Accurate harmonics mitigation for water desalination plant supplied by multi-energy sources based on deep learning algorithms

This research concerns the water desalination plant supplied by multi-energy sources to reduce production costs and minimize environmental impact. The multi-energy system has solar modules added to the grid sources. Adaptive modeling of water desalination plants as distribution systems is a complex task that requires large memory and processing resources. The ETAP simulation program is used to identify the harmonics resulting from the connection between the small photovoltaic (PV) grid and the main grid. It is important to consider that the value of these harmonics varies based on the climate changes such as temperature and solar radiation of the PV modules. The proposed technology aims to identify individual and overall adaptive distortion resulting from various sources of adaptivity. To address this complexity, the current paper presents Long Short-Term Memory Networks (LSTM) relying on a deep learning algorithm predictive model. The effectiveness of the proposed prediction model is evaluated through an assessment study using the coefficient of determination and mean square error. To achieve realistic results and draw accurate conclusions, this research mainly focuses on a real sea water desalination plant supplied from multiple energy sources. The current paper demonstrates the superior performance of the suggested deep-learning prediction model through a comparison with previously published techniques in the literature. The key finding of this research is that the proposed identification models effectively mitigate harmonic distortion and provide accurate harmonic design for the active power filter. The research results indicate a comparison between LSTM networks outperformed, Multiple Linear Regression Analysis (MLRA) and Artificial Neural Networks (ANNs). The minimum Root Mean Squared Error (RMSE) for LSTM was estimated at 0.16e-20, while the RMSE values for MLRA and ANNs were 0.184e-13 and 0.97588, respectively. The results were further documented by comparing the performance of each algorithm with the ETAP simulation results, which demonstrated the alignment between the outcomes obtained using LSTM and the simulated results. Moreover, the study examined Total Harmonic Distortion (THD) under full, half, and light loading conditions, both with and without Active Power Filters (APF), to validate LSTM's capability to predict and mitigate harmonic distortion.