Nov 2026· IEEE transactions on power electronics· Vol 41, pp. 19030-19044· 0 citations· 35 references
Abstract
The dynamic current imbalance between the paralleled SiC mosfets in multichip power modules, which is commonly attributed to the asymmetric module layout, severely limits their current capacity and thermal reliability. Adjusting the connection points of bonding wires is an effective method to mitigate imbalanced dynamic current. However, manual trial-and-error is currently the most common method for optimizing connection points, which is both deficient and inefficient. Existing automated solutions usually rely on a prefitting process based on large datasets, which is time-consuming and impractical for high-dimensional parameter applications. Thus, this article proposes an optimization model to mitigate dynamic current imbalance, which can automatically adjust the connection points of bonding wires without any manual intervention. The reinforcement learning (RL) soft actor-critic algorithm was applied to the power module optimization, eliminating the need for prefitting and enabling high-dimensional parameter optimization. After optimization, nearly complete dynamic current balancing in both high-side and low-side switches in a multichip-paralleled half-bridge power module is achieved, as verified by simulations and experiments. This model achieves true dynamic current balancing automation for the first time, providing an important reference for the application of RL to the automated optimization of multichip power modules.
The problem of Optimal Power Flow (OPF) is central to the secure and economic operation of modern power systems. However, increasing renewable energy penetration, and decreasing system inertia pose significant challenges to conventional optimization-based OPF solvers. While machine learning approaches have demonstrated substantial computational speed-ups, purely data-driven methods often suffer from data dependency, limited generalization, and lack of guaranteed physical feasibility. This paper suggests a physics-informed neural network (PINN) framework for solving the OPF problem in renewable energy-dominated, low-inertia power systems. In contrast to conventional OPF formulations, the model explicitly incorporates a location-aware inertia constraint based on the concept of system inertia strength, which accounts for the electrical distance between generation units and disturbance locations. Simulation results on a 6 GW test system demonstrate high accuracy. The mean absolute error (MAE) for both the training and testing datasets is approximately 0.045% of the total system capacity. The findings demonstrate that the proposed PINN framework is capable of producing highly accurate OPF solutions while ensuring compliance with both physical laws and inertia-related constraints. Overall, the findings highlight the potential of physics-informed learning to enable secure, efficient, and computationally scalable OPF for future low-inertia power systems.
Mahyar Tofighi-Milani, S. Fattaheian‐Dehkordi, F. Rohrhofer et al.· 0 citations
This study addresses energy saving in parallel chiller plants under wide fluctuations of air-conditioning demand. Built-in local controls of individual chillers, including partial-load valve regulation and variable-frequency control, are not sufficient when heterogeneous machines must be staged in parallel. The central algorithmic problem is to learn an accurate mapping from partial load to coefficient-of-performance (COP) so that online grouping and load allocation avoid efficiency cannibalization among mismatched chillers. To solve this problem, the study proposes a physics-informed generative adversarial network digital twin (PI-GAN-DT) that learns unit efficiency curves from sparse manufacturer and field data while enforcing cooling-energy consistency and thermodynamic monotonicity. The learned mapping is then embedded in a supervisory optimization model for optimal chiller sequencing and loading. Two industrial electronic-factory plants are retained only as benchmark cases for empirical verification. Using the archived manufacturer part-load tables and plant load trajectories, the benchmark/simulation study (i.e., results obtained on the archived part-load tables and on a contiguous 84-day campaign rather than on closed-loop deployment) reports daily chiller-energy reductions of 574.2 kWh/day and 673.5 kWh/day in Case A, 2,073.6 kWh/day in Case B at 7 ℃, and an additional 1,601.1 kWh/day when Case B is operated at 9 ℃. These benchmark/simulation results serve as evidence of the effectiveness of the proposed framework; the corresponding closed-loop field savings remain to be confirmed by the A/B field campaign described in the deployment-path discussion. The main contribution of the study is the PI-GAN-DT-based optimization algorithm rather than the benchmark cases themselves.
Tsung-Yin Ou, Yenming J. Chen· Journal of Circuits, Systems...· 0 citations
Dual active bridge (DAB) converters are pivotal in applications requiring bidirectional power flow, such as automotive and aerospace systems. The input series and output parallel (ISOP) configuration addresses the need for higher output currents by reducing the voltage stress on individual modules, yet it faces challenges like slow dynamic response, parameter sensitivity, and difficulties in maintaining output current sharing amid component mismatches and input disturbances. This study establishes an evaluation framework to quantify current-sharing risks in ISOP-DAB systems under extended phase shift (EPS) modulation, analyzing how inductance discrepancies and input voltage dispersion propagate uncertainties and define risk boundaries. By developing a parameter-uncertainty model and applying a multi-objective genetic algorithm, the research concurrently optimizes inductor current stress and the reflux of power. MATLAB simulations validate the theoretical derivations, demonstrating effective current sharing and enhanced performance under multivariable uncertainties. The conclusions confirm the framework's accuracy in guiding the robust design of ISOP-DAB converters, achieving a balance between operational stability and efficiency.
Tianrui Zhao, Jia-Qi Li, Zi-Xuan Yang et al.· International Conference on...· 0 citations
With the increasing dc-bus voltage level and power rating in renewable-energy applications, dual-parallel five-level active neutral-point-clamped (5L-ANPC) inverters have attracted growing attention because they can increase system capacity and improve output current quality. However, under asynchronous strategy, the dual-parallel 5L-ANPC inverters face a large number of space voltage vectors. Meanwhile, the simultaneous requirements of current tracking, CMV reduction, circulating current suppression, and capacitor voltage balancing further increase the difficulty of multi-objective control. To address these issues, this paper proposes a three-layer simplified model predictive control (MPC) strategy. The dual-parallel system is regarded as an equivalent nine-level inverter to improve the output current quality. To avoid the heavy burden caused by the large space voltage vector diagram, the gh coordinate system is adopted to calculate the required vector coordinates online, and low-CMV vector states are further selected to synthesize the reference voltage ( $V_{ref}$ ) without lookup tables (LUT). Then, a dynamic circulating current control strategy is developed to reallocate the vector states of each 5L-ANPC inverter, so that differential-mode current (DMCC) and zero-sequence circulating current (ZSCC) can be suppressed even under conditions of severe inductance mismatch. Finally, redundant switching combinations are used to balance the floating capacitor voltage (FCV) and neutral-point voltage (NPV). Experimental results show that compared with the advanced synchronous MPC strategy, the proposed strategy reduces the CMV peak-to-peak value by half, decreases the CMV RMS by up to 50%, and lowers the output current THD by at least 25%.
This study aims to develop a two-stage Bayesian optimization framework to improve multiphysics electric-machine design, targeting high performance under stringent electromagnetic, thermal, mechanical and economic constraints while keeping finite-element evaluation costs manageable.
Stage I performs a multi-objective tree-structured Parzen estimator (TPE) search to map efficiency–power–cost trade-offs and build a Pareto surrogate. Stage II applies permutation-importance screening and an improved, constraint-aware TPE with dynamic sample filtering to refine key parameters near the feasible boundary.
In a surface-mounted permanent-magnet synchronous motor case, the proposed method produced 48 designs meeting six hard constraints, including = 93% system efficiency and slot fill = 0.85, within 1,147 finite-element runs. It yielded over an order of magnitude more feasible, high-performance designs than single-stage or unrefined two-stage baselines at identical computational budgets.
The framework is the first, to the best of the authors’ knowledge, to integrate explainable permutation importance with dynamically reweighted TPE sampling for constrained electric-machine optimization, simultaneously enhancing feasibility, performance and parameter diversity. It offers a transferable, lightweight template that augments existing multiphysics workflows without altering underlying solvers, supporting better engineering and manufacturing decisions.
Yi-Rong Lin, Chang-Yi Kao, Nien-Yi Jan et al.· Compel· 0 citations
This paper proposes an online state-of-charge (SoC) balancing strategy for a distributed single-stage Modular Multilevel Converter-based Battery Energy Storage System (MMC-BESS). The main contribution is the development of a low-complexity sorting-based selection algorithm integrated into the smart-battery concept, enabling real-time energy redistribution among submodules without requiring additional auxiliary balancing circuits. The proposed method operates jointly with a phase-disposition pulse width modulation (PD-PWM) strategy and is coordinated with conventional grid current control in the synchronous reference frame and circulating current suppression control, thereby ensuring stable converter operation. The complete three-phase MMC-BESS, composed of 18 submodules per arm, was modeled and validated using PSCAD/EMTDC simulations. The performance of the proposed balancing algorithm was evaluated under multiple operating conditions, including active power injection, active power absorption, and reactive power support. Simulation results demonstrate effective SoC equalization across all submodules while maintaining high-quality AC voltage and current waveforms and while ensuring suppression of internal circulating currents. The simulation results confirm the robustness, scalability, and practical applicability of the proposed control strategy, highlighting its potential for improving reliability, extending battery lifetime, and enabling MMC-based BESS to provide ancillary services and support large-scale integration of renewable energy sources.
Juan A. Garay, Pedro M. Almeida, P. G. Barbosa· Eletrônica de Potência· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduAug 18, 2026
A new method for surgically removing training examples from a model reveals that as datasets grow, the link between what a model learns and what it produces dissolves.
