Application of Sliding Window Approach for driving pattern recognition in HEV real-time control

The aim of this paper is to study and implement innovative Genetic Algorithm methods and develop different applications of Sliding Window Approach for driving pattern recognition inside a real-time control model for the optimization and analysis of complex hybrid vehicles, in particular (P)HEVs. The work is divided into different sections dealing respectively with the evolutionary algorithms and the with the improvement of the real time control strategy algorithm. Taking advantage of a Rule based Control Strategy (Clustering Optimization Rule Extraction) in which the optimal discretization of a 3D input domain, constituted by the vehicle velocity, vehicle acceleration and state of charge of battery (SOC) is generated and selected using genetic algorithm technique, a modification of this latter algorithm is proposed, called Stud Genetic Algorithm. Instead of stochastic selection, as common genetic algorithm, the fittest individual which is called Stud, is combined with all the other individuals to generate the new population. The crossover operation is the heart of this algorithm because in this mechanism the Stud G.A. method is performed. The results are then represented for different driving conditions and using different missions with a precise vehicle, hybrid architecture and other parameters. Subsequently, a second analysis shows the influence of the crossover factor over the simulation, in this way the percentage of individuals subjected to crossover or mutation is changing. An enhancement of the Sliding Window Approach (SWA) tool that is an adaptive real time control strategy for driving pattern recognition is developed. The SWA allows to classify current driving mission in a set of previously trained cycle that create the representative driving patterns , constituting the training set after a dynamic programming analysis and a subsequent training and validation with the Clustering Optimization Rule Extraction algorithm using a defined set of hyperparameters that will be properly selected. A subsequent comparison between the two type of previously mentioned controllers is done, in order to analyze the performance of both and understand the behavior switching between different inputs . With the aim of enhance the performance of the driving pattern recognition algorithm, the optimization problem bring the analysis to an accurate selection of the SWA features.