The multiple energy flow (MEF) model is inherently nonlinear, which challenges the fast steady-state analysis and the efficient optimization of the integrated energy system (IES). This paper presents a hybrid mechanism- and data-driven (dual-driven) modeling approach to generate accurate linear models for MEF calculation. A linear regression model for the electricity–heat system is proposed to represent the underlying linear relationship among electrical, thermal, and hydraulic variables. Partial least squares (PLS) regression is used to deal with data collinearity which may lead to over-fitting. Furthermore, a dual-driven modeling approach is proposed to realize the complementarity between mechanism- and data-driven approaches. In this approach, mechanism analysis screens critical features before data processing and provides a fundamental linear model based on physical mechanism. Data-driven regression generates linear models to replace the complicated nonlinear elements and trains an approximation error model to further improve the overall accuracy. Numerical findings on a basic and a complex electricity–heat system confirmed the dual-driven linear model's accuracy advantage over existing linear models. Overall, this study proposes a dual-driven modeling approach that provides high-precision linear models while avoiding the intractable convergence problem in the nonlinear MEF calculation. It also reveals that linear models driven by data and mechanisms can closely fit the energy flows in the nonlinear electricity–heat system.
- Electricity–heat system
- Integrated energy system
- Linear model
- Multiple energy flow
- Partial least squares regression