Abstract: Accurate, reliable and efficient prediction of key variables reflecting the working behaviour of dams plays a vital role in safe construction and operation of dams. To predict and analyse the temperature field of a dam, this paper develops a parallel-over-series fusion model driven by both a mechanism model and a data model, and examines its feasibility by using the internal temperature of the Xiluodu arch dam. In series fusion, the camp chain method and the auxiliary technology of principal component analysis are adopted for back-calculating the thermal conductivities of dam concrete, by which optimised outputs of the mechanism model are obtained. In parallel fusion, the explanatory factor method and the linear-Gaussian model are used to construct a prediction model that enables the optimised mechanism model to be mapped to the dam temperature data. Experimental verification shows this fusion model, benefits from the explanatory power of the mechanism-driven model and the predictive advantage of the linear-Gaussian model. It can predict data fluctuations and has advantages in medium- and long-term forecasting. It reduces the overall prediction error by 81% compared with the series fusion model, while the latter reduces the error by 13% relative to the mechanism model.

Key words: high concrete dam, fusion model, temperature prediction, explanatory factor method, linear-Gaussian model, inverse analysis