaluation of the LSTM prediction model. This is a well-known technique, and the specific method will not be introduced here.

[0112] It should be noted that when a device alarms, continuous operational data for a certain period prior to the alarm needs to be selected for fault analysis. Since the duration of the historical time series of operational data corresponding to each type of historical fault report is a traditionally preset selection duration, in this embodiment, all historical time series of operational data corresponding to all types of historical fault reports with the same fault type as the current fault report are taken to form a dataset. This dataset is then input into a pre-trained model to obtain the data packet ratio and gradient norm during model training and evaluation. In time series prediction, the data packet ratio refers to the proportion of data in different time periods to the total data volume. The data packet ratio reflects the diversity and representativeness of the dataset. A reasonable data packet ratio helps improve the model's generalization ability. By ensuring that the model encounters sufficiently diverse data during training, the bias of the model in practical applications can be reduced. The gradient norm refers to the length or size of the gradient vector during deep learning model training. It can help adjust the learning rate or use gradient pruning techniques to stabilize the training process. Therefore, both packet proportion and gradient norm play crucial roles in the training and evaluation of LSTM prediction models. Packet proportion relates to data quality and distribution, while gradient norm relates to the stability and efficiency of model training. Thus, by analyzing the packet proportion and gradient norm of historical time-series data corresponding to all types of historical fault reports with the same fault type as the current fault report during model training, we can determine whether the dataset obtained under the traditional preset selection duration for historical fault reports with the same fault type as the current fault report is appropriate. If it is inappropriate, the traditional preset selection duration needs to be adjusted, i.e., the time-series expansion requirement of the operational data corresponding to the current fault report needs to be adjusted to obtain the final time-series expansion requirement of the operational data corresponding to the current fault report.

[0113] Obtain the ratio of data packet percentage to gradient norm. The normalized value, denoted as the first ratio, is then multiplied by the product of the first ratio and the time-series extended demand degree of the operational data corresponding to the current fault report. The normalized value is denoted as the final extended demand degree of the time sequence of the current fault report.

[0114] It should be noted that: with and As respectively and The normalized value. If the gradient norm of all historical fault reports of the same type as the current fault report within the reference time before the time of the current fault report is larger during the model training and evaluation process, it indicates that the training effect is worse. Conversely, if the proportion of data packets during the model training and evaluation process is larger, it indicates that specific types of data packets occupy a larger proportion within the reference time before the time of the historical fault report. That is, the proportion of data packets is used as the confidence parameter of the inverse proportional value (i.e., the reciprocal) of the gradient norm. Thus, the first ratio is used as the correction magnitude to adjust the time series expansion requirement of the running data, and the final time series expansion requirement of the running data is obtained.

[0115] Step S004: Based on the final extended demand degree of the operation data time series corresponding to the current fault repair, obtain the target duration; obtain the update time series of all types of operation data with the current fault repair time as the end time and the duration as the target duration, and use it as the time domain matching data for the current fault repair to complete the data integration.

[0116] It should be noted that: further, it is necessary to achieve time-series alignment between discrete maintenance data and continuous operation data of the equipment based on the final extended demand degree of the operation data time series.

[0117] Preferably, in one embodiment of the present invention, the method for obtaining the target duration includes:

[0118] The preset base period (traditional preset selection duration) is 16 hours, and this will be used as an example for explanation.

[0119] The product of the preset basic cycle and the final extended demand degree of the operation data time sequence corresponding to the current fault report is obtained and denoted as the first product.

[0120] When the target time is greater than or equal to the high-probability abnormal time, the sum of the preset basic period and the first product is recorded as the target duration.

[0121] When the target time is less than the high-probability abnormal time, the difference between the preset base period and the first product is recorded as the target duration.

[0122] The system acquires the update time sequence of all types of operational data with the current fault reporting time as the end time and the duration as the target duration. This sequence is used as the time-domain matching data for the current fault reporting, thereby achieving timestamp alignment between discrete maintenance data and continuous operational data, and completing data integration.

[0123] It should be noted that the specific implementation of the complete process for improving data integration and collection for equipment management platforms is as follows: First, for each fault repair record, its time-domain aligned data (i.e., the time-domain matching data for each fault report) is associated with the equipment operation data. Then, a dataset is constructed from all the repair data to be trained and the matched aligned operation data. Half of the dataset is randomly selected as the training set, and the other half as the validation set. During training, the loss function is the mean squared error (MSE) or the mean absolute error (MAE). The Adam optimizer is used, with a learning rate set between 0.001 and 0.01, combined with a momentum factor to avoid local optima. The training data is imported into the local database of the equipment management platform's prediction model. Training is terminated when the error parameter no longer decreases to prevent overfitting; this is a well-known technique. Finally, a data integration and collection method with higher adaptability to equipment management platforms is implemented.

[0124] The present invention also provides a data integration and acquisition system for a device management platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program stored in the memory to implement the steps of the aforementioned data integration and acquisition method for a device management platform.

[0125] This invention is now complete.

[0126] In summary, in this embodiment of the invention, a time-series sequence of operational data for each type is obtained, with the current fault reporting time as the end time and a duration of a preset period. Based on the data changes in each type of operational data time-series sequence, a reference duration is obtained. A historical time-series sequence of operational data for each type, with the time of each historical fault reporting of the same fault type as the current fault reporting time as the end time and a duration equal to the reference duration, is obtained. Based on the differences between each type of operational data time-series sequence and its historical counterpart, the final expansion requirement of the operational data time-series corresponding to the current fault reporting time is determined, thereby obtaining the target duration. Finally, an update time-series sequence of all types of operational data, with the current fault reporting time as the end time and a duration equal to the target duration, is obtained as the time-domain matching data for the current fault reporting time-series, completing data integration. This invention can achieve a more adaptive data integration effect by combining abnormal performance of equipment operational data in actual scenarios, the correlation of multi-source data, and the impact on data training, thus improving data integration efficiency.

[0127] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A data integration collection method for a device management platform, characterized in that, The method comprises the following steps: Obtain each type of operation data time sequence with the time of the current fault repair as the terminal time and a preset period as the time length; obtain the operation mutation time of each type of operation data according to the data change in each type of operation data time sequence; Obtain the reference time length according to the time interval between the operation mutation times of all types of operation data; obtain each type of operation data historical time sequence with the time of each historical fault repair of the same type as the current fault repair as the terminal time and the reference time length as the time length; Determine the final extension demand degree of the operation data time sequence corresponding to the current fault repair according to the difference between each type of operation data time sequence and each type of operation data historical time sequence; Obtain the target time length according to the final extension demand degree of the operation data time sequence corresponding to the current fault repair; obtain all types of operation data update time sequences with the time of the current fault repair as the terminal time and the target time length as the time length as the time domain matching data of the current fault repair, and complete data integration; The specific steps of determining the final extension demand degree of the operation data time sequence corresponding to the current fault repair include the following: In the yth type of operation data historical time sequence corresponding to all historical fault repairs of the same type as the fault type of the current fault repair, obtain the data mean value of the same time in turn to form the yth type of operation data mean value historical time sequence; In the yth type of operation data time sequence corresponding to the current fault repair, obtain the yth type of operation data time sequence segment between the high-probability abnormal time and the current time; Determine the operation data time sequence analysis value degree corresponding to the current fault repair according to the difference between each type of operation data mean value historical time sequence and each type of operation data time sequence segment; Obtain the intermediate time in each type of operation data time sequence corresponding to the current fault repair, denoted as the target time, and obtain the normalized value of the difference between the target time and the high-probability abnormal time, denoted as the extension coefficient corresponding to the current fault repair; The product of the operation data time sequence analysis value degree corresponding to the current fault repair and the extension coefficient is denoted as the operation data time sequence extension demand degree corresponding to the current fault repair; Determine the final extension demand degree of the operation data time sequence corresponding to the current fault repair according to the operation data time sequence extension demand degree corresponding to the current fault repair.

2. The data integration and collection method for a device management platform according to claim 1, wherein, The specific steps of obtaining the operation mutation time of each type of operation data include the following: In each type of operation data time sequence, obtain the mean value of the data value of all times, denoted as the first mean value, and obtain the inverse proportional normalized value of the absolute value of the difference between the data value of each time and the first mean value, denoted as the equipment endurance performance degree of each time; A preset number threshold N is obtained, a local period composed of the first N times closest to the xth time in time and the xth time is obtained, the standard deviation of the data value of all times in the local period is denoted as the volatility of the xth time; The product of the inverse proportional value of the equipment endurance performance degree of each time and the volatility of each time is denoted as the operation abnormality coefficient of each time; Determine the operation mutation time of each type of operation data according to the operation abnormality coefficient of all time points.

3. The data integration and collection method for a device management platform according to claim 2, wherein, The specific steps of determining the operation mutation time of each type of operation data according to the operation abnormality coefficient of all time points include the following: In the time sequence sequence of each type of operation data, use the least square method to perform curve fitting with time points as the horizontal axis and operation abnormality coefficients as the vertical axis to obtain an operation abnormality change curve; On the operation abnormality change curve, obtain the normalized value of the slope of the data point corresponding to the xth time point, denoted as the trend value, obtain the mean value of the data values of all time points before the xth time point, denoted as the second mean value, obtain the mean value of the data values of all time points after the xth time point, denoted as the third mean value, and obtain the product of the absolute value difference between the second mean value and the third mean value and the trend value, denoted as the abnormality mutation possibility of the xth time point; The time point with the largest abnormality mutation possibility is denoted as the operation mutation time of each type of operation data.

4. The data integration and collection method for a device management platform according to claim 1, wherein, The specific steps of obtaining the reference time length include the following: For the operation mutation time of all types of operation data, use the time interval between any two operation mutation time points as the clustering distance, and use the DBSCAN density clustering algorithm for clustering operation to obtain a plurality of clustering clusters and the clustering center time point of each clustering cluster; Obtain the clustering cluster with the largest number of operation mutation time points, denoted as the target clustering cluster; In the target clustering cluster, obtain the inverse proportional normalized value of the time interval between each operation mutation time point and the clustering center time point, denoted as the first interval value, and obtain the product of the first interval value and the number of operation mutation time points, denoted as the abnormality trend factor of each operation mutation time point in the target clustering cluster; Determine the high-probability abnormality time point according to the size of the abnormality trend factor of each operation mutation time point in the target clustering cluster; Obtain the time interval between the high-probability abnormality time point and the current time point, denoted as the reference time length.

5. The data integration and collection method for a device management platform according to claim 4, wherein, The specific steps of determining the high-probability abnormality time point include the following: In the target clustering cluster, the operation mutation time point with the largest abnormality trend factor is denoted as the high-probability abnormality time point.

6. The data integration and collection method for a device management platform according to claim 1, wherein, The specific steps of determining the operation data time sequence analysis value degree corresponding to the current fault repair include the following: Obtain the inverse proportional normalized value of the Pearson correlation coefficient between the yth type of operation data mean historical time sequence and the yth type of operation data time sequence segment, denoted as the first unrelated value; In the yth type of operation data mean historical time sequence and the yth type of operation data time sequence segment, obtain the sum value of the absolute value difference between the two elements with the same sequence number, denoted as the second unrelated value; Obtain the product of the first unrelated value and the second unrelated value, denoted as the current and historical unrelatedness of the yth type of operation data; Obtain the sum value of the current and historical unrelatedness of all types of operation data, denoted as the operation data time sequence analysis value degree corresponding to the current fault repair.

7. The data integration and collection method for a device management platform according to claim 1, wherein, The specific steps of determining the final operation data time sequence expansion requirement degree corresponding to the current fault repair according to the operation data time sequence expansion requirement degree corresponding to the current fault repair include the following: Corresponding to all historical fault reports of the same fault type as the current fault report, all type operation data historical time series are taken as the input of the pre-trained model to obtain the data packet ratio and the gradient norm in the model training and evaluation process; A normalized value of the ratio of the data packet ratio and the gradient norm is obtained, denoted as a first ratio, and a normalized value of the product of the first ratio and the running data time sequence final expansion demand degree corresponding to the current fault report is denoted as the running data time sequence final expansion demand degree corresponding to the current fault report.

8. The data integration and collection method for a device management platform according to claim 1, wherein, The specific steps of the obtaining the target duration include the following: A product of a preset basic period and the running data time sequence final expansion demand degree corresponding to the current fault report is obtained, denoted as a first product; When the target moment is greater than or equal to the high-probability abnormal moment, a sum value of the preset basic period and the first product is denoted as the target duration; When the target moment is less than the high-probability abnormal moment, a difference value of the preset basic period minus the first product is denoted as the target duration.

9. A data integration collection system for a device management platform, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, The computer program is executed by the processor to realize the steps of the data integration and collection method for the device management platform as claimed in any one of claims 1-8.

Images