Related Experiment Videos
Generalized multiple-regression techniques with interaction and nonlinearity for system identification in biological
D A Vaccari1, C Christodoulatos
1Stevens Institute of Technology.
ISA Transactions
|January 1, 1992
Summary
Generalized Multiple Regression (GMR) models offer a robust alternative for complex data modeling, outperforming neural networks and linear models in some applications. While not always the most parsimonious, GMR effectively captures nonlinearities and interactions.
Area of Science:
- Environmental Engineering
- Statistical Modeling
- Water Quality Analysis
Background:
- Traditional time series models like ARIMA may miss complex process behaviors.
- Existing models struggle to effectively capture nonlinearities and interactions among variables.
- There is a need for flexible modeling approaches in environmental data analysis.
Purpose of the Study:
- To introduce and evaluate Generalized Multiple Regression (GMR) models.
- To compare GMR models against ARIMA and neural network models.
- To assess the performance of GMR in modeling environmental processes.
Main Methods:
- Development of Generalized Multiple Regression (GMR) models.
- Application of GMR, ARIMA, and feedforward back propagation neural network models.
- Comparative analysis of model performance using effluent volatile suspended solids and sludge volume index data.
Main Results:
- GMR models demonstrated superiority over linear autoregressive and neural network models for effluent volatile suspended solids.
- Neural network models outperformed linear models in effluent volatile suspended solids modeling.
- GMR and neural network models did not improve upon ARIMA models for sludge volume index.
Conclusions:
- GMR models exhibit robust capability in describing complex environmental data, including nonlinearities and interactions.
- ARIMA models can be parsimonious but may overlook crucial process dynamics.
- GMR presents a valuable, computationally feasible alternative for environmental time series modeling.