A study on black screen fault detection of single-phase smart energy meter based on random forest binary classifier

The intricacies surrounding the induction of black screen faults in smart watt-hour meters pose a significant challenge, particularly given the vast amount of data training samples and the absence of a robust decision correlation process. To address this, a real-time detection method for black screen faults in single-phase intelligent watt-hour meters is proposed, utilizing a random forest two-classifier approach. This method identifies the black screen fault feature by calculating the entropy value of the relevant attributes and selecting those with higher entropy. The attribute data is then discretized using an equidistant division method. A training sample set for black screen fault detection is constructed based on the bagging method, and decision trees are subsequently built and tailored to the identified fault characteristics. Each tree selects the optimal attribute for branching, which ultimately leads to the development of a second random forest classifier. This classifier employs a majority voting method as its decision rule, effectively delivering black screen fault detection results for single-phase smart meters. Notably, the experimental outcomes demonstrate the accuracy of this approach, as it achieved a remarkably low error rate of only 15 out of 1800 black screen fault test samples.