A data optimization method for quality anomaly detection in polypropylene production

Abstract

The present application relates to the field of data optimization, and in particular to a data optimization method and system for detecting abnormal quality of polypropylene production. The method comprises: acquiring data collected from each reactor during polypropylene production, establishing a time window, obtaining local fluctuation data according to a change in local time series data, performing fluctuation analysis on the local fluctuation data, calculating an initial entropy weight of the local fluctuation data, calculating an entropy value sequence of the collected data according to the initial entropy weight, extracting a trend item of the entropy value sequence, calculating a drift coefficient according to a trend change of the trend item and the initial entropy weight, completing data drift judgment according to the drift coefficient, performing data optimization according to the data drift judgment result, and reducing the false detection rate of abnormal quality detection of polypropylene production.

Description

Technical Field

[0001] The present application relates to the field of data optimization, and in particular to a data optimization method for detecting abnormal quality in polypropylene production. Background Art

[0002] The invention with patent number CN113191616A discloses a method for detecting abnormal quality of polypropylene products based on double-layer correlation feature analysis. The method includes: collecting production data of multiple reactors in polypropylene production, calculating a double-layer correlation feature matrix of the production data of multiple reactors, and using the double-layer correlation feature matrix to implement abnormal quality detection of polypropylene production.

[0003] Although the above-mentioned prior art can realize the detection of abnormal quality of polypropylene production, the multiple reactors used in the production of polypropylene are in long-term continuous use. When the multiple reactors are used continuously, the production data of the multiple reactors will drift.

[0004] Although the production data of multiple reactors can be corrected by performing regular maintenance on multiple reactors, if the equipment is not properly maintained during the maintenance of multiple reactors, the production data of the multiple reactors will drift, and false detection will occur when implementing abnormal detection of polypropylene production quality, thereby affecting the production efficiency of polypropylene. Summary of the invention

[0005] In order to solve the technical problems of data drift and inaccurate data in the prior art, the present application provides a data optimization method and system for abnormal detection of polypropylene production quality, so as to improve the accuracy of data during data optimization.

[0006] According to a first aspect of the present application, a data optimization method for detecting abnormal quality of polypropylene production is provided, comprising: acquiring data collected from each reactor during polypropylene production to obtain single-dimensional data during polypropylene production, establishing a time window for each data point in the single-dimensional data, obtaining local time series data and an initial mean value for each data point, subtracting each data point in the local time series data from the initial mean value to obtain a difference sequence of the local time series data, taking an absolute value of each data in the difference sequence to obtain local fluctuation data, calculating adjacent extreme value points of the local fluctuation data, obtaining adjacent extreme value points and adjacent extreme value time series values ​​of each data point in the local fluctuation data along the time direction, calculating the difference value between each data point in the local fluctuation data and its adjacent extreme value points, and the time series interval value with the adjacent extreme value point time series value, and calculating the difference value between each data point in the local fluctuation data and its adjacent extreme value points according to the difference value and the time series value of the adjacent extreme value points. The initial entropy weight of each data point in the local fluctuation data is obtained by calculating the positive correlation between the time series interval value and the time series interval value, and the average of the initial entropy weights of the data points with the same value in the local time series data is calculated as the entropy weight sequence of the data points with the same value in the local time series data. The information entropy is weighted by using the entropy weight sequence to obtain the entropy value of the local time series data, and the entropy value sequence of the single-dimensional data is obtained. The trend item of the entropy value sequence is obtained by using the trend item extraction algorithm, and the change amount of the trend item of the latest collected data is calculated. According to the positive correlation between the change amount of the latest collected data in the trend item and the corresponding initial entropy weight, the drift coefficient of the latest collected data of the single-dimensional data is calculated, and a data drift coefficient threshold is set. According to the relationship between the data drift coefficient and the data drift threshold of any single-dimensional data during polypropylene production, the correction of the collected data during polypropylene production is completed.

[0007] In one embodiment, the single-dimensional data during polypropylene production includes: using a sensor data acquisition system during polypropylene production to obtain acquisition data of a single sensor in each reactor as the single-dimensional data in the acquisition data of each reactor.

[0008] In one embodiment, obtaining the local fluctuation data includes: establishing a time window for each data point in the single-dimensional data, obtaining the The local time series data of the jth collected data of the sensor , and then calculate the The fluctuation value of the mth data in the jth collected data of the sensor :

[0009] ;in, For the The mth data in the local time series data of the jth collected data of the sensor, for The mean of the first collected data to the jth collected data in the local time series data is calculated, and the above difference calculation operation is performed on each data point in the local time series data to obtain local fluctuation data.

[0010] In one embodiment, the entropy value sequence of the single-dimensional data includes: using an extreme point detection algorithm to obtain the extreme point of the local fluctuation data, obtaining the adjacent extreme point of the local fluctuation data according to the time series direction of the collected data, calculating the difference value between the mth data point in the local fluctuation data and its adjacent extreme point, and the time series interval value with the time series value of the adjacent extreme point, and obtaining the mth data point according to the positive correlation between the difference value and the time series interval value. The initial entropy weight of the mth data point in the local fluctuation data of the jth collected data of the sensor :

[0011] ;in, For the The normalized value of the time interval between the mth data point and the adjacent extreme point in the local fluctuation data of the jth collected data of the sensor, For the The normalized value of the difference between the maximum neighboring extreme value point and the minimum neighboring extreme value point of the mth data point in the local fluctuation data of the jth collected data of the sensor is obtained. The above initial entropy weight calculation operation is performed on each data point in the local fluctuation data to obtain the initial entropy weight of each data point in the local fluctuation data. The average of the initial entropy weights of the data points with the same value in the local time series data is calculated as the entropy weight sequence of the data points with the same value in the local time series data. The information entropy calculation is weighted using the entropy weight sequence to calculate the first The entropy value of the local time series data of the jth collected data of the sensor is used as the The entropy value of the jth collected data of a sensor is:

[0012] ;in, is the number of data points in the time window, n represents the number of N from Increment to The sequential traversal of For the The proportion of the nth data with the same value in the local fluctuation data of the jth data collected by the sensor, The function is a logarithmic function with base 2. For the The entropy weight of the nth data with the same value in the local fluctuation data of the jth data collected by the sensor, For the The entropy value of the jth collected data in the collected data of the sensors is Each collected data of each sensor is calculated according to the above entropy value to obtain The entropy value of each collected data of the sensor is used as the The entropy value sequence of the single-dimensional data of each sensor.

[0013] In one embodiment, the drift coefficient of the latest collected data of the single-dimensional data includes: using a trend item extraction algorithm to obtain the trend item of the entropy value sequence, calculating the change amount of the trend item of the latest collected data, calculating the drift coefficient of the latest collected data of the single-dimensional data according to the positive correlation between the change amount of the latest collected data in the trend item and the corresponding initial entropy weight, and obtaining the drift coefficient of the xth latest collected data in the collected data of the i-th sensor ;

[0014] ;in, is the initial entropy weight of the xth collected data point in the time window of the xth collected data of the i-th sensor; is the value of the xth collected data in the trend item of the collected data of the i-th sensor minus The difference between the mean and the mean of .

[0015] In one embodiment, the correction of the data collected during the production of polypropylene includes: setting a data drift coefficient threshold, obtaining the data drift coefficient of any single-dimensional data during the production of polypropylene, when the drift coefficient is greater than or equal to the data drift coefficient threshold, notifying maintenance personnel to perform data correction on the sensor where data drift occurs, and completing data optimization; when the drift coefficient is less than the data drift coefficient threshold, the sensor data of each reactor does not drift, and the abnormal detection result when performing abnormal detection of polypropylene production quality is accurate.

[0016] Beneficial effects:

[0017] By using the windowing method for the collected data, the local fluctuation data of the continuously collected data is obtained, and fluctuation analysis is performed on the local fluctuation data to achieve entropy weighting of the collected data, thereby making the rising changes in the entropy value sequence more obvious, which is conducive to data drift detection of the collected data.

[0018] When using information entropy for fluctuation analysis, it does not rely on the data drift direction of the collected data, nor does it rely on the influence of the data size of the collected data itself. When data drift occurs, there is a trend of increasing entropy value sequence.

[0019] The trend term is used to analyze the drift coefficient of the entropy value sequence. The overall trend of the entropy value sequence is extracted through the trend term to reduce the interference of local information. The drift coefficient analysis is performed in combination with the entropy weight to improve the accuracy of data drift detection.

[0020] By performing data drift detection on the collected data, the quality of the data collected by each reactor during polypropylene production is guaranteed, and the false detection rate of abnormal quality detection in polypropylene production is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become easy to understand. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0022] Figure 1 It is a flow chart of a data optimization method for detecting abnormal quality of polypropylene production according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0024] It should be understood that when the terms "first", "second", etc. are used in the claims, specification and drawings of the present application, they are only used to distinguish different objects, rather than to describe a specific order. The terms "include" and "comprise" used in the specification and claims of the present application indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present application belongs.

[0025] like Figure 1 As shown, an embodiment of a data optimization method for detecting abnormal quality in polypropylene production according to the present application includes steps S1 to S4.

[0026] Step S1, obtaining the collected data of each reactor during polypropylene production, establishing a time window for each data point in the single-dimensional data in the collected data of each reactor, obtaining the local time series data and the initial mean of each data point, and obtaining the local fluctuation data according to the difference between the local time series data and the initial mean of each data point.

[0027] Specifically, since each reactor can work continuously for a long time during polypropylene production, data drift will occur. Through maintenance and overhaul of each reactor equipment, the sensor data that causes data drift can be corrected to ensure the accuracy of the collected data. Therefore, the collected data of each reactor after overhaul is obtained for data drift judgment, so as to achieve data optimization during abnormal detection of polypropylene production quality.

[0028] The sensor data acquisition system during polypropylene production is used to obtain the collected data of a single sensor in each reactor. The collection system is set to collect sensor data once per second as single-dimensional data in the collected data of each reactor. The sensors include but are not limited to temperature sensors, pressure sensors, liquid level sensors, etc.

[0029] Get the Data collected by sensors , the sensor data of each reactor is stable during polypropylene production, that is, is a set of stable time series data. When data drift occurs in a sensor, the data value of the data point causing data drift will gradually deviate from The mean value of When a sensor has data anomaly, the data value of the data point where the data anomaly occurs is relative to The mean value of is a mutation value. With the long-term operation of polypropylene production, the production data with data drift will be misjudged as production data caused by production equipment abnormality during the abnormal detection of polypropylene production quality.

[0030] The creation time is Minute time window, get The jth collected data is in front The collected data within minutes is used as The local time series data of the jth collected data , get The mean of the first collected data to the jth collected data in is used as the starting mean. ,calculate The fluctuation value of the mth data in ,in , similarly, calculate The fluctuation data of each data in is obtained The fluctuation data series , as The local fluctuation data of the local time series data of the jth collected data in Get experience points, .

[0031] In step S1, a time window is established for each collected data point to analyze the fluctuation of the collected data relative to the initial mean value in continuous time. By establishing the time window, the local change analysis of the error of the data in the time window can be performed to determine whether data drift occurs in continuous time for each collected data point.

[0032] Step S2, calculate the sequence of neighboring extreme value points of the local fluctuation data, analyze the sequence of neighboring extreme value points, obtain the entropy weight sequence of data points with the same value in the local time series data, and obtain the entropy value sequence of the single-dimensional data.

[0033] The specific implementation is to use the extreme point detection algorithm to obtain The extreme point sequence in The mth data , along the reverse direction of the time axis, sequentially traverse the extreme value points that are closest to the m data as the adjacent extreme value points. The extreme value point detection algorithm is a well-known technology and will not be described in detail in this application. The reverse direction of the time axis is the direction away from the latest time.

[0034] calculate The difference from the adjacent extreme point , obtain the maximum neighboring extreme point and the minimum neighboring extreme point in the time window of the jth collected data, and calculate the difference between the maximum neighboring extreme point and the minimum neighboring extreme point ,use right Normalize it and get Normalized data , , calculate the time interval between the mth data point and the adjacent extreme point , using the time series length of the time window, Divide by the time series length of the time window to achieve Normalization of Normalized data .

[0035] according to and The positive correlation between and Multiply them together to get the initial entropy weight of the mth data point , .

[0036] Obtain the initial entropy weights corresponding to the same-valued data in the local fluctuation data, calculate the average of the initial entropy weights corresponding to the same-valued data in the local fluctuation data, obtain the entropy weight of each same-valued data in the local fluctuation data, count the entropy weights of all the same-valued data in the local fluctuation data, obtain the entropy weight sequence of the local fluctuation data, use the entropy weight sequence to weight the information entropy, and calculate the first The entropy value corresponding to the jth collected data in the collected data of sensors , and then get the The entropy value sequence of the collected data of the sensors .

[0037] Among them, the data with the same value is the data after deduplication in the local fluctuation data. For example, if there are data 1, 1, 2, 2, 3 in the local fluctuation data, then the data with the same value are 1, 2, 3, and the proportion of each data with the same value is 0.4, 0.4, 0.2.

[0038] The specific formula for weighting the information entropy using the entropy weight sequence to obtain the corresponding entropy value is as follows:

[0039]

[0040] is the number of data points in the time window, n represents the number of N from Increment to The sequential traversal of .

[0041] For the The proportion of the nth data with the same value in the local fluctuation data of the jth collected data in the collected data of the sensor. In the information entropy calculation when data drift occurs, if the data with the same value in the local fluctuation data are chaotic, the number of data with the same value in the local fluctuation data will increase, which will make the data with the same value in the local fluctuation data shift from a skewed distribution to a balanced distribution, resulting in an increase in the entropy value of the local fluctuation data.

[0042] The function is a base 2 logarithmic function.

[0043] For the The entropy weight of the nth data with the same value in the local fluctuation data of the jth data collected by the sensor, is the mean of the entropy weights corresponding to the data points with the same value as the nth data with the same value in the local fluctuation data, The larger the value of , the more dispersed the nth data with the same value in the local fluctuation data is, the greater the probability of data drift, and the greater the entropy weight, which makes the nth The larger the entropy value of the jth collected data among the collected data of sensors.

[0044] For the The entropy value of the jth collected data in the collected data of the sensor is larger, indicating that the The more chaotic the local fluctuation data of the jth collected data among the collected data of the sensors, the greater the probability of data drift.

[0045] In the above step S2, the entropy weight of each data point in the local fluctuation data is calculated, which is used to weight the data with the same value in the entropy value calculation of a single collected data, so that when data deviation occurs, the entropy value sequence changes more obviously, thereby improving the accuracy of data drift judgment and further improving the data optimization effect.

[0046] Step S3, using a trend item extraction algorithm to obtain the trend item of the entropy value sequence, and calculating the drift coefficient of the latest collected data of the single-dimensional data in the trend item.

[0047] In one embodiment, Using the STL algorithm, we can obtain The trend term, seasonal term, and residual term of Trending items .

[0048] In another embodiment, Use the least squares method to perform polynomial fitting and get the fitting function. Substitute the time series of into the fitting function and get the fitting function pair The fitting data of The fitting data is Trending items .

[0049] Get the new collected data of the i-th sensor, record it as the x-th collected data, and set the value of the x-th collected data in the trend item minus The difference between the two is obtained , The larger the value, the greater the change in entropy value, the more likely there is a data drift problem, and the greater the probability that data correction is needed to achieve data optimization.

[0050] Since the entropy weight of the xth collected data is larger, it is also positively correlated with the probability of data drift, and then combined with the initial entropy weight of the xth collected data point in the time window , get the drift coefficient of the single-dimensional data at the xth acquisition data :

[0051]

[0052] is the initial entropy weight of the xth collected data point in the time window. The greater the probability of data drift problem, the greater the probability of needing data correction to achieve data optimization.

[0053] The value of the xth collected data in the trend item minus The difference between the mean and the latter two represents the entropy change of the entropy value sequence at the xth data point. The larger it is, the more obvious the increase in the entropy value sequence is, and the greater the probability of data drift, and the greater the probability of needing data correction to achieve data optimization.

[0054] Represents the drift coefficient of the xth collected data in single-dimensional data, The larger the value, the more likely it is that single-dimensional data will drift in the latest collected data. The larger the value, the greater the probability of data drift, and the greater the probability of needing data correction to achieve data optimization.

[0055] In the above step S3, by obtaining the trend item of the entropy value sequence corresponding to the single-dimensional data, the overall trend of the entropy value sequence is obtained, and the change in the overall trend of the latest collected data in the single-dimensional data is combined with the initial entropy weight of the latest collected data in the single-dimensional data to obtain the drift coefficient of the latest collected data in the single-dimensional data, which is used to realize data drift monitoring of the latest collected data in the single-dimensional data.

[0056] Step S4, setting a data drift coefficient threshold, and completing the correction of the data collected during the production of polypropylene according to the relationship between the data drift coefficient of any single-dimensional data during the production of polypropylene and the data drift threshold.

[0057] The specific implementation is to obtain the collected data of any single sensor in each reactor during polypropylene production, obtain multiple dimensional data in the collected data of each reactor, and obtain the drift coefficient of any dimensional data in the multiple dimensional data in the latest collected data as a drift coefficient sequence.

[0058] When any drift coefficient value in the drift coefficient sequence is greater than or equal to the data drift coefficient threshold, the collected data of the sensor corresponding to the drift coefficient in the drift coefficient sequence that is greater than or equal to the data drift coefficient threshold has a data drift problem. When the data drift problem is monitored, the maintenance personnel are notified to correct the data of the sensor where the data drift occurs, complete the data optimization, and reduce the false detection probability of abnormal detection of polypropylene production quality.

[0059] When any drift coefficient value in the drift coefficient sequence is less than the data drift coefficient threshold, it means that during polypropylene production, the sensor data of each reactor has not experienced data drift, and the abnormal detection result during the abnormal detection of polypropylene production quality is accurate.

[0060] The data drift coefficient threshold in this application takes an empirical value of 0.97.

[0061] At this point, a data optimization method for detecting abnormal quality of polypropylene production is completed. The data drift judgment completed through all the steps of this application can improve the accuracy of data drift judgment when optimizing data for detecting abnormal quality of polypropylene production, and ensure the accuracy of detecting abnormal quality of polypropylene production.

[0062] Although this specification has shown and described a plurality of embodiments of the present application, it is obvious to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will think of many changes, modifications and alternatives without departing from the thought and spirit of the present application. It should be understood that in the process of practicing the present application, various alternatives to the embodiments of the present application described herein may be adopted.

[0063] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A data optimization method for detecting abnormal quality of polypropylene production, characterized in that: Data optimization methods include: S1, acquiring data collected from each reactor during polypropylene production to obtain single-dimensional data during polypropylene production, establishing a time window for each data point in the single-dimensional data to obtain local time series data and an initial mean value for each data point, subtracting each data point in the local time series data from the initial mean value to obtain a difference sequence of the local time series data, taking an absolute value of each data in the difference sequence to obtain local fluctuation data; S2, calculate the neighboring extreme value points of the local fluctuation data, obtain the neighboring extreme value points and the time series values ​​of the neighboring extreme value points of each data point in the local fluctuation data along the time direction, calculate the difference value between each data point in the local fluctuation data and its neighboring extreme value points, and the time series interval value with the time series value of the neighboring extreme value points, obtain the initial entropy weight of each data point in the local fluctuation data according to the positive correlation between the difference value and the time series interval value, calculate the mean of the initial entropy weights of the data points with the same value in the local time series data as the entropy weight sequence of the data points with the same value in the local time series data, use the entropy weight sequence to weight the information entropy calculation, obtain the entropy value of the local time series data, and obtain the entropy value sequence of the single-dimensional data; S3, using a trend item extraction algorithm to obtain the trend item of the entropy value sequence, calculate the change amount of the trend item of the latest collected data, and calculate the drift coefficient of the latest collected data of the single-dimensional data according to the positive correlation between the change amount of the latest collected data in the trend item and the corresponding initial entropy weight; S4, setting a data drift coefficient threshold, and completing the correction of the data collected during the production of polypropylene according to the relationship between the data drift coefficient of any single-dimensional data during the production of polypropylene and the data drift threshold.

2. A data optimization method for detecting abnormal quality of polypropylene production according to claim 1, characterized in that: The single-dimensional data of polypropylene production includes: The sensor data acquisition system during polypropylene production is used to obtain the acquisition data of a single sensor in each reactor as single-dimensional data in the acquisition data of each reactor.

3. A data optimization method for detecting abnormal quality of polypropylene production according to claim 1, characterized in that: The obtaining of local fluctuation data comprises: Calculate the first The local time series data of the jth collected data of the sensor , and then calculate the The fluctuation value of the mth data in the jth collected data of the sensor ; ； For the The mth data in the local time series data of the jth collected data of the sensor; for The mean of the first collected data to the jth collected data; The above difference calculation operation is performed on each data point in the local time series data to obtain local fluctuation data.

4. A data optimization method for detecting abnormal quality of polypropylene production according to claim 1, characterized in that: The entropy value sequence of the single-dimensional data includes: Using an extreme point detection algorithm, the extreme point of the local fluctuation data is obtained, and according to the time series direction of the collected data, the adjacent extreme point of the local fluctuation data is obtained; Calculate the difference between the mth data point and its adjacent extreme value point in the local fluctuation data, as well as the time series interval value with the time series value of the adjacent extreme value point, and obtain the mth data point according to the positive correlation between the difference value and the time series interval value. The initial entropy weight of the mth data point in the local fluctuation data of the jth collected data of the sensor ; ； For the The normalized value of the time interval between the mth data point and the adjacent extreme point in the local fluctuation data of the jth collected data of the sensor, For the The normalized value of the difference between the maximum neighboring extreme value point and the minimum neighboring extreme value point of the mth data point in the local fluctuation data of the jth collected data of the sensor; Performing the above initial entropy weight calculation operation on each data point in the local fluctuation data to obtain an initial entropy weight of each data point in the local fluctuation data; Calculating the mean of the initial entropy weights of the data points with the same value in the local time series data as the entropy weight sequence of the data points with the same value in the local time series data; The information entropy is weighted by using the entropy weight sequence to calculate the The entropy value of the local time series data of the jth collected data of the sensor is used as the The entropy value of the jth collected data of the sensor ; ； is the number of data points in the time window, n represents the number of N from Increment to The sequential traversal of ; For the The proportion of the nth data with the same value in the local fluctuation data of the jth data collected by the sensors; The function is a logarithmic function with base 2; For the The entropy weight of the nth data with the same value in the local fluctuation data of the jth data collected by the sensor; For the The entropy value of the jth collected data among the collected data of sensors; For Each collected data of each sensor is calculated according to the above entropy value to obtain The entropy value of each collected data of the sensor is used as the The entropy value sequence of the single-dimensional data of each sensor.

5. A data optimization method for detecting abnormal quality of polypropylene production according to claim 1, characterized in that: The drift coefficient of the latest collected data of the single-dimensional data includes: Using a trend term extraction algorithm to obtain a trend term of the entropy value sequence; Calculate the change in the trend item of the latest collected data, calculate the drift coefficient of the latest collected data of the single-dimensional data according to the positive correlation between the change in the trend item of the latest collected data and the corresponding initial entropy weight, and obtain the drift coefficient of the xth latest collected data in the collected data of the i-th sensor ; ； is the initial entropy weight of the xth collected data point in the time window of the xth collected data of the i-th sensor; is the value of the xth collected data in the trend item of the collected data of the i-th sensor minus The difference between the mean and the mean of .

6. A data optimization method for detecting abnormal quality of polypropylene production according to claim 1, characterized in that: The correction of the data collected during the production of polypropylene includes: Set the data drift coefficient threshold; Obtain the data drift coefficient of any single dimension data during polypropylene production; When the drift coefficient is greater than or equal to the data drift coefficient threshold, the maintenance personnel are notified to perform data correction on the sensor with data drift to complete data optimization; When the drift coefficient is less than the data drift coefficient threshold, the sensor data of each reactor does not drift, and the abnormality detection result when performing abnormality detection on polypropylene production quality is accurate.

Images