Case Study 1 : Thermal Power Station redesign

This project for British Energy involved the redesign of the direct feed heater layout of a nuclear power station thermal system. The objective was to improve the overall performance of the thermal cycle of the nuclear power plant by optimising feed heater design as well as operation, using a combination of a bespoke Genetic Algorithm, mixed discrete / continuous variable parameter problems and conventional numerical optimisation techniques. 

The project took place during the mid-1990’s and utilised British Energy’s NUMEG suite of thermal system design and operation models to provide an objective function. The work resulted in an increase of predicted net electricity output of over 0.3%. 

In order to address the feed heater layout problem it was first necessary to fully automate the thermal optimisation process which consists of the identification of a boiler steady-state. This originally involved a heuristic exercise taking from a few hours to several days of an experienced engineer’s time to identify initial feasible solutions.

The introduction of a highly exploratory genetic algorithm (GA) utilising appropriate penalty functions to identify feasible solutions in the heavily constrained solution space allowed the subsequent use of a conventional sequential linear optimiser (SLP) commencing from these feasible points. Engineer interaction was eliminated thereby allowing the integration of the GA / SLP optimiser with the feed heater layout problem. 

This problem is particularly complex as it involves the concurrent optimisation of discrete Configuration Variables (number, serial, parallel etc), continuous Operational Control Variables (flow rate etc), and continuous Configuration Related Decision Variables (pipe diameter, length etc).

A bespoke GA was developed that can successfully search this discrete / continuous solution space and, within an acceptable time period, return optimal solutions that result in an overall increase of 0.3% in system power output.

This case study illustrates the flexibility of evolutionary algorithms in terms of the manner in which:

• heavily constrained problem spaces can be successfully negotiated using appropriate penalty functions

• hybridised systems comprising evolutionary algorithms and conventional deterministic optimisers can result in rapid identification of optimal solutions

• evolutionary algorithms can manipulate complex sets of variable types and successfully overcome problem discontinuity caused by discrete design options

• successfully eliminate lengthy engineer-based heuristic searches thereby converting engineer-hours into machine hours in a completely automated process. 

The overall design lead time was reduced on average by around 15%, and the process is now entirely machine-based.