福建春季旱涝灾害BP人工神经网络预测模型

A BP artificial neural network prediction model of spring's drought/flood in Fujian

以动量BP人工神经网络模型对福建3-4月的春季旱涝进行了预测试验。试验结果表明:若前期冷空气偏强,会导致来年春季降水偏多,3个主分量的各因子均达到显著性水平;尽管福建春季降雨量具有明显的非线性演变特征,但从试报的5年来看,神经网络的预测评分可高达90分以上,逐步回归和最优子集最高不超过90分。因此,所提BP神经网络预报模型对福建春季降雨量有较好的预报效果,可供有关部门制定防灾减灾决策参考。

By using BP artificial neural network model, the drought/flood in Fujian Province in spring was predicted experimentally. The result shows that if the cold air is on the strong side in early stage, the spring＇s precipitation will be abundant in next year and the factors of 3 principa components all reach to significant level. In spite of nonlinear change characteristics of the precipitation in Fujian, the prediction score of the BP method can be over 90 for late 5 years which other two methods can not reach. This result means that BP method has good effect than others and spring rainfall in Fujian has an obvious non-linear evolution characteristic.