Deep-Learning-Based Power Generation Forecasting of Thermal Energy Conversion.

Area of Science:

• Energy Systems Engineering
• Artificial Intelligence in Engineering
• Thermodynamics and Heat Transfer

Background:

• Organic Rankine Cycle (ORC) technology efficiently converts low-grade thermal energy into electricity.
• Variations in temperature, pressure, and flow can destabilize ORC power generation.
• Accurate power output prediction is crucial for stable grid integration and operational efficiency.

Purpose of the Study:

• To develop an advanced deep learning methodology for predicting ORC power generation.
• To forecast power output 12 hours in advance for enhanced grid stability.
• To enable early detection of abnormalities in ORC systems.

Main Methods:

• Utilized a deep learning neural network based on Long Short-Term Memory (LSTM) recurrent neural networks (RNN).
• Applied the methodology to real-world ORC data from a steel company case study.
• Compared LSTM performance against traditional time series models like ARIMA and MLP.

Main Results:

• The LSTM-based deep learning model demonstrated superior predictive performance.
• Achieved a 24% reduction in error rate compared to other time series methodologies.
• Validated the effectiveness of the proposed prediction approach.

Conclusions:

• The developed LSTM methodology provides accurate, long-term power generation forecasts for ORC systems.
• This predictive capability supports optimized system warning thresholds and early abnormality detection.
• Offers a novel approach for predictive maintenance and diagnostics in industrial settings.