D3P: Data-driven Demand Prediction for Fast Expanding Electric Vehicle Sharing Systems
summary
Predicts demand for rapidly expanding EV sharing networks and handles stations with little or no history by combining a LSTM temporal encoder for local history and a Graph Convolutional Neural Network for dynamic spatial correlations, feeding a multi-scale predictor that outputs long-term expected and near-term instant demand. Evaluated on Shanghai data with ~4,000 stations, 16,000 vehicles and 7 million trip records, the model achieves up to a three-fold improvement over prior methods and shows robustness to fast deployment and closure events.
AI-written summary · about
related talks
40mDeep Learning and Time Series Forecasting for Smarter Energy
26mAn EV Boom or a Gridlock? A Data-Driven Approach to UAE's Green Energy Transition
18mUsing Transformer Models to Predict Demand at Scale in Production
49mStructural estimation methods to model user behaviour in bike-sharing systems
41mAnalysing and predicting inner-city parking space occupancy
21m"Capacity Prediction" instead of "Capacity Planning":
30mReal-Time ML services: From ideation to production
38mUsing Recurrent Neuronal Networks for Customer Predictions