【摘要】：針對化工過程強非線性和多工況的特性,提出了一種基于BP神經(jīng)網(wǎng)絡(luò)(BPNN)有效非線性融合多關(guān)聯(lián)向量機(MRVM)的建模方法.首先選擇不同的核函數(shù),采用樣本數(shù)據(jù)建立單一RVM子模型;然后利用BPNN的強非線性擬合能力,對各子模型的預(yù)測信息進(jìn)行非線性融合,并采用人工魚群算法(AFSA)對BPNN的初始權(quán)重和閾值進(jìn)行優(yōu)化;最終建立MRVM非線性融合模型.將該建模方法應(yīng)用于甲醇制烯烴生產(chǎn)過程(MTO)乙烯收率預(yù)測研究中,研究結(jié)果表明:與單一RVM模型和最優(yōu)加權(quán)組合模型相比,基于MRVM的非線性融合模型具有更佳的預(yù)測精度.
[Abstract]:A modeling method based on BP neural network (BPNN) effective nonlinear fusion of multi-correlation vector machine (MRVM) is proposed for the characteristics of strong nonlinearity and multi-working conditions in chemical process. Firstly, different kernel functions are selected, and a single RVM submodel is built by using sample data. Then, the prediction information of each sub-model is fused by using the strong nonlinear fitting ability of BPNN. The initial weight and threshold of BPNN are optimized by artificial fish swarm algorithm (AFSA), and the nonlinear fusion model of MRVM is established. The modeling method is applied to the prediction of (MTO) ethylene yield in methanol olefin production process. The results show that the nonlinear fusion model based on MRVM has better prediction accuracy than single RVM model and optimal weighted combination model.
【作者單位】： 浙江工業(yè)大學(xué)化學(xué)工程學(xué)院;浙江省生物燃料利用技術(shù)研究重點實驗室;
【基金】：國家自然科學(xué)基金資助項目(21676251)
【分類號】：TP18;TQ018
，

本文編號：2170924