Study On The Control And Its Optimization For Hotspots And Thermal Uniformity During Microwave Heating Process

Faced with ever-increasing energy consumption and seriously environmental pollution,energy conservation has become one of China's basic national policies.It's necessary to change the existing coal and oil-based fossil fuel heating methods,and clean energy should be widely used immediately.Microwave energy is a kind of clean energy that can be generated by electric energy,and it can be used in a number of industrial heat treatments.Compared with traditional heating methods,microwave energy reveals its excellent energy and time-saving features in many industrial fields.So,more and more researchers and companies pay attention to microwave energy.However,the application of microwave energy needs to solve two major problems: thermal runaway and thermal nonuniformity.Thermal runaway can cause damage to the heating material.Even worse,it may lead to heating chamber explosion.Thermal nonuniformity will influence the final heating result,which may give rise to large temperature difference at heating material different locations.Based on industrial microwave heating characteristics,the priori knowledge of microwave power distribution and material dielectric properties,this paper researches microwave heating process temperature field nonuniform distribution,process identification,hot spot temperature control and multi-objective optimization issues.The aim of this work is to improve temperature field nonuniform distribution and achieve hot spot temperature under control.Heating material temperature field uniformity can be improved by adding microwave input feeds.But the existing researches mainly analyze how to choose the best feed locations.Through adjusting microwave sources input power and phase to achieving heating process temperature field uniformity is little researched.The process of heating material heated by two input sources is a typical issue of heating material heated by multiple input sources.This paper analyzes how to achieve temperature field uniformity under two input sources.Desirable power distribution at any position of the heating material can be acquired by adjusting microwave sources input powers and phases.Then,heating material power distribution uniformity at time dimension can be obtained.This paper proposes a Cuckoo Search combined with sliding mode neural network algorithm to control micorwave source input phase and power,based on which a uniform temperature field rising process can be achieved.In actual applications,as the emergence of unknown distributance,microwave source input power and phase can't accurately equal control algorithm calculated values.So,a simulation with input power randomly jumping around 100 ±40% of the calculated value,phase difference jumping around 100 ±20% of the calculated value and sampling temperature jumping around-0.3-0.3 K of the actual value is made.Simulation results show that temperature uniformity can still be obtained by using Cuckoo Search combined with sliding mode neural network algorithm.We also make a comparison with genetic algorithm,and comparison results show Cuckoo Search can get better input results at shorter calculation time.Multilayer forward static neural network is a common identification method in microwave heating process.But microwave heating is a time-varying process,and a well trained model can't be applied at different applying environment.So,to accurately identify microwave heating process,real time data should be sampled and model parameters should be online updated,which lead to static model inaccuracy.To achieve microwave heating and drying process identification,this paper proposes a novel neural structure,named recurrent self-evolving fuzzy quantum neural network(RSFQNN).RSFQNN can acquire accurate identification result by online updating model parameters and structure through sampling real time process data.Current input power and former state information are used to predict state information at next moment,and identification errors limit in a range of-1~1K in microwave heating and drying process.In applications of dynamic system identification and chaotic series prediction,compared with existing neural networks: recurrent self-evolving fuzzy neural network with local feedbacks(RSEFNN-LF)and interactively recurrent self-evolving fuzzy neural network based on functional-link(IRSFNN-Fu L),under a same training epoch,RSFQNN has the best identification ability.During microwave heating process,under known and unknown priori knowledge,this paper designs two different controllers.Under known priori knowledge,Lambert's law combined with real-time temperature information algorithm is proposed to calculate microwave power distribution.Simulation results show Lambert's law combined with real-time temperature information algorithm can calculate a more accurate result than Lambert's law.Then based on Lambert's law combined with real-time temperature information algorithm,under process parameters approximate known,a model predictive control algorithm is proposed,which can make material temperature well follow reference trajectory.But the general situation is that there is no useful priori knowledge.Microwave power distribution is nonuniform in the microwave reaction cavity,and this time varying system parameters are basically unknown.The commonly used control algorithms contain proportion integration differentiation algorithm,linear tracking algorithm,experience formula,adaptive fuzzy neural network controller and so on.But those algorithms also have many shortcomings,such as,large tracking error,need system parameters,poor generalization ability and need large numbers of training.So,an algorithm with a wide application range,easily determined parameters and high control accuracy is badly in need.This paper proposes a sliding mode radial basis function neural network algorithm to deal with the control problem under single microwave input and microwave combined with convective heat transfer.We also give a fixed learning factor and an adaptive learning factor control law for applications can be performed under same conditions repeatedly and not.For single microwave input,this sliding mode radial basis function neural network algorithm can achieve satisfactory control results under both simulation and actual applications.In actual application,temperature tracking error can gradually converge in the range of 1K with sliding mode radial basis function neural network algorithm online learning.In microwave combined air heat convection multiple inputs simulation,this algorithm also can get suitable microwave power and air heat convection inputs,and sampling temperatures can well follow the reference trajectory.In microwave thermal applications,there exist many control objectives,such as,temperature,energy utilization,moisture and so on.To optimize those multiple objectives and calculate the optimal input power,based on microwave heating time-varying characteristic,this paper proposes a multiple objectives optimization algorithm,named recurrent self-evolving fuzzy neural network based multiobjectives predictive optimization control algorithm.In Red Maple drying simulation experiment,we choose temperature and moisture as the control objectives.By using this optimization algorithm,we can achieve temperature and moisture optimization under control.In actual experiment,we choose lignite as control objective.Through applying this optimization algorithm,lignite temperature tracking error can limit in the range of 2K.