Hybrid Genetic Algorithm: A Robust Parameter Estimation Technique and Its Application to Heavy Duty Vehicles

This study aims at establishing an accurate yet efficient parameter estimation strategy for developing dynamic vehicle models that can be easily implemented for simulation and controller design purposes. Generally, conventional techniques such as Least Square Estimation (LSE), Maximum Likelihood Estimation (MLE), and Instrumental Variable Methods (IVM), can deliver sufficient estimation results for given models that are linear-in-the-parameter. However, many identification problems in the engineering world are very complex in nature and are quite difficult to solve by those techniques. For the nonlinear-in-the-parameter models, it is almost impossible to find an analytical solution. As a result, numerical algorithms have to be used in calculating the estimates. In the area of model parameter estimation for motor vehicles, most studies performed so far have been limited either to the linear-in-the-parameter models, or in their ability to handle multi-modal error surfaces. For models with nondifferentiable cost functions, the conventional methods will not be able to locate the optimal estimates of the unknown parameters. This concern naturally leads to the exploration of other search techniques. In particular, Genetic Algorithms (GAs), as population-based global optimization techniques that emulate natural genetic operators, have been introduced into the field of parameter estimation. In this paper, hybrid parameter estimation technique is developed to improve computational efficiency and accuracy of pure GA-based estimation. The proposed strategy integrates a GA and the Maximum Likelihood Estimation. Choices of input signals and estimation criterion are discussed involving an extensive sensitivity analysis. Experiment-related aspects, such as imperfection of data acquisition, are also considered. Computer simulation results reveal that the hybrid parameter estimation method proposed in this study shows great potential to outperform conventional techniques and pure GAs in accuracy, efficiency, as well as robustness with respect to the initial guesses and measurement uncertainty. Primary experimental validation is also implemented including interpretation and processing of field test data, as well as analysis of errors associated with aspects of experiment design. To provide more guidelines for implementing the hybrid GA approach, some practical guidelines on application of the proposed parameter estimation strategy are discussed.