A novel ensemble machine learning approach for optimizing sustainability and green hydrogen production in hybrid renewable-based organic Rankine cycle-operated proton exchange membrane electrolyser system

Machine learning is a powerful tool in energy systems, yielding significant results across sectors. This study explores machine learning's use in predicting the operation and performance of a hybrid solar biogas fueled organic Rankine cycle system, integrated with proton exchange membrane technology. The organic Rankine cycle model is validated experimentally, and the data is used for training. The study introduces the ensembled transformer-LSTM model, compared with three machine learning models: Recurrent Neural Networks (RNN), Feed forward Neural Networks (FFNN), and long short-term memory networks (LSTM). The transformer-LSTM model effectively captures both long-term and short-term temporal dependencies in the data, which is crucial for accurate predictions in time-series forecasting, effectively capturing temporal dependencies within the data. These models are used to predict organic Rankine cycle and proton exchange membrane electrolyser system outputs and performance metrics are assessed. Traditional models like FFNN struggled with complex relationships due to the lack of internal memory mechanisms, while RNN and LSTM models had limitations in specific scenarios. The transformer-LSTM model consistently showed superior performance across various operating conditions, achieving a significantly lower overall mean absolute percentage error, with 0.74 % compared to 2.1 % for FFNN, 1.69 % for RNN and 0.89 % for LSTM.