Algoma University College

Thesis in Computer Science

Thesis Agreement

Great Lakes Power needs to predict the demand for electricity in the future on a continuous basis.
Such predictions are used to monitor various parameters in the power management.

This project will determine an appropriate neural network model to predict hourly and daily
electrical demands on a continuous basis. The model will be initially trained using 80% of the
available data. The remaining 20% of the data will be used for testing the continuous predictive
ability of the model. In the continuously adaptive predictions, the demand for the next time
period is predicted. Once the actual values of this time period are known they are used to retrain
the network to adapt to the newly available data. It is hoped that such a model will be eventually
tested and implemented in real-time.

The research will involve designing and implementing several models developed using different types
of neural networks such as windowed networks, multi-recurrent or time-delay neural networks. The
results will also be compared with some of the statistical time series analysis techniques such as
autoregressive analysis.