Solar Power Production Prediction in Sweden

As solar power capacity in Sweden expands rapidly, accurate forecasting of aggregated power production becomes critical for maintaining grid stability and efficient energy trading. This thesis develops and evaluates predictive models for hourly national solar power production, with a particular focus on quantifying the impact of using forecast meteorological data rather than observed data. To capture the distinct temporal dynamics of solar energy, a "Two-Stage Temporal Hierarchical" modeling strategy was implemented. This approach decomposes the forecasting problem into a stable, long-term seasonal baseline modeled on daily data, and dynamic short-term deviations modeled on hourly data. Three distinct architectures were evaluated for the hourly component: Multiple Linear Regression (MLR), Extreme Gradient Boosting(XGBoost) and Seasonal Autoregressive Integrated Moving Average Exogenous (SARIMAX). The models were trained and tested using aggregated production data from Eurelectric and high-resolution meteorological data from the MetCoOp Ensemble Prediction System (MEPS), with national weather conditions computed as a population-weighted average of five major Swedish cities. The results demonstrate that the SARIMAX, time series model, achieves the highest predictive accuracy (R2 = 0.9132, RMSE = 0.1717 GWh, MAE = 0.0954 GWh), outperforming both the linear baseline (MLR) and the non-linear XGBoost model. A key finding of this study is that model performance remains remarkably robust when switching from observed to forecast weather data, with negligible degradation in accuracy, RMSE differences of less than 0.005 GWh. Feature importance analysis reveals that this stability is driven by the model’s heavy reliance on autoregressive features (lagged production values) and solar irradiance, which effectively buffer the predictions against meteorological forecast errors. These findings suggest that statistical time-series methods leveraging system inertia are highly effective for national-level solar power production prediction and that weather forecasts are of sufficient quality to support reliable grid planning operations in Sweden.