Employee-Scheduler

This Code can generate shift schedules automatically using the CP-Solver from Google OR-Tools.

Run the Code

Needed Packages:

• Google OR-Tools
  • pip install ortools
  • needed to use the CP-Solver
• Prettytable
  • pip install prettytable
  • needed to print solutions to the console
• openpyxl: pip install openpyxl
  • pip install openpyxl
  • needed to read and write Excel files

Install with:

pip install -r requirements.txt

Alternatively, you can install the package with:

pip install ortools prettytable openpyxl

You can find the entry point in src/main.py

Configure your needs

Input Data

You have to change the input_data_creator.py in the src/model.

The input data is saved in two variables weeks as a list of week objects and teams as a list of team objects.

• Weeks describe the business needs and contain which skill you need on a specific day
  • If you want to generate 4 Week make sure the weeks list contain 4 Week objects
    • Each Week contain a list of Day objects. You should add 7 days to this list. If you dont want to generate a full Week this list can be smaller than 7.
      • Each Day object contain a list of shifts. We used 3 shifts with early-, late and nightshift.
        • Each shift object contain a list of skills that have to be done in this shift.
        • Make sure the skill names are the same as in the teams list
  • You have to configure how many days you actually want to generate in the main function in src/main.py
• Teams describe your resources and contain all employees their skills and availability
  • If you don't have separate teams this list has only one entry.
    • Each team object contain a list of employees.
      • Make sure each employee have a unique name
      • An Employee can be a shift manager or not
      • The variable fixed_skills is used to analyse the business needs and specify a not real employee with a default value of true
      • Each Employee contain a list of skills. Make sure the same skill names are used as in the weeks list.

Make sure that each object has its own unique name. This is needed to uniquely address the variables of employees, skills and working hours. Choosing non-unique names can cause incorrect output or errors.

Rules

Currently, the rules calls have to be manually commented and uncommented whether you want to use the rule or not. This can be done in the main file in src/main.py. look up what each rule do and how it works in the src/rule_builder.py file. The Soft-Constraints need a cost value. to set this value you need a bit of experience, you should use the console output to optimize this values see Understanding Outputs.

Implemented rules

Hard-Constraints:

• *Every defined Need in the teams list is assigned
• *Every employee one assigned to a maximum of one shift per day
• *An employee can not be assigned to a shift with a skill he can't fulfill
• *Each employee can only work with his Team members in a shift. An assignments of two employees from different teams in the same shift is not allowed
• *Each employee can only be assigned to a maximum of 5 workdays a week
• Each employee can only be assigned to a maximum of 5 consecutive work days
• Each employee can only be assigned to a maximum of 10 consecutive work days
• *Each employee works in the same shift a week
• *Each employee have a minimum of two shifts (16 hours) rest time between work shifts
• *Each employee works in a weekly shift cycle Early -> Late -> Night. An Employee worked in Early shifts the first week will be assigned to late shifts the next week and to Night shifts the 3rd week.
• *At every shift there is minimum one shift manager present.
• A function to add absence manually
• A function to add vacation manually
• Each employee works the night shifts in consecutive days
• Every employee always performs the same task every week
• Each employee is ill at a specific time. Based on the parameters number_intervals and number_ill_per_interval(used for simulations)

Soft-Constraints:

• *(3)Each employee should work in a row in consecutive work days
• *(56)Each employee should work night shifts in a row in consecutive work days
• *(10)All employees should work in the same amount og night shifts
• *(10)All employees should work in the same amount og night shifts
• *(10000)Each employee should work a maximum of 5 days a row in consecutive work days
• Each employee should work a maximum of 10 days a row in consecutive work days
• Each employee should have as less skills as possible (used for simulations with the parameter fixed_skills)
• Use as fewer employees as possible (used for simulations with the parameter fixed_skills)

With * marked rules are used for the calculations in the shown example schedules. Additional info for Soft-Constraints after * with the set costs.

Understanding Outputs

This is a generated shift schedule.
in the 2 column are the names of the employees the 1 column shows in which team an employee is and the 3 column shows which skills an employee have.
Shift manager are printed in bold. Following the employee row shows on which day in which shift this employee should work M = Morning, Morning shift, A = Afternoon, Late shift, N = Night, Night shift.
Under the shift you can see which skills this employee should do in this shift.

This is the console output for the shift schedule seen before.
At the Top you see it is the 447th found solution. The Cp-Solver will find valid schedules each better than the schedule before.
This table shows the current cost values for the Soft-Constraints. For example the column sh distr shows how many shifts each employee is assigned to. In the next column you see the cost for this rule. Each cost ist calculated with number_of_assigned_shifts * shift_distribution_cost^2. or generally x * x_cost^2.
All costs are added up and saved to a variable this variable will be minimized by the CP-Solver.

Runtime

This graph shows the time related to the objective value.
The objective value is the sum of all cost for each Soft-Constraint.
You can see that the calculation takes some time. After about 9 hours the solver found a good solution that can be used.
The runtime can be drastically dropped if you don't use that much Soft-Constraints.
Based on previous calculations, the runtime for a schedule with no Soft-Constraint take about 1 to 5 minutes.
A calculation with one Soft-Constraint took about 10 to 30 minutes. You have to build your own experience in runtime for you own schedule problem. Only the given dataset in input_data_creator.py was tested.
We have observed that the more Hard-Constraints and the fewer Soft-Constraints are used, the better the runtime.

This graph shows the runtime for 4 consecutive months. So one month was calculated, based on this plan the second month was generated and so on.