Research On The Portable Block Calibrator Non-linear Control System

Temperature block calibrator system is a typical pure delay, time varying and great inertia non-linear system. Researching control method of the temperature block plays an important part in the process of designing temperature block. Since radial basis function can approach any continuous function in arbitrary precision that the RBF neural network plays an important role within those fields for example, system identification, non-linear system control, pattern classification etc. In order to improve effects of the temperature block control system, build a discrete model about the temperature block; Present a new method about RBF neural network, and simulate some indexes about this new method; Design a calibration software based on the temperature block.Firstly, on account of operating principle of the temperature block, build its discrete model by mechanism analysis method. In view of the previous model, using correlation ratio identification method correcting gain ratio and attenuation coefficient and identifying temperature block’s coefficient of delay. Through the above steps, get a math model of the temperature block system.Secondly, research the latest RBF neural network study methods (RAN method and GIRAN method).Because of those disadvantages that the RAN method is lack of delete rule and has slow speed in network convergence and GIRAN method needs probability distribution functions of the sample input dates in its novelty rule, put forward a RBF neural network study method (SRAN-Simple Resource Allocating Network-method). It contains novelty rule and delete rule, update algorithm about weights of the output and the parameters of the hidden layer neural. In SRAN study method, the number of the hidden layer neural is regulated by the novelty rule and delete rule, the weights of the output is updated by the steepest descent method, the parameters of the neural are not updated ever since their initialization.Then, through computer simulation, compare the SRAN and the LMS method in RBF neural network structure, approach velocity, perfonnances of control and difficulty of the method. After analysis those simulation results, it come to a conclusion that the RBF neural network based on the SRAN method has character such as structure simple, higher approach speed, good performance in control temperature and smaller amount of calculation.Finally, on the basis of the demands of the calibration guideline of the temperature block, design application software for temperature block system using C++Builder2010development environment. Test the function of the software modules by five cases, and the test results indicate that this software has characters such as easy operation、friendly interface and multiple functions.