This concept in general as it applies to microphones is known as beamforming, and is most commonly applied to sound and radio waves. It is mostly aimed at wave forms that carry signals (ie change significantly over time,) but similar concepts could be used in your example.
In general terms, beamforming on the receiving side comes in two flavors. Either you know the location of the signal you want to receive, and you phase-align your sensors to be more sensitive to that direction, or you know what your signal 'looks' like, and you can locate it based on phase differences in your array of sensors.
You are proposing the second application, so the difficult thing for you is knowing how to distinguish one light source from all of the others. For example, if you had only one light in a dark room, it would be easy to identify your target signal. In your situation with multiple lights, if they all produce the same wavelengths, it will be nearly impossible to distinguish them if they are all on at the same time. If they produce different colors, you could possibly distinguish them, but you would still have a different problem.
In order for beamforming based on phase shift to work, you have to be able to identify each received waveform, and index them against each other. There are two problems here. The first, inherent problem is that conventional light bulbs produce essentially the same wave form forever, dimming very slowly, until they burn out. The problem with this is that since the waveform has no distinct changes in it, you can't index it to find time delays. when you see the same signal from two sensors, you might notice that they are 180 degrees out of phase, but you wouldn't be able to tell if they are just half a wavelength different, or 1000.5 wavelengths different. In order to tell how much more the signal has been delayed traveling to one sensor, you need a distinguishable change in the waveform (for example the light getting brighter or changing color.) The second, more pragmatic problem is that I'm not aware of any photodiode that is so responsive that you could actually observe the waveform of light. I'm not an electronics expert though, so they may exist.
So we need a strategy to make the waveform of the light source less uniform, and ideally to add a much lower frequency signal which is easier to detect. Assuming you have control over the lights, this should be easy to do. At its simplest, you could turn on one light at a time, and watch for the first light to hit your sensors. Since you know how far apart your sensors are, and you can easily identify the relative delays between the first and last sensor the light hits, you can use Multilateration to locate the source. If it's not an option to only have one source on at a time, there are other ways to make the source identifiable such as perhaps turning it from 100% brightness to 80% brightness or changing it from white to blue at intervals. How significant these changes will be, and how robustly your system will detect them will depend on the sensitivity of your instruments and the degree of background noise.