Forecasts Are Always Wrong: Why Co-Location Lives or Dies on Real-Time Data

The pitch for co-location is clean: one grid connection, two assets, stacked revenue streams. Solar exports during the day, the battery charges when prices are low and discharges when they're high, and the whole arrangement makes better use of constrained infrastructure. It makes sense on a whiteboard.

The operational reality is considerably less tidy - and a new piece from Arenko's Daniel Moore-Oats, published in PV Tech Power and surfaced by Energy-Storage.news, names the problem directly. Forecasts are always wrong. That's not a complaint about the quality of forecasting tools. It's a structural fact about how co-located assets actually operate, and it has real consequences for whether a project delivers its business case.

The Coordination Problem Nobody Puts in the Deck

When you co-locate solar PV with a BESS, you're not just sharing a cable. You're stacking multiple systems that each have their own logic: generation forecasts, battery optimisation algorithms, on-site control systems, metering infrastructure, and live market signals. Each of these layers generates its own view of what should happen next - and they don't always agree.

Forecasting systems generate multiple views of expected generation; optimisation platforms determine battery behaviour and prices; on-site control systems enforce physical limits; metering systems track real-world performance. That's four distinct layers, each operating on different data, different latency, and different objectives. The gap between what was forecast and what is physically happening on site - right now, in this settlement period - is where co-location projects either capture value or bleed it.

It becomes a constant balancing act between what was forecast to happen, what's physically happening on site, and what the market and grid are asking for at any given moment. At that point, real-time data and system integration stop being 'nice to have' and become fundamental to whether a project truly performs.

This is a grid-planning argument as much as it is a software argument. The grid is asking for dispatchability. Co-location is supposed to provide it. But dispatchability requires knowing, in real time, what capacity is actually available - not what a day-ahead model estimated twelve hours ago.

What Happens When the Assets Don't Talk to Each Other

The coordination problem gets worse when the solar and battery assets are owned or optimised separately - which is more common than the industry likes to admit. A separate SolarCo to hold the solar asset and a BatteryCo to hold the BESS might be an appropriate structure if the developer intends to sell the assets separately, and this separation may enable separate financing packages for each. That's a legitimate commercial rationale. But it creates a structural tension at the operational level.

Misaligned ownership models and optimisation incentives between PV and BESS assets can impact both operational performance and value capture. When the solar operator is optimising for generation yield and the battery operator is optimising for ancillary service revenue, the shared grid connection becomes a contested resource. Decisions about capacity allocation can be contested during periods of high renewables generation, and the commercial structures can often govern the outcome.

The result is curtailment - not because the grid can't absorb the power, but because the two assets are working against each other at the connection point.

The Curtailment Context

This isn't a niche concern. Constraint payments and balancing costs in the UK reached £2.3bn in 2024-2025, according to National Grid ESO data. That's the system-level cost of mismatched generation and grid capacity. Co-location is supposed to be part of the answer - but only if the assets are genuinely coordinated.

Traditional forecasting models rely on historical data and broad assumptions, missing the grid's nuanced and dynamic behaviour. Adding even one new generator to the system can shift curtailment patterns in unpredictable ways. For a co-located project, that means the curtailment assumptions baked into the original investment case can be invalidated by a neighbouring connection that didn't exist when the model was built.

Solar generators are also seeing more frequent periods of price cannibalisation during peak generation hours, with zero and negative pricing no longer an anomaly but an increasingly regular feature of sunny, low-demand days. A battery that can absorb that generation and shift it to a higher-value period is exactly the right tool - but only if it knows, in real time, that the solar asset is about to be curtailed.

What Whole-Site Optimisation Actually Means

The phrase "whole-site optimisation" gets used loosely. What it means in practice is a single system that can see all the relevant data - generation actuals, battery state of charge, live market prices, grid signals, connection constraints - and act on it without waiting for a human to reconcile conflicting instructions from separate platforms.

There is a shift in mindset among operators toward whole-site optimisation, where co-located assets are optimised as a single system. At one UK solar site co-located with BESS, Arenko's Nimbus platform increased grid connection utilisation from 8% to 29%, with export availability around 83%. That's not a marginal improvement. Moving from 8% to 29% utilisation on the same physical connection is the difference between a project that delivers its business case and one that doesn't.

At a co-located UK wind project, wind forecasts are integrated directly into the optimisation process, allowing the system to dynamically allocate export capacity between the wind farm and the battery while continuously trading across markets. Export availability sits at around 70%, but through sophisticated automated trading, the asset still performs in line with industry benchmarks.

The wind case is instructive precisely because wind is harder. Solar has a predictable daily profile. Wind doesn't. Wind-plus-storage projects are less predictable than hybrid solar, and subject to significant export constraints during high winds. If real-time integration can make wind co-location work, it can make solar co-location work - but the integration has to be genuine, not a dashboard that aggregates data after the fact.

Photo: Damien Power / Unsplash

The Question Every Operator Should Be Asking

The next phase of hybrid deployment will be defined less by hardware and more by commercial, operational, and organisational integration. That's the right frame. The hardware decisions - battery chemistry, inverter topology, DC vs. AC coupling - matter, but they're largely solved problems. The unsolved problem is the data layer.

The question for anyone investing in or operating co-location isn't whether real-time data matters - it clearly does - but whether they have the 'digital backbone' to deliver the deep systems integrations and automation needed to act on it. If not, how will they ensure they deliver their co-location business case?

That's a harder question than it sounds. It requires being honest about whether the optimisation platform in use can actually see both assets simultaneously, whether the control system latency is low enough to respond to intraday price moves, and whether the commercial structure between asset owners allows for coordinated dispatch in the first place.

The industry has spent years building the hardware case for co-location. The operational case - the one that determines whether a project actually performs - is built on data infrastructure that most projects are still catching up to. Forecasts will keep being wrong. The question is whether the system around them is fast enough to notice and respond.