Real-time data detection method based on industrial internet

Abstract

The invention discloses a real-time data detection method based on the industrial internet, belongs to the field of industrial calculation models, and adopts an LSTM model to realize full-course remote monitoring of raw materials, product production states and sales data, and performs preprocessing processes such as data cleaning, unbalanced processing and data standardization on original equipment data or software system data. LSTM network training is input, a gradient descent algorithm is used for reducing a loss value and updating network parameters, a TimeGAN model-based data enhancement method is adopted, data feature distribution of a real sample is extracted through a generator, and a data relationship among raw materials, product production states and sales data is generated; according to the invention, a TimeGAN + LSTM combined algorithm is adopted to integrate the connection among yield, capacity and equipment maintenance, raw materials and products are switched, the function of adjusting the production direction is achieved, and automatic monitoring of the production process and adjustment of the production strategy are comprehensively realized through data monitoring.

Description

Technical Field

[0001] The present invention discloses an industrial computing model, and more particularly relates to a real-time data detection method based on the industrial Internet. Background Art

[0002] When industrial equipment is in operation, it collects massive amounts of time series data through sensors, controllers, and instruments for real-time data detection. By analyzing these time series data, it can provide important basis for the control, analysis, decision-making, and planning of the system's operating status.

[0003] However, these data have abnormal problems such as duplicate records, abnormal records, invalid data, inconsistent time scales, and misplaced storage. Due to the complexity of abnormal and erroneous data in multivariate time series data, an abnormal instance often appears between several variables, making it difficult to effectively detect and identify abnormal patterns, thereby affecting the stability of subsequent data; and traditional industrial data classification algorithms are sensitive to parameters and cannot achieve automatic selection of cluster centers. Summary of the invention

[0004] The purpose of the present invention is to provide a real-time data detection method based on industrial Internet in order to solve the above-mentioned problems.

[0005] To achieve the above object, the present invention provides the following technical solution: a real-time data detection method based on the industrial Internet, including an adaptive clustering algorithm and a time series data anomaly detection method, the time series data anomaly detection method is as follows: M1, define a set of time series with a time period of N as , the corresponding sequence point tuple is , and is a real value, the number of sequence elements is an integer, and the time point set of the time series is ; M2, normalizes the time subseries and passes The dimensional length vector re-represents the original time series, that is, The elements of the item are , k is the number of first items of the time subsequence elements, is the last number of elements in the time subsequence; M3, time series of industrial equipment At the point in time The sequence tuple of ; M4, set a time period , the set of sequences stored in a dislocation within a time period is , where m∈ , the misplaced storage is represented by permutation σ, is the image under permutation σ, and ; M5, defines a permutation split into several transformation processes, through the order Transform and introduce elements ,Right now ; M6, Definition There are two transformations on ( , , )and ( , , ),and ,in and All are real number elements; M7, Definition The staggered mode is ,in, ≤m and k≥1, for The conversion order is I= ; M8, models each sequence of time series data and defines each sequence as a binary group γ: (F, T), where F is the indicator set of the sequence model; M9, define the normal mode sequence as ,and for Set the confidence threshold for the two-tuple , if there is , the following conditions are met: Any subsequence in S conforms to the following model, , , and the recursion coefficients have the following relationship: ; in, For sequence by As the starting point, is the subsequence of the termination point, is a data point in the subsequence, and ; M‌‌ 10, each sequence is A subsequence of consecutive time points is used as a sliding window to determine the next time node If the data , then determine is a candidate outlier point; if there are outliers in at least two sequences in the same time period T, it is determined that a time series misalignment occurs in the time period T; where n, m, and j are all positive integer elements.

[0006] Preferably, the time series data has a repair model: K‌‌1, build about Sequential Models of Staggered Patterns , set up a directed bipartite graph ,in, is the source node; is the terminal node; E is The matching coefficient of element V in ; w is The matching probability of element V in ; u i for staggered elements about element V in ; K‌‌2, add a global starting point S and a global ending point t to the constructed bipartite graph G, S points to all vertices with in-degree 0 in G, that is ;make All the vertices in point to t. Match to , the repair equation is:

[0007]

[0008] in, , is the cost function, U and V are vertices extracted from the bipartite graph vertex set E, and is a stream function that satisfies ,and ; K3, where and They are the two longest edge vertices on the bipartite graph G, and they are iteratively selected ,join in , and temporarily delete the vertex and , match all source nodes to the final nodes, and the repair relationship is as follows: ; ; ; use The previous length of data is W Identification The algorithm also considers the abnormal data points whose matching probability in the sequence model is less than the set threshold. Add to array A.

[0009] Preferably, the adaptive clustering algorithm is constructed as follows: Y1, the data sample is represented as follows using the local density ρ and the parent point δ:

[0010] Where n is the number of samples, is the distance between data points i and j, To set the cutoff distance, is the sample density value; Y2, draw a decision diagram with ρ as the x-axis and δ as the Y-axis. The product of the local density and the distance to the parent point is , ; Y3, assuming that the data sample to be queried is p, and the K nearest neighbor relationship of p is as follows: ; in, is the set of K nearest neighbors of p, is the distance between point p and the object that is K farthest from it, is the distance between point p and the object that is qth farthest from it; Y4, the reverse K nearest neighbor of a point p is represented by a set of objects q that are regarded as K nearest neighbors of point p, and the relationship is as follows: ; in, is the distance between a set of points p and the K nearest neighbor objects q.

[0011] Compared with the prior art, the present invention has the following beneficial effects: The time series data anomaly detection method of the present invention performs anomaly detection on the time series data in the real-time data of the industrial Internet, performs dislocation storage and repair through an algorithm model, and uses an adaptive clustering algorithm to generate raw materials, product production status, and data relationships of sales data of the industrial Internet. The TimeGAN+LSTM combined algorithm is used to realize the integration of the relationship between output, production capacity and equipment maintenance, switch raw materials and production products, and achieve the function of adjusting the production direction. Through data monitoring, all-round automated monitoring of the production process and adjustment of the production strategy are realized. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0013] A real-time data detection method based on the industrial Internet includes an adaptive clustering algorithm and a time series data anomaly detection method. The time series data anomaly detection method is as follows: M1, define a set of time series with a time period of N as , the corresponding sequence point tuple is , and is a real value, the number of sequence elements is an integer, and the time point set of the time series is ; M2, normalizes the time subseries and passes The dimensional length vector re-represents the original time series, that is, The elements of the item are , k is the number of first items of the time subsequence elements, is the last number of elements in the time subsequence; M3, time series of industrial equipment At the point in time The sequence tuple of ; M4, set a time period , the set of sequences stored in a dislocation within a time period is , where m∈ , the misplaced storage is represented by permutation σ, is the image under permutation σ, and ; M5, defines a permutation split into several transformation processes, through the order Transform and introduce elements ,Right now ; M6, Definition There are two transformations on ( , , )and ( , , ),and ,in and All are real number elements; M7, Definition The staggered mode is ,in, ≤m and k≥1, for The conversion order is I= ; M8, models each sequence of time series data and defines each sequence as a binary group γ: (F, T), where F is the indicator set of the sequence model; M9, define the normal mode sequence as ,and for Set the confidence threshold for the two-tuple , if there is , the following conditions are met: Any subsequence in S conforms to the following model, , , and the recursion coefficients have the following relationship: ; in, For sequence by As the starting point, is the subsequence of the termination point, is a data point in the subsequence, and ; M‌‌ 10, each sequence is A subsequence of consecutive time points is used as a sliding window to determine the next time node If the data , then determine is a candidate outlier point; if there are outliers in at least two sequences in the same time period T, it is determined that the time series is misaligned in the time period T; where n, m, and j are all positive integer elements; According to the detection method, outliers in the sequence are identified, and then potential misaligned intervals are identified, and the misaligned subsequences in the sequence set where the misalignment occurs in the initially determined misaligned interval I are repaired; Time series data has a repair model: K‌‌1, build about Sequential Models of Staggered Patterns , set up a directed bipartite graph ,in, is the source node; is the terminal node; E is The matching coefficient of element V in ; w is The matching probability of element V in ; u i for staggered elements about element V in; K‌‌2, add a global starting point S and a global ending point t to the constructed bipartite graph G, S points to all vertices with in-degree 0 in G, that is ;make All the vertices in point to t. Match to , the repair equation is:

[0014]

[0015] in, , is the cost function, U and V are vertices extracted from the bipartite graph vertex set E, and is a stream function that satisfies ,and ; K3, where and They are the two longest edge vertices on the bipartite graph G, and they are iteratively selected ,join in , and temporarily delete the vertex and , match all source nodes to the final nodes, and the repair relationship is as follows: ; ; ; use The previous length of data is W Identification The algorithm also considers the abnormal data points whose matching probability in the sequence model is less than the set threshold. Add to array A; By performing statistical analysis on the staggered data, it is possible to identify the staggered sequence fragments. The matching probability is high, but the matching degree with other sequence models is low. Therefore, a bipartite graph is used for fast matching. An array A is maintained for each sequence tuple at time point t to record the sequence of the mismatch. In the abnormal sequence detection stage, Perform detection, and when the same time is detected When the number of elements M in array A is not less than two, the algorithm records these M sequences into array A and creates a matrix A=|A|×|A| for array A. For each staggered data recorded in array A, ,calculate The matching probability of the sequence model to which all elements in array A belong, and right The matching probability of the recorded sequence pattern is recorded in the corresponding position of matrix A. After the matrix is ​​constructed, the repair calculation of the misaligned data is performed according to Construct a bipartite graph G and perform and Perform matching calculations to obtain matching results ,when If it is greater than the set threshold, the algorithm stops outputting the current misaligned sequence data.

[0016] The adaptive clustering algorithm is constructed as follows: Y1, the data sample is represented as follows using the local density ρ and the parent point δ:

[0017] Where n is the number of samples, is the distance between data points i and j, To set the cutoff distance, is the sample density value; Y2, draw a decision diagram with ρ as the x-axis and δ as the Y-axis. The product of the local density and the distance to the parent point is , ; Y3, assuming that the data sample to be queried is p, and the K nearest neighbor relationship of p is as follows: ; in, is the set of K nearest neighbors of p, is the distance between point p and the object that is K farthest from it, is the distance between point p and the object that is qth farthest from it; Y4, the reverse K nearest neighbor of a point p is represented by a set of objects q that are regarded as K nearest neighbors of point p, and the relationship is as follows: ; in, The distance between a set of points p and the K nearest neighbor objects q; The local density of the cluster center is greater than the local density of its neighbors. The distances between different cluster centers are relatively far. Find a data point that is closest to the data point and has a density greater than that of the point, i.e., the parent point. The distance between the two points is the parent point distance of the point. The point with the largest local density in the sample space has a value of δ, which is the distance from it to the farthest point in the sample space because there is no point with a higher local density. The value generally decays exponentially, which is convenient for selecting cluster centers. After determining the cluster centers, different labels are assigned to them. Their labels belong to the cluster to which their parent points belong. If the parent point has no label information, it continues to search for the parent point upwards, and finally completes the label assignment task for all data. Based on this, the LSTM model is used to realize full remote monitoring of raw materials, product production status, and sales data. The original equipment data or software system data is preprocessed by data cleaning, unbalanced processing, data standardization, etc., and then put into LSTM network training. The gradient descent algorithm is used to reduce the loss value and update the network parameters. The data enhancement method based on the TimeGAN model is used to extract the data feature distribution of real samples through the generator, and generate the data relationship between raw materials, product production status, and sales data. The algorithm combined with TimeGAN+LSTM is used to realize the integration of production, production capacity, and equipment maintenance. The connection between switching raw materials and production products is achieved to adjust the production direction. Through data monitoring, the production process is fully automated and monitored and the production strategy is adjusted.

[0018] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0019] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A real-time data detection method based on industrial Internet, characterized in that: It includes an adaptive clustering algorithm and a time series data anomaly detection method, and the time series data anomaly detection method is as follows: M1, define a set of time series with a time period of N as , the corresponding sequence point tuple is , and is a real value, the number of sequence elements is an integer, and the time point set of the time series is ; M2, normalizes the time subseries and passes The dimensional length vector re-represents the original time series, that is, The elements of the item are , k is the number of first items of the time subsequence elements, is the last number of elements in the time subsequence; M3, time series of industrial equipment At the point in time The sequence tuple of ; M4, set a time period , the set of sequences stored in a dislocation within a time period is , where m∈ , the misplaced storage is represented by permutation σ, is the image under permutation σ, and ; M5, defines a permutation split into several transformation processes, through the order Transform and introduce elements ,Right now ; M6, Definition There are two transformations on ( , , )and ( , , ),and ,in and All are real number elements; M7, Definition The staggered mode is ,in, ≤m and k≥1, for The conversion order is I= ; M8, models each sequence of time series data and defines each sequence as a binary group γ: (F, T), where F is the indicator set of the sequence model; M9, define the normal mode sequence as ,and for Set the confidence threshold for the two-tuple , if there is , the following conditions are met: Any subsequence in S conforms to the following model, , , and the recurrence coefficient has the following relationship: ； in, For sequence by As the starting point, is the subsequence of the termination point, is a data point in the subsequence, and ; M‌‌ 10, each sequence is A subsequence of consecutive time points is used as a sliding window to determine the next time node If the data , then determine is a candidate outlier point; if there are outliers in at least two sequences in the same time period T, it is determined that a time series misalignment occurs in the time period T; where n, m, and j are all positive integer elements.

2. A real-time data detection method based on industrial Internet according to claim 1, characterized in that: The time series data has a repair model: K‌‌1, build about Sequential Models of Staggered Patterns , set up a directed bipartite graph ,in, is the source node; is the terminal node; E is The matching coefficient of element V in ; w is The matching probability of element V in ; u i for staggered elements about element V in ; K‌‌2, add a global starting point S and a global ending point t to the constructed bipartite graph G, S points to all vertices with in-degree 0 in G, that is ;make All the vertices in point to t. Match to , the repair equation is: ； ； in, , is the cost function, U and V are vertices extracted from the bipartite graph vertex set E, and is a stream function that satisfies ,and ; K3, where and They are the two longest edge vertices on the bipartite graph G, and they are iteratively selected ,join in , and temporarily delete the vertex and , match all source nodes to the final nodes, and the repair relationship is as follows: ； ； ； use The previous length of data is W Identification The algorithm also considers the abnormal data points whose matching probability in the sequence model is less than the set threshold. Add to array A.

3. A real-time data detection method based on industrial Internet according to claim 2, characterized in that: The adaptive clustering algorithm is constructed as follows: Y1, the data sample is represented as follows using the local density ρ and the parent point δ: ； Where n is the number of samples, is the distance between data points i and j, To set the cutoff distance, is the sample density value; Y2, draw a decision diagram with ρ as the x-axis and δ as the Y-axis. The product of the local density and the distance to the parent point is , ; Y3, assuming that the data sample to be queried is p, and the K nearest neighbor relationship of p is as follows: ； in, is the set of K nearest neighbors of p, is the distance between point p and the object that is K farthest from it, is the distance between point p and the object that is qth farthest from it; Y4, the reverse K nearest neighbor of a point p is represented by a set of objects q that are regarded as K nearest neighbors of point p, and the relationship is as follows: ； in, is the distance between a set of points p and the K nearest neighbor objects q.

Images