Time-series data prediction device of observation value, time-series data prediction method of observation value, and program

1. A time-series data prediction device, including a computer processing unit, for predicting power demand by calculating a predicted value of an observation value based on time-series data of an observation value, comprising:

an acquisition unit that acquires a plurality of observation values from an energy demand notification device that continue at predetermined time intervals, as a prediction data, from time-series data of an observation value of a predetermined observation target and that acquires the plurality of observation values that continue at the predetermined time intervals, as a test data, from the time-series data excluding the prediction data and that acquires a training data from the time-series data excluding the prediction data and the test data, wherein the energy demand notification device acquires the plurality of observation values of energy demand from a sensor attached to a power distribution board;

a prediction model generation unit that generates a prediction model to calculate time-series data, which is an observation value predicted based on the given time-series data, using the training data;

a prediction unit that calculates a predicted value of an observation value using the prediction model generated by the prediction model generation unit and the prediction data acquired by the acquisition unit;

a storage unit that stores a prediction model used in the past; and

an evaluation unit that evaluates a prediction accuracy of the prediction model generated by the prediction model generation unit using the test data, wherein:

the evaluation unit compares at least a predicted value calculated by the prediction model acquired from the storage unit, a predicted value calculated by the prediction model generated by the prediction model generation unit, and a predicted value calculated by an average use prediction model to acquire a predicted value based on an average of time-series data of a predetermined period, using the test data,

the evaluation unit sets a clustering flag and an average flag to off and a prediction model rewrite flag to on when a no-cluster prediction model is selected, the prediction model rewrite flag and the clustering flag to on and the average flag to off when a cluster-specific prediction model is selected, the prediction model rewrite flag, the average flag, and the clustering flag to off when a previous prediction model is selected, and the average flag to on when an average use prediction model is selected, and

the prediction unit calculates a predicted value of an observation value using a prediction model having a best result of the comparison of the evaluation unit and the prediction data acquired by the acquisition unit.

2. The time-series data prediction device of an observation value according to claim 1 ,

wherein the acquisition unit searches for time-series data whose correlation with the prediction data is higher than a predetermined value from the time-series data, and includes the time-series data obtained by a search in the training data.

3. The time-series data prediction device of an observation value according to claim 1 ,

wherein, when the average use prediction model is used in a previous prediction, the acquisition unit includes time-series data used as a prediction data in the previous prediction in the training data.

4. The time-series data prediction device of an observation value according to claim 1 , further comprising:

a deviation amount detection unit that detects a record deviation amount that is a difference between an observation value and the predicted value calculated by the prediction unit using the prediction data,

wherein, when the record deviation amount is smaller than a predetermined value and the prediction model generated by the prediction model generation unit is used to calculate the predicted value, the prediction unit calculates a predicted value of an observation value with the average use prediction model when a difference between the observation value and a predicted value calculated by the average use prediction model using the prediction data used to calculate the predicted value is equal to or less than a predetermined value.

5. The time-series data prediction device of an observation value according to claim 1 , further comprising:

a deviation amount detection unit that detects a record deviation amount that is a difference between an observation value and the predicted value calculated by the prediction unit,

wherein, when the record deviation amount detected by the deviation amount detection unit is larger than a predetermined value, the acquisition unit reselects the training data from the time-series data, and

the prediction model generation unit generates a prediction model using the training data reselected by the acquisition unit.

6. A time-series data prediction method of an observation value executed by a time-series data prediction device, including a computer processing unit, for predicting power demand by calculating a predicted value of an observation value based on time-series data of the observation value, the method comprising:

performing an acquisition process in which an acquisition unit acquires a plurality of observation values from an energy demand notification device that continue at predetermined time intervals, as a prediction data, from time-series data of an observation value of a predetermined observation target and acquires the plurality of observation values that continue at the predetermined time intervals, as a test data, from the time-series data excluding the prediction data and acquires a training data from the time-series data excluding the prediction data and the test data, wherein the energy demand notification device acquires the plurality of observation values of energy demand from a sensor attached to a power distribution board;

performing a prediction model generation process in which a prediction model generation unit generates a prediction model to calculate time-series data, which is an observation value predicted based on the given time-series data, using the training data;

performing a prediction process in which a prediction unit calculates a predicted value of an observation value using the prediction model generated in the prediction model generation process and the prediction data acquired in the acquisition process;

performing a storage process in which a storage unit stores a prediction model used in the past; and

performing an evaluation process in which an evaluation unit evaluates a prediction accuracy of the prediction model generated by the prediction model generation process using the test data, wherein:

the evaluation process compares at least a predicted value calculated by the prediction model acquired from the storage process, a predicted value calculated by the prediction model generated by the prediction model generation process, and a predicted value calculated by an average use prediction model to acquire a predicted value based on an average of time-series data of a predetermined period, using the test data,

the evaluation process sets a clustering flag and an average flag to off and a prediction model rewrite flag to on when a no-cluster prediction model is selected, the prediction model rewrite flag and the clustering flag to on and the average flag to off when a cluster-specific prediction model is selected, the prediction model rewrite flag, the average flag, and the clustering flag to off when a previous prediction model is selected, and the average flag to on when an average use prediction model is selected, and

the prediction process calculates a predicted value of an observation value using a prediction model having a best result of the comparison of the evaluation process and the prediction data acquired by the acquisition process.

7. A program causing a computer, which is used as a time-series data prediction device, including a computer processing unit, for predicting power demand by calculating a predicted value of an observation value based on time-series data of the observation value, to function as:

an acquisition unit that acquires a plurality of observation values from an energy demand notification device that continue at predetermined time intervals, as a prediction data, from time-series data of an observation value of a predetermined observation target and that acquires the plurality of observation values that continue at the predetermined time intervals, as a test data, from the time-series data excluding the prediction data and that acquires a training data from the time-series data excluding the prediction data and the test data, wherein the energy demand notification device acquires the plurality of observation values of energy demand from a sensor attached to a power distribution board;

a prediction model generation unit that generates a prediction model to calculate time-series data, which is an observation value predicted based on the given time-series data, using the training data;

a prediction unit that calculates a predicted value of an observation value using the prediction model generated by the prediction model generation unit and the prediction data acquired by the acquisition unit;

a storage unit that stores a prediction model used in the past; and

an evaluation unit that evaluates a prediction accuracy of the prediction model generated by the prediction model generation unit using the test data, wherein:

the evaluation unit compares at least a predicted value calculated by the prediction model acquired from the storage unit, a predicted value calculated by the prediction model generated by the prediction model generation unit, and a predicted value calculated by an average use prediction model to acquire a predicted value based on an average of time-series data of a predetermined period, using the test data,

the evaluation unit sets a clustering flag and an average flag to off and a prediction model rewrite flag to on when a no-cluster prediction model is selected, the prediction model rewrite flag and the clustering flag to on and the average flag to off when a cluster-specific prediction model is selected, the prediction model rewrite flag, the average flag, and the clustering flag to off when a previous prediction model is selected, and the average flag to on when an average use prediction model is selected, and

the prediction unit calculates a predicted value of an observation value using a prediction model having a best result of the comparison of the evaluation unit and the prediction data acquired by the acquisition unit.