微波加热过程热点与热均匀性控制与优化研究

发布时间：2018-05-11 02:04

本文选题：微波加热 + 温度场均匀性　；参考：《重庆大学》2016年博士论文

【摘要】：面对日益增加的能量消耗以及严重的环境污染,节能减排已成为我国的基本国策,改变现有的以煤、石油为主的化石燃料加热方式,使用清洁能源刻不容缓。微波能是一种清洁能源,可以通过使用电能的方式产生,可用于众多工业热处理领域。相较于传统加热方式,微波能在众多工业领域显露出卓越的节能省时特性,受到越来越多研究人员与公司的重视。但微波能应用需要解决两大问题:热失控与热不均,热失控会导致加热媒质损坏,更严重的情况下会导致加热腔体爆炸,而热不均会影响最终加热效果,导致媒质不同位置温度差异很大。本文的研究工作主要针对工业微波加热特点,基于微波加热过程中微波功率分布和媒质介电特性等先验性知识,分析微波加热过程中温度场非均匀性、媒质温度辨识、热点温度控制以及多目标优化问题,改善微波加热媒质温度场非均匀性,实现热点温度控制。增加微波输入馈口,可以改善加热媒质温度场均匀性,但现有的研究多关注于微波源在反应腔体外壁馈口位置优化选择,对加热过程中微波源输入功率及相位的主动控制实现温度场均匀性较少研究。加热媒质在两个输入源下温升过程是多输入源的一种典型情况,本文对两输入微波源作用下的加热媒质温度场均匀性实现进行了分析。通过改变微波源入出功率和相位,可以在媒质的任意位置得到希望的功率分布,以此可以得到在时间维度上均匀的微波功率分布,基于此设计了布谷鸟搜索结合滑模神经网络的控制算法,对两个微波输入源的功率和相位进行实时控制,实现均匀的温升过程。同时,考虑实际情况下,微波源实际输入功率和相位与控制算法计算值存在误差,对微波源输入功率在计算值100±40%范围内随机变化、相位差在计算值100±20%范围内随机变化以及温度传感器存在-0.3-0.3 K范围内随机变化误差的情况进行了仿真计算,仿真结果表明:布谷鸟搜索结合滑模神经网络算法可以得到媒质温度场均匀的温升过程。通过与遗传算法比较分析可知,布谷鸟搜索算法可以在更短时间内得到更优的输入功率值。在微波加热过程系统辨识研究中,一般采用多层前向静态神经网络。但由于微波加热是时变系统,且一种训练好的模型在媒质加热环境发生变化的情况下难以应用,因此需要实时采样过程数据,导致静态模型难以准确描述微波加热过程。本文提出了一种递归自进化模糊量子神经网络模型,用以对微波常规加热与干燥过程进行系统辨识,该模型通过实时采样微波加热过程数据,实现神经网络参数及结构更新,以得到最佳的辨识结果。递归自进化模糊量子神经网络将微波输入功率及先前状态信息作为输入层用以预测下一时刻的状态数据,在温度辨识中,辨识误差可以控制在1K以内。将递归自进化模糊量子神经网络应用于在动态系统辨识与混沌模型预测,通过与现有的耦合局部反馈递归自进化模糊神经网络与泛函连接交互式递归自进化模糊神经网络比较分析,可以得出该模型在相同的训练周期下,具有更优的辨识能力。在微波加热过程中,在先验知识可用与不可用情况下,本文设计了两种不同的控制算法。在先验知识可用的情况下,提出Lambert定律结合实时温度信息算法计算微波功率分布,仿真结果表明该算法可以得到比Lambert定律准确性更高的计算结果。基于此算法,在过程参数近似已知情况下,进一步提出了模型预测控制算法,实现媒质温度准确跟踪预期轨迹。但更普遍的情况是微波加热过程中无可用先验知识,在反应腔体内部,微波功率通常是非均匀性分布,并且该时变系统过程参数基本上是未知的。现有的控制方法有比例积分微分(PID)控制、线性化跟踪控制、经验公式、自适应神经网络模糊控制器等,但这些算法具有如:误差大、需要系统参数、泛化能力差、需要大量训练等缺点,因此需要研究一种具有更广应用范围、参数容易确定、控制精度较高的控制算法。本文提出滑模径向基神经网络控制算法对单微波输入和微波结合空气热对流输入情况下的控制输入设计问题进行了分析。针对加热过程在相同实验条件下,可以多次重复与难以重复的情况,提出了相应的固定学习速率控制算法及自适应学习速率控制算法。在单微波输入中,该算法在仿真与实际实验中,均获得良好的控制效果,在实际应用中,通过神经网络的学习过程,温度跟踪误差可以逐渐收敛到1K以内。在微波结合空气热对流多变量仿真实验中,该算法可以计算得出合适的微波功率与热对流控制输入值,保证媒质温度准确跟踪预设轨迹。在微波加热过程中,针对控制目标,如:温度、能量利用率、含水率等过程变量,为实现多目标优化控制,确定最优输入功率,本文研究了一种针对微波干燥过程的多目标预测优化算法。根据微波干燥过程的时变特性,提出了基于递归自进化模糊神经网络的多目标预测优化控制算法。在红衫木干燥仿真实验中,选取温度和含水率作为控制对象,通过应用递归自进化模糊神经网络多目标预测优化控制算法,可以实现对温度与含水率的优化控制。在实际实验中,将褐煤作为干燥媒质,选取温度作为被控对象,该优化算法可以将褐煤干燥过程温度误差控制在2K以内。
[Abstract]:In the face of increasing energy consumption and serious environmental pollution, energy saving and emission reduction has become the basic national policy of our country. It is very urgent to change the existing heating mode of fossil fuel based on coal and oil and use clean energy. Microwave energy is a kind of clean energy. It can be produced by means of electric energy and can be used in many industrial heat treatment. Compared to the traditional heating mode, microwave can show remarkable energy saving and time-saving characteristics in many industrial fields. More and more researchers and companies pay more attention to it. But the application of microwave energy needs to solve two major problems: heat out of control and thermal inhomogeneous, thermal runaway can cause heating medium to damage, more serious cases will cause heating chamber explosion. The research work of this paper mainly focuses on the characteristics of the industrial microwave heating. Based on the prior knowledge of the microwave power distribution and medium dielectric properties, this paper analyzes the non uniformity of the temperature field and the medium temperature identification in the process of microwave heating. Hot temperature control and multi-objective optimization problem can improve the non-uniformity of the temperature field of microwave heating medium and realize the hot temperature control. Increasing the microwave input feed inlet can improve the temperature field uniformity of the heated medium. However, the existing research pays much attention to the optimum selection of the feed position of the microwave source in the outer wall of the reaction cavity and the microwave source in the heating process. The active control of the input power and phase is less homogeneous in the temperature field. The temperature rise process of the heated medium under two input sources is a typical case of the multi input source. In this paper, the uniformity of the temperature field of the heated medium under the action of the two input microwave source is analyzed. The power and phase of the microwave source can be changed by changing the power and phase of the microwave source. In any position of the medium, the desired power distribution is obtained to get the uniform microwave power distribution on the time dimension. Based on this, the control algorithm of the cuckoo search and the sliding mode neural network is designed. The power and phase of the two microwave input sources are controlled in real time to realize the uniform temperature rise process. At the same time, the actual situation is considered. The actual input power and phase of the microwave source have error with the calculated value of the control algorithm. The input power of the microwave source is randomly changed in the range of 100 + 40%, the phase difference is randomly changed within the range of 100 + 20% and the temperature sensor has random variation in the range of -0.3-0.3 K. Simulation and simulation are carried out. The results show that the cuckoo search combined with the sliding mode neural network algorithm can get the uniform temperature rise process of the medium temperature field. By comparing with the genetic algorithm, it can be seen that the cuckoo search algorithm can get better input power value in a shorter time. In the study of system identification of microwave heating process, the multi-layer forward static God is generally adopted. But because the microwave heating is a time-varying system, and a good training model is difficult to be applied to the medium heating environment, so it is necessary to sample the process data in real time, which causes the static model to not accurately describe the microwave heating process. In this paper, a recursive self evolution fuzzy neural network model is proposed. The conventional microwave heating and drying process is systematically identified. By sampling the data of the microwave heating process in real time, the model realizes the neural network parameters and the updating of the structure to obtain the best identification results. The recursive self Evolving Fuzzy quantum neural network uses the microwave input power and the previous state information as the input layer to predict the next time. In the temperature identification, the identification error can be controlled within 1K. The recursive self evolving fuzzy neural network is applied to the dynamic system identification and the chaotic model prediction, and is compared with the existing coupled local feedback recursive self evolving fuzzy neural network and functional connection interactive recursive fuzzy neural network. It can be concluded that the model has better identification ability under the same training period. In the microwave heating process, two different control algorithms are designed in the case of prior knowledge availability and unavailability. In the case of prior knowledge, the Lambert law and the real-time temperature information algorithm are proposed to calculate the microwave power points. The simulation results show that the algorithm can obtain more accurate results than the Lambert law. Based on this algorithm, the model predictive control algorithm is further proposed to achieve accurate tracking of the expected trajectory under the approximate known process parameters, but the more common situation is that no prior knowledge is available in the microwave heating process. In the reaction chamber, the microwave power is usually nonuniform, and the parameters of the time-varying system are basically unknown. The existing control methods are proportional integral differential (PID) control, linearized tracking control, empirical formula, adaptive neural network fuzzy controller and so on, but these algorithms have such advantages as large error and need system parameters. This paper proposes a sliding mode radial basis neural network control algorithm for the control input design problem of single microwave input and microwave combined with air heat convection input. The corresponding fixed learning rate control algorithm and adaptive learning rate control algorithm are proposed for the heating process, which can be repeated and difficult to repeat under the same experimental conditions. In the single microwave input, the algorithm has good control effect in both simulation and practical experiments. In practical application, the neural network is used in the actual application. In the learning process of the collaterals, the temperature tracking error can be gradually converged to less than 1K. In the multi variable simulation experiment of microwave combined with air heat convection, the algorithm can calculate the appropriate input values of microwave power and heat convection control, and ensure that the temperature of the medium is accurately tracked by the preset trajectory. In the process of micro wave heating, the control target, such as temperature, can be used in the process of micro wave heating. In order to achieve multi-objective optimization control and determine the optimal input power, a multi-objective optimization algorithm for microwave drying process is studied in this paper. Based on the time-varying characteristics of the microwave drying process, a multi target predictive optimal control algorithm based on the recursive self evolution model paste neural network is proposed. In the drying simulation experiment of red shirt and wood, the temperature and water content are selected as the control object. The optimization control of temperature and water content can be realized by using the recursive self evolving fuzzy neural network multi target predictive optimization control algorithm. In the actual experiment, the lignite is used as the drying medium and the temperature is selected as the controlled object, and the optimization algorithm can be obtained. The temperature error of lignite drying process is controlled within 2K.

【学位授予单位】：重庆大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN015;TP183

【参考文献】