| 摘要 |
<p>PROBLEM TO BE SOLVED: To provide a highly precise water distribution predicting method by selecting and interpolating learning data from result data, learning the connection coefficient of a neural network and generating a neural network superior in generalization ability. SOLUTION: When result data and various setting values are inputted to an input device 10, the inputted values are stored in a working memory 5 through an input control part 6. The respective input values are inputted to a learning data control part 1 and the range of learning data is selected by the date based on result data of the working memory 5 and the various setting values in a condition table, namely, a prediction reference day, a previous prescribed period and a subsequent prescribed period, and result data is outputted to learning data 9. A learning data control part 1 calculates assumed demand water quantity and temperature at every weather division by the weather division and a pseudo data calculation range and against whole data in learning data. Assumed demand water quantity, the temperature and a calendar division, which are obtained, are gathered and are made into one piece of data. Then, it is added to learning data 6.</p> |
