Public Bicycle Demand Forecast Based on PCA-BA-GRNN Model

The rapid development of economy and the extraordinary growth of the vehicle population has brought about serious pollution problems, so some new concepts of ecofriendly commutating evolve. One of the emerging concepts is the bicycle sharing system, which is quickly gaining popularity all over the globe. Most of the companies involved in the system of bike rental are faced with the challenge of balancing between supply and inconsistent demand. Thus, to predict the demand of bike on a particular day or time is very meaningful. This paper focuses on predicting the number of bike rentals for a city bikeshare system in Washington D.C., and presents a combination model named PCA-BA-GRNN. In this model, the dimension of original dataset is reduced by principal component analysis method, and then trained by a generalized regression neural network model which has been optimized the parameter by bat algorithm. The result of PCA-BA-GRNN model compared with the results of other models that are frequently used for bicycle demand prediction. The results show that PCA-BA-GRNN models perform well on the bikeshare data.