Study On Time-varied System Signal Processing Technology Based On PNN

The theory and technology of complex signal processing is an importance research directionof computer applications technology area today. With the developing of signal and informationprocessing research area, and randomicity of affect factor existed many nonlinear system, andcomplexity of information conversion mechanism, intelligence signal and signal processingtechnology is becoming a hot research direction of computer applications technology area.Process Neural Networks (PNN) is a theory and model aim for that the input of manysystem is a procedure which vary with time, and the output of some control signal not onlydepends on the space aggregation effect but also have an intimate dependence with the timecumulative effect. The input and link weight of process neurons are time-varied functions, andadd an aggregation operator to time on the base of space aggregation operation of traditionalneuron, then make the aggregation operation and incentive effect of process neuron can reflectspace summation effect and time cumulative effect of time-varied input signal in the same time,itcan realize the complexity mapping relation between input and output of nonlinear dynamicsystem .In has good adaptability to solve many question that the input and output have therelation with process.This paper mainly aim for the question of build the model of the nonlinear time-variedsystem signal processing, research on the theory and learning algorithm of PNN and PNN applyin time-varied system signal processing technology, integrated the practical application , on thebase of key model and conception of PNN. Introduce the intelligence information processingmethod that used usually, the conception of process neuron and the key model of PNN,researched and built the rational process neural networks and the process neural networks withtime-varied inputs and outputs function, and analyse the character of them. In research oflearning algorithm of PNN, mainly aim for input and link weight of PNN are time-variedfunction and the case that process neuron include the operation of time and space, bring forwardlearning algorithm based on gradient descent and expanding on function orthogonal basislearning algorithm. The algorithm has good effect to process neural network training problemwith continuous time input.During study on the pretreatment method of PNN apply in signal processing, the continuousWalsh conversion method and Fourier conversion method is presented, Research on the PNNused in remove noise in the signals, pattern recognition and signal prediction, the network modeland processing method is presented. This part mainly includes the practical application problemsfor example oil field development procedure simulation,the validity of the process neuralnetworks with time-varied system signal processing is proved.