垃圾协同处置下基于ELM的MISO Hammerstein-Wiener分解炉温度预测控制

ELM Based Temperature Prediction Control for MISO Hammerstein-Wiener Decomposition Furnace under Collaborative Garbage Disposal

  • 摘要: 针对传统的线性模型不足以描述分解炉复杂系统的问题,结合垃圾协同处置的背景,研究了一种基于极限学习机(extreme learning machine,ELM)的MISO Hammerstein-Wiener (multiple-input single-output Hammerstein-Wiener)模型分解炉温度建模及预测控制方法,用以实现分解炉温度的稳定控制。模型以喂煤量和垃圾衍生燃料流量(refuse derived fuel,RDF)为输入、分解炉温度为输出,并且采用ELM拟合非线性环节,ARMAX (autoregressive moving average with extra input)模型来描述动态线性环节,递推最小二乘法辨识出模型混合参数,奇异值分解得到模型的参数估计。分解炉控制方法采用两步法预测控制。首先,建立非线性环节逆模型;其次,采用广义预测控制算法得到中间变量;最后,中间变量经过非线性环节逆模型输出得到模型的控制量。仿真实验表明,ELM的引入提高了模型的拟合精度。与传统的预测控制相比,所提的控制方法稳定性更强、跟随性更好。

     

    Abstract: To address the limitation that traditional linear models are insufficient to describe the complex system of the decomposition furnace, in collaborative garbage disposal, we propose a multiple-input single-output (MISO) Hammerstein-Wiener model for decomposition furnace temperature modeling and predictive control based on an extreme learning machine (ELM). This approach aims to achieve stable temperaturecontrol of the decomposition furnace. The model uses coal feeding rate and refuse-derived fuel (RDF) as inputs, decomposition furnace temperature as the output, and applies ELM to capturing nonlinear relationships while employing an autoregressive moving average with exogenous input (ARMAX) model to describe dynamic linear relationships. We identify the mixed parameters of the model using the recursive least squares method and estimate the model parameters through singular value decomposition. The control method for the decomposition furnace follows a two-step predictive control approach. First, we establish a nonlinear inverse model. Then, we map these intermediate variables through a nonlinear process to determine the control variables of the model. Simulation experiments demonstrate that incorporating ELM improves the fitting accuracy of the model. The proposed approach exhibits greater stability and improved tracking performance compared to traditional predictive control methods.

     

/

返回文章
返回