Surrogate modelling for multi-carrier distribution networks

Abstract

The thesis mainly concentrates on exploring the suitability and applicability of surrogate models for a multi-carrier energy system (MCES). The climate crisis is being paid more attention owing to its notable effects on the environment globally. For the sake of facing the challenge, the European Union (EU) has developed different strategies. One of which is the “2030 Climate Target Plan” published by the European Commission in Brussel in September 2020. It attempts to achieve a greenhouse gas emissions (GHG) reduction target by at least 55% by 2030. Based on the plan, a list of ideas and definitions centered around integrated energy systems has been proposed. To assess energy system integration strategy in a Dutch context, a generic model of the energy system is necessary. To this end, a detailed representative Dutch electricity distribution network model integrating renewable energy and the heat distribution network model are developed. The simulation of this integrated energy system model is computationally expensive due to the inter-dependencies between various energy sectors, dynamic operation of components within individual energy domain, etc. To overcome the computational burden of detailed models, different machine/deep learning-based surrogate models are established for the electrical network and heating network of the energy system, respectively. They include linear regression model, linear regression with chain model, linear support vector with chain model, decision tree model, random forest model, k-nearest neighbour model, multilayer perceptron model as well as long short-term memory model. Their performances are compared by speed-up factor (SUF) and root mean square error (RMSE), and the best model is selected. The results show linear regression model and long short-term memory model have the best performances in the respective electrical network and heating network of the established energy system.