Short Term Traffic Flow Prediction Based On IICA-LSTM-BP Combined Model

The vigorous development of transportation and logistics provides a driving force for social development,but while the increasing number of private cars brings convenience to people’s travel,there are also hidden dangers of unbalanced development with the existing traffic conditions.Improving the accuracy of short-term traffic flow prediction is an important part of urban intelligent transportation system,which can provide data support for traffic guidance and travel planning.Therefore,the research on the optimization of short-term traffic flow prediction model and training algorithm is carried out.The following is the main research results of this thesis.The theoretical basis of short-term traffic flow prediction is studied.Based on the open source traffic data of Seattle road network in the United States,the correlation and periodicity of traffic flow are analyzed,and the volatility and randomness that restrict the prediction accuracy are studied.The research on the characteristics of shortterm traffic flow data provides theoretical support for the construction of prediction model.BP(back propagation)neural network prediction model and long-term and shortterm memory neural network(LSTM)prediction model optimized by imperial competition algorithm(ICA)are established.Through the research on the model principle and parameter transfer process,the ICA-BP and ICA-LSTM prediction models are programmed,and the simulation experiments are constructed.The prediction results are compared with PSO-BP and error back propagation training models under the same data set.The simulation results show that the Empire competition algorithm can successfully improve the prediction accuracy of the model,and the average absolute percentage errors of ICA-BP and ICA-LSTM are reduced by3.94% and 2.93% respectively compared with the error back propagation method.A combined forecasting model of lstm-bp(IICA-LSTM-BP)based on improved imperial competition algorithm is established It improves the ability of the model to deeply mine the characteristics of traffic flow data and the training efficiency of model parameters.The data characteristics obtained from LSTM network training are combined with traffic flow data as the input of BP network The LSTM-BP combination model is established The powerful generalization and fitting ability of LSTM network are deeply explored.ICA is improved and the concepts of adaptive assimilation coefficient and probability factor are proposed.The colonies that improve ICA have three mobile strategies: assimilation,revolution and exploration The movement trend of assimilating colonies to imperialists increases adaptively with the increase of iteration times The selection probability of mobile strategy changes adaptively with the depth of global search and local search.The optimization algorithm strengthens the ability of the colony to search globally and jump out of the local optimal solution in the early stage of the algorithm,as well as the data mining intensity in the later stage of the algorithm.The prediction effect of IICA-LSTM-BP is tested by several groups of simulation experiments.The average absolute percentage error of IICA-LSTM-BP is 11.5%,and the prediction accuracy is better than other models under the same data set.The average absolute percentage error of IICA-LSTM-BP under different road sections,morning and evening peak,working days or weekend data sets is 7%-9%.The first mock exam results show that IICA-LSTM-BP has good performance in prediction accuracy and general applicability.The prediction effect is greatly improved compared with the single model and the prediction model that is not optimized by the ICA.