BQEB ForecastBench: Benchmarking AI Models for Smart Grid Forecasting Using BQEB-Data v1

The transformation of modern power systems through renewable integration, distributed battery storage, transportation electrification, and dynamic market participation has increased the need for accurate and reproducible forecasting models. However, comparisons across forecasting studies are often limited by inconsistent datasets and non-standardized evaluation protocols. This paper introduces BQEB ForecastBench, an open benchmarking framework built on the publicly available BQEB-Data v1 dataset for evaluating artificial intelligence models on smart-grid forecasting tasks. The benchmark investigates two operationally significant tasks: next-hour electricity load forecasting and day-ahead electricity price forecasting. Three baseline models were evaluated under a unified chronological train-validation-test protocol: Linear Regression, Random Forest, and Gradient Boosting. Performance was measured using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and coefficient of determination (R²). For load forecasting, Linear Regression achieved the best performance with RMSE of 9.94 MW and R² of 0.9288. For price forecasting, Linear Regression again produced the lowest RMSE of 4.39 USD/MWh with R² of 0.2078, highlighting the greater stochastic complexity of market prediction relative to demand forecasting. Ensemble methods also showed competitive performance across tasks. By providing transparent protocols, baseline results, and a reusable evaluation framework, BQEB ForecastBench establishes a reproducible foundation for future research in autonomous energy intelligence, smart-grid optimization, and trustworthy AI for power systems.