Probabilistic Multi-Regional Solar Power Forecasting with Any-Quantile Recurrent Neural Networks

TL;DR

This work addresses the uncertainty inherent in solar power by introducing an any-quantile probabilistic forecasting framework for multi-regional PV using the AQ-RNN. The model combines a dual-track RNN with dilated cells, patch-based temporal modeling, and dynamic team ensembles to produce calibrated quantiles at arbitrary levels within a single trained network, while leveraging cross-regional context to enhance system-level robustness. Empirical evaluation on 30 years of hourly PV data across 259 European regions demonstrates consistent improvements in CRPS, calibration, and prediction interval quality over established baselines, with ensemble variants achieving further gains. The approach offers practical benefits for risk-aware energy management and operational decision-making in renewables-dominated grids, without reliance on exogenous weather data in the evaluation setting, and points to promising extensions involving weather inputs and finer spatial resolutions.

Abstract

The increasing penetration of photovoltaic (PV) generation introduces significant uncertainty into power system operation, necessitating forecasting approaches that extend beyond deterministic point predictions. This paper proposes an any-quantile probabilistic forecasting framework for multi-regional PV power generation based on the Any-Quantile Recurrent Neural Network (AQ-RNN). The model integrates an any-quantile forecasting paradigm with a dual-track recurrent architecture that jointly processes series-specific and cross-regional contextual information, supported by dilated recurrent cells, patch-based temporal modeling, and a dynamic ensemble mechanism. The proposed framework enables the estimation of calibrated conditional quantiles at arbitrary probability levels within a single trained model and effectively exploits spatial dependencies to enhance robustness at the system level. The approach is evaluated using 30 years of hourly PV generation data from 259 European regions and compared against established statistical and neural probabilistic baselines. The results demonstrate consistent improvements in forecast accuracy, calibration, and prediction interval quality, underscoring the suitability of the proposed method for uncertainty-aware energy management and operational decision-making in renewable-dominated power systems.

Figures (10)

• Figure 1: Time-series segmentation for constructing input and output sequences.
• Figure 2: Block diagram of the proposed AQ-RNN forecasting model.
• Figure 3: Illustration of the dilated recurrent cell (dRNNCell).
• Figure 4: Primary and context RNN architecture. Blue-highlighted inputs and connections are specific to the primary RNN and are absent in the context RNN.
• Figure 5: Results of the Diebold-Mariano test: Each entry indicates the number of regions (out of 259) for which the model on the y-axis exhibits a significantly lower CRPS than the model on the x-axis, at the significance level $\alpha = 0.05$