Performance Prediction Model Based on Multi-Task Learning and Co-Evolutionary Strategy for Ground Source Heat Pump System

In order to effectively predict the performance of ground source heat pump system, a performance prediction method is proposed in this paper. Based on the basic model of forward neural network, the algorithm predicts the performance data of ground source heat pump system by inputting the time series of system performance and 12 variables including 7 drilling parameters, 2 u-pipe parameters, 2 ground parameters and 1 circulating liquid parameter. The training of the model is divided into three subtasks by the strategy of multi-task learning and co-evolution, where CMA-ES is used as the evolutionary algorithm of the subtask. The experimental results show that the RMSE of the predicted results obtained by the proposed algorithm is less than 0.2, which verifies the effectiveness of the method. At the same time, this algorithm fully considers various influencing factors and has good versatility, which can be used as a reference for the design of ground source heat pump system.