PV simulators comparison: Electrical modeling of the DC side

Overview

This article picks up from the Optical losses article, which covered the chain up to the effective irradiance reaching each PV cell. The DC side begins at the cell and ends at the inverter terminals where AC power is produced.

Three modeling stages are covered:

• DC conversion - the cell's current-voltage characteristic and cell temperature, which together set the DC power available at each time step

• DC-side losses - the deductions between the array output and the inverter input, dominated by ohmic losses on the DC conductors and by inverter clipping

• Inverter conversion - the efficiency of the conversion from DC to AC at the inverter terminals. Clipping is covered under DC-side losses, because it reduces the power drawn from the modules rather than dissipating power in the inverter

The stages are measured against three points in the chain:

1. the theoretical DC power at the array output,

2. the DC power arriving at the inverter input, and

3. the AC power at the inverter terminals.

The inverter sits at the DC/AC boundary: its model takes DC characteristics as input and is covered here for that reason. The AC chain, from inverter output through the transformer to the grid, is covered in Electrical modeling of the AC side.

For the comparison setup (sites, system configurations, version numbers, and bias and RMSE definitions), see Setup and test methodology.

DC conversion

DC conversion turns the effective irradiance reaching each PV cell into DC electricity. The cell's current-voltage (IV) characteristic determines how much electrical power can be extracted at any given operating point. Cell temperature, irradiance level, and the electrical layout of the array all shift this characteristic. There are four areas in which the simulators differ materially:

• Cell model

• Cell temperature

• Transient temperature correction

• IV curve aggregation

Cell model

Two families of approach are used in the compared simulators. Single-diode equivalent circuit models describe the IV curve from physical parameters, either in the canonical De Soto five-parameter formulation (Solargis Evaluate and Solargis Prospect) or in extended versions that add explicit irradiance, temperature, and spectral dependencies for better accuracy across a wider range of conditions (PVsyst and SolarFarmer), at the cost of more parameters per module. Empirical efficiency models instead fit polynomial or simplified functional forms to measured performance, avoiding the diode equation altogether. SAM and pvlib offer models from both families.

Single-diode models are physics-grounded and extrapolate beyond the tested conditions. Empirical models are simpler, but they depend on a populated module database or apply only to standard cell technologies.

Cell temperature

Cell temperature is calculated from ambient temperature, irradiance, and wind speed, and affects DC output through the cell's temperature coefficient. Energy-balance models treat the module as a balance between absorbed solar radiation, electrical extraction, and thermal losses by convection and radiation. Solargis Evaluate uses a modified Duffie and Beckman formulation, Solargis Prospect the same without a wind term, and PVsyst and SolarFarmer an array temperature model with Uc and Uv coefficients. The nominal operating cell temperature (NOCT) approach interpolates linearly from a single reference condition, which is faster but less responsive to varying conditions. SAM offers NOCT alongside a heat transfer model and the Sandia cell temperature model, and pvlib offers both families plus further empirical fits including Faiman, Fuentes, and Ross.

Transient temperature correction

Transient temperature correction is required for sub-hourly simulations, as this resolution exposes module thermal inertia. When irradiance changes rapidly at cloud edges or under partial shading, actual cell temperature lags the equilibrium value, and a steady-state model overstates the swing. A transient correction smooths the temperature response over the preceding minutes. This matters most for sub-hourly simulation at variable-weather sites.

IV curve aggregation

Cell-level resolution computes IV curves per cell and sums them up the electrical hierarchy through substring, module, and string, so bypass diodes and half-cut geometry are handled naturally. Solargis Evaluate and Solargis Prospect work at this resolution. Coarser aggregation works at submodule, module, or string level and represents uneven shading through a generalized mismatch correction factor instead. PVsyst and SolarFarmer aggregate at these levels, with a definable submodule structure in PVsyst, and SAM additionally assumes uniform maximum power point tracking (MPPT) across the whole array. In pvlib the aggregation depends on the user's implementation.

Methodology comparison

Differences between single-diode parameterizations are small at clean Standard Test Conditions (STC) and grow at low irradiance, high temperatures, and partial shading, the regimes that dominate real annual yield. The choice of cell temperature model can shift annual yield by half a percentage point or more. IV aggregation resolution is the main differentiator for systems with significant near shading, half-cut modules, or unusual string configurations. Table 1 lists the models each simulator uses.

Table 1: DC conversion modeling approach per simulator.

Numerical results

Results are computed for the six test sites and four system configurations defined in the Setup and test methodology article.

The results below compare module efficiency (DC power divided by the effective irradiance actually reaching the cell) rather than DC output directly. This approach enables comparison of the DC conversion step only, excluding the effects of the optical and shading differences (already covered in Irradiance modeling and Optical losses).

The DC conversion step combines several sub-models, and each simulator implements a different combination, which produces the widest spread of any stage in this article. Solargis Evaluate computes a higher module efficiency than the other simulators in most cases, and the gap is most consistent on tracker configurations for pvlib and PVsyst (Figure 1). Fixed-tilt configurations behave less uniformly: at Las Vegas and Sulov, some simulators yield more than Solargis Evaluate rather than less, and which ones differ between the bifacial and monofacial cases. RMSE is elevated across all fixed-tilt configurations at those two sites and is the largest in this stage (Figure 2), so the disagreement varies strongly with operating conditions rather than acting as a fixed offset. SAM is closest to Solargis Evaluate overall, but produces the single largest deviation in the stage, at Dharan on the bifacial tracker.

Part of the gap between PVsyst and Solargis Evaluate comes from how each handles electrical mismatch under shading. PVsyst resolves shading and IV curves at submodule and string level, so it applies a parameterized mismatch loss to account for shading that falls unevenly across a module or a string. Solargis Evaluate calculates shading and IV curves per cell, so the same effect emerges from the IV curve summing itself and needs no separate assumption. Figure 1 suggests the PVsyst mismatch assumption may be set too high.

Figure 1: Bias in DC conversion efficiency for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the Solargis Evaluate efficiency, by mounting configuration across the six test sites. Efficiency is DC power divided by the effective irradiance reaching the cell.

Figure 2: RMSE in DC conversion efficiency for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the Solargis Evaluate efficiency, by mounting configuration across the six test sites. Efficiency is DC power divided by the effective irradiance reaching the cell.

DC-side losses

This stage covers the losses the simulators deduct between the array output and the inverter input.

Losses included in this stage

Two mechanisms dominate the losses in this stage. Ohmic losses on the DC conductors are the steady contribution: series wiring within strings, string combiner boxes, and the DC trunk cables. These are usually specified as 1% to 2% at STC, which corresponds to a considerably smaller share of annual DC energy, because the current spends most of the year well below its rated value. Inverter clipping is the variable contribution, and in the bifacial configurations it is several times the cable loss.

There are several other effects the simulators consider between the array output and the inverter input. PVsyst deducts electrical shading mismatch, module quality, and bifacial mismatch. In this comparison these losses were included in the DC conversion stage, where the other simulators account for them, so that a fair comparison can be made. Solargis Evaluate, pvlib, and SAM deduct module quality derating and degradation in this stage where they model them. Both these effects are set to zero in this comparison.

DC cable losses modeling

Most simulators use the same underlying resistive relationship, and the difference is how the loss is entered.

• Physical resistance with simulated current (Solargis Evaluate, Solargis Prospect, PVsyst, optionally pvlib) computes across the wiring network from explicit resistance and the simulated current at each time step. This responds correctly to operating conditions. At low irradiance the current is low and the loss is small, at peak generation the loss is correspondingly higher, and it rises with the square of the current rather than in proportion to it.

• An STC percentage proxy is offered by Solargis Evaluate, Solargis Prospect, pvlib, and PVsyst as a simpler input, with the percentage at STC back-calculated to a resistance.

• Aggregated loss factor is the approach in SAM, which bundles DC wiring with mismatch, diodes and connections, tracking error, nameplate rating, bifacial electrical mismatch, and power optimizer loss into a single DC losses factor per sub-array. The factor is applied as a constant multiplier at each time step, so the deduction does not respond to current.

Clipping

Clipping occurs when the array can deliver more power than the inverter is able to convert. Although this loss is determined by the inverter, it is not a loss inside the inverter. When clipping, the inverter draws less power from the PV modules, moving the array operating point away from the maximum power point. Less power drawn results in lower current in the DC network, so the cable losses fall at the same time. Since both effects happen upstream of the inverter input, the clipping loss is accounted for in this stage.

Clipping triggers depend on how completely a simulator models the inverter operating envelope.

• Comprehensive envelope handling in Solargis Evaluate, PVsyst, and SAM clips at voltage, current, and power limits, with MPPT voltage window enforcement.

• AC-nameplate-only clipping is offered by the Sandia and Anton Driesse models in pvlib, which clip AC output at nameplate but do not enforce voltage windows or current limits. PVWatts enforces neither.

• Hard envelope shutdown is the approach in SolarFarmer: outside the voltage, current, and power bounds the inverter is treated as off.

The test systems in this comparison are small and conservatively designed, and they never reach the inverter voltage or current limits. Clipping in the results below is therefore entirely AC power limitation, and the difference between comprehensive and nameplate-only envelope handling is not exercised. Real projects with wider string voltage ranges or more aggressive electrical design do reach those limits, where the difference becomes material.

For high DC to AC ratio designs, common with single-axis trackers in sunny climates, clipping dominates the DC-side electrical losses, and the difference between comprehensive and nameplate-only handling can amount to several percentage points of clipped energy.

Design-stage features

PVsyst computes DC ohmic losses across the wiring network as seen by each MPPT input, with an internal wiring loss optimization tool. Solargis Evaluate has a default DC cable loss value of 2%, adjustable in the Energy System Designer. Table 2 lists the modeling approach, input method, and clipping behavior for each simulator.

Methodology comparison

Table 2: DC-side losses modeling approach per simulator.

Numerical results

Solargis Evaluate, pvlib, and PVsyst agree closely at this stage. With the exception of SAM on the bifacial tracker configuration, per-configuration median bias stays within half a percent for all simulators, and most site-level values fall inside one percent in either direction (Figure 3). The monofacial configurations clip little, so they isolate the cable loss most cleanly, and there the site-to-site spread is narrowest. The differences are small but not uniform in direction: pvlib and PVsyst compute slightly less total loss than Solargis Evaluate, and SAM slightly more. SAM computes a higher cable loss, which is the expected consequence of applying a constant percentage instead of a current-dependent resistance, because a percentage fixed at standard test conditions overstates the loss at the lower currents that prevail through most of the year.

In the bifacial configurations the clipping loss is much larger, and the agreement between simulators is looser. Clipping is almost entirely a bifacial effect, at roughly 4% of theoretical DC energy against under 1% for the monofacial configurations, because rear-side gain delivers more energy against the fixed inverter power limit (Table 3). Solargis Evaluate and PVsyst agree closely on how much energy is clipped. Both pvlib and PVsyst move from computing slightly smaller DC-side loss than Solargis Evaluate on monofacial to slightly more on bifacial, by roughly half a percentage point, and the spread across sites widens by a factor of two to five. RMSE shows the same effect in its upper tail rather than in its median (Figure 4). The amount clipped follows site irradiance, from 0.4% at Dharan to over 5% at Las Vegas and Pretoria. The agreement between simulators is closest where little energy is clipped and weakest where clipping occurs frequently.

Table 3: Inverter clipping loss per simulator and mounting configuration, as a percentage of theoretical DC energy, averaged across the six test sites.

pvlib does not report clipping as a separate quantity, so the pvlib column is an upper bound that includes the DC cable loss. Subtracting the cable loss implied by the monofacial configurations brings pvlib in line with the other simulators.

SAM computes less clipping than the others, by roughly one percentage point of theoretical DC energy on the bifacial tracker, and this accounts for almost all of its lower DC-side loss on the bifacial configurations. This is a difference in how SAM handles the inverter power limit, not a difference in how it models cable loss. Its median RMSE for bifacial configurations is several times the monofacial value, whereas the medians for pvlib and PVsyst stay level between the two (Figure 4). A small remainder of a few tenths of a percentage point reflects SAM holding the DC current constant during clipped periods, which leaves its cable loss slightly higher than it would otherwise be.

Figure 3: DC-side loss bias for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the array DC output power entering the step, by mounting configuration across the six test sites.

Figure 4: DC-side loss RMSE for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the array DC output power entering the step, by mounting configuration across the six test sites.

Inverter conversion

Inverters convert DC input from the PV strings into AC output at grid frequency. Conversion efficiency varies with DC input voltage, output power level, and ambient or internal temperature. The inverter model determines how accurately these dependencies are captured.

Inverter conversion efficiency accounts for a loss of 1 to 2% of annual generation. Combined with clipping, deducted in the previous stage, the inverter is one of the largest electrical loss contributors.

Clipping is covered in section DC-side losses above. Clipping reduces the power the inverter draws from the modules rather than dissipating power inside the inverter, so the clipped energy is deducted before the inverter input. This stage isolates inverter conversion efficiency alone.

Inverter model

Five model families are used by the compared simulators, and the choice determines how much of the inverter's behavior the simulation can represent.

• Parameter-fitted from measured curves. The Sandia inverter model fits five parameters to efficiency measured across varying DC voltage and output power, and is the most widely validated of the five. It is used by Solargis Evaluate, Solargis Prospect with a simplified parameter set from default presets, pvlib in single and multi-MPPT variants, and SAM through the California Energy Commission (CEC) inverter database.

• Manufacturer curve with a voltage window. The PVsyst grid inverter model reads efficiency curves from the manufacturer .OND component file and enforces an explicit MPP voltage window.

• Measured curves at several voltages. SolarFarmer interpolates CEC measured efficiency curves, three curves of six points each. SAM offers an equivalent part-load curve option.

• Voltage-sensitive alternative. The Anton Driesse model in pvlib suits inverters whose efficiency depends strongly on MPP voltage.

• Efficiency-only. NREL PVWatts, in pvlib and as a simple SAM option, and the maximum-or-weighted efficiency option in SolarFarmer apply a single efficiency with no voltage or current dependency. Fast, but coarse.

The first three families give very similar results for well-characterized inverters in typical operation. Differences grow at operating envelope boundaries and for inverters with strong MPP-voltage dependence.

Temperature derating and startup

PVsyst, SAM, and SolarFarmer derate AC output above defined ambient or internal temperatures. Solargis Evaluate, pvlib, and PVsyst enforce a startup threshold below which the inverter does not deliver AC power, and the minimum DC power bound in SolarFarmer has the same effect. PVsyst additionally supports multi-MPPT and master-slave configurations explicitly. Table 4 lists the inverter model and the derating and startup behavior of each simulator.

Methodology comparison

Table 4: Inverter conversion modeling approach per simulator.

Numerical results

With clipping deducted in the previous stage, this stage isolates conversion efficiency alone, and the simulators agree almost exactly. Bias stays within a third of a percent and RMSE below one percent for every simulator, site, and configuration (Figures 5 and 6). This is among the closest agreement of any stage in the series. PVsyst is marginally closest to Solargis Evaluate, followed by pvlib and then SAM, but the differences are too small to carry practical weight. The result confirms that the inverter conversion models describe efficiency consistently for a well-characterized inverter in normal operation, despite differing in formulation and in parameter source. Where inverter modeling does produce material differences, it does so through clipping and envelope handling rather than through the efficiency curve.

Figure 5: Inverter conversion loss bias for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the DC power at the inverter input, by mounting configuration across the six test sites.

Figure 6: Inverter conversion loss RMSE for pvlib, PVsyst, and SAM against Solargis Evaluate, as a percentage of the DC power at the inverter input, by mounting configuration across the six test sites.

Further reading

Solargis knowledge base

• "Argus electrical simulation overview": Solargis

• "Argus PV simulation chain": Solargis

• "Comparison setup and test methodology": Solargis

• "Electrical modeling of the AC side": Solargis

DC conversion - cell models

• "Improvement and validation of a model for photovoltaic array performance": by De Soto, W., Klein, S. A., Beckman, W. A.

• "Performance assessment of a simulation model for PV modules of any available technology": by Mermoud, A., Lejeune, T.

• "Photovoltaic array performance model" (SAND2004-3535): by King, D. L., Boyson, W. E., Kratochvil, J. A. (link reused from earlier list)

• "From IEC 61853 power measurements to PV system simulations" (SAND2020-3877): by Driesse, A., Stein, J. S.

• "A power-rating model for crystalline silicon PV modules": by Huld, T., Friesen, G., Skoczek, A., Kenny, R., Sample, T., Field, M., Dunlop, E.

• "Simple PV performance equations theoretically well founded on the single-diode model": by Batzelis, E. I.

• "IEC 61853 - Photovoltaic (PV) module performance testing and energy rating": by International Electrotechnical Commission (IEC)

DC conversion - cell temperature

• "Transient weighted moving-average model of photovoltaic module back-surface temperature": by Prilliman, M., Stein, J. S., Riley, D., Tamizhmani, G.

Inverter conversion

• "Performance model for grid-connected photovoltaic inverters" (SAND2007-5036): by King, D., Gonzalez, S., Galbraith, G., Boyson, W.

• "Modeling efficiency of inverters with multiple inputs": by Hansen, C., Johnson, J., Darbali-Zamora, R., Gurule, N.

• "Beyond the curves: modeling the electrical efficiency of photovoltaic inverters": by Driesse, A.

• "PVWatts version 5 manual" (NREL/TP-6A20-62641): by Dobos, A. P.

General simulator comparison

• "Cross-validation of PV system simulation software": by Driesse, A., Patel, N.