Non-intrusive Household Load Identification based on Enhanced Random Forests Learning

Non-intrusive household load identification is an important way to achieve intelligent power consumption management. Random forests (RF), an ensemble learning method, has wide applicability and high robustness, which has been employed in the field of load identification. An enhanced random forests (ERF) algorithm is proposed in this work to solve the problem that the traditional random forest algorithm ignores the difference of ability of decision tree classification and the unfairness of voting. First, the Bayesian information criterion (BIC) is used to detect and identify switching events. Second, the best feature set is selected according to the time-frequency characteristics of the load information. Finally, the enhanced random forest algorithm is used to establish the load identification model. Experimental results show that ERF can achieve higher recognition accuracy of load categories than RF, and the accuracy can reach more than 90%.