![]() |
|||
![]()
|
![]() |
![]() |
![]() |
2. REPRESENTATION OF LINEAR PROGRAMMING MODELS: UNIK-LP2.1. OBJECT-ORIENTED REPRESENTATION OF LINEAR PROGRAMMING MODELSA specific Linear Programming model should be represented in such a way as to be understood by rule-based systems. To accomplish this, the LP model needs to be represented in objects, which is named as semantic form. The object-oriented LP model (named a semantic form) can be transformed to the notational form, modeling language form, and tabular form, as depicted in Figure 1. A specific semantic model is composed of constraints, blocks of terms (BOT), attributes, and indices, as depicted in Figure 1. This idea is implemented in UNIK-LP (Lee and Kim, 1995).
For instance, consider the monthly product mix planning model, which decides the production and inventory levels.
subject to
The attributes imply:
and the indices imply:
The LP model (1)-(4) is a mathematical notational form. However, this form is not compatible with the rule-based systems. So, the model may be represented in an object as follows. {{monthly_product_mix_a IS_A: LP_MODEL DIRECTION: MIN OBJ_FN: (+ production_cost_BOT) CONSTRAINTS: production_capacity_constraint production_sales_balance_constraint CONTEXT: (TIME_UNIT month) (PERIOD multiple) }} The production_capacity_constraint can also be represented in an object. {{production_capacity_constraint IS_A: constraint OPERATOR: LE LHS: (+ processing_time_BOT) RHS: (+ facility_capacity_BOT) INDEX: facility time DOMAIN: (f1,...,fN) (t1,...,tT) }}. In this manner, the variables, constants, and indices can be represented in a top-down manner. {{production_amount IS_A: attribute SYMBOL: X ROLE: variable LINKED_INDEX: product facility time }} {{unit_processing_time IS_A: attribute SYMBOL: a ROLE: constant LINKED INDEX: product facility }} {{product IS_A: index DESCRIPTION: SYMBOL: i LINKED_ATTRIBUTES: production_amount inventory unit_cost unit_processing_time sales_quantity }} Since the semantic form possess more information than the mathematical notational form, we can transform the semantic form to notational form as depicted in Figure 2. The semantic form can be transformed to the aggregated notational form (Figure 3) and individual notational form (Figure 4). Since these models usually have thousands of variables and constraints, the maintenance of the models can be made much easier by using a tool like UNIK-LP. 2.2. KNOWLEDGE ASSISTED MODEL FORMULATIONSince the formulation of the semantic LP model is not an easy task for novice model builders, the function of knowledge-assisted LP model formulation is developed for UNIK-LP. For this purpose, the knowledge base needs to retain the LP model's structure knowledge and domain-specific knowledge (such as information about products and facilities). Using this knowledge, the LP model can be generated simply by identifying the relevant indices, decision variables, and constraints from the menu in a guided manner. For a detailed step-by-step explanation, readers may refer to Lee and Kim (1995).
3. UNIFICATION OF LP MODEL WITH RULE-BASED SYSTEM: UNIK-PMATo unify an LP model with a rule base, the two should share some of the decision variables. The unified model is thus virtually a multiobjective decision-making problem that considers both the numeric objective in the linear programming model and the symbolic objectives in the rule base. To solve the unified model, the Post-Model Analysis procedure is adopted as depicted in Figure 5 (Lee and Hurst, 1988; Lee and Song, 1996; Lee and Song, 1995). In this figure, note that the LP model and rule base are independently prepared. But since the LP model is formulated in semantic form, the LP model can communicate with the rule base at the semantic level. 3.1. POST-MODEL ANALYSIS PROCEDUREThe Post-Model Analysis procedure starts by finding the optimal solution of an LP model. The optimal solution is then inputted into the fact base of the rule-based system for evaluation. If the goals in the rule base are satisfied with the current optimal solution, the current solution can be selected as a satisfactory nondominated solution both for the optimization and rule-based system. However, if the current solution is not satisfactory, the adjusted goal in the rule base should be interpreted as the additional constraints to the LP model. Transforming the rules associated with the adjusted goal to the objects of semantic LP model is the key technological function in this process. For the details, readers may refer to Lee and Song (1996). Finally, the LP model with the additional constraints should be re-solved to assure the adjusted goals.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |