CN-122020457-A - Process measurement point trend diagnosis method and system based on subspace smooth incremental spectrum analysis

CN122020457ACN 122020457 ACN122020457 ACN 122020457ACN-122020457-A

Abstract

The invention relates to a process measurement point trend diagnosis method and system based on subspace smooth incremental spectrum analysis, which comprises the steps of performing incremental processing on a process measurement point historical data sequence, mapping the process measurement point historical data sequence into an incremental subspace matrix, performing smooth processing on each subspace vector in the incremental subspace matrix, performing frequency domain projection analysis on all subspace vectors in the incremental subspace matrix to obtain long-term incremental power spectrum statistical characteristics of a process measurement point, performing short-term data spectrum estimation on the basis of a constructed simulated online subspace to obtain the maximum multiple of error allowance and the corresponding short-term data window length, and performing online diagnosis on real-time process data according to the process measurement point configuration situation and combining the determined long-term incremental power spectrum statistical characteristics, the maximum multiple of error allowance and the corresponding short-term data window length of the process measurement point to obtain a process measurement point trend diagnosis result. The invention can be widely applied to the field of fault diagnosis in the process industry.

Inventors

WANG ZHU
WANG KAIMING

Assignees

中国石油大学(北京)

Dates

Publication Date: 20260512
Application Date: 20260115

Claims (10)

1. The process measurement point trend diagnosis method based on subspace smooth incremental spectrum analysis is characterized by comprising the following steps of: performing incremental processing on the process measurement point historical data sequence, mapping the process measurement point historical data sequence into an incremental subspace matrix, and performing smoothing processing on each subspace vector in the incremental subspace matrix; carrying out frequency domain projection analysis on all subspace vectors in the incremental subspace matrix to obtain long-time incremental power spectrum statistical characteristics of the process measuring points; short-segment data spectrum estimation is carried out based on the constructed simulated online subspace, and the maximum allowable error multiple and the corresponding short-segment data window length are obtained; and according to the configuration condition of the process measurement points, carrying out online diagnosis on the real-time process measurement point data by combining the determined measurement point increment power spectrum statistical characteristics, the maximum allowable error multiple and the corresponding short data window length to obtain a process measurement point trend diagnosis result.
2. The process measurement point trend diagnosis method based on subspace smoothing incremental spectrum analysis of claim 1, wherein the incremental processing and mapping of the process measurement point history data sequence into an incremental subspace matrix, and smoothing of each subspace vector in the incremental subspace matrix, comprises: Performing incremental processing on the historical data sequence of the process measuring point to obtain an incremental data sequence; constructing an incremental subspace matrix by utilizing a segmentation mapping technology based on the incremental data sequence; And adopting an iterative statistical smoothing technology to carry out smoothing treatment on all subspace vectors in the incremental subspace matrix.
3. The process measurement point trend diagnosis method based on subspace smoothing incremental spectrum analysis of claim 2, wherein the smoothing of all subspace vectors in the incremental subspace matrix by using iterative statistical smoothing technique comprises: ① For incremental subspace matrix The first of (3) Subspace vector Calculating the mean value and standard deviation of the model (C); ② For the first Subspace vector Traversing each data point of the database, and carrying out outlier judgment on each data point by using a pre-constructed outlier criterion based on statistical distribution; ③ For data points determined to be outliers Performing median filtering smooth replacement; ④ Using new pairs of data points after replacement Subspace vector Updating to obtain the first Each smooth subspace vector ; ⑤ Repeating the steps ① - ④ until the subspace matrix is incremented No element meeting outlier criterion exists in the matrix to obtain a smooth increment subspace matrix 。
4. The process measurement point trend diagnosis method based on subspace smooth incremental spectrum analysis of claim 3, wherein the frequency domain projection analysis is performed on all subspace vectors in the incremental subspace matrix to obtain the long-term incremental power spectrum statistical characteristics of the process measurement points, and the method comprises the following steps: for the first Each smooth subspace vector Spectrum estimation is carried out by utilizing wiener-Xin Qinding theory to obtain a smooth increment subspace matrix Power spectral density matrix of (2) ; Establishing a frequency domain statistical reference of a process measurement point, and carrying out power spectrum density matrix And (3) carrying out statistical feature calculation to obtain the long-time increment power spectrum statistical feature of the process measuring point, wherein the long-time increment power spectrum statistical feature comprises a reference mean value and a reference standard deviation.
5. The process measurement point trend diagnosis method based on subspace smooth incremental spectrum analysis according to claim 4, wherein the short-segment data spectrum estimation is performed based on the constructed simulated online subspace to obtain the maximum allowable error multiple and the corresponding short-segment data window length, and the method comprises the following steps: Constructing a simulated online increment subspace matrix according to a short-segment data window set by actual online diagnosis, and performing smoothing treatment to obtain a smooth simulated online increment subspace matrix; Carrying out spectrum estimation on each analog online increment subspace vector in the smooth analog online increment subspace matrix to obtain a short-time standard deviation; The obtained short-time standard deviation is compared with the reference standard deviation point by point to obtain error amplification sequences of different frequency points; selecting the maximum value in the error amplification sequences in all frequency points as the maximum allowable multiple of the global error; and checking the obtained maximum allowable error multiple until obtaining the effective maximum allowable multiple and the corresponding short-segment data window length.
6. The process measurement point trend diagnosis method based on subspace smooth incremental spectrum analysis according to claim 1, wherein the online diagnosis of real-time process measurement point data according to the process measurement point configuration situation by combining the determined long-time incremental power spectrum statistical characteristics, the error allowable maximum multiple and the corresponding short-segment data window length of the process measurement point comprises the following steps: according to the configuration condition of the process measuring points, carrying out long-time linear abnormality judgment on the real-time process measuring point data; According to the long-time increment power spectrum statistical characteristics, the maximum allowable error multiple and the corresponding short-segment data window length of the determined process measurement point, trend abnormality judgment is carried out on the real-time process measurement point data; When trend abnormality occurs, extracting local trend of real-time process measurement point data, and judging short-time mutation abnormality; And outputting a diagnosis state code according to the abnormal judgment result, and triggering a corresponding alarm signal.
7. The process measurement point trend diagnosis method based on subspace smooth incremental spectrum analysis according to claim 6, wherein the long-time straight line abnormality judgment is performed on real-time process measurement point data according to the process measurement point configuration condition, and the method comprises the following steps: for real-time process measuring point data sequence Performing a moving average filtering process to obtain a smooth sequence ; Traversing a smooth sequence Statistics satisfy non-decrementing count values And a non-incremented count value ; Counting the value And And comparing the linear motion abnormal value with a preset threshold value, and judging whether the linear motion abnormal value occurs for a long time according to a comparison result.
8. The process measurement point trend diagnosis method based on subspace smooth incremental spectrum analysis according to claim 6, wherein the trend anomaly determination of the real-time process measurement point data according to the long-time incremental power spectrum statistics, the error allowable maximum multiple and the corresponding short-segment data window length based on the determined process measurement point comprises: based on real-time process measurement point data Constructing an online real-time increment subspace matrix, and performing spectrum estimation on each subspace vector of the online real-time increment subspace matrix to obtain a power spectrum density sequence; and taking the maximum allowable error multiple as an initial reference standard of an online diagnosis threshold value, judging the power spectrum density sequence of each subspace vector, and judging whether trend abnormality occurs according to a judgment result.
9. The process measurement point trend diagnosis method based on subspace smooth incremental spectrum analysis according to claim 6, wherein when trend abnormality occurs, extracting local trend of real-time process measurement point data to perform short-time mutation abnormality judgment comprises: For real-time process measurement point data Is truncated and the median value of the beginning segment of the end window is extracted And median value of end segment ; According to the median value of the initial segment And median value of end segment Calculating normalized change rate And judging short-time mutation abnormality according to the calculation result.
10. A process measurement point trend diagnostic system based on subspace smooth delta spectrum analysis, comprising: the incremental subspace construction and vector sequence smoothing module is used for carrying out incremental processing on the process measurement point historical data sequence, mapping the process measurement point historical data sequence into an incremental subspace matrix and carrying out smoothing processing on each subspace vector in the incremental subspace matrix; the subspace incremental spectrum statistical feature mining module is used for carrying out frequency domain projection analysis on all subspace vectors in the incremental subspace matrix to obtain long-term incremental power spectrum statistical features of the process measurement points; The error allowable maximum multiple calculation module is used for estimating a short-segment data spectrum based on the constructed analog online subspace to obtain the error allowable maximum multiple and the corresponding short-segment data window length; And the online trend diagnosis module is used for carrying out online diagnosis on the real-time process measurement point data according to the process measurement point configuration situation and combining the determined measurement point increment power spectrum statistical characteristics, the maximum allowable error multiple and the corresponding short data window length to obtain a process measurement point trend diagnosis result.

Description

Process measurement point trend diagnosis method and system based on subspace smooth incremental spectrum analysis Technical Field The invention relates to the field of fault diagnosis in the process industry, in particular to a process measurement point trend diagnosis method and system based on subspace smooth incremental spectrum analysis. Background The process industry is a national economic prop industry and is an important engine for driving the operation of a modern industrial system. In the large-scale refining and fine chemical production process, the continuous, large-scale and automatic degree of the device are increasingly improved, and complicated physicochemical reactions are accompanied in the production flow. As a core index reflecting the operating state of the device, the process measurement points include conventional process parameters (such as temperature, pressure, liquid level, flow rate, etc.), and also include equipment and environmental parameters, and the stability of these key measurement points is directly related to production efficiency, product quality and energy consumption. Once these stations have an abnormal trend, they are often early signs of equipment failure, control loop failure, or deterioration of process conditions. If the device can not be found and processed in time, the product is unqualified and the energy consumption is increased if the device is light, and the chain reaction is caused if the device is heavy, so that the device is stopped in an unscheduled way, even serious safety and environmental protection accidents are caused, and immeasurable losses are caused to personnel, equipment and environment. Trend diagnosis of process measurement points is an important link for guaranteeing stable operation of chemical processes. In the actual production process, the complex characteristics of non-stability and non-linearity are presented under the influence of multiple factors such as fluctuation of raw material properties, equipment aging, environmental interference and the like. Currently, the actual field is still mainly based on fixed threshold alarm of a Distributed Control System (DCS) and experience inspection of operators. The passive management mode is difficult to capture the fine abnormal trend of early process measurement points, is easy to be interfered by working condition adjustment, and is difficult to meet the urgent demands of modern process industry on digital and intelligent operation and maintenance. At present, scholars at home and abroad have more researches on a process measurement point trend diagnosis method, and the process measurement point trend diagnosis method can be totally divided into three methods based on analytical models, qualitative experience knowledge and data driving. Because of the complex mechanism of chemical processes, it is difficult to build accurate mathematical models, and expert experience is difficult to obtain, data-driven based methods are widely favored because they only use equipment operating data. Under the framework of data driving, the method is mainly divided into methods based on multivariate statistical analysis, artificial intelligence and signal processing. The method based on multivariate statistical analysis realizes diagnosis by mining historical data of process variables and extracting statistical features by using a projection technology. The main representative algorithms include Principal Component Analysis (PCA), partial Least Squares (PLS), independent principal component analysis (ICA), and the like. The method uses the statistic index to diagnose the process state, and can effectively process the correlation among variables. However, the method is mainly based on the static distribution assumption of the data, and dynamic time sequence information and frequency domain energy distribution characteristics contained in the data sequence are often ignored. For common non-stationary oscillation, DC component drift or disturbance of specific frequency in the chemical process, the physical essence of the method is difficult to be revealed by a simple statistical projection method, and the adaptability to nonlinear working conditions is poor. The method based on artificial intelligence mainly utilizes a machine learning algorithm to establish nonlinear mapping of fault characteristics and categories. Mainly comprises an Artificial Neural Network (ANN), a Support Vector Machine (SVM), fuzzy logic and the like. However, the method has obvious limitations in practical engineering application, namely, on one hand, the method belongs to a black box model generally, has poor physical interpretability, and is difficult for operation and maintenance personnel to understand the mechanism behind an alarm, and on the other hand, the method is highly dependent on massive fault sample data with complete marks for training. In actual chemical production, failure samples are often scarce, resulting in limite