特征分离编码的景区短期客流量预测模型

Prediction model of short-term tourist flow in scenic area based on feature separation encoding

为增强景区科学管理、缓解交通压力、减少安全隐患、提升游客体验,提出基于特征分离编码和注意力机制的网络模型(feature separation encoding and attention mechanism network,FSEAMNet)预测景区短期客流量。该模型包含序列到序列(sequence-to-sequence,Seq2Seq)结构,将不同分布规律的特征进行分离并独立编码,融合成最终的编码向量序列。在每个解码时刻,注意力机制将编码向量序列重新组合成一个上下文向量,解码器从上下文向量解码出未来的游客数量。通过真实的景区数据库数据构建训练集、测试集。实验结果表明,与其它模型相较,FSEANet的预测误差最多可下降82.80%,该模型在工程应用案例分析中对未来一周客流量预测的每日相对误差均在10%以下。所提模型能对实际景区未来短期客流量实现较准确的预测。

To enhance scientific management of scenic area,relieve traffic pressure,reduce potential safety hazards and improve tourist experience,a model based on feature separation encoding and attention mechanism network(FSEAMNet)was proposed to predict short-term tourist flow.This model included a sequence-to-sequence structure,which separated and encoded the features of different distribution independently,and they were merged into the final encoding vector sequence.At every decoding moment,the attention mechanism recombined the encoding vector sequence into a context vector.The decoder decoded the number of future tourists from the context vector.The training set and test set were constructed from the real scenic area database data.Experimental results show that,compared with other models,the prediction error of FSEAMNet can be reduced by up to 82.80%.In the case study of engineering application,the daily relative error of the prediction of tourist flow in the next week is less than 10%.The proposed model can achieve more accurate prediction of the future short-term tourist flow of the actual scenic area.