Intelligent hot dry noodles production line monitoring method

Abstract

The present invention relates to the field of data processing, and more specifically, to an intelligent hot dry noodle production line monitoring method, the method comprising: obtaining a data set; calculating the degree of isolation of each dimension data; calculating the abnormal possibility according to the degree of isolation, and removing the dimension whose abnormal possibility is less than a preset possibility threshold; taking the dimension whose abnormal possibility is not less than the preset possibility threshold as the dimension to be analyzed; obtaining possible abnormal data points corresponding to the dimension to be analyzed; performing dimension retention control according to the intersection ratio of possible abnormal data points of two dimensions; and performing abnormal detection on quality monitoring data points based on the retained dimension using an isolation forest algorithm to realize production monitoring of the hot dry noodle production line. The accuracy of abnormal detection is improved by accurately screening out appropriate dimensions.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more specifically, to an intelligent hot-dry noodle production line monitoring method. Background Art

[0002] With the development of science and technology, the trend of unmanned management of hot dry noodle production lines is becoming more and more obvious. To achieve unmanned management, it is necessary to realize intelligent monitoring of hot dry noodle production lines. For example, when an abnormality occurs in the hot dry noodle production line, it is necessary to give people an early warning in time so that people can take timely countermeasures to prevent large economic losses caused by production abnormalities.

[0003] When an abnormality occurs in the hot dry noodle production line, abnormal indicator data will be generated. For example, when an abnormality occurs in the drying station of the hot dry noodles, abnormal humidity or temperature data will be displayed. Therefore, it is possible to determine whether abnormal data occurs through abnormal detection, thereby determining whether an abnormality occurs in the drying station of the hot dry noodles production line.

[0004] As a commonly used anomaly detection algorithm, the isolation forest algorithm has a high detection accuracy. The algorithm selects data of one dimension from all dimensions of the data point as the segmentation dimension, selects a position in the value range of the data of the segmentation dimension as the segmentation point, and uses the segmentation point to segment all data points into two categories, and continuously iterates the segmentation until the cutoff condition is reached. Since abnormal data points will show outliers, the abnormal data points will be quickly separated from other data points through this segmentation method, and the abnormal data point detection is realized through this feature. It can be seen from the implementation process that the quality of dimension selection will affect its anomaly detection accuracy. For example, when drying anomalies occur at the drying station, the temperature and humidity are abnormal, but the surface feed speed and the like will not be abnormal, so the data points at the drying station are segmented using the data of the temperature and humidity dimensions. Therefore, how to select the appropriate dimension so that the abnormal data points can be more easily and accurately segmented becomes the research focus of the present invention.

[0005] A method for monitoring the execution of pipeline instructions is disclosed in a patent application document with publication number CN115454505A. The method in the patent application document only uses an exception analysis algorithm to perform exception analysis on pipeline data. The method in the patent application document does not involve the content of "how to select a suitable dimension", so the method in the patent application document cannot solve the technical problem well. Summary of the invention

[0006] In order to solve the problem of how to select a suitable dimension so as to more easily and accurately segment abnormal data points, the present invention proposes an intelligent hot dry noodles production line monitoring method, which includes the following steps:

[0007] Acquire a data set, wherein the data set includes quality monitoring data points of a hot dry noodle production line;

[0008] The data of one dimension among all the quality monitoring data points in the data set is recorded as the target dimension data, and the degree of isolation of each target dimension data is obtained by dividing the cumulative sum of the distances between each target dimension data and the adjacent target dimension data on both sides by the length of the value range of the target dimension data; the ratio of the maximum degree of isolation of all target dimension data to the mean degree of isolation is recorded as the abnormal possibility, and the dimensions with abnormal possibilities less than the preset possibility threshold are removed;

[0009] The dimension whose abnormal probability is not less than the preset probability threshold is used as the dimension to be analyzed; the data of the dimension to be analyzed is sorted in descending order according to the degree of isolation, and the data before the sudden change in the degree of isolation is recorded as unilateral abnormal data, and the quality monitoring data points containing unilateral abnormal data are used as possible abnormal data points; a dimension is combined with other dimensions with the largest number of intersections of possible abnormal data points, and if the intersection ratio of possible abnormal data points of the two dimensions in the dimension combination is less than the preset ratio threshold, both dimensions are retained; if the intersection ratio of possible abnormal points of the two dimensions in the dimension combination is not less than the preset ratio threshold, only the dimension with the large mean degree of isolation in the dimension combination is retained;

[0010] Based on the retained dimension, the isolation forest algorithm is used to detect anomalies of quality monitoring data points to realize production monitoring of hot dry noodles production line.

[0011] The present invention analyzes the representation of the degree of outlier of abnormal data points by each dimension of all quality monitoring data points in a data set, removes the dimensions with poor representation, thereby preventing the data of the dimensions with poor representation from interfering with anomaly detection and improving the accuracy of anomaly detection; at the same time, the correlation between the remaining dimensions is analyzed, and some correlated dimensions are removed, thereby eliminating the interference of some unnecessary dimensional data and improving the accuracy of anomaly detection; further, when analyzing the representation of the degree of outlier of abnormal data points by each dimension, the degree of isolation, etc. is introduced to accurately describe the degree of outlier of abnormal data points represented by each dimension, thereby improving the accuracy of selected dimensions.

[0012] Preferably, the method for obtaining unilateral abnormal data includes:

[0013] The sequence obtained by arranging the data of the dimension to be analyzed in descending order according to the degree of isolation is recorded as the time series sequence to be analyzed;

[0014] The difference between each data in the time series to be analyzed and the previous data is calculated and recorded as the first difference of each data in the time series to be analyzed. The data position where the first difference is greater than the preset difference threshold is obtained and recorded as the possible mutation position. The possible mutation position that does not appear again is taken as the mutation position, and the data before the mutation position in the time series to be analyzed is taken as the unilateral abnormal data.

[0015] The present invention isolates normal data from abnormal data in the dimension to be analyzed by sorting, thereby providing a basis for subsequent accurate distinction between normal data and abnormal data in the dimension to be analyzed; further, by introducing difference analysis of adjacent data and analysis of mutation situations after each data, mutation positions that can isolate normal data from abnormal data are accurately screened out, providing a basis for accurate abnormal data screening.

[0016] Preferably, the acquiring of the data set includes:

[0017] The quality monitoring data points at each time of a workstation of a hot dry noodle production line are obtained, wherein the quality monitoring data points include quality monitoring data of multiple dimensions; the time series consisting of all the collected quality monitoring data points is divided into a number of periodic segments, each periodic segment is matched with other periodic segments by DTW, and the set consisting of corresponding matching data points in all periodic segments is recorded as a data set.

[0018] The present invention places the corresponding matching data points between the periodic segments in a set, thereby placing data points with similar values ​​together, providing a data basis for subsequent accurate abnormal data analysis.

[0019] Preferably, the step of dividing the time series consisting of all the collected quality monitoring data points into a number of periodic segments includes:

[0020] The time series sequence composed of all the collected quality monitoring data points is recorded as the quality monitoring time series sequence, and the data segment composed of a preset number of data points in the quality detection time series sequence is intercepted as the matching reference segment. The matching reference segment is used to slide on the quality monitoring time series sequence. If the similarity between the data segment to be matched and the data segment at the corresponding sliding position is greater than the preset similarity threshold, the starting data segment of the data segment at the corresponding sliding position is used as the cycle starting point, and the quality monitoring time series sequence is divided into several cycle segments using the cycle starting point.

[0021] According to the characteristic of the data similarity relationship between the periodic segments, the present invention uses a matching analysis method that can reflect the data similarity characteristics to complete the periodic segment division.

[0022] Preferably, the step of combining one dimension with other dimensions having the largest number of intersections of possible abnormal data points comprises:

[0023] Calculate the number of intersection data points of possible abnormal data points in one dimension and possible abnormal data points in other dimensions, and record it as the number of intersections; combine the other dimension with the largest number of intersections of possible abnormal data points with this dimension.

[0024] The present invention combines strongly correlated dimensions to provide a basis for subsequent removal of replaceable strongly correlated dimensions.

[0025] Preferably, the method for obtaining the intersection ratio includes:

[0026] The ratio of the number of intersections of possible abnormal data points in two dimensions in the dimension combination to the mean of the possible abnormal data points in the two dimensions is recorded as the intersection ratio.

[0027] Preferably, the method of performing anomaly detection on quality monitoring data points based on the reserved dimension using an isolation forest algorithm to achieve production monitoring of the hot dry noodle production line includes:

[0028] The isolation forest algorithm selects dimensions from the retained dimensions to perform segmentation detection on the quality monitoring data points until the preset cutoff condition is met, completing the fitting of the anomaly detection model;

[0029] The quality monitoring data points of the workstation at the current moment are input into the anomaly detection model fitted by the corresponding data set to obtain the anomaly detection result, wherein an alarm is issued when the anomaly detection result is abnormal, and no alarm is issued when the anomaly detection result is normal.

[0030] Preferably, the method for acquiring the corresponding data set includes:

[0031] Obtain the data set to which the quality monitoring data point at the previous moment belongs, obtain the time position of each quality monitoring data point in the data set to which the quality monitoring data point at the previous moment belongs, take the next time position of the time position of the quality monitoring data point in the data set to which the quality monitoring data point at the previous moment belongs as the alternative time position, and obtain the data set with the largest number of quality monitoring data points at the alternative time positions among all other data sets as the corresponding data set.

[0032] The present invention has the following beneficial effects:

[0033] The present invention analyzes the representation of the degree of outlier of abnormal data points in each dimension of all quality monitoring data points in the data set, removes the dimensions with poor representation, thereby preventing the data of the dimensions with poor representation from interfering with anomaly detection and improving the accuracy of anomaly detection; at the same time, the correlation between the remaining dimensions is analyzed, and some correlated dimensions are removed, thereby eliminating the interference of some unnecessary dimensional data and improving the accuracy of anomaly detection;

[0034] Furthermore, when analyzing the representation of the degree of outliers of abnormal data points in each dimension, the degree of isolation etc. is introduced to accurately describe the representation of the degree of outliers of abnormal data points in each dimension, thereby improving the accuracy of the selected dimensions. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0036] Figure 1 It is a step flow chart of an intelligent hot dry noodles production line monitoring method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0038] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0039] See also Figure 1 , which shows a flowchart of a method for monitoring production of an intelligent hot dry noodle production line provided by an embodiment of the present invention, the method comprising the following steps:

[0040] S1: Acquire a data set, wherein the data set includes quality monitoring data points of a hot dry noodle production line.

[0041] Preferably, as an example, obtaining a data set includes:

[0042] The quality monitoring data points at each time of a workstation of a hot-dry noodle production line are obtained, and the quality monitoring data points include quality monitoring data of multiple dimensions; for example, the dimensional types of the quality monitoring data points at the drying workstation of the hot-dry noodle production line include but are not limited to the following aspects: noodle thickness, temperature, humidity, air circulation direction, noodle inlet speed, and noodle outlet speed.

[0043] It should be noted that a workstation generally repeats a series of operations, so the data of a workstation has certain periodic characteristics. If there is no abnormality in the workstation, the data of the same operation in different cycles are similar, so the abnormality monitoring of the assembly line can be carried out based on this. First, the data points of the same operation in different cycles are put into a set for analysis.

[0044] The time series consisting of all the collected quality monitoring data points is divided into several periodic segments, and each periodic segment is matched with other periodic segments by DTW. The set consisting of the corresponding matching data points in all periodic segments is recorded as a data set. For example, there are three periodic segments in total. The third data point in one periodic segment matches the fourth data point in the first other periodic segment, and matches the third data point in the second other periodic segment. The set consisting of the third data point in the periodic segment, the fourth data point in the first other periodic segment, and the third data point in the second other periodic segment is recorded as a data set.

[0045] It can be understood that the data points in a data set are corresponding matching data points in different period segments, and thus the data points in a data set are data of the same operation in different periods.

[0046] It should be added that the time series consisting of all the collected quality monitoring data points is divided into several periodic segments, including:

[0047] The time series consisting of all the collected quality monitoring data points is recorded as the quality monitoring time series sequence, and the data segment consisting of the preset number of data points is intercepted in the quality detection time series sequence as the matching reference segment, and the matching reference segment is used to slide on the quality monitoring time series sequence. If the similarity between the data segment to be matched and the data segment at the corresponding sliding position is greater than the preset similarity threshold, the starting data point of the data segment at the corresponding sliding position is used as the cycle starting point, and the quality monitoring time series sequence is divided into several cycle segments using the cycle starting point. This embodiment is described by taking the preset number of 10 as an example, and other embodiments can take other values, as long as the value is guaranteed to be less than the number of data points collected within the beat time. This embodiment is described by taking the preset similarity threshold of 0.8 as an example, and other embodiments can take other values, and this embodiment does not make specific restrictions.

[0048] S2: Record the data of one dimension among all the quality monitoring data points in the data set as the target dimension data, and divide the accumulated distance between each target dimension data and the adjacent target dimension data on both sides by the length of the value range of the target dimension data to obtain the degree of isolation of each target dimension data; record the ratio of the maximum degree of isolation of all target dimension data to the mean degree of isolation as the abnormal possibility, and remove the dimensions whose abnormal possibility is less than the preset possibility threshold.

[0049] It should be noted that abnormal data points generally have outliers, but some dimensions of data can describe the outlier characteristics of abnormal data points, while some dimensions of data cannot. Those dimensions that cannot describe the outlier characteristics of abnormal data points are useless for abnormal data point detection and need to be removed.

[0050] Preferably, as an example, the degree of isolation of each target dimensional data is obtained by dividing the cumulative sum of the distances between each target dimensional data and the adjacent target dimensional data on both sides by the length of the value range of the target dimensional data, including:

[0051] The absolute value of the difference between the target dimensional data and the adjacent target dimensional data on one side is taken as the distance between the target dimensional data and the adjacent target dimensional data on one side, and the cumulative sum of the distances between the target dimensional data and the adjacent target dimensional data on both sides is calculated.

[0052] The difference between the maximum value and the minimum value of the target dimension data is used as the value range length of the target dimension data.

[0053] The degree of isolation of the target dimensional data is obtained by dividing the cumulative sum of the distances between the target dimensional data and the adjacent target dimensional data on both sides by the length of the value range of the target dimensional data.

[0054] It can be understood that the isolation forest algorithm performs data segmentation processing on data points by randomly selecting data as segmentation points within the value range of each dimension. Therefore, the farther the distance between a target dimension data and the target dimension data on both sides, the longer the value length between the target dimension data and the adjacent target dimension data. Therefore, the data between the target dimension data and the adjacent target dimension data is more likely to be extracted as a segmentation point, and thus the target dimension data is more likely to be segmented out separately. Therefore, the degree of isolation calculated in this way can not only reflect the isolation of the data, but also reflect the difficulty of segmenting abnormal data points.

[0055] This embodiment is described by taking the preset possibility threshold of 1.2 as an example. Other embodiments may take other values, and this embodiment does not make any specific limitation.

[0056] S3: Take the dimension whose abnormal possibility is not less than the preset possibility threshold as the dimension to be analyzed; sort the data of the dimension to be analyzed in descending order according to the degree of isolation, and record the data before the sudden change in the degree of isolation as unilateral abnormal data, and take the quality monitoring data points containing unilateral abnormal data as possible abnormal data points; combine a dimension with other dimensions with the largest number of intersections of possible abnormal data points, and if the intersection ratio of possible abnormal data points of the two dimensions in the dimension combination is less than the preset ratio threshold, both dimensions are retained; if the intersection ratio of possible abnormal points of the two dimensions in the dimension combination is not less than the preset ratio threshold, only the dimension with the large mean degree of isolation in the dimension combination is retained.

[0057] It should be noted that, through step S2, the dimensions that cannot characterize the outlier features of the abnormal data points are removed, and the dimensions that can characterize the outlier features of the abnormal data points are retained. Among the retained dimensions, some dimensions have strong correlations, that is, the outlier features of some abnormal data points are reflected in multiple dimensions, and among these multiple dimensions, some dimensions have outlier features, but their characterization capabilities are weak. Retaining these dimensions will increase the difficulty of segmenting these abnormal data points, so it is necessary to remove some dimensions with weak outlier characterization capabilities from the dimensions with strong correlations.

[0058] S30: Dimensions whose abnormal probability is not less than a preset probability threshold are used as dimensions to be analyzed; data of the dimensions to be analyzed are sorted in descending order according to the degree of isolation, and data before a sudden change in the degree of isolation occurs is recorded as unilateral abnormal data, and quality monitoring data points containing unilateral abnormal data are used as possible abnormal data points.

[0059] It should be noted that there is a big difference between the degree of isolation of normal data and abnormal data in the dimension to be analyzed. Therefore, if all the data of the dimensions to be analyzed are arranged according to the degree of isolation, there will be a large mutation in the degree of isolation from abnormal data to normal data. Therefore, possible abnormal data in the dimension to be analyzed can be obtained based on this.

[0060] Preferably, as an example, the data of the dimension to be analyzed are arranged in descending order according to the isolation degree, and the data before the isolation degree suddenly changes is recorded as unilateral abnormal data, including:

[0061] The sequence obtained by arranging the data of the dimension to be analyzed in descending order according to the degree of isolation is recorded as the time series sequence to be analyzed;

[0062] The difference between each data in the time series to be analyzed and the previous data is calculated and recorded as the first difference of each data in the time series to be analyzed, and the data position where the first difference is greater than the preset difference threshold is obtained and recorded as the possible mutation position, and the possible mutation position that does not appear again is taken as the mutation position, and the data before the mutation position in the time series to be analyzed is taken as the unilateral abnormal data. In this embodiment, the upper quartile of the first difference of all data is taken as the preset difference threshold, and other embodiments may take other values, which are not specifically limited in this embodiment.

[0063] It can be understood that, since the degree of isolation between normal data and abnormal data in the dimension to be analyzed is quite different, there is a mutation in the degree of isolation between normal data and abnormal data, and thus the position where the mutation in the degree of isolation exists is obtained by threshold segmentation. However, since some of the degrees of isolation between abnormal data have mutations and some do not, it is impossible to distinguish abnormal data from normal data only by mutation. Since the degree of isolation of abnormal data is relatively large and the degree of isolation of normal data is relatively small, there is a dividing point between normal data and abnormal data in the sequence of data in the dimension to be analyzed according to the degree of isolation, and at the same time, the degree of isolation of normal data is similar, so there is no mutation in the data after the dividing point between normal data and abnormal data, and thus a mutation point without mutation is selected from the mutation points selected above.

[0064] S31: Combine a dimension with other dimensions with the largest number of intersections of possible abnormal data points. If the intersection ratio of possible abnormal data points of two dimensions in the dimension combination is less than a preset ratio threshold, both dimensions are retained; if the intersection ratio of possible abnormal points of two dimensions in the dimension combination is not less than the preset ratio threshold, only the dimension with a large average degree of isolation in the dimension combination is retained.

[0065] Preferably, as an example, combining one dimension with other dimensions having the largest number of intersections of possible abnormal data points includes:

[0066] Calculate the number of intersection data points of possible abnormal data points in one dimension and possible abnormal data points in other dimensions, and record it as the number of intersections; combine the other dimension with the largest number of intersections of possible abnormal data points with this dimension.

[0067] Preferably, as an example, the method for obtaining the intersection ratio includes:

[0068] The ratio of the number of intersections of possible abnormal data points in two dimensions in the dimension combination to the mean of the possible abnormal data points in the two dimensions is recorded as the intersection ratio.

[0069] This embodiment is described by taking the preset ratio threshold value of 0.85 as an example. Other embodiments may take other values, and this embodiment does not make any specific limitation.

[0070] It is understandable that the greater the intersection ratio of possible abnormal data points, the more possible abnormal data points there are in the intersection of the two dimensions, and thus the greater the correlation between the two dimensions, so the abnormal data points can be separated and processed using one of the dimensions. In order to reduce the difficulty of segmentation, the dimension with a large degree of isolation is selected for retention.

[0071] S4: Based on the retained dimension, the isolation forest algorithm is used to perform anomaly detection on quality monitoring data points to realize production monitoring of the hot dry noodles production line.

[0072] The isolation forest algorithm selects dimensions from the retained dimensions to perform segmentation detection on the quality monitoring data points until the preset cutoff condition is met, thereby completing the fitting of the anomaly detection model. In this embodiment, each data point is independently segmented as a preset cutoff condition, and other embodiments may set other cutoff conditions as preset cutoff conditions, which are not specifically limited in this embodiment.

[0073] The quality monitoring data points of the workstation at the current moment are input into the anomaly detection model fitted by the corresponding data set to obtain the anomaly detection result, wherein an early warning is issued when the anomaly detection result is abnormal, and no early warning is issued when the anomaly detection result is normal.

[0074] It should be added that the corresponding data set acquisition method includes:

[0075] Obtain the data set to which the quality monitoring data point at the moment before the current moment belongs, obtain the time position of each quality monitoring data point in the data set to which the quality monitoring data point at the moment before belongs, take the next time position of the time position of the quality monitoring data point in the data set to which the quality monitoring data point at the moment before belongs as the candidate time position, and obtain the data set with the largest number of quality monitoring data points at the candidate time positions from all other data sets as the corresponding data set. For example, if the current time is 15:00, and the moment before the current time is 14:59, the data set to which the quality monitoring data point at 14:59 belongs is recorded as the first data set, the time position of each quality monitoring data point in the first data set is recorded as the candidate time position, and the data set with the largest number of quality monitoring data points at the candidate time positions is taken as the corresponding data set.

[0076] It is understandable that in order to perform anomaly detection on the quality monitoring data point at the current moment, it is necessary to obtain a data set with the same operation as the quality monitoring data point at the current moment, and then perform anomaly detection based on the anomaly detection model fitted by the corresponding data set.

[0077] At this point, this embodiment is completed.

[0078] 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 in the protection scope of the present invention.

Claims

1. An intelligent hot dry noodles production line monitoring method, characterized in that: include: Acquire a data set, wherein the data set includes quality monitoring data points of a hot dry noodle production line; The data of one dimension among all the quality monitoring data points in the data set is recorded as the target dimension data, and the degree of isolation of each target dimension data is obtained by dividing the cumulative sum of the distances between each target dimension data and the adjacent target dimension data on both sides by the length of the value range of the target dimension data; the ratio of the maximum degree of isolation of all target dimension data to the mean degree of isolation is recorded as the abnormal possibility, and the dimensions with abnormal possibilities less than the preset possibility threshold are removed; The dimension whose abnormal probability is not less than the preset probability threshold is taken as the dimension to be analyzed; Arrange the data of the dimensions to be analyzed in descending order according to the degree of isolation, record the data before the mutation of the degree of isolation as unilateral abnormal data, and take the quality monitoring data points containing unilateral abnormal data as possible abnormal data points; combine a dimension with other dimensions with the largest number of intersections of possible abnormal data points, and if the intersection ratio of possible abnormal data points of the two dimensions in the dimension combination is less than the preset ratio threshold, retain both dimensions; if the intersection ratio of possible abnormal points of the two dimensions in the dimension combination is not less than the preset ratio threshold, only retain the dimension with the largest mean degree of isolation in the dimension combination; Based on the reserved dimension, the isolation forest algorithm is used to detect anomalies of quality monitoring data points to achieve production monitoring of hot dry noodles production line; The obtaining of the data set includes: Acquire quality monitoring data points at each time at a workstation of a hot dry noodle production line, wherein the quality monitoring data points include quality monitoring data of multiple dimensions; The time series consisting of all the collected quality monitoring data points is divided into several periodic segments, each periodic segment is matched with other periodic segments by DTW, and the set consisting of the corresponding matching data points in all periodic segments is recorded as the data set.

2. The intelligent hot dry noodles production line monitoring method according to claim 1 is characterized in that: The method for obtaining unilateral abnormal data includes: The sequence obtained by arranging the data of the dimension to be analyzed in descending order according to the degree of isolation is recorded as the time series sequence to be analyzed; The difference between each data in the time series to be analyzed and the previous data is calculated and recorded as the first difference of each data in the time series to be analyzed. The data position where the first difference is greater than the preset difference threshold is obtained and recorded as the possible mutation position. The possible mutation position that does not appear again is taken as the mutation position, and the data before the mutation position in the time series to be analyzed is taken as the unilateral abnormal data.

3. The intelligent hot dry noodles production line monitoring method according to claim 1 is characterized in that: The time series consisting of all the collected quality monitoring data points is divided into several periodic segments, including: The time series sequence composed of all the collected quality monitoring data points is recorded as the quality monitoring time series sequence, and the data segment composed of a preset number of data points in the quality detection time series sequence is intercepted as the matching reference segment. The matching reference segment is used to slide on the quality monitoring time series sequence. If the similarity between the data segment to be matched and the data segment at the corresponding sliding position is greater than the preset similarity threshold, the starting data segment of the data segment at the corresponding sliding position is used as the cycle starting point, and the quality monitoring time series sequence is divided into several cycle segments using the cycle starting point.

4. The intelligent hot dry noodles production line monitoring method according to claim 1 is characterized in that: The combination of a dimension with other dimensions that have the largest number of intersections of possible abnormal data points includes: Calculate the number of intersection data points of possible abnormal data points in one dimension and possible abnormal data points in other dimensions, and record it as the number of intersections; combine the other dimension with the largest number of intersections of possible abnormal data points with this dimension.

5. The intelligent hot dry noodle production line monitoring method according to claim 4 is characterized in that: The method for obtaining the intersection ratio includes: The ratio of the number of intersections of possible abnormal data points in two dimensions in the dimension combination to the mean of the possible abnormal data points in the two dimensions is recorded as the intersection ratio.

6. The intelligent hot dry noodle production line monitoring method according to claim 1 is characterized in that: Based on the reserved dimension, the isolation forest algorithm is used to perform anomaly detection on the quality monitoring data points to realize the production monitoring of the hot dry noodles production line, including: The isolation forest algorithm selects dimensions from the retained dimensions to perform segmentation detection on the quality monitoring data points until the preset cutoff condition is met, completing the fitting of the anomaly detection model; The quality monitoring data points of the workstation at the current moment are input into the anomaly detection model fitted by the corresponding data set to obtain the anomaly detection result, wherein an alarm is issued when the anomaly detection result is abnormal, and no alarm is issued when the anomaly detection result is normal.

7. The intelligent hot dry noodle production line monitoring method according to claim 6 is characterized in that: The method for obtaining the corresponding data set includes: Obtain the data set to which the quality monitoring data point at the previous moment belongs, obtain the time position of each quality monitoring data point in the data set to which the quality monitoring data point at the previous moment belongs, take the next time position of the time position of the quality monitoring data point in the data set to which the quality monitoring data point at the previous moment belongs as the alternative time position, and obtain the data set with the largest number of quality monitoring data points at the alternative time positions among all other data sets as the corresponding data set.

Images