摘要
针对传统GMDH网络建模用最小二乘法辨识参数时容易陷入局部极小导致模型预测效果不理想的问题,提出将模拟退火算法与遗传算法结合起来,并引入到GMDH网络,用模拟退火遗传算法来辨识其部分描述式系数.描述了模拟退火遗传算法,构建了基于该算法的GMDH网络模型,并将该模型应用于泥石流预测的仿真研究,预测平均相对误差达到3.54%.结果表明,该算法既保证了全局寻优又防止了过早收敛,进一步提高了GMDH网络模型的全局与局部寻优能力.
According to the traditional GMDH network modeling with the least square method to recognize parameters, it＇s easy to fall into local minimum, and with the result that the prediction effect is not ideal. This paper puts forward to combine the simulated annealing algorithm and genetic algorithm, and introduces the combined algorithm to the GMDH network which is used to identify some of its description type coefficient. In this paper, it describes the simulated annealing genetic algorithm, and constructs the GMDH network model based on this algorithm, and the model is applied to the simulation of debris flow prediction research, forecast average relative error reached 3.54%. The results show that the algorithm not only ensuring the global optimization but also preventing premature convergence, improve the GMDH network model of global and local searching optimal ability further.
出处
《华中师范大学学报（自然科学版）》
CAS
北大核心
2013年第2期162-166,共5页
Journal of Central China Normal University：Natural Sciences
基金
国家自然科学基金项目(61179064)
关键词
模拟退火算法
遗传算法
自组织
GMDH
预测
simulated annealing algorithm
genetic algorithm
self-organization
The Group Method of Data Handling
prediction
