Advanced Techniques for Joint Weather and Electricity Demand Prediction Survey

Recent advancements in the field of machine learning have exceptionally enhanced the accuracy as well as efficiency of electricity demand and weather forecasting. There is a range of forecasting methods used for predicting electricity demand and weather conditions, including historical approaches like time series analysis and more recently machine learning algorithms. This review aims at summarizing the recent progress that demonstrated a shift from applying simple regression and correlation coefficients to more powerful and flexible algorithms. Findings from the research indicate that Support Vector Regression (SVR), Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM) and Multivariate Adaptive Regression Splines (MARS) are better suited for short-term electricity demand forecasting because of the accuracy. Furthermore, novel approaches such as diffusion models for weather prediction are obvious examples of how generative models are capable of handling uncertainty. Linear methods such as Autoregressive Integrated Moving Average (ARIMA) are still useful for short-term prediction and linear moving average, while modern techniques, especially deep learning and hybrid approaches, are better suited for discovering intricate relations and dealing with different predictors. The idea here is to select the best characteristics of different models for modeling in complex situations. However, some of the difficulties remain. These include data quality issues, interpretation of derived models, and high computational complexity.