CFU-StateDiff: A Lightweight Diffusion Model for State Estimation in the Digital Twins of Coal-Fired Power Units

Authors:
Publisher website
Google Scholar
Abstract
To meet the requirements of a new energy power system for accommodation, coal-fired units need to frequently and rapidly change their load, and precise and rapid estimation of their operating state is crucial for ensuring safety and control performance. However, owing to the nonlinear and strongly coupled characteristics of coal-fired units, existing state estimation methods (such as Kalman filters) have difficulty meeting the new requirements of coal-fired units. Therefore, this paper proposes a novel digital twin modeling method based on a lightweight denoising diffusion probabilistic model (CFU-StateDiff) for state estimation in coal-fired units. First, an integration of expert knowledge and a data-driven approach for state variables (IEDA-SV) is proposed to construct a high-quality state estimation dataset. Second, MobileNet’s depthwise separable convolution is adopted to reduce the model’s parameter count, and the Mish function is introduced to retain negative state information under nondesign conditions and ensure the completeness of information across all operating conditions. Third, a joint optimization strategy combining feature guidance and physical constraints is employed, in which expert features are embedded as conditions into the diffusion sampling process and a physical consistency loss function based on the law of conservation of mass is introduced to ensure that the generated state variables comply with the mass conservation relationship of the working medium. Experimental engineering results reveal that CFU-StateDiff balances rapid state estimation with physical consistency, achieving a 4.38% improvement in the latter over the optimal baseline conditional denoising diffusion probabilistic model (CDDPM).
Funding Information
  • Special Fund for Science and Technology Innovation Teams of Shanxi Province (202304051001004)
  • Major Science and Technology Project of Shanxi Province (202201090301013)
  • Major Science and Technology Project of Shanxi Province (202301020101001)
  • Research Project Supported by Shanxi Scholarship Council of China (2024-125)
  • Foundation of Shanxi Key Laboratory Of Advanced Control and Industrial Intelligence (ACII202508)

This publication has 39 references indexed in Scilit: