Transparent maintenance alerts for rural electrification projects using Electroconvlstm -SHAP hybrids

Maintenance resources and challenges to remote monitoring contribute very much to challenges in building infrastructure reliability and sustainability of rural electrification projects and especially those based on decentralized solar and grid systems. In order to resolve such problems, this paper proposes a new and transparent predictive maintenance warning system based on a hybrid deep learning model known as ElectroConvLSTM. The model manages to capture the existing spatio-temporal dependencies, combining the CNNs and the LSTM networks, which demonstrates the early warning signs that correlate to the anomalies in performance and irregularities in usage. A feature selection step is proposed using the new Mutual Information-driven Golden Beetle Optimization (MIGBO), with the combination of Golden Jackal Optimization (GJO) and Dung Beetle Optimizer (DBO), as well as Mutual Information for . The proposed setting is aimed at increasing model accuracy, and interpretability. The step helps to make the prediction process more precise and triggers fewer warning events only the most relevant variables participate in the process. The suggested system can help run active, data-driven preservation in rural energy structures, presenting understandable and practical warning to parties. The experiments show outstanding results and it is shown that the MCC is 0.99257, the F1-score is 0.99333 and the accuracy is 0.99229 which indicate that this method is effective. The given solution is very helpful in increasing resource availability of operation in terms of reliability, transparency, and trust in rural electrification systems, which have to face constraints of resources.