Long-Term Memory-Based Extreme Learning Machine for Dynamic Industrial Soft Sensing

Addressing the limitation of extreme learning machine (ELM) as a static model that fails to capture temporal information between data points, the long-term memory-based ELM (LM-ELM) model is proposed. LM-ELM extends the traditional ELM model by incorporating a long-term memory subnet, enabling the extraction of both dynamic and static information. The ability of the model to track dynamic industrial data is enhanced by this improvement. Subsequently, a set of industrial process data is used to validate the performance of the model. The results demonstrate superior performance, indicating that the LM-ELM model meets the requirement for high-precision modeling of industrial process data.