How can one tell which PDE's are, roughly speaking, perhaps more interesting to analyse? 

Physical motivation is one reason. For example, the KdV $u_t+u_{xxx} - 6uu_x=0$ for a function $u:\mathbb R\times\mathbb R\to\mathbb R$ is a model for many physical systems, such as the propagation of shallow water waves along water bodies with low height. It is also a completely integrable PDE, which makes its analysis more tempting. But putting aside complete integrability (many studied PDE do not have completely integrable variants) why not study the derivative quasilinear equation $u_t + u_{xxx} - 6uu_{xxx}$ or the another derivative semilinear equation $u_t + u_{xxx} - 6uu_{xx}=0$?

Aside from physical motivation, how can one pick PDE to analyse?

Edit: Just to be clear, I am asking this from the perspective of research. When I say "analyse," I mean "do research on."