Model Estimation for Solar Generation Forecasting using Cloud Cover Data

TL;DR

This work tackles PV power forecasting when site measurements of irradiance and temperature are unavailable, by leveraging cloud cover data and power measurements. It introduces three parametric models grounded in the PVUSA framework and Cloud Cover Factor, with two nonlinear EKF-based estimators (N5/N6) and a linear overparameterized (L) model, all estimated online via EKF/RLS. Across simulated and experimental data, the approach delivers accurate day-ahead and hour-ahead forecasts, outperforming naive predictors and several neural network baselines, and remains computationally efficient for large-scale deployment. The results support scalable, regionally aggregated forecasting for grid operation, with online adaptation capturing seasonal variations in cloudiness effects.

Abstract

This paper presents a parametric model approach to address the problem of photovoltaic generation forecasting in a scenario where measurements of meteorological variables, i.e., solar irradiance and temperature, are not available at the plant site. This scenario is relevant to electricity network operation, when a large number of PV plants are deployed in the grid. The proposed method makes use of raw cloud cover data provided by a meteorological service combined with power generation measurements, and is particularly suitable in PV plant integration on a large-scale basis, due to low model complexity and computational efficiency. An extensive validation is performed using both simulated and real data.

Figures (15)

• Figure 1: Qualitative behavior of the CCF $C(N)$ for $\mu_4>0$ (left) and $\mu_4<0$ (right).
• Figure 2: Day-ahead forecasting. Note that ${\cal K}_{d+1}$ does not cover a complete day period since $\overline{k}_{d+1}$ refers to the last sample pertaining to light hours in day $d+1$.
• Figure 3: Hour-ahead forecasting
• Figure 4: N5 model parameter estimation vs. time (iterations), scenario (i). Black dashed lines are the nominal values of each parameter as reported in \ref{['EQ:PAR_RV']}.
• Figure 5: N5 model parameter estimation (blue line) vs. time (iterations) and using data from scenario (ii) to scenario (v). Black dashed lines are the nominal values of each parameter as reported in \ref{['EQ:PAR_RV']}.