mchlshort / mexnets Goto Github PK

Since running the optimization can take very long times, it may be better to allow the user to terminate the optimization at some point and then still output good results. As opposed to just outputting what has been printed to console.

We need a more systematic and fail-safe way to input these values.

Instead of having the user input the correction factors to initialize the iterative procedure, we should make these automated through the use of a single MINLP and NLP iteration. These will generate the correction factors. This should be a user option as to whether or not to include these auto-generated corrections or to input manually. Saves time for the user as they do not need to generate the initial underestimating correction factors.

A way to generate the SWS, SBS, S&TBS, and T&SBS automatically based on the inputted data. This is a great starting project for someone to learn their way around the software and Python.

Should make sure that the user can select the superstructure options that they want.

A better and more in-depth data validation and error checking system needs to be implemented.

At the moment the second example has very poor solutions and extremely long solve times. There are a number of potential fixes that may work and would be good to get going in Pyomo:

1. Initialize the problem with a smaller subset of the discrete variables to make the B&B less time-consuming
2. Try and get a better implementation for the big-M constraints
3. Check that the model is working properly regarding the lean stream "L" - currently the model does not guarantee that lean process streams are utilized fully. Fixing the L value of the process streams results in infeasible model
4. Input the non-isocompositional mixing equations
5. Try and get BARON or COUENNE working for the problem

These are all potential fixes that would make a good first issue for someone.

Perhaps saving the output as a json is better as we can then process this easily and not have to re-run the model. Dumping all data to json for easier interpretation is one solution.

Currently printing multiple csv documents that log the values between iterations and outputting the values of the best network to the user.

Graphical output would be the best solution. Print tables with optimal exchangers, a diagram of best network and table with all costs summarized.

Need to have a user documentation that explains how to install, how the software works, practice examples, etc.