基于LSTM循环神经网络的泊位需求短时预测研究

Short Term Forecasting of Parking Demand Based on LSTM Recurrent Neural Network

本文提出基于深度学习理论的LSTM(Long Short-Term Memory,长短期记忆)单元的循环神经网络泊位需求预测模型,根据前几个小时泊位需求变化量对后续时间点对应的泊位需求量进行预测。以杭州某大学为实验对象,采用随机两天和特定两天数据进行实践验证。结果显示:采用LSTM循环神经网络模对区域内泊位需求进行预测能够比传统方法在结果上更加接近实际值,并且精度较为满意,表明该预测方法可行有效。

A Recurrent neural network berth demand forecasting model based on LSTM(Long Short-Term Memory)unit is proposed in this paper.The berth demand corresponding to the following time points is forecasted according to the change of berth demand in the first few hours.Taking a university in Hangzhou as an experimental object,the data are tested by random two days and given two days.The results show that the LSTM Recurrent neural network model can be used to predict the regional berth demand more close to the actual value than the traditional method,and the accuracy is satisfactory,indicating that the prediction method is feasible and effective.