基于大数据分析的多维公路货运价格预测问题研究

Research on multi-dimensional highway freight pricing prediction based on big data analysis

公路货运价格由于受油价和货运周转量的动态变化影响,无法实现精确预测,一直困扰着运输行业.采用时间序列预测与BP神经网络相结合的方法进行预测研究:首先,将订单中的经纬度信息利用高德地图API接口转变成具体地址,从地理信息编码提取省、市信息,对货运订单进行跨区域分类;然后,运用指数平滑法和ARIMA模型分别对油价及货运周转量动态特征进行预测;再对运输距离、油价和货运周转量数据进行标准化操作,用处理后的数据对BP神经网络进行训练,得到训练完成的神经网络;最后,使用测试样本对模型进行测试,将加入时间序列预测的BP神经网络模型与不加入时间序列预测的模型相比较,发现前者的预测误差减少37.2%.

The price of road freight is affected by many factors,especially the dynamic changes of oil price and freight turnover.The inability to accurately predict has always plagued the transportation industry..This paper adopts the method of combining time series prediction with BP neural network.Firstly,the latitude and longitude information in the order is transformed into a specific address by using Amap API interface,the provincial and municipal information is extracted from the geographic information coding,and the freight order is cross-regional classification;then the exponential smoothing method and the ARIMA model are used to predict the dynamic characteristics of oil price and freight turnover respectively;next the transportation distance,length,oil price and freight turnover data are to be standardized,and the processed data is used to test BP neural network obtained;last,the model is tested with the samples.It finds the BP neural network model with time series prediction reduces the prediction error by 37.2%,compared with the model without time series prediction.