抄録
Diesel particulate filter system is very effective for reducing the amount of particulate matter in exhaust emission of diesel engine.This system catches particulate by its filter and the filter is regenerated when it is fully loaded.To control effectively its filter regeneration, an accurate and reliable estimation of particulate accumulation is required.In this case it is desirable to employ an estimation method capable to compensate unknown factors in filter dynamics and operating conditions.In our previous work, the estimation problem was solved by feedforward neural network (FFNN), widely known of its capability to model nonlinear system.However using FFNN encounters both training convergence and structure design problems.This work is designed to solve both problems by employing a novel compact mehtod capable to conduct training while determine network efficient structure automatically.The experimental results verify the effectiveness of the method where an efficient structure is automatically obtained with desired training convergence.
