Prediction of Ammonia Concentration Under The Influence of Temperature and Humidity Based on LSTM-FNN Algorithm

Ammonia and human production activities are closely related to human prolonged exposure to a certain concentration of ammonia will make life and health suffered certain hazards, so for the detection of low concentration of ammonia sensor demand. Most of the ammonia sensors on the market are affected by temperature and humidity, which will change their responsiveness and reliability, resulting in zero drift of the sensor. Aiming at the problem that ammonia sensor detection is susceptible to temperature and humidity interference, this paper investigates a model of NH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf> concentration prediction algorithm based on LSTM-FNN (Long Short Term Memory Network-Feed Forward Neural Network), which is designed to improve the measurement accuracy of NH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf> sensors under different temperature and humidity conditions. In this work, six graphene NH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf> sensors with high sensitivity and selectivity to ammonia were fabricated to form a sensor array, and their response data to 10 to 100 ppm ammonia were collected under different temperature and humidity environments as inputs to the model, and finally a high-precision ammonia concentration prediction model was obtained after model training and optimisation. The experimental results show that after 1000 iterations using the LSTM-FNN model, the prediction error is as low as 1.60263 ppm and the R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> can reach 98.47% compared with other algorithms.