AI Solar, Battery & EV Optimisation | SolarMind Guide

How does AI energy optimisation work for homes with solar, batteries and EVs?

AI energy optimisation software decides, every half hour, whether a home battery should charge from the grid, hold its charge, power the house or export to the grid. It also decides when an EV should charge. It bases those decisions on forecast solar generation, expected household demand and the prices on your tariff. Whether it saves you money depends mainly on four things. These are your tariff type, the size of your battery relative to your usage, how much your consumption shifts from day to day, and whether your inverter can be controlled reliably.

For homes on a fixed single-rate tariff with a small battery, the gains are usually modest. For homes on a half-hourly or time-of-use tariff with a battery, an EV and a variable routine, the decisions get complicated quickly. That is where software tends to outperform fixed schedules set by hand.

Why fixed battery schedules leave value on the table

Most battery systems ship with simple modes. Common ones are self-consumption, a timed charge window overnight, or a forced export period. These work reasonably well when prices and weather are predictable. They struggle when either changes.

A typical example is a homeowner who sets the battery to charge fully from the grid every night at the cheap rate. On a sunny spring day, the battery is then already full by morning. The solar generation it could have stored is exported at a low export rate instead. On a dull winter week, the same schedule may not charge enough to cover the expensive evening peak. Neither outcome is a fault. The schedule simply cannot see what tomorrow looks like.

Half-hourly tariffs make this more noticeable. Agile-style tariffs publish a new price for each of the 48 half-hour slots in a day, and those prices change daily. Some days include very cheap or even negative-price periods. Others have sharp evening peaks. A schedule chosen once and left alone cannot respond to that pattern. Many people also forget to update their settings when the seasons change.

What the software actually decides each half hour

An optimisation platform repeats the same decision loop across the day. It looks ahead at prices, weather and likely demand, then picks the action for the next slot that gives the best expected outcome over the whole horizon. It does not simply pick the cheapest action in the moment.

The main decisions for each slot are these.

• Hold

  Keep the stored energy back because a more expensive period, or a better use of it, is coming later.

• Export

  Sell stored or surplus solar energy when the export price is high enough to beat using it later.

• Charge the EV

  Schedule vehicle charging into the cheapest or sunniest periods while still meeting the departure time you set.

• Charge from grid

  Store energy when import prices are low enough to justify it, allowing for round-trip losses.

• Discharge to the home

  Cover household demand from the battery when import would otherwise be expensive.

The look-ahead is what matters most. Discharging now might seem sensible, but the system should hold if a higher price is due in two hours. Charging from the grid overnight should be skipped if a strong solar day is forecast. Getting these trade-offs right, day after day, is where the value comes from.

Rule-based control versus reinforcement learning

There are two broad approaches. Rule-based systems follow logic that people have written, such as "charge if the price is below a threshold" or "export if the battery is above a set percentage". These are transparent and predictable. Their weakness is that every rule is a compromise. A threshold that suits January can be wrong in June.

Reinforcement learning works differently. An agent is trained in a simulated environment over a very large number of scenarios. It learns which actions lead to lower overall cost, given prices, generation, demand and the physical limits of the battery. SolarMind uses this approach, with a policy trained to make charge, hold and export decisions on a half-hourly basis. The benefit is that the system can handle the interactions between solar, battery, EV and tariff without needing a separate rule for every situation.

Neither approach is magic. A learned policy is only as good as its inputs and its training. Poor solar forecasts, inaccurate consumption estimates or wrong tariff data will lead to poor decisions whatever the method. Learned systems can also be harder to explain. That is why good software should show you what it did and why, so you can check it.

The factors that decide whether it is worth it

Optimisation software adds value on top of hardware you already have. It cannot make up for a system that is badly matched to the home. Before expecting meaningful savings, it helps to look honestly at your setup.

• Tariff type

  Savings depend heavily on price variation. A flat-rate import tariff with a low fixed export rate leaves little room to optimise. Time-of-use and half-hourly tariffs leave much more.

• Household routine

  Homes where occupancy, cooking, laundry and EV use change a lot from day to day benefit more from a system that adapts than homes with very regular patterns.

• EV charging flexibility

  An EV is often the largest flexible load in the home. If the car is regularly plugged in for long periods, there is more scope to move charging into cheap or sunny slots.

• Inverter and battery control

  The software needs reliable control of the inverter, either through a local integration or a manufacturer's interface. Slow, limited or unreliable control reduces what can be achieved.

• Solar array size and orientation

  The balance between self-consumption and export depends on how much generation you have and when it peaks. East-west arrays and south-facing arrays produce different daily profiles.

• Battery capacity relative to usage

  A battery that is small compared with evening demand will be emptied every day whatever the software does. A very large battery in a low-usage home may sit partly unused.

The honest answer for some homes is that a well-chosen tariff and a sensible manual schedule capture most of the available value. Software becomes more useful as the number of variables grows.

Common mistakes to avoid

Several problems come up repeatedly when homes move to automated battery control. Most are easy to avoid if you know to look for them.

• Forgetting round-trip losses

  Energy is lost on the way into and out of a battery. A small price gap between charging and discharging can disappear once those losses are counted.

• Ignoring battery warranty terms

  Check how your manufacturer defines warranted throughput and cycles. Heavy grid trading can use up that allowance faster than self-consumption alone.

• Running two controllers at once

  If the inverter's own schedule, a manufacturer app and third-party software all try to control the battery, they will conflict. Pick one system to be in charge.

• Leaving no reserve for power cuts

  If your system offers backup power, decide on a minimum charge level to keep in reserve and make sure the software respects it.

• Not checking the export arrangement

  Export payments depend on your export tariff and on having the right metering in place. Confirm your export is registered and paid before relying on export revenue.

• Trusting savings figures without checking the method

  Savings should be compared against a clear baseline that counts solar self-consumption properly. Figures that compare against an unrealistic scenario can be misleading in either direction.

How heat pumps and other loads fit in

Batteries and EVs are the most obvious flexible loads, but they are not the only ones. Heat pumps, hot water cylinders and immersion diverters can also be shifted in time, within limits. Pre-heating a well-insulated home before an expensive evening period, or heating hot water during a cheap or sunny slot, uses the building and the cylinder as thermal storage.

The limits matter here. Comfort comes first, and a poorly insulated home loses stored heat quickly. That reduces the benefit of shifting heating. Heat pump control also has to respect the system's design, including flow temperatures and defrost behaviour, so any optimisation should work with the heat pump's controls rather than against them. In practice, the larger and more varied the set of flexible loads, the more an integrated approach helps compared with controlling each device separately.

How to judge whether a platform is working for you

Give any optimisation system a fair trial across different weather and price conditions. A sunny fortnight and a grey fortnight can produce very different results, and a single week tells you little.

Look for a clear daily breakdown of what the system did. That means when it charged, held, discharged and exported, and the prices at each point. Compare costs against a realistic baseline, such as your previous schedule or a simple self-consumption mode on the same tariff. If the decisions regularly look wrong, such as exporting just before an expensive peak, it is worth investigating the inputs, especially the tariff data and consumption estimates.

Finally, keep your own control. You should be able to set an EV departure time, a minimum battery reserve and any periods you do not want the system to touch. Good automation should take the routine decisions off your hands while leaving you in charge of the ones that matter to you.