AI Energy Management · ASEAN Baseline

Edge AI that keeps optimizing even when the internet doesn't.

GridMind AI runs forecasts 4 hours ahead, every 15 minutes, on-edge. Reduces diesel 18-40% in hybrid PV+BESS+Genset systems. Battle-tested in 693+ sites across 18 countries since 1991.

9–13%
Load forecast error · live sites
18-40%
Diesel Saved
24/7
Offline Edge
4 hr
Forecast Ahead
15 min
Cycle

Diesel saving is a design figure for a hybrid PV+BESS+Genset site — what a given site gets depends on its load, how the PV and battery are sized, and how much the generator ran before. The forecast error above is measured at live sites.

The Problem

3 gaps every microgrid operator faces today

Current EMS solutions miss the mark in three specific ways. GridMind AI was built to close each gap.

Gap 01

Reactive, not predictive

No forecasting. No tariff-aware dispatch. Operators react to load swings instead of anticipating them — missing TOU peaks and demand charge optimization.

Gap 02

Vendor lock-in

Cross-brand devices (Huawei, Sungrow, BYD, Leonics) cannot work together. Each vendor's app sees only its own equipment — fragmented operations, multiple dashboards.

Gap 03

No auditable carbon data

Raw energy data exists, but isn't converted into auditable carbon metrics. ESG reports become manual spreadsheet exercises. CBAM compliance is out of reach.

The Solution

Four things the system does for your site — every 15 minutes

Each one closes a gap you are paying for today — and they reinforce each other in one loop that keeps running 24/7, even with no connection to the cloud.

01

Your existing inverters keep working

No need to replace hardware to adopt the platform. Huawei, Sungrow, BYD, LEONICS or any other brand already on site is read into one common data format — and a new model is added as a config file, not a development project.
Solves: Gap 02 · Protocols: Modbus RTU/TCP, CAN, Analog
02

You see the next 4 hours before they happen

Parallel models predict solar generation and load, 4 hours ahead, every 15 minutes. Load forecast measured at 9–13% wMAPE at live sites. Runs entirely on-edge.
Solves: Gap 01 · Runs on: edge controller, no cloud
03

Every kWh is stored and used at the cheapest hour

Dispatch is decided against your real MEA/PEA bill — TOU, Maximum Demand, Ft and AFA — so the battery and genset move to cut cost, not just to balance power. Peak shaving, load shifting and capacity firming all come out of that one decision.
Solves: Gap 01 · Output: Modbus commands
04

Your carbon report is ready when the auditor asks

The kWh the site already logs becomes an auditable carbon figure — no separate data-collection exercise at year end. Built on TGO and IEA emission factors, and CBAM-ready for exporters shipping to the EU.
Solves: Gap 03 · Standards: TGO/IEA · CBAM · T-VER
CLOSED-LOOP PLATFORM · 15 MIN CYCLEM-01 · HALIntegration layerM-02 · FORECASTPredictive Edge-AIM-04 · CARBONESG / CBAMM-03 · DISPATCHTOU / MD / Ft / AFADataForecastDispatchVerify↻ every 15 min
Outcome 01 · Your existing equipment

Keep the inverters you already own

Adopting the platform does not mean re-buying your fleet, and it does not tie your next purchase to one supplier. Adding a device is a configuration file — no code change, no integration project.

HuaweiModbus TCPSungrowModbus RTULEONICSModbus RTUBYDCAN busBrand NAny protocolHAL LIBRARY · HARDWARE ABSTRACTION LAYERProtocol translatorRTU / TCP / CAN→ unifiedRegister mapperBrand-specific→ standardDevice profile libraryJSON configper brand/modelNormalized dataUNIFIED DATA BUSStandard format: kW · kWh · SOC% · V · A · Hz · temp — brand-independentPredictive Edge-AIForecast · M-02Tariff + dispatchTOU/MD · M-03Carbon + reportingESG/CBAM · M-04
✓ IEEE 1547-2018 DER interoperability ✓ Modbus RTU / TCP ✓ CAN bus ✓ Analog 4-20 mA ✓ Digital I/O dry contact
Outcome 02 · What changes on site

You act on the next 4 hours, not the last 5 minutes

Every 15 minutes the system projects solar generation and site load 4 hours ahead, then dispatches battery and genset against that projection — automatically, on site, with no operator in the loop.

Because solar is known ahead

Fewer genset starts, less wasted charge

  • The battery charges before a cloud front arrives — not after output has already dropped.
  • The genset does not start for a dip that will last ten minutes.
  • Export and curtailment are decided ahead of the limit, not corrected after it.
Because load is known ahead

The peak is shaved before it forms

  • Demand charge is set by one 15-minute peak a month — the system reserves battery capacity for it in advance.
  • Capacity is not spent early on a rise that does not matter.
  • When the genset does run, it runs at an efficient load point instead of as a reflex.
FORECASTNext 4 hourssolar + loadDECISIONCharge · discharge · holdre-decided every 15 minRESULTDiesel down 18–40%peak shaved · no operator↻ the same loop runs again every 15 minutes
Real-World Proof

Internet went down 6 hours at Sabah, Malaysia.

A cloud-dependent EMS site running in parallel — vs GridMind AI on edge. The difference is measured in liters of diesel.

PROOF BOX · INTERNET DOWN 6 HOURS · SABAH, MAY 2025

Cloud-Dependent EMS

  • AI optimization stopped working
  • 22 optimization cycles lost
  • 120 liters of diesel wasted
  • Dispatch fell back to manual operation

GridMind AI · Edge

  • AI kept running — 100% continuous
  • 0 optimization cycles lost
  • 0 liters of diesel wasted
  • Dispatch fully autonomous on edge

Edge AI isn't an optional feature. It's the new baseline for clean energy in ASEAN — where internet is unreliable, but customers still need 24/7 optimization.

On-Grid Application

On-grid too — cuts TOU cost and Maximum Demand, recomputed every 15 minutes

A rule-based EMS is lookahead-free — it hoards the battery because it cannot know when solar will refill it. GridMind AI predicts solar and load hours ahead: it discharges into the morning peak, lets free midday solar refill, and holds one flat import cap all evening. Same hardware, more value per kWh of battery.

One simulated on-grid day — GridMind dispatch

Linfox-class site · 300 kWp PV · 750 kW / 1,566 kWh BESS · TOU 4.18 / 2.60 THB/kWh. Midday PV refills the battery to 90%, which then discharges into the evening peak to hold grid import down — protecting the Maximum Demand charge.

GridMind AI on-grid dispatch — one simulated dayA 15-minute profile of load, PV, grid import and battery state-of-charge for a Linfox-class on-grid site. Midday PV refills the battery to 90%; the battery then discharges into the evening peak, holding grid import down.TOU PEAK 09:00–22:0002505007501000kWSoC0%50%100%00:0006:0012:0018:0024:00Free midday PV → battery to 90%Battery shaves evening peakPVLoadGrid importBattery SoC

Source: Solar Insight Pro 15-minute simulation (GridMind ON), a representative clear-sky weekday. Illustrative of dispatch behaviour — not a measurement of a specific installed site.

Dispatch 01

Dynamic Peak Shaving

The import cap ("shave line") is recomputed every 15 minutes from the remaining-window forecast — not a fixed annual target.
Result: Flat cap, never empty early
Dispatch 02

Smart TOU Arbitrage

Charges off-peak only what tomorrow's plan will burn beyond forecast solar — no blind charge-to-full timers.
Result: Buys only what it needs
Dispatch 03

PV Double-Cycling

Discharges into the morning peak when midday surplus is forecast to refill the battery — turning solar that would have been curtailed into peak-rate savings. How much is recovered is set by the site itself: the shape of the load, the PV size relative to that load, and the battery headroom left at midday.
Result: Recovers curtailed energy
Dispatch 04

Demand Charge Protection

Holds one flat import cap across the whole 09:00–22:00 window — protecting Maximum Demand (THB/kW), not just energy (kWh).
Result: MD held to plan
Trust 01

Self-Learning Bias Correction

Checks forecast against actual every day and auto-corrects seasonal drift — no manual re-tuning by an engineer.
Result: Adapts within 3 days
Trust 02

Explainable Decisions

Every action is logged with its reason, and the dashboard overlays the forecast and the live shave line on the same chart as reality. Forecast error (wMAPE) is tracked on screen daily.
Result: No black box · audited accuracy
+55%
Battery energy delivered vs tuned rule-based EMS
687 → 684 kW
Maximum Demand held level while delivering that extra battery energy
−149 MWh/yr
Peak-rate grid imports (300 kWp simulated site)
9–13%
Load forecast wMAPE measured at live sites

Source: GridMind production measurement across 3 sites live 24/7 in Thailand and the Philippines (2026), plus a full-year 15-minute simulation of a 300 kWp / 1,566 kWh on-grid site against an auto-tuned rule-based EMS, at measured forecast-error levels.

Rule-based EMS vs GridMind AI — same hardware, one full year

On the identical site, GridMind AI extracts 55% more energy from the same battery and buys 149 MWh/year less peak-rate power — while holding Maximum Demand equal. Forecast-driven dispatch simply uses the asset you already paid for, better.

Rule-based EMS versus GridMind AI — annual outcomes on the same siteOn the same battery hardware, GridMind AI delivers 55% more battery energy per year and imports 13% less peak-rate grid energy than a tuned rule-based EMS, with Maximum Demand held equal.Battery energy used / year273Rule-based EMS422GridMind AI+55% MWhmore value from the same batteryPeak-rate grid imports / year1,129Rule-based EMS980GridMind AI−13% MWh−149 MWh of expensive peak energySame site · 300 kWp PV + 750 kW / 1,566 kWh BESS · Maximum Demand held equal (687 vs 684 kW)

Source: full-year 15-minute Solar Insight Pro simulation of the same site — rule-based EMS (fixed peak-shaving target) vs GridMind AI (forecast-driven), at measured forecast-error levels.

Simulated Case · Demand Charge

The solar saved nothing on demand charge — the peak was at 21:45

Without storage, this C&I site hit its monthly maximum demand at 21:45 in all twelve months — after sunset. Adding solar moved the demand charge by exactly zero. Every baht of reduction came from the battery, and how much came from deciding when to use it.

Annual demand charge by system configuration, full-year simulation
Configuration Demand charge / year vs no PV
No PV 917,962
PV only 917,962 no change
PV + BESS · rule-based EMS 813,490 −104,471 (−11.4%)
PV + BESS · GridMind AI 620,654 −297,307 (−32.4%)

Same battery, same inverter, same tariff — only the decision-maker changed. GridMind AI took 192,836 THB/year more off the demand charge than the rule-based EMS, a 23.7% lower bill on identical hardware.

How much of this transfers to your site depends on when your peak falls. A plant that peaks at midday gets much of the reduction from PV directly. This one peaked after dark — which is exactly where storage and forecast-driven dispatch do all of the work.

Source: full-year 15-minute Solar Insight Pro simulation — assumed load profile of about 500 kW, Solcast irradiance at Bang Bo, Samut Prakan, 1 Jan – 31 Dec 2025, at the site's TOU maximum-demand rate. Simulated figures for a separate site from the 300 kWp study above — not metered results, and not to be read together.

Live Monitoring

GridMind AI Dashboard — what the operator sees

The AI's forecast and its grid-import cap are drawn on the same chart as reality — so every dispatch decision is visible and auditable, not a black box.

GridMind AI  ·  Site: On-Grid C&I · 300 kWp PV · 1,566 kWh BESS (illustrative)
GridMind ON System OK Live · 13:45:12
Solar Now
638 kW
Forecast → 655 kW
Load Now
296 kW
Forecast → 288 kW
Grid Import
0 kW
Cap (shave line): 360 kW
Battery SoC
90 %
Ready for evening peak
Savings Today
฿9,418
vs no-BESS baseline
Today — 15-min Profile with AI Plan Overlay kW · TOU peak window 09:00–22:00
Load PV Generation Grid Import GM Forecast PV GM Forecast Load GridMind Shave Line ESS Charge ESS Discharge
GridMind AI dashboard — one day of dispatch: forecast PV and load, the 360 kW shave line held through the 09:00-22:00 TOU peak, and battery charge and discharge windows. 050100Battery SoC %free PV refill 38→90%planned drain to 12%
AI Forecast — Ahead Horizons
HorizonSolar kWLoad kWGrid kW
+15 min6123010
+30 min5713220
+1 h4683590
+2 h236418182
+4 h21598360
Forecast Accuracy (wMAPE)
Solar · today10.8%
Load · today8.9%
Solar · 7-day12.3%
Load · 7-day10.1%
GridMind Decision Log
06:15 PLAN Overnight buy 0 kWh — PV covers today's burn
09:00 DISPATCH Morning discharge 103 kW — refill forecast OK
10:30 CHARGE PV surplus → BESS 415 kW (headroom reserved)
13:45 HOLD SoC 90% — reserving for 16:30–22:00 window
13:45 CAP Evening shave line = 360 kW (recomputed)

How to read it: the black stepped line is the grid-import cap GridMind promised for the peak window — recomputed every 15 minutes from the AI forecast — and the red Grid line rides exactly on it through the evening. The morning discharge at 09:00 was taken because the dotted forecast showed midday surplus would refill the battery for free. Illustrative dashboard on a simulated on-grid day — not a live screenshot of a specific installed site.

Real Deployments

693+ sites · 18 countries · 35 years

Selected projects where GridMind AI runs in production today.

🇲🇾 Sabah, Malaysia

TNBES Off-Grid Hybrid

4 sites · Solar + Diesel + Battery

GridMind AI keeps optimizing dispatch — even during the 6-hour internet outage of May 2025. Reduced diesel by 35%.

4
Sites
35%
Diesel ↓
24/7
Offline
🇹🇭 Bangkok, Thailand

BCP Solar Rooftop

126 kWp · MEA grid-tied C&I

GridMind AI optimizes TOU dispatch and demand charge reduction for a Bangkok C&I site on MEA tariff.

126
kWp Solar
TOU
MEA-optimized
MD reduced
🇻🇺 Port Vila, Vanuatu

Iririki Island Resort

NODEX microgrid · 10× Power Nodes

NODEX with 10× Power Nodes + Storage Nodes. GridMind AI manages BESS arbitrage across resort load profile.

N+1
Redundancy
10×
Power Nodes
Island
Off-grid
Why LEONICS

35 years of ASEAN microgrid data — tuned to sites like yours

  • 35-year proprietary datasetsince 1991, across 693+ sites in 18 countries. No competitor has this depth for ASEAN climate and tariff conditions.
  • Hardware + Software full-stackwe make the BESS, the PCS, the inverter AND the AI. End-to-end ownership — no integration friction.
  • Thailand-rooted, ASEAN-tuneddeep knowledge of TOU MEA/PEA, BOI tax depreciation (150% × 3yr), AFA, Ft, Demand Charge. Not generic global software.
  • Off-grid firstdesigned for sites where internet is unreliable. Edge AI is the baseline, not an optional add-on.
  • Standards & certificationsISO 9001:2015 · ISO 14001 · UL1741SA · IEC62477-1 · CSA22.2 · IEEE 1547-2018 DER · TGO/IEA · CBAM · T-VER

Track record cloud-only EMS can't match

693+
Reference sites
100+
MWh BESS
18
Countries
35
Years · 1991
Procurement Q&A

Questions your procurement team will ask — answered

Vendor lock-in?
No. HAL Library handles multi-brand devices via JSON profiles. Add Huawei, Sungrow, BYD, or any new vendor without code changes. IEEE 1547-2018 compliant.
Data sovereignty?
Edge-first. All data is processed and stored locally. Cloud sync is optional — your operational data stays on-site. ESG aggregations can be cloud-uploaded only with your consent.
What if internet goes down?
Nothing breaks. Edge AI continues 100% offline. Proven at Sabah (6-hour outage — 0 cycles lost, 0 diesel wasted). Forecast models, dispatch logic, Modbus commands — all on edge.
Replace SCADA?
No. GridMind AI is the application layer above SCADA. It reads from SCADA/PLC, optimizes, and writes dispatch commands. Your existing SCADA keeps doing what it does best.
Security & compliance?
Industrial-grade. Industrial SBC + real-time OS. Encrypted Modbus/TCP. ISO 9001 + 14001. IEEE 1547-2018 DER. CBAM and T-VER ready ESG reporting on TGO/IEA factors.
Pricing model?
Two models. Model A — Edge Controller retrofit (plug into existing PCS + cloud subscription). Model B — Full turnkey (GridMind + LEONICS Modular PCS, single SKU). Free 60-min technical workshop + free Solar Insight Pro feasibility study before any quote.

Your next site — proof of value in 4 weeks.

Free 60-minute Technical Workshop. By the end you'll have: pilot scope (up to 3 devices), deployment plan, draft quote — plus a free Solar Insight Pro feasibility report.