Loading...

Estimation Influential Parameters in Operation of the Bushehr Nuclear Power Plant using Neural Network

Ghanbari, Mohammad | 2016

618 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 49324 (46)
  4. University: Sharif University of Technology
  5. Department: Energy Engineering
  6. Advisor(s): Ghofrani, Mohammad Bagher; Moshkbar Bakhshayesh, Khalil
  7. Abstract:
  8. Given many computing errors in current systems, a method appears necessary for predicting the nuclear parameter quickly and accurately. In this thesis, a neural network was used to predict safety in a nuclear power plant in order to develop an operating aid tool for preventive measures.First, some studies were conducted on appropriate feature selection for training neural networks. Some case studies have also been carried out on parameter prediction through soft computing in a power plant. In the next section, an expert judgment was taken into account to select DNBR (Departure from Nucleate Boiling Ratio) as a criterion for safety evaluation in the exploitation of a nuclear power plant (PWR) and for prediction with the help of a neural network. The unusual value of this parameter is regarded as a threat to safety barriers such as fuel pods. Then three appropriate routes were obtained from the reference to the final safety report of Bushehr Nuclear Power Plant to extract data required to train the neural network and predict DNBR. In the third chapter, a new method was developed for appropriate feature selection to train the neural network, decrease computations, and increase computing speed. The proposed method was validated by comparing it with the features selected by particle swarm optimization. The results indicated the efficiency of the proposed method in appropriate feature selection and computing cost reduction.
    In the next section, an improved multilayer neural network was used for the main components such as the way of selecting initial weights, the way of updating weights, the type of cost function, and the type of activation function to predict DNBR. The results were presented in the fourth chapter. Based on different error criteria, the results indicated that the parameter was predicted by the neural network developed at the maximum 10% error with a higher accuracy compared with the predictor systems based on the current model at a 30-40% error. Moreover, the prediction error was lower when the features selected by the new method were used to train the neural network in comparison with the case in which feature selection was done by the particle swarm optimization.The final chapter includes brief findings, innovation, and the research evaluation
  9. Keywords:
  10. Feature Selection ; Neural Network ; Particles Swarm Optimization (PSO) ; Soft Computation ; Parameter Estimation ; Bushehr Nuclear Power Plant

 Digital Object List

 Bookmark

No TOC