摘要: |
【目的】建立地下水位预测的正则化RBF网络模型,为区域地下水资源的利用、规划和管理提供决策依据。【方法】以MATLAB7.0为平台,用函数newrb创建正则化RBF网络模型,基于宝鸡峡灌区B210号观测井1983-2009年的地下水位埋深资料,对网络模型进行训练后再用测试集检验,分别绘制训练集与测试集的拟合曲线,同时计算实测值与预测值间的相对误差(RE)、平均绝对偏差(MAD)和均方误差(MSE),并将其与BP网络模型的相应值进行对比。【结果】正则化RBF网络模型和BP网络模型的相对误差均小于5%,平均绝对偏差分别为0.53和0.85,均方误差分别为0.54和1.15,相比之下,正则化RBF网络模型的预测精度更高。【结论】训练样本和测试样本的合理选取为时间序列的拟合扩展了思路,良好的泛化能力使正则化RBF网络模型在区域地下水位预测中具有一定的可行性。 |
关键词: 正则化 RBF网络模型 径向基函数 地下水 水位预测 |
DOI: |
分类号: |
基金项目:国家科技支撑计划项目(2006BAD11B05);国家自然科学基金项目(50879071) |
|
Application of regularized RBF network model in the groundwater level prediction |
|
Abstract: |
【Objective】Establishing regularized RBF network model for groundwater level prediction can provide strategic decision for groundwater use,planning and management.【Method】Regularized RBF network model was built employing newrb function in MATLAB7.0 for well B210 in Baojixia irrigation area based on the groundwater level depth data from 1983 to 2009.The training sets and testing sets were used to train and test the network respectively.Corresponding fitting curve was plotted as well.Meanwhile,relative error(RE),mean absolute deviation (MAD) and mean square error (MSE) between predicted and measured values were all calculated and the comparison was addressed with BP network model.【Result】RE of both regularized RBF and BP network model is less than 5%,MAD is 0.53 and 0.85,MSE is 0.54 and 1.15 respectively.By contrast,the precision of regularized RBF network model about predicted values is much higher.【Conclusion】Selecting training sample and testing sample reasonably has provided a new way for time series simulation.Regularized RBF network model is viable in forecasting regional groundwater table due to its good generalization. |
Key words: regularization RBF network model redial basis funcion groundwater water level forecast |