I've listened to a few recordings of this, here are my ideas of what is going on. Annotated score is attached. As theoretical analysis is akin to performance interpretation, somebody else may come up with a completely different yet valid explanation.
Using the score provided above, Bar 20-22. Short tonicization into Cm, where the Dm7 and G chords are sourced from. The G chord does not resolve, as it is a half cadence. Immediately return to the home key of Eb.
Bar 28-30. Again a move from Eb to Cm. This time a deceptive cadence from G to Ab.
Bar ??-??. The third time is actually the focus of this group of cadences, it is the arrival into the chorus with the G finally resolving into Cm.
These three cadences work as a compositional unit, and all suggest a resolution to Cm. The first instance leaves the resolution up in air, the second gives us more Cm tonality but avoids the Cm chord in favour of a deceptive resolution to Ab. The third finally gives the ear the resolution from G to Cm and times it with the start of the chorus. This gives the chorus entrance a very prominent place in the harmonic movement with the V-I cadence.
Here's a recording we can listen to hear these three cadences, along with the equivalent bar numbers from the example provided above.
0:46 Half cadence (Bar 20-22)
1:01 Deceptive cadence (Bar 28-30)
1:16 ii-V-I cadence (next page)