Study On Identification Of Wood Drying Model Based On Neural Networks

With the deficient of forest resources in our country, how to use resources reasonably and effectively become an advancing front subject which has been concerned by the researcher of wood science. Wood drying is the important step during the process of the wood processing. It is urgent to improve the wood drying technology and build up automatic drying control system.The process of the wood drying is a complicated system which is a nonlinear multivariable and coupling system, it is time-variable and uncertain. It's difficult to build up an exact model using classical system identification method, but the system identification based on Neural Networks can directly learn the input and output data of the system and make the object function to get minimal value, then generalize the relation which hide in the input and output data of the system, the relation is the model of system. In this thesis, the wood drying model are built based on Neural Networks.In the thesis, the temperature-humidity control model and dry standard model are built using Time Delay Neural Network (TDNN) and Dynamical Recurrent Neural Network (DRNN) which is suitable for identification and control. The former is the model between the control signal and the temperature-humidity, this model will provide rule for control the temperature and humidity in the drying kiln; the latter is the model between the temperature-humidity and moisture content of timber, this model realize the mathematic model for dry standard. Because the inverse model of system is often used in many control methods, in the thesis, the temperature-humidity control inverse model and dry standard inverse model are built too.The structures of the Neural Networks and the learning algorithms were given, in addition, we verify the accuracy of the model, the simulation results show that the model is effective and reliable; at the same time, we compare the results of two kinds of Neural Networks, and make the conclusion that identification effectiveness based on TDNN is better than DRNN. In fact, as the identification model of system, Neural Networks is the physics implementation for actual system, and can be used in on-line control. The methods of identification can be popularized to other similar systems.