Short-term smart grid energy forecasting using a hybrid deep learning method on univariate and multivariate data sets

Transforming energy power systems into integrated smart grids is pivotal in addressing climate change challenges. Accurate forecasting of energy demand within such grids is essential for optimizing operations, incorporating renewable energy sources, and enhancing grid stability. However, many traditional or simpler machine learning approaches struggle to harness the interdependencies between the various energy sources. This limits their predictive performance and scalability. This paper presents a novel hybrid approach for short-term smart grid energy forecasting in the load and wind speed datasets, utilizing meteorological factors to enhance prediction accuracy. To reduce the difficulty caused by energy volatility, noisy measurement, non-linearity, and improve short-term forecasting, the proposed method combines an optimal intrinsic mode function decomposition method, a bidirectional long short-term memory (BiLSTM) model with an attention mechanism and integrating deep neural network (DNN) layers, while employing Bayesian optimization with Gaussian processes for hyperparameter tuning. The model effectively captures complex temporal and spatial dependencies in the univariate and multivariate data with different time horizons. Results demonstrate significant improvements in predictive accuracy, with the proposed model outperforming conventional single models and hybrid approaches across all forecast horizons. The proposed model also achieves a significant improvement in computational efficiency compared to the default decomposition method. Achieving an average reduction in computational time of 65.78 % for load data and 49.23 % for wind speed data. An improvement rate of up to 34.17 % for two-day forecasts and 5.37 % for seven-day forecasts for load data in terms of root mean square error. The findings suggest that the proposed method can be a reliable tool for smart grid operators and energy planners. This will aid decision-making processes related to grid operation and renewable energy integration.