Related Experiment Video
Updated: Jun 10, 2025
10:50
Computational Modeling of Retinal Neurons for Visual Prosthesis Research - Fundamental Approaches
Published on: June 21, 2022
1.7K
Soft sensor modeling method and application based on TSECIT2FNN-LSTM
Huangtao Dai1, Taoyan Zhao2, Jiangtao Cao1
1School of Information and Control Engineering, Liaoning Petrochemical University, Fushun, 113001, China.
Scientific Reports
|October 10, 2024
Summary
This study introduces a novel soft sensor model, the TSK-type-based self-evolving compensatory interval type-2 fuzzy Long short-term memory (LSTM) neural network (TSECIT2FNN-LSTM), to improve accuracy in complex industrial processes by better handling variable coupling and parameter sensitivity.
Area of Science:
- Process Control
- Artificial Intelligence
- Fuzzy Systems
Background:
- Complex industrial processes often suffer from low soft sensor modeling accuracy due to multi-variable coupling and parameter sensitivity.
- Existing models struggle to effectively capture long-term dependencies in sequential data, limiting predictive performance.
Purpose of the Study:
- To develop an advanced soft sensor model, the TSK-type-based self-evolving compensatory interval type-2 fuzzy Long short-term memory (LSTM) neural network (TSECIT2FNN-LSTM), to enhance prediction accuracy in complex processes.
- To integrate the strengths of LSTM networks for sequential data with interval type-2 fuzzy logic for improved handling of uncertainty and complex relationships.
Main Methods:
- The TSECIT2FNN-LSTM model combines LSTM's gate mechanism for long-term dependencies with an interval type-2 fuzzy inference system.
- A self-evolving structure learning algorithm, based on firing strength, dynamically generates rules to optimize network architecture.
- Parameter learning employs gradient descent, with unique backpropagation of error terms through time and to upper layers to boost accuracy and memory.
Main Results:
- The TSECIT2FNN-LSTM model demonstrated superior prediction accuracy in predicting alcohol concentration in wine and nitrogen oxide emissions in gas turbines.
- Experimental results confirmed the model's effectiveness in overcoming challenges posed by multi-variable coupling and parameter sensitivity.
- The enhanced error propagation mechanism improved both prediction accuracy and the model's memory capabilities.
Conclusions:
- The proposed TSECIT2FNN-LSTM soft sensor model offers a significant advancement in accurately modeling key variables in complex processes.
- The integration of LSTM and interval type-2 fuzzy logic provides a robust framework for handling sequential data with inherent uncertainties.
- The model's self-evolving structure and enhanced learning mechanism contribute to its superior performance over existing methods.
