A Neuro-Fuzzy Approach for Modelling Electricity Demand in Victoria
Title: A Neuro-Fuzzy Approach for Modelling Electricity Demand in Victoria
Research Question: Can soft computing techniques and statistical models improve the accuracy of electricity demand forecasting compared to traditional methods?
Methodology:
1. Data Collection: The researchers used historical electricity demand data from the State of Victoria, Australia, covering ten months from 27th January to 30th November 1995. They also collected associated data such as minimum and maximum temperatures, time of day, season, and day of the week. 2. Data Preprocessing: The data was preprocessed to remove outliers and normalize the features. 3. Model Development: Three forecasting models were developed: * Evolving Fuzzy Neural Network (EFuNN): An advanced soft computing technique that combines the strengths of fuzzy logic and neural networks. * Artificial Neural Network (ANN) with Scaled Conjugate Gradient Algorithm: A type of ANN that uses a gradient-based optimization algorithm to train the network. * ARIMA Model: A popular statistical technique based on Box-Jenkins methodology for time series analysis. 4. Model Training and Testing: The models were trained using 3 randomly selected samples containing 20% of the data during the period 27th January 1995 to 28th November 1995. The trained models were then tested to predict the demand for the period (29th-30th) November 1995. 5. Evaluation: The forecasting accuracy of each model was evaluated using mean absolute percentage error (MAPE), mean absolute error (MAE), and mean squared error (MSE).
Results:
1. EFuNN: The EFuNN model showed the best performance, with a MAPE of 4.04%, followed by the ANN with a MAPE of 4.58%, and the ARIMA model with a MAPE of 5.34%. 2. Comparison with VPX Forecasts: The EFuNN model outperformed the VPX forecasts, with a MAPE of 4.04% compared to VPX's MAPE of 5.65%.
Implications:
The researchers found that the EFuNN model performed better than the ANN, ARIMA model, and the VPX forecasts. This suggests that soft computing techniques and statistical models can improve the accuracy of electricity demand forecasting, which is crucial for the efficient operation of the power grid and competitive pricing in the energy market. The results also indicate that the EFuNN model could be a valuable tool for energy companies and policy makers in their decision-making processes.
Link to Article: https://arxiv.org/abs/0405010v1 Authors: arXiv ID: 0405010v1