Agentic Workflow Using RBA$_θ$ for Event Prediction

TL;DR

This paper reframes wind-power forecasting as an event-centric, frequency-aware problem, arguing that ramp events—defined by onset, magnitude, duration, and direction—drive operational risk more than pointwise trajectory accuracy. It develops a multi-architecture framework built on enhanced RBA$_\theta$ event semantics, wavelet-based multi-resolution decomposition, and Hawkes causal priors, enabling direct event prediction, sequence modelling, and physically coherent trajectory reconstruction. An agentic orchestration layer selects among four complementary workflows (trajectory-first, direct-event, event-aware LSTM, and event-aware Transformer) using context-aware utilities and experience replay, which improves robustness and cross-site transfer. Results show ramp dynamics concentrate in mid-frequency bands (notably $D_3$), reconstruction achieves high fidelity (e.g., $R^2$ ≈ 0.90) while maintaining operational relevance, and zero-shot transfer across wind farms remains feasible with strong event-forecast performance. Overall, the event-first paradigm with frequency awareness and adaptive workflow selection offers a transferable, operator-aligned alternative to traditional trajectory-centric wind-power prediction, with practical implications for reliability, cost, and grid stability.

Abstract

Wind power ramp events are difficult to forecast due to strong variability, multi-scale dynamics, and site-specific meteorological effects. This paper proposes an event-first, frequency-aware forecasting paradigm that directly predicts ramp events and reconstructs the power trajectory thereafter, rather than inferring events from dense forecasts. The framework is built on an enhanced Ramping Behaviour Analysis (RBA$_θ$) method's event representation and progressively integrates statistical, machine-learning, and deep-learning models. Traditional forecasting models with post-hoc event extraction provides a strong interpretable baseline but exhibits limited generalisation across sites. Direct event prediction using Random Forests improves robustness over survival-based formulations, motivating fully event-aware modelling. To capture the multi-scale nature of wind ramps, we introduce an event-first deep architecture that integrates wavelet-based frequency decomposition, temporal excitation features, and adaptive feature selection. The resulting sequence models enable stable long-horizon event prediction, physically consistent trajectory reconstruction, and zero-shot transfer to previously unseen wind farms. Empirical analysis shows that ramp magnitude and duration are governed by distinct mid-frequency bands, allowing accurate signal reconstruction from sparse event forecasts. An agentic forecasting layer is proposed, in which specialised workflows are selected dynamically based on operational context. Together, the framework demonstrates that event-first, frequency-aware forecasting provides a transferable and operationally aligned alternative to trajectory-first wind-power prediction.

Figures (8)

• Figure 1: Conceptual comparison between the traditional trajectory-first forecasting pipeline and the proposed event-centric paradigm. The classical approach forecasts the full time series and detects events afterward, whereas the proposed system predicts event semantics first and reconstructs the trajectory, improving interpretability, transferability, and operational relevance
• Figure 2: Wavelet-based multi-band decomposition of the input signal, showing approximation ($cA_i$) and detail ($cD_i$) components across scales wang_wavelet
• Figure 3: Conceptual overview of the agentic forecasting system developed where $\{W_x\}$ are designed workflows
• Figure 4: Full pipeline of Approach 3
• Figure 5: DWT decomposition of the wind-power time series into approximation and detail bands.