NDEKF Neural Network Applied to Electronically Controlled Fuel Injection System

The electronically controlled fuel injection system in locomotive diesel is a complicated nonlinear system. So we lead the NARMAX (nonlinear auto-regressive moving average with exogenous inputs) neural network into its model. In order to overcome the deficiency that the neural network structure relies on one´s own personal experience, we used the pruning based on the Hession matrix to optimize the network structure. NDEKF (node-decoupled extend Kalman filter) which was adopted to train networks converges more quickly than the back-propagation algorithm does and assists in the avoidance of local minimum. The experiments showed that the hybrid neural networks of the nonlinear auto-regressive with exogenous outputs are very close to the actual results, and the inputs can identify object ranks precisely.