中文摘要：

针对BP神经网络收敛速度慢和易于陷入局部极值的缺陷,提出以双极性sigmoid函数作为隐层节点的激活函数的洪水灾害评价前向神经网络模型。采用对洪水灾害指标值进行规范变换,使规范变换后的各指标皆＂等效＂于同一个规范指标;构建并优化得出适用于洪水灾害评价的任意2个指标规范值的前向神经网络模型（NV-FNN（2-2-1））和任意3个指标规范值的前向神经网络模型（NV-FNN（3-2-1））;对洪水灾害指标较多的前向神经网络建模,则可通过将多指标的前向神经网络模型分解为以上两种简单模型的组合表示。模型用于中国45个洪水灾情案例分析,其评价结果与实况相符合,验证了模型的有效性。与其他方法的评价结果比较表明：该模型由于不受指标数多少的限制,实用范围广泛。

英文摘要：

For the defects of slow convergence and falling into local optimum of BP neural network,a feed forward neural network model was developed for evaluating the flood disasters. This model used the bi-sigmoid function as the activation function of hidden nodes. This study set the normalized transformations for each index value of flood disaster in order to make the all normalized indexes be equivalent to a certain normalized index. Furthermore,two optimized types of feed forward neural networks models were built： the NV-FNN（ 2-2-1）,which was suitable for the case involved any 2 normalized index values of the flood disaster evaluation,and the NV-FNN（ 3-2-1）,which was suitable for the case involved 3normalized index values of the flood disaster evaluation. For the potential case with more indicators of flood disasters,it can be divided into several NV-FNN（ 2-2-1） and（ or） NV-FNN（ 3-2-1） models. The models were used to analyze 45 cases of flood disaster in China,the evaluation results,which were consistent with the actual situations,proved the validity of the models. Comparisons of evaluation results with other methods showed that this models had a wide range of applications because of no restriction on the number of indicators.