特征选择中的一类遗传序优化算法

针对模式识别中特征子集的选择存在组合优化的情况,采用了一类遗传序优化算法,以序优化思想为指导,对传统遗传算法中的参数及操作进行改进.利用有限的计算资源寻求复杂特征选择问题的近似最优解,从而有效地降低了特征维数,加快搜索到最优特征子集的速度,为特征选择问题提供了一种新的处理方法.通过Matlab仿真,对美国加州电力市场的实际数据进行预测.结果表明,月平均误差不超过2.62％,因此该算法在电力负荷预测方面有很强的实用性.

A class of genetic ordinal optimization for feature selection

For feature subset’s choice existence combination in the view of the pattern recognition,a new processing method of feature selection is presented to sequence the guidance of optimization of traditional genetic algorithm,the parameters and operation to improve,which uses the limited computing resource to seek the approximate optimal solution of the complex feature selection question,effectively reduce the feature dimension and speed up the search to the speed of the optimal feature subset.Through the simulation of Matlab to the United States,the real data of the California power market forecast,the results show that the average monthly error less than 2.62%,so the algorithm in power load forecasting has strong practicability.