"How do you know if your solar plant is performing optimally?" We asked this question to an audience of asset managers and operations and maintenance professionals at RE+ 2024, and received half a dozen different answers.

Despite half a decade of work in the renewable energy industry, we are still consistently surprised by the lack of consensus on this fundamental question. Before we can meaningfully discuss the health of a single project, much less an entire portfolio, we need to agree on a yardstick.

While there's no shortage of performance metrics in the industry, they're not all created equal. We've watched operators, investors, and maintenance teams struggle to align on what exactly each KPI measures and when it should be used.

In our experience at Wattch, an effective KPI for performance management should:

1. Disambiguate resolvable and unresolvable energy losses
2. Respond rapidly, providing certainty in less than a day
3. Clearly identify "healthy" performance without requiring additional context

We've taken the liberty of grading the most widely used KPIs based on these criteria.

Now, let's dive into the most common questions we receive about each KPI.

Specific Yield: F

Specific yield, measured in kilowatt-hours per kilowatt peak (kWh/kWp), is a basic calculation that divides total energy output by the DC capacity of the system:

Specific Yield = Actual Energy Output / DC Nameplate Capacity

For example, if a 100 kW system generates 1 Megawatt-hour, the specific yield would be 10 hours.

Unfortunately, this metric has serious limitations. Most importantly, it's entirely dependent on weather conditions - even a perfectly healthy system will show low yield during cloudy periods. Additionally, specific yield fails to account for unavoidable losses like clipping or inverter efficiency.

This KPI appears to originate from Solar Resource Maps that track average annual solar insolation for different regions. While useful for site selection and financial feasibility analysis, it's ineffective for evaluating operating solar plants.

Using specific yield to assess performance is like trying to evaluate a car by looking only at its total mileage while ignoring terrain, traffic, and route.

Performance Ratio (PR): F

Performance Ratio attempts to account for weather variability by incorporating measured Plane of Array (POA) insolation:

PR = Actual Energy Output / (Plane of Array Insolation × DC Capacity)

Despite this improvement over Specific Yield, PR has significant limitations. Most notably, it ignores temperature effects. As a rule of thumb, PV module performance degrades by 0.3% - 0.5% per degree Celsius. This means a system in Arizona will consistently show worse PR values than an identical system in Minnesota, purely due to temperature differences.

Weather Corrected Performance Ratio (WCPR): D

Operators often try to address PR’s shortcomings with Weather-Corrected PR (WCPR), which incorporates temperature data. Unlike previous KPIs, this calculation requires time-series data rather than simple daily/monthly/annual totals.

WCPR = (Σ MeasuredEnergy(t)) / (Σ (POAInsolation(t) × DCCapacity × (1 + γ(Tcell(t) - Tref))))

When selecting the Tref component in the WCPR equation, there are generally two options:

1. The original NREL paper defining WCPR uses the average temperature over the year in question. This approach ensures annual PR and WCPR have the same value, while monthly WCPR remains stable regarding temperature. While theoretically interesting, this can be cumbersome in practice.
2. The more common approach uses a constant 25°C, the Standard Test Condition (STC) temperature for PV modules. This is the method implemented by every PV monitoring and reporting software platform we're aware of.

Weather-corrected or not, an "ideal" PR will always be less than 100% due to unavoidable system losses and design factors. The expected PR value for a given site requires context about its configuration - while 80% PR might be perfectly acceptable for some sites, at others it could represent thousands of dollars in lost revenue.

While PR has become an industry standard, particularly in performance guarantee contracts, its limitations make it poor for operational insights. Most major monitoring platforms provide this metric by default, but that convenience shouldn't be mistaken for effectiveness.

Baseline Energy Performance Index (BEPI): C

BEPI, commonly called "Performance vs. Budget," compares actual production against output predicted by your performance model (typically PVsyst):

BEPI = Actual Energy Output / Predicted Energy Output

BEPI's prominence in solar is primarily driven by financial considerations. Performance predictions are used to secure project financing, and maintaining a minimum BEPI is often required in loan agreements.

Consider this scenario: A solar project's financing requires maintaining a BEPI of 90%. The predicted energy values come from PVsyst simulations using typical meteorological year (TMY) data - essentially 20-30 year weather averages for the project location. During unusually sunny periods, a site might achieve 100% BEPI while having significant operational issues. Conversely, during poor weather, a perfectly healthy system could show an alarmingly low BEPI and risk violating loan covenants.

Like Specific Yield, BEPI ignores real-time weather conditions, making it ineffective for operational performance assessment. While asset managers must track BEPI to satisfy financial partners, it shouldn't guide day-to-day operations and maintenance decisions.

Availability: B

Availability (or uptime) measures the percentage of time that a solar system is operational and capable of generating power. The basic formula appears simple:

Availability = Operating Time / Total Available Production Hours

However, the industry lacks consensus on precisely how to calculate either component. Common considerations include:

• Weighting uptime by inverter capacity (AC or DC)
• Communication vs. production availability
• Excluding nighttime and force majeure events
• Time bucket granularity

Regardless of calculation method, availability can be misleading. A site running at half health might report 100% availability simply because it's producing some power. Alternatively, a site with intermittent production in the very early morning could show much lower availability than the total energy might suggest.

Availability is best understood as a measure of O&M team responsiveness - how quickly are total outages and communication failures addressed? While valuable for operations teams, it's inadequate for performance optimization since it doesn't capture partial losses or efficiency issues.

Energy Performance Index (EPI): A

At Wattch, Energy Performance Index (EPI), sometimes called "Performance vs. Expected," is our preferred KPI and the metric we recommend customers use to track, trend, and analyze their portfolios. EPI answers one of the most fundamental questions in Asset Management - "Am I leaving value on the table?"

At its core, EPI simply compares actual production against the "expected" or ideal production based on real-world meteorological conditions. While the formula looks deceptively simple:

EPI = Actual Energy / Expected Energy

The complexity lies in calculating the expected energy value.

It often seems there are as many opinions on calculating Expected Energy as there are Asset Managers, and like KPIs, each has its pros and cons. While we can't cover them all, here are the three most common methodologies, plus our approach at Wattch.

Linear Model

The simplest approach assumes a set of "linear" behaviors from any given solar power plant, then uses straightforward arithmetic to calculate expected power output:

Expected Power = min((POA Irradiance / 1000 W/m²) × DC Capacity × [1 + γ(Tcell - 25°C)], AC Capacity)

Where γ is the temperature coefficient of the module used at the site.

While easy to implement, this method typically achieves accuracy within 8-10% at best. Many customers find this surprising, but solar modules and inverters behave far more complexly than this model represents. For reliably diagnosing EPI swings of less than 10%, a more detailed and accurate model is needed.

That said, we find that even with a basic linear model, EPI remains more useful than any previously discussed KPIs.

ASTM Method

The ASTM Method, derived from ASTM Standard E2848 (ca. 2011) for Capacity Testing, uses a statistical rather than physical model:

Expected Power = A + B × Irradiance + C × Temperature + D × Wind Speed

Where A, B, C, and D are coefficients calculated using time series data regression.

This model does not account for clipping and assumes entirely linear meteorological impacts on PV output. To help correct these issues, the standard guides users to discard data points where irradiance is below 400 W/m² or inverters are within 2% of nameplate output.

For a modern commercial PV system with a 1.3:1 DC:AC ratio, this means discarding over 60% of irradiance input range. Consequently, collecting enough data for valid regression can take days or weeks.

Given these drawbacks, we generally recommend against using the ASTM model for Expected Energy calculations.

"Adjusted" PVsyst

PVsyst is arguably the leading tool for designing and analyzing theoretical PV solar installations. It combines a sophisticated physical model with extensive Typical Meteorological Year (TMY) data to produce "Predicted" output for a site (see BEPI above).

PVsyst uses what we call a "diode level" model, relying on the Single Diode Model of photovoltaic performance. This diode-level simulation, plus robust handling of shading, ohmic losses, optical refraction, and other unavoidable losses, produces highly accurate results for output power under typical weather conditions.

The challenge becomes reconciling these predicted outputs with real-world weather data. There are two common approaches:

1. Computing the ratio of TMY irradiance to measured actual irradiance and using that ratio to "adjust" PVsyst energy values. When operators request "weather adjusted" output, this is typically what they mean. However, this methodology assumes linear behavior, and we strongly recommend against it in practice.

Expected Energy = Predicted Energy * (Actual Insolation / Predicted Insolation)

2. Re-running PVsyst simulations using measured rather than TMY meteorological data as input. While rare in practice, we fully support this approach and trust its output validity.

Digital Twin

Finally, we arrive at Digital Twins. This term is quite overloaded in the energy industry, so let me start by clarifying my 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.

Unlike statistical or simplified physical models, a well-implemented Digital Twin captures the complex interactions between modules, inverters, and environmental conditions. This granular modeling enables accurate calculations and precise identification of underperforming components. As the physical system ages or undergoes modifications, the Digital Twin adapts to reflect these changes, ensuring calculations remain accurate throughout the asset's lifetime.

One of the many values produced by a robust solar Digital Twin is Expected Energy, enabling simple and highly accurate EPI calculation. However, the applications extend far beyond just EPI - from predictive maintenance to design optimization.

We'll share more details about Wattch's Digital Twin in a future blog!

The Bottom Line

Performance metrics in solar energy are essential tools for operational decision-making. While each KPI has its place, understanding their strengths and limitations is crucial for effective asset management.

When assessing solar performance, remember that sophisticated, weather-normalized metrics like EPI provide the clearest picture of system health. Traditional KPIs like Performance Ratio and availability should be evaluated cautiously and with full site context.

Our shared goal is maximizing the value of every solar asset through accurate performance assessment and timely issue identification. While implementing sophisticated metrics may require additional investment, the potential returns through improved system performance make it worthwhile for any serious solar portfolio.