A new automotive fuel-injection controller using the cerebellar model articulation controller (CMAC) neural network is developed and implemented to maintain the engine air-to-fuel ratio at its stoichiometric value. The CMAC controller requires minimal a priori knowledge of the engine dynamics because it can learn the dynamics and adapt to changing conditions in real time. The CMAC controller is experimentally evaluated on a research vehicle in a configuration fully compatible with production hardware. Initial training followed by continual adaptation allows the CMAC controller to maintain desired performance under previously inexperienced driving conditions.
展开▼