Mod enhancement: labor rules in workforce mod

[ronaldvdv]

Workforce scheduling is interesting and complex especially when labor rules come into play. Different countries, industries, labor unions and individual companies each have their own rules so supporting all possible rule structures will be too ambitious. However, by including a generic framework that supports several different rule types, we can still provide something that is easy to use and understand but also showcases how any labor rule can be implemented.

Example rules
Below is an inventory of different rules that happen in reality. Not all of them may be suitable for inclusion in this optimod.

1. Junior employee can have no more than 5 night shifts in a row
2. No more than 3 evening or night shifts per (rolling or calendar) week
3. No more than 2 weekend shifts per (rolling or calendar) month
4. No more than 2 weekends per month having at least one shift
5. 4. At least 14 hours of rest after a shift of type X
6. At least one block of 60 hours rest per (rolling or calendar) week
7. A series of night shifts should be followed by at least 48 hours of rest

Max shift count
Rules 1-3 are similar in the way they can be represented in the MIP.

• Each rule applies to a subset or possibly all employees; the rule is checked for employees individually
• Each rule applies to multiple subsets of days in the planning horizon (e.g each period of 7 days, or all weekend days in a month)
• For each applicable employee an subset of days, the rule is to be validated against a subset of shifts in those days
• Rule 4 is complex because we'd need to model whether any shift is assigned in each weekend.

Min rest time
Rules 5-7 require the addition of shift start and end times. Once we have those:

• Rule 5 can be checked by generating all combinations of shifts that are less than 14 hours apart and impose a constraint that only one of those can be assigned to any employee
• Rule 6 and 7 require more work, because the rest requirement does not apply after every assigned shift, so we need additional variables and constraint to properly model them.

I'd argue that from the above set, we should at least have rules 1-3 and rule 5. Curious to hear other opinions!

[ronaldvdv]

We could see if we can roughly match the features of https://developers.google.com/optimization/service/reference/rest/v1/scheduling/solveShiftScheduling#schedulingconstraint