An Improved Invasive Weed Optimization Algorithm Applied To Thermal Process Identification

This paper analyzes the dominant factors of main steam temperature changing process and its model is established.Some disturbance experiments in the traditional methods may be dangerous to the actual working conditions and are difficult to be realized.With the constant improvement of automation of the thermal power plant,the DCS system has stored a large number of rich historical data.The utility of intelligent swarm algorithms to mine these data to establish the object model is a new trend.In view of the large inertia,large delay and non-linearity of thermal process in power plant,the least-squares and other traditional identification methods are not accurate enough and the parameters estimation is greatly disturbed by noise in some circumstances.This paper proposes a improved Invasive Weeds Optimization(IWO)algorithm which has powerful global search ability for thermal process identification.The main contents and conclusions of this paper include:1.The source,principles and implementations of IWO algorithm are introduced.The Differential Evolution algorithm is combined with IWO to yield DE-IWO algorithm.The performances of the two algorithms are compared using benchmark such as Griewank and transfer functions.The experimental results show that DE-IWO algorithm has advancement both in convergence and accuracy.2.After the normalization of the standard deviation of the historical data derived from the field DCS system,the boxplot is used to qualitatively analyze the dominant factors affecting the main steam temperature.Nine single-in-single-out models are established for model identification and verification of time-independent data.Meanwhile,the results show that the secondary chilled water flow and the water-coal ratio are the two most important factors of temperature.3.The principal component analysis(PCA)is used to quantitatively analyze the influence of each variable in the field historical data on the main steam temperature.And the main component contains abundant characteristics of the original information.So it is considered as the model input to establish SISO and MISO model.DE-IWO algorithm and least-squares algorithms are used to estimate the model parameters and the average absolute output errors of the model are compared.The results show that the former has higher recognition accuracy and the corresponding average absolute error is smaller.4.Based on the result of principal component analysis,a linear regression model and a nonlinear model are respectively established to describe the thermal process.The experimental results show that the linear regression model has a large error in verification of data in uncorrelated time.And the latter achieves a good identification result.However,it is not as good as the accuracy of linear model based on DE-IWO algorithm.In summary,principal component analysis and DE-IWO algorithm applied to the main steam temperature system identification yield the best outcome.