Genetic Algorithm-Based On-Line Optimization of a Speed Controller for a Combustion Engine

This paper introduces a new approach to optimal automatic parameterization of automotive engine controllers based on Genetic Algorithms. A set of control parameters is computed by minimization of a given objective function. The objective embodies the requirements for the controller to be optimized, e.g. fast and accurate compensation of set point changes and disturbances. The Genetic Algorithms evolve a generation of parameters by means of selecting the fittest ones, recombining and mutating them until the objective function is minimized and the optimal solution is found. The optimization algorithm is embedded into a Matlab tool determining the best parameters on-line in the car without human assistance. In an iterative procedure measurements of engine variables are used to evaluate the value of the objective function on-line. A new generation of parameters is set up and immediately employed in the engine management system. The proposed approach has successfully been applied to automatically optimize PI-control gains, filter constants and initialization values of an engine speed controller in selected operating points. Results of experiments in a car demonstrate the effectiveness of the approach to automatic tuning of automotive controllers by Genetic Algorithms.