基于BP神经网络的公路风吹雪雪深预测模型

Prediction model of snow depth of snowdrift on highway based on Back Propagation Neural Network

摘要:

公路风吹雪雪深预测作为国际雪冰学领域的研究热点和难点问题，一直未能很好解决。以白茫雪山防雪走廊段安装的自动气象站和当地气象局提供的气象资料为基础，提取了对公路风吹雪雪深有影响的4种因素（降雪量、大气温度、风速和湿度）的指标值，建立了基于Back Propagation Neural Network的公路风吹雪雪深预测模型。对研究区5次降雪过程中所记录的199组数据进行训练学习，用20组数据来验证建立的模型，验证结果表明20h累计雪深预测值的误差在10%以内，85%的雪深预测值误差在20%以内，因此所建立的模型具有很强的泛化能力和较高的精度。并对降雪量、大气温度、风速和湿度这4个因素进行了敏感性分析，表明雪深与降雪量成正比，与其他3个因素成反比，其中降雪量对雪深的影响最大，风速次之，湿度最小。

Abstract:

As the research focus in international snow and ice field, snow-depth prediction of snowdrift on highway still has not been well solved. Based on meteorological data provided by automatic weather stations installed along the anti snow corridor on White Snow Mountain and meteorological bureau, index values of four factors (snowfall, air temperature, wind speed and humidity) which have influence on snow depth of snowdrift on highway are extracted and prediction model of snow depth of snowdrift on highway based on BP Neural Network is established. 199 sets of data during five snowfall in study area are used to train network and establish model, then use 20 sets of data to validate the model. Validation results show relative error of accumulated snow-depth predictions in 20 hours is less than 10% and 85% of relative error of snow-depth predictions is less than 20%. Therefore, the model has strong generalization ability and high accuracy. Sensitivity analysis of snowfall, air temperature, wind speed and humidity indicates that snow depth is directly proportional to snowfall and inversely proportional to other three factors, wherein snowfall has the greatest impact on snowdepth, followed by wind speed, humidity minimum.