Enhancing Power Generation Forecasting in Smart Grids Using Hybrid Autoencoder Long Short-Term Memory Machine Learning Model

Abstract

This study explores the implementation of advanced machine learning techniques to enhance the integration of renewable energy into smart grids, focusing specifically on predicting solar power generation for the upcoming year. Three distinct machine learning models are employed: the Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM) and a hybrid model that combines Autoencoder and Long Short-Term Memory (AE-LSTM). Using real-time solar power production data spanning a year, these models are trained and evaluated using mean absolute error (MAE) and mean squared error (MSE) as performance metrics. Results highlight the superior accuracy of the hybrid AE-LSTM model compared to the LSTM model as well as Bi-LSTM model, attributed to its capability to capture intricate temporal patterns and correlations within the data. This research underscores the significant potential of machine learning techniques, particularly the hybrid AE-LSTM approach, in facilitating the seamless integration of renewable energy resources into smart grids, contributing to more efficient and environmentally conscious power systems. Furthermore, a comprehensive analysis reveals the hybrid AE-LSTM model's capacity to produce superior predictions due to its additional training, solidifying its advantage over models solely reliant on the other model's architecture. In summary, this study demonstrates the effectiveness of advanced machine learning methodologies in revolutionizing renewable energy integration, with the hybrid AE-LSTM model standing out as a promising avenue for enhanced prediction accuracy.

Machine learning autoencoder‐based parameters prediction for solar power generation systems in smart grid

“…Among the three ML models employed for solar power production forecasting, the hybrid Autoencoder-Long Short-Term Memory (AE-LSTM) model demonstrated higher accuracy, outperforming the LSTM and Bi-LSTM models. This advantage is attributed to the hybrid model's capacity to identify complex temporal patterns and relationships in the data, as well as its further training, which solidified its superiority over other models [29]. Similarly, the hybrid method combining an LSTM neural network and an autoencoder for solar energy forecasting outperformed stateof-the-art models, demonstrating its superior predictive capabilities by effectively capturing both temporal and spatial features in the data [30].…”

“…The encoder is simply used by us to obtain the original data's properties once the autoencoder's training is complete so as to improve the data's internal structure. The Equations ( 7) and ( 8) for encoding and decoding are as follows [29]:…”

“…Equation ( 15) is the formula to determine the MAE between the target variable's real and predicted scores is [29][30][31].…”

“…In contrast, the CNN LSTM model achieved an MAE of 0.14611, while the LSTM model had a MAE of 0.16188. In the same way, Equation ( 16) is the formula to determine the MSE between the target variable's real and predicted scores [29][30][31].…”