Noise-containing nonlinear process fault detection model construction method and detection method thereof

Abstract

The invention relates to the technical field of fault detection methods based on manifold learning, in particular to a noise-containing nonlinear process fault detection model construction method and a detection method thereof. The construction method comprises the following steps: carrying out standardization processing on a sample data set to obtain a calibration data set of the sample data set; calculating the standard deviation and the standard residual error of the calibration data set according to the calibration data set, removing noise data in the calibration data set to obtain a normal data set, and calculating a low-dimensional representation data set of the normal data set; and according to the low-dimensional representation data set, establishing a fault multi-classification model by using a support vector machine algorithm. The method is based on mathematical statistics, and regression analysis is carried out on manifold characteristics of original data by using standardized residual errors, noise is identified and effectively isolated, the accuracy of a fault detection model is improved, accurate acquisition of nonlinear principal elements of sample data is realized by improving an Isomap algorithm, and the noise robustness of the isometric mapping algorithm is enhanced.

Description

technical field [0001] The invention relates to the technical field of fault detection methods based on manifold learning, in particular to a method for constructing a fault detection model of a nonlinear process with noise and a detection method thereof. Background technique [0002] Modern industrial systems are large in scale and increasingly complex, making the monitoring data in the chemical production process often present high-dimensional and strongly nonlinear characteristics. Reducing the complexity of process data and accurately obtaining the nonlinear pivot of monitoring data provides an important guarantee for real-time and efficient fault detection in industrial processes. Now widely used linear feature extraction algorithms, such as principal component analysis, linear discriminant method, and multidimensional scale analysis, cannot obtain nonlinear feature information in data, which can easily cause loss of key information, which is not conducive to improving ...

AI Technical Summary

Problems solved by technology

[0006] In the actual production process, when the system equipment is running normally or fails, there are often strong magnetic fields, high temperatures or other unknown irregular disturbances around the equipment sensor, which makes the data transmitted by the sensor contain abnormal or noise data. When the data is used as system operating status data for system fault detection, the irregular characteristics of abnormal or noise data often lead to low accuracy of fault detection results and a high rate of missed detection, which is far from meeting the safety requirements of actual industrial processes

Images