Google Maps KPIs were successfully used to predict routes for EVs, taking into account the state of charge, temperature, wait time at the charging station, and distance.  

The Problem

The objective of this project was to utilize the deep reinforcement learning method to find route from point A to B considering the EV state of charge, temperature, wait time at charging station and distance as well.

We have successfully implemented the reinforcement learning method called DQN on google map. The DQN algorithm predicts the action (direction) that should be taken on coordinates. The project utilizes Google’s Maps APIs to get steps to reach the next position on map, and the time it will take to reach that position. The EV would move towards the direction predicted by DQN algorithm and APIs would be used to navigate. At any point if the state of charge is below a certain threshold, the EV would start looking for nearby charging stations through Maps API and would move towards that as its second objective; after recharging it will continue to its main end point. The battery discharge rate is based on the temperature and the EV would dynamically calculate the time required for charging to full capacity. Also, the random wait time on charging station has been added.

The model has been trained on a fraction of episodes of what reinforcement learning algorithms requires it to because of cost and time of APIs but still it shows potential of reaching its goal. The Red pointers are the steps taken by agent to find its way till the end point, Green represents start position, Dark Blue is the end, and Light Blue are points where it was charged. The Blue line is for comparison with graph-based approach (Dijkstra) to find the shortest path.

Case Studies

This is some text inside of a div block.