Addison, J. F. Dale and Hunter, Andrew and Bass, Jeremy and Rebbeck, Matt (2000) A neural network version of the measure correlate predict algorithm for estimating wind energy yield. In: 13th International Congress on Condition Monitoring and Diagnostic Engineering Management, 3-8 December, 2000, Houston, USA.

PDF AddisonMeasureCorrelatePredict.pdf - Whole Document Restricted to Repository staff only Download (30Kb) | Request a copy

Abstract

We have investigated the feasibility of using neural networks to make predictions of long term energy yield at a potential wind farm site. This paper considers the effectiveness of neural networks in predicting wind speed at a target site from wind speed and direction measurements at a reference site. The technique is compared with the standard Measure Correlate Predict (MCP) algorithm used in the wind energy industry. Improvements of predictive accuracy in the region of 5%-12% can be achieved. Best results are obtained using multilayer perceptron networks with a large number of hidden units, with extensive Quasi-Newton (BFGS) training. Experiments have been conducted using contemporaneous measurements, and time shifted wind speed (previous and next hour) as inputs. Performance is consistently improved by using time-shifted inputs. However, the improvement in performance has to be offset against the financial penalty incurred in purchasing time series data for input.

Item Type:	Conference or Workshop Item (Paper)
Additional Information:	We have investigated the feasibility of using neural networks to make predictions of long term energy yield at a potential wind farm site. This paper considers the effectiveness of neural networks in predicting wind speed at a target site from wind speed and direction measurements at a reference site. The technique is compared with the standard Measure Correlate Predict (MCP) algorithm used in the wind energy industry. Improvements of predictive accuracy in the region of 5%-12% can be achieved. Best results are obtained using multilayer perceptron networks with a large number of hidden units, with extensive Quasi-Newton (BFGS) training. Experiments have been conducted using contemporaneous measurements, and time shifted wind speed (previous and next hour) as inputs. Performance is consistently improved by using time-shifted inputs. However, the improvement in performance has to be offset against the financial penalty incurred in purchasing time series data for input.
Keywords:	neural networks, wind energy, measure correlate predict, algorithm
Subjects:	G Mathematical and Computer Sciences > G730 Neural Computing
Divisions:	College of Sciences > Faculty of Science > Lincoln School of Computer Science
Depositing User:	Tammie Farley
Date Deposited:	25 Sep 2010 19:59
Last Modified:	13 Mar 2013 08:32
URI:	http://eprints.lincoln.ac.uk/id/eprint/1892

Actions (login required)

View Item