In our previous blog post on solar KPIs, we introduced the concept of digital twins as an important tool for assessing system performance and provided a high-level definition:
A digital twin combines a comprehensive model of a system's interconnected components with a real-time, physics-based simulation of their expected behaviors.
In this post, we pull back the curtain on the physics and engineering that make digital twins possible and discuss why we believe they are the key for successful asset management.
In turn, we will examine each of the three essential components of the Wattch Digital Twin: 1) the static inputs and time series data that feed a digital twin, 2) the physics-based calculations that power the models, and 3) the outputs that enable best-in-class performance metrics and alerts.
Finally, we will explain how digital twins can help enhance your team’s diagnostic capabilities to reduce truck rolls, eliminate manual data analysis, and streamline corrective action.
Understanding these technical foundations will help you better evaluate software solutions and maximize the performance of your solar investments as you scale your fleet.
In general, digital twin accuracy is bound by input data quality, or, as we say at Wattch, "garbage in, garbage out." Accurate and complete data enables precise outputs, while poor inputs compromise the entire model.
There are four key types of data that digital twins require to function effectively. In the following sections, we will explore each in detail.
Accurate weather data forms the foundation of any solar performance model. While this requirement appears straightforward, small implementation details can have a significant impact.
Sensor manufacturers offer a wide range of equipment capturing different meteorological parameters at varying levels of precision, but in our experience, installation quality has a bigger impact on model performance than equipment cost.
For example, a $4,000 Class A pyranometer installed with improper leveling or non-representative shading will consistently provide incorrect readings. Conversely, a $1,000 Class C pyranometer, properly installed and maintained, can deliver reliable, actionable data with more than enough accuracy for a many-megawatt PV system.
While this may seem obvious, in practice we find that improper sensor installation is nearly always the root cause of major modeling discrepancies.
It is also worth calling out that while on-site sensors provide the most accurate data, satellite irradiance services can provide a decent approximation of these measurements when physical sensors are impractical.
The primary meteorological data series used in the Wattch Digital Twin are the plane of array irradiance and back of module temperature. Any additional irradiance and temperature measurements are used to approximate these two values. In future articles, we will talk more about how to balance precision and cost-effectiveness in meteorological data.
You can't model what you don't know. A digital twin needs the real composition of a site—array geometry, inverter specs, and string configurations—all as installed, not as designed.
During construction, field conditions drive changes. Strings get shortened to avoid obstacles, wiring routes change for structural constraints, and equipment gets substituted due to supply chain realities. And many times, these changes are small enough that they are neither reported nor recorded.
Where many traditional models lack the fidelity to identify these small discrepancies, a physics-based digital twin reveals them immediately. When string voltages don't match theoretical calculations, owners have a choice: treat discrepancies as commissioning issues to be corrected in the field, or as acceptable losses to “mark down” and incorporate into a baseline. Either decision is valid, but making that decision requires visibility into what's actually happening on your site.
The assumed goal for any PV power plant is simple: maximize output. But that's increasingly not the reality on today's grid.
The industry is seeing rapid proliferation of flexible interconnection agreements, export limitations, hybrid assets with storage, and fully dispatchable solar installations. In these cases, the PV system isn't always trying to find the Maximum Power Point—it's targeting specific operational outcomes like active power limits, reactive power support, or voltage regulation. The plant might be curtailing to 80% of capacity during peak export hours, providing grid services, or coordinating with co-located storage to optimize revenue.
This fundamentally changes what "expected performance" means. A digital twin that assumes maximum output will flag perfectly normal curtailment as underperformance, creating false alarms and eroding confidence in your monitoring system. Instead, the digital twin needs to consume these control setpoints in real time to maintain an accurate model.
Comprehensive component specifications are required for physics-based modeling, far above and beyond what is included in a typical data sheet. Details like efficiency curves, turn-on voltages, and per-MPPT current limits are essential for a comprehensive model of system behavior.
In our experience, independent third-party testing data is the best source for this type of information. For modules, Wattch prioritizes California Energy Commission (CEC) independent test data over manufacturer specs, though PAN files serve as a good fallback. Inverter modeling follows similar principles, though CEC data for inverters requires frequent augmentation with in-field measurements and observations.
When these standard data sources are insufficient, Wattch works directly with Original Equipment Manufacturers to obtain these detailed operational parameters. At the time of writing, our equipment library contains over 4,000 models for customer use, with ongoing updates as new equipment enters the market and additional performance data becomes available.
A large part of what differentiates digital twins is the use of first-principles physics rather than linear approximations or comparative analyses. This approach drives higher accuracy and better diagnostic capabilities compared to traditional top-down or linear models.
Before discussing how digital twin outputs are calculated, it’s important to dispel some common misconceptions about them. Digital twins differ fundamentally from conventional performance modeling approaches in that they are:
Not adjustments to predicted models. Digital twins do not compare project performance predictions made through PVsyst or another modeling tool to operational data.
Not statistical models based on historical data. Digital twins do not extrapolate current expected performance from past behavior.
Not a comparative analysis between components. Digital twins do not assume “similar” components should perform identically. Each component is modeled individually based on its specific characteristics and operating conditions.
All three of the above approaches are quite popular, but share a fundamental flaw: they bake your prior assumptions into your future predictions. Digital twins calculate expected performance from only fundamental physical principles and ground truth information.
The good news for solar asset managers is that historical, regressive, or statistical approaches aren’t necessary. The behavior of the silicon semiconductor is one of the most thoroughly researched and well-understood topics in the history of materials science. Decades of academic research provide a comprehensive theoretical foundation for photovoltaic modeling, with extensive literature documenting these physical relationships.
Wattch hasn’t invented any proprietary physics models (yet). Instead, we leverage established academic research. Our contribution lies in commercializing these calculations for day-to-day operation on customer data, making sophisticated, computationally intensive modeling accessible through easy-to-use interfaces.
One of the most common questions we are asked is, “How does your digital twin actually work?” What follows is a simplified overview of the bottom-up modeling process used in the Wattch Digital Twin.
Step 1: Calculate the location of the sun. The very first step is determining where the sun is in the sky at any given moment. This calculation is critically important for accurate modeling, but almost comically obtuse mathematically—involving corrections for Earth's orbital eccentricity, atmospheric refraction, and the derivation of solar time. If you're curious about the gory details, you can read about the implementation in NREL/TP-560-34302.
Step 2: Determine plane-of-array irradiance. For the first meteorological input, the model must determine how much solar energy reaches each cell second-by-second. Sensors mounted in the Plane-of-Array (PoA) provide a direct measurement and require no further calculations, while Global Horizontal Irradiance (GHI) measurements require decomposition and transposition into Direct Normal Irradiance (DNI) and Diffuse Horizontal Irradiance (DHI) components. This conversion process introduces increased uncertainty, making PoA sensors preferable to GHI sensors when practical.
Optical effects in the glass layer of the module are also considered at this stage. Refraction changes the effective angle of the incident irradiance, resulting in a change in performance commonly referred to as the Incidence Angle Modifier (IAM).
Step 3: Analyze shading impacts. Our approach distinguishes between far shading and near shading effects. Far shading from distant objects like mountains affects irradiance sensors and solar modules identically, and generally can be ignored. Near shading from adjacent structures requires detailed modeling to calculate partial shading effects based on solar position and object geometry.
Step 4: Calculate the cell temperature. To determine the cell temperature, we compute a thermal gradient based on module material type and back-of-module sensor readings, when available. Otherwise, we calculate cell temperature from irradiance, ambient temperature, and wind speed measurements using established thermal models.
Step 5: Calculate the IV curve of each cell. We apply the single diode model to determine current-voltage relationships for individual cells based on temperature and irradiance conditions. Module degradation is taken into account to reduce voltage/current as the cells age. This step is the core physical model used by the Wattch Digital Twin.
Step 6: Aggregate IV curves in parallel and series. Our model then combines individual cell IV curves into modules, modules into strings, and strings into inverter MPPT inputs, maintaining the electrical relationships between system components. This is the step for which an accurate site configuration is most important.
Step 7: Simulate MPPT behavior. We then model inverter maximum power point tracking algorithms to determine optimal operating points on aggregated IV curves.
Step 8: Model inverter behavior. We apply efficiency curves, clipping limits, and curtailment setpoints to each inverter based on its configuration and component data. Inverter efficiency varies with input/output voltage and current rather than remaining constant, requiring dynamic efficiency calculations. Both DC clipping on a per-input basis and AC clipping are modeled when power exceeds inverter capacity.
Step 9: Calculate conductor losses. In this process, we account for ohmic losses between inverters and the point of interconnection. These losses scale with current squared (P = I²R), increasing disproportionately at higher power levels.
While determining expected energy production is the primary objective of any performance model, intermediate calculations also provide significant diagnostic value. For every real data point you can measure, a digital twin can provide an "ideal" analog for comparison. Expected string voltage and current, along with IV curves, enable groundbreaking performance analysis.
A key insight is that while string voltage varies mostly with changes in temperature, string current is primarily influenced by irradiance.
Current deviations suggest soiling, unexpected shading, or module degradation issues may be present. Voltage deviations may be a sign of failed modules, active bypass diodes, or construction errors in string configuration. These diagnostic capabilities enable detailed analysis before the truck leaves the warehouse.
Some traditional performance models typically rely on heavy-handed, conservative loss estimates. For example, 2-3% is a common "loss factor" for ohmic losses which is applied at the end of the calculation on an energy basis.
With a digital twin, ohmic losses should be calculated second-by-second based on actual current—losses scale with the square of current—and compared to real-world observations of voltage drop within your monitoring system.
Because a digital twin calculates everything over time and with component-specific losses, it maintains granular detail throughout the course of a day. Not only is this dynamic modeling a more accurate approach, it helps you identify issues that only occur during specific parts of the day. A string with intermittent shading might perform perfectly in the morning but underperform every afternoon. Traditional models average this out and may miss the pattern entirely.
While conservative "loss factors" are perfect for financial and underwriting models, their use in O&M can mask significant, resolvable performance issues in real-world projects. Modeling every detail with a digital twin uncovers the small losses that can add up to big savings.
Traditional, top-down, linear models often come with error bounds of nearly 10%, while modern single-diode based modeling can be as accurate as 1-2%. This precision enables confident identification of genuine underperformance issues and accurate quantification of performance gaps. This quantification translates directly into a better understanding of project financials and allows for more informed decision making.
Beyond O&M Applications
A good digital twin isn't just for operations and maintenance—it becomes an improved input for forecasting models, battery storage dispatch algorithms, and grid service optimization. Traditional solar forecasting relies on historical production patterns, but a physics-based model can predict output under novel weather conditions or update immediately after equipment changes.
Digital twins represent a fundamental advancement in solar performance monitoring methodology. Rather than relying on statistical comparisons or historical extrapolations, this approach models actual power generation from first principles, making it both more accurate and more useful for issue diagnosis and resolution. When underperformance occurs, you can be confident in both the magnitude of issues and the likely causes of these issues.
The underlying physics may be complex, but the operational benefits are clear: enhanced visibility and precision, accelerated problem resolution, and improved confidence in solar performance data.
For information about implementing the Wattch Digital Twin for your solar portfolio, contact us at hello@wattch.io.