Pressure retarded osmosis (PRO) is highly investigated in the literature as one of the blue energy techniques. The PRO membrane plays a key role in harvesting the osmotic energy from a salinity gradient resource and process optimization. Therefore, a well-selected membrane will improve the power density generated in the PRO process to meet a designed power density threshold required for an economic salinity gradient power plant. In order to select a proper membrane for the PRO process, it is crucial to know its intrinsic properties, such as the membrane water permeability, salt permeability, and structural parameters, that impact the process performance. Determining the membrane's exact intrinsic parameters in a full-scale PRO module is challenging and time-consuming, and assuming constant parameters will compromise the accuracy of the results and power generation in the PRO process. This study employs artificial neural networks and Boosting-based tree models to predict the intrinsic parameters of the PRO membrane based on the minimum theoretical power density that could be predetermined and was assumed to be 5 W/m2 in this study. The Random Forest and XGBoost algorithms demonstrate superior predictive power (R2 = 0.97) compared to the other examined machine learning algorithms. The results reveal that machine learning algorithms can provide significant predictive power for the membrane's intrinsic parameters and power density based on the input parameters. Additionally, the algorithms were used to evaluate the feature importance of each input parameter affecting the power density of the pressure retarded osmosis membrane.
Read full abstract