Energy At Risk

ACME Solar
EAR Report

Solar Performance & Anomaly Analysis

ACME Solar Installation
28 January 2026
01 Nov 2025 – 28 Jan 2026
2,310 kW (2.31 MW)
Energy At Risk

ACME Solar — EAR Executive Dashboard

2,310 kW capacity · 7 inverters Analysis period: 01 Nov 2025 – 28 Jan 2026
Performance Ratio
74.1%
Target: 80% · Below Target
Daytime Availability
96.2%
Target: 97% · Below Target
Energy Shortfall
32.2%
291,279 kWh Lost
Capacity Utilisation
67.8%
7 Inverters Monitored
Historical Production vs Physics Expected Yield Evaluation Period: 01 Nov 2025 – 28 Jan 2026 Expected Physics Yield Actual Solar 350k 250k 150k 50k 0 kWh Nov 2025 242,100 kWh Dec 2025 208,200 kWh INV-3 & 4 Trip Jan 2026 164,052 kWh 291,279 kWh Lost Shortfall (−32.2%)
01

Executive Summary

Finding: System Performance Ratio Below Contractual Target

ACME Solar delivered a fleet Performance Ratio of 74.1% over the evaluation period—5.9 percentage points below the 80.0% IEC target. Four of seven inverters consistently failed contract specifications.

Finding: Daytime Availability Deficit Masks Daylight Failures

While OEM SCADA recorded 97.8% 24-hour availability, daytime operational availability dropped to 96.2% (below the 97.0% SLA). Inverters registered 77.6 cumulative hours in "Shutdown: on fault" during peak generation windows.

Finding: Behind-the-Meter Double Penalty Explains the 16% Bill Variance

Because ACME Solar operates behind the meter, ungenerated solar energy is automatically replaced by municipal grid draw at peak Time-of-Use rates (R5.10 / kWh average). This translates to R94,264 in period losses and an annualized cash drain of R1,146,878 ($69,264 USD).

02

Performance Ratio Analysis

Reference Yield (Irradiance)
700,074 kWh
Theoretical potential per pyranometer data
Actual Fleet Generation
518,487 kWh
Audited revenue meter production
Performance Ratio Gap
−5.9 pp
74.1% actual vs 80.0% IEC target
Per-Inverter Performance Ratio vs IEC 0.80 Contract Target IEC 0.80 Target 0.0 0.4 0.6 0.8 1.0 INV-1 82.1% INV-2 78.4% INV-3 61.2% INV-4 64.8% INV-5 81.9% INV-6 65.3% INV-7 83.7% −R 320,400/yr (INV-3) ≥ 80% Target Marginal (75–79%) Sub-Target (Costing Cash)
Inverter Telemetry ID Reference (kWh) Actual (kWh) PR (%) Status Annualized Impact
INV-1 INV-...54495195 100,010 82,108 82.1% Compliant +R 24,100
INV-2 INV-...54495196 100,010 78,407 78.4% Marginal −R 18,200
INV-3 INV-...54495190 100,010 61,206 61.2% Critical −R 320,400
INV-4 INV-...54495192 100,010 64,806 64.8% Critical −R 290,100
INV-5 INV-...54495191 100,010 81,908 81.9% Compliant +R 21,800
INV-6 INV-...54495194 100,010 65,306 65.3% Critical −R 284,500
INV-7 INV-...54495193 100,014 83,710 83.7% Optimal +R 42,300
03

Availability Analysis

Daytime Technical Availability
96.2%
Target: 97.0% (O&M Contract SLA)
Recorded Fault Hours
77.6 hrs
All classified as "Shutdown: on fault"
24-Hour Nominal Uptime
97.8%
High nominal uptime masks daytime solar trips
Why Nominal Availability Misleads Asset Managers

Inverters idling normally at night inflate the 24-hour metric to 97.8%. When evaluated strictly during productive daytime irradiance hours (>50 W/m²), availability falls to 96.2%. The 77.6 fault hours coincided with peak tariff periods, magnifying financial loss.

04

Physics-Based Performance Analysis

Physics Expected Production
905,631 kWh
Irradiance × 2,310 kW × 85% system efficiency
Actual Plant Yield
614,352 kWh
Total revenue meter energy delivered
Midday Clipping & Droop Gap
36.1%
Expected 2,194 kW peak vs 1,402 kW actual
Generation Gap Decomposition 291,279 kWh Unharvested (32.2%)
Actual Generated: 614,352 kWh (67.8%) Lost to Hardware Faults: 291,279 kWh (32.2%)
05

Anomaly Detection Analysis (Physical World Model)

Critical State Events
931
Shutdown & Trip Events
Underperformance Anomalies
11,809
>20% Below Expected Manifold
Thermal Warnings
1,798
IGBT Heatsink >75°C
Zero Production Windows
104
Daylight Hours Missed

Asoba's self-supervised Joint-Embedding Predictive Architecture (JEPA) builds a continuous operating manifold for each asset, separating ambient weather effects from internal electrical degradation. The world model proved that 92.6% of lost solar yield was internal hardware underperformance, while cloud cover and weather variation accounted for just 7.4%.

Asset Anomalies Primary Signature Isolated Hardware Culprit
INV-3 (54495190) 2,412 MPPT 2 Voltage Collapse Blown DC string fuse & open combiner circuit
INV-4 (54495192) 2,218 Midday Thermal Throttling Clogged heatsink fins & cooling fan assembly 2 stall
INV-6 (54495194) 2,054 Intermittent Ground Trip Low insulation resistance ($R_{iso} < 50 ext{ k}\Omega$) on String 14
INV-2 (54495196) 1,890 Partial String Undercurrent Localized module soiling & diode degradation
06

Financial Impact & Three Recovery Scenarios

An executive committee does not require ambiguous engineering charts; they require a costed business case. Ona translates electrical physics into three capital-gated recovery scenarios (directly modeled on the Sun International Sibaya Casino methodology), enabling asset managers to calibrate maintenance interventions against verified payback schedules:

SCENARIO 1
50% Restoration
R573,439
$34,632 USD / year recovered
  • Energy: +112,439 kWh recovered
  • Scope: Inverter 3 (INV-3) triage
  • Focus: MPPT 2 string fuses & combiner block
PAYBACK: < 30 DAYS
SCENARIO 2
75% Restoration
R860,158
$51,948 USD / year recovered
  • Energy: +168,658 kWh recovered
  • Scope: Inverters 3, 4 & 6 sub-array overhaul
  • Focus: Fan assembly, fuses & ground isolation
PAYBACK: 60 – 90 DAYS
The 3-Year Commercial Horizon Impact

Leaving this 5.9 pp performance gap uncorrected represents a 3-year cumulative loss of R3,440,634 ($207,792 USD) in permanent grid replacement penalties. Repairing Inverter 3 alone captures over 40% of the site's total financial upside.

07

Corrective Actions: JEPA → OEM Manual RAG

The Autonomous Intelligence Pipeline: Telemetry → Manual RAG → Work Order

Instead of generic advice ("inspect all cables"), Ona's JEPA model pinpoints the exact anomalous latent state, queries our vector database of OEM service manuals, and synthesizes part numbers, torque specifications, and step-by-step diagnostic procedures directly for your field technicians.

Selected scenario: 3 · Full Fleet Overhaul

Section 6 sets the operational policy: 100% of annualised R 1,146,878 leaves no targeted inverter loss. Section 7 only items inverters with explicit OEM manual citations: INV-3 = R 320,400, INV-4 = R 290,100 and INV-6 = R 284,500 (R 895,000 total). The remaining R 251,878 (21.9%) is the INV-2 / marginal / availability tail not covered by the same three part-replacement scopes; need either additional corrective scopes, availability-tariff intervention, or residual operational-loss accounting rather than another generic BOM line.

Selected work-order recovery: R 895,000 (all three cited OEM interventions)

Work Order 01: INV-3 (54495190) — MPPT 2 String Fuse Replacement
Financial Recovery: R320,400 / year · Priority: Immediate (7 Days)
Huawei SUN2000-330KTL-H1 Manual §6.4 · Alarm ID 2001

Identified Fault: DC String Undercurrent / Open Circuit on MPPT 2.

Required Part: 15A 1500V DC gPV Fuse (10×85mm) — Part #PV-15A10F.

Torque Spec: DC terminal disconnect screws: 2.5 N·m.

Estimated Labor: 1.5 technician hours on site.

Technician Action Checklist:
  1. De-energize DC disconnect switch on Inverter 3. Verify zero voltage with 1,500V rated multimeter.
  2. Extract fuse holders for Strings 4 & 5 on MPPT 2; test continuity with digital multimeter.
  3. Replace blown fuse element with certified 15A 1500V gPV element (Part #PV-15A10F).
  4. Retorque combiner terminal screws to exactly 2.5 N·m; re-energize DC switch and confirm string current >8.4A.
Work Order 02: INV-4 (54495192) — Heatsink De-Soiling & Fan Unit 2 Replacement
Financial Recovery: R290,100 / year · Priority: 14 Days
Huawei SUN2000 Series Guide §8.1 · Alarm ID 2032

Identified Fault: External Fan Module 2 Stall causing thermal curtailment above 28°C ambient.

Required Part: External Fan Assembly Unit 2 — Part #FAN-330K-MOD.

Torque Spec: Fan bracket M4 mounting screws: 1.2 N·m.

Estimated Labor: 2.0 technician hours.

Technician Action Checklist:
  1. Power down inverter; lock out AC and DC disconnect breakers.
  2. Unscrew fan baffle tray; inspect internal impeller blades for foreign debris or bearing seizure.
  3. Unplug quick-disconnect power harness; mount replacement fan assembly (Part #FAN-330K-MOD).
  4. Clean heatsink radiator channels using 4-bar dry compressed air; re-engage power and trigger manual fan test via commissioning app.
Work Order 03: INV-6 (54495194) — String 14 Low Insulation Resistance Isolation
Financial Recovery: R284,500 / year · Priority: 30 Days
Huawei SUN2000 Series Guide §7.2 · Alarm ID 2064

Identified Fault: Ground Insulation Resistance Fault ($R_{iso} < 50 ext{ k}\Omega$) causing midday inverter tripping.

Required Tool / Part: 1,000V DC Insulation Resistance Tester (Megger) + MC4 EVO2 Connectors.

Acceptance Spec: $R_{iso} > 1.0 ext{ M}\Omega$ across all DC conductors to earth.

Estimated Labor: 3.0 technician hours.

Technician Action Checklist:
  1. Disconnect all DC inputs on Inverter 6; apply 1,000V test voltage between positive/negative conductors and station ground.
  2. Isolate fault to String 14 (measured at 32 kΩ); inspect rooftop cable run between Modules 16 and 22 for pinching against sharp purlin edges.
  3. Cut damaged section, crimp new MC4 EVO2 connectors with calibrated ratchet crimper, and seal inside UV-resistant conduit.
  4. Retest to confirm insulation resistance exceeds 10 MΩ before reconnecting to inverter.
08

Methodology & Standards Compliance

IEC 61724-1 Class A Performance Ratio

Performance Ratio is computed per the international standard $PR = Y_f / Y_r$, where $Y_f$ is the specific final energy yield (kWh/kWp) and $Y_r$ is the reference yield calculated by integrating hourly global tilted irradiance (GTI) measured by calibrated on-site reference sensors.

Joint-Embedding World Model (JEPA)

Self-supervised representation learning predicts future latent representations from historical sensor observations across 5-minute intervals. Anomalies are flagged when cosine distance between predicted and actual embeddings exceeds 3.0 standard deviations from normal operational baseline.