Solar Thermal Process Parameters Forecasting for Evacuated Tubes Collector (ETC) Based on RNN-LSTM

Abstract

Solar Heat for Industrial Process (SHIP) systems are a clean source of alternative and renewable energy for industrial processes. A typical SHIP system consists of a solar panel connected with a thermal storage system along with necessary piping. Predictive maintenance and condition monitoring of these SHIP systems are essential to prevent system downtime and ensure a steady supply of heated water for a particular industrial process. This paper proposes the use of recurrent neural network-based predictive models to forecast solar thermal process parameters. Data of five process parameters namely - Solar Irradiance, Solar Collector Inlet & Outlet Temperature, and Flux Calorimeter Readings at two points were collected throughout a four-month period. Two variants of RNN, including LSTM and Gated Recurrent Units, were explored and the performance for this forecasting task was compared. The results show that Root Mean Square Errors (RMSE) between the actual and predicted values were 0.4346 (Solar Irradiance), 61.51 (Heat Meter 1), 23.85 (Heat Meter 2), Inlet Temperature (0.432) and Outlet Temperature (0.805) respectively. These results open up possibilities for employing a deep learning based forecasting method in the application of SHIP systems. ABSTRAK: Penggunaan sumber bersih seperti Tenaga Solar dalam Proses Industri (SHIP) adalah satu kaedah alternatif untuk menhasilkan tenaga yang boleh diperbaharui bagi mengurangkan kesan gas rumah hijau yang terhasil dari proses industri. Sistem SHIP biasanya mengandungi panel solar dan sistem penyimpanan haba yang berhubung melalui paip yang sesuai. Penyelengaraan secara berkala diperlukan bagi memastikan sistem ini sentiasa membekalkan tenaga solar pada kadar bersesuaian dan bekalan tenaga solar yang terhasil berterusan dan tidak menjejaskan sistem pemanasan air bagi sesuatu proses industri. Kajian ini mencadangkan penggunaan model ramalan rangkaian neural berulang bagi meramal parameter proses pemanasan solar. Kelima-lima parameter proses iaitu – Iradiasi Solar, Suhu Saluran Keluar & Masuk Pengumpul Solar dan Bacaan Kalorimeter Fluks pada dua tempat diambil sepanjang empat bulan (dari Julai 2021 sehingga Oktober 2021). Dapatan menunjukkan dua varian RNN termasuk LSTM dan Unit Berulang dapat dibanding prestasinya bagi tugas ramalan ini. Dapatan kajian menunjukkan Ralat Punca Min Kuasa Dua (RMSE) antara bacaan sebenar dan ramalan adalah masing-masing 0.4346 (Iradiasi Solar), 61.51 (Meter Terma 1), 23.85 (Meter Terma 2), Suhu Salur Masuk (0.432) and Suhu Salur Keluar (0.805). Ini membuka peluang kajian mendalam berdasarkan kaedah ramalan dalam aplikasi sistem SHIP.

Keywords

• Solar energy
• Machine learning techniques
• Artificial Neural Networks (ANNs), Stock Market, Prediction, Stock Exchange, Backpropagation, FeedForward.