关键词: 碳中和, 平行控制系统, 长短期记忆神经网络, 注意力机制, 灰度关联分析法, 虚拟系统, 循环流化床, 床温

Abstract:

To pursue the carbon peaking and carbon neutrality,with the accelerating development of renewable energy oriented power system and rising of environment protection awareness, thermal power units have to operate under extreme operating conditions. But the traditional mechanism modelling can hardly realize the accurate prediction on CFB boiler bed temperature. Based on parallel control theory,a real system can operate following the guidance of its virtual counterpart based on computational experiments. The virtual system takes Temporal Pattern Attention for Multivariate Time Series Forecasting （TPA-LSTM）model which can improve the traditional LSTM model's recognition ability of the temporal segments in industrial process by introducing temporal attention as the attention mechanism. Taking gray correlation analysis method to screen the data of the real system can improve the accuracy of the computational experiments of the virtual system. The analysis results show that the mean absolute deviation and mean absolute percent error of the predicted bed temperature can be reduced to 0.131 7 ℃ and 0.014 29%. The model realizes the accurate prediction on CFB boiler bed temperature.

Key words: carbon neutrality, parallel control system, LSTM neural network, attention mechanism, grey correlation degree analysis, virtual system, circulating fluidized bed, bed temperature