Bagging-based neural network ensemble for load identification with parameter sensitivity considered

Abstract

Extensive installation of measuring devices in power systems promotes the application of the artificial intelligence (AI) in load identification. However, the convergence problems of training and the relatively low accuracy hinder the AI method from further development. In this study, a neural network ensemble method considering parameter sensitivity is proposed to solve these problems. In this method, with distributed generation considered, the parameters of the load model are classified according to the response uniqueness, and identified separately by multiple base learners according to its features. Additionally, ensemble algorithm is introduced for the higher accuracy, and Bagging strategy is used to ensure the diversity of learners through sampling the train set. Numerical simulations on a real power grid system validate the applicability of the proposed method in the field of parameter identification.

Keywords

• Parameter identification
• Ensemble learning
• Bagging
• Parameter sensitivity
• Generalized composite load model