Model based and AI range estimation for small EV vehicles

The excessive growth of greenhouse gases, among the most dangerous to the Earth and human health itself, has forced the development of green technologies with the aim to solve this issue. Since one of the most negative factors, which has a big impact to the increasing of the pollution is the transports field, both public and private, the carmakers have decided to introduce, in their production lines, electric vehicles (EV), both hybrid (HEV) and fully electric (BEV). Despite the goal is to reduce the greenhouses gases, the introduction of zero emission vehicles is not the solution to the problem, but only a starting point. Since the battery is one of the two sources of energy in HEV or the unique in BEV, it is necessary to have a good estimation of the state of charge (SOC). This is a fundamental parameter used in the battery management system (BMS) of the EV’s battery; in fact, it reflects a lot of battery performances, can guarantee battery protection, prevents overcharging and overdischarging, improves and extends the battery life, but it can be used also to implement control strategies. Moreover, one of the most important problems associated to the limited spread of electric vehicles is the range anxiety. Among the possible resolutions of this problem, there is the development of a system able to compute an estimation, accurate and robust, of the SOC level of consumption of the battery system. Nevertheless, finding a good estimation of the SOC is a still open challenge, due to the complexity of the problem. The aim of this thesis work, carried out in collaboration with Bylogix S.r.l., is to develop two different types of algorithms, both written in Python language, that provide an estimation of the SOC that the vehicle will be expected to consume. In the design process of the two algorithms and, mainly, in the second one, real data acquired by a Tesla Model 3 with a 75 kWh battery pack are used; the data were provided by the company. The first algorithm was designed on a model-based approach; therefore, it is based on the physical knowledge of the problem and on the equations that allow to calculate the power consumption delivered by the battery pack for the vehicle motion, starting from the longitudinal vehicle dynamics equation. Then, the energy and the associated SOC consumption, needed to complete the travel, are derived. Velocity, altitude profile, and weather information are the input data used by the algorithm. The algorithm is then tested with respect to real drive cycles, extracted from the database given by the company. The second algorithm, on the other hand, was designed following a data-driven approach, including an artificial neural network model. The network, based on a feed-forward architecture, computes an estimation of the power consumption delivered by the battery pack, using velocity, acceleration, and slope road angle as inputs. After, the power quantity returned by the neural network is integrated, obtaining the energy, and thus, the SOC estimation is finally computed. The training, validation, and testing processes of this algorithm are performed using the same database provided. Moreover, both the algorithms were designed to work offline, so are able to provide an estimation of the SOC consumption before the trip begin. To achieve this purpose, an algorithm that predicts the velocity profile was implemented, using road information and weather data. These data are obtained from APIs services and online databases, such as OpenStreetMap.