Research On Short-term Traffic Flow Forecasting Method By Mutual Information Feature Selection And ACSA-WNN

Short-term traffic flow forecasting is the prerequisite for traffic guidance and control to intelligent transportation.It can improve the performance of the intelligent transportation system.There are many traffic variables which are related to short-term traffic flow forecasting.But using excessive traffic variables will introduce noise and increase the amount of computation,which will affect the real-time and lead to the forecasting accuracy decline.So,how to select traffic variables is worth studying.In addition,short-term traffic flow is a highly complex nonlinear system with random mutability.BP neural network has the self-learning ability and self-organization,and can approximate any nonlinear mapping.Therefore,it is suitable for short-term traffic flow forecasting.However,BP neural network is sensitive to the initial parameters value.The convergence rate is slow,and is easy to fall into the local optimal.Aiming at these problems,some studies are carried out,and the main work and contribution are as follows:1.Most of the current prediction methods select traffic variables according to the experience and trial-error.It lacks strict theoretical basis and has certain blindness,so it is likely to fall into local optimum.In this thesis,a mutual information feature selection technique is proposed for this problem.Firstly,the mutual information is used to evaluate the relevance and redundancy of traffic variables.Secondly,the feature selection is used to select the relevant variables and filter out the redundancy between the selected variables.Finally,the experimental results indicate that the proposed method reduces the cost of calculation and also avoid the blindness of the variable selection to a certain extent.Besides,forecasting precision is slightly improved.2.The BP neural network is sensitive to the initial value of the parameter,which is easy to fall into the local optimum.It's polarity excitation function is smoother and the approximation ability is poor.In this thesis,wavelet analysis theory and clonal selection algorithm are used to comprehensively improve the performance of BP neural network.The wavelet function has the characteristics of fast attenuation and multi-scale resolution,and it is suitable for dealing with the oscillation data.As the excitation function of BP neural network,it can improve the approaching ability and convergence speed of BP neural network.Clonal selection algorithm is a kind of global search technology,which has the characteristics of fast convergence and strong robustness.It is used to optimize the parameters of neural network to avoid the neural network falling into local optimum.The experimental results show that the proposed method has higher accuracy than the BP neural network and is also a feasible method.