Neural Network Sensitivity and Interpretability Predictions in Power Plant Application
Résumé
Machine learning (ML) models such as Deep Neural Networks (DNN) have become increasingly ubiquitous due to their accuracy and flexibility. However, the lack of interpretability and explainability is why they are uncommon in engineering applications. Meanwhile, the research community has identified interpretability as a hot research topic, leading to confusion in various communities. This paper discusses a methodological framework to define and enhance interpretability in the prediction application of Neural Networks. The methods to deal with this problem are (i) Sensitivity Analysis (SA) for Neural Network prediction (model-specific interpretation tool) and (ii) model-agnostic methods. The latter tools could be used for any ML model prediction. In this study, we enhance the interpretability of the Neural Network predictions with the help of SA and model-agnostic methods. In order to visualize the inputs' impacts on prediction results, Partial Dependence Plots (PDP), Individual Conditional Expectation (ICE), and Accumulated Local Effects (ALE) are used and compared. The prediction of the electrical power (PE) output of a combined cycle power plant (CCPP) has been chosen to demonstrate the feasibility of these methods under real operating conditions. The results show that the most influential input parameter among ambient temperature (AT), atmospheric pressure (AP)), Vacuum (V) and relative humidity (RH) is AT. The visualization outputs allow us to identify the direction (positive or negative) and the form (linear, nonlinear, random, stepwise, ...) of the relationship between the input variables and the model's output.
Origine : Fichiers produits par l'(les) auteur(s)