Full text
GenEneCast Forecast Report Forecast period: October 2024 to March 2025 Monthly Forecast vs Previous Year In the upcoming months, Rio de Janeiro is expected to see varied shifts in energy consumption compared to the corresponding periods last year. Starting with October 2024, there is a significant forecasted decrease in energy use by 31.6%, followed by November and December, which also show considerable reductions of 24.5% and 26.0%, respectively. However, a turnaround is anticipated in January 2025, where energy usage is predicted to increase by 6.5% over the previous year. This rise is short-lived as February and March 2025 are both expected to see decreases again, specifically by 10.5% and 2.8%, respectively. This pattern indicates a fluctuating energy demand as the year transitions from 2024 into 2025. The local characteristics of Rio de Janeiro, such as its tropical climate and the pattern of urban energy consumption, play a crucial role in understanding these forecasted changes. The significant decreases in energy usage towards the end of 2024 could be attributed to milder weather conditions reducing the need for cooling systems, which are typically a major energy drain in this warm region. The slight increase in January 2025 might be linked to the height of summer, when temperatures peak and the demand for air conditioning surges, especially in densely populated urban areas. The subsequent decreases in February and March could be associated with the end of the hottest period and the beginning of milder autumn conditions. Additionally, seasonal variations in tourism may impact energy consumption, with peaks likely during the warmer months when tourists flock to the city, and declines as visitor
numbers drop. The urban dynamics, including population density and the concentration of commercial activities, significantly influence the overall energy demand patterns observed in the region. Forecast Error Interpretation The LSTM component of the GenEneCast model has produced a set of residual corrections that vary considerably, ranging from a decrease of about 82 units to an increase of approximately 155 units. This variation suggests that while the model can at times significantly adjust its predictions to be closer to actual values, it can also deviate quite a bit. The model's validation performance, with a Root Mean Square Error (RMSE) of 0.96 and a Mean Absolute Error (MAE) of 0.77, indicates a generally high level of accuracy in its forecasts. These low error values imply that the model is reliable and can be trusted in making decisions based on its predictions, as it typically forecasts energy consumption with minimal deviation from the actual observed amounts. This reliability is crucial for decisionmakers who rely on precise forecasts for planning and operational efficiency. Model Summary GenEneCast Configuration Level: Medium accuracy Holt-Winters - Seasonality type: mul Holt-Winters - Seasonal periods: 12 LSTM - Time steps: 12 LSTM - Neurons: 60 LSTM - Activation: tanh LSTM - Dropout: 0.2 LSTM - Validation folds: 3 LSTM - Repeats per fold: 3 Validation Metrics Validation RMSE: 0.96 Validation MAE: 0.77
Residual Plots CPU Usage Wall-clock time: 617.04 seconds CPU time used: 831.08 seconds CPU usage (overall): 134.69% Final Forecast Month HW Forecast LSTM Residual Hybrid Forecast Oct 2024 451.92 -81.92 370.00 Nov 2024 476.26 -46.80 429.46 Dec 2024 649.34 -33.27 616.06 Jan 2025 667.86 +155.30 823.16 Feb 2025 832.94 +35.61 868.54 Mar 2025 821.49 -2.94 818.54
