Big data-driven methods for gas turbine performance prediction and energy-saving control

Abstract

Gas turbines are subject to fluctuations in performance and increased energy consumption in varying loads, environmental disturbances, and long term operating conditions. Traditional empirical control methods can not make full use of the temporal correlation information in multi-source operation data. Based on Distributed Control System (DCS), Supervisory Control and Data Acquisition (SCADA) and environment monitoring data, a multi-dimension time series model including power, fuel flow, exhaust temperature, compressor pressure, and ambient temperature has been built. The model input is generated by the data alignment, the anomaly correction, the normalization, and the sliding window. The Transformer model is used to extract long-term dependencies during gas turbine operation, enabling predictions of power, heat efficiency, and exhaust temperature. On this basis, the prediction results are combined with real time operation parameters to form a state vector. Simulation results show that the proposed approach is superior to Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU) Temporal Convolutional Network (TCN), in the prediction precision, error stability, and energy saving control efficiency, and can maintain relatively stable operating conditions and optimize energy efficiency under variable load conditions.