Summary: | 碩士 === 元智大學 === 工業工程與管理學系 === 106 === Nowadays, Energy saving and carbon reduction is a serious issue that people should care about. In Taiwan, the largest proportion of air pollution comes from cars/motorcycles, so if we could replace most cars/motorcycles with “shared bike”, it would have a massive contribution for improving the environment.
Besides, if we could help the supervisor to predict the demand of the “share bike” accurately and plan a profitable business strategy under reasonable cost, it would increase the consumer’s willingness to use the share bike with lower price and more convenience.
This thesis is basically using the “Recurrent Neural Net” (RNN) to analyze and predict the “share bike” requirement, it has a dependency on all past states and its result can be adjusted constantly by actual input and expected output, therefore the predictable deviation is smaller than the traditional time series prediction method and the machine learning time series prediction method. Besides, every neuron in RNN could process data faster than above traditional method as well.
However, RNN normally has vanishing gradient problem, yet the long short-term memory (LSTM) could easily solve the problem and also have the advantage that RNN has, thus for the vanishing level in this thesis would use LSTM to analyze.
|