The technique of computer simulation has been an important tool now in analyzing andresearching all kinds of systems, especially complex systems. Because it's economy, trusty,easy to be implemented and can be used repeatedly, it has been an efficient means in analyzing,designing, experimenting and evaluating actual systems, and has extensive applications in thefields of science and engineering.With the continually widening of the application fields of computer simulation, theproblems people encountered become more and more complex. There are two kinds ofproblems at present. One is that the objects to be modeled are complex, and have many kinds ofindetermination and the non-linearity character which is hard to describe. The other is thatpeople's requirements to system modeling have been higher and higher, so it is necessary toimprove the describing ability of system models and the flexibility, the generality and theintelligence level of modeling methods.The Artificial Neural Networks(ANNs) is a theory and technic developing rapidly in thefield of computer intelligence. Because it is not necessary to create accurate physical models ormathematic models before used to solve problems, ANNs is applied to computer simulation.The feed forward multi-layer ANNs can approximate any continual function. and its ability ofdescribing real systems is equal to a differential equation or differential equations, so it is asystem modeling method with strong applicabilities. At the same time, the method has theability of getting knowledge by learning from surroundings, so it has good adaptabilities tosystem simulation.In this paper, aiming at some practical problems in dynamic system simulation, the modelsand algorithms of ANNs adapted for the modeling of dynamic systems are studied. A processneural networks with time-varying inputs and outputs and a discrete process neural networksare created. Their characters are analyzed and proved, and their learning algorithms are given.Then, the applications in the dynamic system simulation of time-delay neural networks,Hopfield neural networks, partial recurrent neural networks, the process neural networks withtime-varying inputs and outputs, the discrete process neural networks are shown.The contents of this paper are as follows:The 1st chapter is the introduction. The developments and applications of systemsimulation are expounded. The methods of system simulation modeling now exists aresummarized and their disadvantages are analyzed. Then the basic concepts or models of systemsimulation, ANNs, process neural networks are expound. The items and the contents of thisresearch are proposed.The 2nd chapter is the summarization of the basic theories, characters, methods of systemsimulation and ANNs. The system simulation based on ANNs are expounded and the feasibilityof system simulation with ANNs. The three ways of modeling with ANNs used in systemsimulation are summarized. And a general procedure of simulating with ANNs is proposed.The 3rd chapter studies the simulation models, the adaptabilities of simulation models, themethods the models are used in system simulation, and the algorithms of the time-delay neuralnetworks, entire recurrent neural networks, and partial recurrent neural networks.The 4th chapter studies the process neural networks with time-varying inputs and outputsand the discrete process neural networks. To the the process neural networks with time-varyinginputs and outputs, its topological structure is given, its continuity, approximation, andcalculating capability is proved, and a learning algorithm combining base functions expandingand grads descending is proposed. To the discrete process neural networks, the topologicalstructure is given, too. And a learning algorithm combining grads descending and datacalculating is proposed. The concrete implementing procedures of the two algorithms are alsoshown in this chapter.The 5th chapter gives an application of process neural networks in dynamic systemsimulation. Aiming at the Parallel connection model and the Multiple-series connection model,two neural networks models of system simulation are created, and their algorithms are pointedout. At last, the simulating results are shown. |