If you're looking to measure things near or below the limits of human hearing, I'd recommend looking at microphones designed for infrasound rather than low sensitivity microphones designed to go into the ultrasonic range. PCB and Brüel & Kjær are the big names here.
Measuring infrasound is pretty difficult, unfortunately. Most audio gear has a high pass filter, which makes sense for most uses but not this. Also, pseudo-noise from air currents is often a huge source of error.
Edit: Also, note that I mean doing hardware amplification of the electrical signal in the wire coming directly from the motion of the magnet of the speaker.
Microphone size doesn't really matter that much for low frequency sound. The Earthworks M50 suggested above? I wouldn't recommend it for this application because if you're going to spend a lot of money, it might as well be on something well-suited to the application. However, the approach it takes works and can be done for much less money. Cheap omnidirectional electrets go down pretty far[1], and you don't need the ruler flat response of the M50. You just need amplification circuitry that doesn't filter out the low frequency stuff you want.
DSP is commonly used to correct the frequency response of microphones. However, measuring frequency response is difficult and expensive.
[1] http://volcanomodels.sr.unh.edu/jbj/MICROPHONES/microphone_l...
However, please read my post above on what I actually found.
And yes, I don't expect comparing amplitude to be useful--sound localization almost always uses TDOA/phase difference for that reason.
On Youtube, it yields several people video recording them, from Russia to Asia: https://www.youtube.com/results?search_query=strange+sounds+...
This reminds me of a YouTube video I found from someone who thought they found a UFO near the Wal-Mart parking lot one night. It was the red blinking light on top of a water tower. People are very good at convincing themselves of things.
We only have their word to go by, and it's not much use given their leaning toward supernatural and conspiracy.
Next step would probably to get a few of these, see if you can record the hum on a few of them, and then start experimenting.
The theory behind triangulation is simple. But in practice, it's probably really hard to make sure all microphones have the same sensitivity. The standard approach depends on sounds being quieter when they are far away, but if one mic is naturally quieter (deeper inside the house, or more absorbant material, or just more obstacles in the way) that no longer works.
The best approach would be using a phase difference to measure. Exploiting the delay due to the speed of sound. That means you need some way to synchronize the recordings. Presuming a 20hz signal, you'd need your synchronization to be better than 0.05 seconds. This also requires the hum waveform to be easy to match. That is, you'd need to be able to tell the phase offset by matching parts of the sound.
You could also try a 'big data' approach, just cataloging all reported cases and looking for patterns. That requires dealing with very noisy data though, as reported cases will be inconsistent. This is due to people not always being home, not everyone hearing it, and no verification of reports.
edit: Interesing idea: use the 'bit data' approach together with wind speed and direction. The sound should be easier to hear when the wind blows it towards you. With enough data and corresponding wind data, you might have something. Does require compensating for correlation between time of day and the wind. If wind is west in the night, and less people hear things when the wind is west, that might just be because they are sleeping.