Due to the nonlinear nature of power converters, unpredictable operating conditions and the need to preserve physical consistency, using neural network models to predict performance beyond training data presents a significant challenge. The ability to extrapolate circuit performance lays the foundations for a robust machine learning platform for electronics design automation and circuit synthesis. In this paper, a framework for extrapolation of circuit dynamics beyond the range of the training data is proposed. To achieve this objective, a heterogeneous-physics-informed graph neural network is developed. In this method, heterogeneous graphs of electronic circuits are used to create a unique representation for each converter topology, which are then fed as inputs to a physics-informed graph neural network, thus allowing simultaneous learning of converter dynamics while enforcing the physical circuit laws. Furthermore, a detailed analysis of the impact of activation functions on mapping nonlinear circuit behavior is studied to identify and select the optimal activation function shape. The proposed approach is validated on different DC-DC converter topologies, achieving accurate interpolation and extrapolation performance.

Machine learning (ML) has the potential to revolutionize electronic design automation (EDA), but mainly lacks the scalable and adequate representation of electric circuits, and requires diverse datasets that capture the performance of various electrical circuits. Literature shows methodologies that fall short on representing unlimited complexity circuits, in which models are hard coded to a limited certain circuit order, and are more focused on using a fixed circuit topology rather than providing a general purpose representation. This paper proposes a generalizable and scalable circuit graph representation based on bond graphs modeling as heterogeneous graph. Additionally, an analytical comparison between mapping the bond graph as either homogeneous or heterogeneous graph is shown. By analyzing the graph in spectral domain, the frequency component distribution of node features is found to impact the performance of Graph Neural Networks (GNNs) unique to the graph representation used. The analysis discusses inherent limitations of GNNs, and reveals that GNNs designed for heterogeneous graphs can more effectively capture system dynamics & variations. Finally, different case studies are presented utilizing heterogeneous graph neural network (Hetero-GNN) for different ML tasks including dynamics prediction and converter classification.

Virtual Oscillator Control (VOC) represents a decentralized control technique in the time domain, offering promising prospects such as fast dynamic response, accurate power sharing, and synchronization among paralleled inverters in islanded mode. However, it requires inner voltage-current loops for voltage regulation at the inverter output. These inner loops have conventionally relied on linear controllers such as Proportional–integral(PI) or Proportional-Resonant (PR), contributing to complexity and limiting dynamic response. This paper introduces an alternative to conventional linear loops by employing Finite-set Model Predictive Control (FS-MPC). By integrating the VOC and FS-MPC, the voltage at the output of the inverter can be predicted, and the error of tracking reference be minimized. Also, within the proposed method, a multi-objective optimization problem can be formulated enabling over-current limiting capabilities. It also facilitates a simplified control structure, broad bandwidth, and faster dynamic response by removing the need for PWM and inner linear controllers. Simulation results validate the efficacy of the proposed approach for single and multiple paralleled inverters.

This paper introduces a neural network based approach for parameter identification of a half-bridge LLC Resonant Converter. By using a set of measured inputs, the system can characterize the component values, and difficult to measure switch characteristics including the junction temperature or gate-source capacitance. This capability is useful for reliability, health monitoring and model based control applications. The range of the Mean Absolute Error (MAE) obtained for all parameters predictions was from 0.142% to 4.12%, with an average MAE of 1.57% over all parameters.

Machine learning (ML) has the potential to revolutionize the design and optimization of electronic systems, but mainly requires high-quality and diverse datasets that capture the features and performance of various electrical circuits. Previously, there were no generic method applicable to any electric circuit whether continuous circuit like resonant LC circuits or switching circuits like buck, boost and buck-boost that allows circuit connections and components separate representation when circuits are fed to ML models. Moreover, all other methods and models are inefficiently built to process either limited complexity circuits, i.e. limited to a certain circuit order, or use a fixed circuit topology so that the generated dataset is of a definite size. In this paper, a novel circuit representation that is based on heterogeneous graph is presented, which is a graphical representation of the energy flow in a physical system including electrical, mechanical and even chemical interactions. Moreover, heterogeneous graph neural network (Hetero-GNN) is applied to the proposed representation of heterogeneous graphs representing electric circuits, allowing for the different ML tasks on different electric circuit and converter applications.

In wireless power transfer systems, many parameters affect the performance of WPT systems. One of the critical parameters in the wireless power transfer system is the coupling factor that relies on the primary and the secondary coils and the mutual inductance between these coils. Therefore, coupling factor estimation is a challenging issue in wireless power transfer. Many papers concentrate on this issue and explain ways to estimate the coupling factor between the primary and the secondary sides. Using the primary and secondary currents is conventional methods to estimate the coupling factor, but these methods have significant disadvantages. These methods must have many system parameters, complicating the control system. In this paper, a new method of coupling factor estimating addresses this problem in a wireless power transfer system with series-series topology. In this method, a modified 3rdharmonic current is used for impedance and coupling factor estimation. While the conventional methods need all the system parameters, this method requires the primary inductance and input impedance, the inverter output voltage, and the primary current. It reduces the complexity of the control system and coupling factor estimation error. The viability of a proposed method is verified with an experimental prototype.

A regression model based on graph neural network, tailored for electric circuit dynamics prediction is introduced, providing converter performance predictions on converter circuit level and internal parameter variations. Regardless of the number of components or connections present in a converter circuit, the proposed model can be readily scaled to incorporate different converter circuit topologies. Moreover, the model can be used to analyse converter circuits with any number of circuit components and any control parameters variation. To enable the use of machine learning methods and applications, all physical and switching circuit properties such as converter circuits operating in continuous conduction mode or discontinuous conduction mode are accurately mapped to graph representation. Three of the most common converters (Buck, Boost, and Buck-boost) are used as example circuits applied to model and the target is to predict the gain and current ripples in inductor. The model achieves 99.51% on the $$R^2$$ measure and a mean square error of 0.0263.

The frequency instability due to the lack of inertia in the inverter-dominant power system can be alleviated by applying an appropriate virtual inertia (VI) control. A modified dispatchable virtual oscillator control (dVOC), which is a combination of conventional dVOC and an extra loop for providing VI is proposed for frequency stability improvement. This controller, in addition to providing advantages of conventional dVOC such as fast dynamic response and accurate power sharing among inverters, improves the frequency response by adding VI to system. However, a constant VI control is not an efficient approach as it is not able to distinguish various system operating conditions requiring large or small inertia. Thus, in this work an adaptive VI control is proposed in which the control coefficients are dynamically updated in response to different operating conditions. This method reduces the exchanged power during inertial response by accelerating transient response. Also, it does not require frequency measurement using phase-locked loop (PLL) in VI loop. Therefore, this approach not only improves transient response of the system but also removes possible instability due to using PLL. The controller behavior is analyzed and validated for different cases under large transient conditions.