JP2020123198A - Forecast system and forecast method - Google Patents

Abstract

To provide a forecast system and a forecast method that reduce an error between an estimated value and a measured value, thus improving the forecast accuracy.SOLUTION: An electrical power forecast system comprises: a coefficient output unit for outputting a coefficient indicating an error between an estimated value obtained by performing an estimation by a predetermined estimation method based on one or more predetermined parameters and an actual value corresponding to the estimated value; and a forecast unit for inputting a predetermined parameter for a predetermined period in a neural network that has learned based on at least one parameter of the predetermined parameters to output a post-learning coefficent, and calculating a forecast value for a forecast day based on the output post-learning coefficient and an estimated value for the forecast day obtained by performing an estimation by the predetermined estimation method.SELECTED DRAWING: Figure 2

Description

The present invention relates to a prediction system and a prediction method.

For example, there is known a power generation amount prediction system that predicts the power generation amount of the prediction target solar power generation device based on the fluctuation of the power generation amount of the solar power generation device installed around the prediction target solar power generation device. (For example, Patent Document 1).

JP, 2013-51326, A

The power generation amount prediction system disclosed in Patent Document 1 includes a prediction target fluctuation pattern indicating a fluctuation state of power generation amount in a prediction target solar power generation device, and a sun installed at a point around the prediction target solar power generation device. The future power generation amount in the prediction target solar power generation device is predicted by collating with the peripheral fluctuation pattern indicating the fluctuation state of the power generation amount in the photovoltaic power generation device. According to this power generation amount prediction system, it is possible to predict the power generation amount of the photovoltaic power generation device in consideration of the influence of clouds that flow in the wind, but error factors due to equipment deterioration and weather conditions are not considered. However, there is a possibility that the power generation amount cannot be accurately predicted.

The main invention for solving the above-mentioned problem is a coefficient output for outputting a coefficient indicating an error between an estimated value estimated by a predetermined estimation method based on one or more predetermined parameters and an actual value corresponding to the estimated value. Part, and the neural network trained based on at least one or more of the predetermined parameters, inputs the predetermined parameter in a predetermined period, outputs the coefficient after learning, and outputs the learning A prediction unit that calculates a predicted value on the predicted date based on the coefficient afterwards and an estimated value estimated by the predetermined estimation method on the predicted date.
Other features of the present invention will be apparent from the accompanying drawings and the description of the present specification.

According to the present invention, the error between the estimated value and the actually measured value can be reduced, so that the prediction accuracy is improved.

It is a figure which shows an example of a structure of an electric power system. It is a figure which shows an example of a structure of a power prediction system. It is a table which shows an example of past DB. It is a table which shows an example of equipment DB. It is a table which shows an example of coefficient DB. It is a table which shows an example of a prediction DB. It is a conceptual diagram which shows a neural network. It is a flowchart which shows an example of the process which estimates a prediction date. It is a graph which shows an example of a snow cover coefficient.

At least the following matters will be made clear by the description in the present specification and the accompanying drawings. In the following description, parts denoted by the same reference numerals represent the same elements, and the basic configurations and operations are the same.

===Power System 1===
FIG. 1 is a diagram illustrating an example of the configuration of the power system 1. As shown in FIG. 1, the power system 1 includes a power facility 100, a meteorological observation device 200, and a power prediction system 10. Each component is connected via the network 400.

The power facility 100 is, for example, a power transmission/distribution facility or a power generation facility. Below, the electric power equipment 100 is demonstrated as a solar power generation equipment. The solar power generation facility transmits various types of information regarding power to the power prediction system 10 via the network 400.

The meteorological observation instrument 200 is a measuring instrument related to meteorology and includes, for example, a pyranometer, a thermometer, and an anemometer. The meteorological observation instrument 200 transmits, to the power prediction system 10, for example, solar radiation amount information indicating the solar radiation amount, temperature, and temperature information indicating the wind speed. When the power equipment 100 is, for example, a wind power generation equipment, the meteorological observation instrument 200 may include measurement of the wind direction. The meteorological observation instrument 200 may output various information to the power prediction system 10 as a digital signal via the network 400, or may output directly as an analog signal.

The power prediction system 10 is a system that predicts a power generation output of a photovoltaic power generation facility in the future based on a coefficient (adjustment coefficient) indicating a difference between an estimated value and an actual value related to power. The power prediction system 10 acquires information for calculating coefficients such as solar radiation amount information, temperature information, and wind speed information from the meteorological observation device 200. The power prediction system 10 may be configured to obtain the solar radiation amount information, the wind speed information, and the weather information from the JMA database 300 via the network 400.

As a method of estimating the power generation output of the photovoltaic power generation facility 100 in the future, it is possible to estimate based on weather data and panel capacity. However, with this method, it is difficult to accurately determine the degree to which the meteorological data influences the calculation of power generation output, and the overloading and the panel load caused by the small capacity of the inverter to the size of the panel capacity. Due to the fact that the decrease in the power generation output due to the deterioration of is not taken into consideration, the estimation accuracy of the power generation output may decrease.

On the other hand, in the power prediction system 10 of the present embodiment, by multiplying the estimated value by a coefficient for suppressing an error between the estimated value of the generated output and the actual value corresponding to the estimated value, Enables accurate prediction. The power prediction system 10 will be described in detail below.

===Power Prediction System 10===
FIG. 2 shows a configuration example of the power prediction system 10. As shown in FIG. 2, the power prediction system 10 includes an input unit 11, a control unit 12, a storage unit 13, and a communication unit 14.

The input unit 11 receives various inputs to the power prediction system 10.

The control unit 12 includes an arithmetic processing unit 121 such as a CPU or MPU and a memory such as a RAM. The arithmetic processing unit 121 operates various functional units by executing a program stored in the storage unit 13 based on various inputs. This program may be stored in a recording medium such as a CDROM or distributed via the network 400 and installed in a computer. The memory is for temporarily storing various kinds of information necessary for calculation during execution of processing in the power prediction system 10 program.

The storage unit 13 is configured by a storage device such as a hard disk, and stores various programs necessary for execution of processing in the control unit 12, data necessary for execution of various programs, and the like. In the present embodiment, it is desirable that the storage unit 13 has a past database (DB) 131, a facility DB 132, a coefficient DB 133, and a prediction DB 134. Note that each DB is shown as a relational database as an example.

In the past DB 131, information on past achievements for calculating the estimated value of the power generation output is registered. As shown in FIG. 3, for example, the past DB 131 has fields indicating “date and time”, “point”, “solar radiation amount”, “temperature”, “wind speed”, “power generation output”, and “estimated value”. There is.

Information on the photovoltaic power generation facility 100 for calculating the estimated value of the power generation output is registered in the facility DB 132. As shown in FIG. 4, for example, the facility DB 132 stores “point”, “tilt angle”, “installation azimuth angle”, “type” of the photovoltaic power generation facility 100, “panel capacity” of the photovoltaic power generation facility 100, and power. It has a field that shows the "power conditioner capacity" that indicates the capacity of the conditioner, and a fixed coefficient (hereinafter, "fixed coefficient") that is generally used according to the type of equipment, etc. to consider changes over time and loss. There is.

In the coefficient DB 133, an adjustment coefficient indicating an error between the estimated value and the actual value is registered. As illustrated in FIG. 5, for example, the coefficient DB 133 has fields indicating “time point”, “actual value”, “estimated value”, and “adjustment coefficient”.

In the prediction DB 134, the adjusted adjustment coefficient for calculating the predicted value from the estimated value is registered. As illustrated in FIG. 6, for example, the prediction DB 134 has fields indicating “prediction time”, “estimated value”, “correction adjustment coefficient”, and “prediction value”.

The communication unit 14 has a communication interface for connecting the power prediction system 10 to the private line VPN or the network 400. The communication unit 14 is, for example, a LAN card, an analog modem, an ISDN modem, or the like, and is an interface for connecting these to the processing unit via a transmission path such as a system bus.

As shown in FIG. 2, the arithmetic processing unit 121 includes a reception unit 121a, an estimation unit 121b, a coefficient calculation unit 121c, a learning unit 121d, a re-learning unit 121e, and a prediction unit 121f as functional units.

In the present embodiment, as will be described later, the machine based on the slope solar radiation amount, the temperature, the wind speed (predetermined parameter) for calculating the power generation output of the photovoltaic power generation facility 100, and the adjustment coefficient corresponding to these Calculate the predicted value using learning. Here, machine learning includes supervised learning, unsupervised learning, and reinforcement learning, and the prediction value may be calculated using any of them. However, in the present embodiment, as an example, there is a supervised learning using a neural network. The calculation of the predicted value by learning will be described below.

The reception unit 121a is a functional unit that registers various received data in the past DB 131.

The estimation unit 121b is a functional unit that calculates an estimated value of the power generation output of the photovoltaic power generation facility 100 based on the weather observation data registered in the past DB 131. Specifically, the estimation unit 121b calculates the estimated value of the power generation output, for example, based on the amount of solar radiation, the temperature, the wind speed, the panel capacity, and the system output coefficient of the photovoltaic power generation facility 100 at a predetermined time in the past. .. The method of estimating the power generation output estimated value is not particularly limited, and a method of estimating using the following formulas (1) to (4) can be considered as an example.

Tpa=Ta+(A/(B* ^V0.8 +1)+2)*Ga-2...(1)
(Tpa: solar cell panel temperature (°C), Ta: outside air temperature (°C), A: coefficient (for example, roof-mounted type "50"), B: coefficient (for example, roof-mounted type "0.38"), V: wind speed (M/s), Ga: amount of solar radiation on the inclined surface (kW/m ² ))

Kpt=1+αmax×(Tpa-25) (2)
(Kpt: temperature correction coefficient, αmax: maximum output temperature coefficient (1/°C), Tpa: solar cell panel temperature (°C))

System output coefficient=Kloss×Kpt (3)
(Kloss: a fixed coefficient that is generally used according to the type of equipment to consider changes over time (dirt, deterioration), wiring resistance loss, inverter loss, etc., Kpt: temperature correction coefficient)

Estimated value (Pw) = amount of solar radiation on inclined surface (Ga) x system output coefficient x panel capacity (4)

The coefficient calculation unit 121c is a functional unit that calculates an adjustment coefficient indicating an error between the estimated value of the power generation output of the photovoltaic power generation facility 100 and the actual value of the power generation output actually generated at a predetermined time. The adjustment coefficient calculated by the coefficient calculation unit 121c is, for example, the actual value divided by the estimated value, and is a coefficient for eliminating the low accuracy of the system output coefficient and the reduction of the accuracy of the estimated value due to overloading of the panel. Is. The coefficient calculation unit 121c registers the adjustment coefficient in the coefficient DB 133.

The coefficient calculation unit 121c acquires the actual value and the estimated value in the predetermined period from the past DB 131, and extracts the acquired actual value and the estimated value that are larger than the threshold value obtained by multiplying the power conditioner capacity by the predetermined coefficient. To do. The predetermined coefficient is, for example, a ratio indicating the output lower limit of the rated output of the photovoltaic power generation facility (panel). Then, the coefficient calculation unit 121c extracts the adjustment coefficient corresponding to the extracted actual value and estimated value from the coefficient DB 133. As described above, by excluding the actual value and the estimated value on the day when the power generation output is small, the actual value and the estimated value in the operating state of the power generation output level or higher at which the error of the estimated value of the photovoltaic power generation facility 100 becomes a problem. Since the adjustment coefficient can be calculated based on the value, the prediction accuracy can be improved.

The learning unit 121d is a functional unit that learns a neural network. This learning is supervised learning that uses training data that is a pair of input data and a label. Supervised learning is used to predict a particular outcome for given input data. In the present embodiment, for example, the input data is the amount of solar radiation on the inclined surface, the temperature, the wind speed, and the date and time, and the label is the adjustment coefficient obtained by dividing the actual value of the power generation output corresponding to the input data by the estimated value.

In this embodiment, the neural network is described as a multi-layer perceptron having one hidden layer, but it may be one having a plurality of hidden layers or a simple perceptron. A multilayer perceptron neural network having one hidden layer will be described below.

FIG. 7 is a conceptual diagram showing a neural network. As shown in FIG. 7, the nodes 01 to 04 indicate input data (input feature amount). A line connecting the nodes 01 to 04 and the nodes 11, 12, and 13 indicates a learned coefficient (hereinafter, referred to as “learning coefficient”) (weight). The nodes 11, 12 and 13 are so-called hidden layers. Similarly, the line connecting the nodes 11, 12, 13 and the node 21 indicates the learned learning coefficient (weight). The node 21 shows output data. The output data is the sum of the input data and the learning coefficient (weight).

The learning unit 121d gives the input data and the label to the neural network, and changes the learning coefficient so that the error between the label and the output data is reduced. As described above, the learning unit 121d causes the neural network to learn the characteristics of the training data and recursively acquires the neural network that outputs the optimum adjustment coefficient from the input data (hereinafter, referred to as “learned model”). ..

Here, the activation function for adjusting the electric signal propagating between the layers may be any function that can convert any real number into a non-linear form. For example, the activation function is a hyperbolic tangent function, a ReLU function, or the like.

The learning unit 121d registers, in the storage unit, the learned model constructed by the supervised learning up to that point. Thereby, the re-learning unit 121e described later can acquire the learned model and execute the fine tuning.

The re-learning unit 121e is a functional unit that performs fine tuning using the learning coefficient of the learned model. In the present embodiment, the training data learned by the learning unit 121d is, for example, data for one year, whereas the training data learned by the re-learning unit 121e is, for example, the latest two weeks immediately before the predicted date. In this way, re-learning with the training data of the latest past, which is the same weather condition as the prediction date, improves the prediction accuracy as compared with the case where the learning model is generated only with the training data for one year.

The predicting unit 121f inputs the amount of solar radiation on the inclined surface, the air temperature, and the wind speed in a predetermined period into the learned model, outputs the adjustment coefficient after learning, and when the adjustment coefficient after learning satisfies the predetermined condition. Is a functional unit that calculates the predicted value by multiplying the estimated value on the predicted date by the adjustment coefficient after learning. Here, the predetermined condition means that the predetermined condition falls between a predetermined upper limit threshold value and a predetermined lower limit threshold value. By using the adjustment coefficient after learning that falls within such a fixed range, the abnormal value can be eliminated, and thus the prediction accuracy can be improved.

If the adjustment coefficient after learning does not satisfy the predetermined condition, the prediction unit 121f further corrects the adjustment coefficient after learning so as to fall between the upper threshold and the lower threshold. Correction of the adjustment coefficient after learning is, for example, correction of the adjustment coefficient after learning to the upper limit threshold value or the lower limit threshold value, or to an intermediate value between the upper limit threshold value and the lower limit threshold value.

=== Operation of Power Prediction System 10 ===
FIG. 8 is a flow showing an example of processing of the power prediction system 10. A process in which the power prediction system 10 calculates a predicted value of the power generation output of the photovoltaic power generation facility will be described with reference to FIG. 8.

First, the reception unit 121a of the power prediction system 10 acquires various weather data observed by the weather observation device 200 in advance (S10), and registers actual values in the past DB 131 (S11). Further, the power prediction system 10 has a facility DB 132 built in advance.

The estimation unit 121b of the power prediction system 10 calculates an estimated value of the power generation output of the photovoltaic power generation facility based on various past weather information of the past DB 131 and the facility information of the facility DB 132. The power prediction system 10 registers the calculated estimated value and the actual value corresponding to the estimated value in the coefficient DB 133 (S12). This estimated value is an estimated value in the past.

In the following, as an example to facilitate understanding, as shown in FIG. 5, the actual values at time points 1 to 5 in the predetermined photovoltaic power generation facility 100 are “105 kW”, “200 kW”, “323 kW”, and “323 kW”, respectively. 441 kW” and “513 kW”, and the estimated values at time points 1 to 5 are “150 kW”, “250 kW”, “380 kW”, “490 kW”, and “540 kW”, respectively.

The coefficient calculation unit 121c of the power prediction system 10 calculates the adjustment coefficients for time points 1 to 5 and registers them in the coefficient DB 133 (S13, S14). Specifically, the adjustment coefficient at time point 1 is “0.7” obtained by dividing the actual value “105 kW” at time point 1 by the estimated value “150 kW” at time point 1. Similarly, the adjustment coefficient at time point 2 is “0.8”, the adjustment coefficient at time point 3 is “0.85”, the adjustment coefficient at time point 4 is “0.9”, and the adjustment coefficient at time point 5 is “0”. .95".

The coefficient calculation unit 121c refers to the coefficient DB 133 and extracts, from the actual value and the estimated value, a predetermined condition, that is, a value that is larger than a threshold value obtained by multiplying the power conditioner capacity by a predetermined coefficient in the present embodiment ( S15). Hereinafter, it is assumed that the power prediction system 10 extracts the past year's actual value and estimated value as an example. It should be noted that the more past data there is, the more desirable it is.

The learning unit 121d of the power prediction system 10 uses the adjustment coefficient corresponding to the extracted actual value and estimated value as a label, and learns the neural network using the slope solar radiation amount, air temperature, and wind speed corresponding to the adjustment coefficient as input data. Then, the learned model is generated (S16). The learning unit 121d assigns weights to the neural network with random numbers before starting learning. Further, the condition for the learning unit 121d to end the supervised learning for the neural network can be arbitrarily determined. For example, when the supervised learning is repeated for a predetermined number of days, the supervised learning is ended. Further, when the value of the error between the output data of the neural network and the label becomes a predetermined value or less, the supervised learning may be ended.

Next, the operator designates a prediction date in the power prediction system 10 (S17).

The re-learning unit 121e of the power prediction system 10 fine-tunes the learned model with the training data for the most recent two weeks of the predicted date (S18, S19). This can improve the prediction accuracy of the learned model. Even in the training data, similar to the extraction work in the coefficient calculation unit 121c, a predetermined condition among the actual value and the estimated value, that is, a threshold value obtained by multiplying the power conditioner capacity by a predetermined coefficient in the present embodiment is used. The larger one is extracted and used.

The prediction unit 121f of the power prediction system 10 inputs the slope solar radiation amount, the temperature, and the wind speed to the learned model that has been re-learned, and calculates the adjustment coefficient after learning (S20).

The prediction unit 121f determines whether or not the learned adjustment coefficient is between the predetermined upper limit threshold and the predetermined lower limit threshold (S21). In the following description, the upper limit threshold value is "0.95" and the lower limit threshold value is "0.85" as an example. As shown in FIG. 6, when the adjustment coefficient after learning is “0.9”, the prediction unit 121f determines that it is within the range. Here, for example, when the adjustment coefficient after learning is “0.8”, the prediction unit 121f determines that the adjustment coefficient after learning is out of the range, and sets the learned adjustment coefficient to the lower threshold “0. It is corrected to 85". Further, for example, when the adjustment coefficient after learning is “0.98”, the prediction unit 121f determines that the adjustment coefficient after learning is out of the range, and sets the adjustment coefficient after learning to the upper threshold “0.95”. To be corrected. Although the adjustment coefficient after learning is corrected to the lower limit threshold value or the upper limit threshold value, the adjustment coefficient may be corrected to fall between the lower limit threshold value and the upper limit threshold value.

The estimation unit 121b of the power prediction system 10 calculates an estimated value of a predetermined solar power generation facility 100 on the designated prediction date (S22). The prediction unit 121f multiplies the calculated estimated value by the adjustment coefficient after learning to calculate the predicted value of the predicted date (S23). The prediction unit 121f registers the predicted value in the prediction DB 134 (S24).

=== Other Embodiments ===
The prediction unit 121f calculates the predicted value by multiplying the estimated value by the adjustment coefficient as described above, and further, the snow cover coefficient for reflecting the decrease in the power generation output due to the snow on the panel of the photovoltaic power generation facility 100. You may multiply by. The snowfall coefficient shows a value as shown in FIG. 9, for example. In FIG. 9, for example, the vertical axis represents “snow cover coefficient” and the horizontal axis represents “snow depth”. As shown in FIG. 9, as for the snow accumulation coefficient, the snow accumulation coefficient becomes smaller between the predetermined snow depths (L and H) as the snow depth becomes larger. The prediction unit 121f determines whether or not there is snow within a predetermined time in the past, and when it determines that there is no snow, determines that there is no snow on the panel and sets the snow coefficient to "1". When it is determined that there is snow, it is determined that there is snow on the panel, and the prediction unit 121f uses a statistical method from, for example, snow depth information, meteorological information, facility information and constant information at the present or future time. , Calculate the snow cover factor.

In addition, the prediction unit 121f may calculate the predicted value by subtracting the electric power for self-consumption consumed by the photovoltaic power generation facility 100 from the predicted value calculated as described above.

Further, although the predetermined power facility 100 is described as the solar power generation facility 100, it may be a wind power generation facility, a power demand facility, or the like. In other words, equipment that can calculate the estimated value by calculating the estimated value, calculating the adjustment coefficient using the estimated value and the actual value that satisfy the predetermined conditions, and calculating the learned adjustment coefficient using the neural network If

Further, in FIGS. 1 and 2, the power prediction system 10 is shown to be configured by one device, but the present invention is not limited to this. The power prediction system 10 may be configured by a plurality of devices, or may be configured on the cloud.

In addition, the power prediction system 10 is a system that predicts the power generation output of the solar power generation facility 100 in the future based on a coefficient (adjustment coefficient) indicating the difference between the estimated value and the actual value of the power generation output of the solar power generation facility 100. However, the present invention is not limited to this. The system may be any system as long as it can obtain an estimated value and an actual value, and for example, a system that predicts the number of products manufactured by manufacturing equipment, or general power equipment such as power demand equipment and power consumption equipment. Good.

Further, although the learning unit 121d has been described as learning the neural network by using the solar radiation amount on the slope, the temperature, and the wind speed as input data, the learning unit is not limited thereto, and various parameters such as humidity and weather are used as input data. You may let them learn.

Further, the learning unit 121d may normalize the input data and convert it into a range of 0 to 1 for use. In the normalization, each feature amount is divided by the maximum value of the same feature amount. This makes it possible to match the difference in scale between the respective feature amounts.

Further, the learning unit 121d may convert the input data so as to follow the standard normal distribution so that the average of each feature amount (input data) is “0” and the variance is “1”. By setting the distributions of the respective feature amounts in this manner, it is possible to evenly see the sensitivity of the respective feature amounts to the movement.

Further, the learning unit 121d has been described as building a model of one neural network by inputting the date and time, but a learned model for each prediction time, for example, when predicting every hour, 12:00. A learned model may be constructed at each time, such as a learned model of 12:00 for prediction and a learned model of 13:00 for prediction of 13:00. This can reduce the influence of input values at other times on the predicted value at that time.

=== Summary ===
The power prediction system 10 according to the present embodiment includes an error between an estimated value estimated by a predetermined estimation method based on one or more predetermined parameters (solar radiation, temperature, wind speed, etc.) and an actual value corresponding to the estimated value. A coefficient output unit 121c for outputting a coefficient indicating the above, and a neural network learned based on at least one predetermined parameter (solar radiation, temperature, wind speed, etc.) among predetermined parameters, and a predetermined parameter in a predetermined period. And a prediction unit 121f that outputs the adjustment coefficient after learning and calculates the prediction value on the prediction date based on the output adjustment coefficient after learning and the estimated value on the prediction day. .. Thereby, the error between the estimated value and the actual value can be reduced.

In addition, the power prediction system 10 according to the present embodiment uses the predetermined parameter used to calculate the estimated value in order to output the adjusted coefficient after learning indicating the error between the estimated value and the actual value corresponding to the estimated value. The learning unit 121d for learning the neural network based on at least one of the parameters is further provided. As a result, the neural network can be learned quickly and efficiently.

Further, the power prediction system 10 according to the present embodiment includes the re-learning unit 121e that fine-tunes the neural network learned by the learning unit 121d in a predetermined period based on a predetermined parameter in a period shorter than the predetermined period. Further prepare. As a result, the learning accuracy is improved because the model learned by the learning unit 121d is relearned.

In addition, the coefficient calculation unit 121c of the power prediction system 10 according to the present embodiment is a coefficient that indicates an error between the estimated value of the electric power equipment 100 estimated by a predetermined estimation method and the actual value corresponding to the estimated value. To calculate. This enables stable operation of the electric power system by predicting the output of the electric power equipment that is easily affected by equipment deterioration and weather conditions.

Further, the predetermined power facility 100 for which the power prediction system 10 according to the present embodiment predicts the power generation output is a solar power generation facility, and the predetermined parameter includes at least the amount of solar radiation. As a result, the difference between the estimated value and the actual value of the power generation facility using natural energy can be reduced.

In addition, the prediction unit 121f of the power prediction system 10 according to the present embodiment displays the adjustment coefficient after learning, the estimated value of the predicted date, and the snowfall coefficient indicating the influence of snowfall on the panel of the predetermined power facility 100. Based on this, the predicted value on the predicted day is calculated. As a result, the predicted value is calculated in consideration of the snowfall on the panel, so that a more accurate predicted value can be calculated.

Further, the coefficient calculation unit 121c of the power prediction system 10 according to the present embodiment extracts an estimated value that is equal to or greater than a predetermined threshold value from the estimated values, extracts an actual value that is equal to or greater than the predetermined threshold value from the actual value, and extracts the actual value. A coefficient indicating the difference between the estimated value and the actual value is calculated. As a result, the error between the estimated value and the actual value is further reduced because the estimated value and the actual value in the operating state of the power generation output level or above, where the error in the estimated value of the predetermined power facility 100 becomes a problem, are used for prediction. It can be reduced.

Note that the above-described embodiments are for facilitating the understanding of the present invention and are not for limiting the interpretation of the present invention. The present invention can be modified and improved without departing from the spirit thereof, and the present invention also includes equivalents thereof.

10 power prediction system 100 power equipment 121c coefficient calculation unit 121d learning unit 121e re-learning unit 121f prediction unit

Claims (8)

A coefficient output unit that outputs a coefficient indicating an error between an estimated value estimated by a predetermined estimation method based on one or more predetermined parameters and an actual value corresponding to the estimated value,
The neural network trained on the basis of at least one or more of the predetermined parameters, inputs the predetermined parameters in a predetermined period, outputs the coefficient after learning, the output after learning the Based on a coefficient and an estimated value estimated by the predetermined estimation method on the predicted date, a prediction unit that calculates a predicted value on the predicted date,
A prediction system comprising: The neural network based on at least one of the predetermined parameters so as to output the coefficient indicating an error between the estimated value estimated by the predetermined estimation method and the actual value corresponding to the estimated value. The prediction system according to claim 1, further comprising: a learning unit that learns. The re-learning unit for fine tuning the neural network learned by the learning unit in a predetermined period based on the predetermined parameter in a period shorter than the predetermined period. Prediction system described. The coefficient output unit outputs a coefficient indicating an error between an estimated value related to electric power estimated by a predetermined estimation method in electric power equipment and an actual value corresponding to the estimated value. The prediction system according to claim 3. The power facility is a photovoltaic power generation facility,
The prediction system according to claim 4, wherein the predetermined parameter includes at least solar radiation amount information indicating a solar radiation amount. The prediction unit, based on the coefficient after the learning, an estimated value estimated by the predetermined estimation method of the prediction day, and a snow cover coefficient indicating the effect of snow on the panel of the power equipment, the prediction The prediction system according to claim 4, wherein a prediction value in a day is calculated. The coefficient calculation unit extracts an estimated value of a predetermined threshold value or more among the estimated values estimated by the predetermined estimation method, extracts the actual value of a predetermined threshold value or more of the actual value, and the extracted The prediction system according to any one of claims 1 to 6, wherein a coefficient indicating an error between an estimated value and the actual value is output. A coefficient output step of outputting a coefficient indicating an error between an estimated value estimated by a computer based on one or more predetermined parameters by a predetermined estimation method, and an actual value corresponding to the estimated value,
The neural network trained on the basis of at least one of the predetermined parameters is input with the predetermined parameter for a predetermined period, and the coefficient after learning is output. A prediction step of calculating a predicted value on the predicted day based on the coefficient and an estimated value estimated by the predetermined estimation method on the predicted day;
Prediction method to realize.

Images