Research of the identification using neural network and performance forecast for a chain automatic lubricating device

The method combined digital simulation and neural network training was brought forward to research the subject of system modeling and performance forecast in automatic lubricating device of a motorcycle chain. In order to solve the low convergence speed of backpropagation forward BP Network in reverse transmit and easily to get into local minimum, the method of L-M optimized algorithm was used. A BP network model of a motor chain's automatic lubricating system with 4-inputs, 1-output, 3-layers and 25-neural cells was designed. The result shows this network is able to forecast the system performance under a variable of input, and to analyze the dynamic system performance while the change of structure parameter of the system. It is very convenient for the system modeling, analyzing, and designing in variable parameters. It also cuts down the development cost and decreases the blindness of design.