Multi-modal chemical process fault detection method based on improved t-SNE

Abstract

The invention relates to a multi-modal chemical process fault detection method based on improved t-SNE. The method comprises the steps: S1, collecting multi-modal chemical process original data Xo and carrying out standardization processing to obtain high-dimensional data X; S2, calculating a mahalanobis distance DM between the high-dimensional data points; S3, performing feature extraction on the high-dimensional data X by adopting an improved t-distribution random neighbor embedding method t-SNE to obtain a low-dimensional matrix Y; S4, solving a mapping matrix A projected from the high-dimensional space to the low-dimensional space; S5, solving a residual space E of the training data; S6, solving a feature space and a residual space of the online data; S7, using a local outlier factor LOF algorithm to construct an LOF statistic; S8, calculating the LOF statistical magnitude and the corresponding control limit of the training data; and S9, calculating the LOF statistic of the online data, and carrying out real-time fault detection. Compared with the prior art, the multi-modal process monitoring requirement can be met, and the accuracy is high.

Description

technical field [0001] The invention relates to the field of chemical production process monitoring, in particular to a multimodal chemical process fault detection method based on improved t-SNE. Background technique [0002] With the support of distributed control system technology, Internet of Things, big data and other technologies, the chemical production process is becoming more and more automated and intelligent. Among them, process monitoring is the key link to ensure safe and efficient production, and fault detection is the most basic and most important link in process monitoring. Based on the data-driven process monitoring technology developed on the basis of multivariate statistical process monitoring (MSPM) theory, effective information can be mined from massive process data to realize the modeling of chemical process, so as to achieve the purpose of monitoring. [0003] In the process of chemical production, due to factors such as seasonal changes and demand flu...

AI Technical Summary

Problems solved by technology

However, when the classic t-SNE algorithm measures the distance between data points, it uses the calculation of the Euclidean distance, and then converts the Euclidean distance into a probability. This method is only suitable for the modeling of single-mode chemical processes.

After the modeling is completed, the corresponding statistics are constructed to realize fault detection. The traditional statistics T 2 and SPE need to meet the requirement that the data obey the normal distribution, but the chemical process data after dimensionality reduction usually have non-Gaussian characteristics, use T 2 and SPE statistics will affect the estimation of control limits and the final fault detection effect

Images