WATER LEVEL PREDICTION DEVICE AND WATER LEVEL PREDICTION PROGRAM

To provide a prediction device that properly extracts data for training without taking features of a river etc., into consideration when using machine learning to predict a water level of the river etc., (especially, close to a water level peak as an announcement standard for evaluation advisory etc.) with high precision.SOLUTION: A water level prediction device comprises: a training data extraction unit 1 which extracts training data L1 etc., having high correlation with a water level prediction point and a water level at certain time needed to learn the water level prediction point and a water level at the certain time from water level data WL in a predetermined past period without depending upon features of the water level at the water level prediction point; a model construction unit 2 which uses the training data L1 etc., to construct a water prediction model M1 etc.; and a predicted water level data output unit 4 which inputs test data T for water level prediction to the water level prediction model M1 etc. constructed by use of the training data L1, etc., and outputs predicted water level data P.SELECTED DRAWING: Figure 1 【課題】機械学習を用いて、河川等の水位（特に、避難勧告等の発令基準となる水位ピーク付近）を高精度に予測するにあたり、河川等の特徴を考慮することなく、訓練用データを適切に抽出する予測装置を提供する。【解決手段】水位予測地点及びある時刻の水位の学習に必要な、水位予測地点及び当該時刻の水位との相関が高い訓練用データＬ１等を、水位予測地点の水位の特徴に依存しないで、過去の所定期間の水位データＷＬから抽出する訓練用データ抽出部１と、訓練用データＬ１等を用いて、水位予測モデルＭ１等を構築するモデル構築部２と、水位予測のためのテストデータＴを、訓練用データＬ１等を用いて構築された水位予測モデルＭ１等に入力し、予測水位データＰを出力する予測水位データ出力部４と、を備える。【選択図】図１