| Title | Regional Evaluation and Statistical Modeling of Required Lengths for a Borehole Heat Exchanger Considering Varied Geology in Japan |
|---|---|
| Authors | Yoshitaka SAKATA, Takao KATSURA and Katsunori NAGANO |
| Year | 2020 |
| Conference | World Geothermal Congress |
| Keywords | geologic database, ground thermal conductivity, Heat pump simulation, optimization, gaussian process regression |
| Abstract | Required lengths of borehole heat exchanger for a ground-source heat pump system is dependent on various factors regarding to building, heat pump, geology and other facilities. Especially in Japan, geology and climate conditions are variable among each site along the long North-South land. The purpose of this study is to evaluate the required sizes over the land considering the variable geology and climate conditions, and to construct a statistical model to determine the required size directly from selected parameters without numerical simulation. For this purpose, this study constructed the geologic database, which provides ground thermal conductivity at any location and depth as probability-weighted averages. The probability of each soil/rock type was estimated by indicator kriging. The heating/cooling loads were calculated from climate data in an assumed residence of 120 m2 in total area. The ground-source heat pump system was assumed to be composed of 10kW heat pump and one borehole heat exchanger. The length of a borehole heat exchanger was optimized as the least for satisfying the criteria in terms of fluid temperature and COP. The heat pump simulation was performed in a 10km regular grid of Japan, resulting in its nation-scale map and a dataset for statistical modeling. As a result, the required total lengths were found to vary between 150 m, with differences among adjacent grids arising owing to the sharply varying climatic and geologic conditions. The lengths appeared normally distributed with an average of 98 m. This study compared simple estimation models employing liner regression and Gaussian process regression. The latter was much more effective in providing reasonable and practically useful estimates; however, its relative error of more than 20% was not negligible in several grid points. |