An EMS schedules BESS charging and discharging within power, state-of-charge and site constraints. Price or solar forecasts can improve that schedule, but an AI label does not prove savings. A purchasing decision needs a comparison with ordinary schedule-based control using the same inputs.

What the algorithm needs

Provide the prices actually payable under your contract, a load profile, PV forecasts and the minimum emergency reserve. If your electricity price is fixed, wholesale volatility does not automatically create a benefit. Exporting electricity also requires the appropriate contractual arrangement and permitted operating mode. A forecast cannot create an export right or additional connection capacity.

Test the result, not the presentation

Use a period that was not used to tune the algorithm. Compare actual costs with and without storage, including cycle losses, auxiliary loads, software fees and battery wear. Then test forecast errors: less sunshine, a lower selling price or an outage that starts earlier than expected. Keep the assumptions and dispatch log so that a different supplier can reproduce the comparison.

Protect the reserve first

Tariff dispatch must not consume energy reserved for critical loads. Specify minimum SOC, a manual dispatch lock, operation without internet and access to command logs. Purchase functions that can be demonstrated at acceptance, rather than a promise of future doubled profits.

Identify the decision that forecasting will change

An equipment buyer should distinguish data collection, forecasting and dispatch. An accurate price forecast cannot produce a useful schedule if the controller misses a production shift or a battery outage. Conversely, a straightforward rule may work well at a site with a stable operating pattern. Compare actual cost and compliance with operating constraints rather than the apparent sophistication of the model.

List the inputs that genuinely require forecasting. A planned furnace schedule may already be known, whereas solar output remains uncertain. Contract prices may be published before a decision or established through a different process. For each input, record its source, publication time and acceptable delay. The calculation must not use information that the real operator would not yet have possessed.

Run a historical test without future information

Ask the supplier to separate the data chronologically. One period can be used to configure the model; a later period should test it without access to future observations. Randomly mixing adjacent intervals can produce an attractive result that does not represent real operation. A reproducible report identifies the dates, algorithm version and information available at each decision point.

Request three comparisons: the site without storage, the same battery following an agreed simple schedule, and that battery using forecast-based dispatch. Load, prices and billing rules must remain identical. The difference between the last two cases indicates the contribution of the more elaborate controller. Savings created by purchasing a battery should not all be credited to the software.

Include unfavourable periods in the report. Average savings can conceal an exhausted reserve before an outage or an import-limit violation. Count such events and show their duration and largest deviation. For a production business, the consequences of a small number of failed operating decisions may outweigh a modest improvement in average energy cost.

Correct forecasts still need a complete energy balance

Consider an illustrative delivery of 100 kWh after charging. With 90% round-trip efficiency, the system needs approximately 111.1 kWh of purchased energy. At an illustrative charging price of UAH 4/kWh, this costs UAH 444.4. If the displaced purchase would cost UAH 5/kWh, the gross difference is about UAH 55.6, not UAH 100. Auxiliary consumption, battery wear and software charges have not yet been deducted.

Negative wholesale prices do not automatically create income for a consumer. The contract formula, variable billing components and permitted operating arrangement determine what the business actually pays. Feed the controller the applicable site price rather than an unadjusted market chart. Otherwise, a technically accurate optimiser may minimise a number that does not match the electricity bill.

Check the treatment of solar energy as well. Energy diverted to storage may displace immediate on-site consumption or an available export payment. Its opportunity cost depends on the real alternative during that interval. Assigning zero cost to every solar kilowatt-hour can overstate the value of storage even when the weather forecast is excellent.

Treat reserve energy as an enforceable constraint

Define the energy required for critical loads separately from commercial cycling. If the requirement changes by shift or season, the rule must remain understandable to the operator. Establish who can change it, how changes are recorded and whether economic discharge can be suspended before an expected interruption risk. An attractive dashboard is not a substitute for these controls.

Test the loss of a forecast feed, delayed prices, internet failure and invalid meter readings. Agree the fallback behaviour for each event. Holding the last command indefinitely may be inappropriate. The response needs to suit the actual architecture and must not bypass local BMS or PCS protection. Forecast optimisation should remain subordinate to the equipment’s operating limits.

A reserve expressed only as a percentage can also obscure changing available energy. Ask how the control strategy accounts for measured capacity, auxiliary loads and the specified delivery boundary. The objective is the agreed service for the critical load, not simply a visually reassuring SOC number on the screen.

Acceptance checks for software functions

CheckEvidence to request
Input dataSource, time zone, units and quality flags
Schedule executionRequested and measured power on the same time axis
Reserve constraintHistory showing compliance with the agreed minimum
Financial resultReproducible calculations using the full billing formula
Communication failureA demonstrated transition to the agreed local mode
Version changeConfiguration history and a practical recovery procedure

Ask for exportable data that an independent reviewer can analyse. A chart image is not equivalent to a time series with units and timestamps. Establish recording frequency, retention period and data ownership. Losing access to history when a subscription ends affects both financial review and the investigation of warranty events.

Check timestamps before blaming the forecast

Meter data can describe the interval just completed while a command marks the next interval’s start. Time-zone or daylight-saving differences can therefore make correct dispatch appear late. Align the records first, then distinguish a poor forecast from a command limited by temperature or SOC. The explanation should follow measured inputs, requested action and actual equipment response on one timeline.

Read the commercial scope as carefully as the algorithm

Separate one-off licensing, recurring charges, external data feeds, remote support and future integration changes. Identify which functions remain available locally after a subscription ends. Ask who confirms compatibility between a software update and the installed equipment, and what support period is included. That period should make sense alongside the intended ownership life of the BESS.

For a USK Energy enquiry, provide interval demand, the actual price formula, solar characteristics and the required reserve. Include a simple reference schedule against which forecast-based dispatch will be assessed. The procurement objective is a measurable improvement from a specific supplied configuration, supported by accessible command history, transparent operating limits and an understandable cost over the planned period.

Your inputs

Price difference after losses

Change the inputs and save the result to discuss the equipment package.

Charging energy required—kWh
Displaced price at energy-cost break-even—UAH/kWh
Energy-cost difference only—UAH

E is delivered AC energy; efficiency covers the full cycle at the same boundary. Δ = E × displaced price − E / efficiency × charging price. Initial prices are illustrative. Excludes CAPEX, lifetime consumption, service, taxes and charges not included in your input prices. A positive result does not establish payback.

Before requesting a quotation

Mark questions for which you already have a confirmed answer. Save the list and send unresolved points to the supplier.

Before requesting a quotation
0 / 3
How to check controls before ordering

Request an operating log for a specific load profile. It should expose the command, measured power, SOC, limits and reasons for stopping. A polished dashboard does not show what happens when communication fails or a forecast is wrong. Local protection takes priority over tariff optimisation; the minimum emergency reserve needs a separate setting.

Clarify data ownership, export options and the functions that remain available after a subscription ends. Remote access needs separated roles, an audit trail and a way to revoke access. Make failure-scenario tests an acceptance criterion rather than a promise of a future software update.

USK Energy / B2B

From calculation to equipment

Preparing an equipment purchase? Send the load, backup duration or specification. USK Energy will clarify the configuration, compatibility and delivery terms.