Abstract: Abstract: Combining similarity search with neural network, a method of predicting water level was proposed. We calculate similarity distance of water level series between fifteen days before predicting day and months of forty-nine years before predicting year whose hydrological characteristics was similar to predicting months, then seek the most similar water level series of each year. Lastly, the most similar series of forty-nine years and the water level after the last day of them were used as training set of BP neural network. The experimental results show that predictions of the method combining similarity search with BP neural network are more accurate and their fluctuations are within the scope permitted by state.

Key words: hydroinformatics, similarity search, BP neural network, water level prediction