Forecasting energy demand and generation using time series models: A comparative analysis of classical, grey, fuzzy, and intelligent approaches

Efficiency and sustainability in the contemporary power systems cannot be achieved without accurate forecasting of energy consumption and production regardless of whether the energy source is conventional or renewable. The current research offers a systematic analytical review of 145 published articles that are devoted to the use of time series models to predict energy demand, consumption, and production. The review relies on 80 studies and compares eight predictive models. Eight major categories of models were used in the reviewed studies: Moving Average (MA), Exponential Smoothing (ES), ARMA, ARIMA, Case-Based Reasoning (CBR), Fuzzy Time Series (FTS), Gray Prediction Model (GPM), and Prophet forecasting model. The analysis was based on the research using real-time data obtained with the help of smart meters, solar power plants, national demand records, and databases of energy institutions. A quantitative analysis was done to compare the performance of the models based on accuracy indicators (RMSE, MAE, MAPE, R²) where the results were arranged in tables and presented using standardized classifications that highlight the differences between the models in terms of forecast horizon, accuracy, and applicability. The results indicated that classical approaches like ARIMA and ES are best suited in linear or regularly seasonal situations, but gray models are the best in situations where data is scarce, especially in developing nations. Fuzzy models, such as FTS and Prophet, were more accurate in cases with complicated, multi-seasonal trends. The current paper is distinguished among the existing reviews in that it provides a comparative quantitative analysis and a qualitative classification of forecasting models along with highlighting the latest research trends. It also suggests an effective framework that links the characteristics of data to the performance of the model, thus helping the researchers and decision-makers to choose the most suitable model under certain contextual limitations. Unlike other reviews that tend to mix the time-based forecasting algorithms with machine learning and deep learning methods without distinguishing methodologically between them, the current paper is more focused and specific in the field of energy time-series forecasting as it only considers the time series models of energy data.