水库控制流域汛期分期的有效聚类分析

Effective fuzzy cluster method for dividing reservoir flood season

受诸多因素影响的水库汛期分期其实质是一个聚类数目未确知的高维时间序列聚类问题,它要求聚类方法具有能处理高维和时序性数据,且能同时回答聚为几类最为合理的能力。鉴于目前常规聚类方法不同时具备这些能力,在模糊C-均值聚类和紧密与分离聚类有效函数的基础上,提出了能够处理高维时序聚类问题的动态模糊C-均值聚类分析方法和相应的时序聚类有效性函数,耦合二者建立了适用于汛期分期的有效模糊聚类分析方法,提出采用实码加速遗传算法优化求解,克服了模糊C-均值聚类方法常规迭代优化求解对初值敏感的困难,并给出了完备的建模步骤和模型的合理性检验。最后,将模型应用于滦河流域潘家口水库汛期分期中,得出了合理的结论。

The division of reservoir flood season is a multi-dimension time series cluster question of the unknown number of cluster, so the cluster methods are in demanded to hold the capacities of dealing with multi-dimension, time series and duster validity. But no one can meet all these needs. Therefore, based on the fuzzy C-means duster method and the compact and separate duster validity function, a dynamic fuzzy C-means clustering method and corresponding cluster validity function based time are put forward, and a set of self-contained modeling steps is taken coupling the two methods above. There is a complex nonlinear optimal question in the model, and the regular optimal method is very difficult to solve it. By using the real coding based accelerating genetic algorithm to solve it is very effective and can overcome the initialization value sensitivity difficulty from the regular iterative optimal method used for the fuzzy C-means duster. Its test and application results show that it is very effective.