Shift Scheduling Problem

Consider the following shift scheduling problem, which aims to find a schedule that minimizes the total worker cost.

  • There are 6 workers and a planning horizon of 31 days, from day 1 to day 31. For simplicity, we assume that all workers are off on day 0 and day 32.
  • Exactly 4 workers must be scheduled on each day from day 1 to day 31.
  • The following constraints must be satisfied for each worker:
    • works for either 20 or 21 days,
    • works no more than 6 consecutive days,
    • works no fewer than 3 consecutive days,
    • has no isolated day off; days off must be consecutive.

QUBO formulation for the shift scheduling problem

The QUBO formulation uses a $6\times 33$ matrix of binary variables $X=(x_{i,j})$ ($0\leq i\leq 5, 0\leq j\leq 32$) where worker $i$ works on day $j$ if and only if $x_{i,j}=1$.

Since all workers are off on day 0 and day 32, we fix

\[\begin{aligned} x_{i,0}=x_{i,32}=0 & &(0\leq i\leq 5). \end{aligned}\]

The constraints are formulated as follows.

Daily staffing constraint

Exactly 4 workers must be scheduled on each day:

\[\begin{aligned} \sum_{i=0}^{5} x_{i,j} = 4& &(1\leq j\leq 31) \end{aligned}\]

Total working days constraint

Each worker must work for either 20 or 21 days:

\[\begin{aligned} 20\leq \sum_{j=0}^{32} x_{i,j} \leq 21& &(0\leq i\leq 5) \end{aligned}\]

Maximum consecutive working days constraint

No worker may work for more than 6 consecutive days:

\[\begin{aligned} x_{i,j}x_{i,j+1}x_{i,j+2}x_{i,j+3}x_{i,j+4}x_{i,j+5}x_{i,j+6} = 0 & &(0\leq i\leq 5, 0\leq j\leq 26)\\ \end{aligned}\]

Minimum consecutive working days constraint

Each working period must consist of at least 3 consecutive working days:

\[\begin{aligned} \bar{x}_{i,j}x_{i,j+1}x_{i,j+2}\bar{x}_{i,j+3} = 0 & &(0\leq i\leq 5, 0\leq j\leq 29)\\ \bar{x}_{i,j}x_{i,j+1}\bar{x}_{i,j+2} = 0 & & (0\leq i\leq 5, 0\leq j \leq 30) \end{aligned}\]

No isolated day off constraint

No worker may have a single day off between two working days:

\[\begin{aligned} x_{i,j}\bar{x}_{i,j+1}x_{i,j+2} = 0 & &(0\leq i\leq 5, 0\leq j\leq 30)\\ \end{aligned}\]

Total worker cost

Let $C=(c_i)$ be a cost vector, where $c_i$ denotes the daily cost of assigning worker $i$. The total worker cost is formulated as:

\[\begin{aligned} \sum_{i=0}^5\sum_{j=0}^{32} c_i x_{i,j} \end{aligned}\]

This objective function is minimized subject to the constraints described above.

QUBO++ program for the shift scheduling

The shift scheduling problem defined above can be formulated and solved using QUBO++ as follows:

#include <qbpp/qbpp.hpp>
#include <qbpp/easy_solver.hpp>

int main() {
  const size_t days = 31;
  const auto worker_cost = qbpp::int_array({13, 13, 12, 12, 11, 10});
  const size_t workers = worker_cost.size();

  auto x = qbpp::var("x", workers, days + 2);

  auto workers_each_day = qbpp::vector_sum(x, 0);
  auto each_day_4_workers = qbpp::toExpr(0);
  for (size_t j = 1; j <= days; ++j) {
    each_day_4_workers += workers_each_day[j] == 4;
  }

  auto workers_working_days = qbpp::vector_sum(x);
  auto work_20_21_days = qbpp::sum(20 <= workers_working_days <= 21);

  auto no_more_than_6_consecutive_working_days = qbpp::toExpr(0);
  for (size_t w = 0; w < workers; ++w) {
    for (size_t j = 0; j <= days - 5; ++j) {
      no_more_than_6_consecutive_working_days +=
          x[w][j] * x[w][j + 1] * x[w][j + 2] * x[w][j + 3] * x[w][j + 4] *
          x[w][j + 5] * x[w][j + 6];
    }
  }
  auto no_less_than_3_consecutive_working_days = qbpp::toExpr(0);
  for (size_t w = 0; w < workers; ++w) {
    for (size_t j = 0; j < days - 1; ++j) {
      no_less_than_3_consecutive_working_days +=
          ~x[w][j] * x[w][j + 1] * x[w][j + 2] * ~x[w][j + 3];
    }
    for (size_t j = 0; j < days; ++j) {
      no_less_than_3_consecutive_working_days +=
          ~x[w][j] * x[w][j + 1] * ~x[w][j + 2];
    }
  }

  auto no_single_day_off = qbpp::toExpr(0);
  for (size_t w = 0; w < workers; ++w) {
    for (size_t j = 0; j <= days - 1; ++j) {
      no_single_day_off += x[w][j] * ~x[w][j + 1] * x[w][j + 2];
    }
  }

  auto total_worker_cost = qbpp::sum(worker_cost * workers_working_days);

  auto constraints = work_20_21_days + no_less_than_3_consecutive_working_days +
                     no_more_than_6_consecutive_working_days +
                     no_single_day_off + each_day_4_workers;
  auto f = total_worker_cost + 10000 * constraints;

  qbpp::MapList ml;
  for (size_t i = 0; i < workers; ++i) {
    ml.push_back({x[i][0], 0});
    ml.push_back({~x[i][0], 1});
    ml.push_back({x[i][days + 1], 0});
    ml.push_back({~x[i][days + 1], 1});
  }
  f.simplify_as_binary();

  auto g = qbpp::replace(f, ml);
  g.simplify_as_binary();
  workers_working_days.replace(ml);

  auto solver = qbpp::easy_solver::EasySolver(g);
  auto sol = solver.search({{"time_limit", 5.0}, {"target_energy", 0}});
  for (size_t i = 0; i < workers; ++i) {
    std::cout << "Worker " << i << ": " << sol(workers_working_days[i])
              << " days worked: ";
    for (size_t j = 1; j <= days; ++j) {
      std::cout << sol(x[i][j]);
    }
    std::cout << std::endl;
  }
  std::cout << "Workers each day        : ";
  for (size_t d = 1; d <= days; ++d) {
    std::cout << sol(workers_each_day[d]);
  }
  std::cout << std::endl;

  auto sol_f = qbpp::Sol(f).set(ml).set(sol);

  std::cout << "Total worker cost: " << sol_f(total_worker_cost) << std::endl;
  std::cout << "Constraints violations: " << sol_f(constraints) << std::endl;
}

In this program, the variables and expressions are defined as follows:

  • x: A $6\times 33$ matrix of binary variables,
  • workers_each_day: An array containing the column-wise sums of x, representing the number of workers assigned to each day.
  • each_day_4_workers: A constraint expression that attains a minimum value of 0 if and only if exactly four workers are assigned to each day.
  • workers_working_days: An array of row-wise sums of x, representing the total number of working days for each worker.
  • work_20_21_days: A constraint expression that attains a minimum value of 0 if and only if each worker works for either 20 or 21 days.
  • no_more_than_6_consecutive_working_days: A constraint expression that attains a minimum value of 0 if and only if no worker works for 7 or more consecutive days.
  • no_less_than_3_consecutive_working_days: A constraint expression that attains a minimum value of 0 if and only if every working period consists of at least 3 consecutive working days.
  • no_single_day_off: A constraint expression that attains a minimum value of 0 if and only if no worker has a single day off between two working days.
  • constraints: The sum of all constraint expressions.
  • total_worker_cost: An expression representing the total worker cost.

QUBO construction and solution

By summing total_worker_cost and constraints with a penalty factor of 10000, we obtain an expression f, which represents a QUBO formulation of the shift scheduling problem.

A qbpp::MapList object ml is used to fix the values of the variables corresponding to day 0 and day 32. Applying the qbpp::replace() function to f with ml yields a new expression g.

The Easy Solver is then applied to g, and the resulting solution is stored in sol. The obtained solution is as follows:

Worker 0: 20 days worked: 0001111001110011111001111001111
Worker 1: 20 days worked: 1111001111110001111110011110000
Worker 2: 21 days worked: 0000111100111110011111100111111
Worker 3: 21 days worked: 1111110011111100111000111111000
Worker 4: 21 days worked: 1111100111001111000111000111111
Worker 5: 21 days worked: 1110011110001111100111111000111
Workers each day        : 4444444444444444444444444444444
Total worker cost: 1465
Constraints violations: 0

We observe that a feasible shift schedule with a total worker cost of 1465 is obtained, and all constraints are satisfied.

シフトスケジューリング問題

以下のシフトスケジューリング問題を考えます。この問題は、総労働者コストを最小化するスケジュールを見つけることを目的とします。

  • 6人の労働者と、1日目から31日目までの31日間の計画期間があります。簡単のため、すべての労働者は0日目と32日目は休みであると仮定します。
  • 1日目から31日目まで、各日にちょうど4人の労働者をスケジュールする必要があります。
  • 各労働者について以下の制約を満たす必要があります:
    • 20日間または21日間勤務する、
    • 連続勤務は6日以内、
    • 連続勤務は3日以上、
    • 孤立した休日がない(休日は連続でなければならない)。

シフトスケジューリング問題のQUBO定式化

QUBO定式化では、$6\times 33$ のバイナリ変数行列 $X=(x_{i,j})$ ($0\leq i\leq 5, 0\leq j\leq 32$) を使用します。労働者 $i$ が $j$ 日目に勤務するのは $x_{i,j}=1$ のときかつそのときに限ります。

すべての労働者は0日目と32日目は休みであるため、次を固定します:

\[\begin{aligned} x_{i,0}=x_{i,32}=0 & &(0\leq i\leq 5). \end{aligned}\]

制約は以下のように定式化されます。

日次配置制約

各日にちょうど4人の労働者をスケジュールしなければなりません:

\[\begin{aligned} \sum_{i=0}^{5} x_{i,j} = 4& &(1\leq j\leq 31) \end{aligned}\]

総勤務日数制約

各労働者は20日間または21日間勤務しなければなりません:

\[\begin{aligned} 20\leq \sum_{j=0}^{32} x_{i,j} \leq 21& &(0\leq i\leq 5) \end{aligned}\]

最大連続勤務日数制約

どの労働者も7日以上連続で勤務してはなりません:

\[\begin{aligned} x_{i,j}x_{i,j+1}x_{i,j+2}x_{i,j+3}x_{i,j+4}x_{i,j+5}x_{i,j+6} = 0 & &(0\leq i\leq 5, 0\leq j\leq 26)\\ \end{aligned}\]

最小連続勤務日数制約

各勤務期間は少なくとも3日連続の勤務日で構成されなければなりません:

\[\begin{aligned} \bar{x}_{i,j}x_{i,j+1}x_{i,j+2}\bar{x}_{i,j+3} = 0 & &(0\leq i\leq 5, 0\leq j\leq 29)\\ \bar{x}_{i,j}x_{i,j+1}\bar{x}_{i,j+2} = 0 & & (0\leq i\leq 5, 0\leq j \leq 30) \end{aligned}\]

孤立休日禁止制約

どの労働者も2つの勤務日の間に1日だけの休日を持ってはなりません:

\[\begin{aligned} x_{i,j}\bar{x}_{i,j+1}x_{i,j+2} = 0 & &(0\leq i\leq 5, 0\leq j\leq 30)\\ \end{aligned}\]

総労働者コスト

$C=(c_i)$ をコストベクトルとし、$c_i$ は労働者 $i$ を配置する1日あたりのコストを表します。 総労働者コストは次のように定式化されます:

\[\begin{aligned} \sum_{i=0}^5\sum_{j=0}^{32} c_i x_{i,j} \end{aligned}\]

この目的関数を上記の制約のもとで最小化します。

シフトスケジューリングのQUBO++プログラム

上で定義したシフトスケジューリング問題は、QUBO++を用いて以下のように定式化・求解できます:

#include <qbpp/qbpp.hpp>
#include <qbpp/easy_solver.hpp>

int main() {
  const size_t days = 31;
  const auto worker_cost = qbpp::int_array({13, 13, 12, 12, 11, 10});
  const size_t workers = worker_cost.size();

  auto x = qbpp::var("x", workers, days + 2);

  auto workers_each_day = qbpp::vector_sum(x, 0);
  auto each_day_4_workers = qbpp::toExpr(0);
  for (size_t j = 1; j <= days; ++j) {
    each_day_4_workers += workers_each_day[j] == 4;
  }

  auto workers_working_days = qbpp::vector_sum(x);
  auto work_20_21_days = qbpp::sum(20 <= workers_working_days <= 21);

  auto no_more_than_6_consecutive_working_days = qbpp::toExpr(0);
  for (size_t w = 0; w < workers; ++w) {
    for (size_t j = 0; j <= days - 5; ++j) {
      no_more_than_6_consecutive_working_days +=
          x[w][j] * x[w][j + 1] * x[w][j + 2] * x[w][j + 3] * x[w][j + 4] *
          x[w][j + 5] * x[w][j + 6];
    }
  }
  auto no_less_than_3_consecutive_working_days = qbpp::toExpr(0);
  for (size_t w = 0; w < workers; ++w) {
    for (size_t j = 0; j < days - 1; ++j) {
      no_less_than_3_consecutive_working_days +=
          ~x[w][j] * x[w][j + 1] * x[w][j + 2] * ~x[w][j + 3];
    }
    for (size_t j = 0; j < days; ++j) {
      no_less_than_3_consecutive_working_days +=
          ~x[w][j] * x[w][j + 1] * ~x[w][j + 2];
    }
  }

  auto no_single_day_off = qbpp::toExpr(0);
  for (size_t w = 0; w < workers; ++w) {
    for (size_t j = 0; j <= days - 1; ++j) {
      no_single_day_off += x[w][j] * ~x[w][j + 1] * x[w][j + 2];
    }
  }

  auto total_worker_cost = qbpp::sum(worker_cost * workers_working_days);

  auto constraints = work_20_21_days + no_less_than_3_consecutive_working_days +
                     no_more_than_6_consecutive_working_days +
                     no_single_day_off + each_day_4_workers;
  auto f = total_worker_cost + 10000 * constraints;

  qbpp::MapList ml;
  for (size_t i = 0; i < workers; ++i) {
    ml.push_back({x[i][0], 0});
    ml.push_back({~x[i][0], 1});
    ml.push_back({x[i][days + 1], 0});
    ml.push_back({~x[i][days + 1], 1});
  }
  f.simplify_as_binary();

  auto g = qbpp::replace(f, ml);
  g.simplify_as_binary();
  workers_working_days.replace(ml);

  auto solver = qbpp::easy_solver::EasySolver(g);
  auto sol = solver.search({{"time_limit", 5.0}, {"target_energy", 0}});
  for (size_t i = 0; i < workers; ++i) {
    std::cout << "Worker " << i << ": " << sol(workers_working_days[i])
              << " days worked: ";
    for (size_t j = 1; j <= days; ++j) {
      std::cout << sol(x[i][j]);
    }
    std::cout << std::endl;
  }
  std::cout << "Workers each day        : ";
  for (size_t d = 1; d <= days; ++d) {
    std::cout << sol(workers_each_day[d]);
  }
  std::cout << std::endl;

  auto sol_f = qbpp::Sol(f).set(ml).set(sol);

  std::cout << "Total worker cost: " << sol_f(total_worker_cost) << std::endl;
  std::cout << "Constraints violations: " << sol_f(constraints) << std::endl;
}

このプログラムでは、変数と式は以下のように定義されています:

  • x: $6\times 33$ のバイナリ変数行列。
  • workers_each_day: x の列方向の和を含む配列で、各日に配置された労働者数を表します。
  • each_day_4_workers: 各日にちょうど4人の労働者が配置されているときかつそのときに限り最小値0をとる制約式。
  • workers_working_days: x の行方向の和の配列で、各労働者の総勤務日数を表します。
  • work_20_21_days: 各労働者が20日間または21日間勤務しているときかつそのときに限り最小値0をとる制約式。
  • no_more_than_6_consecutive_working_days: どの労働者も7日以上連続で勤務していないときかつそのときに限り最小値0をとる制約式。
  • no_less_than_3_consecutive_working_days: すべての勤務期間が少なくとも3日連続の勤務日で構成されているときかつそのときに限り最小値0をとる制約式。
  • no_single_day_off: どの労働者も2つの勤務日の間に1日だけの休日を持っていないときかつそのときに限り最小値0をとる制約式。
  • constraints: すべての制約式の和。
  • total_worker_cost: 総労働者コストを表す式。

QUBO構築と求解

total_worker_costconstraints をペナルティ係数10000で加算することにより、シフトスケジューリング問題のQUBO定式化を表す式 f を得ます。

qbpp::MapList オブジェクト ml は、0日目と32日目に対応する変数の値を固定するために使用されます。 qbpp::replace() 関数を fml を適用することで、新しい式 g が得られます。

次に、Easy Solver が g に適用され、得られた解が sol に格納されます。 得られた解は以下のとおりです:

Worker 0: 20 days worked: 0001111001110011111001111001111
Worker 1: 20 days worked: 1111001111110001111110011110000
Worker 2: 21 days worked: 0000111100111110011111100111111
Worker 3: 21 days worked: 1111110011111100111000111111000
Worker 4: 21 days worked: 1111100111001111000111000111111
Worker 5: 21 days worked: 1110011110001111100111111000111
Workers each day        : 4444444444444444444444444444444
Total worker cost: 1465
Constraints violations: 0

総労働者コスト 1465 の実行可能なシフトスケジュールが得られ、すべての制約が満たされていることがわかります。