| CPC H02J 3/004 (2020.01) [H02J 2203/20 (2020.01)] | 8 Claims |
|
1. A method of predicting power generation using meta-learning, the method comprising:
(a) receiving, by a main server, power generation data and weather data of one or more power plants;
(b) scaling, by the main server, the power generation data using respective capacities of the one or more power plants to generate power generation scaling data;
(c) scaling, by the main server, the weather data using maximum and minimum weather values during a predetermined period to generate weather scaling data;
(d) generating, by the main server, a weather-power generation input sequence using the weather scaling data corresponding to a first past period and a prediction period that are continuous to each other, and the power generation scaling data corresponding to a second past period that is longer than the first past period; and
(e) inputting, by the main server, the weather-power generation input sequence into a power generation prediction model to acquire a power generation output sequence corresponding to the prediction period; and
(f) operating a given power plant, other than the one or more power plants, according to the power generation output sequence acquired in the step (e),
wherein respective element data constituting the weather-power generation input sequence are generated by combining the weather scaling data at a specific time point with the power generation scaling data at a time point prior to the specific time point,
wherein a length of the second past period is equal to a sum of the first past period and the prediction period, and an end point of the second past period is equal to a start point of the prediction period,
wherein the step (e) comprises assigning, by the main server, greater weights to the weather-power generation input sequence as being closer to first element data including weather scaling data at the start point of the prediction period,
wherein the weather data is a plurality of different types of weather data, and
wherein, prior to the step (c), the main server is configured to generate the respective plurality of weather data at the start point of the prediction period as a plurality of reference data, generate a plurality of reference ranges for the respective plurality of weather data with respect to the plurality of reference data, extract all past periods corresponding to the plurality of reference ranges for the plurality of weather data, generate a plurality of maximum past start points for all the past periods for the respective plurality of weather data, calculate an average value of the plurality of maximum past start points, and set the calculated average value as the start point of the predetermined period in the step (c).
|