The B.P.Network's Application In Multivariable Regress Analysis

Artificial neural network is abbreviated as A.N.N.. It is an engineering system to simulates the constitution structure and intelligence behavior that of realizing and understanding being brain's organization as well as action mechanism.Artificial neural network is different from present computer, it is a type of non-lineal processing cell. Neural cell produced signal when the result from processing all of input information is lager than a limited value. So artificial neural network is a dynamic system with highly non-lineal continued time.The key of artificial neural network is the suitable connection around processing unit.Neural units making up of artificial neural network is such not as disorder as they were imaged, but they were classified. The every layer's unit receives signals from the down and sends information to upper. The input layer is responsible of receiving information, the output layer yields final result needed, there are a few of layer between input and output layer, they are called hide layer generally. Therefore we could not predict how input signals are transmitted around these units. Each neural cell receives signal from down firstly, it process the signal secondly and yields result, finally it sends information to upper layer. There the key is the each cell's weight-efficiency.In order to obtain the weight-efficiency needed, the artificial neural network should be trained.B.P.,that is Error Back Propagation Neural Network, was published by Rumelhart in 1986. It is an adjusting artificial neural network with themselves toward potential relationship between input and output.The arithmetic of Error Back Propagation Neural Network is adjusting the weight-efficiency by means of the difference between practical and ideal output. With a lot of repeat, the difference between practical and ideal output could be minimal.The construction, training arithmetic, action mechanism and application in multivariable regress of B.P.(Back Propagation) is studied in detail in this paper. In order to strengthen B.P. arithmetic's convergence and converging speed, the B.P. arithmetic is improved. An example of applying B.P. to real work is demonstrated.