Research On Industrial Furnace Temperature Control Based On Neural Network Predictive Control

Industrial furnace is a very important equipment in industrial production,and its temperature control is particularly important in the production process.However,the industrial furnace temperature is a nonlinear,coupled and delay system strongly.It is difficult for that the system is identified its accurate model to guarantee the precise temperature.The predictive control method does not rely on an accurate model.It adopts rolling optimization and uses rolling limited time period optimization instead of static optimization,which is more on line with actual production conditions.Neural network is a model constructed based on historical data,which can accurately approximate uncertain nonlinear models.Neural network prediction models can be well applied in predictive control as their prediction models.Based on the above situation,the main research work in this thesis is as follows:(1)The neural network has a strong ability to approximate non-linearity.However,during the training process,if the traditional back propagation(BP)algorithmis used for training,it will be affected by the initial value of the network weight and threshold during the training process,and the training may fall into the local extreme value.Therefore,this paper uses genetic algorithm(GA)to initialize network weights and thresholds to improve the model’s convergence and accuracy.(2)According to the characteristics of Newton-Raphson(N-R)that it is easy to fall into local extremes,this article uses the fusion particle swarm optimization(PSO)and N-R algorithm to implement multi-step scrolling predictive control.Firstly,using the PSO algorithm to perform preliminary optimization as the initial value of the N-R to iteratively calculate the control amount in the step size.Secondly,according to the basic principles of PSO,use the control input in this step as the initial value of the global optimal position for the next step of PSO.This not only solves the problem of N-R depending on the initial value,but also reduces the cost of PSO operation time,and improves accuracy and speed.(3)In the actual production process,because the system has different constraints,it increases the difficulty of control.This paper uses the interior point method to convert constraints,and constructs penalty terms,then changes the original performance index function,next combines PSO and N-R algorithms to achieve rolling optimization,which is more in line with production conditions.(4)The fusion algorithm is applied to practice,and the data is generated with the help of an industrial furnace model in the form of numerical simulation.Then we treat the model as a black box.We assume the model is unknown,and use a neural network to approximate the mode.Using digital simulation simulation to show the advantages of the improved fusion algorithm of the model proposed in this paper.Finally,combined with the actual industrial furnace project requirements,we design the human-computer interaction interface by using the MATLAB GUI development kit.Through the interface,that can achieve easy-to-operate human-computer interaction.