A Data-Driven Self-Supervised LSTM-DeepFM Model for Industrial Soft Sensor

Soft sensor, as an important paradigm for industrial intelligence, is widely used in production to achieve efficient monitoring and prediction of status including product quality. Data-driven soft sensor methods have attracted attention, which still challenges because complex data with diverse characteristics, nonlinear relationships, massive unlabeled samples. In this article, a data-driven self-supervised long short-term memory–deep factorization machine (LSTM-DeepFM) model proposed framework mainly pretraining finetuning stages explore characteristics. the stage, LSTM-autoencoder first unsupervised pretrained. Then, based on two mask strategies, LSTM-deep can interdependencies between features well dynamic fluctuation time series. relying pretrained representation, temporal, high-dimensional, low-dimensional be extracted from LSTM, deep, FM components, respectively. Finally, experiments real-world mining dataset demonstrate that method achieves state art comparing stacked autoencoder-based models, variational semisupervised parallel DeepFM, etc.