Research On Target Gather Prediction Algorithm Based On Information Fusion

Gather prediction is to predict the possible crowd gathering phenomenon in the city.Gather prediction can help the government deal with the gathering problems,and enhance the coping ability of various departments.Gather prediction can significantly improve the level of urban governance.The urban scene has a strong regularity of life,complex road structure,and factors such as geography,holidays,and weather have a large impact on user behavior patterns.And the multi-objective interaction is complex,and pedestrians,vehicles and public transportation need to be considered simultaneously.The existing methods have the problems of insufficient consideration of external knowledge and inaccurate portrayal of multi-user association relationships,which lead to low model prediction accuracy.Based on the above-mentioned problems,this paper conducts research around the target gather prediction algorithm based on information fusion as follows.1.Thesis proposes a trajectory prediction model based on spatio-temporal gated recurrent unit(STGRU).The existing trajectory prediction algorithms ignore external knowledge such as geographic environment,holiday and weather information affects the movement patterns.To this end,a fused road network gating structure is proposed to achieve the fusion of geographic environment feature extraction and external knowledges,so as to achieve more accurate user behavior pattern modeling and reduce the number of model parameters.It is shown through experiments that considering the road network structure and external knowledge can effectively improve the trajectory prediction performance,and the performance improvement is up to 18.1% compared with the existing methods.2.Thesis proposes a gather prediction model based on spatial feature enhancement.The spatial feature enhancement method based on Hilbert curves and graph embedding is proposed to address the problems of the existing spatio-temporal feature mining algorithms in label design and explicit feature extraction methods.The model enhances the spatial relationship between labels by using Hilbert curves for label design,and realizes real-time road network structure feature extraction in multi-user scenarios by fusing graph embedding methods.It is shown through experiments that the model can effectively improve the modeling ability of multi-user spatio-temporal features with the highest performance improvement of 22.6%.3.Complete the gather prediction alert system based on city flow.In order to verify the practicality of the algorithm,the gather prediction alert system based on city flow is designed and implemented in conjunction with the actual demand,which realizes the multi-objective gather prediction at the city level and provides favorable support for the management control of the city.