Volume 17, 1998, No. 2-3


Special Issue on Intelligent Technologies for Nuclear and Electric Power Systems

Guest Editor: Ephraim Nissan

 

FuelGen: effective evolutionary design of refuellings for pressurized water reactors

J. Zhao, B. Knight, E. Nissan, A. Soper

Abstract. The paper describes the design of an efficient and robust genetic algorithm for the nuclear fuel loading problem (i.e., refuellings: the in-core fuel management problem) - a complex combinatorial, multimodal optimisation., Evolutionary computation as performed by FuelGen replaces heuristic search of the kind performed by the Fuelcon expert system (CAI 12/4), to solve the same problem.

In contrast to the traditional genetic algorithm which makes strong requirements on the representation used and its parameter setting in order to be efficient, the results of recent research results on new, robust genetic algorithms show that representations unsuitable for the traditional genetic algorithm can still be used to good effect with little parameter adjustment.  The representation presented here is a simple symbolic one with no linkage attributes, making the genetic algorithm particularly easy to apply to fuel loading problems with differing core structures and assembly inventories. A nonlinear fitness function has been constructed to direct the search efficiently in the presence of the many local optima that result from the constraint on solutions.

 

fuzzy-logic control applications to the belgian reactor 1 (BR1)

Da Ruan, Xiaozhong Li

Abstract. Fuzzy-logic control (FLC) applications in nuclear industry present a tremendous challenge. The main reason for this is the public awareness of the risks of nuclear industry and the very strict safety regulations in force for nuclear power plants. The very same regulations prevent a researcher from quickly introducing novel fuzzy-logic methods into this field.  On the other hand, the application of FLC  has, despite the ominous sound of the word "fuzzy" to nuclear engineers, a number of very desirable advantages over classical methods, e.g., its robustness and the capability to include human experience into the controller.

In this paper  we report an on-going R&D project for controlling the power level of the Belgian Reactor 1 (BR1) at the Belgian  Nuclear Research Centre (SCK·CEN). The project started in 1995 and aims to investigate the added value of FLC for nuclear reactors. We first review some relevant literature on FLC in nuclear reactors, then present the state-of-the-art of the BR1 project. After experimenting FLC under off-line test cases at the BR1 reactor, we now foresee a new development for a closed-loop FLC as an on-line operation of the BR1 reactor. Finally, we present the new development for the closed-loop FLC at BR1 with an understanding of the safety requirements for this real FLC application in nuclear reactors.

 

sensor data validation for nuclear power plants through bayesian conditioning and dempster's rule of combination

A.F. Dragoni, P. Giorgini

Abstract. Sensor data fusion and interpretation, sensor failure detection, isolation and identification are extremely important activities for the safety of a nuclear power plant. In particular, they become critical in case of conflicts among the data. If the monitored system's description  model is correct and its components work properly, then incompatibilities among data may only be attributed to temporary deterioration or permanent breakage of one or more sensors.  This paper introduces and discusses three simple ideas:

1. classical "model-based diagnosis" can be extended straightforwardly to encompass the sensors models into the system's description in order to diagnose even their own faults

2. from the "log-file" of the diagnosed minimal conflicts among the sensors, one can draw interesting conclusion regarding their relative reliability (e.g., through Bayesian conditioning)

3. the estimated reliability of the sensors is useful when assessing (e.g. through Dempster's Rule of Combination) the actual state of the monitored physical system, even in cases of conflicting data.

These ideas lead to the conception of a distributed monitoring system able to attach to each sensor a statistically evaluated relative reliability, which is especially useful for devices situated in dangerous zones or areas, difficult to reach inside huge and complex power plants.

 

Neurofuzzy approaches and their application to nuclear power systems

R.E. Uhrig, L.H. Tsoukalas

Abstract. Neurofuzzy approaches (NFA) utilize a variety of neural and fuzzy synergisms that exploit a measured tolerance for imprecision and uncertainty for the purpose of enhancing flexibility and tractability in models and systems.  It is theoretically expected  and empirically conformed that neurofuzzy approaches when appropriately structured allow for improved control over the modeling economy or parsimony resulting in easier to develop and modify systems.  Hence, they hold considerable promise for significant enhancements in the control and safety of nuclear plant appurtenances, components and systems. Two nuclear power system applications are presented in this paper. The first is in the reactor control area. It uses neural networks to predict power trajectories and fuzzy rules that incorporate such predictions in proactive or anticipatory strategies in order to improve power manoeuvres during reactor startup. The second is in the area of safety, where neural mappings are used to produce fuzzy values for epistemic variables. The methodology is extending the notion of measurement to variables with functional or operational significance and hence is preferred to as virtual measurement; it is applied to flow visualization and holds considerable promise for improving diagnostics and hence safety in nuclear reactors.

 

Gaussian mixture densities for classification of nuclear power plant data

Y. Bengio, F. Gingras, B. Goulard, J.-M. Lina, K. Scott

Abstract. In this paper we are concerned with the application of learning algorithms to the classification of reactor states in nuclear plants. Two aspects must be considered: (1) some types events (e.g., abnormal or rare) will not appear in the data set, but the system should be able to detect them, (2) not only classification of signals but also their interpretation are important for nuclear plant monitoring. We address both issues with a mixture of mixtures of Gaussians in which some parameters are shared to reflect the similar signals observed in different states of the reactor. An EM algorithm for these shared Gaussian mixtures is presented.  Experimental results on nuclear plant data demonstrate  the advantages of the proposed approach with respect to the above two points.

 

optimizing neural network models: motivation and case studies

S.A. Harp, T. Samad

Abstract. Practical successes have been achieved  with neural network models in a variety of domains, including energy-related industry. The large, complex design space presented by neural networks is only minimally explored in current practice. The satisfactory results that nevertheless have been obtained testify that neural networks are a robust modeling technology; at the same time, however, the lack of a systematic design approach implies that the best neural network models generally  remain undiscovered for most applications.  This paper first presents an experimental study that demonstrates the complex interdependencies between various parameters of neural models.  We then present an approach, based on genetic algorithms, for designing optimized neural networks for specific applications. Two case studies are discussed n which the benefits of a systematic design method are exemplified. These studies are on real data sets that are relevant to the power industry. The flexibility of genetic optimization also permits some novel twists on neural modeling: input selection, and the synthesis of network architectures well suited  for problem classes can be directly addressed.

 

a hybrid neuro-genetic approach to flow calculation based on the representation of an electrical power system by critical switches

D. Arjona

Abstract. This paper is intended to present an approach to decision making in the operation of electrical power systems that will use a simple genetic algorithm as a teacher for the process of supervised learning of a feedforward, backpropagation artificial neural network. The fitness function used in the genetic algorithms is based on a load flow program and used to determine the optimal condition of the critical switches of the system. Reward and penalty functions are applied to it in order to emphasize environmental, economic, security, robustness, public policy and other considerations as they are pre-determined by the philosophy of operation of the utility. These considerations (policies) become a part of the training set and operation of the neural network.

The fitness function used by the genetic algorithm in order to rank the possible solutions is based on a load flow program. The binary nature of the genetic algorithm is particularly appropriate for the operation of switches.

The result of the methodology is the equivalent of an on-line implicit load flow program used to redesign the configuration of the system in real time by opening and closing critical switches that are placed along the power system.

Experiments leading towards the development of this methodology using real data from the Peninsular Control Area (The Yucatan Peninsula) of the National Mexican Interconnected Power Grid are presented.

Concepts of electrical power engineering are presented as a general reference for the reader of this document who is not a specialist on that field.


Go To Contents