Existing time-series data-driven approaches for converter modeling are data-intensive, uninterpretable, and lack out-of-domain extrapolation capability. Recent physics-informed modeling methods combine physics into da...
详细信息
Existing time-series data-driven approaches for converter modeling are data-intensive, uninterpretable, and lack out-of-domain extrapolation capability. Recent physics-informed modeling methods combine physics into data-driven models using loss functions, but they inherently suffer from physical inconsistency, lower modeling accuracy, and require resource-intensive retraining for new case predictions. Consequently, catering for the challenges in current data-driven and physics-informed models, this article proposes a physics-in-architecture recurrent neural network (PA-RNN) for the time-domain modeling of power converters. The proposed PA-RNN consists of a physics-in-architecture core and a data-driven core in parallel. The physics-in-architecture core rigorously integrates circuit physical laws into its customized recurrent neural architecture by leveraging numerical differentiation, while a gated recurrent unit with layer normalization serves as the data-driven core to compensate for converter behaviors not characterized by physics. The PA-RNN modeling process is explained in detail with a design case. As 1-kW hardware and comprehensive algorithm experiments have verified the superiority of PA-RNN. Overall, PA-RNN is explainable and data-light as well as possesses good domain transfer capability to assess out-of-domain scenarios without training. This article envisions to democratize artificial intelligence for the modeling of power electronics systems.
In modulation optimization, power converter modeling is pivotal for performance evaluation. However, mainstream knowledge-based approaches suffer from low accuracy and heavy computation burden, while emerging data-dri...
详细信息
In modulation optimization, power converter modeling is pivotal for performance evaluation. However, mainstream knowledge-based approaches suffer from low accuracy and heavy computation burden, while emerging data-driven methods are data-intensive and opaque black-box models. Even state-of-the-art physics-informed artificial intelligence (AI) is improper for modulation optimization due to resource-intensive retraining for new predictions. Hence, a physics-in-architecture recurrent neural network (PA-RNN), which customizes recurrent neurons to integrate physical laws into the structure, is proposed, tailoring for modulation optimization of power converters. The PA-RNN model reveals diverse circuit insights, exhibiting data-light merit and on-call prediction capability. Modulation optimization via PA-RNN involves two stages. First, PA-RNN constructs converter models in the time domain for direct performance evaluation. Second, an optimization algorithm interacts with the PA-RNN model to minimize current stress while realizing full-range soft switching. Two design cases are presented: first, modeling buck converters;second, optimizing dual-active-bridge converters under a triple phase-shift modulation or a five-degree-of-freedom modulation. Algorithm experiments and 1-kW hardware experiments have comprehensively validated the merits and feasibility of the proposed PA-RNN. Broadly speaking, this article strives to increase the penetration of AI in power electronics.
The emerging gray-box modeling for power converters effectively mitigates model discrepancies seen in traditional physics-based white-box models while offering a data-light, explainable alternative to data-driven blac...
详细信息
The emerging gray-box modeling for power converters effectively mitigates model discrepancies seen in traditional physics-based white-box models while offering a data-light, explainable alternative to data-driven black-box models. However, a significant challenge remains existing gray-box modeling approaches suffer from poor generalization to out-of-domain topologies. This limitation necessitates rebuilding or retraining the model when a new topology is encountered, hindering widespread adoption. Catering for these challenges, this article proposes a generic gray-box modeling approach tailored for the dual-active-bridge (DAB) converter topology family, which is based on a proposed topology transferrable physics-in-architecture mixture density network (T(2)PA-MDN). As the core part, the T(2)PA network retrofits recurrent neurons to embed circuit physics seamlessly via discretized numeric methods, enabling efficient topology transfer. Moreover, a probabilistic mixture density network (MDN) quantifies ambient fluctuations using a mixture of Gaussian distributions, mitigating model discrepancies. The proposed modeling methodology is demonstrated with three topology transfer design cases, in which the model is trained on a nonresonant DAB with merely a five-time series and is easily transferred to resonant, multilevel, and multiport topologies with no extra data or training. Algorithm analysis and 2-kW hardware experiments have verified the feasibility and the superiority of T(2)PA-MDN. This research aims to pioneer a new direction for the future gray-box modeling of power converters, toward generalization across diverse topologies but effectiveness.
暂无评论