An artificial intelligence based control strategy for hybrid vehicles

Internal combustion engine (ICE) vehicles make up the vast majority of the worldwide transportation system. ICE vehicles use petroleum resources as an energy source and emit a lot of carbon emissions, causing global warming and environmental pollution. One of the main reasons for the development of hybrid electric vehicles is to reduce carbon emissions minimized and carbon emissions are reduced. There are many studies in the literature for the energy control method of hybrid electric vehicles. In this study, a new rule-based control method is proposed to minimize fuel consumption and carbon emission values in driving cycles in the experimental model of parallel hybrid electric vehicle. The proposed control method allows the internal combustion engine to operate at the optimum point. In addition, the electric motor (EM) is engaged more frequently at low speeds and the electric motor can also operate as a generator. Then, fuel consumption values of a parallel hybrid electric fuel vehicle were calculated in Matlab simulation environment using ECE-15 (Urban Cycle), EUDC (Urban Cycle) and WLTC (Worldwide Harmonized Light Vehicles Test Cycle) and NEDC (New European Driving Cycle) driving cycles. By selecting 55% and 65% battery values close to the minimum and maximum values, the change in the fuel consumption of the vehicle at different SOC values was observed. The fuel efficiency of the parallel hybrid electric vehicle with 55% SOC was 9.6% in the EUDC cycle, 29.3% in the ECE-15 cycle and 17% in the NEDC driving cycle. The fuel efficiency of the parallel hybrid electric vehicle with 65% SOC was 35.7% in the NEDC driving cycle, 50.1% in the EUDC cycle, 52.3% in the ECE-15 driving cycle, and 49.35% in the WLTC driving cycle thanks to the energy control management strategy. The proposed control strategy ensures continuous control of the base battery state and optimum efficiency of the internal combustion engine by operating the drive systems at the highest level. In this study, a new dataset was obtained by creating a simulation model of the parallel hybrid vehicle in Matlab environment in ECE-15, EUDC and NEDC driving cycles and also at various SOC levels. This data set has been created by considering all components affecting the vehicle drive system. In this thesis, seven different machine learning methods (ML) were used. These methods are regression, random tree, random forest, decision stump, M5P, REP Tree and multilayer perceptron. Fuel forecasting was performed with seven different machine learning methods (ML). Using the obtained data set, these methods were compared in terms of performance criteria. Correlation coefficient (R^2), Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Two different test datasets were created to prevent machine learning from memorization. For both test datasets, the fuel consumption model was trained and tested using the correlation coefficient R^2, mean absolute error (MAE) and root mean squared error (RMSE) performance criteria. R^2 values of 0.97 and 0.90 were obtained for the random forest algorithm for both test data. These values mean that the random forest algorithm achieves 97% and 90% accuracy in fuel consumption estimation. The smallest MAE values of 0.0011 and 0.003 were obtained in the random forest algorithm for both test data. In addition, the smallest RMSE values of 0.0284 and 0.0138 were obtained in the random forest algorithm for both test data. According to the results obtained, the random forest algorithm was found to be the most appropriate method for estimating the fuel consumption model for parallel hybrid electric vehicle in terms of R^2, MAE and RMSE