Computational Methods for the Operational Surveillance of Nuclear Power Plant

Research student:  Nicholas Matcham
Supervisors:Adrian Hopgood
 Martin Weller
PhD awarded:27 September 2005

Abstract from thesis

The primary concern of nuclear power plant utilities is the safe operation of their plant. However, in complex dynamic engineering systems such as nuclear power plant, operations staff face demanding situations from the dynamics of operational faults. To maintain safe plant operation, operations staff are required to monitor variation in system parameters, interpret plant behaviour and diagnose any operational problems that occurred.

The research addressed in this thesis concentrates on the development and implementation of computational methods that can be used for the operational surveillance of nuclear power plant. The application of these methods is investigated to produce an overall surveillance model capable of providing operations staff with accurate information on the current state of the plant under their control and indications of deviations from normal operating conditions.

The operational surveillance models developed incorporate three main computational methods to determine periods of steady state operation, define the plant operational state and assess the significance of any change in plant operational state. The determination of steady state operation uses a statistical test for level stationarity. The definition of the plant operational state is derived from a neural network design that combines a conventional multilayer perceptron network with a mixture density model trained using a fivefold non-linear cross validation method. The predicted plant variable outputs from the network are processed by two established methods from statistical process control, the Shewhart control and the cumulative summation models, that enable the plant operational state to be defined.

The overall surveillance model is assessed using real plant data from two case studies. Each case study is designed to assess the performance of the model when processing data from an actual plant transient and the criteria introduced for assessment is used primarily to evaluate the ability of the model in identifying the initial time point and duration of the transient and in providing early warning information on potential adverse operating conditions.

The work of this thesis has resulted in the development of an adaptive surveillance model capable of identifying the transition from steady state operation to transient operation that assists operations staff with the safe and economic operation of their plant. A method for identifying a period of steady state operation is developed and used successfully to discriminate between steady state operation and transient operation. The model developed avoids the use of conceivable operating transients profiles produced during design and actual plant data is used exclusively for transient identification. However, the general reliability of the model can only be fully assessed by prolonged operational use. Whilst the overall research effort has been directed at implementation of the methods to nuclear power plant, the methods are readily adaptable to any other industrial process.

Return to Adrian Hopgood's page.

Valid HTML 4.01! Owner:
  Adrian Hopgood
  March 2019