JP6086875B2 - Power generation amount prediction device and power generation amount prediction method - Google Patents

Description

  The present invention relates to a power generation amount prediction apparatus and a power generation amount prediction method.

In an energy management system (EMS), it is important to accurately predict the amount of power generated and the demand for power in order to operate the components efficiently.
When predicting the amount of power generated by natural energy at the time of such prediction, for example, a method of predicting the amount of power generated using, for example, weather prediction data provided by the Patent Office or the like is generally used. For example, when predicting the amount of power generated by a wind power generator, it is common to use weather prediction data such as wind speed and direction.

Patent Document 1 describes that process characteristics are predicted using multiple regression analysis and a neural network.
In the process characteristic prediction method described in Patent Document 1, a state quantity and an output quantity are sampled at predetermined time intervals as variables representing the process state. Then, the multiple regression analysis is performed by the multiple regression analysis unit using the sampling values from each sampling time point to the past predetermined period, thereby obtaining a partial regression coefficient. Then, by classifying the partial regression coefficient patterns with a neural network, the variation pattern of the output value of the process is predicted.

  With respect to this method, in Patent Document 1, the result of multiple regression analysis based on the data of the past predetermined period of each variable is not directly input to the neural network for each variable representing the state of the process. Since the value obtained by the above is input to the neural network, the amount of input data to the neural network is greatly reduced while handling the same data. By performing multiple regression analysis on the input data to the neural network as a preprocessing, the scale of the neural network can be greatly reduced, and the time required for internal processing of the neural network can be greatly reduced. It is said that it can be done.

JP-A-5-204407

In the method of predicting the amount of power generated by natural energy by receiving the distribution of weather prediction data, the prediction accuracy of the power generation amount depends on the update frequency of the weather prediction data. The distribution of weather forecast data is generally updated at a certain time interval, for example, every 3 hours or every 6 hours. This time interval may reduce the power generation prediction accuracy. In order to make more detailed predictions, it is conceivable to process the distribution data using a weather model, but analysis using the weather model requires a large-scale parallel computer, etc., which increases equipment costs and operational costs. End up.
Moreover, although patent document 1 describes combining a neural network and multiple regression analysis, a specific technique is not described. In particular, Patent Document 1 does not describe a specific method for predicting in more detail the amount of power generated by natural energy.

  The present invention provides a power generation amount prediction apparatus and a power generation amount prediction method capable of predicting in more detail the amount of power generated by natural energy without requiring a large-scale facility such as a large-scale parallel computer.

A power generation amount prediction apparatus according to a first aspect of the present invention is a power generation amount prediction apparatus that predicts a power generation amount of a power generation device that generates power using natural energy, obtains a weather measurement value at a predetermined time, and A smoothing unit that smoothes the change in the predetermined time, and the weather measurement value in the predetermined time , both the data before smoothing by the smoothing unit and the data after smoothing are input values to the neural network model And learning data for providing the neural network model with learning data in which the power generation amount actual value of the power generation device corresponding to the input value is an output value from the neural network model , and learning processing of the relationship between the input value and the output value A neural network processing unit for performing the measurement, the meteorological measurement value at the most recent predetermined time, and the change at the predetermined time. A power generation amount prediction unit that inputs an input value indicating the smoothed actual weather measurement value to the neural network model subjected to the learning process and predicts a power generation amount of the power generation device at a prediction target time; Are provided.

  A plurality of neural network models having different initial states, wherein the neural network processing unit performs the learning process on each of the plurality of neural network models, and the power generation amount prediction unit includes the plurality of neural network models. The power generation amount predicted value of the power generation device may be obtained from the output value.

A power generation amount prediction method according to a second aspect of the present invention is a power generation amount prediction method of a power generation amount prediction device that predicts the power generation amount of a power generation device that generates power using natural energy, and obtains a measured weather value at a predetermined time. Smoothing the meteorological measurement value with respect to the change at the predetermined time; and for the meteorological measurement value at the predetermined time , both the data before smoothing at the smoothing step and the data after smoothing are neuralized An input value and an output value are given to the neural network model as learning values that are input values to the network model, and the actual power generation amount value of the power generation device corresponding to the input value is an output value from the neural network model. A neural network processing step for performing a learning process of the relationship between And an input value indicating the actually measured weather value smoothed with respect to the change in the predetermined time is input to the neural network model subjected to the learning process, and the power generation amount of the power generation device at the prediction target time is calculated. And a power generation amount prediction step for predicting.

  According to the power generation amount prediction apparatus and the power generation amount prediction method described above, it is possible to predict the power generation amount due to natural energy in more detail without requiring a large-scale facility such as a large-scale parallel computer.

It is a schematic block diagram which shows the function structure of the electric power generation amount prediction apparatus in one Embodiment of this invention. It is explanatory drawing which shows the outline | summary of the process in which the electric power generation amount prediction part in the embodiment acquires an electric power generation amount prediction value. It is a graph which shows the example of the input data which the electric power generation amount prediction part in the embodiment uses. It is explanatory drawing which shows the example of the simulation result of electric power generation amount predicted value. It is explanatory drawing which shows the example of the several neural network model from which the initial state which the neural network model part in the same embodiment implement | achieves differs.

Hereinafter, the present invention will be described through embodiments of the invention, but the following embodiments do not limit the invention according to the claims. In addition, not all the combinations of features described in the embodiments are essential for the solving means of the invention.
Moreover, although the case where the electric power generation amount of a wind power generator is estimated below is demonstrated to an example, the application range of this embodiment is not restricted to the electric power generation amount prediction of a wind power generator. For example, the wind power generator according to the present embodiment can be used for predicting the power generation amount of various natural energies whose power generation amount can change depending on weather conditions such as solar power generation and wave power generation.

FIG. 1 is a schematic block diagram illustrating a functional configuration of a power generation amount prediction apparatus according to an embodiment of the present invention. In the figure, the power generation amount prediction apparatus 100 includes a communication unit 110, a display unit 120, a storage unit 180, and a control unit 190. The control unit 190 includes a smoothing unit 191, a neural network model unit 192, a neural network processing unit 193, and a power generation amount prediction unit 194.
The communication unit 110 is connected to the wind power generator 910 and the wind direction anemometer 920 via the communication network 930. The power generation amount prediction device 100, the wind power generation device 910, and the wind direction anemometer 920 constitute the wind power generation system 1.

The wind power generator 910 receives power to generate power. The wind power generator 910 corresponds to an example of a power generator.
The wind direction anemometer 920 is provided in the vicinity of the wind power generator and measures the wind force and the wind direction of the wind used for power generation.
The communication network 930 mediates communication between the wind power generator 910 and the communication unit 110 and communication between the wind direction anemometer 920 and the communication unit 110. Various communication networks can be used as the communication network 930. For example, the Internet may be used as the communication network 930, or the communication network 930 may be configured with a dedicated line.

  The power generation amount prediction device 100 predicts the power generation amount of the wind power generation device 910. The generated power here may be generated power, or may be generated power per predetermined unit time (for example, one hour). The power generation amount prediction apparatus 100 includes an information processing apparatus such as a personal computer (PC).

  The communication unit 110 communicates with the wind power generator 910 and the wind direction anemometer 920 via the communication network 930. In particular, the communication unit 110 periodically obtains (receives) measured values of wind force and wind direction from the wind direction anemometer 920, for example, every 10 minutes. The measured values of wind force and wind direction acquired by the communication unit 110 correspond to examples of measured weather values. In addition, the communication unit 110 periodically acquires the actual power generation amount value from the wind power generator 910, for example, every 10 minutes.

  The memory | storage part 180 is implement | achieved using the memory | storage device with which the electric power generation amount prediction apparatus 100 comprises, and memorize | stores various information. In particular, the storage unit 180 stores a history of measured values of wind force and wind direction acquired by the communication unit 110 from the anemometer 920 and a history of actual power generation values acquired by the communication unit 110 from the wind power generator 910. In addition, the storage unit 180 stores data obtained by the smoothing unit 191 performing multiple regression analysis for a predetermined time (for example, 3 hours) in the history of measured values of wind force and wind direction.

  The control unit 190 executes various functions by controlling each unit of the power generation amount prediction apparatus 100. The control unit 190 is realized, for example, when a CPU (Central Processing Unit) included in the power generation amount prediction apparatus 100 reads out and executes a program from the storage unit 180.

The smoothing unit 191 acquires a weather measurement value at a predetermined time, and smoothes the obtained weather measurement value with respect to a change at the predetermined time. For example, the smoothing unit 191 smoothes the time series data of the actual measurement values by performing multiple regression analysis for the latest three hours of the wind direction and the actual wind power values stored in the storage unit 180.
Hereinafter, data obtained by the smoothing unit 191 performing multiple regression analysis is referred to as “smoothed data”.

The neural network model unit 192 implements a neural network model.
The method of realizing the neural network model by the neural network model unit 192 may be a hardware method or a software method. That is, the neural network model unit 192 may include a neural network configured by hardware. Alternatively, the neural network model unit 192 may perform calculation (simulation) of the neural network with software.

The neural network processing unit 193 includes an input value indicating a measured weather value and a smoothed measured weather value at a predetermined time, an output value indicating an actual power generation amount value of the wind turbine generator 910 corresponding to the input value, and a neural network. A learning process of the relationship between the input value and the output value is performed using the neural network model realized by the model unit 192.
For example, the neural network processing unit 193 includes data for 3 hours among the wind direction and the measured value of wind force stored in the storage unit 180, and smoothed data obtained by performing the multiple regression analysis on the data by the smoothing unit 191. Are used as input data to the neural network model. In addition, the neural network processing unit 193 uses the power generation amount actual value after a predetermined time from the end time of the input data (for example, 1 hour after the end time of the 3 hours) as correct data. The neural network processing unit 193 gives the input data and the correct answer data to the neural network model to learn the weighting coefficient in the neural network.

The power generation amount prediction unit 194 predicts the power generation amount of the power generation device at the prediction target time.
More specifically, the power generation amount prediction unit 194 inputs input data to the neural network model (the neural network model of the neural network model unit 192) that has undergone learning processing by the neural network processing unit 193, and the neural network model Get the output value from. At that time, the power generation amount predicting unit 194 calculates the meteorological measurement value at the most recent predetermined time (that is, the meteorological measurement value from the predetermined time to the present) and the meteorological measurement value smoothed with respect to the change at the predetermined time. Is used as input data.
Then, the power generation amount prediction unit 194 outputs the output value from the neural network model to the display unit 120 as the power generation amount prediction value at the prediction target time (for example, after one hour in the above learning example).

The display unit 120 has a display screen such as a liquid crystal panel, for example, and displays various types of information. In particular, the display unit 120 displays the power generation amount predicted value output from the power generation amount prediction unit 194.
However, the method by which the power generation amount prediction apparatus 100 outputs the power generation amount prediction value is not limited to the method that the display unit 120 displays. For example, the communication unit 110 may transmit the power generation amount prediction value to another device.

FIG. 2 is an explanatory diagram illustrating an outline of processing in which the power generation amount prediction unit 194 obtains a power generation amount prediction value.
In the processing shown in the figure, the power generation amount prediction unit 194 obtains the history of wind direction measurement values (for example, the latest three hours) stored in the storage unit 180 and inputs it to the neural network of the neural network model unit 192. . The power generation amount prediction unit 194 inputs data obtained by smoothing the wind direction measurement value history by the smoothing unit 191 to the neural network of the neural network model unit 192.
Further, the power generation amount prediction unit 194 acquires the history of wind force measurement values stored in the storage unit 180 (for example, the latest three hours) and inputs it to the neural network of the neural network model unit 192. In addition, the power generation amount prediction unit 194 inputs data obtained by smoothing the history of the wind force measurement values by the smoothing unit 191 to the neural network of the neural network model unit 192.

Then, the power generation amount prediction unit 194 acquires an output value from the neural network as a power generation amount prediction value.
FIG. 2 shows a smoothing unit code 191 and a neural network model unit code 192 for the multiple regression analysis and the neural network, respectively.

  Note that the neural network model realized by the neural network model unit 192 is not limited to the three-layer perceptron model shown in FIG. 2, and can be various neural network models. For example, the number of layers in the neural network model realized by the neural network model unit 192 may be two layers or four or more layers. Further, the neural network model unit 192 may realize a feedback type neural network model.

In addition, in the example of FIG. 2, one input node is shown for each data type for easy viewing of the figure, but the neural network model realized by the neural network model unit 192 has input nodes corresponding to the number of data. Have.
Specifically, the neural network model unit 192 includes an input node for each sampling time in the wind direction actual value history. The same applies to the multiple regression analysis result of the wind direction measurement value history, the wind speed measurement value history, and the multiple regression analysis result of the wind speed measurement value history.
For example, a random value is used as the initial value of the weighting coefficient of the neural network. Alternatively, the neural network processing unit 193 may set an initial value input by the user.

FIG. 3 is a graph illustrating an example of input data used by the power generation amount prediction unit 194. In the figure, the horizontal axis indicates time, and the vertical axis indicates wind speed.
In the figure, a line L11 indicates the actual wind speed value. A line L12 indicates data obtained by performing multiple regression analysis on the measured wind speed from time T11 to time T12 by the smoothing unit 191. A point P11 indicates an output from the neural network of the neural network model unit 192.
In FIG. 3, the history of the wind speed actual measurement value and the multiple regression analysis data of the history of the wind speed actual measurement value among the input data used by the power generation amount prediction unit 194 are shown. The same applies to the history of the wind direction actual measurement value and the multiple diffraction analysis data of the history of the wind direction actual measurement value.

At time T12, the power generation amount prediction unit 194 includes a smoothing unit 191 that overlaps the measured wind direction value and the measured wind speed value from time T11 to time T12 (for example, for three hours), the measured wind direction value, and the measured wind speed value. The data subjected to the regression analysis is input to the neural network model unit 192.
Then, the power generation amount prediction unit 194 acquires the output from the neural network indicated by the point P11 from the neural network model unit 192. The output indicates a predicted power generation amount of the wind turbine generator 910 at time T13, which is a future time (for example, one hour later) at time T12.

The neural network processing unit 193 also smoothes the wind direction actual measurement value and the wind speed actual measurement value for the same amount of time as the power generation amount prediction unit 194 (for example, three hours), and the wind direction actual measurement value and the wind speed actual measurement value, respectively. Is obtained as input data to the neural network model unit 192. Furthermore, the neural network processing unit 193 generates a power generation amount actual value at a time interval similar to the power generation amount predicted value acquired by the power generation amount prediction unit 194 (for example, power generation one hour after the final time of actual measurement data used as input data). (Actual amount value) is acquired as correct answer data.
Then, the neural network processing unit 193 outputs the obtained input data and correct answer data to the neural network model unit 192 to learn the weighting coefficient in the neural network.

FIG. 4 is an explanatory diagram illustrating an example of the simulation result of the predicted power generation amount. In the figure, the vertical axis indicates the elapsed time from the reference time, and the vertical axis indicates the power generation amount.
A line L21 indicates the actual power generation amount. A line L22 indicates a power generation amount predicted value obtained by a process combining multiple regression analysis and a neural network as in the power generation amount prediction apparatus 100. A line L23 indicates a power generation amount prediction value obtained using a neural network without performing multiple regression analysis.

In the figure, the power generation amount predicted value (line L22) when the multiple regression analysis and the neural network are combined is more than the power generation amount predicted value (line L23) when the neural network is used without performing the multiple regression analysis. Is close to the actual power generation amount (line L21). Specifically, the mean square deviation of the error (difference from the measured power generation value) when combining multiple regression analysis and a neural network is the error when using a neural network without performing multiple regression analysis. The value was smaller than the mean square deviation.
Further, when the multiple regression analysis and the neural network are combined, a smooth predicted value is obtained as compared with the case where the neural network is used without performing the multiple regression analysis.

  As described above, the smoothing unit 191 acquires a weather measurement value at a predetermined time, and smoothes the obtained weather measurement value with respect to a change at the predetermined time. In addition, the neural network processing unit 193 uses an input value indicating a measured weather value and a smoothed measured weather value at a predetermined time, and an output value indicating an actual power generation amount value of the power generator corresponding to the input value. Thus, the neural network model realized by the neural network model unit 192 is made to learn the relationship between the input value and the output value. Then, the power generation amount prediction unit 194 uses the input value indicating the measured weather value at the latest predetermined time and the measured weather value smoothed with respect to the change at the predetermined time as the neural network model subjected to the learning process. The power generation amount of the wind power generator 910 at the prediction target time is predicted.

Thereby, the power generation amount prediction apparatus 100 can predict the power generation amount due to natural energy in more detail without requiring large-scale facilities such as a large-scale parallel computer.
Specifically, the power generation amount prediction apparatus 100 acquires weather measurement data at a time interval shorter than the update time interval (for example, every 3 hours) of the weather prediction data, and based on the obtained data, the weather prediction data It is possible to obtain a predicted power generation amount for each hour that is more detailed than the update time interval.
In addition, since the power generation amount prediction device 100 uses the actually measured weather value instead of the weather prediction data, it is not necessary to implement a process according to the format of the weather prediction data when the power generation amount prediction device 100 is designed or manufactured. . In this respect, the development cost of the power generation amount prediction apparatus 100 can be reduced.

In addition, the power generation amount prediction apparatus 100 uses the actually measured weather value smoothed by the smoothing unit 191 as an input to the neural network. As a result, the power generation amount prediction apparatus 100 reduces the influence of short-term fluctuations in the actually measured weather value and the influence of noise included in the actually measured weather value, and thus predicts the power generation amount with higher accuracy than when smoothing is not performed. Get the value.
In particular, since the smoothing unit 191 performs the smoothing process, the power generation amount prediction apparatus 100 acquires data indicating the trend (change rate) of the actually measured weather value, and based on the data indicating the trend, the accuracy is higher. The predicted power generation value can be obtained.

In addition, the power generation amount prediction unit 194 does not use all past weather measurement values, but performs power generation amount prediction using data of the latest predetermined time (for example, 3 hours), so that the power generation amount prediction processing time is It is relatively short.
Further, it is conceivable that the tendency of the wind direction and the wind speed changes in about 1 to 2 hours. The power generation amount prediction unit 194 can predict the power generation amount with higher accuracy by performing the power generation amount prediction using the weather measured values for the latest three hours.

Note that the output of the neural network depends on the initial value of the weighting coefficient. Therefore, the neural network model unit 192 may acquire a power generation amount predicted value based on a more appropriate initial value by realizing a plurality of neural network models having different initial states.
In this case, the neural network processing unit 193 performs a learning process on each of the plurality of neural network models. Further, the power generation amount prediction unit 194 obtains a power generation amount prediction value of the wind turbine generator 910 from the output values of the plurality of neural network models.

FIG. 5 is an explanatory diagram showing an example of a plurality of neural network models with different initial states realized by the neural network model unit 192.
In the example shown in the figure, the neural network model unit 192 implements three neural network models, and different weighting coefficient initial values are set for each neural network. Note that the number of neural network models realized by the neural network model unit 192 is not limited to three, and may be two or four or more.

The neural network processing unit 193 performs the learning process described above for each of these neural networks.
The power generation amount prediction unit 194 provides input data to each of these neural networks in the same manner as described above. Then, the power generation amount prediction unit 194 calculates a power generation amount prediction value of the wind power generator 910 based on the output from each of the neural networks.

For example, the power generation amount prediction unit 194 performs a multiple regression analysis on the output from each of the neural networks, and acquires the analysis result as a power generation amount prediction value. Further, for example, the power generation amount prediction unit 194 excludes, as abnormal data, data that is more than 3σ (σ indicates a standard deviation) from the average value among outputs from each of the neural networks. Then, the power generation amount prediction unit 194 acquires the average of the remaining prediction results as the power generation amount prediction value.
Alternatively, the power generation amount prediction unit 194 may select one of each of the neural networks and obtain it as a power generation amount prediction value.

As described above, the neural network model unit 192 includes (realizes) a plurality of neural network models having different initial states. Then, the neural network processing unit 193 performs a learning process on each of the plurality of neural network models. Then, the power generation amount prediction unit 194 obtains a power generation amount prediction value of the wind turbine generator 910 from the output values of the plurality of neural network models.
As a result, the power generation amount prediction apparatus 100 can reduce the influence of the weighting coefficient initial value on the output of the neural network. In this respect, the power generation amount prediction apparatus 100 can obtain the power generation amount prediction value with higher accuracy.

Note that the power generation amount prediction apparatus 100 may perform the power generation amount prediction of the solar power generation facility.
Here, in the prediction of the amount of power generated by conventional photovoltaic power generation facilities, information such as weather forecast data (sunny, cloudy, rainy weather, etc.) is used to estimate how much sunlight reaches the photovoltaic power generation panel. , Predict power generation. However, such weather data is vague and it is difficult to predict the actual amount of sunlight.

  Therefore, the power generation amount prediction apparatus 100 predicts the power generation amount based on, for example, the measured temperature value of the photovoltaic power generation panel instead of the weather forecast data. Also in this case, similarly to the above, the smoothing unit 191 smoothes the temperature measurement value, and the neural network processing unit 193 and the power generation amount prediction unit 194 use the temperature measurement value and the smoothed temperature measurement value as the neural network. Used as input to

In this way, the power generation amount prediction apparatus 100 improves the prediction accuracy of the power generation amount by using the measured temperature value of the photovoltaic power generation panel that can be quantitatively evaluated instead of the weather forecast data that is difficult to quantitatively evaluate. There is expected.
Further, the power generation amount prediction apparatus 100 is expected to further improve the power generation amount prediction accuracy in that not only the observed data itself but also the smoothed change rate of the data can be taken into consideration.

Also in this case, similarly to the above, the power generation amount prediction apparatus 100 can predict the power generation amount by natural energy in more detail without requiring large-scale facilities such as a large-scale parallel computer.
In addition, since the power generation amount prediction device 100 uses the actually measured weather value instead of the weather prediction data, it is not necessary to implement a process according to the format of the weather prediction data when the power generation amount prediction device 100 is designed or manufactured. . In this respect, the development cost of the power generation amount prediction apparatus 100 can be reduced.

In addition, the power generation amount prediction apparatus 100 uses the actually measured weather value smoothed by the smoothing unit 191 as an input to the neural network. As a result, the power generation amount prediction apparatus 100 reduces the influence of short-term fluctuations in the actually measured weather value and the influence of noise included in the actually measured weather value, and thus predicts the power generation amount with higher accuracy than when smoothing is not performed. Get the value.
In particular, since the smoothing unit 191 performs the smoothing process, the power generation amount prediction apparatus 100 acquires data indicating the trend (change rate) of the actually measured weather value, and based on the data indicating the trend, the accuracy is higher. The predicted power generation value can be obtained.

In addition, the power generation amount prediction unit 194 does not use all past weather measurement values, but performs power generation amount prediction using data of the latest predetermined time (for example, 3 hours), so that the power generation amount prediction processing time is It is relatively short.
As described above, the neural network model unit 192 may acquire a power generation amount prediction value based on a more appropriate initial value by realizing a plurality of neural network models having different initial states.

Note that a program for realizing all or part of the functions of the power generation amount prediction apparatus 100 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed. You may process each part by. Here, the “computer system” includes an OS and hardware such as peripheral devices.
Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.
The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. The program may be a program for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in a computer system.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes design changes and the like without departing from the gist of the present invention.

DESCRIPTION OF SYMBOLS 1 Wind power generation system 100 Power generation amount prediction apparatus 110 Communication part 120 Display part 180 Storage part 190 Control part 191 Smoothing part 192 Neural network model part 193 Neural network processing part 194 Power generation amount prediction part 910 Wind power generation apparatus 920 Wind direction anemometer 930 Communication network

Claims (3)

A power generation amount prediction device that predicts the power generation amount of a power generation device that generates power using natural energy,
A smoothing unit that obtains a weather measurement value at a predetermined time and smoothes the weather measurement value with respect to a change at the predetermined time;
With respect to the actually measured weather value at the predetermined time , both the data before smoothing by the smoothing unit and the data after smoothing are input values to the neural network model, and the power generation amount results of the power generation device corresponding to the input values A neural network processing unit that performs learning processing of a relationship between an input value and an output value by giving the neural network model learning data whose value is an output value from the neural network model;
An input value indicating the actual weather measurement value at the most recent predetermined time and the actual weather measurement value smoothed with respect to the change at the predetermined time is input to the neural network model subjected to the learning process, and predicted A power generation amount prediction unit that predicts a power generation amount of the power generation device at a target time;
A power generation amount prediction apparatus comprising: It has a plurality of neural network models with different initial states,
The neural network processing unit performs the learning process on each of the plurality of neural network models,
The power generation amount prediction unit obtains a power generation amount prediction value of the power generation device from output values of the plurality of neural network models.
The power generation amount prediction apparatus according to claim 1. A power generation amount prediction method of a power generation amount prediction device that predicts a power generation amount of a power generation device that generates power with natural energy,
Obtaining a meteorological measurement value at a predetermined time, and smoothing the meteorological measurement value with respect to a change at the predetermined time;
For the measured weather value at the predetermined time , both the data before smoothing in the smoothing step and the data after smoothing are input values to the neural network model, and the power generation amount of the power generator corresponding to the input values giving learning data for the actual value and the output value from the neural network model in the neural network model, the neural network processing step of performing a learning process of the relationship between the input values and output values,
An input value indicating the actual weather measurement value at the most recent predetermined time and the actual weather measurement value smoothed with respect to the change at the predetermined time is input to the neural network model subjected to the learning process, and predicted A power generation amount prediction step for predicting a power generation amount of the power generation device at a target time; and
A method for predicting the amount of power generation.

Images