Murdoch University Research Repository

Welcome to the Murdoch University Research Repository

The Murdoch University Research Repository is an open access digital collection of research
created by Murdoch University staff, researchers and postgraduate students.

Learn more

The use of artificial neural networks in the control of remote area power supply systems

Cheok, Kenneth K.L. (1998) The use of artificial neural networks in the control of remote area power supply systems. PhD thesis, Murdoch University.

[img]
PDF - Whole Thesis
Available Upon Request

Abstract

Remote Area Power Supply systems, or (RAPS), are increasingly being used to supply energy to remote areas not connected to the utility grid. Technological advances in recent years have made the inclusion of photovoltaics, wind turbines, batteries and inverters a more cost effective option to the traditionally used diesel generator systems. In RAPS the use of such renewables involves a large initial capital cost. Frequent mismatch between the changing load demand and renewable supply may result in low system efficiency and wastage of renewables. For renewables to prove viable the operation of the system must maximise the use of the renewable component. Such an optimisation problem requires accurate short term forecasts of the load, together with the solar and wind resource.

This thesis describes a novel method of incorporating Artificial Neural Networks (ANNs) to provide load forecasts to optimise the operation of larger hybrid RAPS systems. The key inputs to the ANN are investigated and identified for short term load forecasting. The forecasts are then compared to traditional statistical methods of load forecasting. The ANN is modified to make it adaptive to seasonal data and used to forecast the load for up to 24 hours ahead in three RAPS systems in Western Australia over a one year period. The accuracies of the ANN forecasts are analysed for each location.

A new predictive control strategy, known as the Predictive Control Strategy, is developed to improve the operation of a parallel hybrid RAPS system. An existing RAPS simulation model, RESIM, which utilises conventional control, is modified by incorporating predictive control algorithms into its structure. Initially, perfect forecasts of the load and weather are applied in a hypothetical operating environment created by RESIM. The system control actions and component performances are compared to the simulation results using a conventional control strategy without forecasts. The perfect simulation forecasts may be used to determine the theoretical upper limit of the benefits of the performance and cost savings that may be possible using such predictive control methods.

The ANN is then incorporated into the simulation model to provide the load forecasts. The control actions of this predictive control strategy are subsequently analysed by simulating over one year of data from two of the three locations, Yalgoo and Gascoyne Junction, and comparing the results to the conventional approach.

The results of the work in this thesis indicate that, compared to an equivalent RAPS system using conventional control, the Predictive Control has a potential to decrease operating costs of the diesel generator, while incurring a penalty in the form of increased battery replacement costs. However, an appropriately sized RAPS system using conventional control is shown to be very efficient in itself, and the benefits gained using this predictive approach are modest. Application of a life cycle cost methodology and a cost sensitivity analysis show that the overall economic benefit of such a predictive approach is dependent largely on the cost values used for the diesel fuel and battery components of the RAPS system.

Item Type: Thesis (PhD)
Murdoch Affiliation: Division of Science
Notes: Note to the author: If you would like to make your thesis openly available on Murdoch University Library's Research Repository, please contact: repository@murdoch.edu.au. Thank you.
Supervisor(s): Pryor, Trevor and Cole, Graeme
URI: http://researchrepository.murdoch.edu.au/id/eprint/52365
Item Control Page Item Control Page