FORECASTING SHORT-TERM SEASONAL ELECTRICITY CONSUMPTION USING GENETIC ALGORITHM-BASED MODIFIED GREY MODELLING

Abstract

Electricity consumption contributes significantly to the global increase in total energy consumption, which is closely correlated with economic growth. It is therefore necessary to conduct research predicting electricity consumption. The primary research objectives are to develop proposed modified grey models for forecasting seasonal electricity consumption across various countries, to determine the best-optimized parameters for the background value and fractional order using genetic algorithms (GA), and to validate the strength of the proposed models across different simulated data scenarios. Previous studies in this field have encountered significant challenges, such as the inability of existing data grouping-based grey forecasting models, DGGM(1,1), and data grouping-based three‐parameter whitenization grey model, DGTWGM(1,1), to accurately capture seasonal fluctuations and the lack of optimization in model parameters, which often resulted in suboptimal forecasting performance. This research addresses these challenges by introducing innovative modifications to the grey modelling framework. The proposed models, data grouping-based fractional grey model, DGFGM(1,1), data grouping-based optimized fractional grey model, DGF2GM(1,1), data grouping-based fractional three‐parameter whitenization grey model, DGFTWGM(1,1), and data grouping-based optimized fractional three‐parameter whitenization grey model, DGF2TWGM(1,1), incorporate advanced data grouping and parameter optimization techniques, significantly enhancing their ability to handle seasonal variations and complex data patterns. By using a data grouping method, the entire quarterly and monthly time series was split into four and 12 groups, respectively, each of which only contains time series data from the same quarter and month. Models were then developed using the new set of four quarters and 12 months, each of which has specific seasonal characteristics. The prediction results from four quarters and 12 months for these models were integrated into a whole quarterly and monthly time series in chronological order to reflect seasonal differences. The mean absolute percentage error (MAPE) and root mean square error (RMSE) criteria were used to assess the forecasting accuracy for each forecasting method. The study utilized both the forecasting accuracy for simulated data, generated using seasonal autoregressive integrated moving average (SARIMA) models and real-world electricity consumption data. Simulated data allowed for controlled experimentation to evaluate the models under various conditions, ensuring their robustness and adaptability, while the real data provided a practical application context, validating the models' effectiveness in real-world scenarios. The results show that the proposed DGF2GM(1,1) and DGF2TWGM(1,1) models provide the best-fitting effect and are preferred for simulation and prediction to forecast quarterly and monthly electricity consumption in different countries. The two proposed models, optimized through genetic algorithms, significantly enhance forecasting performance for quarterly and monthly electricity consumption, making them highly effective tools for predicting electricity consumption trends with great accuracy.