Digital Twin

The Digital Twin builds a virtual model of each solar plant down to the individual string and continuously compares what your plant actually produces against what physics says it should produce. This lets the platform tell you not just that something is underperforming, but which component, by how much, and why — a real outage, a degraded string, a stuck logger, or merely a communication gap.

Digital Twin Concept

The Digital Twin understands your plant as a hierarchy of components — feed-in meters, inverters, junction boxes, strings, and panels — and reasons about each one individually as well as about how they roll up into the whole plant.

  • Component-Level Model: Every component of the plant has a digital counterpart with its own expected behavior
  • Physics-Based Expectations: Expected production is computed from solar geometry, weather, and validated panel and inverter models — not from machine-learning guesses
  • Self-Calibrating: The model continuously learns each component's real behavior over a rolling window, so "expected" tracks reality rather than a fixed nameplate value

Solar Only Today

The Digital Twin analyzes solar photovoltaic (PV) plants. Wind and battery analysis are planned (see below). Other plant types can still collect and store data, but the per-component analysis described here applies to solar.

How the Digital Twin Works

The Digital Twin runs two distinct engines against your plant's measured data:

  • String Analysis — discovers and validates each string's configuration and characteristics (orientation, panel count, inverter clipping, shadowing, and performance). You trigger this when a plant is onboarded or whenever the configuration changes.
  • Watchdog Monitoring — runs automatically every night to evaluate the health of every component, separate genuine faults from communication gaps, and quantify energy losses with a confidence rating.

Two Engines, Two Triggers

String Analysis is run on demand (and once on startup); it is not scheduled on its own. Watchdog Monitoring is scheduled — it runs nightly per plant without any action from you. Don't expect a freshly onboarded plant's string configuration to refresh on its own until you trigger an analysis.

Expected production is built from a deterministic clear-sky model (solar position and irradiance for the plant's location and date), a validated single-diode panel model, and a validated inverter model. There are no machine-learning predictors and no interpolated "shadow" of the plant — every expectation is traceable to physics and to the plant's own measured behavior.

Data Validation

Before any value is trusted, the Digital Twin checks it for plausibility:

  • Physical Plausibility Checks: values outside physically possible ranges are flagged (and a string can be marked as having a measurement defect)
  • Relational Validation: child measurements are checked against their parent's total, so an inverter whose strings sum to an implausible figure is detected
  • Reference Comparison: each component is compared against its physics-based expectation and its own recent behavior

These checks prevent erroneous readings from being miscounted as real performance problems.

Self-Calibration

Rather than assume every component performs exactly to nameplate, the Digital Twin maintains a per-component correction factor that it updates over a rolling multi-day window. This keeps each component's expected production aligned with how it actually behaves.

  • The correction is updated only for healthy components and only on non-anomalous days — if a day is too disturbed (for example, more than a third of daylight shows anomalies), calibration is skipped so bad data never poisons the model
  • It is applied according to the day's weather (clear, hazy, or cloudy)

Give a New Plant a Few Days

Because the model calibrates against real behavior over a rolling window, a newly onboarded plant needs several days of clean data before its expected-production figures fully settle. Early numbers may be conservative until calibration completes.

Reference Irradiance — No On-Site Sensor Required

The Watchdog derives a measured-irradiance baseline from the best-producing strings in the plant itself, rather than relying on a physical pyranometer. This reference irradiance is what the rest of the analysis compares against, and it also doubles as the marker for "this day has been processed."

Weather and Shutdowns Are Excluded

Periods affected by snow, dew, fog, network outages, or controlled shutdowns are excluded for all components, so weather is never miscounted as a component fault. Where measured irradiance is ambiguous, the system falls back to standardized weather codes to confirm whether a shutdown or weather event occurred.

Solar Plants

For solar PV plants, the Digital Twin provides component-level analysis and monitoring across the entire direct current (DC) and alternating current (AC) power chain.

Component Hierarchy

Understanding how a solar plant is structured is essential to understanding how the Digital Twin monitors and analyzes it. A typical solar installation consists of several layers of components, each playing a specific role in converting sunlight into grid-ready electricity.

Abstracted Structure

The component hierarchy below shows the abstracted structure the Digital Twin uses for a solar plant. Real-world installations may be more complex; the Digital Twin condenses them to this model.

The Power Flow Path

Let's follow the path that electrical energy takes through a solar installation:

Note: Dashed lines indicate optional components or alternative paths. A string may connect directly to an inverter or through a junction box (GAK).

Component Descriptions

This entire flowchart represents a solar plant. Each component in the power flow serves a distinct purpose:

1. Solar Panel

  • The individual photovoltaic modules that convert sunlight into electricity
  • Multiple panels are connected in series to form a string
  • Panel specifications (power rating, efficiency) determine string performance

2. String Measurement Point

  • The point where one or more rows of solar panels are measured
  • A string is a series of solar panels connected together electrically
  • Typically consists of 10-30 panels wired in series
  • Each string produces DC power based on its configuration and orientation

3. Junction Box / GAK (Optional)

  • Groups multiple strings together electrically before they reach the inverter
  • "GAK" stands for "Generator-Anschluss-Kasten" (Generator Connection Box in German)
  • Not all installations use junction boxes; strings may connect directly to inverters

4. Inverter

  • Converts DC (direct current) power from the solar panels into AC (alternating current) power for the grid
  • Multiple inverters may exist in a plant, each handling a portion of the installation
  • Tracks conversion efficiency and operational status

5. Feed-in Meter

  • Measures the total energy exported to the electrical grid
  • Located at the grid connection point
  • Provides the authoritative measurement of total production

The Digital Twin creates a virtual model of this entire power flow and continuously monitors each component's performance.

String Analysis

String Analysis examines each PV string to validate its configuration and surface performance issues. You trigger it for a plant once it is onboarded, and re-run it whenever the configuration changes.

Best Run on Clear-Sky Summer Days

This analysis works best on clear-sky days, ideally in summer, so that all performance issues are visible. Run in winter, shaded rows may not be analyzable. String Analysis is triggered on demand or once at startup — it is not scheduled to repeat on its own.

Orientation Analysis:

  • Detects each string's effective orientation (azimuth) from its production pattern
  • Compares measured against theoretical production for different orientations
  • Flags deviation from the configured orientation
  • Normalizes data so performance differences don't distort the result

Panel Count Analysis:

  • Estimates the actual panel count per string from voltage and irradiance behavior across the day
  • Accounts for degradation and temperature, and requires ambient temperature and wind-speed weather data
  • Validates the configured count against the estimate and flags missing or non-functioning panels
  • Can identify when multiple rows are connected through a single measurement point

Inverter Limit Analysis:

  • Detects when DC power exceeds the inverter's AC capacity (clipping)
  • Quantifies clipping duration and the associated energy loss
  • Accounts for clipping so it is not mistaken for a fault

Shadowing Analysis:

  • Detects sunrise and sunset row-shadowing against the clear-sky expectation
  • Reports the morning and evening shadow periods and the resulting shadow loss

Performance Analysis:

  • Compares measured against simulated energy using the validated panel model
  • Produces a string performance ratio and tracks degradation across repeated analyses

Each string analysis carries a confidence rating and a per-string status: completed, zero current (no usable data — the string can be marked hidden), uncertainty high, insufficient data, or measurement defect (values outside physical limits).

Panel Configuration Requirement

String Analysis relies on a valid panel configuration for the plant. This is set automatically when the whole plant uses a single panel type. If multiple panel types are mixed across the plant, the per-type assignment cannot be detected automatically and must be configured manually.

Watchdog Monitoring

The Watchdog evaluates the health of every component in the plant and runs automatically every night, so you don't have to schedule or trigger anything. Each plant is assigned its own time slot in the early-morning window to spread the workload, and the plant is evaluated for the previous day.

It Backfills and Self-Heals

The first time a plant is monitored, the Watchdog backfills history (up to roughly the last six months) so you immediately have a full picture. On every following night it processes any days it has not yet covered — so a temporary gap in coverage fills itself in automatically. Nightly monitoring begins once enough of the plant's strings have had their configuration checked.

The Watchdog uses a two-pass approach so it does not raise false alarms when a single component drops out for a non-critical reason such as a network outage or a missing reading:

Bottom-Up Monitoring:

  • Evaluates each level from strings up to the feed-in meter
  • Compares actual production against the physics-based expectation
  • Excludes network outages, data gaps, and weather conditions (snow, dew, fog) and controlled shutdowns
  • Aggregates child components to evaluate their parents

Top-Down Inference:

  • When a component has missing data but its parent is healthy, the parent's production is used to infer the child's likely state
  • This is how an inverter that is genuinely fine but temporarily unreachable is correctly flagged as a communication issue, not an outage

Real Faults vs. Data-Collection Issues

A central job of the Watchdog is telling apart a genuine fault from a mere reporting gap:

  • Outage / Degradation: the component really produced less than expected — this counts as an energy loss
  • Data-Collection Issue: the parent is healthy but the child's data is missing or implausibly low — this is treated as a communication or logging problem and is not counted as a loss

This distinction prevents a disconnected logger from being reported as lost production, and sets the right expectation for what an operator should actually act on. For the full list of component states and what each means, see Component Evaluation.

Loss Confidence

Every energy-loss figure the Watchdog reports is assigned a confidence level — high, medium, or low — based on how consistent the underlying data was and how certain the gap classification is. The buckets always add up to the total loss, so you can see at a glance how much of a reported loss is solidly established versus uncertain. For how losses are computed and attributed, see Loss Detection.

Wind Turbines

Digital Twin analysis for wind energy installations is planned. Wind plant data can be collected today, but the per-component analysis described above does not yet apply to wind.

Planned Capabilities:

  • Turbine-level performance monitoring
  • Power curve validation
  • Wind resource correlation
  • Nacelle and rotor component tracking
  • Availability and downtime analysis
  • Vibration and operational anomaly detection
  • Yaw alignment validation
  • Grid integration monitoring

Planned Component Hierarchy:

  • Wind Farm (top level)
  • Individual Turbines
  • Generators
  • Control Systems
  • Environmental Sensors (anemometers, wind vanes)

Wind monitoring will follow similar patterns to solar plants, with physics-based models adapted to wind energy principles.

Battery Energy Storage

Digital Twin analysis for battery storage systems is planned. Battery plant data and the component hierarchy can be collected today, but the analysis engine does not yet evaluate battery behavior.

Planned Capabilities:

  • State of Charge (SOC) tracking and validation
  • State of Health (SOH) monitoring
  • Charge/discharge efficiency analysis
  • Thermal management monitoring
  • Cell balance analysis
  • Cycle counting and degradation tracking
  • Round-trip efficiency calculation
  • Grid services performance (frequency response, peak shaving)
  • Battery Management System (BMS) integration

Planned Component Hierarchy:

  • Battery Energy Storage System (BESS)
  • Battery Banks/Racks
  • Individual Modules/Cells
  • Power Conversion System (PCS)
  • Thermal Management
  • BMS

Battery monitoring will validate electrochemical behavior and ensure safe, efficient operation.

Operational Benefits

The Digital Twin gives you:

  • Trustworthy Monitoring: implausible readings are flagged and excluded, so what you see reflects the plant, not the data link
  • Root-Cause Clarity: deviations are attributed to a specific component and a specific cause — outage, degradation, stuck logger, inactive endpoint, or communication gap
  • Quantified, Rated Losses: every loss comes with a confidence level so you know how much to rely on it
  • Hands-Off Coverage: nightly evaluation and automatic backfill keep the picture current and fill gaps without operator effort
  • No On-Site Irradiance Sensor Needed: the plant's own best strings provide the irradiance baseline

Integration with Other Features

The Digital Twin works alongside other platform capabilities:

Technical Implementation

For the engineering view of how the Digital Twin runs, see Digital Twin Architecture.

  • Component Evaluation — the component states the Digital Twin can determine and what each one means
  • Loss Detection — how energy losses are quantified, attributed, and confidence-rated
  • Efficiency Detection — string and performance-ratio analysis for spotting underperformance
  • Real-Time Monitoring — the live metric pipeline the Digital Twin reasons over
  • Events — machine-detected signals raised from Digital Twin findings
MIT Licensed | Copyright 2026 Mirox Verwaltungs GmbH