基于优化LSTM模型的停车泊位预测算法

Parking prediction algorithm based on optimized LSTM model

针对道路停车泊位数预测准确性不高、预测误差较大的问题,提出一种基于循环神经网络LSTM模型的停车泊位预测算法,研究从历史停车数据中挖掘知识并预测不同时段内的停车泊位数。首先,建立一种优化的基于LSTM和双向LSTM网络的LSTM模型,通过双向LSTM网络对上一层的LSTM网络预测后的时间序列再进一步进行学习训练,以有效克服预测误差大的缺点;其次,结合正向LSTM和逆向LSTM具有的捕获数据时序性和长程依赖性的优势,进一步提高预测结果的精确度。利用不同实测道路停车场的数据对所提算法的有效性进行验证,结果表明,在同等条件下,所提算法的准确度和效率均优于LSTM模型算法,预测精度和训练速度均有较大提高。

Concerning the problem of low accuracy and large prediction error fluctuation in prediction of parking space, a parking space prediction algorithm based on optimized LSTM model was proposed to excavate knowledge from historical parking data and forecast the parking volume in different time periods. Firstly, an optimized LSTM model based on LSTM and Bi-LSTM (Bi-directional LSTM) network was constructed to further study and train, the time series results predicted by the previous LSTM layer through Bi-LSTM, which can effectively ovcome the defect of large fluctuation of prediction error. Then, forward LSTM and reverse LSTM were combined with the advantage of capturing data timing and long-range dependencies, further improving the accuracy of prediction results. In order to verify the effectiveness of the proposed algorithm, it was applied to different measured road parking lots. Under the same conditions, compared with the algorithm based on LSTM model, the proposed algorithm has higher accuracy and efficiency. The conclusion shows that the prediction precision and training speed have been greatly improved.