Study On Intelligent Modeling Methods Of Omethoate Synthesis Process

Omethoate synthesis is a typical batch process, with its temperature plant having characteristics of non-linear,time-delay and uncertainty, etc, it is difficult to use conventional method to identify its structure. In recent years, intelligent theory methods represented by neural networks, fuzzy logic theory and genetic algorithms, and the integrated approaches of them, provide an effective way to build up models of the complex object with the above-mentioned characteristics. In this paper, referencing to the self-learning and adaptability capacity of neural networks, the ability of fuzzy logic to integrate expertise knowledge, and the optimize characteristics of genetic algorithm, with the complex nonlinear systems as the target object, we do some jobs in the following aspects:First, we make an overview of the research of system identification; analyze the characteristics and the application range of the various identification methods. And then discuss the role and application prospects which played by the intelligent identification methods in the modeling of the complex non-linear object.Then, on the base of identification of the temperature object using basic BP network, we found that there is large errors in the initial stages of response, which may be due to the neural network is easily trapped into local minima. Therefore genetic algorithm with global search feature is brought in to optimize the weight of the neural net, and is applied into the modeling of the reactor temperature. Compared the simulation results with the basic BP net, we found that although the stable phase in the temperature fit better, it does no good to the large errors in the initial stage. Studies in other literatures proved that with same training structure and capacity, the performance of fuzzy neural network is much better than traditional BP network, mainly reflected in decline in the average error and maximum error, decreases in the training time. Therefore, traditional fuzzy neural network is introduced here to identify the temperature model, and the result achieved is pretty good, however the large error of the initial stage is still not greatly improved.The reason probably is the temperature during the batch production process of Omethoate synthesis reflects the dynamic behavior of the system, while traditional neural networks and fuzzy neural networks are all static neural network, so just doing the static identification is not enough, through which the characteristic of the system cannot be fully reflected. Therefore, in this paper, we add a first order time delay link to the basic BP neural network and traditional fuzzy neural network so as to constitute a dynamic recurrent BP neural network with a delay and deviation units and a dynamic recurrent fuzzy neural network, and apply them to the modeling of the temperature. The results of the simulation show that dynamic recurrent network can reach higher accuracy and better generalization ability than static neural network, since they take full advantage of the current data and history data.