A new systematic and flexible method for developing hierarchical decision-making models


BELDEK U., LEBLEBİCİOĞLU M. K.

TURKISH JOURNAL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCES, cilt.23, ss.279-297, 2015 (SCI İndekslerine Giren Dergi) identifier identifier

  • Cilt numarası: 23 Konu: 1
  • Basım Tarihi: 2015
  • Doi Numarası: 10.3906/elk-1302-3
  • Dergi Adı: TURKISH JOURNAL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCES
  • Sayfa Sayıları: ss.279-297

Özet

The common practice in multilevel decision-making (DM) systems is to achieve the final decision by going through a finite number of DM levels. In this study, a new multilevel DM model is proposed. This model is called the hierarchical DM (HDM) model and it is supposed to provide a flexible way of interaction and information flow between the consecutive levels that allows policy changes in DM procedures if necessary. In the model, in the early levels, there are primary agents that perform DM tasks. As the levels increase, the information associated with these agents is combined through suitable processes and agents with higher complexity are formed to carry out the DM tasks more elegantly. The HDM model is applied to the case study 'Fault degree classification in a 4-tank water circulation system'. For this case study, the processes that connect the lower levels to the higher levels are agent development processes where a special decision fusion technique is its integral part. This decision fusion technique combines the previous level's decisions and their performance indicator suitably to contribute to the improvement of new agents in higher levels. Additionally, the proposed agent development process provides flexibility both in the training and validation phases, and less computational effort is required in the training phase compared to a single-agent development simulation carried out for the same DM task under similar circumstances. Hence, the HDM model puts forward an enhanced performance compared to a single agent with a more sophisticated structure. Finally, model validation and efficiency in the presence of noise are also simulated. The adaptability of the agent development process due to the flexible structure of the model also accounts for improved performance, as seen in the results.