Federated Learning Based Scheduling Method Of Idle Mobile Charging Station

With the rapid growth of automobile ownership,excessive emissions of car exhaust have caused great damage to atmosphere and affected human health dramatically.To address these problems,electric vehicles have become increasingly popular.Different from fossil-fueled cars,electric vehicles powered by onboard batteries have the following benefits: low energy consumption,zero emissions,low noise,etc.Hence,it is of great significance to develop and popularize the electric vehicles to mitigate pollution issues.Currently,the main problem to be resolved by electric vehicles is still the endurance.Electric vehicle drivers typically have "mileage anxiety" when carrying the long-distance travels.Currently,the widespread deployment of fixed charging piles to provide charging services has become the main charging solution.However,the large-scale deployment of fixed charging piles requires expensive and intricate preliminary planning.Due to these deficiencies,mobile charging stations have become an alternative solution.With great flexibility and low deployment cost,mobile charging stations can provide fast charging services based on charging demand at different locations,and become the complementary form of fixed charging piles.Typically,mobile charging stations will only travel towards the electric vehicles for charging after being requested.This method is not efficient,resulting in long idle time of mobile charging stations,low charging efficiency,long waiting time of electric vehicles,and delayed responses to charging demand.To solve the above issues,this thesis proposes a Federated Learning based Scheduling Method of Idle Mobile Charging Station(FL-SMIM).Due to the high mobility,the mobile charging stations could travel among different regions to achieve the balance between charging supply and charging demand.Within each region,mobile charging stations provide charging services according to charging demand and move towards the charging locations.Therefore,a cloud-edge-end collaborative mobile charging station scheduling framework is designed.In this framework,an edge server is deployed in each region,and the deep neural network model Bi-LSTM is trained jointly by the edge server and mobile charging stations.Then,mobile charging stations will move towards the locations of future charging demand predicted by the Bi-LSTM.This scheme could significantly improve the charging efficiency and reduce the waiting time of EV owners.Besides,the edge servers regularly collect the charging demand and operating statuses of mobile charging stations in different regions.Then,the cloud server migrates mobile charging stations among regions based on the information provided by the edge servers.In this thesis,the problem of inter-regional migration is modeled as bipartite graph matching problem and solved by Kuhn-Munkres algorithm.The experiment results demonstrate that the scheduling efficiency of mobile charging stations and the charging experience of EV owners can be improved through the efficient collaboration of cloud-edgeend framework and federated learning model.In addition,considering the real application,due to the high communication cost in federated learning,this thesis also proposes an adaptive communication interval mechanism to adjust the intervals for model parameters aggregation,which effectively improves the communication efficiency.In this thesis,a prototype system of mobile charging stations scheduling is developed based on IOS system and django framework.Based on this system,the mobile charging stations can train models and predict the future trajectories,and the operators can observe the statuses of charging loads and historical scheduling in regions.The test results show that the prototype system can visually display the relevant information,and can store and manage the historical data efficiently,which is available for further analysis and processing.To improve the charging efficiency,the cloud-edge-end collaboration and federated learning are adopted in the scheduling of mobile charging stations.However,the communication instability and the participation of fixed charging piles are temporarily ignored.Therefore,in future work,the fixed charging piles should be incorporated into the charging system,and a reliable collaboration mode should be designed.