Integrated energy system optimal operation using Data-Driven district heating network model

A feasible measure to improve the volatility of the power grid is using heat storage capacity of a district heating network (DHN) in the optimal operation of an integrated energy system (IES), but producing a high computational complexity problem in solving partial differential equations (PDEs) describing the thermodynamics and heat transfer behavior of DHN system. A data-driven model called Gaussian interpolated spatiotemporal Volterra model is proposed to fit input-output characteristics of the DHN system, in which both the diameter and length parameter of pipes in DHN are selected as operating variables of the weighting functions. The model can be estimated through operation data, and the thermodynamic model and parameters of DHN are no longer required. The schedule of an integrated energy system with a 9-bus power grid and 14-node DHN is investigated. The temperature predictive error of the DHN model is less than 5%. The optimization results demonstrate the superiority of wind power accommodation by adding heat storage capacity as new dispatch variables to the operation schedule of the IES, and the computational burden of optimization is effectively reduced by avoiding the PDEs solving problem of DHN.

Automated summary

The use of heat storage capacity of a district heating network (DHN) can improve the volatility of the power grid, but solving the partial differential equations (PDEs) describing the thermodynamics and heat transfer behavior of DHN system leads to a high computational complexity problem. A data-driven model called Gaussian interpolated spatiotemporal Volterra model is proposed, which can fit the input-output characteristics of the DHN system without requiring the thermodynamic model and parameters of DHN. By adding heat storage capacity as new dispatch variables to the operation schedule of the integrated energy system, the optimization results demonstrate the superiority of wind power accommodation and effectively reduce the computational burden by avoiding the PDEs solving problem of DHN.