The application discloses a kind of water
power simulation systems and its application based on
physical information neural network.It includes: obtaining target river
terrain characteristics and hydrological boundary
condition monitoring data, constructing space-time coordinate input
data set, and implementing the geometric constraint pre-training of section hydraulic parameter;Water
power simulation core model based on
physical information neural network is constructed, and normalized space-time coordinates are mapped into
water level and flow state variables;Saint-Venant equation set describing one-dimensional unsteady flow dynamics is used as physical priori knowledge, and its
mass conservation equation and
momentum conservation equation are used as residual form as soft constraint embedded model
loss function, and
network parameter training is carried out in combination with multi-objective collaborative optimization mechanism;Directly output high-precision
water level and flow prediction results of any space-time position of target river.The application effectively suppresses the non-physical high-
frequency oscillation generated by pure data-driven model in complex unsteady flow environment by combining physical mechanism constraint with
deep learning, while maintaining high computing efficiency, significantly improves the physical consistency and generalization prediction ability of water
power simulation.