A Novel Method to Reduce Event Variables in Continuous-time Formulation for Short-term Scheduling

GUILLERMO A. DURAND; J. ALBERTO BANDONI

Glasgow (Escocia)

Congreso; 7th World Congress of Chemical Engineering; 2005

European Federation of Chemical Engineering - Institution of Chemical Engineers

There are many cases in the chemical and pharmaceutical industries where a large-scale scheduling problem has to be solved in order to accommodate production to many restrictions, such as due dates, limited equipment and resources. Obtaining an optimal solution can be a daunting task, even with high computing power. Continuous-time formulation based on events, called Resource Task Network (RTN), developed by Ierapetritou and Floudas (1998), has proven to be an effective method to face short-term scheduling problem, rendering MILP models with good computer performance. The size of the models is determined mainly by the quantity of tasks, equipment and event variables (time periods to be scheduled within the horizon time). However, when event variables surpass the number of 20 and the number of task is around 10-20, the problem can become computationally expensive. This work presents a new method, consisting on a set of constraints that takes profit of the production recipe to fix the value of some event variables, thus reducing their quantity and improving the integrality gap, leading to a better performance in short-term scheduling problems. As this paper will show, this technique can be used within the approaches by Ierapetritou and Floudas (1998), and Maravelias and Grossmann (2003). There are many cases in the chemical and pharmaceutical industries where a large-scale scheduling problem has to be solved in order to accommodate production to many restrictions, such as due dates, limited equipment and resources. Obtaining an optimal solution can be a daunting task, even with high computing power. Continuous-time formulation based on events, called Resource Task Network (RTN), developed by Ierapetritou and Floudas (1998), has proven to be an effective method to face short-term scheduling problem, rendering MILP models with good computer performance. The size of the models is determined mainly by the quantity of tasks, equipment and event variables (time periods to be scheduled within the horizon time). However, when event variables surpass the number of 20 and the number of task is around 10-20, the problem can become computationally expensive. This work presents a new method, consisting on a set of constraints that takes profit of the production recipe to fix the value of some event variables, thus reducing their quantity and improving the integrality gap, leading to a better performance in short-term scheduling problems. As this paper will show, this technique can be used within the approaches by Ierapetritou and Floudas (1998), and Maravelias and Grossmann (2003). There are many cases in the chemical and pharmaceutical industries where a large-scale scheduling problem has to be solved in order to accommodate production to many restrictions, such as due dates, limited equipment and resources. Obtaining an optimal solution can be a daunting task, even with high computing power. Continuous-time formulation based on events, called Resource Task Network (RTN), developed by Ierapetritou and Floudas (1998), has proven to be an effective method to face short-term scheduling problem, rendering MILP models with good computer performance. The size of the models is determined mainly by the quantity of tasks, equipment and event variables (time periods to be scheduled within the horizon time). However, when event variables surpass the number of 20 and the number of task is around 10-20, the problem can become computationally expensive. This work presents a new method, consisting on a set of constraints that takes profit of the production recipe to fix the value of some event variables, thus reducing their quantity and improving the integrality gap, leading to a better performance in short-term scheduling problems. As this paper will show, this technique can be used within the approaches by Ierapetritou and Floudas (1998), and Maravelias and Grossmann (2003). There are many cases in the chemical and pharmaceutical industries where a large-scale scheduling problem has to be solved in order to accommodate production to many restrictions, such as due dates, limited equipment and resources. Obtaining an optimal solution can be a daunting task, even with high computing power. Continuous-time formulation based on events, called Resource Task Network (RTN), developed by Ierapetritou and Floudas (1998), has proven to be an effective method to face short-term scheduling problem, rendering MILP models with good computer performance. The size of the models is determined mainly by the quantity of tasks, equipment and event variables (time periods to be scheduled within the horizon time). However, when event variables surpass the number of 20 and the number of task is around 10-20, the problem can become computationally expensive. This work presents a new method, consisting on a set of constraints that takes profit of the production recipe to fix the value of some event variables, thus reducing their quantity and improving the integrality gap, leading to a better performance in short-term scheduling problems. As this paper will show, this technique can be used within the approaches by Ierapetritou and Floudas (1998), and Maravelias and Grossmann (2003). There are many cases in the chemical and pharmaceutical industries where a large-scale scheduling problem has to be solved in order to accommodate production to many restrictions, such as due dates, limited equipment and resources. Obtaining an optimal solution can be a daunting task, even with high computing power. Continuous-time formulation based on events, called Resource Task Network (RTN), developed by Ierapetritou and Floudas (1998), has proven to be an effective method to face short-term scheduling problem, rendering MILP models with good computer performance. The size of the models is determined mainly by the quantity of tasks, equipment and event variables (time periods to be scheduled within the horizon time). However, when event variables surpass the number of 20 and the number of task is around 10-20, the problem can become computationally expensive. This work presents a new method, consisting on a set of constraints that takes profit of the production recipe to fix the value of some event variables, thus reducing their quantity and improving the integrality gap, leading to a better performance in short-term scheduling problems. As this paper will show, this technique can be used within the approaches by Ierapetritou and Floudas (1998), and Maravelias and Grossmann (2003). Debido a errores del Comité Organizador el trabajo presentado no fue incluido en las actas del congreso. Como confirmación de su aceptación, se adjuntan en el archivo .pdf el e-mail de aceptación y una copia del programa on-line del congreso.