Probabilistic Forecasting for Demand of a Bike-Sharing Service Using a Deep-Learning Approach

Efficient and sustainable bike-sharing service (BSS) operations require accurate demand forecasting for bike inventory management rebalancing. Probabilistic provides a set of information on uncertainties in forecasting, thus it is suitable use stochastic management. Our research objective to develop probabilistic time-series BSS demand. We an RNN–LSTM-based model, called DeepAR, the station-wise bike-demand problem. The deep-learning structure DeepAR captures complex patterns correlations between stations one trained model; therefore, not necessary demand-forecasting models each individual station. makes parameter forecast estimates distribution target values prediction range. apply estimate parameters normal, truncated negative binomial distributions. dataset from Seoul Metropolitan City evaluate model’s performance. create district- station-level forecasts, comparing several statistical methods; as result, we show that outperforms other models. Furthermore, our district-level evaluation results all three distributions are acceptable forecasting; however, normal tends overestimate At station level, performs best, with least errors out tested