High-Precision Prediction of NO_x, NO₂ and C₆H₆ by Multiple Gas Sensors Using a Novel Cascaded MLP-LSTM Model

Gas sensing applications provide real-time, accurate monitoring crucial for public health and environmental sustain ability, especially outdoors. Outdoor gas sensing faces challenges from sensor distortions due to temperature, humidity, and cross-sensitivity interference. This study employs a sensor array with a machine learning algorithm that combines a Multi-Layer Perceptron (MLP) with a Long Short-Term Memory (LSTM) model, tailored for edge applications to compensate for cross-sensitivity and temperature-humidity variations. The features of MLP stacked with LSTM highlight the breakthrough growth of such network models in gas concentration estimation. In the Experimental results showed significant improvement in gas estimation accuracy with the proposed MLP-LSTM model, achieving R2 values of 0.812, 0.803, 0.644, and 0.997 for CO, NOx, N02, and C6H6, respectively.