[Paper Review] Using Artificial Neural Network Techniques for Prediction of Electric Energy Consumption

Hasan M. H. Owda, Babatunji Omoniwa+2 more|arXiv (Cornell University)|Dec 6, 2014

Energy Load and Power Forecasting16 references3 citations

TL;DR

This paper proposes a feedforward artificial neural network (ANN) model trained on monthly electric energy consumption data from the Gaza Strip (1994–2013) to forecast future demand. Using 2-Fold and K-Fold cross-validation, the model achieves low error rates (MSE: 1.21–1.88%, MAPE: 120.91–187.65%) with high accuracy, demonstrating ANN's effectiveness for energy forecasting in data-scarce, high-uncertainty environments.

ABSTRACT

Due to imprecision and uncertainties in predicting real world problems, artificial neural network (ANN) techniques have become increasingly useful for modeling and optimization. This paper presents an artificial neural network approach for forecasting electric energy consumption. For effective planning and operation of power systems, optimal forecasting tools are needed for energy operators to maximize profit and also to provide maximum satisfaction to energy consumers. Monthly data for electric energy consumed in the Gaza strip was collected from year 1994 to 2013. Data was trained and the proposed model was validated using 2-Fold and K-Fold cross validation techniques. The model has been tested with actual energy consumption data and yields satisfactory performance.

Motivation & Objective

• To develop a reliable forecasting model for electric energy consumption in the Gaza Strip, where demand growth exceeds supply and historical data is limited.
• To address the limitations of traditional statistical methods like regression and ARIMA, which often fail due to instability and poor handling of nonlinear relationships.
• To evaluate the performance of artificial neural networks in capturing complex, nonlinear patterns in energy consumption influenced by weather, population, and GDP.
• To validate the model using robust cross-validation techniques (2-Fold and K-Fold) to ensure generalization and avoid overfitting.
• To provide a practical, data-driven tool for energy operators to improve planning, maximize profit, and enhance consumer satisfaction in energy-scarce regions.

Proposed method

• The study uses a feedforward backpropagation neural network trained on monthly electric energy consumption data from 1994 to 2013 in the Gaza Strip.
• Input features include historical load, mean temperature, relative humidity, population, and GDP per capita, which are known to influence energy demand.
• The model employs the Levenberg–Marquardt algorithm for efficient training and weight optimization to minimize prediction error.
• Performance is evaluated using multiple error metrics: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE).
• 2-Fold and K-Fold cross-validation are applied to assess model robustness, with the validation set used only during training and the test set reserved for final evaluation.
• The model is trained on 19 years of data and tested on actual 2012 and 2013 consumption data to validate predictive accuracy.

Experimental results

Research questions

• RQ1 Can an artificial neural network effectively model and predict electric energy consumption in a high-uncertainty, data-limited environment like the Gaza Strip?
• RQ2 How does the performance of the ANN model compare across different cross-validation techniques (2-Fold vs. K-Fold) in terms of prediction accuracy and stability?
• RQ3 To what extent do weather variables and socioeconomic factors (population, GDP) improve the predictive power of the ANN model compared to time-series-only approaches?
• RQ4 What is the model’s predictive accuracy when tested on unseen data from 2012 and 2013, and how do error metrics (MSE, RMSE, MAE, MAPE) reflect its reliability?
• RQ5 Can the proposed ANN model outperform traditional statistical methods such as ARIMA or regression in forecasting energy consumption under real-world uncertainties?

Key findings

• The ANN model achieved a Mean Squared Error (MSE) of 1.21% and a Root Mean Squared Error (RMSE) of 1.10% using 2-Fold cross-validation for 2012 data, indicating strong predictive accuracy.
• For 2013, the model recorded an MSE of 1.74% and RMSE of 1.32% under 2-Fold cross-validation, with MAPE values of 173.76%, showing consistent performance across years.
• K-Fold cross-validation yielded higher error rates (MSE: 3.87%, MAPE: 386.54%) in 2012, suggesting potential overfitting or instability in the K-Fold configuration, possibly due to data partitioning.
• The model demonstrated high consistency in forecasting monthly consumption, with forecasted values closely matching actual data across most months in 2012 and 2013.
• The use of weather and socioeconomic variables as inputs significantly enhanced model performance compared to time-series-only models, which often fail to capture external influences.
• The results confirm that ANNs are effective for energy forecasting in regions with high uncertainty and limited data, outperforming traditional regression and time-series models in this context.