基于神经网络和粒子群算法产流量模型构建

为了研究坡面产流量与各影响因子间定量关系,分析野外多尺度人工径流小区实测数据,采用人工神经网络及粒子群算法,建立了坡面产流量预测模型,产流量与坡长、坡宽、坡度、前期土壤含水量间可采用二次抛物线关系进行描述,与植被覆盖度、降雨量间分别采用幂函数和线性函数进行描述.另外采用加权相对差距和法确定了产流量BP神经网络模型最优拓扑结构及网络参数,建立了产流量BP神经网络模型,该模型模拟值与实测相对误差在±20%以内,预测精度较高.同时基于产流量与各单因子定量关系,建立了产流量经验模型,采用粒子群算法推求了模型未知参数,该经验模型相对误差主要在±30%以内,其精度略逊于BP神经网络模型.

Study on runoff production model by artificial neural network and particle swarm algorithm

Two runoff production models are built by using artificial neural network and particle swarm algorithm, respectively, to study the relationships between runoff production and related factors based on existing multiple scale field data. The results showed that the relationships between runoff production and slope length, width, gradient and antecedent soil water are parabolic, but those between runoff production and vegetation cover and rainfall can be described with power and linear functions, respectively. The optimum topological structure and the network parameters of the BP model for predicating runoff production are determined by means of weighted summation of relative difference, and the relative error between the model estimate and observed data is within ?20%, showing a good prediction accuracy. According to the quantitative relationship between runoff production and each factor, the empirical model of runoff production is established, and the particle swarm algorithm is used to determine model parameters. Its relative error of the empirical model is ?30%, slightly poorer than the BP model.