Session: 02-02: AI for Energy Sustainability II

Paper Number: 156886

156886 - Time-Series Ai Forecasting of Wind-to-Hydrogen Production 

Abstract: 

The transition to solutions offering carbon-neutral energy has created, in due course, an uptrend in reliance on renewable sources like wind power and poignantly raises the need for efficient systems of hydrogen production. This paper established a study related to integrating advanced AI techniques of Long-Short Term Memory (LSTM) and Bidirectional-LSTM (Bi-LSTM) with Permanent Exchange Membrane (PEM) electrolyser for better hydrogen production. The research herein will be performed to develop an optimized wind energy forecast and its application in hydrogen generations by leveraging strengths from such AI models.

A complete dataset was used composed of wind speed, active power, and wind direction readings; hence, proper preparation of quality inputs included normalization and the transformation of wind direction into sine and cosine components to embed the circular characteristics of this latter variable. In this dataset, the LSTM and Bi-LSTM models were trained and evaluated. The Bi-LSTM model performed the best in predictions, since it captures information both forward and backward in time. For the performance evaluation of these models, RMSE was used, with Bi-LSTM reaching a testing RMSE lower compared to the LSTM.

The wind energy forecast was then coupled through a PEM electrolyzer model that embodied critical electrochemical parameters along with activation, ohmic, and concentration overpotentials. Optimization of the key design variables such as membrane thickness and operating temperature came into view to obtain maximum efficiency. The results showed that the optimum efficiency of 63.45% may be obtained at a membrane thickness of 0.00254 cm (N211) and an operating temperature of 70°C.

Besides, applying the forecasts obtained by Bi-LSTM increased the monthly hydrogen production by 5–8% compared to the forecasts made by LSTM.

These findings are thus in line with the literature and support the strength of the Bi-LSTM in modeling complex temporal dependencies, further demonstrating the efficiency of the PEM model with respect to state-of-the-art benchmarks. Coupling AI-driven forecasting and optimization of the electrochemical system, the present work underlines an integrated approach to foster renewable energy consumption and hydrogen production. In agreement with previous studies, scalability, and reliability of PEM electrolyzers are also underlined, capable of converting renewable energy into sustainable hydrogen. Future research may involve the use of other different hybrid AI architectures, such as combinations of LSTM variants with other machine learning models, while real-time integration of wind farms with dynamically optimizing PEM systems is also a venue of future research. It forms part of a worldwide contribution toward reaching a sustainable energy future by showing how AI and electrochemical innovations are together going to move the renewable energy landscape forward and speed up the transition toward a carbon-neutral economy.

Presenting Author: Doha Bounaim Mississippi State University

Presenting Author Biography: Doha Bounaim currently completes the final year of her Master of Science in Mechanical Engineering from Mississippi State University, expected to graduate in December 2024. The major area of her research is green hydrogen forecasting and renewable energy systems. In January 2025, she will begin her PhD in Mechanical Engineering, with a primary focus on applications of machine learning. She is committed to innovative solutions in the fields of energy and sustainability.