摘要
传统的数据处理群方法(Group method of data handling,GMDH)在结构上具有自组织和全局选优的特性,非常适合进行非线性数据的拟合。但由于在传统GMDH网络建模是用最小二乘法来辨识参数,常常使得模型预测效果不理想。遗传算法是一种有效的搜索和优化方法,它具有自适应搜索、渐进式寻优、并行式搜索、通用性强等特点,论文将遗传算法引入GMDH网络,用遗传算法辨识部分描述式的系数,建立了基于遗传算法的GMDH网络模型。并将该模型应用于一组实测时间序列的预测研究,计算机仿真结果表明,模型预测效果令人满意。
Traditional GMDH network has self-organization and overall selection features structurally, thus it is very suitable for the forecasting of nonlinear data. Since traditional GMDH network modeling uses least square method to identify parameters, model prediction results are unsatisfactory. The genetic algorithm is an effective searching and optimized method with the characteristics of adaptive search, gradual optimization, parallel search, and strong general features. The genetic algorithm is introduced into GMDH network to identify the coefficients of some descriptive formulae and a GMDH network model is established based on genetie algorithm. The model is used to forecast a group of measured time series. Simulation results show that the predicting result of the model is satisfactory.
出处
《数据采集与处理》
CSCD
北大核心
2009年第6期820-824,共5页
Journal of Data Acquisition and Processing
基金
教育部博士点基金(20060286005)资助项目
江苏省高校自然科学基金(07KJD580085)资助项目
