Multivariate monitoring time series regression prediction method for energy chemical production system

Abstract

The invention discloses a multivariate monitoring time sequence regression prediction method for an energy chemical production system, and the method comprises the steps: measuring the coupling relation of multivariate monitoring time of the energy chemical system through employing a symbol transfer entropy calculation method, and determining an auxiliary monitoring time sequence which facilitatesthe regression prediction precision of a target monitoring time sequence according to the strength degree of the coupling relationship; on this basis, a deep echo state network model is adopted as aregression prediction model, and in order to improve the nonlinear mapping capability of the model, a LeakyReLU activation function is adopted to optimize the activation function of the regression prediction model, so that the influence of the irrelevant monitoring time sequence on the regression prediction of the target monitoring time sequence is avoided. The nonlinear mapping capability of themodel is stronger, and the differential evolution algorithm is adopted to perform hyperparameter optimization of the regression prediction model, so that the regression prediction precision of the multivariate monitoring time sequence is higher, and an effective scheme is provided for realizing detection and data reconstruction of an abnormal instrument. The invention has great significance for maintaining safe and stable operation of an energy chemical production system.

Description

technical field [0001] The invention belongs to the field of regression prediction of monitoring time series of energy and chemical production systems, and in particular relates to a multivariate monitoring time series regression prediction method of energy and chemical production systems. Background technique [0002] In the energy and chemical production system, the monitoring data of various measuring instruments, sensors and other measuring equipment is the basis for equipment status monitoring and precise production control. However, many measuring equipment in the system often send false data to the control system due to failures. Thus affecting the smooth operation of the system, it is of great significance to maintain the safe and stable operation of the system to realize the detection of abnormal instruments and data reconstruction based on the multivariate monitoring time series regression prediction method of the energy and chemical production system. [0003] At ...

AI Technical Summary

Problems solved by technology

However, the existing research on regression prediction of multiple monitoring time series in energy and chemical production systems does not have a strict selection method for auxiliary monitoring time series, and the regression prediction models used are generally based on principal component regression analysis, Kalman filter, etc. Models for Mapping Capabilities

However, in practical engineering applications, the strength of the coupling relationship between the monitoring time series in the energy and chemical production system is very different. Some monitoring time series have a tight coupling relationship, while others have only a weak coupling relationship. The regression prediction of time series, the latter is the opposite; more importantly, the monitoring time series in the energy and chemical production system has the characteristics of chaos, nonlinearity, non-stationarity, etc. Unsatisfactory results obtained on regression prediction tasks

Images