RER Renewable Energy Research  français
   Overview       Expertise       Publications       Resources        Photos    
    All       By Topic       Articles       Reports       Non-technical       Presentations    

Verification of the PVToolbox against Monitored Data, WatsunPV, HOMER, and SOMES for the Gwaii Haanas PV System

Michael M.D. Ross
GPCo Inc.

Full Text Report
Link to GPCo Inc
Link to CETC-Varennes

Note on Authorship:

This technical report was authored by the principal of RER Renewable Energy Research when he was an employee of GPCo Inc.


Research conducted on behalf of the Photovoltaics and Hybrid Systems Program at the CETC-Varennes (Natural Resources Canada) with partial funding from the Panel on Energy Research and Development (PERD).


Ross, Michael M. D. Verification of the PVToolbox against Monitored Data, WatsunPV, HOMER, and SOMES for the Gwaii Haanas PV System. Report to CETC-Varennes (Natural Resources Canada). Varennes, Québec, 2003.


Data from the Gwaii Haanas photovoltaic array/battery/genset hybrid power system was used to investigate the accuracy of the PVToolbox in comparison with other simulation tools. The comparison was done over one year, during which the genset was not operated.

The simulation tools that use a time-base of one hour or finer all produce comparable estimates of the daily average insolation in the plane of the array for Gwaii Haanas. While it is in agreement with these tools during the summer months, RETScreen PV suggests much higher insolation levels than these tools for the winter months. For January, for example, its estimate of the average daily insolation in the plane of the array is nearly three times that of the PVToolbox. Watsun PV sometimes calculates spikes at the beginning and end of the day, presumably because it permits unrealistically high beam ratios. PVToolbox and the other tools limit the beam ratio to physically reasonable levels.

The PVToolbox, SOMES, and Watsun PV generally predict similar average daily array output power (i.e., before the charge controller). During the winter months, the PVToolbox slightly underestimates Watsun PV and SOMES. During the winter months, RETScreen PV estimates array output power far in excess of that predicted by the other tools. Homer, on the other hand, generates estimates of the summer array output power that far exceed those of the other tools. This probably stems from Homer’s use of a maximum power point tracker, as opposed to a direct connection, between the array and the battery.

The PVToolbox, SOMES, Homer, and Watsun PV were compared to monitored data for the useful power output of the array (i.e., once the charge controller had limited charge to the battery). On both an annual basis (Table S1) and a monthly basis (Table S2), the PVToolbox was the most accurate tool according to every measure of error.

Data SourceAverage Array Output Power
Relative Error
Monitored Data1087NA
Table S1 Monitored and Simulated Daily Useful Array Output Power (Average over Entire Year)

Error MeasurePVToolboxWatsunPVHomerSOMES
Mean Monthly % Error-0.3%30.8%10.0%-3.8%
Maximum Magnitude Monthly % Error32.5%91.4%54.8%-34.8%
RMS Monthly % Error4.1%11.5%8.3%6.3%
Mean Monthly Absolute Error40 Wh/day128 Wh/day75 Wh/day68 Wh/day
Table S2 Comparison of Errors for the Simulated Useful Array Power Output

Under sunny skies, all simulations appear to greatly overestimate the available array current. This may be a problem with the underlying models, but it is more likely to be due to the particular site conditions at Gwaii Haanas. With coniferous forest to east, west and north, the array likely receives less diffuse irradiation from these directions and may also be partially shaded at certain times of the day.

PVToolbox indicates a much deeper discharge during the winter months than either Homer or SOMES. Part of this is due to the difference in the state of charge at the end of the summer (the PVToolbox indicates that the battery is never fully charged). But a major part of this stems from the PVToolbox underestimating the useful array power during this period at the same timer that Homer and SOMES overestimate it. During the winter, the PVToolbox shows the system being near failure, while the other tools suggest that the system is oversized. In reality, the situation is probably between the two, although the very low voltages exhibited by the real system in late January suggest that the PVToolbox may not be too far off the mark.

[Figure showing relative error of different simulation packages by month.]
Figure S1 Relative Error in the Simulated Average Daily Useful Array Power Output for Each Month

According to the PVToolbox, the battery is never fully charged by the array; rather, it reaches a maximum state-of-charge of about 87%. All other tool indicate that the battery is fully charged. During those times when the other simulation tools indicate that the battery is fully charged, the PVToolbox shows the battery being drawn down slightly during the night, charging up in the first hours of the morning, and then having the current cut when the voltage reaches the temperature adjusted setpoint of 14.0 V; the float voltage (13.5 V plus temperature adjustment) is so low that only a small current enters the battery during the day. If the day was very long, this current might eventually fully charge the battery, but the limited daylight hours never permit the state-of-charge to exceed 87%.

This behaviour predicted by the PVToolbox is quite a radical departure from the other tools. It does, however, seem reasonable based on laboratory tests of the Global Yuasa battery. Furthermore, when the PVToolbox was run with both the bulk charge and float setpoints raised to 15 V, a 100% state of charge was achieved. These setpoints correspond to what was used in the laboratory to achieve full charge. It is very significant and exciting that the PVToolbox may be able to model effects of this type.

While running the five tools used in this comparison it became clear that a major advantage of the PVToolbox as a research tool is its flexibility and open architecture. In numerous areas, the other tools were unable to do exactly what was required, while the PVToolbox could be adjusted to suit our requirements in terms of input data, types of loads, types of output, components included and their configuration. Furthermore, being able to “look” into components to see how they were operating (or why they were crashing) was very helpful with the PVToolbox, but not possible with the other tools. On the other hand, the PVToolbox was orders of magnitude slower than the other tools.

Other than the PVToolbox, only WatsunPV permits the user to specify parameters describing the voltage/current behaviour of the battery. In the simulations discussed in this report, however, it was found that problems with the Watsun PV battery model prevented it from being “calibrated” with charge and discharge curve data from the Global Yuasa battery.

While the PVToolbox model is semi-empirical, it at least attempts to have some basis in the underlying physical reality of the battery. In contrast, the Watsun PV model is largely an exercise in curve-fitting. In many ways it is much simpler than the PVToolbox model, but because it ignores the underlying reality, it makes assumptions that may work reasonably well some of the time, but can not be hoped to work in the general case.