Circuit Structure Dependent Multi-Variable Regression Model Based Predictions for DC-DC Converters

• Published in: Conference proceedings - IEEE Applied Power Electronics Conference and Exposition pp. 2703 - 2708
• Main Authors: Khamis, Ahmed K., Agamy, Mohammed
• Format: Conference Proceeding
• Language: English
• Published: IEEE 19.03.2023
• Subjects: Bond Graph; Circuit synthesis; DC-DC power converters; Electric circuit; Graph Neural Networks (GNN); Machine Learning; Measurement uncertainty; Network topology; Neural networks; Power electronics; Predictive models; Regression Model
• ISSN: 2470-6647
• DOI: 10.1109/APEC43580.2023.10131352

The new regression model based on converter circuit structure is introduced, allowing for predictions on converter circuit level and internal parameter predictions. The proposed model is applicable to any converter circuit regardless of the number of components or connections present in the circuit. This is achieved by properly mapping all physical and switching circuit properties whether the converter circuits working in continuous conduction mode (CCM) to graph representation, so that machine learning (ML) algorithms and applications can be easily applied. This is useful in applications such as AI based circuit synthesis, automatic circuit layout generation, AI circuit design automation and many others. Three of the most common converters (Buck, Boost, and Buck-boost) are used as example circuit applied to model and the target is to predict the gain and current ripples in inductor. The model achieves 99.88% on the R^{2} measure and a mean square error (MSE) of 0.023.