论文部分内容阅读
冬季冰期水温的预报是冰情预报的基础。将基于自适应网络的模糊推理系统(ANFIS)和Levenberg-Marquardt算法改进的BP神经网络模型ANN应用到黄河冬季水温预报中,通过分析模型结构特性、水文数据及其相关预报因子的特点,确定模型合理的输入参数。在ANFIS和ANN模型输入因子和预见期相同的条件下,预报黄河最北端三湖河口、头道拐、巴彦高勒3个水文站冬季结冰期的水温。两种模型预报结果的优劣通过确定性系数、均方根误差和相关系数3种参数的比较进行评定。通过12组参数预报结果的比较和特性评定,自适应网络的模糊推理系统预报结果均比神经网络模型预报结果好。研究表明:基于自适应网络的模糊推理系统这一新的理论能够适合冬季结冰期水温预报的特点,预报精度得到普遍的提高。
Winter glacial water temperature forecast is the basis of ice forecast. The ANN ANN based on BP neural network model based on adaptive fuzzy inference network (ANFIS) and Levenberg-Marquardt algorithm is applied to forecast the winter water temperature in the Yellow River. By analyzing the characteristics of model structure, hydrological data and its related predictors, the model Reasonable input parameters. Under the same input factor and forecast period of ANFIS and ANN models, the water temperature of three hydrological stations in the northernmost part of the Yellow River during the winter icing period was predicted at Sanhuhekou, Toudaoguai and Bayannigal. The advantages and disadvantages of the two models’ forecasting results are compared by three kinds of parameters: deterministic coefficient, root mean square error and correlation coefficient. Through the comparison and characterization of 12 sets of parameter forecast results, the results of fuzzy inference system of adaptive network are better than those of neural network model. The research shows that the new theory of fuzzy inference system based on adaptive network can adapt to the characteristics of winter ice temperature forecast, and the prediction accuracy is generally improved.