基于权值优化神经网络的交通拥堵程度决策方法

Decision method of traffic congestion based on weight-optimized neutral network

针对现有的交通拥堵程度决策方法在证据不确定和不完备情况下评价准确率低的问题,提出了一种基于权值优化神经网络方法。首先,采用二次型隶属度描述了各类交通拥堵证据的不确定性与不完备性。其次,由信息熵计算证据的贡献度,并作为优化神经网络的输入层权值。接下来,由初始权值与前序时刻的梯度和自适应更新各隐层与输出层的神经元权值,以期降低不确定和不完备证据给整个网络带来的累积误差。最后,结合实际交通状况进行算例分析,验证了该方法的准确性与收敛性。结论分析表明,提出的方法能作出准确的交通拥堵程度决策。

To deal with the low accuracy of existing decision methods of traffic congestion under the condition of the uncertain and incomplete evidences,this paper proposed a weight-optimized neutral network.At first,this paper applied the quadratic-form membership function to present the characteristics of evidences.It achieved the evidence contributions by using the information entropy,which could be considered as the weights on input layer in the neutral network.Moreover,the weights on the hidden and output layers were adaptively computed on the basis of initial weight and previous gradient sum.It reduced the accumulated error from the uncertain and incomplete evidences.Finally,the numerical example proved both accuracy and convergence of the proposed method in terms of actual traffic environment.The results indicate that the proposed method can make the stable decision on traffic congestion.