|Pratama Mahadika||Department of Electrical Engineering, Faculty of Engineering, Universitas Indonesia, Kampus UI Depok, Depok 16424, Indonesia|
|Aries Subiantoro||Department of Electrical Engineering, Faculty of Engineering, Universitas Indonesia, Kampus UI Depok, Depok 16424, Indonesia|
|Benyamin Kusumoputro||Department of Electrical Engineering, Faculty of Engineering, Universitas Indonesia, Kampus UI Depok, Depok 16424, Indonesia|
As one part of the advanced driver assistance systems (ADAS), adaptive cruise control (ACC) is introduced to reduce the possibility of traffic accidents by controlling the throttle and the pressure on the brakes to maintain a safe distance from the vehicle in front. Generally, linearized model-based controllers are used in the ACC. In this paper, a new approach to ACC’s inner loop is developed by designing the controller using neural network predictive control (NNPC) which integrates the capability of artificial neural networks (ANN) to imitate vehicle characteristics and model predictive control (MPC) to obtain the minimized quadratic error between future reference trajectories and predicted outputs. Two separate control loops will be used: an outer loop based on a decision algorithm, and the PI controller, which will give the inner loop a speed reference to maintain the safe distance from the vehicle in front. NNPC is used in the inner loop to manipulate throttle and brake pressure on the brakes in order to control the speed of the following vehicle. Simulations will be carried out using software-in-the-loop (SIL) between CarSim and Simulink. The ANN model is identified and verified to mimic the nonlinearity behavior of the vehicle model using the mean square error (MSE) parameter. The results of this study are that the ANN model is able to imitate the vehicle dynamic with MSE equal to 0.0095, and the controller can maintain a safe distance while having a smooth response.
Adaptive cruise control; Artificial neural network; Dynamic vehicle model; Neural network predictive control
In recent years, trends in improving driving safety have become an important concern for the automotive industry because traffic accidents are major concerns faced by drivers. These problems can be avoided by introducing some forms of driver assistance to prevent accidents. In fact, 50% of accidents that occur are rear-end collisions. That’s why advanced driver assistance systems (ADAS) are developed by automobile manufacturers to make driving safer. The National Transportation Safety Board said that active safety systems are 50% more effective in reducing death rates in accidents compared to passive safety systems such as airbags (ACEA, 2018). As part of the ADAS system, active cruise control (ACC) was developed in early 1990. The ACC system is capable of adjusting vehicle speed while maintaining a safe distance from the vehicle in front. This system modulates the throttle valve and brake pressure to reduce or accelerate the vehicle to the desired speed and distance. Radar, laser, and other sensory devices are used to measure the distance from vehicles in front, so the ACC system can choose a proper driving mode.
Several methods for the ACC system have been developed in four-wheeled vehicles. Classical methods such as PID and fuzzy have been developed since a few decades ago. Rout and his colleagues utilized a PID controller that was optimized with genetic algorithm (GA) to produce optimal PID tuning (Rout et al., 2016). Shakouri developed the ACC system with the gain-scheduling method using PI and LQ controllers to manipulate throttle valve and brake pressure (Shakouri et al., 2011), and Pananurak and his colleagues used fuzzy logic algorithms for ACC systems (Pananurak et al., 2009). Some predictive methods began to be developed because they resulted in better control of vehicle dynamics. Shakouri also developed the two-loop ACC system by utilizing the model predictive control (MPC) for throttle and brake control as inner loop control and PI as a speed controller for outer-loop control (Shakouri and Ordys, 2014). After that, Naus and his colleagues utilized implicit MPC and multi-parametric quadratic programs for online identification of ACC systems (Naus et al., 2010). Miftakhudin and colleagues developed a multistage MPC system with constraints for the ACC controller to achieve a smooth response (Miftakhudin et al., 2019). On problem shared by all the research mentioned above, is that the controller uses the linearization method in modeling the longitudinal motion of four-wheeled vehicles. These methods limit the controller’s ability to work only in a specified range and are difficult to obtain for a large working range.
From many control methods that have been developed, artificial neural network (ANN) has not been widely applied in automotive controllers, especially in ACC systems, even though ANN is widely known for its ability to capture nonlinear phenomena. For that reason, the main contribution of this work is to make a controller that integrates the ability of ANN to capture nonlinear dynamics of moving vehicles and the predictive ability of MPC to control ACC systems. This method, called neural network predictive control (NNPC), began development 1996 when Soloway started working on neural generalized predictive control that combines ANN for model identification and generalized predictive control (GPC) for the controller (Soloway and Haley, 1996). These methods created a new problem for minimization of the cost function in GPC. Newton-Raphson was used to compute optimization problems numerically to obtain optimal control sequences for the controller. This paper used a slightly modified method, quasi-Newton, to compute the control sequence for the controller and Broyden Fletcher Goldfarb Shanno (BFGS) algorithm to solve inverse Hessian matrices that appear in Newton-based optimization. In this research, simulation will be carried out in software-in-the-loop (SIL) between MATLAB and CarSim.
Based on the data and simulation implemented in CarSim and MATLAB, the ACC system with NNPC can track the leader vehicle’s speed and keep the safe distance desired with a relatively small error in distance. This research has limitations in acceleration of the leader vehicle. A rapid change in speed will need new training data sets for the ANN model. This method also has a drawback in the working range because the ANN model can only work accurately if there is sufficient data training, and is hard to implement in a wide working range. It would need vast data and take huge computational effort to train. For future work, another ANN model can be implemented to switch between a vehicle’s speed range and acceleration to be able to work in different scenarios.
This research is funded by a Research Grant from Publikasi Terindeks Internasonal (PUTI) 2020 Universitas Indonesia.
ACEA, 2018. Active Vehicle Safety Most Effective, New Analysis of Accident Data Shows ACEA—European Automobile Manufacturers' Association. Available Online at https://www.acea.be/press-releases/article/active-vehicle-safety-most-effective-new-analysis-of-accident-data-shows/ Accessed on June 25, 2020
Chong, E.K.P., Zak, S.H., 2013. An Introduction to Optimization. 2nd Edition. New York: John Wiley & Sons, Inc
Gao, Z., Wang, J., Hu, H., Yan, W., Wang, D., Wang L., 2016. Multi-Argument Control Mode Switching Strategy for Adaptive Cruise Control System. Procedia Engineering, Volume 137, pp. 581–589
Martinez, J.J., Canudas-de-Wit, C., 2007. A Safe Longitudinal Control for Adaptive Cruise Control and Stop-and-Go Scenarios. IEEE Transactions on Control Systems Technology, Volume 15(2), pp. 246–258
Miftakhudin, M.I., Subiantoro, A, Yusivar, F., 2019. Adaptive Cruise Control by Considering Control Decision as Multistage MPC Constraints. In: IEEE Conference on Energy Conversion (CENCON), Yogyakarta, Indonesia, pp. 171–176
Naus, G.J.L., Ploeg, M.J.G., Van de Molengraft, W.P.M.H., Heemels, M, Steinbuch., 2010. Design and Implementation of Parameterized Adaptive Cruise Control: An Explicit Model Predictive Control Approach. Control Engineering Practice, Volume 18(8), pp. 882–892
Pananurak, W., Thanok, S, Parnichkun, M., 2009. Adaptive Cruise Control for an Intelligent Vehicle. In: IEEE International Conference on Robotics and Biomimetics, Bangkok, pp. 1794–1799
Rout, M.K., Sain, D., Swain, S.K., Mishra, S.K., 2016. PID Controller Design for Cruise Control System using Genetic Algorithm. In: International Conference on Electrical, Electronics, and Optimization Techniques (ICEEOT), Chennai, pp. 4170–4174
Shakouri P., Ordys A., Laila D.S., 2011. Adaptive Cruise Control System: Comparing Gain-Scheduling PI and LQ Controllers. In: The 18th IFAC World Congress, August 28–September 2, Milano, Italy
Shakouri, P., Ordys, A., 2014. Nonlinear Model Predictive Control Approach in Design of Adaptive Cruise Control with Automated Switching to Cruise Control. Control Engineering Practice, Volume 26, pp. 160–177
Soloway, D., Haley, P.J., 1996. Neural Generalized Predictive Control. In: Proceedings of the 1996 IEEE International Symposium on Intelligent Control, IEEE
Sørensen, P.H., Nørgaard, M., Ravn, O., Poulsen, N.K.,1999. Implementation of Neural Network Based Non-Linear Predictive Control. Neurocomputing, Volume 28(1-3), pp. 37–51
Subiantoro, A., Fauzan, M., Feri, Y., 2018. Adaptive Cruise Control based on Multistage Predictive Control Approach. In: The 4th International Conference on Nano Electronics Research and Education (ICNERE)
van den Bleek, R.A.P.M., 2007. Design of a Hybrid Adaptive Cruise Control Stop-&-Go System. TNO Science & Industry Business Unit Automotive Department of Integrated Safety, Technische Universities’ Eindhoven Department of Mechanical Engineering Control System Technology Group