The employment of solar space heating is a significant measure to achieve carbon neutrality. However, the design of the solar space heating system usually leads to the dilemma of a trade-off between the economic benefit and environmental protection, which cannot be solved by single-objective optimization. In this work, in order to design a solar space heating system of a bungalow equipped with radiant floor heating, multi-objective optimization of the solar collector area and the volume of the water thermal energy storage (TES) tank is performed by coupling TRNSYS simulation, machine learning and genetic algorithm (GA). Simulation for the entire winter operation of the solar space heating system is conducted in TRNSYS, and a database with various combinations of the design variables are generated, based on which an artificial neural network (ANN) is constructed to capture the mapping of the design variables and the system performance indicators. After proper training and validation, the ANN model with satisfactory forecasting precision is utilized as the objective function for optimizing the system. GA is employed for searching the solutions of the multi-objective optimizations. This data-driven approach combining ANN and GA is helpful to achieve the optimal solutions. For the solar space heating system, the CO2 emission reduction can benefit most from the enlargement of the solar collector area and the TES tank volume, while the payback period reaches its minimum when both the solar collector area and the TES tank volume are small. By comparing single-objective with multi-objective optimizations, it is found that the single-objective optimization cannot reveal the interactions among the CO2 emission reduction, annualized life cycle cost (ALCC) and payback period (PBP) and that the optimization focusing on a single objective probably leads to apparent defects and inapplicability in reality. Better balance among the CO2 emission reduction, ALCC and PBP can be achieved from the Pareto optimums of the multi-objective optimization rather than the single-objective optimization. According to the TOPSIS evaluation, the most appropriate design scheme is obtained from the multi-objective optimization and has the potential to reduce the carbon emission by 5480.6 kgCO(2)eq/year while the ALCC and payback period are only 33.40 yen /(m(2) year) and 3.70 years, respectively.(c) 2022 Elsevier Ltd. All rights reserved.
