We present MSPLib, a library of multistage stochastic programming problems to measure the computational performance of different implementations of stochastic dual dynamic programming (SDDP). The MSPLib contains large real-world instances as well as synthetic problems that are difficult to solve. We use the library to test prevailing implementations, including MSPPy, QUASAR, and SDDP.jl.
Each file in the MSPLib library follows the following convention:
If you want to obtain a particular file - problem model, lattice, first stage solution, bounds plot, etc - check the Problem Instance Chart above for the file name of the document following the convention.