Performance analysis of Solar Heat Industrial Process (SHIP) system in malaysia and energy output forecasting based on optimized deep learning technique [EMBARGOED]

Abstract

Industries commonly use natural gas boilers to fulfil the hot water and steam requirement of the factories. The emission of greenhouse gasses increase in the environment due to the vast usage of natural gas boilers. Therefore, solar heat for industrial process (SHIP) systems are being introduced around the globe to supply hot water and steam in the processing activities of factories. In the first part of this work, the performance of the SHIP system is evaluated, which was installed at the oleochemical factory's rooftop in Johor Bahru, Malaysia. The SHIP system comprises 75 evacuated tube collector (ETC) solar thermal panels. The performance is evaluated on monthly and quarterly data monitored from January 2021 to December 2021. First, a comprehensive analysis is conducted on nine performance parameters; useful energy, delivered energy, supply temperature, collector inlet temperature, collector outlet temperature, temperature difference, collector efficiency, system efficiency, and system losses. Secondly, an hour ahead forecasting of solar thermal output is performed quarterly for the SHIP system over the same period (JAN-DEC 2021), based on forecasting accuracy measurement parameters such as RMSE, MSE, r and R2. A deep learning method (LSTM-RNN) is proposed and compared with other machine learning (ANN, CNN & NARX) for an hour ahead forecasting of SHIP system output on a quarterly basis for one-year data. Finally, particle swarm optimization (PSO) is used as a hybrid forecasting technique (PSO-LSTM) to tune the hyperparameters of the developed deep learning method to enhance its forecasting accuracy and compared with LSTM-RNN, SSA-LSTM, and GA-LSTM. Performance analysis findings show that the SHIP system performs better, with a monthly average efficiency of 56.38% and 51.92% for useful and delivered energy, respectively. Moreover, the SHIP system has avoided 104.74 tons of CO2 emissions in one year. On the other hand, forecasting results show that the proposed deep learning technique (LSTM-RNN) has presented lower (RMSE, MSE) and higher (r and R2) compared to other techniques. The developed PSO-LSTM hybrid method gave an average reduction of (30.75% and 51.10%) and (25.99% and 44.18%) in the (RMSE and MSE) for useful and delivered energy, respectively. Furthermore, the proposed PSO-LSTM method showed higher r and R2 than GA-LSTM and SSA-LSTM. In addition, the proposed deep learning and hybrid models (PSO-LSTM) are found robust and flexible in predicting power output for the SHIP system over one year.