| 摘要: |
| 【目的】建立一种对资料要求较少、精确度较高的非点源污染负荷多变量灰色神经网络预测模型，为有限资料条件下非点源污染负荷的预测提供支持。【方法】针对GM(1，N)模型在原始数据变化幅度较大且趋势不明显时预测效果较差的不足，提出并建立用人工神经网络对GM(1，N)模型的残差系列进行修正的改进模型，并将其应用于华县站总氮非点源污染负荷的预测。【结果】在华县站总氮非点源污染负荷预测中，灰色+BP神经网络组合模型拟合预测效果较好，建模阶段和检验阶段的确定性系数（Nash-suttcliffe模拟效率系数）分别为1.00和0.93，优于单独灰色模型或神经网络模型的预测效果。【结论】研究建立的多变量灰色神经网络模型综合了灰色理论和神经网络的优点，提高了模拟精度，为有限资料条件下非点源污染负荷的预测提供了一种有效的方法。 |
| 关键词: 非点源污染 负荷预测 多变量 GM(1,N)模型 BP神经网络 RBF神经网络 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金项目(50909080,50979090)；国家水体污染控制与治理科技重大专项河流主题“渭河水污染防治专项技术研究与示范课题”(2009ZX07212-002-005-02) |
|
| Multi-variable grey-neural network model for non-point source pollution load prediction |
|
LI Jiake1, LI Yajiao2, LI Huaien,et al3
|
|
1.Key Lab of Northwest Water Resource and Environment Ecology of Ministry of Education ,Xi’an University of Technology;2.School of Architecture and Civil Engineering ,Xi’an University of Science and Technology;3.Key Lab of Northwest Water Resource and Environment Ecology of Ministry of Education , Xi’an University of Technology
|
| Abstract: |
| 【Objective】This paper is to establish a multivariable grey neural network model for non-point source pollution load prediction,which needs less data input and the accuracy is higher.It can provide support for non-point source pollution load prediction under limited data condition.【Method】GM(1,N) model is limited to forecast when the input data is various and random.Artificial neural network model is used to adjust errors of GM(1,N) and the modified model is presented in this paper.The modified model is used to predict non-point source pollution (NSP) load.【Result】The prediction effect of the combined model of the grey and the BP neural network is pretty good,and the deterministic coefficients of fitting stage and testing stage are 1.00 and 0.93 respectively.The effect is better than that of the grey model or the neural network model.【Conclusion】The established model combines the virtues of the grey theory and neural network,improves the prediction accuracy of the model and can be used to predict the NSP load under limited data condition. |
| Key words: non-point source pollution load prediction multi-variable GM(1,N) model BP neural network RBF neural network |
