基于神经网络的SBS改性沥青材料结构性能预测

摘要

对SBS改性沥青混合料进行试验对比研究,通过MATLAB软件建立数学模型,以人工神经网络方法对SBS改性沥青混合料的结构性能进行预测,结果表明：在计算中,由于样本的数量对结果有一定的影响,存在一定误差,可以通过加大样本的数量对数学模型进行训练来获得更高的预测精度;为了使目标误差调小,可以尝试采用减小目标误差的方法,虽然加大预测样本的相对误差,但是可以使网络的泛化和推广能力上升,所以要选择能使训练样本和预测误差样本同时满意目标误差值作为训练参数。%The paper undertakes the research on the experimental comparison of the SBS modified asphalt mixture,establishes the mathematic model by MATLAB,forecasts the structural performance of SBS modified asphalt mixture by the manual nerve network method,proves by the results that some errors exist in the network model in the calculation,which is related to the number of samples,so the higher forecasting accuracy can be realized by better trainings on the mathematic models with increased samples,and the forecasting by the nerve network model can reduce the relative errors of the trained samples,so the aim errors can be smaller,but it can enlarge the relative errors in the forecasting samples,so the generalization of the network and its extension capacity can be moved upward,so the aim error which can meet the training and forecasting errors at the same time can be used as the training parameter.