Data-Driven Based Dynamic State Estimation Method for Regional Integrated Energy Systems Incorporating Multi-Dimensional Generation-Grid-Load Characteristics

The regional integrated energy system (RIES) has emerged as a critical focus in energy systems research. The comprehensive incorporation of renewable energy and inherent multi-energy flow interconnection within RIES markedly elevates the complexity of “generation-load” balance regulation. Traditional model-driven dynamic state estimation methods, however, are constrained by fundamental limitations—complex modeling, inadequate representation of multi-energy flow interdependencies, and poor computational efficiency. This study proposes a data-driven dynamic state estimation method for RIES, utilizing multi-dimensional “generation-grid-load” characteristic information as its primary input and employing a synergistic framework of Empirical Mode Decomposition-Singular Value Decomposition (EMD-SVD) alongside an enhanced Bidirectional Long Short-Term Memory (BiLSTM) network. EMD-SVD preprocesses raw data to remove noise and extract essential features, while the enhanced BiLSTM serves a dual purpose: it first attains high-precision photovoltaic output prediction and multi-energy load forecasting and subsequently evaluates the node states of the multi-energy flow coupling system. A case study on a practical coupled RIES, comprising a 33-node power system, 7-node gas system, and 6-node thermal system, demonstrates that the proposed method achieves high estimation accuracy and remarkable computational efficiency while effectively addressing the inherent limitations of conventional model-driven approaches.