Industrial system nonlinear process monitoring method based on space-time local information neural network

Through the method based on spatiotemporal local information neural network, the problem that traditional technology is difficult to extract local information of industrial system data is solved, and more efficient nonlinear feature extraction and fault detection are achieved, ensuring the safety of industrial processes.

CN119087939APending Publication Date: 2024-12-06BEIJING UNIV OF CHEM TECH
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411241092.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

Traditional multivariate statistical process monitoring technology is difficult to fully extract low-dimensional manifold embedded information in industrial system data, especially local information, which leads to insufficient monitoring of equipment safety and product quality.

Method used

Using a method based on spatiotemporal local information neural network, the space-time neighbors of data are selected through spatiotemporal sliding window technology, and the STLNN model is constructed for nonlinear feature extraction, and the model parameters are optimized using loss function and gradient descent method, and the local neighborhood is maintained self-encoding. and kernel density estimation method to monitor the operating status of industrial systems.

Benefits of technology

It improves the nonlinear feature extraction capability, retains local information of the original data, improves the accuracy and interpretability of industrial system fault detection, and ensures the safe and reliable operation of industrial processes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119087939A_ABST
    Figure CN119087939A_ABST
Patent Text Reader

Abstract

The invention provides an industrial system non-linear process monitoring method based on a space-time local information neural network (space-time local information neural network). The industrial system non-linear process monitoring method based on the space-time local information neural network is provided. According to the method, a manifold learning technology and a neural network framework are combined, nonlinear features are extracted from original data more effectively, space-time local information of the nonlinear features is reserved, and the detection rate and accuracy of industrial system faults are improved. By introducing a time sliding window and a local neighborhood preserving auto-encoder method, the time-space local neural network can more accurately identify abnormities and faults in the industrial process, thereby ensuring the safety and reliability of industrial production.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of engineering detection, and in particular to a method for monitoring nonlinear processes of industrial systems based on spatiotemporal local information neural networks. Background Art

[0002] With the continuous development of industrial technology, especially the increasing requirements for equipment safety and product quality in complex industrial fields such as the chemical industry, industrial process monitoring has become an indispensable part. Although traditional multivariate statistical process monitoring (MSPM) techniques such as principal component analysis (PCA) and partial least squares (PLS) can process high-dimensional data, it is difficult to fully extract the low-dimensional manifold embedded information in the data, that is, local information. In recent years, methods based on multivariate learning and neural networks have received widespread attention.

[0003] In order to solve the above problems, the applicant proposes a method for monitoring nonlinear processes in industrial systems based on spatiotemporal local information neural networks. Summary of the invention

[0004] The purpose of the present invention is to provide a method for monitoring nonlinear processes in industrial systems based on a spatiotemporal local information neural network to solve the problems in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for monitoring a nonlinear process of an industrial system based on a spatiotemporal local information neural network, comprising the following steps:

[0006] Data selection and preprocessing: Collect process data from industrial systems and perform standardized processing;

[0007] Time and space neighbor selection: Use the time and space sliding window technology to select the time and space neighbors of the data to retain the local information of the original data;

[0008] Constructing a spatiotemporal local neural network: taking the selected spatiotemporal neighbors as input, and performing nonlinear feature extraction through the STLNN model;

[0009] Training the STLNN model: Use the loss function to constrain nonlinear features and use the gradient descent method to optimize model parameters until the loss function converges;

[0010] Monitoring process: The trained STLNN model is used to monitor the real-time process data of the industrial system, evaluate the operating status through T2 and SPE statistics, and set thresholds to identify anomalies and faults.

[0011] Optionally, the spatiotemporal sliding window technique can simultaneously consider the spatial and temporal proximity of data to form a spatiotemporal neighbor set.

[0012] Optionally, the spatiotemporal local neural network model includes an input layer, a hidden layer and an output layer, wherein the hidden layer is used to extract nonlinear features of the data and to maintain local information of the data through a local neighborhood preserving autoencoder method.

[0013] Optionally, the loss function includes a reconstruction error term and a local loss term, which are used to constrain the spatiotemporal local neural network model to retain the local structure of the original data during the feature extraction process.

[0014] Optionally, the gradient descent method is implemented by the Adam optimizer in PyTorch to iteratively update the parameters in the STLNN model.

[0015] Optionally, the T2 and SPE statistics are used to monitor the operating status of the feature space, and their thresholds are determined by a kernel density estimation method to determine whether an abnormality or failure occurs in the industrial system.

[0016] According to the method described in any one of claims 1 to 6, the method is particularly suitable for industrial systems in the chemical industry, can improve the detection rate of equipment failures and abnormal working conditions, and ensure the safe and reliable operation of industrial processes.

[0017] Beneficial effects: 1. The present invention integrates the manifold learning method into the neural network framework, improving the ability to extract nonlinear features from raw data.

[0018] 2. By designing a local neighborhood preserving autoencoder method, the local information of the original data is retained and the interpretability of the reduced dimensionality data is improved.

[0019] 3. Considering the temporal and spatial information of the data at the same time improves the fault detection rate of the nonlinear process of the industrial system. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a diagram of the spatiotemporal local neural network structure of an embodiment of the present invention;

[0021] Figure 2 It is a convergence diagram of the loss function of an embodiment of the present invention;

[0022] Figure 3 It is an overall flow chart of an embodiment of the present invention;

[0023] Figure 4 This is a dimensionality reduction effect diagram of an embodiment of the present invention;

[0024] Figure 5 This is a monitoring result diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following describes the preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.

[0026] Example 1

[0027] Data selection and preprocessing process

[0028] The study selected two classic data sets that reflect the manifold structure and an actual process data set of a thermal power plant. The effectiveness of the proposed method was verified by analyzing the actual monitoring data of the power plant, and 13 variables were selected in the present invention. The faults that occurred in the thermal power plant were that the A pulverizing system was abnormal and the A coal feeder was broken. The inspectors found that the coal output of the 1A coal feeder dropped rapidly to 0. The 1A raw coal bin was immediately vibrated, but it was found to be ineffective. It was immediately reported that a foreign object was stuck in the 1A coal feeder pipeline and it needed to be closed, and then manual load reduction was performed, and the coal mills of other units were started at the same time. A few hours later that day, the measurement point returned to normal. The data collected during the normal operation of the power plant a few hours before the failure was used as training data. The test data uses a period of time before the failure occurred. After selecting the data set, it is subjected to standardization operations.

[0029] Temporal and spatial neighbor selection

[0030] The present invention defines the selection result of the nearest neighbor in the W matrix. Unlike the LPP method, the LPP only covers the structure of the space because the W matrix only considers the Euclidean distance and ignores and omits the information of time. A new sliding window constraint is added to the W matrix of the present invention so that it can include information from time and space. Therefore, the selection of the nearest neighbor in the two methods has also changed. The definition of the W matrix is:

[0031]

[0032] Among them Ψ i For x i 's space-time neighbors.

[0033] The figure below shows the difference between the neighbor point selection of the W matrix and the LPP matrix in the present invention. It can be seen from the figure that the neighbors of each data point in the LPP are scattered, and the selection of the neighbors of the newly invented W matrix is ​​centralized and regular, which can better retain local information in time and space.

[0034] Training phase

[0035] The total cost function of the present invention is constructed as

[0036]

[0037] Where E1 represents the reconstruction error of the neural network, E 2 Represents the constraint on the nearest neighbor. Using this constraint, the data is guaranteed to maintain the same local neighborhood structure in the input space and feature space. This preserves the local neighborhood information in the spatiotemporal local neural network.

[0038] In the training phase, w and b are randomly assigned, and then the gradient descent method is used to iteratively update the parameters to minimize the objective function, which is calculated as

[0039]

[0040] Where η is the learning rate.

[0041] The gradient descent method in parameter update in the present invention is implemented by the optimizer in PyTorch, and the Adam optimizer is used here to complete this work. The Adam optimizer is a gradient-based optimization algorithm that is widely used in some deep learning algorithms. The model is iterated in a loop, and the parameters w and b in the model are updated by back propagation until convergence. In the spatiotemporal local neural network, as the number of iterations increases, the loss function eventually tends to converge.

[0042] T2 and SPE are two commonly used statistics in process monitoring. They help identify abnormal operating processes. The spatiotemporal local neural network method uses T2 statistics and SPE statistics to monitor the operating status of the feature space.

[0043] T2 and SPE are

[0044]

[0045] SPE(i)=∥x i -x i ∥ 2

[0046] This paper selects the commonly used kernel density estimation (KDE) to determine the thresholds T2limit and SPElimit. When a new sample arrives, the corresponding statistics are calculated. When any indicator of the new sample exceeds the corresponding threshold, it is considered that a fault has occurred. The complete fault detection algorithm is shown as follows.

[0047] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0048] 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 method for monitoring nonlinear processes in industrial systems based on spatiotemporal local information neural networks, characterized in that: The following steps are involved: Data selection and preprocessing: Collect process data from industrial systems and perform standardized processing; Time and space neighbor selection: Use the time and space sliding window technology to select the time and space neighbors of the data to retain the local information of the original data; Constructing a spatiotemporal local neural network: Taking the selected spatiotemporal neighbors as input, nonlinear feature extraction is performed through the spatiotemporal local neural network model; Training the spatiotemporal local neural network model: using the loss function to constrain the nonlinear features and using the gradient descent method to optimize the model parameters until the loss function converges; Monitoring process: Use the trained spatiotemporal local neural network model to monitor the real-time process data of the industrial system, evaluate the operating status through T2 and SPE statistics, and set thresholds to identify anomalies and faults.

2. The method according to claim 1, characterized in that The spatiotemporal sliding window technique can simultaneously consider the spatial and temporal proximity of data to form a spatiotemporal neighbor set.

3. The method according to claim 1, characterized in that The spatiotemporal local neural network model comprises an input layer, a hidden layer and an output layer, wherein the hidden layer is used to extract the nonlinear characteristics of the data and to maintain the local information of the data through a local neighborhood preserving autoencoder method.

4. The method according to claim 1, characterized in that The loss function includes a reconstruction error term and a local loss term, which are used to constrain the spatiotemporal local neural network model to retain the local structure of the original data during the feature extraction process.

5. The method according to claim 1, characterized in that The gradient descent method is implemented through the Adam optimizer in PyTorch to iteratively update the parameters in the spatiotemporal local neural network model.

6. The method according to claim 1, characterized in that The T2 and SPE statistics are used to monitor the operating status of the feature space, and their thresholds are determined by a kernel density estimation method to determine whether an abnormality or failure occurs in the industrial system.

7. The method according to any one of claims 1 to 6, characterized in that The method is particularly suitable for industrial systems in the chemical industry, and can improve the detection rate of equipment failures and abnormal working conditions, and ensure the safe and reliable operation of industrial processes.

Citation Information

Cited By

  • Industrial process monitoring method

    CN120276381A