Neural networks in multiphase reactors data mining: feature selection, prior knowledge, and model design
|Advisor:||Grandjean, Bernard; Larachi, Faïcal|
|Abstract:||Artificial neural networks (ANN) have recently gained enormous popularity in many engineering fields, not only for their appealing “learning ability, ” but also for their versatility and superior performance with respect to classical approaches. Without supposing a particular equational form, ANNs mimic complex nonlinear relationships that might exist between an input feature vector x and a dependent (output) variable y. In the context of multiphase reactors the potential of neural networks is high as the modeling by resolution of first principle equations to forecast sought key hydrodynamics and transfer characteristics is intractable. The general-purpose applicability of neural networks in regression and classification, however, poses some subsidiary difficulties that can make their use inappropriate for certain modeling problems. Some of these problems are general to any empirical modeling technique, including the feature selection step, in which one has to decide which subset xs ⊂ x should constitute the inputs (regressors) of the model. Other weaknesses specific to the neural networks are overfitting, model design ambiguity (architecture and parameters identification), and the lack of interpretability of resulting models. This work addresses three issues in the application of neural networks: i) feature selection ii) prior knowledge matching within the models (to answer to some extent the overfitting and interpretability issues), and iii) the model design. Feature selection was conducted with genetic algorithms (yet another companion from artificial intelligence area), which allowed identification of good combinations of dimensionless inputs to use in regression ANNs, or with sequential methods in a classification context. The type of a priori knowledge we wanted the resulting ANN models to match was the monotonicity and/or concavity in regression or class connectivity and different misclassification costs in classification. Even the purpose of the study was rather methodological; some resulting ANN models might be considered contributions per se. These models-- direct proofs for the underlying methodologies-- are useful for predicting liquid hold-up and pressure drop in counter-current packed beds and flow regime type in trickle beds.|
|Document Type:||Thèse de doctorat|
|Open Access Date:||11 April 2018|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.