基于IA-PNN的电能质量复合扰动特征选择及参数优化

Power Quality Complex Disturbance Feature Selection and Parameter Optimization Based on IA-PNN

提出了一种基于S变换(ST)和免疫算法(IA)优化的概率神经网络(PNN)算法的电能质量复合扰动特征选择和参数优化混合方法。首先,引入了线型判别分析(LDA),将其最优方向作为特征权重对基于ST提取的原始特征进行降序重排,通过将特征选择过程线性化以方便进行优化;其次,基于特征选择和PNN窗宽调节因子的优化需要,改进了IA的亲和度计算方法;最后,使用改进后的IA进行PQD信号特征选择和PNN参数混合优化,并依据结果重构了IA-PNN分类器,并对含随机噪声PQD信号进行分类。试验结果证明,相比原PNN分类器,优化后的IA-PNN的分类精度得到了有效提高,同时总运行时间得到了降低,与DT和KNN相比,IA-PNN在性能上均有一定优势,体现出了新方法的有效性。

This paper proposes a complex power quality disturbance(PQD) feature selection and parameter optimization hybrid method based on S-transform(ST) and immune algorithm(IA) optimized probabilistic neural network(PNN) algorithm. Firstly, linear discriminant analysis(LDA) is introduced, and the optimal direction is used as the feature weight to descend the original features based on ST in order to linearize the feature selection process. Secondly, based on the need of feature selection and window width adjustment factor optimization of PNN, the affinity calculation of IA is improved. Finally, the IA algorithm is used to optimize the feature selection and PNN parameters of PQD signals. The IA-PNN is reconstructed according to the optimization result, and PQD signals with random 50-20 dB noise are classified. Experiments show that compared with the original PNN classifier, the classification accuracy of the optimized IA-PNN is effectively improved, and the total running time is reduced. Compared with DT and KNN, IA-PNN has advantages in performance, reflecting the effectiveness of the new method.