Mismatch analysis of rooftop photovoltaics supply and farmhouse load: Data dimensionality reduction and explicable load pattern mining via hybrid deep learning

Establishing a new type of electricity system based on rooftop photovoltaics (PV) can facilitate the energy transition in rural China. Research on the mismatch between the PV supply and rural household demand is vital to the widespread adoption of PV microgrid systems. Currently, typical load patterns (TLPs) in rural areas lack accurate characterization and mismatch assessment methods disregard PV curtailment. Therefore, this study proposes a hybrid deep learning-based analytical framework to quantify short-term mismatches between PV power generation and TLPs throughout the day and applies it to a real rural dataset. This study employs the variational autoencoder (VAE) model for dimensionality reduction and feature extraction of high-resolution load data and compares it with traditional methods. In addition, we employed the k-medoids method to uncover TLPs and utilized decision trees to enhance interpretability. The results show that (1) The VAE model exhibits superior dimensionality reduction and feature extraction capabilities on both public and measured datasets and compared to other models, it can reconstruct peak loads more effectively. (2) Three types of TLPs were identified within the rural dataset, with the outdoor average daily wet-bulb temperature being the major influencing factor. (3) Significant differences existed in the mismatch levels between the three types of TLPs and PV power generation. The Lorenz curves and Gini coefficients can effectively quantify the mismatch between PV power generation and TLPs. The proposed framework provides theoretical support for optimizing PV microgrid systems design in rural areas and developing demand-side response strategies.