Modelling the Performance Prediction of Free-Piston Stirling Engine/Reverse Osmosis Desalination System Based on Improved BiLSTM Network

Owing to the exponential advancement of AI technologies, the application scope of machine learning has significantly expanded across diverse disciplines. In desalination system performance prediction, only the traditional shallow artificial neural network (ANN) has been applied, in order to optimize the predictive performance of the model with greater precision, an improved BiLSTM (DBO-BiLSTM) network model that optimises the Dung Beetle Search Algorithm (DBO) to bi-directional long and short-term memory network (BiLSTM) is proposed for free-piston Stirling engine (FPSE)/ Reverse Osmosis Desalination System (SWRO) performance prediction. The system consists of a FPSE power source connected to a single-bar reciprocating booster pump, which generates a pressure difference through the volume change in the pump chamber, and then squeezes seawater through a reverse osmosis membrane for seawater desalination. The prediction model is based on the expansion chamber temperature, compression chamber temperature, inflation pressure, power piston spring stiffness, and distribution piston spring stiffness of the FPSE as inputs, and the water production of the system as output. Based on the dung beetle optimisation algorithm, a conventional gradient descent optimiser is used as an internal optimiser for the bidirectional long- and short-term memory network model. The proposed prediction model was compared with ANN, LSTM, BiLSTM and DBO-LSTM on different accuracy metrics. The results show that the prediction model for reverse osmosis desalination system based on improved BiLSTM has higher accuracy.