KYAMOS research portfolio
BRIDGE2HORIZON/0823 · Research & development

SMART-HYBRID-RENEW

Explore solar, wind, battery storage and electricity-network control

A research platform connecting renewable generation, battery energy accounting and electricity-network calculations. Explore the tested examples, prototype control strategies and assumptions behind the results.

Solar evaluation conditions
14
Separate from eight calibration conditions in the reported laboratory-reference preset. Outdoor accuracy remains open.
Example network sizes
2 + 308
Connection points in the networks used for GPU-assisted numerical comparisons with a CPU reference.
Smart charging controller
Prototype
Four rule-based strategies explore price differences, peak demand, solar self-use and combined selection.

Evidence checked 20 November 2026. Numerical examples and software checks have the limits stated below.

What the project does

SMART-HYBRID-RENEW studies how solar, wind, batteries and the electricity network interact. The software brings together solar-panel calculations, battery charge tracking, renewable dispatch and power flow: the calculation of power and voltage across a network.

The project includes software checks, comparisons with reference calculations and studies of simple rules alongside learned control. Its wind input shares development work with OPTI-WIND-AI. A shared component does not provide a separate wind-validation result.

The results below describe development examples. A successful calculation checks the software under its stated assumptions; a real installation also needs matched equipment, measurements, operating limits and customer data.

Solar simulator

A solar-panel electrical model with a clearly stated laboratory-reference scope.

The single-diode model represents the electrical behaviour of a solar panel and adjusts its calculations for temperature. Named presets preserve the relationship between model settings and their reference conditions.

The 14 November software report records six passing solar checks. Its laboratory-reference preset contains eight calibration conditions and 14 separate evaluation conditions. These checks cover model identity, reproduction of reference values, intermediate predictions and refusal of unsuitable inputs.

Measured current-voltage curves from the Sandia PV-IV-EL dataset support further model-development studies. The inspected analysis contains 616 measurement records. The number of records alone does not establish accuracy for an installed panel or for changing outdoor conditions.

Solar model: demonstrated scope and remaining work
ItemCurrent evidenceLimit
Electrical modelSingle-diode calculation with temperature translationDepends on the selected panel settings and input quality
Laboratory-reference preset8 calibration conditions; 14 separate evaluation conditionsReference-condition agreement is not outdoor validation
Software checks6 passing checks reportedSoftware behaviour, rather than six field experiments
Measured-data study616 Sandia measurement records analysedNo matched outdoor accuracy result follows from this count
Outdoor predictionEarlier measured comparisons retain substantial disagreementAccuracy for a real installation remains unresolved

A later analysis estimates that different panel settings could reduce outdoor error to about 4%. No new outdoor test establishes that figure, so it is not presented as achieved prediction accuracy.

Battery and renewable dispatch

Track stored energy, charge losses and the effect of accepted or refused actions.

Battery calculations account for stored energy, charging and discharging, power limits, reserve levels and conversion losses. Example control rules use renewable surplus and time windows to choose when to hold, charge or discharge.

The reported software checks include full-day energy accounting, rejected actions, network calculation failures and preservation of state after a failed step. These are useful checks of a simulation workflow; they do not establish the response or safety of a physical battery and converter.

  • Example capacity studies include 100, 197 and 394 kilowatt-hours of storage.
  • Energy ledgers distinguish charging cost, discharged energy, losses and the energy left in the battery.
  • Network-connected examples check voltage and loading constraints within the represented steady-state model.
  • A battery can reduce peak demand while increasing total imported energy because charging and discharging lose energy. Compare both outcomes with idle operation.

The 14 November summary reports 207 passing SMART software checks. Its listed component counts total 191, so the overall count has not been independently reconciled and is not used as proof of complete system validation.

GPU-assisted power flow

Graphics-card calculations compared with a processor-based reference on example networks.

The Newton power-flow calculation runs on a GPU, a graphics processing unit used for parallel calculation. Network conversion and result extraction run on the CPU, the ordinary processor. The adapter rejects unsupported equipment and does not silently replace a failed graphics-card calculation with a processor calculation.

The report describes ten numerical comparison cases across 2-point and 308-point networks. Four cases use the 308-point development network, covering import, export, curtailed solar generation and added heater loads. Other cases exercise a smaller network and changes to its inputs.

The checks compare voltages and calculated power with a reference implementation. They also check unsupported inputs, changed network parameters and failed calculations. This provides evidence about the stated calculation workflow, rather than measured behaviour of a real electricity grid.

Numerical power-flow evidence
MeasureReported valueMeaning
Network sizes2 and 308 connection pointsDevelopment examples, including four 308-point operating cases
Numerical comparisons10 cases across both network sizesNot ten independent physical-grid experiments
Solve tolerance1e-10 to 1e-11 MVAThe stopping threshold for the power calculation
Largest reported bus difference9.85e-10 MVADifference in a bus result, measured in megavolt-amperes
Largest reported line difference6.14e-8 MVADifference in a line result, measured in megavolt-amperes
Result comparison settingsAbsolute and relative tolerances of 1e-7Separate from the solve stopping threshold

These are balanced, steady-state examples: they calculate a settled operating point. They do not establish fast fault response, frequency stability, real-grid accuracy or a speed advantage over the reference.

Illustrative battery economics

Explore how assumed electricity-price differences change a simplified storage calculation.

Charging at a lower price and discharging at a higher price can create value, but the result depends on the actual contract, usage, losses and equipment costs. The project includes simplified annual scenarios to explore those relationships.

The examples below use a 197 kilowatt-hour battery with a 50 kilowatt power limit, a 10%-90% allowed charge range and 90% charging efficiency. They start half full and apply a prescribed daily charging schedule over 8,760 hourly steps. The prices are assumed inputs, not verified tariffs offered in any country.

Annual value means the credited discharge value minus charging cost, before upkeep or other costs. Every discharged unit is credited at the assumed high price. The calculation does not establish whether a customer could use or sell all of that energy at that price.

  • Schedule A charges from 22:00 to 08:00 and discharges from 08:00 to 22:00.
  • Schedule B charges from 23:00 to 07:00 and discharges from 12:00 to 17:00.
  • Schedule C charges from 21:00 to 06:00 and discharges from 06:00 to 09:00 and 17:00 to 21:00.
  • Schedule D charges from 22:00 to 08:00 and discharges from 16:00 to 22:00.
  • The simplified calculation does not enforce equal starting and ending stored energy. It does not include battery ageing, installation, financing, tax, replacement costs or actual grid constraints.
  • Separate peak-reduction and solar-self-use studies do not establish combined savings. The same battery capacity cannot be counted as fully available to several strategies at once.
Illustrative annual scenarios: 197 kWh, 50 kW battery
Assumed scheduleLow price (EUR/kWh)High price (EUR/kWh)Annual value before upkeep
A0.120.28EUR 8,435.36
B0.140.38EUR 12,916.20
C0.150.45EUR 16,289.10
D0.180.58EUR 21,856.88

These are illustrative model outputs, not customer savings, investment returns or payback promises. A useful site estimate needs current import and export terms, measured demand and generation, equipment specifications and a single feasible operating schedule.

Smart charging prototype

Four rule-based strategies provide a starting point for control research.

The standalone controller accepts forecast lists and returns a proposed battery action. Its saved demonstration uses hourly mock inputs. Live data connections, correct 15-minute energy updates and complete equipment and network constraints still need qualification before operational use.

Prototype implementation does not establish annual combined savings or a guaranteed response time. It provides a place to compare strategies and identify the additional checks needed for a real control system.

Prototype control strategies
StrategyPurposeCurrent scope
Price arbitrageCharge at a lower assumed price and discharge at a higher oneRules using a supplied price forecast
Peak reductionDischarge when demand crosses a thresholdA threshold-based development strategy
Solar self-useStore surplus solar generation for later demandRules using supplied solar and demand forecasts
Combined selectionChoose between the strategies using estimated valueA heuristic selector; optimality is not established

Public discussions should distinguish a proposed command, an accepted simulation action and a measured hardware response. They are different forms of evidence.

Learning versus simple rules

Use equal assumptions and clear references to assess whether a learned policy helps.

The project includes learning infrastructure and comparisons with fixed control rules. Existing development studies do not establish a consistent advantage for learned policies over simple rules.

A useful comparison checks the same inputs, battery limits, network constraints and final stored energy. Peak reduction, imported energy, losses and a chosen score should be reported separately so that an improved score is not mistaken for lower electricity bills.

  • Keep training examples separate from the examples used to assess performance.
  • Retain adverse results and refused actions when comparing approaches.
  • Test on representative new inputs before drawing conclusions about a real installation.

Deliverables & current status

Software delivery and scientific acceptance are separate milestones.

D12

Client graphical platform software

Linux engineering release prepared; full platform scope open

A shared application package and saved-result workflow are included in the delivery records. Complete Windows operation, the full server workflow and authorised acceptance remain separate tasks.

D13

Solar Simulator

Software package prepared; outdoor accuracy unresolved

The version 2.1 document records model and laboratory-reference work. Its proposed outdoor improvement and economic claims require the additional evidence described on this page.

D14

Wind Farm Simulator

Shared engineering package prepared; requirements remain open

The delivered wind dependency shares work with OPTI-WIND-AI. The package records remaining atmospheric, terrain, turbine-to-electricity and acceptance requirements; it is not a second independently validated wind implementation.

D15

Literature review of AI optimisation technologies

Methods review prepared

The review covers learning methods, battery control and optimisation approaches. A completed review document does not establish learned-policy superiority or validate projected financial outcomes.

D16

AI prototype for power-flow optimisation and grid balancing

Prototype and numerical evidence package prepared

Development examples cover battery accounting, dispatch, network comparisons and control studies. Physical operation, live control performance, customer savings and full acceptance remain unestablished.

D6 · Product forum

The project discussion area is prepared. Public posting opens after a moderator is assigned and topic, reply and moderation checks pass. Creating a page alone does not complete D6.

Open the project forum

Evidence & remaining work

This summary distinguishes saved development evidence from measured performance at a real installation. Supporting technical records can be discussed with the project team; confidential project files are not published here.

Solar software and reference conditions

14 November 2026

Six passing software checks and a preset with eight calibration and 14 evaluation conditions are reported. The same record identifies outdoor prediction as unvalidated.

Battery and network software report

14 November 2026

The summary records passing battery-accounting, dispatch and network checks. The stated overall total and listed component counts do not reconcile, so no verified total is claimed here.

Graphics-card numerical comparisons

14 November 2026

Reported comparisons span 2- and 308-point example networks. The code distinguishes solve tolerance from result agreement and excludes a real-site or speedup claim.

Assumed-price battery studies

15 November 2026

Saved annual outputs support the four illustrative price scenarios shown above. They do not establish current customer tariffs, combined-strategy savings or real-world payback.

What the results do not establish

  • Outdoor solar accuracy for a matched installed panel, changing weather and the associated converter remains unresolved.
  • The development electricity networks and steady-state calculations do not establish physical-grid accuracy, fast fault response or frequency stability.
  • The standalone smart charging controller remains a prototype; live operation and complete battery and grid constraints need further checks.
  • Financial scenarios use assumed prices and simplified operation. Customer savings, combined benefits and investment payback are not established.
  • Neither shared wind components nor successful software checks establish complete project acceptance or a general advantage for learned control.

Talk with the project team

Project coordinator: Antonis Papadakis, KYAMOS LIMITED. Programme: BRIDGE2HORIZON/0823.

For desktop demonstration access, installation instructions or a project-specific evaluation, contact KYAMOS. Keep customer data and confidential proposals out of public discussions.