Research On Pattern Recognition In Dynamic Systems Based On Process Neural Networks

The problem of pattern recognition in dynamic systems is very common in the field of scientific research.Against the problem of time-varying signal in artificial neural network, domestic and foreign scholars in recent years have made a number of discriminant models based on traditional BP neurons, but when using these models in resolving the process inputs and time sequence-dependent issues , they usually made the time into space (time series),and then link to external delay between input and output to achieve the dynamic mapping, which constitute a discrete time of the network to complete the cycle of the signal pattern recognition process. However, the nature of these models is also based on traditional neurons, not change their own information-processing mechanism.Process neural networks is a new type of artificial neural network in the last few years, and its input and output as well as the weight of connections can be both time-varying functions, process neural networks based on polymerization computing space of traditional BP neurons increasing. The cumulative effect of the timing operator in the mechanism of the process model with the characteristics of the direct extraction capacity, the time-varying signal characteristics of the process of memory directly reflected in the network structure, the form of operator time polymerization and network connectivity functions on the weight. Therefore, using process neural network can create a single model of time-varying signal pattern recognition, because of the flexibility of process neural network models, its method can be generally applied to the specific application of pattern recognition in dynamic systems.Against different problems in dynamic pattern recognition system, papers set up a process-based neural networks, fuzzy process neural networks, support vector machines of process of the dynamic model and pattern recognition methods, and made analysis in these models of information processing and problem solving mechanism, then proposing different network model algorithm,in particular the establishment of a new algorithm based on Legendre. Different simulation results verified the various models and the effectiveness of the algorithm created in papers.Papers presented an overview of pattern recognition and artificial neural network the basic principles and methods for different dynamic pattern recognition problem,Detail describe the process neural network model, fuzzy process neural network model based on function-start the process of Support Vector Machine model , and the corresponding model and the achievement of the learning, and the corresponding model and the achievement of the learning algorithm, at the same time for each model and algorithm made the results of the simulation and analysis.