An adaptive backpropagation algorithm for long-term electricity load forecasting.

Area of Science:

• Electrical Engineering
• Computer Science
• Data Science

Background:

• Artificial Neural Networks (ANNs) are commonly used for electricity load forecasting.
• Traditional ANNs exhibit inaccuracies in long-term predictions due to accumulated errors and insufficient training data.
• Existing methods struggle with the dynamic behavioral shifts between training and future datasets.

Purpose of the Study:

• To develop an improved ANN model for accurate long-term electricity load demand forecasting.
• To introduce an Adaptive Backpropagation Algorithm (ABPA) that addresses limitations of traditional ANNs.
• To enhance forecasting by incorporating adjustment factors for dataset behavioral differences.

Main Methods:

• Developed an improved ANN model incorporating an Adaptive Backpropagation Algorithm (ABPA).
• Utilized a Multi-Layer Perceptron (MLP) architecture as a baseline, enhancing the traditional Backpropagation Algorithm (BPA).
• Integrated adjustment factors into forecasting formulations to account for deviations between training and future data.

Main Results:

• The proposed ABPA achieved highly accurate long-term forecasts.
• Demonstrated minimum Mean Squared Error (MSE) of 1,195,650 and Mean Absolute Percentage Error (MAPE) of 0.045.
• The adaptive algorithm outperformed traditional regression and advanced methods like Recurrent Neural Networks (RNNs).

Conclusions:

• The Adaptive Backpropagation Algorithm (ABPA) significantly improves the accuracy of long-term electricity load demand forecasting.
• The proposed method effectively handles behavioral differences between historical and future datasets.
• The enhanced ANN model offers a robust solution for reliable long-term energy demand prediction.