DESS: Trade mode not smart in cloudy days, while green mode charges to early on sunny days

Been wrestling with something similar on my static caravan setup and figured others here might have thoughts.

Running Victron kit with a forecasting-based charge strategy, and the tension between "trust the algorithm" and "trust your eyes" is real. On a decent day in summer, the system backs off overnight grid charging because it knows the solar will deliver — fair enough, it usually does. But then you get one of those classic British "well, it looked promising at 6am" days where the clouds roll in by 9 and never leave, and suddenly you're sitting on a half-empty Fogstar battery bank with no cheap-rate window left to top up.

The flip side is just as frustrating. On a genuinely clear forecast day, the system sometimes pushes a green charge too early — before you hit cheap overnight rates — and you end up paying more than necessary for electrons you didn't need to import at that hour.

What I haven't found is a clean middle ground. The forecast data is only ever an approximation, especially here in the UK where weather prediction beyond about 48 hours is arguably optimistic at best.

A few questions I keep coming back to:

• Does anyone manually override based on Met Office cloud cover data the evening before?
• Is there a minimum state-of-charge floor you set to protect against the cloudy-day scenario?
• Anyone using Victron's ESS with a tighter SOC buffer specifically for autumn and winter months?

Curious whether this is a settings problem, a forecast data quality problem, or just an inherent limitation of predictive systems in a famously unpredictable climate.

@EcoFlow_Master yeah this is the classic forecasting dilemma isn't it 😄

Had exactly this on my shepherd's hut setup last winter. What helped me was tweaking the ESS minimum SoC dynamically rather than relying on a single fixed threshold.

I run a Node-RED flow pulling in tomorrow's irradiance forecast from Open-Meteo (free API, UK coverage is decent). If forecast is poor, minimum SoC stays high overnight. Sunny day predicted? Let it drop lower, knowing it'll fill by midday.

Victron's built-in DESS is honestly a bit blunt for UK weather — too optimistic on marginal days. Rolling your own logic via VenusOS + Node-RED gives you much finer control.

Worth checking Fogstar cells too if you're near capacity limits — tighter cell matching helps when you're cycling more aggressively through dynamic strategies.

Really interesting thread this - I've been down this rabbit hole myself on my narrowboat.

What worked for me was layering a minimum SOC buffer on top of the forecast logic. So even if DESS thinks a sunny day is coming and wants to discharge aggressively overnight, I've set a hard floor of around 30% that it simply won't breach. Means on those frustrating grey days where the forecast was optimistic, you're not caught completely flat-footed.

@EcoFlow_Master the early green mode charging on genuinely sunny days is slightly annoying but honestly I'd rather lose a bit of grid-free potential than get stranded. Have you tried tightening your forecast confidence threshold in the settings? On mine a small adjustment made a noticeable difference to how cautiously it interprets marginal forecasts.

@BirchLover curious what your shepherd's hut setup ended up settling on?

Has anyone tried setting a minimum SoC floor that adjusts based on the forecast confidence interval rather than the forecast value itself?

So on a cloudy day where the model is uncertain, you'd hold more reserve — not because the forecast says low generation, but because you don't trust the forecast enough to run lean.

On my van build I just do this manually (hardly sophisticated), but I'm wondering if something like Victron's ESS assistant combined with Node-RED could automate it based on, say, the Met Office DataPoint API uncertainty bands.

@EcoFlow_Master what forecasting source are you pulling from — Solcast, or something else? That might matter quite a bit here. Some services give you P10/P90 bounds which seem more useful than a single figure for this kind of decision.

@VanKev that's exactly where I landed after a winter of frustration. I run a Fogstar Drift 280Ah LiFePO4 bank with Victron Cerbo GX, and I ended up scripting a Node-RED flow that pulls the Solcast API confidence bands directly. When the P90/P10 spread is wide (i.e., the forecast is unreliable), it raises my overnight floor from 20% to around 45%. Narrow spread on a clear anticyclone day? Floor drops back down and DESS gets more headroom to work properly.

The key insight for me was treating forecast uncertainty as a first-class input rather than just the headline figure. UK maritime climate makes the headline number almost meaningless half the time — it's the confidence interval that tells you whether to trust it.

Anyone else using Solcast's probabilistic output this way, or are people still just using the single deterministic figure?