Variable values are wiped clean after each solve in order to conserve memory. References to previous variable values are now handled through the results parameters, e.g. "r_state". Additionally, initializing of the various sets, parameters etc. is now handled through the "Option clear" command, as I suspect it to be faster than the old method of resetting the values over the defined sets.

Before these changes, GAMS would occupy more and more memory the longer the model was set to run. This resulted in significant reduction of performance during simulations spanning e.g. a whole year with a large system, as every single variable value for each forecast time-step would be kept in memory, regardless of whether it was necessary for the current solve. In order to avoid this, temporary timeseries and variable values are now wiped clean between solves, and the desired results are committed to the results parameter arrays.

The results arrays will also bloat the memory requirements of each solve, but not as much as the unaltered variable values (and bounds). I imagine it would be possible to write these into some external file during the solve loop for safekeeping, and periodically wipe the results arrays as well in order to conserve memory for the calculations proper. However, this writing could slow the loop by itself, and might not be necessary for the current project work.

It seems that the equation values are still left in the debug file, and are presumably still carried along with the solves. Have to see if further changes are required to avoid unnecessary memory use.