چكيده به لاتين
Budget and resource limitations have increased consentration on maintenance and repair management instead of new constructions, which has not been seriously considered in Iran, at least in practical issues. The first step in maintenance management is the correct and principled estimation of costs and manage them with the aim of improving the planning and proper management of projects, economic review, budget allocation, etc., which requires a suitable method for costs prediction. Therefore, the aim of this thesis is to examine the maintenance cost estimation models along with the methods and applications of each one. Here, by designing and using three widely useful and valid models including parametric models (regression), artificial neural network and case-based reasoning, the maintenance costs in 10 concrete structural residential towns in Tehran have been predicted. Minitab software was used for the regression method and a linear model was obtained based on the effective factors. The neural network algorithm was designed with MATLAB software, with two layers and a logarithmic activation function. Finally, the case-based reasoning algorithm was designed with Python software and the particle swarm algorithm was used for weighting. The neural network with a correlation coefficient of 0.924 was more accurate in prediction. However, the case-based reasoning method also shows a similar accuracy with a value of 0.903, and since it requires less data than the neural network, it can be a very suitable alternative in this field. The lowest coefficient would be related to the regression method with the value of 0.805, which is less accurate than the other two methods although it is acceptable. The results have shown that the application of maintenance cost prediction models is possible and can be developed by modern meta-exploration models with a satisfy accuracy.
