Robust State Estimation of Integrated Electric-Heat System Based on Physics-Guided Deep Learning With Maximum Correntropy Criterion

Designing an accurate and robust state estimation (SE) framework for integrated electric-heat systems (IEHSs) in the presence of non-Gaussian noise (or outliers) is a crucial challenge. While conventional deep learning (DL) models have shown promise for SE tasks, their performance can be significantly degraded by noisy measurements. In this article, a novel robust physics-guided DL (PGDL) model, termed Ph-maximum correntropy criterion (MCC)-convolutional neural network (CNN)-long short-term memory (LSTM), is proposed to address this challenge. First, to improve the robustness of the original CNN-LSTM with mean squared error criterion, we integrate the correntropy with high-order moment of error into the CNN framework, forming an innovative robust DL model termed MCC-CNN-LSTM. This model excels in achieving superior SE outcomes by effectively handling complex error distributions inherent in the data. Second, the physical laws governing the IEHS are further incorporated into the MCC-CNN-LSTM model to enhance the estimation accuracy, where the estimated state values obtained by MCC-CNN-LSTM are brought into the IEHS physical model to calculate the estimated value of the measurements. Then the cumulative error between the estimated and the actual measurements, along with the error between estimated and actual state values, is used as a learning criterion to guide the learning trajectory of the MCC-CNN-LSTM model. Finally, extensive numerical simulations are conducted on an IEHS test setup comprising a 33-node power system and a 17-node Barry thermal system, and the results demonstrate that the proposed method can achieve superior estimation accuracy and robustness compared to various existing methods, particularly in the presence of non-Gaussian measurement challenges.