Fig 1.
The struct of small watersheds’ external relationship KG and entity of WCE’s internal information KG by LLM.
Table 1.
Inference rules and explanations of the watershed-internal KG.
Fig 2.
Overall scheme of multi-source heterogeneous data processing.
Fig 3.
The forecasting model based on LSTM with LLM and watershed’s KG.
Table 2.
LSTM model’s equations.
Table 3.
Parameter’s interpretation.
Table 4.
Collection of data on the impact factors of mountain floods.
Fig 4.
Rainfall and water level data.
(a) Qixi Reservoir’s forecasting points from February 25th, 2021 to May 25th, 2021. (b) Qiaodongcun’s forecasting points from May 29, 2020 to July 9, 2020.
Fig 5.
The WCE’s partial KG including watershed relationship and internal attribute.
Table 5.
Association point and relationship.
Table 6.
Model’s measure methods.
Table 7.
The models’ training parameters.
Fig 6.
Forecasting method’s training and validation phase of Qixi Reservoir forecasting point.
(a) LSTM’s forecast curve by using watershed-internal KG and LLM. (b) RNN’s forecast curve by using watershed-internal KG and LLM. (c) GRU’s forecast curve by using watershed-internal KG and LLM.
Fig 7.
Forecasting method’s training and validation phase of Qiaodongcun forecasting point.
(a) LSTM’s forecast curve by using watershed-internal KG and LLM. (b) RNN’s forecast curve by using watershed-internal KG and LLM. (c) GRU’s forecast curve by using watershed-internal KG and LLM.
Table 8.
Comparison of key indicators for model forecasting.
Table 9.
Characteristics comparison of structure of forecasting.