Convolutional neural network-long short term memory optimization for accurate prediction of airflow in a ventilation system

Poor airflow ventilation systems fetch a progressively critical challenge for many working areas, which transmits many calamitous physical consequences on operatives’ health and quality of work. However, accurate monitoring and prediction of the ventilation systems airflow remain challenging due to the multiple properties and non-linear characteristics in time and space. Machine learning and deep neural network techniques have recently received significant consideration for their real-world applications in numerous areas. This affluence key feature is a deep neural network motivated by the data handling in biological brains. In this article, we applied one of the representative deep neural network techniques, i.e., 1D-CNN with LSTM, to predict the variation in the airflow of the ventilation system. These utilize CNN advantage, which effectively extracts the systems feature, whereas the LSTM can imitate the long-term sequential progression of input time-series data. It provides SHAP analysis that can be used to understand the output of the proposed model to forecast the airflow of the ventilation system. This method computes an approximation of the influence of individual features for predicting the non-linear element. Subsequently, five models, i.e., CNN, LSTM, 1D-CNN-LSTM, ANN, and LR, are used to predict the ventilation system's airflow. The result shows that the proposed model 1D-CNN-LSTM accuracy and loss are 96.7 %, and 0.01348 provides the finest result compared to others. The aim of this research lies in the application of a complex model to interpret the airflow of the ventilation system. It is of great interest as it consents us to comprehend how a model behaves and enables us to take pre-emptive methods to improve working efficiency.