基于模糊神经网络的模型预测硅单晶直径控制  

Model prediction of monocrystalline silicon diameter control based on fuzzy neural network

作　　者：彭鑫 高德东 王珊 徐圣哲 PENG Xin;GAO Dedong;WANG Shan;XU Shengzhe(College of Mechanical Engineering,Xining 810016,China)

机构地区：[1]青海大学机械工程学院,青海西宁810016

出　　处：《青海大学学报》2023年第4期92-99,共8页Journal of Qinghai University

基　　金：青海省科学技术厅项目(2022-GX-C07);西宁市科学技术局项目(2021-Y-01)。

摘　　要：为了生产大尺寸、高质量、低能耗的直拉单晶硅棒,文中通过在直拉硅单晶生产车间采集的大量数据,通过模糊神经网络(Fuzzy Neural Networks,FNN)和模型预测控制(Model predic-tive control,MPC)进行建模,得到了等径阶段的直径预测模型和直径控制模型.直径预测模型测试结果表明:平均相对误差仅为0.0287%,预测精度极高.通过仿真并与常规PID控制进行对比分析得出,MPC比常规PID控制的调节速度快53.66%,并且MPC的控制过程非常稳定,其超调量基本为0;由加热器功率调控变化过程可知,MPC减少了调节过程的能耗,并提高了热场稳定性及单晶硅棒质量.通过预测模型建立的直径控制模型能提高控制精度并促进硅单晶高质量生产.In order to produce large-sized,high-quality and low-energy consumption Czochralski monocrystalline silicon rods,a large amount of data collected in the production workshop are used to make models using Fuzzy Neural Networks(FNN)and Model Predictive Control(MPC)to obtain diameter prediction and control models for equal diameter stage.The test results of diameter prediction model indicate that the average relative error is only 0.0287%,indicating extremely high prediction accuracy.According to the simulation and comparative analysis with conventional PID control,it is found that the adjustment speed of MPC is 53.66%faster than that of conventional PID control,and the control process of MPC is stable with an overshoot value of 0.From the process of regulating the power of the heater,it can be seen that MPC reduces energy consumption and improves thermal field stability and the quality of monocrystalline silicon rods.The control accuracy and high-quality production of monocrystalline silicon can be promoted by the diameter control model established through predictive models.

关 键 词：直拉硅单晶 模糊神经网络 直径预测 模型预测控制 直径控制 

分 类 号：TP273.1[自动化与计算机技术—检测技术与自动化装置]