A Soft-Sensor Modeling Method for Debutanizer Tower Based on Aliesn Online Learning Algorithm

Abstract

The invention discloses a debutanizer soft sensor modeling method based on ALIESN online learning algorithm. This method normalizes the key variables to obtain auxiliary variables and leading variables, and then performs dynamic modeling to transform the auxiliary variables and leading variables with only static characteristics of time series into ALIESN input variables with dynamic characteristics; then the ALIESN input variables For the first n data of , use the ALIESN ridge regression offline learning algorithm to conduct offline training and learning, and obtain the output weight of the neural network offline training; finally, use the above value as the initial weight of online learning, and use the n+1th ALIESN input variable The online output weight training starts from the data, and the final network output weight and predicted output variable are obtained. The learning algorithm of this method is simple, fast, low cost, less training data, and strong dynamic nonlinear system approximation ability, etc., the speed of network output weights is fast, the prediction accuracy of butane content is improved, and it meets the requirements of real-time learning in the process of fine chemical industry.

Description

technical field [0001] The invention relates to a soft sensor modeling method for a debutanizer tower based on an augmented leakage integral echo state network (ALIESN) fast online learning algorithm, belonging to the field of soft sensor, specifically a debutanizer based on an ALIESN online learning algorithm Soft-sensing modeling method for alkane towers. Background technique [0002] In the chemical process, real-time online monitoring of important variables plays an extremely important role in product quality control. However, these variables are often in the environment of high temperature, strong nonlinearity, strong radiation, and rapid changes in the chemical process, and cannot be directly, timely, and accurately measured and collected. Therefore, the application of soft sensor technology was born. [0003] In the past ten years, soft sensor technology has developed rapidly, especially the soft sensor modeling method based on neural network, which has been applied ...

AI Technical Summary

Problems solved by technology

However, in the existing chemical process based on neural network soft sensor modeling methods, there are generally complex algorithms, large data sets are required for learning and training, the sampling time of input and output variables is long, the learning time is long, and the learning cost is high. Problems such as fast and real-time attainment of high learning accuracy have caused the monitoring of important variables to fail to meet the relevant accuracy requirements of fine chemicals

Images