Improving Power Grid Disaster Resilience through Predictive Analytics Based on Data-Driven Photovoltaic Power Plant and Distribution Network Power Forecasting

The resilience of power grids in disaster-prone areas is a critical and growing concern for maintaining stable, secure, and reliable electricity distribution. Recent advancements in data-driven forecasting methods, especially in the context of photovoltaic (PV) power plants and distribution networks, offer strong potential to enhance grid operations during extreme weather or seismic events. Traditional power grid forecasting methods often rely on rule-based models and limited historical data. These approaches struggle to capture the complex, nonlinear, and dynamic interactions between renewable energy sources—such as PV systems—and grid behavior under stress or failure. As a result, they frequently face challenges in scalability, accuracy, and real-time adaptability during high-impact disaster events. To overcome these limitations, this paper presents a new predictive analytics framework. It uses advanced machine learning techniques to improve power grid resilience. The framework focuses on accurately forecasting both the power output from distributed PV plants and the variable load of the distribution network. The proposed method integrates real-time data from multiple heterogeneous sources, including weather forecasts, sensor networks, and grid telemetry. It applies algorithms such as decision trees, deep neural networks, and ensemble learning to extract meaningful patterns. By combining extensive historical data with real-time updates, the models produce robust, adaptive, and precise power predictions. Experimental results show that the data-driven approach significantly outperforms traditional forecasting techniques, with reductions in Mean Absolute Error (MAE) by up to 40% and improvements in R 2 by over 25% across multiple datasets. While these gains may appear modest, they are practically meaningful in the context of disaster-resilient grid forecasting, where even small reductions in prediction error can lead to more efficient resource allocation, better load balancing, and enhanced disaster response. Compared to traditional models like ARIMA and Prophet, which are still widely used in resource-constrained environments, the proposed approach offers higher predictive accuracy, lower latency, and faster adaptation to unforeseen grid failures. These improvements are critical for enhancing disaster preparedness and response capabilities, demonstrating the practical relevance of this method for power grid resilience in extreme conditions.