Research on Trajectory Prediction Method of Mobile Pollution Source Based on Hybrid Genetic Particle Swarm and optimized Extreme Learning Machine

Abstract

In order to accurately predict the trajectory of mobile pollution sources such as motor vehicles in real time, a trajectory prediction method based on hybrid genetic particle swarm optimization and optimized extreme learning machine (HGPSO-OELM) is proposed in this paper. optimized Extreme Learning Machine (OELM) avoids the disadvantage of traditional Extreme Learning Machine (ELM) which has poor generalization performance for small data sets. However, due to the random assignment of input weights and hidden layer node biases parameter groups, the prediction accuracy is affected. Therefore, finding the optimal parameter group can improve the prediction accuracy of vehicle trajectory. By introducing the hybrid genetic particle swarm optimization (HGPSO) algorithm, the optimal parameters of OELM model are dynamically optimized, which overcomes the randomness of the model establishment. Only a small number of hidden layer neurons are needed to achieve better prediction performance and improve the generalization of the network. Using the vehicle GPS trajectory data provided by ACM SIGS PATIAL GIS 2012, this paper chooses different historical data lengths based on ELM, OELM, HMM, GA-BPNN, LSTM, HGPSO-OELM and other methods to compare the one-step prediction performance under different historical data lengths. The experimental results show that the proposed HGPSO-OELM algorithm has higher prediction accuracy and real-time performance. The single-step prediction accuracy is the best when the length of historical data sequence is 20, and the multi-step time series prediction is realized under this length.