Joint time sequence prediction and abnormal early warning method for multi-dimensional monitoring indexes

Abstract

The invention discloses a multi-dimensional monitoring index combined time sequence prediction and abnormal early warning method, which relates to the technical field of equipment abnormal early warning, and comprises the following steps: carrying out calibration calculation on real-time detection data according to a correlation coefficient of data of each dimension and time sequence delay duration of detection equipment; obtaining a reference calculation value corresponding to each piece of dimension data; determining to generate an early warning signal according to the reference calculation numerical value corresponding to each dimension data, selecting the longest historical operation duration of the machine equipment as the calibration operation duration, setting a data acquisition time point by taking the longest historical operation duration as the reference, acquiring multi-dimension data, obtaining node numerical values of each dimension data at different time points through analysis, and calculating the early warning signal according to the node numerical values. And determining the correlation coefficient of each dimension data and the operation duration and the time sequence delay duration of each detection device, and finally determining and generating an early warning signal based on the reference calculation value corresponding to each dimension data, thereby realizing the early precise early warning of the machine equipment abnormity.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of equipment anomaly early warning, and particularly relates to a joint time sequence prediction and early anomaly warning method for multi-dimensional monitoring indexes. BACKGROUND

[0002] With the increasing complexity of modern industrial equipment and systems, it is particularly important to monitor the running state in multiple dimensions. In industrial production, machine equipment and other periodic operation equipment are the core, and their stable and reliable operation is directly related to production efficiency and product quality. Traditional equipment monitoring systems usually rely on single-dimensional monitoring indexes (such as temperature, pressure, vibration, etc.) and use a simple threshold alarm mechanism for real-time monitoring. This mode provides basic information about the equipment state to a certain extent, but its limitations are increasingly evident in the face of increasingly complex equipment failure modes and higher production efficiency requirements.

[0003] Most of the existing methods directly analyze the raw data collected by the detection equipment without considering the delay of the collected signals caused by the differences in hardware performance and transmission paths of different detection equipment, ignoring the dynamic nature of the running state of machine equipment over time, resulting in the hysteresis of time sequence data not being effectively handled. For example, vibration sensor data transmission delay, temperature sensor delay, causing multi-dimensional data to be misaligned on the time axis, making it impossible to achieve accurate time sequence correlation analysis. At the same time, the coupling effect between the dimensional data and the running time of the machine equipment is ignored, that is, there are differences in the normal parameter range of the machine equipment at different running stages, such as just starting, stable running, and continuous running after normal running. Using fixed thresholds or static models to judge these uncalibrated raw data seriously affects the accuracy and timeliness of early anomaly warning, easily leading to a decrease in warning sensitivity, false negatives or false positives, and thus affecting the accuracy and timeliness of the warning. Therefore, a joint time sequence prediction and early anomaly warning method for multi-dimensional monitoring indexes is proposed. SUMMARY

[0004] The purpose of the present application is to provide a joint time sequence prediction and early anomaly warning method for multi-dimensional monitoring indexes, which solves the technical problem that the existing monitoring system directly uses the raw values of each detection equipment to monitor the data, ignoring the data acquisition delay of each detection equipment and the coupling effect between the dimensional data and the running time of the machine equipment, resulting in inaccurate reflection of the actual running state of the equipment and thus affecting the accuracy and timeliness of the warning.

[0005] A joint time sequence prediction and early anomaly warning method for multi-dimensional monitoring indexes, comprising the following steps:

[0006] Step 1: Select the maximum value of the running time of the machine equipment in multiple historical running times as the calibrated running time of the machine equipment.

[0007] Step two: set multiple data acquisition time points according to the calibration running time length, and acquire multi-dimensional data of the machine equipment at different data acquisition time points through multiple detection devices to obtain multiple sets of multi-dimensional data sets;

[0008] Step three: analyze the multi-dimensional data at each data acquisition time point to obtain the node values corresponding to different dimensional data at each data acquisition time point respectively;

[0009] Step four: analyze the correlation between each node value and the running time length of the machine equipment to obtain the correlation coefficient between each dimensional data and the running time length;

[0010] Step five: obtain and analyze the signal receiving time points and data acquisition time points of different detection devices at different data acquisition time points; according to the analysis result, obtain the time sequence delay time length corresponding to each detection device and the data acquisition time point respectively;

[0011] Step six: according to the correlation coefficient of each dimensional data and the time sequence delay time length of each detection device, the real-time detection data is calibrated and calculated to obtain the reference calculation value corresponding to each dimensional data respectively;

[0012] Step seven: the sum of the product of the reference calculation value Kj corresponding to each dimensional data respectively and the preset parameter coefficient βj corresponding to each dimensional data is taken as the comprehensive judgment value of the machine equipment, when the comprehensive judgment value is greater than the preset warning threshold Y5, an early warning signal is generated, otherwise, no processing is performed.

[0013] As a further scheme of the application: the specific way of obtaining the node values corresponding to different dimensional data at each data acquisition time point respectively is:

[0014] S1: randomly select one from each data acquisition time point as an analysis time point;

[0015] S2: randomly select one from different dimensional data as analysis dimensional data;

[0016] S3: obtain the mean value of each analysis dimensional data, and obtain the absolute value of the difference between each analysis dimensional data and the mean value, compare the absolute value with the preset value Y1 to obtain the node value J11 at the analysis time point;

[0017] S4: repeat steps S2-S3, that is, obtain the node values Jj1 corresponding to each dimensional data respectively at the analysis time point, wherein j is different dimensional data;

[0018] S5: repeating steps S1-S4, that is, obtaining the node value Jjr of each dimension data respectively corresponding to different data collection time points, wherein r is different data collection time points, r=1, 2, …, a2, a2 is the total number of data collection time points, and a2 is a positive integer and greater than 8.

[0019] As a further scheme of the present application, the specific way of comparing and analyzing the absolute value of the difference with the preset value Y1 is:

[0020] The number of analysis dimension data with the absolute value of the difference greater than the preset value Y1 is marked as b, when b is greater than the preset value Y2, then the mean value of the maximum value and the minimum value in each analysis dimension data is taken as the node value J11 of the analysis dimension data at the analysis time point, and when b is less than or equal to the preset value Y2, then the mean value of each analysis dimension data is taken as the node value J11 of the analysis dimension data at the analysis time point.

[0021] As a further scheme of the present application, when the total number of analysis dimension data is even, the preset value Y2 is half of the total number of analysis dimension data, and when the total number of analysis dimension data is odd, the preset value Y2 is half of the total number of analysis dimension data plus one.

[0022] As a further scheme of the present application, the specific way of obtaining the correlation coefficient between each dimension data and the running time is:

[0023] S01: selecting the analysis dimension data in step S2 as target dimension data;

[0024] S02: obtaining the machine equipment running time Tr respectively corresponding to each data collection time point of the target dimension data, that is, the cumulative running time from the start of the machine equipment running to the rth data collection time point, correlating the node value J1r respectively corresponding to different data collection time points of the target dimension data with the machine equipment running time Tr respectively corresponding to each data collection time point, and then using the Pearson correlation coefficient calculation formula to obtain the correlation coefficient R between the target dimension data and the equipment running time, and analyzing the correlation coefficient R to obtain the correlation coefficient X1 between the target dimension data and the equipment running time;

[0025] S03: repeating steps S01-S02, that is, obtaining the correlation coefficient Xj between each dimension data and the equipment running time.

[0026] As a further scheme of the present application, the specific way of analyzing the correlation coefficient R is:

[0027] When the correlation coefficient R is greater than Y3 +When the correlation coefficient R is less than Y3

[0028] When the correlation coefficient R is less than Y3 - When the correlation coefficient R is less than Y3

[0029] When the correlation coefficient R is less than Y3 - When the correlation coefficient R is less than Y3 + When the correlation coefficient R is less than Y3

[0030] As a further scheme of the present application, the specific way of obtaining the time sequence delay duration corresponding between each detection device and the data collection time point is as follows:

[0031] S11: randomly selecting one from different detection devices as a target device;

[0032] S12: obtaining the signal receiving time point and the data collection time point of the target device at different data collection time points, and obtaining the absolute value of the difference between the signal receiving time point and the data collection time point, which is the time difference value Er of the target device at different data collection time points, obtaining the discrete value U of the time difference value Er and analyzing it to obtain the time sequence delay duration F1 of the target device;

[0033] S13: repeating steps S11-S12, so as to obtain the time sequence delay duration Fj corresponding between each detection device and the data collection time point.

[0034] When the discrete value U is greater than the preset value Y4, the mean value of the maximum value and the minimum value in the time difference value Er is the time sequence delay duration F1 of the target device, and when the discrete value U is less than or equal to the preset value Y4, the mean value of the time difference value Er is the time sequence delay duration F1 of the target device.

[0035] As a further scheme of the present application, the specific way of obtaining the time sequence delay duration corresponding between each detection device and the data collection time point is as follows:

[0036] The real-time detection data of each dimension data is marked as Hj, and the real-time detection data Hj of each dimension data is subtracted by the product between the corresponding time sequence delay duration Fj and the correlation coefficient Xj, thereby obtaining the reference calculation value corresponding to each dimension data.

[0037] Compared with the prior art, the present application has the following beneficial effects:

[0038] (1) The present application directly reflects the linear influence trend of the running time length on different dimension data by evaluating the correlation coefficient between each dimension data and the running time length of the equipment, accurately quantifies the internal influence degree and direction of the machine equipment running time length on each monitoring indicator, that is, the natural growth or decline trend over time, provides a basis for data calibration;

[0039] (2) The present application calculates the time delay length of each detection equipment by comparing the signal receiving time point and the data acquisition time point, accurately measures the inherent time delay of different detection equipment in the data acquisition process, makes up for the problem of ignoring sensor response time in traditional monitoring, and clearly defines the time difference between collected data and actual physical event occurrence, providing accurate basis for subsequent time alignment of data and key parameters for data calibration;

[0040] (3) The present application realizes the calibration calculation of real-time detection data of each dimension data by obtaining the corresponding reference calculation value of each dimension data, eliminates the systematic trend deviation caused by the running time length of machine equipment and the time sequence deviation caused by the time sequence delay of detection equipment, so that the reference calculation value used for early warning calculation and analysis is closer to the real state of machine equipment at the actual physical time point, improving the prediction and early warning accuracy;

[0041] (4) The present application determines the generation of early warning signal through the corresponding reference calculation value of each dimension data, effectively solves the problem that the existing monitoring system directly uses the original data of detection equipment, ignores the data acquisition delay and the coupling effect of each dimension data and equipment running time length, and the data cannot accurately reflect the actual running state of equipment, affecting the accuracy and timeliness of early warning, effectively reduces the non-planned downtime, significantly improves the reliability and production efficiency of equipment operation, and reduces the prediction error and false alarm caused by data quality problems. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The figure is a schematic diagram of the method framework structure of the present application. DETAILED DESCRIPTION

[0043] The technical solutions of the present application will be described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0044] Embodiment one: please refer to Figure 1 The present application provides a joint time sequence prediction and early warning method for multi-dimensional monitoring indicators, which specifically includes the following steps:

[0045] Step one: obtain multiple historical running time lengths of the machine equipment, the historical running time length refers to the single running time length of the machine equipment, and the calibration running time length of the machine equipment is obtained according to the single running time length corresponding to the multiple historical running data of the machine equipment, and the specific mode is:

[0046] Obtain multiple historical running time lengths of the machine equipment, and take the maximum value in the multiple historical running time lengths as the calibration running time length of the machine equipment; select the longest historical running time length as the calibration running time length, so as to ensure that the subsequent data is obtained as much as possible within the running time;

[0047] It should be noted that the running time length of the machine equipment refers to the running time length of the machine equipment in a single independent running cycle, rather than the cumulative running time length of the machine equipment, which belongs to periodic running type monitoring, and the data structure is constructed and aligned around a single running cycle;

[0048] Taking the single running cycle of the machine equipment as the core, the longest historical running time length is taken as the calibration running time length, so as to ensure that the data acquisition covers the complete running cycle, which is suitable for the running characteristics of periodic equipment and avoids the problem of fuzzy cycle boundary in cumulative running time length analysis, and lays a targeted foundation for subsequent data alignment and time sequence analysis. Compared with the general continuous time sequence data processing framework, it is more focused on the particularity of periodic scene.

[0049] Step two: set multiple data acquisition time points according to the calibration running time length, the time intervals between each data acquisition time point are the same, set multiple machine equipments to run according to the calibration running time length, and acquire multiple sets of multi-dimensional data sets based on the data acquisition time points by collecting and acquiring multi-dimensional data of each machine equipment at different data acquisition time points through different detection devices;

[0050] Within the determined calibration running time length, multiple data acquisition points are set in an equal interval manner, and multi-dimensional data is collected from multiple machine equipments in parallel at these preset time points by using different detection devices, so as to construct a comprehensive, detailed and time-aligned equipment running state. It should be noted that the default model of the multiple machine equipments is the same, and the use environment is also the same.

[0051] Step three: analyze the multi-dimensional data at each data acquisition time point, and then obtain the node values corresponding to different dimensional data at each data acquisition time point, and the specific mode is:

[0052] S1: randomly select one from each data acquisition time point as an analysis time point;

[0053] S2: randomly select one of the different dimension data as the analysis dimension data;

[0054] S3: obtain the mean value of each analysis dimension data, and obtain the absolute value of the difference between each analysis dimension data and the mean value, mark the number of analysis dimension data whose absolute value of the difference is greater than a preset value Y1 as b, when b is greater than a preset value Y2, then take the mean value of the maximum value and the minimum value of each analysis dimension data as the node value J11 of the analysis dimension data at the analysis time point, when b is less than or equal to the preset value Y2, then take the mean value of each analysis dimension data as the node value J11 of the analysis dimension data at the analysis time point, wherein the specific value of the preset value Y1 is determined by relevant personnel according to actual needs, and the specific value of the preset value Y2 is half of the total number of analysis dimension data when the total number of analysis dimension data is even, and is half of the total number of analysis dimension data plus one when the total number of analysis dimension data is odd;

[0055] For example, the analysis dimension data of each machine equipment at the analysis time point is marked as Bi, the mean value Bp of Bi is obtained, wherein i represents different analysis dimension data, i = 1, 2, …, a1, a1 is the total number of analysis dimension data, a1 is a positive integer and a1 is greater than 4, the number b of analysis dimension data satisfying |Bi-Bp|>Y1 is obtained, when b is greater than a preset value Y2, then J11=(B max +B min ) / 2, when b is less than or equal to the preset value Y2, then J11=Bp; when a is even, Y2=a / 2; when a is odd, Y2=(a+1) / 2;

[0056] For different time points and different dimension data, by comparing the number of abnormal values b with the threshold value Y2 which is about half of the total data, the node value calculation method is dynamically selected, that is, the mean value or the mean value of the maximum and minimum values: when the number of abnormal values exceeds half, it is considered that the overall discreteness of the data is too high, and the mean value of the extreme values is used to replace the mean value to reduce the abnormal interference; otherwise, the mean value is directly used. This processing method is different from simply excluding abnormal values, which may lose effective information, and is also different from the conventional method of using median or truncated mean, but dynamically adjusts according to the proportion of abnormal values, and takes into account data stability and anti-interference.

[0057] S4: repeat steps S2-S3, that is, obtain the node value Jj1 corresponding to each dimension data at the analysis time point, wherein j is different dimension data;

[0058] S5: Repeat steps S1-S4, i.e. obtain the node values Jjr corresponding to each dimension data at different data collection time points respectively, where r is different data collection time points, r = 1, 2, …, a2, a2 is the total number of data collection time points, a2 is a positive integer and a is greater than 8;

[0059] The data collected by multiple machines at the same time point and in the same dimension are processed to obtain the most representative node values. The average value is calculated, and a conditional judgment based on the preset thresholds Y1 and Y2 is introduced to robustly process abnormal or discrete values in the data. When the data dispersion is large, the average of the maximum value and the minimum value is used to avoid excessive pulling of the average value by extreme abnormal values. When the data dispersion is small, b≤Y2, the average value is directly used to maintain the integrity of the data. This processing method can effectively suppress the influence of measurement noise and accidental abnormal data on the analysis results, improve the anti-interference ability of the node values, make them less affected by individual abnormal measurement values, improve the representativeness of the data: the obtained node values can more truly reflect the overall trend or reference state of the dimension data at a specific time point, providing high-quality input for subsequent analysis: the processed node values are more stable and reliable, providing high-quality input for the correlation analysis in step four and the training of the subsequent prediction model.

[0060] Step four: analyze the node values corresponding to each dimension data at different data collection time points respectively, and obtain the correlation coefficient between each dimension data and the machine equipment running time according to the analysis results, in particular:

[0061] S01: select the analysis dimension data in step S2 as the target dimension data;

[0062] S02: obtain the machine equipment running time Tr corresponding to the target dimension data at each data collection time point, i.e. the cumulative running time from the start of the machine equipment running to the rth data collection time point. Associate the node values J1r corresponding to the target dimension data at different data collection time points with the machine equipment running time Tr corresponding to each data collection time point to form the node coordinates Dr(Tr, J1r) corresponding to the target dimension data at each data collection time point, and then obtain a series of target dimension data coordinate points. Use the Pearson correlation coefficient calculation formula to obtain the correlation coefficient R between the target dimension data and the equipment running time;

[0063] For example:

[0064] The correlation coefficient R between the dimension data and the equipment running time, where is the average value of Tr, The mean value of Tr is calculated by the Pearson correlation coefficient, which is an existing and mature calculation method, and thus, no further description is provided herein.

[0065] When the correlation coefficient R is greater than Y3 + , the ratio Er between each J1r and Tr is obtained, and the absolute value of the mean value of the ratio Er is taken as the correlation coefficient X1 between the target dimension data and the equipment running time length.

[0066] When the correlation coefficient R is less than Y3 - , the product between the absolute value of the mean value of the ratio Er and -1 is taken as the correlation coefficient X1 between the target dimension data and the equipment running time length.

[0067] When the correlation coefficient R is greater than or equal to Y3 - and less than or equal to Y3 + , 0 is taken as the correlation coefficient X1 between the target dimension data and the equipment running time length.

[0068] Y3 is a preset value, and the specific value is determined by relevant personnel according to actual needs.

[0069] S03: repeating steps S01-S02, so as to obtain the correlation coefficient Xj between each dimension data and the equipment running time length.

[0070] The systematic influence of the machine equipment running time length on each dimension data is quantified. First, the mature Pearson correlation coefficient is used to evaluate the linear correlation strength between the dimension node value and the running time length. On this basis, the correlation coefficient is innovatively introduced. When the correlation is strong and positive, Xj takes the absolute value of the mean value of the ratio, indicating the trend rate of the dimension value increasing with the increase of the running time length. When the correlation is strong and negative, Xj takes the negative value of the absolute value of the mean value of the ratio, indicating the trend rate of the dimension value decreasing with the increase of the running time length. When the correlation is not significant, Xj is set to 0. This processing converts the abstract correlation into a quantifiable and physically meaningful change rate, which directly reflects the linear influence trend of the running time length on different dimension data and accurately quantifies the internal influence degree and direction of the machine equipment running time length on each monitoring indicator. From experience judgment to data-driven quantitative analysis, it can identify the natural growth or decline trend of some parameters over time in the normal operation process of the equipment, providing a basis for data calibration. The correlation coefficient Xj obtained is a key parameter for real-time data calibration calculation in subsequent step six, which is used to compensate or eliminate the systematic deviation caused by the running time length.

[0071] Step five: obtain and analyze the signal receiving time points and data collection time points of different detection devices at different data collection time points; obtain the corresponding time delay duration between each detection device and the data collection time point according to the analysis result, specifically as follows:

[0072] S11: randomly select one from different detection devices as a target device;

[0073] S12: obtain the signal receiving time points and data collection time points of the target device at different data collection time points, and obtain the absolute value of the difference between the signal receiving time points and the data collection time points, which is the time difference value Er of the target device at different data collection time points;

[0074] Obtain the discrete value U of the time difference value Er, when the discrete value U is greater than the preset value Y4, the mean value of the maximum value and the minimum value of the time difference value Er is the time delay duration F1 of the target device, when the discrete value U is less than or equal to the preset value Y4, the mean value of the time difference value Er is the time delay duration F1 of the target device;

[0075] S13: repeat steps S11-S12, that is, obtain the corresponding time delay duration Fj between each detection device and the data collection time point, it needs to be noted that different detection devices correspondingly collect data of different dimensions; and each detection device corresponds to different dimensional data one by one;

[0076] Solve the signal transmission and data processing delay problem of the detection device itself, compare the signal receiving time points and the data collection time points, calculate each time difference value Er, analyze the discrete value U of Er and combine the preset threshold Y4, calculate the time delay duration Fj of each detection device, which means not only the delay is identified, but also the stability of the delay is evaluated, the inherent time delay of different detection devices in the data collection process is accurately measured, the problem of ignoring the sensor response time in traditional monitoring is solved, the time difference between the collected data and the actual physical event is determined, which provides accurate basis for subsequent time alignment of data, provides key parameters for data calibration, and the time delay duration Fj obtained is another key parameter for real-time data calibration calculation in subsequent step six, which is used to correct the data value, so that it is closer to the state when the actual event occurs.

[0077] Step six: when the real-time detection data of each dimension data needs to be analyzed, according to the correlation coefficient Xj between each dimension data and the machine equipment running time length and the time delay duration Fj between each detection equipment and the data collection time point, the corresponding reference calculation value of each dimension data is obtained, the calibration calculation of the real-time detection data of each dimension data is realized, and the accuracy of data monitoring is further improved. The specific mode is:

[0078] The real-time detection data of each dimension data is marked as Hj, and the real-time detection data Hj of each dimension data is subtracted by the product of the corresponding time delay duration Fj and the correlation coefficient Xj, and then the corresponding reference calculation value Kj of each dimension data is obtained, that is, Kj = Hj - Fj x Xj. The product of the time delay duration Fj and the correlation coefficient Xj represents the dimension data change caused by the time correlation of the time delay duration data between each detection equipment and the data collection time point;

[0079] The systematic trend deviation caused by the machine equipment running time length and the time sequence deviation caused by the detection equipment time delay are eliminated, so that the obtained reference calculation value Kj is closer to the real state of the machine equipment at the actual physical time point, which is the key to realize high-precision joint time sequence prediction and accurate abnormal identification.

[0080] When the detection equipment has a delay Fj, it means that the corresponding detection equipment actually collects data at a later time. If the data increases with time (i.e. Xj>0), the data will increase by Fj x Xj during the delay period, so the corresponding reference calculation value needs to be subtracted;

[0081] If the data decreases with time (i.e. Xj<0), it means that the data will decrease by |Xj| x Fj during the delay period, so subtracting a negative number is equivalent to adding the decreased part to restore the reference value at the preset time point. If it is independent of time (i.e. Xj=0), the delay does not affect the data value, and the reference calculation value is the real-time node value;

[0082] Through the above calibration, the time sequence delay of the detection equipment and the correlation deviation of the data with the running time length can be eliminated, the reference calculation value corresponding to the preset collection time point is obtained, and the accuracy of data monitoring is significantly improved, which lays a reliable data foundation for subsequent time sequence prediction and abnormal warning;

[0083] By setting multiple data collection time points, collecting multi-dimensional data such as temperature, pressure, vibration, etc. from different machine devices at each data collection time point through each detection device, and performing time series analysis on these data, at each data collection time point, the collected data is analyzed in multiple dimensions, the mean value of each data is calculated, and the deviation of each data value from the mean value is judged. Through this joint time series analysis, the internal correlation between the data can be identified, and fault prediction can be performed based on this;

[0084] Due to the delay of the collected signals of the equipment, a time delay correction mechanism is adopted. Before data analysis, the error caused by time delay is corrected to ensure that all data points reflect the true state of the equipment at a certain time. By analyzing the time difference between the signal receiving time point and the data collection time point, the data is calibrated to eliminate the influence of time delay. The problem of inaccurate data caused by time delay in the prior art is solved, so that the real-time data can reflect the true state of the equipment, and the early warning system can respond in time, thereby reducing the risk of failure.

[0085] The systematic trend deviation caused by the running time and the time series deviation caused by the sensor delay are eliminated, so that the baseline calculation value for early warning calculation and analysis is closer to the true state of the machine equipment at the actual physical time point, improving the prediction and early warning accuracy. The calibrated baseline data is a more reliable basis for joint time series prediction and early abnormal early warning, significantly reducing the false positive rate and the false negative rate, improving the accuracy of the prediction model and the timeliness of the early warning, realizing data standardization, making the data at different time points, different equipment and different dimensions more comparable when comparing and analyzing, and providing data guarantee for building a unified health assessment model.

[0086] Embodiment two: as embodiment two of the present application, compared with embodiment one, the technical solution of the present embodiment is only different from embodiment one in that the present embodiment further comprises step seven;

[0087] Step seven: according to the baseline calculation value corresponding to each dimension data, an early warning signal is generated, and the specific method is:

[0088] The sum of the products between the baseline calculation value Kj corresponding to each dimension data and the preset parameter coefficient βj corresponding to each dimension data is taken as the comprehensive judgment value of the machine equipment. The comprehensive judgment value is compared with the preset early warning threshold Y5. When the comprehensive judgment value is greater than the preset early warning threshold Y5, an early warning signal is generated. Otherwise, no processing is performed. The specific value of the threshold Y5 is determined by relevant personnel according to actual needs;

[0089] By selecting the longest historical running time of the machine equipment as the calibration running time, the data collection time points are set based on the calibration running time and multi-dimensional data are collected, the node values of each dimension data at different time points are obtained through analysis, and then the correlation coefficient of each dimension data and the running time and the time delay length of each detection equipment are determined, finally the real-time data are calibrated based on the correlation coefficient and the time delay length and the early warning signal is determined, effectively solving the problem that in the existing monitoring system, the data acquisition delay is ignored due to the direct use of the original data of the detection equipment, the coupling effect of each dimension data and the equipment running time, the data cannot accurately reflect the actual running state of the equipment, and the accuracy and timeliness of the early warning are affected, through the joint time sequence analysis and calibration of the multi-dimensional data, the early and accurate early warning of the machine equipment anomaly is realized, the maintenance personnel can receive the warning in the early stage of fault development, and sufficient time is obtained for inspection, diagnosis and preventive intervention, effectively reducing the unplanned downtime, and significantly improving the reliability and production efficiency of the equipment running.

[0090] Embodiment three: as embodiment three of the present application, compared with embodiment one and embodiment two, the technical scheme of the present embodiment is to combine the schemes of embodiment one and embodiment two.

[0091] The above formulas are dimensionless numerical calculations, the formula is obtained by software simulation of a large amount of data to obtain a formula of the latest real situation, and the preset parameters and threshold values in the formula are set by the person skilled in the art according to the actual situation.

[0092] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators, characterized in that, Includes the following steps: Step 1: Select the maximum value among the machine's historical running times as the machine's calibrated running time; Step 2: Set multiple data acquisition time points according to the calibration runtime, and collect multi-dimensional data of the machine at different data acquisition time points through multiple detection devices to obtain multiple sets of multi-dimensional data sets; Step 3: Analyze the multi-dimensional data at each data collection time point to obtain the node values ​​corresponding to different dimensions of data at each data collection time point; Step 4: Analyze the correlation between the values ​​of each node and the runtime of the machine equipment to obtain the correlation coefficient between the data of each dimension and the runtime; Step 5: Acquire and analyze the signal reception time and data acquisition time of different detection devices at different data acquisition time points; based on the analysis results, obtain the corresponding timing delay duration between each detection device and the data acquisition time point; Step 6: Based on the correlation coefficients of the data in each dimension and the timing delay of each detection device, perform calibration calculations on the real-time detection data to obtain the baseline calculation values ​​corresponding to each dimension of the data. Step 7: Calculate the baseline values ​​corresponding to each dimension of data to determine and generate an early warning signal.

2. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 1, characterized in that, The specific method for obtaining the node values ​​corresponding to different dimensions of data at each data collection time point is as follows: S1: Randomly select one of the various data collection time points as the analysis time point; S2: Randomly select one dimension from the different dimensions of data as the analysis dimension; S3: Obtain the mean of the data for each analysis dimension, and obtain the absolute value of the difference between the data for each analysis dimension and the mean. Compare and analyze the absolute value of the difference with the preset value Y1 to obtain the node value J11 at the analysis time point. S4: Repeat steps S2-S3 to obtain the node values ​​Jj1 corresponding to each dimension of data at the analysis time point, where j represents different dimension data; S5: Repeat steps S1-S4 to obtain the node values ​​Jjr corresponding to each dimension of data at different data collection time points, where r is the different data collection time point, r = 1, 2, ..., a2, a2 ​​is the total number of data collection time points, a2 is a positive integer and a is greater than 8.

3. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 2, characterized in that, The specific method for comparing and analyzing the absolute value of the difference with the preset value Y1 is as follows: The number of analysis dimension data whose absolute difference is greater than the preset value Y1 is marked as b. When b is greater than the preset value Y2, the average of the maximum and minimum values ​​in each analysis dimension data is taken as the node value J11 of the analysis dimension data at the analysis time point. When b is less than or equal to the preset value Y2, the average of each analysis dimension data is taken as the node value J11 of the analysis dimension data at the analysis time point.

4. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 3, characterized in that, When the total number of data points in the analysis dimension is even, the preset value Y2 is half of the total number of data points in the analysis dimension. When the total number of data points in the analysis dimension is odd, the preset value Y2 is half of the total number of data points in the analysis dimension plus one.

5. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 2, characterized in that, The specific method for obtaining the correlation coefficients between data in each dimension and runtime is as follows: S01: Select the analysis dimension data from step S2 as the target dimension data; S02: Obtain the machine running time Tr corresponding to the target dimension data at each data collection time point, that is, the cumulative running time from the start of machine operation to the r-th data collection time point. Correlate the node value J1 r corresponding to the target dimension data at different data collection time points with the machine running time Tr corresponding to each data collection time point. Then use the Pearson correlation coefficient calculation formula to obtain the correlation coefficient R between the target dimension data and the machine running time. Analyze the correlation coefficient R to obtain the correlation coefficient X1 between the target dimension data and the machine running time. S03: Repeat steps S01-S02 to obtain the correlation coefficients Xj between the data of each dimension and the device runtime.

6. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 5, characterized in that, The specific method for analyzing the correlation coefficient R is as follows: When the correlation coefficient R is greater than Y3 + When, the ratio Er between each J1 r and Tr is obtained, and the absolute value of the mean of the ratio Er is used as the correlation coefficient X1 between the target dimension data and the device runtime. When the correlation coefficient R is less than Y3 - When the mean absolute value of the ratio Er is multiplied by negative one, the correlation coefficient X1 between the target dimension data and the device runtime is used. When the correlation coefficient R is greater than or equal to Y3 - And less than or equal to Y3 + In this case, 0 is used as the correlation coefficient X1 between the target dimension data and the device runtime.

7. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 3, characterized in that, The specific method for obtaining the timing delay duration between each detection device and the data acquisition time point is as follows: S11: Randomly select one of the different detection devices as the target device; S12: Obtain the signal reception time and data acquisition time of the target device at different data acquisition time points, and obtain the absolute value of the difference between the signal reception time point and the data acquisition time point. Use this as the time difference Er of the target device at different data acquisition time points, obtain the discrete value U of the time difference Er, and analyze it to obtain the timing delay duration F1 of the target device. S13: Repeat steps S11-S12 to obtain the timing delay Fj between each detection device and the data acquisition time point.

8. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 7, characterized in that, The specific method for obtaining the timing delay duration of the target device is as follows: When the discrete value U is greater than the preset value Y4, the average of the maximum and minimum values ​​in the time difference Er is used as the timing delay duration F1 of the target device. When the discrete value U is less than or equal to the preset value Y4, the average of the time difference Er is used as the timing delay duration F1 of the target device.

9. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 7, characterized in that, The specific method for obtaining the baseline calculation values ​​corresponding to each dimension of data is as follows: The real-time detection data of each dimension is labeled as Hj. The real-time detection data of each dimension is subtracted from the product of the corresponding time delay duration Fj and the correlation coefficient Xj, thereby obtaining the benchmark calculation value corresponding to each dimension.

10. The method for joint time-series prediction and early warning of anomalies based on multi-dimensional monitoring indicators according to claim 9, characterized in that, The specific method for determining the generation of early warning signals is as follows: The sum of the products of the baseline calculated value Kj corresponding to each dimension of data and the preset parameter coefficient βj corresponding to each dimension of data is used as the comprehensive judgment value of the machine. When the comprehensive judgment value is greater than the preset warning threshold Y5, an early warning signal is generated; otherwise, no action is taken.

Images