Optimal management for innovative electric vehicle charging station with machine learning forecasting modules

In many situations, the resources in the organizations are focused on the improvement of processes in order to reduce their costs. Seen in this way, the use of an energy management system can offer a reduction of costs in both the economic and technical aspects. From the technical point of view, we extend the useful life of the elements of the system, while from the economic point of view, we will try to minimize the costs of complying with the constraints of the problem. In order to develop the optimization model included in the energy management system, some inputs are necessary, such as, for example, the demand of the system or the price of energy for the next period of time considered. In this work, two families of machine-learning models are applied to forecast system demand, one based on decision trees and the other based on neural networks. The use of these models is currently being extended due to the improvement of the computing and processing capacity of the computers. In addition to this phenomenon, the growth of the available data volume allows to have a very broad knowledge of the behavior of a recurrent process, making these models able to learn the behavior of the series. The objective sought in the present study is to optimize a system integrated by an electric vehicle charger, an energy storage system, a building and a cogeneration plant. For this, an optimization model focused on reducing system costs has been developed, deciding in each period of time what amount of energy to acquire, at what level to start the cogeneration plant and what to do with the potential energy. The use of a cogeneration plant in the system makes it necessary to take into account the generation of heat derived from the ignition of the machine when modeling the problem. Finally, in order to know the amount of energy, both electrical and thermal that will be required, a machine-learning model obtained from comparing different models and selecting the one that minimizes a given metric are used