Research On Forecasting Model Of Hot Spots In Characteristic Towns Based On Deep Learning

Same as city,characteristic towns also face the problems caused by urbanization,such as population explosion,traffic jam,increasing energy consumption,environmental degradation,behindhand planning and so on.Short-term prediction of hot spots in characteristic towns is conducive to solving the problem of crowd gathering safety and traffic jam.Otherwise,it can remind merchants to be prepared to provide better services.Therefore,this thesis that based on deep learning method,constructs the short-term prediction model of hot spots in characteristic towns,and visualize the prediction results.The main work of this thesis is as follows:(1)In response to the problem that existing methods of hotspot prediction are mainly oriented to cities,not suitable for characteristic towns,this thesis combined with actual situation of the characteristic towns builds model for the traffic flow.According to the thesis,the definition of regional traffic is given.Then,the method that collect and process data of region flow,weather and holiday in the characteristic towns is determined.Based on studying the variation of data,the influencing factors of the region flow are summarized.These data and influencing factors provide the basis for the design and construction of the prediction model.(2)For the problem that single neural network model can not learn the temporal and spatial correlation at the same time,by integrating four residual networks with exactly same structure and one fully connected network,the model named Quad-ResNet for short-term prediction of hotspots in characteristic towns is constructed.Meanwhile,the overall framework of the model along with the network structure of each part and the fusion method are discussed in detail.Then,the appropriate pre-processing methods for converting data of region flow,weather and holiday into forms acceptable for the model are selected.By this means,the training method and prediction method of the model are determined.(3)Obtain the optimal parameters for the model performance by performing a series of parameter selection experiments.And on this basis,a short-term prediction comparison experiment of hot spots is designed.The Quad-ResNet model is tested on the same dataset with LSTM,CNN and ST-ResNet models.Learning the results would prove the validity of Quad-ResNet model.(4)Through the statistical analysis to the data,measuring standard for hot spots is built up.The hot spots visualization system is implemented based on Node.js,Vue.js and Echarts.js programming,which then presents the results of prediction model in the form of heat map.Eventually,the Quad-ResNet is used to predict hot spots in characteristic towns with real data.According to the actual situation and the heat maps,the results of prediction are analysed.