A Demand-Side Load Event Detection Algorithm Based on Wide-Deep Neural Networks and Randomized Sparse Backpropagation

Event detection is an important application in demand-side management. Precise event algorithms can improve the accuracy of non-intrusive load monitoring (NILM) and energy disaggregation models. Existing be divided into four categories: rule-based, statistics-based, conventional machine learning, deep learning. The rule-based approach entails hand-crafted feature engineering carefully calibrated thresholds; accuracies statistics-based learning methods are inferior to due their limited ability extract complex features. Deep models require a long training time hard interpret. This paper proposes novel algorithm for smart homes based on wide that combines convolutional neural network (CNN) soft-max regression (SMR). model extracts power series patterns utilizes percentile information series. A randomized sparse backpropagation (RSB) weight filters proposed robustness standard wide-deep model. Compared wide-deep, pure CNN, SMR models, hybrid powered by RSB demonstrates its superiority terms accuracy, convergence speed, robustness.