Volume 40: Energy Transitions toward Carbon Neutrality: Part III

A dynamic prediction method for the outlet fluid temperature of the large-scale borehole thermal energy storage system based on the multi-channel parallel neural network model Pengchao Li, Fang Guo, Xuejing Yang, Xudong Yang

Abstract

Borehole thermal energy storage (BTES) is a technology in which the thermal energy generated during non-heating seasons may be collected and stored in the soil for extraction in the heating season. However, the average soil temperature of BTES continues to decay with the heat extraction process, resulting in a serious mismatch between the heat extraction and the actual heat load. The variable flow operation of the BTES system allows for flexible adjustment of the heat extraction, increasing the heat flexibility of the BTES system. However, traditional heat transfer models of the BTES cannot quickly and accurately predict the outlet fluid temperature dynamically, making it difficult to match real-time heat load requirements through online regulation of the BTES system. The paper proposes a dynamic prediction method for outlet fluid temperature of the BTES system based on the multi-channel parallel neural network model. To train the neural network model, fluid temperature, flow rate, and multiple sets of soil temperature monitoring results from a large BTES project in Chifeng lasting 11,947 hours were used as the dataset. Randomly divide the dataset into 60% as the training dataset, 20% as the validation dataset, and 20% as the test dataset. The input layer of the basic model contains inlet fluid temperature, flow rate, and multiple sets of soil temperature; the outlet water temperature of the BTES is the output layer. The input features of the advanced model also include the inlet temperature, outlet temperature, and flow rate of the previous moment (hour). After training, the variance of the prediction error for the outlet temperature of the basic model and the advanced model is 0.93 (℃)2 and 0.27 (℃)2, respectively. The advanced model can rapidly and accurately predict the outlet temperature of the BTES, which implies that by continuously iterating with the model, the optimal flow rate can be found to match heat extraction with the real-time heat load.
The influence of changing the heat extraction flow rate of the BTES was also evaluated. The heat extraction of the BTES system increases rapidly with the heat extraction flow rate and then levels off, which emphasizes the importance of variable flow operation for the flexible operation of BTES systems.

Keywords borehole thermal energy storage, multi-channel parallel neural network model, outlet fluid temperature, heat extraction, dynamic heat load matching