Cap-Aware Boosting for Green Day-Ahead Electricity Price Forecasting
Organizations: AMPIN Energy Transition
Abstract
Accurate forecasting of Green Day-Ahead Market (GDAM) electricity prices is essential for market participants in the Indian power sector. We present a LightGBM-based forecasting framework for Indian Energy Exchange GDAM prices, which clear in 96 fifteen-minute intervals per day. The proposed system explicitly models price-cap behavior by separating capped-price events from normal market conditions. A classification head estimates the probability of the regulatory price cap (Rs 10,000/MWh), while a regression head predicts uncapped prices in logarithmic space. Final forecasts are obtained through probabilistic blending of these components and are accompanied by nine quantile forecasts for uncertainty estimation. Evaluated on 257 trading days between 1 January and 14 September 2026, the model achieves a mean absolute error (MAE) of 397.9 Rs/MWh and a mean absolute percentage error (MAPE) of 9.57%, substantially outperforming a Chronos-2 baseline (583.3 MAE) and a persistence baseline using the previous day's profile (589.1 MAE). We further introduce leak-safe variants that incorporate real-time market signals, unmatched sell volumes, and solar generation proxies. In a recent out-of-sample evaluation, the quantile-clock variant achieved the best short-term performance with an MAE of 789 Rs/MWh during 25-28 October, while the enhanced capped-price model achieved an MAE of 716 Rs/MWh across 21 evaluation days through 16 November. These results demonstrate the effectiveness of regime-aware probabilistic forecasting for high-volatility electricity markets.
Figures & tables
| Step | Name | Change from the row above |
|---|---|---|
| 0 | InputFix | Prod v2. Half-life 180 days. Log-price tree on every row. Cap blend. RTM only through . |
| 1 | Hurdle | Price tree trained only below the cap. Floor classifier at 1,005. Half-life 45 days. Weight . |
| 2 | Morning RTM | Blocks 0–35 of (00:00–08:45). Blocks 36–95 of that morning stay missing. |
| 3 | Auction shape | Lagged unmatched sell, GDAM–DAM spread, neighbour blocks, slope. |
| 4 | Clear sky | Three-hub GHI, west minus south, 7-day and 28-day drift. |
| 5 | exp5 | Steps 1–4 together. Issued price is still the cap blend, Eq. ( 13 ). |
| Setting | InputFix | exp5 | exp7 |
|---|---|---|---|
| Trees | 350 | 350 | 350 |
| Learning rate | 0.05 | 0.05 | 0.05 |
| Leaves | 63 | 63 | 63 |
| Min. child samples | 40 | 40 | 40 |
| Row / column subsample | 0.9 / 0.8 | 0.9 / 0.8 | 0.9 / 0.8 |
| Seed | 42 | 42 | 42 |
| Model | MAE | RMSE | MAPE (%) | WAPE (%) | Bias |
|---|---|---|---|---|---|
| Prod v1 | 397.9 | 849.1 | 9.573 | 7.677 | -6.1 |
| Prod v2 | 395.5 | 855.6 | 9.628 | 7.630 | 4.4 |
| Chronos-2 | 583.3 | 1,196.4 | 13.749 | 11.254 | 33.7 |
| Previous day | 589.1 | 1,281.6 | 14.037 | 11.367 | -13.4 |
| Previous week | 1,241.0 | 2,194.9 | 27.844 | 23.945 | -90.8 |
| Comparator | 95% interval | v1 wins / 257 | |
|---|---|---|---|
| Prod v2 | 2.4 | [-7.8, 12.6] | 136 |
| Chronos-2 | -185.4 | [-250.4, -127.7] | 202 |
| Previous day | -191.2 | [-240.1, -144.9] | 196 |
| Previous week | -843.1 | [-1,086.5, -620.5] | 236 |
| Model | Solar MAE | Non-solar MAE | Solar MAPE (%) | Non-solar MAPE (%) |
|---|---|---|---|---|
| Prod v1 | 337.7 | 463.3 | 11.474 | 7.507 |
| Prod v2 | 339.7 | 456.1 | 11.667 | 7.412 |
| Chronos-2 | 455.6 | 722.0 | 15.125 | 12.254 |
| Model | Pinball loss | Coverage (%) | Width | Point outside (%) |
|---|---|---|---|---|
| Prod v1 | 171.6 | 81.21 | 1,394.6 | 0.31 |
| Prod v2 | 192.0 | 83.22 | 3,565.6 | 0.48 |
| Chronos-2 | 234.2 | 78.30 | 1,595.9 | 0.07 |
| Model | MAE | MAPE (%) | Solar MAPE (%) |
|---|---|---|---|
| Chronos-2 (slim) | 2641.1 | 109.95 | 171.77 |
| InputFix Prod v2 | 1018.5 | 23.48 | 20.68 |
| exp5 ( v3_full ) | 827.7 | 19.20 | 15.86 |
| exp7 ( seasonal_master ) | 789.2 | 16.69 | 15.05 |
| Model | MAE | MAPE (%) | Solar MAPE (%) |
|---|---|---|---|
| Chronos-2 (slim) | 2000.4 | 108.67 | 182.19 |
| exp7 ( seasonal_master ) | 823.8 | 20.80 | 25.27 |
| exp5 ( v3_full ) | 716.0 | 19.07 | 21.90 |
| Model | MAE | MAPE (%) | Solar MAPE (%) |
|---|---|---|---|
| Chronos-2 (slim) | 1417.7 | 97.89 | 168.23 |
| InputFix Prod v2 | 708.1 | 26.12 | 35.68 |
| exp7 ( seasonal_master ) | 696.7 | 21.69 | 30.46 |
| exp5 ( v3_full ) | 606.7 | 20.27 | 27.51 |
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
| Model | History begins | Usable rows begin | Rows | Inputs |
|---|---|---|---|---|
| Prod v1 | 1 May 2022 | 21 May 2022 | 126,816 | 49 features |
| Prod v2 | 1 April 2022 | 21 April 2022 | 129,696 | 97 features |
| Chronos-2 | 1 April 2022 | 1 April 2022 | 131,616 | Univariate |
| Setting | Value |
|---|---|
| Estimators per ensemble | 350 |
| Learning rate / leaves | 0.05 / 63 |
| Minimum samples per leaf | 40 |
| Row subsampling / frequency | 0.9 / 1 |
| Column subsampling | 0.8 |
| Objectives | L1 point; binary event; quantile |
| Month | v1 MAE | v2 MAE | C-2 MAE | v1 MAPE | v2 MAPE | C-2 MAPE |
|---|---|---|---|---|---|---|
| Jan | 300.2 | 311.8 | 357.3 | 7.08 | 7.28 | 7.96 |
| Feb | 238.7 | 238.3 | 325.0 | 5.85 | 5.72 | 7.27 |
| Mar | 412.7 | 419.8 | 767.1 | 8.43 | 8.68 | 14.98 |
| Apr | 329.9 | 352.5 | 472.2 | 8.53 | 9.06 | 12.09 |
| May | 443.1 | 423.0 | 672.4 | 10.71 | 10.36 | 17.16 |
| Jun | 484.7 | 455.4 | 550.7 | 11.05 | 11.75 | 12.93 |
| Comparator | Block days | Lower | Upper | |
|---|---|---|---|---|
| Prod v2 | 7 | 2.4 | -8.1 | 12.8 |
| Prod v2 | 14 | 2.4 | -7.8 | 12.6 |
| Prod v2 | 28 | 2.4 | -8.6 | 13.2 |
| Chronos-2 | 7 | -185.4 | -244.2 | -133.2 |
| Chronos-2 | 14 | -185.4 | -250.4 | -127.7 |
| Chronos-2 | 28 | -185.4 | -253.9 | -122.9 |
| Actual price | Model | Blocks | MAE | MAPE | Bias | Coverage |
|---|---|---|---|---|---|---|
| Prod v1 | 464 | 225.6 | 30.60 | 195.0 | 64.01 | |
| Prod v2 | 464 | 282.2 | 37.28 | 265.7 | 53.23 | |
| Chronos-2 | 464 | 370.5 | 50.96 | 358.1 | 68.10 | |
| Prod v1 | 6398 | 374.4 | 3.74 | -374.4 | 87.53 | |
| Prod v2 | 6398 | 344.9 | 3.45 | -344.9 | 94.22 | |
| Chronos-2 | 6398 | 562.9 | 5.63 | -562.9 | 88.23 |
Electricity Price Forecasting
Electricity price forecasting predicts wholesale and retail prices to support trading, grid operations, and energy planning.
See all 25 papers