DeepANFIS: An Interpretable and Uncertainty-Aware Neuro-Fuzzy Framework for Energy Disaggregation in Smart Microgrids

Abstract

Abstract Non-Intrusive Load Monitoring (NILM) facilitates appliance-level energy disambiguation based on aggregate power consumption measurements and is an important facilitator of energy efficiency and demand-side management. Moreover, state-of-the-art NILM methods tend to have less interpretability, sensitivity to the sampling rate, and difficulty in providing reliable uncertainty information, which hinders their effectiveness in practical implementations. To address these issues, this work presents a new Deep Adaptive Neuro-Fuzzy Inference System called DeepANFIS for joint appliance state identification and power forecasting with enhanced interpretability and uncertainty awareness. The proposed system comprehensively incorporates modular convolutional feature learning with multiple resolutions, fuzzy memory networks with temporal dependencies, hierarchical fuzzy inference, and evidential deep learning. Extensive experiments conducted on both UK-DALE and REDD datasets illustrate the efficacy of DeepANFIS under high-frequency and low-frequency schemes. With a sampling frequency of 6 seconds on UK-DALE, DeepANFIS attains an F1-score of 0.7122 with an MAE of 25.63 W, and with an F1-score of 0.8371 with an RMSE of 78.03 W with a sampling frequency of 15 minutes. On REDD, DeepANFIS reports a nearly perfect F1-score of 0.9869 with an accuracy of 0.9992 and a low MAE of 2.56 W. As part of cross-dataset analysis, evaluation on other datasets shows excellent F1-score classification capability with a mere decrement of 2.4%, but power estimation transfer remains a challenge. Ablative analysis further validates the significance of both temporal memory and fuzzy processing in DeepANFIS. The efficacy of DeepANFIS is thus established with respect to a non-interpretable, uncertainty-conscious model for NILM.