Predictive analysis of load at EVCS with deep learning algorithms

This paper considers Electric vehicle charging station(EVCS) as load and predicts the total charging power demand of an EVCS with deep learning algorithms. For pragmatic approach different categories of electric vehicles like electric bikes, electric scooters, electric cars and electric trucks are considered. Their arrival estimate is compared with non-homogeneous Poisson process and it comes under category of M1/M2/C queuing method. The EVCS load is a function of number of vehicles charging at a time, total charging current for each category at a particular time, arrival rate and time of vehicles. These independent parameters are related to the total charging power demand at an EVCS. Different Neural network(NN) optimization techniques like simple gradient descent, full batch gradient descent, stochastic batch gradient descent, adagrad and adam are used for better model representation of EVCS load which can be used for further analysis.