Machine Learning-Based Restoration Forecast with Predictive Power Outage for Diverse Power Outage Scenarios

Efficiently restoring power outages (PO) is vital for ensuring uninterrupted electricity distribution, as it profoundly impacts economic development, infrastructure, businesses, and people’s well-being. Accurately predicting restoration timelines after a blackout is crucial for effective operational planning and power system resilience. This is especially important for developing countries where frequent and impactful blackouts, caused by technical faults in diverse equipment across multiple locations, often persist for extended durations. While previous studies have focused either on predictive power outage (PPO) or restoration forecasting separately, considering both aspects is essential for formulating resilience measures. Machine learning (ML) applications, including artificial intelligence (AI), have seen a notable rise in PO restoration forecasting. However, few studies have examined the combination of ML restoration forecasting and PPO for resilience planning. This study presents an integrated framework that combines ML restoration forecasting with PPO, leveraging solar self-sufficient sustainable energy resources (SSER). A comprehensive ML workflow preprocesses data and employs a neural network (NN) utilizing nonlinear autoregression with external inputs (NARX) to accurately forecast spatiotemporal solar features, facilitating efficient PO restoration. The framework’s validity is demonstrated using one year of historical solar weather and PO interruption data from a 33kV business hub (BH) operated by the Ibadan Electricity Distribution Company in Nigeria. Results indicate outstanding performance in forecasting spatiotemporal solar features for PO, with NARX regression R-values of 99.51%, 99.43%, and 99.48% during training, validation, and testing, respectively. The minute mean square errors (MSE) are remarkably low, reaching values as low as 7.8360e-06, confirming the high accuracy and reliability of the proposed ML approach.