Soft Sensor Modeling of Acrylic Acid Yield Based on Autoencoder Long Short-Term Memory Neural Network of Savitzky–Golay and ReliefF Algorithm

Abstract

Acrylic acid yield (AAY) is a key quality index in production process of acrylic acid. Meanwhile, AAY has been considered as direct characterization of productivity. Aiming at the difficulty of online measurement of AAY in acrylic acid process, a soft sensing model of AAY based on autoencoder long short-term memory neural network (AE LSTM NN) applying Savitzky–Golay and ReliefF method is presented in this paper. Firstly, Savitzky–Golay method with denoising effect is adopted to remove industrial noise in measurement. Then, ReliefF algorithm is developed to compress characteristic variables from the result of denoising. Finally, AE LSTM is employed to predict the AAY in acrylic acid process. In contrast to LSTM, support vector machine, and artificial neural network, the root mean square error (RMSE) of the provided method is 0.0954, mean absolute error (MAE) is 0.0757, mean absolute percent error (MAPE) is 0.09%, and maximum absolute error (MaxAE) is 0.3236, which shows validity and superiority.