Variable weight PSO-Elman neural network based road adhesion coefficient estimation
The variable weight particle swarm optimization(PSO)-Elman neural network was proposed for road adhesion coefficient estimation,in order to address the issue that the unstable weight update of traditional neural network leads to the poor accuracy in the estimation of road adhesion coefficient.The neural network model was constructed on the basis of a seven degrees-of-freedom vehicle dynamic model.The particle swarm algorithm was applied in the Elman neural network model to reduce the training absolute error.The linear decreasing weight strategy was used to change the weight of the particle swarm algorithm,which was useful for balancing the particle's global and local search ability.Thus,the optimization of network weight arrays were realized.Then,the correlation curve of optimal slip ratio and rood adhesion coefficient was fitted by the Fourier approximation method.Theoretical analysis and simulation verification proved that this method can improve the estimation accuracy of road adhesion coefficient.Simulation results showed that,under both fixed and unfixed adhesion coefficient roads,the root-mean-square error of the road adhesion coefficient obtained by the proposed variable weight PSO-Elman neural network estimation method was reduced by 35.62%and 19.20%on average compared with that of the traditional Elman neural network.Furthermore,the anti-lock control effect was also effectively improved.
vehicle stateroad adhesion coefficientneural networkparticle swarmanti-lock control
NETL
NSTL
万方数据
维普
