Steam turbine power prediction based on encode-decoder framework guided by the condenser vacuum degree.

Area of Science:

• Mechanical Engineering
• Power Systems Engineering
• Artificial Intelligence

Background:

• Steam turbines are critical components in thermal power plants, requiring accurate power output prediction.
• Existing prediction methods often overlook the crucial coupling relationship between steam turbines and condensers.
• Accurate prediction is vital for efficient and stable power plant operation.

Purpose of the Study:

• To propose a novel approach for steam turbine power prediction that incorporates the condenser's influence.
• To explore and leverage the coupling relationship between steam turbines and condensers for improved forecasting.
• To develop a model that enhances the accuracy of steam turbine power output predictions.

Main Methods:

• A novel encode-decoder framework guided by condenser vacuum degree (CVD-EDF) was developed.
• A long-short term memory (LSTM) network encoded historical condenser operation data.
• An attention mechanism and convolutional neural network (CNN) captured local and global information within the encoder.

Main Results:

• The CVD-EDF model demonstrated significant improvements over existing methods on real-world power plant data.
• The proposed method achieved a 32.2% improvement in Root Mean Square Error (RMSE) and a 37.0% improvement in Mean Absolute Error (MAE) compared to LSTM at one-minute intervals.
• The model effectively captured the complex coupling between the condenser and steam turbine operations.

Conclusions:

• The CVD-EDF approach successfully addresses the challenge of steam turbine power prediction by considering condenser dynamics.
• The integration of condenser vacuum degree significantly enhances prediction accuracy.
• This method offers a promising solution for more reliable power output forecasting in thermal power plants.