An approach of applying machine learning for range prediction for LD, HD commercial electrical trucks energy management [articol]

Abstract

This paper investigates the range anxiety problem in the electric truck (light and heavy duty) commercial truck. Predicting range is popular in passenger car. The necessity of predicting range in passenger car is connected to customer delight and comfortability. Whereas necessity of predicting range in commercial vehicle is connected to Target Cost of Ownership (TCO) recovery especially for fleet business. TCO recovery is mainly connected to capital cost and running cost. Capital cost is based on vehicle overall cost in which battery cost is the predominant one. Running cost is based on the distance covered by charging in which mileage is the predominant factor. The range of the EV truck depend on road profile, battery (parameters like SOC, SOH), driver driving behavior, particularly this research is focused on commercial vehicle payload, tire pressure & rolling resistance. This paper has two step approach. First step is to realize the range estimation related to tire pressure and rolling resistance. The second step is correlating this range estimation with respect to payload of the vehicle. In passenger car payload is more or less fixed. However, in the commercial vehicles kind of load carrying medium duty or heavy-duty vehicle where load carrying is based on the trip and delivery schedule. The load on the vehicle influences the rolling resistance and mileage of the vehicle. In more, precise the paper describes the deeper thinking approach to estimate the range accurately by considering rolling resistance with respect to tire pressure and pay load in the truck while driving. This is in addition to the traditional data considerations like weather conditions, driving behavior, other power train and battery related information like (torque demand, Battery SOC, SOH and aging). In order to achieve effective range prediction method, it is proposed to use machine learning algorithm to estimate and find the remaining distance to reach (along with respect to payload). Indirectly the model would identity any deviation in the tire pressure, which increases the rolling resistance because of which the energy depletion will be faster than the expected or idle conditions. Online fleet data will be used as a training data to train the model. With matured model, predicting the range in advance can be done by feeding trip and delivery schedule. In addition, attempt is made to implement the scalable battery based on the trip delivery schedule. Hence, TCO recovery will be faster through optimum battery estimation before the actual trip take place.