The Wood Drying Kiln Neural Networks Control And Simulation

This paper aims at designing a kind of controller for complicated control objects on wood drying and other related things, then on the basis of which, it sets up neural net model for miniature JBGZ-l.8 wood drying kiln. Single nerve cell adaptive PID control is used for algorithm design and model test in the paper, meanwhile, the designing of software and hardware and the relevant system implement are also researched deeply. System differentiating and analysis of neural net is to study input and output data of system directly, in which the aim of study is to make error function smallest, then to sum up the relation included between input data and output data. It is not algorithm form, neural net itself is analysis model, whose adjustable parameter is reflected by connection weight in the net. The method of the work need not set up analysis format based on actual system mathematical model, so the process of system modeling before analysis can be omitted. In fact, as analysis model of actual system, neural net is also a physics implement for actual system and can be used in on-line control. The amount of calculation is very big in the course of adaptive control of neural net, because multi-layer forward direction net uses type S action function. Of quick process control, since there is short of relevant ufility type neural net hardware support, it is difficult to make adaptive control on multi-layer neural net in use during real-time on-line control. To adapt quick process neural control, the paper uses a kind of single adaptive nerve cell, which can make use of the advantage of neural net and also adapt quick process real-time control. The single nerve cell adaptive PID controller is varied coefficient and multiple proportional differential integral, meanwhile, the study algorithm is adaptive, so it is nonlinear. The non- linear approach capability and self-learning capability of neural net are very strong. As the structure of it is single, and the stability of system and astringency of study are very good, so no model direct adaptive control of neural net has stronger develop potential.