JP2018036196A - Wind speed prediction device, wind speed prediction system, wind speed predication method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately predict a confidence interval of predication error in prediction of wind speed.SOLUTION: A wind speed prediction device for predicting a wind speed is configured to: input learning data indicating an actual value that a wind speed at a point in the past is measured and a predication value that the actual value is predicted; at least calculate, based on the learning data, a model for wind speed at a prediction point when a predetermined time elapses from a current point, and a confidence interval of predication error at the predication point, based on the model; calculate an evaluation index of the confidence v of prediction error; and determine, based on the evaluation index, a period data used for prediction, out of the learning data.SELECTED DRAWING: Figure 8

Description

  The present invention relates to a wind speed prediction apparatus, a wind speed prediction system, a wind speed prediction method, and a program.

  A method is known in which a computer or the like calculates and predicts electric power generated in wind power generation based on weather conditions or the like.

  For example, a method for accurately predicting wind power generation output is known. Specifically, first, the wind power generation output prediction device performs collective learning on a plurality of physical models and a plurality of statistical models using past terrain conditions, weather conditions, and actually measured values, and then a plurality of physical models and a plurality of statistics. Determine the final model that best combines the models. Next, the wind power generation output prediction device uses the final model, and finally predicts the wind power generation output at the target point at a specific time based on the topographic conditions, weather conditions, and actual measurement values at a specific time after the past. Calculate the value. As described above, there is known a method in which the wind power generation output prediction apparatus can accurately predict the wind power generation output by using the optimized final model (for example, Patent Document 1).

  In addition, a method is known in which a computer or the like calculates and predicts the generated power based on the cube of the wind speed.

  In addition, the natural energy amount prediction apparatus first acquires an explanatory variable related to a change in the natural energy amount and calculates a probability distribution. Then, the natural energy amount prediction device calculates the confidence interval of the predicted value of the natural energy amount based on the probability distribution. Thus, a method for accurately predicting the predicted value of the amount of energy per unit time obtained from natural energy including the distribution of prediction errors is known (for example, Patent Document 2).

JP 2007-233639 A Japanese Patent Laid-Open No. 2014-21555

  However, the conventional method does not consider the fluctuation characteristics of the wind speed due to the season or the like. That is, since the wind speed has a variation characteristic depending on the season and the like, if the wind speed is predicted without considering the variation characteristic, the confidence interval of the prediction error may not be accurately predicted.

  One aspect of the present invention has been made in view of such problems, and an object thereof is to accurately predict a confidence interval of a prediction error in wind speed prediction.

In order to solve the above-described problems and achieve the object, the wind speed prediction apparatus for predicting the wind speed in one embodiment of the present invention is
An input unit for inputting learning data indicating an actual value obtained by measuring the wind speed at a time point in the past from the present time and a predicted value obtained by predicting the actual value;
Based on the learning data, a calculation unit for calculating at least a wind speed model at a prediction time after a predetermined time from the current time, and a prediction error confidence interval at the prediction time based on the model;
An evaluation index calculation unit for calculating an evaluation index of the confidence interval of the prediction error;
And a determination unit that determines period data to be used for prediction among the learning data based on the evaluation index.

  According to the present invention, it is possible to accurately predict a confidence interval of a prediction error in wind speed prediction.

It is a block diagram which shows an example of the hardware constitutions of the wind speed prediction apparatus in one Embodiment of this invention. It is a flowchart which shows an example of the whole process by the wind speed prediction apparatus in one Embodiment of this invention. It is a figure which shows an example of the learning data in one Embodiment of this invention. It is a schematic diagram which shows the example of whole process in one Embodiment of this invention. It is a figure which shows an example of the confidence area of the prediction error of the learning data in one Embodiment of this invention. It is a figure which shows the example of selection of the learning data based on the confidence interval of the prediction error in one Embodiment of this invention. It is a figure which shows an example of the effect in one Embodiment of this invention. It is a functional block diagram which shows an example of a function structure of the wind speed prediction apparatus in one Embodiment of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

1. 1. Hardware configuration example of wind speed prediction device 2. Example of overall processing by wind speed prediction device Example of functional configuration of wind speed prediction device

≪ １． Example of hardware configuration of wind speed prediction device ≫
The wind speed prediction apparatus is an information processing apparatus such as a PC (Personal Computer). Hereinafter, an example in which the wind speed prediction apparatus is a PC will be described.

  FIG. 1 is a block diagram illustrating an example of a hardware configuration of a wind speed prediction apparatus according to an embodiment of the present invention. Specifically, the PC 10 includes a CPU (Central Processing Unit) 101, a storage device 102, a network I / F (interface) 103, an input I / F 104, and an output I / F 105. That is, the wind speed prediction apparatus is a computer that executes a program to execute processing.

  The CPU 101 is an arithmetic device that performs calculations for realizing various processes and various controls and processes various data. Further, the CPU 101 is a control device that controls the hardware of the PC 10.

  The storage device 102 stores data, programs, setting values, and the like used by the PC 10. The storage device 102 is a so-called memory or the like. Note that the storage device 102 may include an auxiliary storage device such as a hard disk.

  The network I / F 103 transmits and receives various data and the like to and from devices connected via a network such as a LAN (Local Area Network). For example, the network I / F 103 is a connector or the like for connecting a NIC (Network Interface Controller) and a LAN cable. Note that the network I / F 103 is not limited to an I / F that uses a network, and may be an I / F that transmits and receives to / from an external device through a cable, wireless, or a connector.

  The input I / F 104 is an interface with a user who uses the PC 10. Specifically, the input I / F 104 inputs various operations performed by the user. For example, the input I / F 104 includes an input device such as a keyboard and a connector that connects the input device to the PC 10.

  The output I / F 105 is an interface with a user who uses the PC 10. Specifically, the output I / F 105 outputs processing results and the like of various processes performed by the PC 10 to the user. For example, the output I / F 105 is an output device such as a display and a connector for connecting the output device to the PC 10.

  Note that the PC 10 may further include an auxiliary device that assists the processing by each hardware resource. Further, the PC 10 may further include an apparatus inside or outside in order to perform various processes in parallel, redundantly, or distributedly. Further, the PC 10 may be composed of a plurality of information processing apparatuses.

≪ 2. Example of overall processing by wind speed prediction device ≫
FIG. 2 is a flowchart showing an example of overall processing by the wind speed prediction apparatus according to the embodiment of the present invention. For example, the PC performs processing as illustrated.

≪ Example of learning data input (step S101) ≫
In step S101, the PC inputs learning data DL. The learning data DL includes a result of measuring the wind speed (hereinafter referred to as “actual value”) at a time point in the past from the present time when the wind speed is predicted. Further, the learning data DL includes a result of predicting the actual value (hereinafter referred to as “predicted value”). For example, the learning data DL, that is, the actual value and the predicted value are the following data.

  FIG. 3 is a diagram illustrating an example of learning data according to an embodiment of the present invention. In the figure, the horizontal axis is time, and the vertical axis is wind speed. In this example, the current time point when the prediction is performed is a current time TC.

  The learning data DL shown in the figure is data of an example in the case where the actual value is measured at six time points in the past from the current time point TC. Specifically, in the illustrated example, each actual value is measured at the time of the first time T1, the second time T2, the third time T3, the fourth time T4, the fifth time T5, and the sixth time T6. Suppose that In the figure, each actual value RES measured at each time point is indicated by a point. The actual value RES may be a value measured irregularly.

  As shown in the figure, a predicted value CAL obtained by predicting the actual value RES is input to the learning data DL. That is, the predicted value CAL is the result of predicting the wind speed at each time point in the past. As described above, the learning data DL is data in which the actual value measured at a plurality of time points and the predicted value obtained by predicting the actual value are input in advance. For example, the PC inputs learning data DL for one year. Hereinafter, an example in which the learning data DL is data for one year will be described.

<< Example of construction of learning data (step S102) >>
Returning to FIG. 2, in step S102, the PC constructs learning data. Specifically, first, the PC divides the learning data. The PC divides the learning data so that each of the divided data (hereinafter referred to as “period data”) is evenly within a predetermined period. For example, when the predetermined period is set as “1 month” in advance, the learning data for one year is divided into period data in units of one month, and each period data is The data shows the actual value and the predicted value at the time. When divided in units of one month in this way, twelve period data are generated by division from learning data for one year. Hereinafter, it is assumed that the learning data is divided into “m” period data. The predetermined period may be, for example, 1 minute unit, 1 hour unit, 1 day unit, 1 week unit, or season unit. The predetermined period is preset in the PC by a user operation or the like, for example.

  In the following description, the period data is represented by the following formula (1).

なお、上記（１）式に示す期間データのうち、最も現時点に近いデータ、すなわち、最新のデータを添え字の「１」とする。一方で、上記（１）式に示す期間データのうち、最も現時点に遠いデータ、すなわち、最も古いデータを添え字の「ｍ」とする。

Of the period data shown in the above equation (1), the data closest to the current time, that is, the latest data is set as the subscript “1”. On the other hand, among the period data shown in the above equation (1), the data farthest from the current time, that is, the oldest data is set as the subscript “m”.

  Next, the PC selects one or a combination of two or more of the plurality of period data generated by the division. Hereinafter, it is assumed that a plurality of pieces of period data are selected by the PC in the order from the current time point.

<< Example of model calculation based on learning data (step S103) >>
Returning to FIG. 2, in step S103, the PC calculates a model based on the learning data. That is, the PC calculates a model using one or a plurality of period data among the period data divided in step S102.

  Hereinafter, the current time point when the prediction is performed is “i”. Then, the PC predicts the wind speed at the predicted time point after a predetermined time from the current time point of “i”. In the following description, the predetermined time unit is referred to as “step”. The predicted time point is referred to as “k step” ahead. Specifically, assuming that the predetermined time is in units of “1 hour”, “1 step” becomes “1 hour”. Therefore, in this example, for example, the predicted time point after 5 hours is the time point “5 steps” ahead.

  Then, at the past “i” time point, the prediction error at the prediction time point is calculated as in the following equation (2) based on the prediction value and the actual value at the prediction time point “k steps” ahead.

また、上記（２）式で示す予測誤差の標準偏差は、下記（３）式のように計算される。

Further, the standard deviation of the prediction error shown by the above equation (2) is calculated as the following equation (3).

そして、現時点から「ｋステップ」先の予測値についての標準偏差は、例えば、下記（４）式に示すモデルによって計算できる。

Then, the standard deviation of the predicted value “k steps” ahead from the present time can be calculated by, for example, a model shown in the following equation (4).

以上のように、学習データに基づいて、モデルが計算される。そして、上記（４）式に示すように、モデルが計算されると、予測値の標準偏差が計算できる。このようにして、学習データに基づいて、学習が行われる。

As described above, the model is calculated based on the learning data. Then, as shown in the above equation (4), when the model is calculated, the standard deviation of the predicted value can be calculated. In this way, learning is performed based on the learning data.

<< Calculation example of evaluation index (step S104) >>
Returning to FIG. 2, in step S104, the PC calculates an evaluation index for each model. For example, the evaluation index is PICP (Prediction Interval Coverage Probability). Also, PICP is a value calculated by the following equation (5), for example.

例えば、まず、上記（４）式に示すモデルに基づいて、標準偏差が計算されると、「１σ」の下限値及び上限値のそれぞれの推定値が計算できる。そして、上記（５）式に示すように、実績値が、下限値及び上限値の範囲、すなわち、予測誤差の信頼区間に入ったか否かによって、「ｃ_ｉ（ｋ）」が「１」又は「０」に定まる。次に、各「ｃ_ｉ（ｋ）」をサンプル数「Ｎ」で除算した値の総和を計算すると、ＰＣは、上記（５）式のように、ＰＩＣＰを計算することができる。

For example, first, when the standard deviation is calculated based on the model shown in the above equation (4), the estimated values of the lower limit value and the upper limit value of “1σ” can be calculated. Then, as shown in the above equation (5), “c _i (k)” is “1” or “2” depending on whether the actual value has entered the range of the lower limit value and the upper limit value, that is, the confidence interval of the prediction error. Set to “0”. Next, when the total sum of values obtained by dividing each “c _i (k)” by the number of samples “N” is calculated, the PC can calculate the PICP as shown in the above equation (5).

  More detailed calculation method of PICP is, for example, "Abbas Khosravi, Saeid Nahavandi and Doug Creighton," Construction of Optimal Prediction Intervals for Load Forecasting Probrem ", IEEE Transactions on Power Systems, Vol.25, Issue 3, pp.1496-1503 (2010) ".

These estimated values are “l _j ” and “u _j ” in the above equation (5). On the other hand, with the standard deviation of “1σ”, the probability that the actual value is theoretically included in the confidence interval of the prediction error is about 68.27%. Further, the theoretical value of PICP is expressed by the following equation (6) for each integer multiple of the standard deviation.

したがって、ＰＣは、上記（５）式によって計算されるＰＩＣＰと、上記（６）式とのノルム「Ｅ（ｋ）」を下記（７）式のように最小化する問題を計算する。

Therefore, the PC calculates the problem of minimizing the norm “E (k)” between the PICP calculated by the above equation (5) and the above equation (6) as the following equation (7).

なお、上記（７）式において解が存在するには、学習期間の候補は、有限個である条件となる。そして、有限個の学習期間の候補を「ｍ」個用意し、用意された「ｍ」個の学習期間の下で、ＰＣは、それぞれのＰＩＣＰを計算する。このように、上記（７）式の問題を計算すると、ＰＣは、ノルム「Ｅ（ｋ）」が最小となる予測誤差の信頼区間を計算できる。すなわち、上記（７）式に示す問題を計算すると、ＰＣは、上記（６）式に示す理論値に最も近い分布となる期間データを選定できる。

In addition, in order for a solution to exist in the above equation (7), the learning period candidate has a finite number of conditions. Then, “m” candidates for a finite number of learning periods are prepared, and the PC calculates each PICP under the prepared “m” learning periods. Thus, when the problem of the above equation (7) is calculated, the PC can calculate the confidence interval of the prediction error that minimizes the norm “E (k)”. That is, when the problem shown in the above equation (7) is calculated, the PC can select the period data having a distribution closest to the theoretical value shown in the above equation (6).

<< Judgment example of whether or not “k = k _max ” (step S105) >>
Returning to FIG. 2, in step S <b> 105, the PC determines whether or not “k = k _max ”. If the PC determines that “k = k _max ” is not satisfied (NO in step S105), the PC proceeds to step S106. On the other hand, when the PC determines that “k = k _max ” (YES in step S105), the PC proceeds to step S107.

“K _max ” is the maximum value of the step for performing prediction. Specifically, when prediction is performed up to “80” steps ahead, “k _max = 80”.

<< Calculation Example of “k = k + 1” (Step S106) >>
In step S106, the PC calculates “k = k + 1”. That is, in step S106, the PC repeats steps S103 to S105 until “k” becomes “+1” and “k” becomes “k _max ”.

<< Judgment example of whether or not "i = m" (step S107) >>
In step S107, the PC determines whether or not “i = m”. If the PC determines that “i = m” is not satisfied (NO in step S107), the PC proceeds to step S108. On the other hand, if the PC determines that “i = m” (YES in step S107), the PC proceeds to step S109.

<< Calculation Example of “i = i + 1” (Step S108) >>
In step S108, the PC calculates “i = i + 1”. That is, in step S108, the PC sets “i” to “+1” and repeats steps S102 to S107 until “i” becomes “m”.

As described above, the model and the evaluation index are calculated for each step and for each learning data in steps S105 to S108. Then, when “k = k _max ” and “i = m”, which are determination conditions set in advance, are satisfied, the iteration of calculating the model and the evaluation index is completed.

  The above process can be illustrated as follows, for example.

  FIG. 4 is a schematic diagram showing an example of overall processing in an embodiment of the present invention. In the figure, the horizontal axis is time, and the vertical axis is “i”, that is, the processing results of steps S102 to S104 shown in FIG. Accordingly, in the figure, it is assumed that steps S102 to S104 are repeatedly performed from the top (from “i = 1”) toward the bottom (that is, until “i = m”) on the vertical axis.

  For example, as shown in the figure, learning data DL including past actual values and predicted values from the current TC is input to the PC in advance (step S101 shown in FIG. 2). As illustrated, the learning data DL is divided every predetermined period, and the period data DT is generated in advance by the division (step S102 shown in FIG. 2).

  In the illustrated example, first, in “i = 1”, the PC selects the period data DT for the first period TM1 from the learning data DL, and uses the period data DT for the first period TM1 to determine the first model f1. Calculation is performed (step S103 shown in FIG. 2). In this way, at “i = 1”, the PC calculates the first evaluation index EV1 based on the first model f1, for example, using the equation (5) (step S104 shown in FIG. 2). .

  Next, in step S108 shown in FIG. 2, the PC sets “i = 2”. In “i = 2”, the PC first selects the period data DT as in “i = 1”. In the illustrated example, the PC selects the period data DT for the first period TM1 and the second period TM2 from the learning data DL (step S102 shown in FIG. 2). Next, the PC calculates the second model f2 using the period data DT of the first period TM1 and the second period TM2 (step S103 shown in FIG. 2). In this way, at “i = 2”, the PC calculates the second evaluation index EV2 based on the second model f2 using, for example, the above equation (5) (step S104 shown in FIG. 2). .

  Hereinafter, as illustrated, the PC repeatedly performs the same processing as “i = 1” or the like until “i = m”.

  It is desirable that the PC calculates each model for each prediction time point, that is, determines the period data used for each prediction for each prediction time point. Specifically, the period data used to calculate the model at the step “k = 1” and the period data used to calculate the model at the step “k = 2” are determined separately. Is desirable. Therefore, it is desirable that the calculation of the model and the like be performed independently for each step. Specifically, as illustrated in FIG. 2, it is desirable that the model is calculated in step S103 every time “k = k + 1” in step S106.

  As described above, when each period data is selected for each step, the period data to be used may be different. Therefore, a model generated by calculation may be a different model for each step. In this way, when optimum period data is determined for each prediction time point, the PC can calculate a model more suitable for each prediction time point.

<< Example of determining period data based on confidence interval of prediction error (step S109) >>
Returning to FIG. 2, in step S109, the PC first determines a confidence interval of the prediction error. The confidence interval of the prediction error is determined as follows, for example.

  FIG. 5 is a diagram illustrating an example of a confidence interval for a prediction error in an embodiment of the present invention. For example, when the standard deviation is calculated based on the model based on the model as shown in the above equation (4), the upper limit value and the lower limit value of the prediction error confidence interval are determined. The section is determined.

  As shown in FIG. 5A, for example, the confidence interval of the prediction error where the standard deviation is “1σ” includes the actual value with a probability of about 68.27%. The confidence interval of the prediction error can also be illustrated as shown in FIG. 5B, for example. As shown in the figure, the confidence interval for the prediction error is a width centered on the average value of the prediction error determined by the upper limit value and the lower limit value. Note that the predicted value and the average value of the prediction error are agreement.

  Then, as shown in FIG. 5B, at the prediction time point “k steps” ahead of the current TC, the prediction value is calculated so that the confidence interval of the determined prediction error is obtained. Learning data is selected as follows.

  FIG. 6 is a diagram showing an example of selecting learning data based on the prediction error confidence interval in the embodiment of the present invention. In the figure, the vertical axis is PICP and the horizontal axis is step “k”. The figure shows an example of a prediction error confidence interval in which the standard deviation is “1σ”.

  As shown in the figure, the period data of the learning data is the latest “15 days”, the latest “30 days”, the latest “45 days”, the latest “60 days”, and the latest “75 days” from the current TC. An example of data will be described. This example shows the PICP calculation result for the model calculated based on each period data at each step “k”.

  Then, based on the above equation (7), among the calculation results of each PICP, the period data that is the calculation result closest to the theoretical value is selected for each step. That is, the selection result DR is a combination of period data having a PICP closest to the theoretical value at each step.

  Specifically, when the step is “1” (that is, “k = 1”), if period data for “30 days” is used, PICP is the closest value to the theoretical value. In this example, in step “1”, a selection result DR is selected in which period data for “30 days” is selected. Next, when the step is “2” (that is, “k = 2”), using the period data for “45 days”, the PICP is the closest value to the theoretical value. In the example, in the step “2”, the period data for “45 days” is selected as the selection result DR. In this way, in each step, any period data is selected based on the PICP, and the selection result DR is determined.

  Using the period data determined as described above, the wind speed is predicted by the PC in each step. In other words, in the illustrated example, in step “1”, the wind speed is predicted using the latest “30 days” of the learning data. As described above, when learning data determined based on an evaluation index such as PICP is used, the PC can accurately predict a confidence interval of a prediction error.

  The selected period data may not be continuous data from the present time. That is, for example, the PC may not select data for the most recent period from the current time. For example, as for the latest “1 day”, that is, yesterday's period data is not selected, and the latest period data is “15 days” from the latest “2 days” to “16 days”, etc. Selections that are not included may be made. When the latest period data is selected, the influence of seasonality may increase. Therefore, in order to reduce the influence of seasonality, the PC may perform selection not to select data for the most recent period from the current time.

  FIG. 7 is a diagram illustrating an example of an effect in one embodiment of the present invention. First, as shown in the figure, PICP is calculated, and period data DT in which PICP is closest to the theoretical value is selected. Next, the PC predicts the wind speed using the selected period data DT. In this way, the PC can accurately predict the confidence interval of the prediction error.

  Note that the evaluation index is not limited to PICP. For example, the evaluation index may be NMPIL (Normalized Mean Prediction Interval Length) or the like. When PICP is used, the PC can evaluate the reliability of the distribution, whereas when NMPIL is used, the PC can evaluate the degree of separation of the distribution. Calculation method and the like of NMPIL is, for example, "Abbas Khosravi, Saeid Nahavandi and Doug Creighton," Construction of Optimal Prediction Intervals for Load Forecasting Probrem ", IEEE Transactions on Power Systems, Vol.25, Issue 3, pp.1496-1503 ( 2010) ”.

≪ 3. Example of functional configuration of wind speed prediction device ≫
FIG. 8 is a functional block diagram illustrating an example of a functional configuration of the wind speed prediction apparatus according to the embodiment of the present invention. For example, the PC 10 has a functional configuration including an input unit FN1, a calculation unit FN2, an evaluation index calculation unit FN3, and a determination unit FN4 as illustrated.

  The input unit FN1 inputs learning data DL indicating the actual value and the predicted value. For example, the input unit FN1 is realized by the network I / F 103 (see FIG. 1) or the input I / F 104 (see FIG. 1).

  The calculation unit FN2 first calculates a model based on the learning data DL. Next, the calculation unit FN2 calculates a confidence interval of each prediction error at each prediction time based on the model. For example, the calculation unit FN2 is realized by the CPU 101 (see FIG. 1) or the like.

  The evaluation index calculation unit FN3 calculates each evaluation index of the confidence interval of the prediction error calculated by the calculation unit FN2. For example, the evaluation index calculation unit FN3 is realized by the CPU 101 (see FIG. 1) or the like.

  The determination unit FN4 determines period data used for prediction in the learning data DL based on the evaluation index calculated by the evaluation index calculation unit FN3. For example, the determination unit FN4 is realized by the CPU 101 (see FIG. 1) or the like.

≪ Summary ≫
First, the PC 10 uses the input unit FN1 to input learning data DL indicating a wind speed actual value measured in advance and a predicted value obtained by predicting the actual value. When the learning data DL is input, as shown in FIG. 4, the PC 10 can calculate a model by the calculation unit FN2. Next, when the model is calculated, the PC calculates a standard deviation or an integer multiple of the standard deviation based on the model, and calculates a prediction error confidence interval.

  Subsequently, the PC 10 calculates the evaluation index by the evaluation index calculation unit FN3 based on, for example, the above equation (5) based on the model. Thus, when the evaluation index is calculated, the PC predicts the period data that is the PICP closest to the theoretical value as shown in FIG. 6 by the determination unit FN4 based on the above equation (7), for example. You can decide to use it.

  Thus, if the period data used for prediction can be determined, the PC can accurately predict the confidence interval of the prediction error in the wind speed prediction.

  If the confidence interval of the prediction error has good accuracy, it can be used as a guideline for determining the capacity of the storage battery in wind power generation, for example. Specifically, since the storage battery stores the electric power generated by wind power generation, it is necessary to determine the capacity that can be stored according to the amount of power generation at the time of installation. That is, the capacity of the storage battery may be determined according to the predicted maximum wind speed or the like. Therefore, when the confidence interval of the prediction error is predicted with high accuracy, it is possible to reduce a case where a storage battery having a large capacity is unnecessarily selected.

  Note that all or part of each processing according to an embodiment of the present invention is performed by a low-level language such as an assembler, a high-level language such as C language, Java (registered trademark) language, or a program described in combination of these. It may be realized. That is, the program is a computer program for causing a computer such as an information processing apparatus or an information processing system to execute each procedure related to the wind speed prediction method.

  Further, all or a part of each process according to an embodiment of the present invention is processed by a wind speed prediction system having one or more information processing devices, and all or a part of the process is processed in parallel, distributed, redundant, or a combination thereof. May be.

  Moreover, each process which concerns on one Embodiment of this invention is not restricted to the order shown in figure. For example, some or all of the processes may be processed in different orders, in parallel, distributed, or omitted.

  The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the above-described embodiments, and various modifications or changes can be made within the scope of the gist of the present invention described in the claims. Is possible.

10 PC
FN1 Input unit FN2 Calculation unit FN3 Evaluation index calculation unit FN4 Determination unit DL Learning data DT Period data DR Selection result CAL Predicted value RES Actual value

Claims (7)

A wind speed prediction device for predicting wind speed,
An input unit for inputting learning data indicating an actual value obtained by measuring the wind speed at a time point in the past from the present time and a predicted value obtained by predicting the actual value;
Based on the learning data, a calculation unit for calculating at least a wind speed model at a prediction time after a predetermined time from the current time, and a prediction error confidence interval at the prediction time based on the model;
An evaluation index calculation unit for calculating an evaluation index of the confidence interval of the prediction error;
A wind speed prediction apparatus including a determination unit that determines period data to be used for prediction out of the learning data based on the evaluation index.   The wind speed prediction apparatus according to claim 1, wherein the evaluation index is PICP or NMPIL.   The wind speed prediction apparatus according to claim 1, wherein the determination unit determines the period data for each prediction time point.   The wind speed prediction device according to any one of claims 1 to 3, wherein the prediction error confidence interval is a range determined based on a standard deviation of the prediction error or an upper limit value and a lower limit value that are integer multiples of the standard deviation. . A wind speed prediction system having one or more information processing devices and predicting wind speed,
An input unit for inputting learning data indicating an actual value obtained by measuring the wind speed at a time point in the past from the present time and a predicted value obtained by predicting the actual value;
Based on the learning data, a calculation unit for calculating at least a wind speed model at a prediction time after a predetermined time from the current time, and a prediction error confidence interval at the prediction time based on the model;
An evaluation index calculation unit for calculating an evaluation index of the confidence interval of the prediction error;
A wind speed prediction system including a determination unit that determines period data used for prediction among the learning data based on the evaluation index. A wind speed prediction method performed by a wind speed prediction device that predicts wind speed,
Input procedure for inputting learning data indicating a predicted value obtained by predicting the actual value and the actual value obtained by measuring the actual wind speed from the present time by the wind speed prediction device;
The wind speed prediction device calculates at least a wind speed model at a prediction time after a predetermined time from the current time based on the learning data, and a prediction error confidence interval at the prediction time based on the model. Procedure and
The wind speed prediction apparatus calculates an evaluation index calculation procedure for calculating an evaluation index of a confidence interval of the prediction error;
A wind speed prediction method, wherein the wind speed prediction device includes a determination procedure for determining period data used for prediction among the learning data based on the evaluation index. A program for causing a computer for predicting wind speed to execute a wind speed prediction method,
An input procedure for inputting learning data indicating an actual value obtained by measuring the wind speed at a time point in the past from the present time and a predicted value obtained by predicting the actual value;
A calculation procedure in which the computer calculates at least a wind speed model at a prediction time point after a predetermined time from the current time based on the learning data, and a prediction error confidence interval at the prediction time point based on the model; ,
An evaluation index calculation procedure in which the computer calculates an evaluation index of a confidence interval of the prediction error;
The program for making the said computer perform the determination procedure which determines the period data used for prediction among the said learning data based on the said evaluation parameter | index.

Images