Battery dispatch optimization runs on forecasts. The question is not whether to use forecasts, but which forecasts, for which decisions, over which time horizons. Getting this mapping right is the practical work of building a dispatch system that performs well across the full range of operating conditions a BESS will encounter.
Day-ahead and intraday forecast horizons serve different dispatch decisions with different requirements. Treating them as interchangeable, or using one where the other is appropriate, produces a system that works adequately in routine conditions and breaks down during the events that matter most for asset economics.
What Day-Ahead Forecasting Drives
The day-ahead forecast covers the next 24 to 48 hour window, typically initialized at midnight or early morning and used to build the operating plan before the market's day-ahead submission cutoff. In Japan's JEPX market, this is the forecast that feeds the supply plan submission by noon for the next operating day.
Day-ahead forecast skill requirements are dominated by two questions: What will total daily generation volume be? And when will the generation peak, plateau, and ramp down? For dispatch, the precise timing of the peak matters because it determines when the battery should be at low SOC (ready to charge from solar surplus) versus high SOC (positioned to discharge into peak-price hours).
Day-ahead forecasts have a structural advantage in one respect: they can use the full suite of numerical weather prediction (NWP) model output, including ensemble spread, to characterize forecast uncertainty. An ensemble of NWP runs through the day-ahead horizon gives the dispatch optimizer a probabilistic view of tomorrow's generation profile, which it can use to hedge between scenarios. A high-confidence clear-sky day calls for an aggressive arbitrage plan; a high-uncertainty partly-cloudy forecast calls for a more conservative SOC trajectory that preserves flexibility.
Where Day-Ahead Forecasts Fall Short
The limitation of day-ahead forecasts is that they are initialized before the operating day begins and rely on NWP model output that becomes increasingly stale as the day progresses. A forecast initialized at 00:00 UTC that was skillful at the time of JEPX submission may diverge significantly from actual conditions by 14:00 if a cloud system moved through faster or slower than the model predicted.
This divergence creates a well-known dispatch problem: the battery is executing a plan that was optimal given the morning's information, but actual conditions have changed enough that the plan is no longer optimal. The battery may be charging when it should be holding, or discharging when it should be building headroom for an incoming surplus.
Day-ahead forecast skill degrades more quickly in certain weather regimes. Partly-cloudy transitional weather, frontal passages, and sea-breeze effects (which are significant for coastal solar sites in western Japan) all produce intraday cloud variability that global NWP models represent poorly at the sub-10 km scale. For sites in these environments, the gap between day-ahead plan and actual conditions is wider on average, and the intraday correction layer matters more.
What Intraday Forecasting Drives
Intraday forecasting covers the zero to six hour window, refreshed at 30 to 60 minute intervals throughout the operating day. The primary dispatch decisions it feeds are: whether to revise the JEPX intraday supply plan submission for upcoming delivery hours, and whether to pre-position the BESS differently from the day-ahead schedule in anticipation of a forecast change over the next few hours.
For curtailment pre-positioning, the intraday forecast is particularly critical. The decision to pre-charge the BESS before a curtailment window opens needs to happen early enough that the battery can reach the target SOC before the curtailment instruction arrives. A three-hour intraday forecast showing high surplus probability for the 13:00 to 16:00 window gives the dispatch optimizer time to begin pre-charging from 10:00, even if the day-ahead plan had originally scheduled a midday discharge.
Intraday forecast accuracy benefits from satellite nowcasting. Japan Meteorological Agency's Himawari satellite provides high-resolution visible imagery at short intervals, and cloud motion vector extrapolation from this data produces irradiance nowcasts that outperform NWP for the zero to three hour window. An intraday dispatch system that integrates satellite nowcasting operates with materially better short-range information than one relying solely on the stale morning NWP initialization.
The Handoff Between Horizons
One non-obvious design decision in a combined day-ahead and intraday system is where the handoff between the two forecast sources happens. A naive approach uses day-ahead NWP for everything beyond six hours and intraday nowcasting for everything under six hours, with a hard boundary. A better approach uses a blended ensemble where the weight shifts from nowcasting toward NWP gradually as the horizon extends, based on the relative skill of each method for the site's specific climate conditions and season.
The skill-weighted blend matters because NWP performance varies significantly by season in Japan. During winter, when cloud cover patterns are more driven by synoptic-scale weather systems, NWP performs better at medium ranges. During summer, when cumulus convection and mesoscale cloud dynamics dominate, NWP degrades more quickly at ranges beyond a few hours. A fixed-boundary handoff ignores this seasonality and will underperform against a dynamically weighted blend.
SOC Management: Day-Ahead Sets the Trajectory, Intraday Corrects It
The relationship between the two forecast horizons is most clearly visible in SOC management. The day-ahead plan establishes a target SOC trajectory for the operating day: where the battery should be at each hour, what its floor and ceiling should be given the contracted regulation reserve, and what the end-of-day target is to set up the following day.
Intraday forecasting is how that trajectory gets corrected when actual conditions diverge from the day-ahead plan. If the morning's solar generation is 30% below the day-ahead forecast due to unexpected cloud cover, the battery may arrive at noon at higher SOC than planned. The intraday optimizer recognizes this, adjusts the afternoon discharge schedule to manage the surplus SOC, and revises the JEPX intraday position accordingly.
Without intraday correction, the battery executes the day-ahead plan regardless of how conditions have changed. On days with large forecast errors, this produces a meaningfully worse outcome: the battery ends the day at the wrong SOC, the JEPX settlement position may be in deficit, and the next morning's starting position is suboptimal. Over a year of operation, the cumulative effect of poor intraday correction is measurable in settlement costs and missed revenue.
Which Horizon Matters More
We are often asked whether we focus more on day-ahead or intraday forecasting. The honest answer is that the relative value depends on what the operator is optimizing for. If the primary objective is JEPX arbitrage and the grid environment is reasonably predictable, day-ahead forecast quality dominates. If the primary objective is curtailment avoidance in a high-penetration area like Kyushu, where curtailment events are driven by fast-changing conditions, intraday forecast quality matters more.
In practice, both matter, and the dispatch system needs to run both well. The day-ahead layer sets up the plan; the intraday layer keeps the battery positioned correctly when reality diverges from the plan. Operating with only one is like navigating with a map but no ability to recalibrate when you realize the map is wrong.
