Iterative resource allocation (solve)

[jetuk]

There has been a request to add the ability to iterative re-solve the resource allocation (LP) procedure in each time-step. This would be for cases where the allocation (e.g. a parameter) is dependent on the current time-step's flow. Currently, this is only possible by using previous time-step's data as an approximation, or estimating current time-step's expected flow by manually calculating flows in a custom parameter. The latter approach is difficult and can not know exactly what the allocation will be.

Therefore, one method would be to re-solve the LP with updated information / estimates from the first solve. And continue this iteration until convergence (or a limit is reached).

Requirements

• A new Metric variant that refers to current flow, or makes an explicit separate between previous and current flow.
• Implement the solve iteration loop.
• Handle iterative state changes.
  • Current mutable state is updated during solve.
  • Need to handle storing previous time-step's flows.
  • Storing an iterative sequence of flows from each iteration (or could just be current and last iteration?).
  • Rewinding storage changes or only updating storage once iteration is complete.
• Estimate convergence as the difference between flows on successive time-steps? Perhaps only those flows which are being used in Metrics.
• Solver settings to configure the resolve feature, maximum iterations, convergence, etc.