Dual-optimization method for improving accuracy in GA-CBR cost estimating model

다국어 초록 (Multilingual Abstract)

As large amount of time and resources used in completing a construction project, cost estimating is consistently carried out in the all of the stages. In particular, the early stage of cost estimating is very important in the decision-making as it determines the success and failure of the project. Case-based reasoning is widely used as an effective methodology for early cost estimation in construction projects. It has the advantages that it can infer persuasive and accurate answers relatively fast, easy to maintain, and the accuracy increases as it used. So many researchers have been conducting research to improve the accuracy and usability of cost estimating models by using case-based reasoning.

The accuracy of the case-based reasoning cost estimating model has been affected by retrieving and adaptation. Case retrieval has a significant effect on the performance of CBR models. Retrieval accuracy depends on how many attributes are used, what kinds of attributes are used, and how the attribute weights are assigned in a model. However, existing methods used only a limited number of qualitative variables by applying a subjective weight assignment method or excluded these from their model. This can limit the number of variables which can be used, or the difference between qualitative attributes can be disregarded. Additionally, since the problem description is not completely same as previous cases, old solutions should be adjusted to fit new situations. There are several adaptation methods for CBR, it is necessary to improve the estimation accuracy by applying suitable adaptation methods for the cost estimating model. The other method of adaptation is using several cases for problem solving. If only one case is used to solve the problem, it cannot reflect the good traits of other similar cases. The weighted mean method is most widely utilized because it can reflect the difference of case similarities to deduce a solution by giving higher weights to more similar cases. However, there is a disadvantage that the difference of weights is calculated relatively small.

In an effort to address these problems, this research attempts to present the dual-optimization method for considering qualitative variables in CBR cost estimating models based on a genetic algorithm. This method can assign not only the attribute weights, but also the quantified values of the qualitative attributes. This method is able to apply more attributes than existing methods through quantification of the qualitative variables. Additionally, this research suggested two adaptation methods for the GA-CBR cost estimating model. The retrieving error adaptation is the method to calculate the differences and to adjust the solutions of retrieved cases. By reflecting estimation error caused by differences between target and retrieved cases, a retrieved solution is adjusted to be more appropriate. Furthermore, the improved weighted mean method was suggested to alteration method for multiple cases adaptation. By assigning weights according to the similarity distribution of retrieved cases, the improved weighted mean method can increase the influence of more similar cases.

To validate the proposed methods, three kinds of validations were conducted to estimate the construction cost of military barracks and public apartment projects. The results of validation 1 indicate that the dual-optimization method is improved in terms of accuracy and stability compared to previous methods. In validation 2 and 3, the estimation accuracy is increased when the proposed adaptation methods are used to the GA-CBR cost estimating model. These validation results support that the proposed methods can be utilized for construction cost estimation to make better decision.

This research has significant in that it suggests three new methods to improve the disadvantages of existing methods. Consequently, the new GA-CBR cost estimating model can make more accurate cost estimation compared to other methods. It is expected that the proposed cost estimating model and methods will support stakeholders of construction project to make better decision at early stage of project. Moreover, the dual-optimization is a general-purpose method; it is expected to be more readily applied to a problem of other fields. In the dual-optimization, despite of its excellent performance, the more qualitative variables and values are utilized, the longer the length of the chromosome to be optimized and the longer the calculation duration. This research hopes that this problem will be solved by advance of computer science and improvement of its algorithm.

목차 (Table of Contents)

• Chapter 1. Introduction 1
• 1.1 Research Backgrounds 1
• 1.2 Problem Statement 3
• 1.3 Research Objective 7
• 1.4 Research Scope and Process 10
• Chapter 2. Preliminary Research 13
• 2.1 Early Stage Cost Estimation Methods 14
• 2.1.1 Traditional Approaches 18
• 2.1.2 Artificial Intelligence Approaches 20
• 2.2 Case Based Reasoning (CBR) 23
• 2.2.1 Overview of CBR 23
• 2.2.2 CBR Cycle 26
• 2.2.3 CBR Advantages and Limitations 30
• 2.3 Model Components 34
• 2.3.1 Case Retrieval Methods 34
• 2.3.2 Case Adaptation Methods 42
• 2.3.3 Genetic Algorithm (GA) 45
• 2.3.4 Type of Variables 53
• 2.3.5 Construction Cost Index 55
• 2.4 Literature Review 58
• 2.4.1 Calculating Weights in CBR Model 58
• 2.4.2 Representation of Qualitative Attributes 63
• 2.5 Summary 67
• Chapter 3. Establishment of Dual-Optimization and Adaptation Methods 71
• 3.1 Dual-Optimization Method 72
• 3.1.1 Algorithms of Dual-Optimization 72
• 3.1.2 Process of Dual-Optimization 80
• 3.1.3 Advantages and Disadvantages of Dual-Optimization 82
• 3.2 Case Adaptation Methods 85
• 3.2.1 Retrieving Error Adaptation Method 88
• 3.2.2 Improved Weighted Mean Method for Multiple Case Adaptation 92
• 3.3 Summary 96
• Chapter 4. GA-CBR Cost Estimating Models with Dual-Optimization 98
• 4.1 Case Base Establishment 99
• 4.1.1 Data Modeling 100
• 4.1.2 Data Analysis 102
• 4.2 Calculating Weights using Dual-Optimization 110
• 4.2.1 Model 1: Military Barrack Projects 110
• 4.2.2 Model 2: Public Apartment Projects 116
• 4.3 Cost Estimating Model 121
• 4.3.1 System Architecture 121
• 4.3.2 Cost Estimating Process 123
• 4.4 Summary 127
• Chapter 5. Model Validations 128
• 5.1 Validation Methods 129
• 5.2 Validation 1: Dual-optimization Method 135
• 5.2.1 Military Barracks Projects 136
• 5.2.2 Public Apartment Projects 144
• 5.3 Validation 2: Retrieving Error Adaptation 153
• 5.3.1 Military Barracks Projects 154
• 5.3.2 Public Apartment Projects 156
• 5.4 Validation 3: Improved Weighted Mean Method 158
• 5.4.1 Military Barracks Projects 159
• 5.4.2 Public Apartment Projects 161
• 5.5 Summary 163
• Chapter 6. Conclusions 166
• 6.1 Summary of Research 166
• 6.2 Research Contributions 168
• 6.3 Limitations and Further Studies 170
• Bibliography 171
• Appendix 182
• Appendix 1. Grossary of Acronyms 182
• Appendix 2. Case Base of Military Barrack Projects 183
• Appendix 3. Case Base of Public Apartment Projects 190
• 초록 196