Statistical characterisation of the real transaction data gathered from electric vehicle charging stations

Marco Giacomo Flammini, Giuseppe Prettico, Andreea Julea, Gianluca Fulli, Andrea Mazza, Gianfranco Chicco
2019 Electric power systems research  
Despite the many environmental benefits that a massive diffusion of electric vehicles (EVs) could bring to the urban mobility and to society as a whole, numerous are the challenges that this could pose to the electricity distribution grid, particularly to its operation and development. While uncoordinated management of EVs can lead to load imbalances, current or voltage variation excess and steep power requests, properly designed and well-coordinated integration approaches can in contrast
more » ... n in contrast provide flexibility, hence value, to the whole electrical system. Such step can be achieved only if real data are available and real drivers' behaviours are identified. This paper is based on a real dataset of 400,000 EV charging transactions. It shows and analyses an important set of key figures (charge time, idle time, connected time, power, and energy) depending on driver's behaviour in the Netherlands. From these figures, it emerges a key role of the uncertainty of the relevant variables due to the drivers' behaviour. This requires a statistical characterisation of these variables, which generally leads to multimodal probability distributions. Thereby, this paper develops a Beta Mixture Model to represent these multimodal probability distributions. Based on the emerged statistical facts, a number of results and suggestions are provided, in order to contribute to the important debate on the role of EVs to move to a fully decarbonised society. T provide increases, adding more options to modify the demand patterns, as well as the provision of reserves [7] . The EV aggregations (in both G2V and V2G paradigms) are considered as possible source of operational flexibility. Another option is Vehicle-to-building (V2B), in which the EV batteries can be used also as electrical energy storage units for the local building (e.g., offices and in general workplaces), without sending power to the external grid. Workplace EV charging is addressed in Ref. [8] by taking into account the uncertain EV charging demand, also considering different options of G2V for general users and possible V2B only for the local employees. An interesting classification of the flexibility sources is presented in Ref. [6], by dividing them into different categories based on whether they are used (actual flexibility) or not (potential flexibility) and how they can be obtained (for example, within an ancillary service market). Flexibility has also been defined in Ref. [9] for the collective EV charging demand, adapting the model introduced in Ref. [10] for aggregate residential users, based on the statistical properties of the demand variations in time. The most common option to provide EV charging flexibility is to use automated charging management systems [11] , also considering the specific standards for limiting the EV charging rate [12] . The effect of the EV charging is more important at the distribution system level, due to the fact that the "dumb" charging strategy can lead the operation of the system to be out of the acceptable range [13] . However, an intelligent management of the EV charging process would allow the exploitation of the potential EV flexibility [14] . In this framework, optimisation of the EV charging schedule is addressed in Ref. [15] by adding a conditional value-at-risk term to the objective function. Furthermore, the optimisation model presented in Ref. [16] handles both the renewable energy sources (RES) curtailment and the EV charging in such a way that the overall strategy becomes economically convenient; one limitation in this contribution is the description of the departure and arrival times by means of Gaussian distributions, not obtained directly from an analysis of EV information, but starting from data regarding Internal Combustion Engine (ICE) cars. The Gaussian distribution is used in other contributions (e.g., [17] ), but is generally not suitable to model the statistical properties of the relevant variables that characterise the EV mobility. The scarcity of available real data regarding EVs and their charging stations has pushed researchers to define ad-hoc probability distributions for a number of variables used in the study of EV integration in the electrical networks, in some cases starting from travel survey data on sets of individual vehicles (not only EVs). Finding out suitable probability distributions for these variables is an open challenge, and has to be driven by real data. The lognormal distribution is adopted in Ref. [22] to represent EV-related random variables (arrival time, departure time, and initial state of charge). Other approaches consider EV-related patterns taken from specific databases without performing a detailed statistical characterisation. The contribution presented in Ref. [18] studies 255 charging stations in the UK, considering weather data, and the EV charging demand patterns are clustered with a k-means algorithm. The work of Kara et al. [19] identifies the benefits, for load aggregators and the distribution grid, of applying smart charging driven by time-of-use pricing to 2000 non-residential EV charging stations. The ElaadNL database has been analysed also in Ref. [20] to quantify the demand response potential of consumers coordinated with EV charging, and in Ref. [21] by developing eight indicators to allow a comparison among EV public charging infrastructures. Concerning ancillary services, the EV contribution to the frequency control (in particular to the inertial control) has been analysed in Ref. [23], with the implementation of a proper EV battery control; however, it is necessary to determine the "availability" of this flexibility, by investigating real data linked with the real behaviour of the drivers. The paper [24] tries to cover this gap, by analysing 390k transactions, and,
doi:10.1016/j.epsr.2018.09.022 fatcat:57zfwvwwpne3bc6nwxme3wzpc4