Research On Prediction Model Of Online Car-hailing Supply And Demand Based On Attention Mechanism

The imbalance between the supply and demand of online car-hailing is a hot issue in the transportation field in recent years.In response to this problem,relevant departments require online car-hailing platforms to establish a supply and demand calculation model,scientifically regulate the scale and structure of transportation capacity,and avoid imbalances between supply and demand.The existing online car-hailing supply and demand prediction models fail to fully consider the temporal and spatial dynamic correlation,which makes the model lose some important information to a certain extent,and reduces the accuracy of the model's prediction.Therefore,this article aims to build a prediction model of online car-hailing supply and demand from the perspective of temporal-spatial dynamic correlation to make up for the deficiencies of existing research.The main research contents are as follows:(1)Construct a forecasting model for the supply and demand of online car-hailing.On the one hand,the paper introduces the temporal attention mechanism to learn the key temporal information of the influence of different features on the predicted target.On the other hand,the paper introduces the spatial attention mechanism to learn the key spatial information of the influence of different features on the predicted target.Finally,combining temporal and spatial attention mechanisms,the paper constructs the Temporal?Spatial?AM-LSTM model.(2)Compare the prediction performance of RNN,LSTM,RNN-LSTM and Temporal?Spatial?AM-LSTM models through experiments.The experimental results show that the average absolute error of the Temporal?Spatial?AM-LSTM model is16.069.Compared with other control models,the model constructed in this paper has the smallest prediction error and the best performance.(3)Forecast the supply and demand based on the relevant data of Didi online carhailing.Set the time window to 144 and the prediction step size to 12,and the prediction results show that: 1)The predicted value is consistent with the actual value,the prediction error is within an acceptable range,and the model performs well.2)According to the temporal attention distribution matrix,it can be seen that the attention distribution value of the same time period on different dates is the largest.Secondly,the attention distribution value of 4 adjacent time periods in history is relatively large.Therefore,in practical applications,selecting the supply and demand data and their associated data of the same time period on different dates and 40 minutes in history for supply and demand prediction can effectively improve the prediction performance of the model.3)According to the spatial attention distribution matrix,this paper captures the dynamic dependencies between regions,improves the generalization ability of the model in space,and proves the necessity of introducing the spatial attention mechanism into the model.According to the research results of the paper,online car-hailing companies can implement real-time dynamic vehicle scheduling,tariff adjustment and driver compensation adjustment strategies;traffic management departments can formulate time and space precise management methods and measures;drivers and passengers can also reduce unnecessary cruise and travel time.As a result,the imbalance between supply and demand in the online car-hailing market will be effectively alleviated.