چكيده به لاتين
Efficiency improvement is one the most important tasks of managers and decision makers. First step for efficiency improvement is efficiency evaluation. While different methods are applied for efficiency evaluation previously, the most important method is "Data Envelopment Analysis (DEA)". This method is based on mathematical programming. Different inputs and outputs are considered in DEA method to calculate relative efficiency of Decision-Making Units (DMUs).Primary DEA models are only suitable for simple one-stage DMUs, because they ignore internal structure of units. "Network DEA (NDEA)" method is developed and applied in literature to deal with DMUs with complex structure.
Data uncertainty is ignored in standard NDEA model. Hence, based on robust optimization approach, novel NDEA models are developed in this thesis to deal with data perturbation and uncertainty. In this concept, three class of models are developed: 1) Robust NDEA (RN) models for efficiency evaluation, 2) Robust super-efficiency NDEA (RSN) models and 3) Robust super-efficiency NDEA (FRSN) models with fuzzy perturbation parameter. According to an uncertainty set induced approach, six models are developed in each class. Following uncertainty sets are considered in each class: Box, Ellipsoidal, Polyhedral, Box+Ellipsoidal, Box+Polyhedral and Box+Ellipsoidal+Polyhedral.
Developed models are applied in electricity energy and water industries. RN models are used to efficiency evaluation and ranking of 16 regional electricity power companies in Iran and the efficiency of the entire networks of electricity power, involving generation, transmission and distribution stages is measured. Each stage has inputs and outputs and specific energy is transferred between staged. RSN and FRSN models are applied to efficiency evaluation and ranking of 35 provincial water supply and distribution companies. Validation of results is investigated by experts of Tavanir and Water resources management companies. Results of developed models are more reliable and more suitable for real world application. Furthermore, discrimination power of developed models is much better of standard NDEA models.
