Comparison of algorithms

How the two algorithms from this chapter also behave under shadowed conditions will be clarified in this chapter. For this purpose, the PV emulator is used again, but with the shadowed configuration of two different current sources and the bypass diodes, for this the setting of the current sources is kept. The time for the experimental frame shall be $t=10min$ for both algorithms and the settings of the PV emulator remain the same, also both algorithms are tested once with and without shadowing. A logging function is added to both algorithms, seen in chapter saving data, which always logs which point on the PV characteristic is being sampled. In order to better nastand the progression and search of the algorithms for the MPP, the operating points are shown together with the characteristic curves from the last chapter. The first experiment is performed with the PO-algorithm and a unshaded solar module, the red operating points are shown in figure. As expected, the algorithm scans up to the global MPP, and oscillates around this point in small steps.

Perturb and Observe basic

image

Perturb and Observe shaded

image

The algorithm can solve this task satisfactorily and shows how easy the search for the MPP can be implemented. However, in another experiment with a shaded solar module, a weakness of this algorithm becomes apparent.

As seen in figure, the algorithm dwells in the local MPP at $V_{pv}=23V$ and will never find the global MPP because its decision making is based on the slope of the PU-characteristic. The global MPP is at $U_{pv}=13V$ and is the much more powerful operating point.

Next, the IU scanner algorithm is tested with the normal IU characteristic of the PV emulator without any shading. As visible in basic figure at the red operating points the MPP of this characteristic curve is well hit with some small deviations, the deviations are attributed to smaller measurement deviations and the heating of the diodes in the PV emulator.

The test with the VI-scanner and a shadowing of the PV emulator shows that by scanning the characteristic line several times, the global more powerful MPP is easily found. The red operating points from shaded figure coincide well this time with the global MPP of the shaded PV emulator.

VI-Scanner basic

image

VI-Scanner shaded

image

In summary, the PO algorithm is clearly inferior to the IU scanner algorithm in terms of the global MPP. In return, the PO can react faster to short-term shifts of the characteristic curve, because here the MPP is not set for the next 60 seconds.


Copyright © 2023 Arne Christian Schmidt. Distributed by an CC BY-NC 4.0.