JP2014054048A - Power prediction method, device, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a power prediction method, device, and program which can be easily used by an operator.SOLUTION: A power prediction method comprises: a step of previously dividing one region whose total generated power output is to be predicted into a plurality of points; a step of previously collecting observation data having correlation with generated power output for each of points P1 to Pn; a step of previously calculating correlation values showing correlations between two respective points by using the observation data; a step of collecting power data showing generated power output of each of the points P1 to Pn multiple times; and a step of performing statistical analysis on the power data while reflecting one or more of the correlation values. Thereby, a prediction value of generated power output in the region is calculated.

Description

  Embodiments described herein relate generally to a power prediction method, a power prediction device, and a power prediction program for predicting power generation output or power demand in one region.

  Power generation using natural energy such as sunlight and wind power has a drawback that the power generation amount fluctuates greatly due to weather fluctuations. Therefore, when a large number of natural energy generators are connected to the power system, it is difficult to adjust the balance between the power demand and the power generation output, and the power system may become unstable.

  On the other hand, natural energy generators are attracting more and more attention due to rising fuel costs and increasing awareness of environmental protection. This is because the natural energy generator does not require fuel costs and does not emit greenhouse gases. Therefore, it is an urgent task to establish a method for accurately predicting the power generation output of a natural energy generator in preparation for the future when a large number of natural energy generators will be linked to the power system.

  In addition, power demand has changed significantly due to weather fluctuations, making it difficult to adjust the balance between power demand and power output. For example, if the temperature fluctuates greatly, the operating rate of the air conditioner will increase greatly. Therefore, it is necessary to accurately predict power demand.

  Here, for the power system operator, a large number of power generations in a specific area under jurisdiction are generated rather than the individual power generation output of the natural energy generator or the individual power demand required by the electric equipment as a load. The total power output of the machine and the net power demand are important. The net power demand is the demand that the main power supplier should respond to, which is the total power demand minus the output of natural energy power generation.

  A conventional method for predicting the total power output in a region is shown in FIG. As shown in FIG. 1, conventionally, there has been proposed a method for predicting the total power output of the entire region by individually predicting the power output of each natural energy generator existing in the region and summing the results. It was.

  Specifically, a curve model indicating the correspondence between the power generation output and the solar radiation intensity is created from past data representing the maximum power generation output for each time point P1 to Pn and the solar radiation intensity at that time. The predicted power generation outputs Xs1 to Xsn at the points P1 to Pn are obtained from the forecast of the solar radiation intensity (see, for example, Patent Document 1). Then, the total power generation output Xtot in the region is obtained by summing up these predicted power generation outputs. The curve model may be common to all points P1 to Pn, or may be generated individually according to the difference between the past maximum power generation output and the solar radiation intensity at that time, and the characteristics and trends of the point Pi. is there.

JP 2007-173657 A

  It can be said that the conventional method for predicting the power generation output at each point P1 to Pn using an individual curve model has been devised to obtain the most probable output for the natural energy generator at each point P1 to Pn. However, satisfactory accuracy is not obtained for the total power generation output Xtot of the region predicted using this conventional method. In other words, there are many large gaps between the value obtained by summing the predicted values of the power generation outputs Xs1 to Xsn of the points P1 to Pn using the conventional method and the predicted value of the total power generation output amount Xtot in the region, In addition, there is uncertainty about the degree of separation, whether the separation is larger or smaller.

  For example, assume that the result of predicting the output of the solar power generator at the point P1 is Xs1. In this case, the predicted value of the total output Xtot of each of the P1 to Pn-th solar power generators is the same for external and internal factors such as local weather conditions, panel power generation output characteristics, panel orientation, etc. If there is, it should be approximately n × Xs1. This is because all curve models are almost the same.

  However, when the power generation output of the solar cell generator at the point P1 tends to be small due to external and internal factors, for example, when the power generation output of the solar cell generator at the point P1 is large, the result of the simulation by the above conventional method The total value of the power generation output predicted values Xs1 and Xs2 and the total power generation output in the area where the points P1 and P2 are totaled may be greatly separated. Furthermore, when the weather conditions vary and the relevance of the power generation output at each of the points P1 to Pn is poor, the total output of the area is smaller or larger than n × Xs1, how much is smaller It becomes unclear how large it will be.

  The cause is presumed to be uncertainty due to weather conditions. The predicted solar radiation intensity, wind direction, wind force, humidity, temperature, etc. for a region are not uniformly observed in the region, but change instantaneously. In some cases, a cloud smaller than the entire region passes through the region and casts a shadow on the solar panel locally and temporally. Therefore, although the power generation outputs Xs1 to Xsn predicted at the respective points P1 to Pn are to be expressed as probability distributions, the above-described conventional method that uniformly determines the most probable output is the power generation at each point P1 to Pn. It seems that the separation between the total of the predicted values of the outputs Xs1 to Xsn and the predicted value of the total power generation output amount Xtot in the region is increased, and the difficulty of predicting the degree and direction of the separation is increased.

  Therefore, it cannot be said that the conventional method is easy to use for the operator of the power system. There is an urgent need to devise a new idea and a method for realizing it in order to predict the total value of local power generation output or net power demand. Embodiments of the present invention have been proposed to solve such a problem, and an object thereof is to provide a power prediction method, a power device, and a power prediction program that are easy to use for an operator.

  An embodiment for achieving the above object is a power prediction method for calculating a predicted value of a total power generation output of a region where a plurality of natural energy generators are installed, and divides the region into a plurality of points. In addition, for each point, observation data having a correlation with the power generation output at that point is collected, the one area is divided into a plurality of points, and power data indicating the power generation output at each point And using the observed data, a correlation value indicating a correlation between the two points is calculated, and statistical analysis of the power data is performed while reflecting one or more correlation values. Thus, a predicted value of the total power generation output of the one region is calculated.

  Another embodiment is a power prediction method for calculating a predicted value of net power demand in a region where a plurality of natural energy generators are installed, wherein the region is divided into a plurality of points. For each point, observation data having a correlation with the power generation output or power demand at that point is collected, the one area is divided into a plurality of points, and the power generation output or power demand is determined for each point. Power data is collected, a correlation value indicating the correlation between the two points is calculated using the observation data, and the power data indicating the power generation output and the power data indicating the power demand are positive or negative. And calculating the predicted value of the net electric power demand in the one region by performing the statistical analysis that is mixed while being distinguished while reflecting one or more correlation values.

  Further, another embodiment is a power prediction method for calculating a predicted value of power demand for each time zone in a region where a natural energy generator is installed, the power data indicating the power generation output and power demand in the region. Collecting for each time zone, using the power data, calculating correlation values indicating the correlation between the power generation output and power demand in two different time zones, and performing statistical analysis of the power data of the power generation output , By calculating one or more of the correlation values related to the power generation output, to calculate a predicted value of the power generation output for each time zone of the region, and to perform a statistical analysis of the power data of the power demand By performing while reflecting one or more of the correlation values, calculating a predicted value of power demand for each time zone of the region, and adding the predicted value of power generation output and power demand for each time zone, Features.

  These embodiments can also be understood as a power prediction apparatus and an electronic prediction program.

The conceptual diagram which shows an example of the conventional electric power prediction method. It is a flowchart which shows the procedure of the electric power prediction method which concerns on this embodiment. FIG. 2 is a conceptual diagram showing a power prediction method according to the first embodiment. (A) is a figure which shows the past data Xp, (b) is a figure which shows distribution of the electric power generation amount of the point 1, (c) is a figure which shows distribution of the average electric power generation output of each point regarding Example 1. . (A) is a figure which shows the past data Xp, (b) is a figure which shows distribution of the electric power generation amount of the point 1, (c) is a figure which shows distribution of the average electric power generation output of each point regarding Example 2. . (A) is a figure which shows the past data Xp, (b) is a figure which shows distribution of the electric power generation amount of the point 1, (c) is a figure which shows distribution of the average electric power generation output of each point regarding Example 3. . (A) is a figure which shows the past data Xp, (b) is a figure which shows distribution of the electric power generation amount of the point 1, (c) is a figure which shows distribution of the average electric power generation output of each point regarding Example 4. . (A) is a figure which shows the past data Xp, (b) is a figure which shows distribution of the electric power generation amount of the point 1, (c) is a figure which shows distribution of the average electric power generation output of each point regarding Example 5. . (A) is a figure which shows the past data Xp, (b) is a figure which shows distribution of the electric power generation amount of the point 1, (c) is a figure which shows distribution of the average electric power generation output of each point regarding Example 6. . It is a figure which concerns on the modification and shows the approximation by the least square method of (sigma) i, j. (A) is a figure which shows the case where electric power demand and electric power generation output are made into the past data Xp, (b) is a figure which shows distribution of the electric power generation amount of the point 1, (c) is the sum total of each point. It is a figure which shows distribution of electric power generation output. (A) is a figure which shows the case where electric power demand and electric power generation output are made into the past data Xp, (b) is a figure which shows distribution of the electric power generation amount of the point 1, (c) is a total of each point, and relates to Example 8. It is a figure which shows distribution of electric power generation output. FIG. 6A is a diagram illustrating power demand in each time zone, and FIG. 5B is a diagram illustrating a demand scenario in each time zone according to the third embodiment. FIG. 7A is a diagram illustrating a power generation output in each time zone, and FIG. 5B is a diagram illustrating a power generation output scenario in each time zone according to the third embodiment. (A) is a figure which shows the net demand scenario of each time slot | zone concerning 3rd Embodiment, (b) is a figure which shows the frequency distribution of the net electric power demand at 12:00. According to the third embodiment, (a) is a diagram showing a temperature scenario in each time zone, (b) is a diagram showing a power generation output scenario in each time zone, and (c) is a frequency of net power demand at 12:00. It is a figure which shows distribution. It is a figure which shows the structure of the electric power prediction apparatus which concerns on embodiment. It is a figure which shows the network structure to which the electric power prediction apparatus was connected. It is a flowchart which shows operation | movement of an electric power prediction apparatus.

(concept)
First, the inventors greatly increase the predicted value and the actual measurement value of the total power generation output Xtot in the region by the conventional method of predicting the power generation outputs Xs1 to Xsn of the points P1 to Pn using individual prediction curve models. We have eagerly investigated the cause of many breaks. As a result, the inventors believe that, due to the characteristics of the conventional method, the points P1 to Pn do not affect each other, and independent events are scheduled to occur at the points P1 to Pn. Thought.

  That is, in the power prediction method according to the present embodiment, it is assumed that the power generation outputs at the points P1 to Pn have an influence on each other, and the correlation between the points P1 to Pn with the predicted value of the total power generation output Xtot in the region. Reflect. The correlation may be found directly from the power generation output, or may be other various elements having a correlation with the power generation output.

  Various other elements having a correlation with the power generation output are, for example, weather conditions, elements that are affected by weather conditions, or elements that produce differences in weather conditions. Examples of weather conditions are temperature, wind speed, humidity, precipitation, cloud cover, solar radiation, and the like. Examples of factors that are affected by weather conditions include the charge / discharge amount of a storage battery and the charge / discharge amount of an electric vehicle. A factor that produces a difference in weather conditions is, for example, the distance between two points.

  Specifically, as shown in FIG. 2, as a first step, the correlation coefficient between the observation data is calculated using the observation data at the points P1 to Pn. The observation data is correlated with the power generation output, such as the power generation output itself, power demand, temperature, wind speed, humidity, precipitation, cloud cover, solar radiation, etc., storage battery charge / discharge amount, electric vehicle charge / discharge amount, etc. Or a numerical value indicating the distance between two points.

  In this first step, if the numerical values between the observation data are the same, instead of calculating the correlation coefficient, it is possible to calculate and use the covariance that can represent the correlation in the same manner as the correlation coefficient. The result can be obtained. Values indicating correlation such as correlation coefficient and covariance are collectively called correlation values. The maximum number of correlation values can be calculated by n × (n−1) / 2, where n is the number of local points. In consideration of accuracy and processing load, correlation values may be calculated for all combinations of two points, or correlation values between one point may be uniformly applied between all points.

  Next, as a second step, the power data is statistically analyzed using the calculated correlation value, and the total power generation output Xtot in the region is predicted. The power data is past data of the power generation outputs Xp1 to Xpn at the points P1 to Pn.

  Examples of statistical analysis methods include single regression analysis, multiple regression analysis, least squares method, and neural network. In the regression analysis, the correlation value is used as an explanatory variable, and the local power generation output is used as an explanatory variable. The power generation output in the area can be obtained as a total value of past data. When a neural network is used, some correlation values are used as learning data. In general, since the correlation value at a point with a long distance is close to 0, only data at a point with a short distance may be used. In addition, weather data such as temperature may be added to these explanatory variables and learning data.

  For example, as a second step, a plurality of scenarios representing the power generation outputs Xs1 to Xsn at the points P1 to Pn are created. The scenario is collective data obtained by combining simulated values of the power generation outputs Xs1 to Xsn at the points P1 to Pn. The scenario is created by reflecting the correlation value between the points calculated in the first step and the statistical value derived from the past power generation output at the point Pi.

  Reflecting the correlation value means that each correlation value between each two points calculated from each value of the point Pi and each value of the point Pj included in each scenario is the correlation value calculated from the observation data. It means being equivalent or approximate. When one correlation value is uniformly applied, each correlation value between each two points calculated from each value of the point Pi and each value of the point Pj included in each scenario is the one correlation. Equivalent or approximate to the value.

  The statistical value derived from the past power generation output at the point Pi is an average value, standard deviation, distribution shape (kurtosis, skewness), etc., regarding the past power generation output Xpi at the point Pi. It means that the average value, standard deviation and distribution shape of each value of each point Pi included in the scenario are the same as or approximate to the average value and standard deviation derived from the past power generation output.

  In the case of the prediction method for creating a scenario, as a third step, the power generation outputs Xs1 to Xsn at the points P1 to Pn are totaled for each scenario, and the totals are collected into a frequency distribution. Thereby, the operator of the power system can obtain the predicted range and accuracy of the total power generation output in the region.

  This power prediction method can predict local net power demand and fluctuations in power demand for each local time zone without changing the calculation process, simply by changing the contents of observation data and power data. it can.

  That is, it is assumed that the power generation output and power demand at each point P1 to Pn have an influence on each other, and the correlation between the points P1 to Pn is reflected in the predicted value of the net power demand in the region. The net power demand is the demand that the main power supplier should respond to, which is the total power demand minus the output of natural energy power generation. In this case, the observation data and the power data are collective data in which the power generation output and the power demand at each of the points P1 to Pn are distinguished and mixed with each other.

  The correlation may be found directly from the power generation output and the power demand, or may be various other elements having a correlation with the power generation output and the power demand. Similarly, various other factors having a correlation with the power generation output or the power demand are factors that influence the weather condition, an element depending on the weather condition, or a difference in the weather condition.

  Also, assuming that the power generation output or power demand for each local time zone has an influence on each other, the predicted value of power generation output or power demand for each local time zone is correlated with the predicted value for each local time zone. reflect. The net power demand is the demand that the main power supplier should respond to, which is the total power demand minus the output of natural energy power generation. In this case, the observation data is power generation output and power demand data for each time zone in the area.

  Hereinafter, in the first embodiment, a further specific example will be described in which the past power generation output at each of the points P1 to Pn is used as observation data and power data, and the total power generation output in the region is predicted. 2nd Embodiment demonstrates the further specific example which estimates the net electric power demand of an area | region using the electric power generation output and electric power demand of each point P1-Pn as observation data and electric power data. In the third embodiment, a change in the daily power output in the region, a change in the daily power demand in the region, a change in the net daily power demand in the region, and the generated power output narrowed by weather conditions A further specific example of prediction of power demand will be described.

(First embodiment)
An example of calculating the predicted value of the total power generation output in the region will be described in detail with reference to the scenario creation method shown in FIG. A case where covariance is used instead of the correlation coefficient is taken as an example. The area to be calculated is divided in advance from points P1 to Pn by the number n of points. It is assumed that a natural energy generator of the same rating is installed at each of the points P1 to Pn.

  First, past data Xp = (Xp1, Xp2, Xp3,..., Xpn) indicating past power generation outputs at the respective points P1 to Pn is observed. In the past data Xp, Xpi indicates the past power generation output at the point Pi. This past data Xp is used as power data, and also as observation data used for calculating statistical information and calculating correlation values at the point Pi.

  The past data Xp is observed twice or more. This is for obtaining statistical information and an average value necessary for calculating a correlation value. The observation of the past data Xp is desirable in the same time zone on different days. Two or more sets of past data Xp represent the characteristics of each point Pi, that is, statistical information such as an average value, standard deviation, and distribution shape of past power generation output Xi at the point Pi is latent.

  Next, a correlation coefficient matrix (or variance-covariance matrix) between points is created using two or more sets of observed past data Xp. Moreover, the average value of the past power generation output Xpi at each point Pi is also calculated.

  Here, a variance covariance matrix Σ as shown in the following formula (1) is created. The variance-covariance matrix Σ has n × n components, var (Xi) is the variance of the past power generation output Xpi at the point Pi, and cov (Xi, Xj) is the past power generation at the points Pi and Pj. The covariance of the outputs Xpi and Xpj.

  Once the variance-covariance matrix Σ is calculated, next, a large number such that the correlation value between the points is approximately equal to the element value of the variance-covariance matrix Σ, which is equal to the statistical information for each point P1 to Pn. A set of random number data Xs is generated. If Xsi is a simulated power generation output at the point Pi, the random number data Xs = (Xs1, Xs2, Xs3,..., Xsn). The generation of a large number of sets of random number data Xs means that a large number of scenarios are generated.

  It is assumed that a large number of sets of random number data Xs = (Xs1, Xs2, Xs3,..., Xsn) follow an n-dimensional normal distribution. Then, the probability density function f (Xs, μ, Σ) satisfied by the large number of sets of random number data Xs can be expressed as in the following formula (2). Μ in the equation is set data having an average value μi of each point Pi that can be calculated from two or more sets of past data Xp as an element, and μ = (μ1, μ2, μ3,..., Μn). By generating a large number of sets of n random numbers according to the probability distribution of the following formula (2), n sets of scenarios are generated.

  Finally, the respective element values of the random number data Xs = (Xs1, Xs2, Xs3,..., Xsn) are summed to obtain the total power generation output Xtot for each scenario, and the frequency for each numerical section of the total power generation output Xtot By generating the distribution, the predicted value of the total power generation output Xtot can be obtained as a probability distribution.

(Correlation coefficient application example)
Next, a plurality of examples will be presented regarding a method for obtaining a correlation coefficient between two points. Examples 1 to 4 are examples equivalent to the case where the correlation coefficient between a pair of two points is uniformly applied between all two points, and are shown in FIGS. 4 to 7 respectively. Examples 5 and 6 are cases where all of the correlation coefficients between the two points are reflected in the scenario creation, and are shown in FIGS. 8 and 9, respectively.

Example 1
Condition: Number of points 100, correlation coefficient 0.9, number of scenarios 100
In Example 1, as shown to (a) of FIG. 4, about the area containing a total of 100 points P1-P100 where the natural energy generator which has the same rated output was installed, the correlation between each point 100 scenarios were generated with the number uniformly assumed to be 0.9. In the figure, a continuous line represents one scenario. And as shown to (c) of FIG. 4, the average value of the electric power generation output Xs1-Xs100 of the points P1-P100 was calculated | required for every scenario, and the average value was put together in frequency distribution.

  As a comparison object, as shown in FIG. 4B, the frequency distribution of each power generation amount Xs1 at the point 1 included in each scenario is summarized. In FIG. 4C, the average of the elements of the random number data Xs is obtained for each scenario and summarized in the frequency distribution for easy comparison with the frequency distribution of the power generation amount Xs1 at the point 1. When the average value is multiplied by n, the total power generation output Xtot is obtained.

(Example 2)
Condition: Number of points 100, correlation coefficient 0.3, number of scenarios 100
The second embodiment differs from the first embodiment in that the correlation coefficient is regarded as 0.3. That is, in Example 2, as shown to (a) of FIG. 5, about the area containing a total of 100 points P1-P100 where the natural energy generator which has the same rated output was installed, between each point. After assuming that the correlation coefficient was uniformly 0.3, 100 scenarios were generated. And as shown to (c) of FIG. 5, the average value of the electric power generation output Xs1-Xs100 of the points P1-P100 was calculated | required for every scenario, and the average value was put together in frequency distribution. Further, as a comparison target, as shown in FIG. 5B, the frequency distribution of each power generation amount Xs1 at the point 1 included in each scenario is summarized.

(Example 3)
Condition: Number of points 1000, correlation coefficient 0.9, number of scenarios 100
The third embodiment is different from the first embodiment in that an area including 1000 points is assumed, and the correlation coefficient applied uniformly is the same as that of the first embodiment. That is, in Example 3, as shown to (a) of FIG. 6, about the area containing a total of 1000 points P1-P1000 where the natural energy generator which has the same rated output was installed, between each point. After assuming that the correlation coefficient was uniformly 0.9, 100 scenarios were generated. And as shown to (c) of FIG. 6, the average value of the electric power generation output Xs1-Xs1000 of the points P1-P1000 was calculated | required for every scenario, and the average value was put together in frequency distribution. Further, as a comparison target, as shown in FIG. 6B, the frequency distribution of each power generation amount Xs1 at the point 1 included in each scenario is summarized.

Example 4
Conditions: 1000 points, 0.1 correlation coefficient, 100 scenarios
The fourth embodiment is different from the first embodiment in that an area including 1000 points is assumed and the correlation coefficient that is uniformly applied is set to 0.1. That is, in Example 4, as shown to (a) of FIG. 7, about the area | region which includes a total of 1000 points P1-P1000 where the natural energy generator which has the same rated output was installed, between each point. After assuming that the correlation coefficient was uniformly 0.1, 100 scenarios were generated. And as shown to (c) of FIG. 7, the average value of the electric power generation output Xs1-Xs1000 of the points P1-P1000 was calculated | required for every scenario, and the average value was put together in frequency distribution. Further, as a comparison target, as shown in FIG. 7B, the frequency distribution of each power generation amount Xs1 at the point 1 included in each scenario is summarized.

(result)
As can be seen from Example 1 in FIG. 4 and Example 3 in FIG. 6, when the correlation coefficient between the points is as large as 0.9, the frequency distribution of the power generation output Xp1 representing each point and The frequency distribution of the total power generation output Xtot in the region is similar.

  In the case of Example 1, the average power generation output Xs1 at point 1 was 6.07 MWh and the standard deviation σ was 0.96 MWh, whereas the average power generation output in the region was 6.09 MWh and the standard deviation σ was 0.00. 94 MWh. In the case of Example 3, the average power generation output XP1 at point 1 was 6.08 MWh and the standard deviation σ was 0.99 MWh, whereas the average power generation output in the region was 6.09 MWh and the standard deviation σ was 0.93 MWh.

  In contrast to Examples 1 and 3, as can be seen from Example 2 in FIG. 5 and Example 4 in FIG. 7, when the correlation coefficient between points is as small as 0.3 or 0.1, It can be seen that the prediction accuracy of the total power generation output Xtot in the region increases. That is, fluctuations between points are absorbed by averaging, and the standard deviation is reduced. In the case of Example 2, the average power generation output Xs1 at point 1 was 6.03 MWh and the standard deviation σ was 0.92 MWh, whereas the average power generation output in the region was 6.05 MWh and the standard deviation σ was 0.00. It became small at 55 MWh. In the case of Example 4, the average power generation output Xs1 at point 1 was 5.96 MWh and the standard deviation σ was 1.01 MWh, while the average power generation output in the region was 6.03 MWh and the standard deviation σ was 0.00. It was further reduced to 31 MWh.

  As described above, according to this power prediction method, if the correlation between the points P1 to Pn is known, the past total power generation output Xtot in the area can be substantially reproduced, and the total power generation output Xtot of the total power generation output Xtot can be reproduced from the reproduction result. The predicted value can be obtained with accuracy. Therefore, it is an easy-to-use power prediction method for the operator. In particular, when the correlation coefficient between the points P1 to Pn is small, the standard deviation of the total power generation output in the region is small, and the prediction accuracy is improved.

In Example 5 of FIG. 8 and Example 6 of FIG. 9, the correlation coefficient between the points P1 to Pn is obtained using the distance between the points as a variable, and each correlation coefficient is reflected in the scenario. That is, the correlation coefficient ρ _{i, j} between the point Pi and the point Pj is calculated by, for example, the following equation (3). In the following formula (3), i and j are point numbers. The point numbers are assigned in order of increasing distance from one reference point. abs (i−j) is the absolute value of i−j.

In the case of the point number i = j (the same point), the above equation (3) indicates that the correlation coefficient is 1, and the point number i and j are separated from each other and abs (i−j) increases. The relationship number ρ _{i, j} is set to be small. The correction constant D in the equation (3) is a parameter that determines the magnitude of the correlation. The larger the correction constant D is set, the closer the correlation coefficient ρ _{i, j} between short distances approaches 1.

(Example 5)
Condition: Number of points 100, correction constant D = 100, number of scenarios 100
In the fifth embodiment, as shown in FIG. 8A, the probability density function f (Xs) for an area including a total of 100 points P1 to P100 where the natural energy generators having the same rated output are installed. , Μ, Σ) were used to generate 100 scenarios. The correlation coefficient between each point was calculated by the above equation (3), and the correction constant D = 100. And as shown to (c) of FIG. 8, the average value of the electric power generation output Xs1-Xs100 of the points P1-P100 was calculated | required for every scenario, and the average value was put together in frequency distribution. Further, as a comparison target, as shown in FIG. 8B, the frequency distribution of each power generation amount Xs1 at the point 1 included in each scenario is summarized.

(Example 6)
Condition: Number of points 100, correction constant D = 10, number of scenarios 100
In the sixth embodiment, as shown in FIG. 9A, the probability density function f (Xs) is obtained for an area including a total of 100 points P1 to P100 where the natural energy generators having the same rated output are installed. , Μ, Σ) were used to generate 100 scenarios. The correlation coefficient between each point was calculated by the above formula (3). The difference from the fifth embodiment is that the correction constant D = 10. And as shown to (c) of FIG. 9, the average value of the electric power generation output of the points P1-P100 was calculated | required for every scenario, and the average value was put together in frequency distribution. Further, as a comparison target, as shown in FIG. 9B, the frequency distribution of each power generation amount Xsi at the point 1 included in each scenario is summarized.

(result)
As shown in Examples 5 and 6, when the correlation coefficient is obtained by reflecting the correlation between points, it can be seen that the prediction accuracy of the total power generation output Xtot in the region is considerably improved. In the case of Example 5, the average power generation output Xs1 at point 1 was 6.09 MWh and the standard deviation σ was 1.04 MWh, whereas the average power generation output in the region was 6.08 MWh and the standard deviation σ was 0.00. 84 MWh. In the case of Example 6, the average power generation output Xs1 at point 1 was 6.19 MWh and the standard deviation σ was 1.05 MWh, whereas the average power generation output in the region was 6.14 MWh and the standard deviation σ was 0.41 MWh.

  Furthermore, it can be seen that if the correction constant D is set to a small value so that the calculated correlation coefficient is small while observing the correlation between points, the prediction accuracy of the total power generation output Xtot in the region is further improved. . In the case of Example 5 with the correction constant D = 100, the standard deviation σ of the local average power output was 0.84 MWh, whereas in the case of Example 6 with the correction constant D = 10, the average power output of the region The standard deviation σ decreased to 0.41 MWh.

  As described above, according to this power prediction method, if the correlation value is calculated so as to more closely observe the correlation between points, a more accurate predicted value of the total power output Xtot can be obtained. In addition, if the correction constant of the calculation result is determined so that the calculated correlation value becomes small while observing the correlation between points, a more accurate predicted value of the total power output can be obtained.

(Modification)
The correlation coefficient ρ _{i, j} between the point Pi and the point Pj may be obtained by reflecting the relationship between the actually measured correlation coefficient and the distance between the points to improve accuracy. That is, the correlation coefficient ρ _{i, j} between the point Pi and the point Pj may be calculated by the following equation (4).

In the formula, d _{i, j} is an actually measured distance between the point Pi and the point Pj. Further, as shown in FIG. 10, the correction constant A and the correction constant D are obtained by fitting the measured correlation coefficient and the distance between points to the following equation (4) using the least square method.

(Second Embodiment)
Next, a power prediction method according to the second embodiment will be described. In the first embodiment, the contents of the observation data and the past data Xp are the power generation output, the random number data Xs is generated as the scenario of the power generation output, and the total power generation output Xtot of the region is obtained from the scenario. On the other hand, the power demand required by the load installed at each of the points P1 to Pn can be considered as a negative power generation output. If the power demand is positive, the power generation output can be considered as a negative power demand. In other words, the power generation amount prediction method presented in the first embodiment can change the contents of observation data and past data Xp to power demand, or simply mix power demand with observation data and past data Xp. This is a prediction method for predicting power demand.

  In the second embodiment, using the past data Xp in which the power demand and the power generation output of each point P1 to Pn are mixed as observation data and power data, random data Xs that is a scenario of net power demand is generated. The local total power demand Xtot is obtained. As a result, it is possible to easily predict the net power demand necessary for the main power supplier, which is obtained by subtracting the sales of natural energy generators installed in homes, buildings, and factories.

(Example 7)
In FIG. 11A, as Example 7, the power demand at each point P1 to P7 and the power generation output at each point P8 to P10 existing in the region are mixed in the past data Xp, and this past data Xp is observed data. The correlation between two points is calculated using the equation (1) as follows, and a large number of scenarios are generated using the probability density function f (Xs, μ, Σ) represented by the equation (2). The number of points is 10, the power demand is from point P1 to P7, and the power generation output is from point P8 to P10. In this calculation, in order to predict the demand amount, the demand amount is a positive value and the power generation output is a negative value. In Example 7, the correlation coefficient between each point was uniformly regarded as 0.9.

  And as shown to (c) of FIG. 11, the average value of value Xs1-Xs10 of the points P1-P10 was calculated | required for every scenario, and the average value was put together in frequency distribution. As a comparison target, as shown in FIG. 11B, the frequency distribution of each demand amount Xs1 at the point 1 included in each scenario is summarized.

  Note that the points P1 to P7 and the points P8 to P10 do not have to be different points. In some cases, both natural energy power generation and load are installed at one point.

(Example 8)
In (a) of FIG. 12, as Example 8, the electric power demand of each point P1-P7 which exists in an area and the power generation output of each point P8-P10 are mixed in the past data Xp, and Formula (1) is used. The correlation between the two points is calculated, and a large number of scenarios are generated using the probability density function f (Xs, μ, Σ) expressed by Equation (2). The number of points is 10. The points P1 to P7 are demand, and the points 8 to 10 are power generation outputs. In this calculation, in order to predict the demand amount, the demand amount is a positive value and the power generation output is a negative value. In Example 7, the correlation coefficient between each point was uniformly considered as 0.1.

  And as shown to (c) of FIG. 12, the total value of value Xs1-Xs10 of the points P1-P10 was calculated | required for every scenario, and the total value was put together in frequency distribution. As a comparison target, as shown in FIG. 11B, the frequency distribution of each demand amount Xs1 at the point 1 included in each scenario is summarized.

(result)
As shown in Example 7 in FIG. 11, even when the power demand and the power generation output are divided into positive and negative values and treated as data having the same content in the calculation, if the correlation coefficient is increased, the load shows the maximum value The power generation output is also close to the maximum value. Therefore, for example, if the temperature is high, it is possible to reproduce a general phenomenon that the amount of operation of the air conditioner increases, the weather is likely to be good, and the power generation output of solar power generation increases.

  As shown in FIG. 11 (c), the net power demand (total value) was a positive value, and the standard deviation was within 1.07 MWh with respect to the average of 2.69 MWh. On the other hand, as shown in FIG. 12C, it can be seen that if the correlation coefficient is small, the net power demand will be negative with a probability of about 20%. Therefore, when the correlation coefficient is high, it is possible to approximate the net demand amount by treating the power demand and the power generation output as data having the same contents in calculation.

(Third embodiment)
The electric power prediction method which concerns on 3rd Embodiment is shown. The specific calculation method is the same as in the first embodiment and the second embodiment, and detailed description thereof is omitted.

  As shown in FIG. 13 (a), if the concept of a point is replaced with a time zone, and the power demand in each time zone in the region is taken as observation data and power data, as shown in FIG. 13 (b), Multiple sets of scenarios are generated that show changes in daily demand in the region.

  On the other hand, as shown in FIG. 14 (a), if the concept of a point is replaced with a time zone, and the power generation output in each time zone in the region is taken as observation data and power data, as shown in FIG. 14 (b). In addition, a large number of sets of scenarios showing changes in the daily power output in the region are generated.

  Further, as shown in FIG. 15 (a), a large number of scenarios showing changes in daily demand shown in FIGS. 13 (b) and 14 (b) and changes in daily power output are shown. If a large number of scenarios are added together, a large number of net power demand scenarios are generated. At the time of addition, each element value of a large number of scenarios indicating changes in the daily power output is converted into a negative value.

  Then, as shown in FIG. 15B, by extracting the arbitrary time included in each scenario of the net power demand, for example, the value at 12 o'clock and collecting it in the frequency distribution, the net at that time The range and accuracy predicted as the power demand can be obtained.

  Further, as shown in FIG. 16, if the aperture is narrowed down by the weather condition, it is possible to calculate the power generation output and demand in the area under the weather condition.

  For example, the correlation between the temperature and the power generation output of the solar power generator is calculated in advance. And as shown to (a) of FIG. 16, by calculating the said Formula (1) and (2) using the temperature of each time slot | zone of a area, many sets which show the temperature change of the day in a area | region Generate a scenario. Furthermore, as shown in FIG. 16 (b), a large number of sets of scenarios indicating the power generation output of the solar power generator in the area for one day are generated.

  Next, using the correlation coefficient between the temperature and the power generation output of the solar power generator, each element value in which the temperature at 10:00 falls between 17.5 ° C. and 18.5 ° C. is extracted from each scenario of the temperature. . Furthermore, element values that fall within the range of the extracted element value and the correlation coefficient calculated in advance are extracted from each scenario of the power generation output of the solar power generator.

  And as shown in (c) of Drawing 16, when element values extracted from each scenario of the power generation output of a solar power generator are put together in frequency distribution, temperature will be 17.5 ° C-18.5 ° C at 10:00. The range that the predicted value of the power generation output can be taken with probability is obtained. From (c) of FIG. 16, when the temperature at 10:00 is the above, the solar power generation amount at 12:00 is an average of 8.55 kW, a standard deviation of 1.37 kW, and a distribution having a peak near the maximum output. You can see that

  As described above, when the past data Xp is the power demand and power generation output in each time zone, it is also possible to predict the net power demand in each time zone. Further, by using the temperature data and the power generation output of the solar battery generator as the past data Xp, it is also possible to predict the power generation output of the solar battery generator in an arbitrary temperature range in an arbitrary time zone.

(Constitution)
FIG. 17 shows a configuration of a power prediction apparatus that implements the method according to each embodiment as described above. The power prediction apparatus 1 is composed of a single computer or a plurality of computers connected to a network. The maximum number of data that can be handled by one PC at a time depends on the memory and calculation speed, but currently the upper limit is about 10,000 points. However, by dividing the area to be calculated into smaller areas and considering a total of about 1000 points as one point, about 1 million points can be handled by one PC. It is possible to use a separate computer for the calculation of each small area. When 100 PCs are used, about 100 million points can be handled with 1 million points × 100 units, so that the number of points is practically sufficient.

  The storage device such as the HDD or SSD of the power prediction apparatus 1 stores an OS and a power prediction program. The power prediction program is appropriately expanded in the RAM and processed by the processor, thereby obtaining the collection unit 11. The database 12, the correlation coefficient calculation unit 13, the statistical analysis unit 14, the evaluation calculation unit 15, and the output unit 16 function.

  The collection unit 11 includes a network adapter, and collects past data Xpi from a smart meter included in a generator or a load, for example, using a communication protocol such as TCP / IP. The past data Xpi collected from the generator is the power generation output, and the past data Xpi collected from the load is the power demand.

  An example of the configuration of a network to which the power prediction apparatus 1 is connected is shown in FIG. The power prediction apparatus 1 is connected to various generators 2 and loads 3 connected to a local power system through a network 4 capable of data communication. The generator 2 is, for example, a wind power generator, a solar power generator, or an electric vehicle. The load 3 is, for example, a building electrical facility, a household electrical facility, or a factory electrical facility. The collection unit 11 collects power generation output and power demand from these generators 2 and loads 3. Moreover, when predicting the power generation output and power demand for a specific weather, the collection unit 11 collects weather data from a server connected to another network.

  The database 12 mainly includes storage means such as an HDD and an SSD, and stores past data Xpi received by the collection unit 11. When storing the past data Xpi, the spot number i of the spot Pi where the past data Xpi is generated is associated with information for distinguishing the occurrence date and time. In the case where the power generation output and power demand at each time zone at one point are used as the past data Xpi, information for distinguishing the time number i from the occurrence date is associated with the past data Xpi.

  The correlation coefficient calculation unit 13 mainly includes a processor and calculates a correlation coefficient. Specifically, the past data Xpi stored in the database 12 is read, and the correlation coefficient between the two points is calculated. Here, for example, var (Xi) and cov (Xi, Xj) corresponding to each component of the number of points n × n are calculated in order to generate the variance-covariance matrix Σ of Expression (1).

  When the correlation coefficient is uniformly applied as an input value, the past correlation data Xpi and Xpj associated with two predetermined spot numbers i and j are read and the corresponding correlation coefficients are input. . When calculating the correlation coefficient based on the distance between two points according to the point number, Equation (3) is sequentially calculated while changing the value i and the value j from 1 to the number n of points. When calculating the correlation coefficient according to the distance between two points, information indicating the distance between the points i and j is stored in advance in the database, and the information is read and calculated. When the correlation coefficient is calculated using a temperature or the like different from the power generation output or power demand, the server providing the temperature or the like is accessed via the network and stored in the database 12.

  The statistical analysis unit 14 mainly includes a processor, and generates a large number of scenarios by statistical analysis. Specifically, the statistical information of the point Pi such as the average value and the standard deviation is calculated from the past data Xpi stored in the database 12, and a lot of random number data Xs on which the statistical information and the variance-covariance matrix Σ are projected are calculated. Generate a pair. When generating the random number data Xs, for example, the probability density function f (Xs, μ, Σ) of Expression (2) is calculated.

  The evaluation calculation unit 15 is configured mainly including a processor, calculates a total power generation output Xtot from each scenario, and summarizes it in a frequency distribution. That is, the evaluation calculation unit 15 calculates the total power generation output Xtot = Xs1 + Xs2 +... For each random number data Xs, and classifies each total power generation output Xtot corresponding to the magnitude of the value.

  The output unit 16 mainly includes a processor, a printer, and a display, and displays the frequency distribution collected by the evaluation calculation unit 15 on a paper medium or a screen. The output unit 16 displays graphs as shown in FIGS. 4 to 16 so as to be visually recognizable.

  The operation of the power prediction apparatus 1 is shown in FIG. As shown in FIG. 19, first, the collection unit 11 collects observation data and power data at points P1 to Pn via the network 4 (step S01) and stores them in the database 12 (step S02). Next, the correlation coefficient calculation unit 13 calculates a correlation value between the respective observation data by calculating Expression (1) using the observation data at the points P1 to Pn (Step S03).

  Then, the statistical analysis unit 14 creates a large number of scenarios by calculating Expression (2) using the correlation value and the statistical value indicated by the power data as input values (step S04). Then, the evaluation calculation unit 15 calculates a predicted value such as the total power generation output of the region from the scenario (Step S05), summarizes it in the frequency distribution (Step S06), and the output unit outputs the predicted value and the frequency distribution (Step S05). S07).

(Summary)
As described above, the power prediction apparatus 1 functions as the collection unit 11, the correlation coefficient calculation unit 13, the statistical analysis unit 14, and the evaluation calculation unit 15 by appropriately processing the power prediction program.

  The collection unit 11 collects observation data having a correlation with the power generation output for each point that divides one region, and calculates the power generation output for each point that divides one region when calculating the predicted value of the power generation output of the region. Collect power data indicating. When calculating the predicted value of power demand in one region, the collection unit 11 collects observation data having a correlation with the power generation output or power demand for each point that divides one region, and also points that divide one region Collect data showing power generation output and power demand every time.

  Further, when calculating the predicted value of the power generation output for each time zone at the point, the collection unit 11 collects observation data having a correlation with the power generation output at the point for each time zone, and indicates the power generation output at the point. Collect power data by time of day. Further, when the predicted value of the power demand for each time zone at the point is calculated, the collecting unit 11 collects observation data having a correlation with the power demand at the point for each time zone, and also indicates the power indicating the power demand at the point. Collect data by time of day.

  Then, the correlation coefficient calculation unit 13 calculates a correlation value indicating the correlation between the two points or the correlation between two different time zones using the observation data, and the statistical analysis unit 14 calculates the power data. Statistical analysis is performed while reflecting one or more correlation values. Thus, it is possible to calculate a predicted value of local power output, a predicted value of power demand in one region, a predicted value of power generation output for each time zone of a point, or a predicted value of power demand for each time zone of a point. it can.

  This predicted value is calculated by assuming that each of the local points P1 to Pn and each time zone at one point has an influence on each other by the introduction of the concept of the correlation value. Prediction accuracy can be increased.

  In addition, the correlation coefficient calculation unit 13 prepares a correction constant for reducing the value of the correlation value in advance, and calculates the correlation value using the observation data and the correction constant. As a result, it was confirmed that the prediction accuracy was dramatically increased. Further, the correlation coefficient calculation unit 13 determines a relational expression between the distance between the two points and the actually measured value of the correlation value by least square fitting, and calculates the correlation value by the least square fitting. Thereby, the correlation value can be obtained by approximating the actual measurement value more, and the prediction error can be reduced.

  In addition, the statistical analysis unit 14 generates a large number of random data that reproduces the average, standard deviation, and correlation value between each two points of the past data for each point, and generates a predicted value based on the large number of random data. Calculated. Furthermore, the statistical analysis unit 14 generates a large number of sets of random data that reproduces the average, standard deviation, and correlation values between two different time periods of the past data for each time zone at one point. The predicted value was calculated based on the above. As a result, the predicted values based on each random number data can be collected into the frequency distribution, and the range and probability that the predicted values can take can be calculated, which is very convenient for the main power supplier.

(Other embodiments)
In the present specification, an embodiment according to the present invention has been described. However, this embodiment is presented as an example, and is not intended to limit the scope of the invention. Specifically, the embodiment as described above can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. This embodiment and its modifications are included in the scope of the present invention and the gist thereof, and are also included in the invention described in the claims and the equivalent scope thereof.

  For example, the area is divided into the large area, the medium area, the small area, and the small area, the small area is defined as the medium area, and the medium area is defined as the large area. By considering the total power output of the middle region as the sum of the power output of each small region and the total power output of the large region as the sum of the power output of each middle region, it is the same in the large region, middle region, and small region. The total power output can also be calculated by

  Further, the distribution shape of each of the points P1 to Pn does not need to be a normal distribution as in the formula (2), and may be a distribution function such as a β distribution or an actually measured distribution shape, and a random number according to these distribution shapes may be used. It may be generated.

  Furthermore, the calculation method of the total value using the correlation value between the points and the correlation value between the time zones is not only the total power generation output of the region and the net power demand, It is possible to calculate and sum sales, other quantities related to electricity.

DESCRIPTION OF SYMBOLS 1 Electric power prediction apparatus 11 Collection part 12 Database 13 Correlation coefficient calculation part 14 Statistical analysis part 15 Evaluation calculation part 16 Output part 2 Generator 3 Load 4 Network

Claims (11)

A power prediction method for calculating a predicted value of total power generation output in a region where a plurality of natural energy generators are installed,
The one area is divided into a plurality of points, and for each point, observation data having a correlation with the power generation output at that point is collected,
Dividing the one area into a plurality of points, collecting power data indicating the power generation output for each point,
Using the observation data, calculate a correlation value indicating the correlation between the two points,
Performing a statistical analysis of the power data while reflecting one or more correlation values, thereby calculating a predicted value of the total power generation output of the one region;
A power prediction method characterized by the above. A power prediction method for calculating a predicted value of net power demand in a region where a plurality of natural energy generators are installed,
The one area is divided into a plurality of points, and for each point, observation data having a correlation with the power generation output or power demand at that point is collected,
Dividing the one area into a plurality of points, collecting power data indicating power generation output or power demand for each point,
Using the observation data, calculate a correlation value indicating the correlation between the two points,
By performing statistical analysis in which the power data indicating the power generation output and the power data indicating the power demand are mixed in a positive and negative manner while reflecting one or more correlation values, the net value of the one region is obtained. To calculate the predicted value of electricity demand for
A power prediction method characterized by the above. A power prediction method for calculating a predicted value of power demand for each time zone in a region where a natural energy generator is installed,
Collect power data indicating the power generation output and power demand in the region for each time period,
Using the power data, calculate a correlation value indicating each correlation between power generation output and power demand in two different time zones,
By performing statistical analysis of the power data of the power generation output while reflecting one or more of the correlation values related to the power generation output, a predicted value of the power generation output for each time zone of the region is calculated,
By performing statistical analysis of the power data of the power demand while reflecting one or more of the correlation values related to the power demand, a predicted value of the power demand for each time zone of the region is calculated,
Summing the predicted values of power generation output and power demand for each time period;
A power prediction method characterized by the above. The observation data is power generation output, power demand, temperature, wind speed, humidity, precipitation, cloud cover, solar radiation, storage battery charge / discharge amount, electric vehicle charge / discharge amount, or distance between two points,
The power prediction method according to claim 1 or 2. Preparing in advance a correction constant for reducing the correlation value as a whole, and calculating the correlation value using the observation data and the correction constant;
The power prediction method according to any one of claims 1 to 3. The observation data is a measured value of a distance between two points and a correlation value,
Determining a relational expression between the distance between the two points and the measured value of the correlation value by least square fitting, and calculating a correlation value by the relational expression;
The power prediction method according to claim 1 or 2. In the statistical analysis,
Generate multiple sets of random data that reproduces the average, standard deviation, and correlation values between each two points for power data for each point,
Calculating a predicted value based on the multiple sets of random number data;
The power prediction method according to claim 1 or 2. In the statistical analysis,
Generating a large number of sets of random data that reproduces the average, standard deviation, and correlation values between two different time zones of power data for each time zone at a single point;
Calculating a predicted value based on the multiple sets of random number data;
The power prediction method according to claim 3. Calculating a range that the predicted value can take by collecting the predicted values based on each random number data in a frequency distribution;
9. The power prediction method according to claim 7 or 8, wherein: A power prediction device that calculates a predicted value of total power generation output in a region where a plurality of natural energy generators are installed,
First collection means for collecting, for each point, observation data having a correlation with the power generation output at each point dividing the one area;
A second collecting means for collecting, for each point, power data indicating the power generation output at each point dividing the one area;
A calculation means for calculating a correlation value indicating a correlation between the two points using the observation data;
Statistical analysis of the power data is performed while reflecting one or more correlation values, statistical analysis means for calculating a predicted value of the power generation output of the one region,
Providing
A power prediction apparatus characterized by the above. A power prediction program for causing a computer to calculate a predicted value of power generation output in one area where a plurality of natural energy generators are installed,
Computer
First collection means for collecting, for each point, observation data having a correlation with the power generation output at each point dividing the one area;
A second collecting means for collecting, for each point, power data indicating the power generation output at each point dividing the one area;
A calculation means for calculating a correlation value indicating a correlation between the two points using the observation data;
Statistical analysis of the power data is performed while reflecting one or more correlation values, statistical analysis means for calculating a predicted value of the power generation output of the one region,
Make it work,
A power prediction program characterized by

Images