Spatio-Temporal Consistent Soft Sensor Modeling and Monitoring of Thermal Power Plants Based on Physical Knowledge

Data-driven soft sensors have been widely applied in thermal power plants. However, the consistency of spatio-temporal features and the explicit interpretability in data-driven soft sensors remain challenges. In this light, this work introduces a novel method termed spatio-temporal consistent and interpretable model (STCIM). First, local and far topological spatio-temporal features are captured by a spatio-temporal consistent feature extraction block, in which a local spatio-temporal feature extraction block and a far topological alignment block are designed. Then, the spatio-temporal consistent features are projected into the latent space, which is guided by the first principle and explicitly assigned physical meanings. During the training process, the latent space is integrated into the loss function to predict the state variables and the output and maintain the interpretability of the model. The efficacy of the proposed STCIM is demonstrated through the modeling of two generated datasets and a real-life dataset of coal-fired power plants. The corresponding experiments show: 1) The generalization of STCIM outperforms other methods, especially in different operation situations. 2) The far topological spatio-temporal consistency is vital for feature alignment. 3) The hyper-parameters of physics-informed interpretable latent space loss function decide the performance of state variable estimation, especially the variables whose slight changes will affect the output.