作　　者： ; ;

机构地区： 中国科学院深圳先进技术研究院

出　　处： 《集成技术》 2018年第6期39-48,共10页

摘　　要： 可用停车位预测是智能停车诱导系统的关键问题之一。当前基于神经网络的预测技术在较短预测周期内,预测准确度的平均绝对误差在10左右,但随着预测步长或周期的增加,预测精度急剧下降。针对这一问题,该文提出了一种在中长预测时间周期内可保持数据变化特征的泊位预测方法。该方法使用模糊信息粒化获取特征数据集,通过训练长短期记忆神经网络预测未来的特征数据集,基于数据插值方法重建出整个区间可用停车位的连续变化曲线。仿真结果表明,该方法在相同预测步长的可用车位预测上,比传统预测方法具有更高的预测精度;在保持相近预测精度的条件下,比传统预测方法具有更高的计算效率。 Prediction of available parking spaces is the critical technique in the intelligent parking guidance system.The prediction technology based on neural network can achieve high accuracy in short-term prediction. And existing techniques can reach an average absolute prediction error of about 10.However,with the increase of prediction steps or time-span,the prediction accuracy will decrease dramatically.To solve this problem, a prediction method that can keep the characteristics of data changes in the long-span is introduced in this paper.The method uses the fuzzy information granulation to obtain the feature data sets.Then,a long short- term memory network is trained to predict the future feature data sets.Finally,an interpolation procedure is applied to reconstruct the curve of the parking space.The simulation results show that the proposed method can achieve better prediction accuracy and higher computation efficiency when compared with traditional prediction methods.

关 键 词： 停车诱导系统 模糊信息粒化 长短期记忆神经网络 三次样条插值

领　　域： []