Machine learning-based performance prediction for ground source heat pump systems

Abstract

The coefficient of performance (COP) prediction of heat pump units and ground source heat pump (GSHP) systems are required for effective evaluation, optimization, and fault diagnosis of GSHP systems. COP prediction models of the heat pump unit and the GSHP system are developed in this work using machine learning methods, including extreme learning machine (ELM), support vector machine (SVM), and the back propagation neural network (BPNN). Ten different operation parameters are used as feature variables of the model, and the correlation analysis method and least absolute shrinkage and selection operator (LASSO) approach are used to optimize the feature variables. The prediction performance of ELM models with different hidden layer nodes are then analyzed and compared. Results show that the prediction accuracy of the ELM model is better than other models. Five feature sets of variables are obtained, where the COP of the heat pump unit prediction models established using feature set 2 are better, and the optimal number of hidden layer nodes is 9 for the ELM model. Additionally, the COP of the GSHP system prediction models established using feature set 2 are better, and the optimal number of hidden layer nodes is 14 for the ELM model.