Probabilistic Wind Power Forecasting using Adam-OBL Optimized DeepAR Model

Abstract

Accurate forecasting of wind power generation is crucial for effective planning and scheduling in modern power systems. This paper presents a probabilistic wind power forecasting framework based on the Deep Autoregressive (DeepAR) model whose hyperparameters are selected by a proposed Adam-Opposition Based Learning optimization algorithm (Adam-OBL). The proposed model is benchmarked against a modified Grasshopper Optimization Algorithm (MGOA) optimized DeepAR, a vanilla DeepAR, an equal-budget random search, and a classical probabilistic persistence baseline. These optimizer-driven approaches automatically find the best hyperparameters for tuning Deep Neural Networks (DNNs), preventing time-consuming and inefficient manual intervention. Unlike traditional deterministic approaches, the proposed method captures the inherent uncertainty and stochastic behavior of wind power generation by providing probabilistic forecasts. The methodology includes data pre-processing, model implementation, testing, and evaluation. Monte Carlo sampling improves accuracy, while validation-based conformal scaling calibrates prediction intervals. The results show the effectiveness of the Adam-OBL optimized DeepAR framework.