Performance prediction of air-cooled unit based on hierarchical attention residual network

The cooling fan is a key component to ensure the normal operation of complex systems under rated load. It is of great significance to analyze and predict its degradation trend. The existing methods mainly have deficiencies such as insufficient capture of temporal dependencies and single prediction information. For this purpose, this paper proposes a performance prediction method for cooling fans based on Hierarchical Attention Residual Temporal Network (HARTNet). First, the attention module composed of a fully connected layer and a Softmax activation function is used to dynamically weight different features. Then, the parallel time feature extraction model with attention mechanism is used to represent the long-term and medium-term dependency times. After that, shared feature layer is used to simultaneously accomplish the tasks of fault classification and performance degradation prediction. Finally, experimental verification on the air-cooled unit has demonstrated the effectiveness and practicability of the proposed method.