KR20140021179A - Method and apparatus for predicting daily solar radiation level - Google Patents

Abstract

A method for predicting a daily solar radiation level according to the embodiments of the present invention includes the steps of: providing a case database to express each case by a dependent variable including the daily solar radiation level according to the subdivisions of the seasons and a plurality of independent variables including geographical attributes and meteorological attributes measured according to the subdivisions of the seasons and a test case to which independent variable test values are given; dividing the meteorological attributes and the daily solar radiation level according to the subdivisions of the seasons in the cases of the case database into groups according to the subdivisions of the seasons; and estimating a daily solar radiation level prediction value based on a search result according to the subdivisions of the seasons searched from the group according to the subdivisions of the seasons in the case database with regard to the given independent variable test values. [Reference numerals] (AA) Grasp insolation properties; (BB) Meteorological observation document DB; (CC) Grouping by season; (DD) Prediction model; (EE) Insolation by season

Description

METHOD AND APPARATUS FOR PREDICTING DAILY SOLAR RADIATION LEVEL}

TECHNICAL FIELD The present invention relates to solar light utilization technology, and more particularly, to solar radiation prediction technology.

The United Nations Framework Convention on Climate Change (UNFCCC) was signed in 1992 and published in March 1994 as a result of several years of calls for global awareness of global warming. Since then, as part of efforts to reduce greenhouse gas emissions, the main culprit of global warming, many countries have implemented policies to disseminate renewable energy such as solar energy.

In many developed countries, there is a power generation differential support system, with a mandatory solar energy supply requirement ranging from 20% to 50% by each target year.

In the case of Korea, solar energy facilities have been actively spreading recently with an average annual growth rate of about 130%. Educational facilities, such as schools, are growing by about 800%.

Technically, unlike other renewables such as wind, wave and tidal power, which require large sites or facilities, energy production sites are far from electricity energy demands such as urban centers, and where efficiency is more difficult, solar energy In buildings, power energy can be produced in various forms and scales, solar power can be utilized, mass production can be achieved through modularization, and efficiency is greatly improved by large-scale investment and technology development.

Typically, photovoltaic power generation system is called PV (Photovoltaic) system. The development economics of PV systems are, in short, influenced by the way they cope with changes in the intensity of sunlight during the day, but are more dependent on the daily solar radiation (DSR).

In the past, efforts have been focused on improving PV materials, improving power conversion efficiency, improving power storage devices, and solar tracking devices, but in reality, the question of where to install the solar power generation system is This is more important than the increase in efficiency gained by improvements in these technologies.

In addition, if a site needs to install a photovoltaic power generation system, it is a problem of predicting the scale to be installed in order to obtain a target amount of power, or, conversely, predicting the amount of power that can be obtained by installing a specific scale. The problem can be raised. All of these problems require accurate knowledge of solar radiation to provide a solution.

Therefore, in order to maximize the economics of PV systems, the amount of solar radiation must be accurately identified.

However, solar radiation is greatly influenced by latitude, which determines the angle of incidence of sunlight, or weather conditions, which determine how much sunlight is absorbed in the atmosphere until it reaches the earth's surface.

In Korea, meteorological surveys are conducted at 24 weather stations, and monthly solar radiation figures for the past 30 years have been built into the database. However, insolation data from regions without meteorological stations are virtually impossible to obtain unless surveyed for a long time.

Accordingly, the necessity of predicting the amount of insolation in an area where there is no data on insolation has emerged. Several prediction methodologies have been applied to predict the amount of solar radiation. An artificial neural network (ANN) model, a geographic information system (GIS) model, and a satellite-based model have been proposed.

However, ANN modeling has good prediction accuracy, but it does not explain the basis of the prediction result. GIS model or satellite-based model is based on the map distance between the prediction area and the actual measurement area. It does not take into account the disadvantage of low reliability.

The problem to be solved by the present invention is to provide a solar radiation prediction method and apparatus that can explain the basis of the prediction result while having a high prediction accuracy.

Seasonal solar radiation prediction method according to an aspect of the present invention,

Provides test cases with independent variable test values, each with a case database represented as a plurality of independent variables containing geographic and seasonal observed meteorological properties, and a dependent variable containing seasonally observed solar radiation. Making;

Dividing the seasonal meteorological attributes and the seasonal insolation of the cases in the case database into seasonal clusters; And

And estimating a solar radiation forecast value per season based on a seasonal search result retrieved from a seasonal cluster in the case database with respect to the given independent variable test values.

In one embodiment, the search of the case database,

Determining a filtering range based on respective dependent variable predictions and error rates based on at least two or more quantitative prediction methodologies;

Optimizing at least parameters for Case Based Reasoning using a Gene Algorithm; And

Among search cases obtained from case-based reasoning based on optimized parameters, the method may include outputting search cases whose dependent variable values fall within the filtering range as similar cases.

In one embodiment, the at least two quantitative prediction methodologies can include MRA (multiple regression analysis), ANN (artificial neural network).

In one embodiment, the filtering range is

It may be a cross-range between a maximum of lower limit values of prediction ranges obtained from the at least two prediction methodologies and a minimum of upper limit values of prediction ranges obtained from the at least two prediction methodologies.

In one embodiment, the filtering range is

An enlarged crossover that extends an intersection range between a maximum of lower limits of prediction ranges obtained from the at least two prediction methodologies and a minimum of upper limits of prediction ranges obtained from the at least two prediction methodologies by a tolerance range. It can be a range.

In one embodiment, the at least parameters for case-based reasoning are:

Minimum criterion for scoring attribute similarity (MCAS), range of nth attribute weight (RAC), range of case selection (RCS) as parameters for case-based reasoning, and TRCRMA (Tolerance range of cross range) as parameters for determining filtering range between the predicted values of MRA and ANN models).

In one embodiment, the season may be any one of the month, season, semi-annual.

Seasonal solar radiation prediction apparatus according to another aspect of the present invention,

A case database in which each case is represented as a plurality of independent variables including geographic attributes and meteorological attributes observed per season and dependent variables including solar radiation observed per season;

A cluster division unit for dividing the seasonal meteorological attributes and the solar radiation in each season into the clusters by season; And

If a test case provided with independent variable test values is provided, the solar radiation prediction unit may estimate an insolation dependent variable predicted value for each season based on a result retrieved from the case database with respect to the given independent variable test values.

In one embodiment, the solar radiation prediction unit

A quantitative prediction analysis unit that calculates dependent variable prediction values, error rates, and prediction ranges with respect to the cases of the case database based on at least two or more quantitative prediction methodologies;

A case-based reasoning model for searching for cases similar to test cases using case-based reasoning among the cases of the case database;

A parameter optimizer which optimizes parameters for the case-based reasoning model using a genetic algorithm; And

And a filtering range that has a filtering range determined based on the dependent variable prediction values, the error rates, and the prediction ranges, and outputs similar cases by filtering the cases searched in the case-based reasoning model.

The at least two quantitative prediction methodologies may include MRA (Multiple Regression Analysis), ANN (Artificial Neural Network).

In one embodiment, the filtering unit

Determine a cross-range between the maximum of the lower bounds of the prediction ranges obtained from the at least two prediction methodologies and the minimum of the upper bounds of the prediction ranges obtained from the at least two prediction methodologies as the filtering range. It can work.

In one embodiment, the filtering unit

Filtering the enlarged crossover range by enlarging the crossover range between a maximum of the lower bounds of the prediction ranges obtained from the at least two prediction methodologies and a minimum of the upper bounds of the prediction ranges obtained from the at least two prediction methodologies by an acceptable threshold rate. Can be determined to determine the range.

In one embodiment, the season may be any one of the month, season, semi-annual.

According to the method and apparatus for predicting insolation of the present invention, it is possible to predict, with high prediction accuracy, insolation which shows various aspects influenced by the climatic / geographical characteristics.

According to the solar radiation prediction method and apparatus of the present invention, the average prediction accuracy and the standard deviation are greatly improved based on the clustering based on the monthly data, and the basis of the prediction result can be explained with the past cases.

According to insolation prediction method and apparatus of the present invention ensures a high prediction accuracy because of the large-scale power generation projects long-term energy costs that may _{occur:: (Life Cycle CO 2 LCCO2} ) (LCEC Life Cycle Energy Cost), environmental impact Uncertainty and errors can be reduced.

1 is a conceptual diagram schematically illustrating a method of predicting insolation according to an embodiment of the present invention.
2 is a flowchart specifically illustrating a method of predicting insolation according to an embodiment of the present invention.
3 is a flowchart illustrating specific steps of the hybrid case-based reasoning methodology applied in the solar radiation prediction method according to an embodiment of the present invention.
4 is a conceptual diagram illustrating a chromosome of a genetic algorithm (GA) in the method of predicting insolation according to an embodiment of the present invention.
5 is a block diagram illustrating an apparatus for predicting insolation according to an embodiment of the present invention.
6A through 6D are seasons illustrating a geographical distribution of actually measured solar radiation amount.
7 is a diagram illustrating a monthly distribution of measured solar radiation amount.

For the embodiments of the invention disclosed herein, specific structural and functional descriptions are set forth for the purpose of describing an embodiment of the invention only, and it is to be understood that the embodiments of the invention may be practiced in various forms, The present invention should not be construed as limited to the embodiments described in Figs.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. The same reference numerals are used for the same constituent elements in the drawings and redundant explanations for the same constituent elements are omitted.

1 is a conceptual diagram schematically illustrating a method of predicting insolation according to an embodiment of the present invention.

Referring to FIG. 1, the method of predicting insolation according to the present invention may first start from a process of identifying characteristics of insolation from various aspects based on insolation measurement data, weather data, and geographic information.

The solar radiation and other weather data are then collected and databaseed as case data. At this time, insolation and other weather data may be clustered on a monthly basis.

Next, the seasonal insolation of the test point can be predicted through case based reasoning on the clustered case data for each season, for example, monthly.

The predicted solar radiation can then be used to predict power production or plant size.

The characteristics of the solar radiation can be roughly identified as follows.

In Korea, various weather information including solar radiation is measured at 24 stations operated by the Korea Meteorological Administration. Using the weather analysis program ArcGIS 9.3, seasonal average values of insolation data for one year from March 2011 to February 2012 were analyzed at 24 stations in Korea, and the analysis results are represented on a map. 6d.

FIG. 6A is measured from March to May 2011, FIG. 6B is measured from June to August 2011, FIG. 6C is measured from September to November 2011, and FIG. 6D is measured from December 2011 to February 2012. The approximate seasonal average of the insolations.

The solar radiation map represented in Figs. 6A to 6D may be considered to be able to sufficiently predict the solar radiation with such a map. However, the solar radiation maps of FIGS. 6A to 6D are only contour maps which are mathematically analyzed by treating data of the stations as height values, and thus only a rough trend can be seen because the geographical requirements are not considered.

Since the Korean Peninsula belongs to a temperate climate with four distinct seasons, it is hot and humid under the influence of the southeast season in summer and cold and dry under the influence of the northwest season in the winter. In addition, the three sides are the sea, 70% of the country is mountains, and the topographical features of the ridges of the Taebaek Mountains extend from the north to the south. The combination of these climate / terrain features complicates the seasonal solar radiation characteristics of the Korean peninsula.

At first glance, insolation is likely to be most affected by latitude, but in reality it can be more affected by weather conditions as well as latitude. 6A to 6D, in the spring (FIG. 6A) and autumn (FIG. 6C), the average amount of solar radiation tends to increase toward the south, but in summer (FIG. 6C) and winter (FIG. 6D), the average amount of solar radiation in the southeast or east region. This seems to be much larger. Therefore, it is not enough to simply use latitude or geographic information for the prediction of insolation and play an important role in meteorological phenomena.

Table 1 below analyzes the correlation between insolation and major meteorological phenomena in relation to the insolation and meteorological data that are the basis of FIGS. 6A to 6D.

division Monthly average
Sunshine Monthly average
Cloud Monthly average
Temperatures Seasonal
Average
Insolation spring
(March-May) .445 -.341 .744 summer
(June to August) .857 -.804 -.132 autumn
(September to November) .195 .023 .735 winter
(December-February) .570 -.449 0.070

Referring to Table 1, it can be seen that the correlation between the solar radiation and each of the major meteorological phenomena varies greatly from season to season. In general, if the absolute value of the correlation coefficient is 0.5 or more, there is a strong correlation.

During spring, the correlation between insolation and mean temperature is high (0.744), while the sunshine and cloudiness are low. This correlation can be confirmed in FIG. 6a that the higher the temperature is, the greater the amount of sunshine.

In contrast, in summer, solar radiation showed a low correlation with the average temperature, but with a high correlation of 0.857 and 0.804, respectively. Summer on the Korean Peninsula is influenced by southeastern summer winds and shows high temperature and humid climates, and due to the rainy season, the sunshine rate decreases and the cloudiness affects the amount of insolation. This trend is confirmed in FIG. 6B.

In autumn, insolation again correlates well with the mean temperature (0.735). Because of the high sunshine rate and low cloudiness throughout the country, it has little effect on the amount of insolation.

In winter, solar radiation has a slightly higher correlation with sunshine (0.570) and is almost unaffected by temperature.

In this way, the correlation between the insolation and the main geographic information and the major meteorological factors were analyzed and its validity. As a result, various attributes affecting the average insolation are summarized as shown in Table 2 below.

variable property Explanation Independent variable Geographical
factor Hardness () ° Tokyo Latitude () ° North Altitude () m Meteorological
factor Monthly average sunshine rate ()% Monthly average cloud volume 1 to 10 Average monthly clarity days ( ) Work Monthly average temperature () ° C Monthly average relative humidity ()% Monthly average wind speed () m / s Dependent variable - Insolation () kWh / ㎡ / day

Referring to Table 2, there may be largely geographic and meteorological factors that affect solar radiation. Geographical factors include longitude, latitude, altitude, and meteorological factors such as sunshine, cloud cover, sunny days, temperature, relative humidity, and wind speed.

Topographic Factors Factors such as islands, basins, mountain ranges, and fog areas clearly affect solar radiation but cannot be nominally and quantified collectively, since the effects appear to be meteorological phenomena. These can be replaced by meteorological factors.

Using these influencing attributes as independent variables and solar radiation as the dependent variable, we can infer historical cases that are considered most similar to given independent variable test values, and from the solar radiation data of those historical cases, The amount of insolation can be predicted.

Below, based on a case database of meteorological observations collected over the past 10 years from meteorological stations in 15 regions, each month's observational data is taken according to the classification of independent and dependent variables in Table 2. Case-based reasoning was performed. The case database can consist of 15 regions * 12 months * 10 years = 1,800 cases.

On the other hand, the solar radiation data has been analyzed as having a seasonal characteristics, based on the analysis results, it can be considered an efficient clustering method of the solar radiation data.

To this end, a brief look at Figure 7, Figure 7 is a diagram illustrating a monthly distribution of the measured solar radiation.

In FIG. 7, a distribution of data is visually determined by using a schematic diagram called a boxplot. Box plots can visually display the distribution of data with five statistical indicators: minimum observations, low quadrants, median, high quadrants, and maximum observations. As is apparent from FIG. 7, the amount of insolation shows a big difference from month to month, and it shows that periodicity is repeated every month.

Statistical validity of these monthly cycle characteristics can be supported by time series analysis. A time-series analysis of the measurements shows that the absolute value of the t-statistics for the autocorrelation coefficient of time lag 12 at the 5% significance level is The warning level is higher than 1.25, indicating that there is seasonality in the time series. In other words, the observed data of 15 regions show seasonality with a 12-month time difference, and it can be determined that there is a monthly cycle characteristic.

Such monthly cycle characteristics can also be effectively used in the present invention as follows. According to an embodiment, such periodicity is not only a monthly cycle but also a spring / summer / autumn / winter season cycle or other summer / winter cycle, so that it is a period of a certain period, for example, a monthly cycle, a seasonal cycle or a semiannual cycle. May be used.

2 is a flowchart specifically illustrating a method of predicting insolation according to an embodiment of the present invention.

Referring to FIG. 2, the method of predicting insolation firstly, in step S21, examples of test case projects, for example, annual insolation, may be defined in various aspects including target attributes for the purpose of prediction. Each of them is identified and the correlation between these various attributes is analyzed.

For example, as illustrated in Table 1, Table 2, FIGS. 6A-6D and 7, analyze geographic attributes and climatic attributes, select geographic and climatic attributes highly correlated with insolation, Periodic characteristics of solar radiation by season, for example monthly, seasonal, semi-annual, etc. can be identified.

In step S22, in step S21, specific geographic and seasonal climatic attributes are set as independent variables, and solar insolation for each season is set as dependent variables, and past seasonal actual cases are represented as independent variables and dependent variables. You can build a case database and prepare a test case in which test values of independent variables are given seasonally about where you want to predict the amount of insolation.

For example, the longitude, latitude, altitude, etc. of Table 2, monthly average sunshine rate, cloud cover, sunny days, temperature, relative humidity, wind speed, etc. may be independent variables, and the monthly average solar radiation amount to be predicted may be a dependent variable.

In this case, actual cases of the case database may include at least independent variable values, such as monthly average sunshine rate data, and may include solar radiation dependent variable values.

The test case for the location where the solar radiation is to be predicted should be provided monthly for the test values, eg measured values or estimates, for all independent variables except solar radiation. In this case, the longitude, latitude, and altitude can be obtained from geographic data of the National Geographic Institute, for example, and the monthly average sunshine rate, cloud cover, cheonmyeong days, temperature, relative humidity, and wind speed can be provided from, for example, observation data of the Korea Meteorological Agency.

We can now drive case-based reasoning (CBR) -based prediction models to predict solar radiation dependent variables.

At this time, as described above, the cases in the case database have periodicity by season, and by using these features, clustering cases by season and searching for similar cases by season can more accurately predict solar radiation dependent variables by season.

Therefore, in step S23, the cases of the case database are clustered by season. Depending on the embodiment, such clustering may be performed not only monthly but for example seasonally or semi-annually.

In step S24, the solar radiation dependent variable prediction value is estimated based on the results retrieved from the case database with respect to the seasonal independent variable test values of the given test case. The specific process of searching the case database will be described in detail later with reference to FIGS. 3 and 4.

Specifically, given the test values of the independent variables in the test case, first, the case-based reasoning methodology searches the clusters for each season in the case database and extracts similar cases by season.

Depending on the embodiment, the extraction of such similar cases may be carried out not only monthly but for example seasonally or semi-annually.

Subsequently, the solar radiation dependent variable predicted value of the test case is estimated for each season from the extracted solar radiation dependent variable values of the similar cases for each season. Estimation of the solar radiation dependent variable predicted value may be calculated as a statistical representative value of the arithmetic mean, median, mode, weighted average, etc. of the extracted solar radiation dependent variable values of similar cases.

According to an embodiment, the estimation of the solar radiation dependent variable prediction value may be performed not only monthly but for example seasonally or semi-annually.

For example, the test case is to predict the amount of insolation to introduce a PV system at an elementary school in Seongbuk-gu, Seoul, 126.71 degrees long, 37.39 degrees north latitude, and 86m above sea level. Monthly weather information for the region is, for example, October, with an average monthly sunlight rate of 62% in October, monthly average cloudiness (minimum 1, maximum 10) of 3.8, monthly average Qingming Day 9, monthly average temperature 14.2 degrees, monthly average The relative humidity is 55% and the monthly average wind speed is 2.1 m / s.

Similar instances of October retrieved from the case database by the case-based reasoning methodology for the October test values of these test cases can be illustrated in Table 3 below.

variable Case number Hardness
(° C) Latitude
(N) Altitude
(m) Monthly average
Sunshine
(%) month
Average
Cloud
(1 to 10) Monthly average
Clarity
(Work) month
Average
Temperatures
(° C) month
Average
opponent
Humidity
(%) month
Average
Wind speed
(m / s) Insolation
(kWh /
m ² / day) Test case - 126.71 37.39 86 62.0 3.8 9 14.2 55 2.1 3.50 Case Study 1 80 126.57 37.36 86 55.8 3.9 13 14.5 61 2.2 3.42 Search case 2 72 126.57 37.36 86 42.1 3.7 11 12.8 60 2.0 3.50 Search case 3 70 126.29 36.46 26 57.0 3.9 10 13.7 73 2.4 3.57 Case Study 4 139 127.44 37.54 77 54.2 4.0 6 13.3 74 0.9 3.50 Search case 5 2 128.53 37.45 26 58.1 3.8 10 14.0 55 2.9 3.47 Search case6 24 128.37 35.53 58 60.8 3.7 15 15.7 59 2.0 3.56

From the monthly mean insolation values of six similar cases, the monthly mean insolation of test cases can be calculated from statistical representative values such as arithmetic mean, median, mode, weighted mean, etc. When calculated as an arithmetic mean, the monthly mean insolation estimates for the test cases can be estimated to be 3.50 kWh / m 2 / day.

The maximum generation amount per unit area can be predicted from the solar radiation prediction value and the facility performance value. If the target power generation is fixed, the area of the PV power plant can be calculated. If the site area is fixed, the maximum generation of the entire power plant can be predicted.

In an optional step (S25), it is possible to determine whether or not to invest in the power generation equipment based on the above-expected insolation forecast, it is possible to predict the size and economic feasibility of the equipment to be constructed.

The following is a brief description of the case-based reasoning methodology. The case-based reasoning methodology applied in the present invention is another quantitative prediction technique that generally shows high accuracy and low deviation in order to improve low accuracy and large deviation, which are general disadvantages of case-based reasoning, while maintaining the advantage of presenting evidence through cases. It is a hybrid case-based reasoning methodology (Hybrid CBR) that combines multiple regression analysis (MRA), artificial neural network (ANN), and genetic algorithm (Genetic Algorithm).

In general, case-based reasoning (CBR) is a methodology for selecting project cases with high similarity through comparative analysis between projects based on project characteristics. The core of the CBR methodology is attribute similarity (AS), attribute weight (AW), and case similarity (CS). The case similarity CS between any real case and the test case may be quantified as the sum of the products of the attribute similarities AS and the attribute weights AW.

Where AS is attribute similarity, AW is attribute weight, CS is case similarity, n is the number of attributes, and m is the number of cases.

Attribute similarity (AS) is based on the difference between the same kind of independent variables between the actual case and the test case for each project characteristic, that is, each of the independent variables (which in some cases may also include dependent variables). Is calculated. If the data of the independent variable is a nominal, the attribute similarity is 1 if they are identical to each other and 0 otherwise. If the data of the independent variable is numeric, the attribute similarity AS may be expressed as in Equation 2 below.

f _AS means attribute similarity calculation formula, AV _{Test_Case} is a specific attribute value of a test case, and AV _{Retrieved_Case} is a specific attribute value of a retrieved case. f _AS is closer to 100% as the attributes are similar. MCAS is the minimum criterion for scoring the attribute similarity, and the attribute similarity value remains valid only if the attribute similarity is higher than MCAS, otherwise the value is discarded. MCAS may be set to 10%, for example, and may be optimized as described below.

The case similarity (CS) may quantify how similar the test case is to each search case by using attribute similarity and attribute weight of all independent variables between the test case and the search case, as shown in Equation 3 below.

f _CS is a function for calculating case similarity (CS), f _AW is a function or result for calculating attribute weight (AW), f _AS is a function or result for calculating attribute similarity (AS), n is the number of attributes. The attribute similarity AS is calculated as in Equation 2, and the attribute weight AW may be empirically determined or mathematically optimized through other optimization methods. Case similarity CS may be illustrated as a value obtained by dividing a weighted attribute similarity obtained by multiplying the attribute similarity by the attribute weight divided by the sum of the attribute weights.

This general case-based reasoning methodology provides similar cases with high case similarity, and can produce prediction results through statistical processing of similar cases. Thus, the CBR methodology not only predicts the results but also suggests similar cases to support them, thus explaining how such predictions were obtained. Accumulation of cases can also improve prediction accuracy.

However, the general CBR methodology tends to be relatively less accurate than other quantitative prediction methodologies, such as methodologies such as MRA (Multiple Regression Analysis) and ANN (Artificial Neural Network).

On the other hand, quantitative prediction methodologies such as MRA and ANN cannot provide similar cases based on high prediction accuracy.

In order to overcome these problems, the inventors devised a hybrid CBR methodology in which MRA, ANN, and GA (Gene Algorithm) are combined. According to the hybrid CBR methodology devised in the present invention, it is possible to provide all the advantages of each methodology such as high prediction accuracy of MRA and ANN and similar case evidence presentation of CBR.

FIG. 3 is a flowchart illustrating specific steps of a hybrid case-based reasoning methodology applied in a method of calculating an optimal input resource of a test case project taking an example of calculating an optimal scale of a highway rest area according to an embodiment of the present invention.

Referring to FIG. 3, in step S31, at least two or more quantitative prediction methodologies, for example, MRA and ANN, are applied to m cases in a case database, so that the predicted values PV of the dependent variable are applied. Calculate.

In step S32, the error rate of the dependent variable prediction, e.g., the mean absolute error percentage MAPE, from each difference between the actual values AV of the dependent variable and the previously calculated predictions PV in m cases : mean absolute percentage error) can be calculated as in Equation 4.

Where f _MAPE is a function to calculate MAPE, AV is the actual value of the dependent variable, PV is the predicted value of the dependent variable, and m is the number of cases.

Since the error rates obtained are based on a large number of real cases, it is reasonable to assume that the same methodology will produce the same or similar error rates for new test cases. Also, since the error rate of the MRA or ANN methodology is usually smaller than the error rate of the CBR methodology, if the actual value of the case found to be similar to the CBR methodology is outside the range of the predicted value considering the error of the MRA or ANN methodology, the so-called prediction range, It is desirable to remove CBR cases from similar case candidates, and in this specification, the setting of filtering ranges based on these prediction ranges and the removal of inappropriate CBR cases using the same are called filtering. The specific filtering operation is described in the steps below.

In step S33, the respective dependent variable predictions are calculated for the test case based on at least two or more quantitative prediction methodologies, for example MRA and ANN.

In step S34, based on the error rate MAPE calculated in step S32 and the prediction value PV of the dependent variable calculated in step S33, the prediction range in which the actual value may exist in each prediction methodology ( PR) are respectively calculated.

Specifically for example, it is possible to define the predicted range (PR _ANN) in the prediction range (PR _MRA) and ANN methodology in MRA methods as in the following equation (5).

In Equation 5, each prediction range PR is defined to have an upper limit and a lower limit by an average absolute error percentage MAPE around each prediction value PV, but an upper limit and a lower limit may be defined by other parameters according to embodiments. It may be.

In step S35, a filtering range is determined based on at least two or more quantitative prediction methodologies, for example, a prediction range such as MRA and ANN. Here, the filtering range is applied to the filtering of the search cases to be performed later in step S37.

Specifically, for example, the prediction range of the _MRA (PR _MRA ) and the prediction range of the _ANN (PR _ANN ) may be slightly different, and the range where the prediction ranges of the two methodologies overlap each other, that is, the cross-range, is represented by the following equation. It can be determined as the filtering range as shown in 6 and 7.

The filtering range (CRMA: cross-range between predicted values of MRA and ANN models) of Equation 6 is between the maximum of the lower limits of the MRA's predicted range (PR _MRA ) and the ANN's predicted range (PR _ANN ) and the minimum of the upper limits. By way of example, it may be specified as a cross range of.

The extended filtering range CRMA * of Equation 7 may be exemplarily specified as an extended crossing range enlarged by a tolerance range of CRMA (TRCRMA) for the filtering range CRMA of Equation 6.

This specified filtering range can be applied to cases retrieved through the CBR methodology for the test case.

Depending on the embodiment, in step S33, the prediction range itself of one CA estimation methodology may be used as a filtering range and applied to search cases.

In step S36, which may be optional, at least a plurality of parameters for case-based reasoning are optimized using genetic algorithm.

Referring to FIG. 4 for a while, FIG. 4 is a conceptual diagram illustrating a chromosome of a genetic algorithm (GA) in a method for calculating an optimal input resource of a test case project using an example of calculating an optimal scale of a highway rest area according to an embodiment of the present invention. .

According to an embodiment, the chromosomes of the genetic algorithm may include parameters for case-based reasoning, such as minimum criterion for scoring attribute similarity (MCAS), range of nth attribute weight (RAW), range of case selection (RCS), Parameters for determining the filtering range may include, for example, a tolerance range of cross range between predicted values of MRA and ANN models (TRCRMA).

For example, if MCAS is too high, cases of low similarity only in certain attributes but high similarity in other attributes may be dropped. For example, if the RAW is too high, the impact of certain attributes may be overestimated. If the RCS is too small, fewer cases may be retrieved, making it difficult to calculate the predicted value of the dependent variable or present past cases. Too large TRCRMA can lead to poor prediction accuracy.

Thus, genetic algorithms track changes in population over several generations by selecting, breeding, mutating and replacing chromosomal individuals with these parameters as genes with the highest predictive accuracy. And, as the chromosomes belonging to the largest population, optimized parameters can be presented.

Referring back to FIG. 3, in step S37, search cases are extracted through case-based reasoning, and in step S38, only search cases in which dependent variable values are among the extracted search cases are within a filtering range are filtered. You can output it as a similar case.

According to an embodiment, the case-based reasoning of step S37 may be performed, for example, by applying empirical parameters, without going through the parameter optimization of step S36.

5 is a block diagram illustrating an apparatus for predicting insolation according to an embodiment of the present invention.

Referring to FIG. 5, the solar radiation estimation apparatus 50 is an apparatus for implementing the solar radiation estimation method of the present invention described with reference to FIGS. 1 to 4, 6, and 7, and includes a case database 51 and a cluster division unit ( 52, the solar radiation hybrid prediction unit 53 may be included.

The case database 51 stores measurement cases expressed as values of independent variables and dependent variables from seasonal weather observation data that have been observed for years. In this case, the measured cases may be stored.

The cluster divider 52 divides the cases in the case database 51 into seasons by season, for example, monthly, seasonal, semi-annual, and clusters. Clusters each divided by the cluster divider 52 are provided to the solar radiation hybrid predictor 53.

The solar radiation hybrid prediction unit 53 searches for cases similar to the test cases having independent variable test values of the given geographic and seasonal weather observation data among the cases of the case database 51, and depends on the solar radiation by season of the similar cases found. Estimates the seasonal insolation dependent variable predictions for the test cases based on the variable values.

In detail, the solar radiation hybrid prediction unit 53 includes quantitative prediction analysis units 531 which each perform a plurality of different quantitative prediction methodologies, a case-based reasoning model 532, and error rates and predictions of the respective quantitative prediction methodologies. A parameter optimizer 533 which filters the search results of the case-based reasoning model 532 and the parameters of the case-based reasoning model 532 and / or the filter unit 533 with a filtering range determined from the range. 534 may each include.

The quantitative prediction analyzer 531 performs quantitative prediction through, for example, MRA and ANN methodologies.

The filtering unit 533 may, for example, have an error rate and a prediction range of MRA and ANN methodologies, optionally further have a tolerance parameter, determine a predetermined filtering range, and use the filtering range of the case-based reasoning model 532 as a filtering range. Filter the search results.

In this case, the filtering unit 533 intersects the maximum of the lower limit values of the prediction ranges obtained from the at least two prediction methodologies in the quantitative prediction analysis unit 531 and the minimum of the upper limit values of the prediction ranges obtained from the at least two prediction methodologies. The range may be determined as the filtering range, or the enlarged cross range in which such an intersecting range is enlarged by an allowable threshold may be determined as the filtering range.

The parameter optimizer 534 may optimize various parameters, such as various parameters of the case-based reasoning model or tolerance parameters of the filter unit, for example, through a genetic algorithm.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, Modification is possible. Accordingly, the spirit of the present invention should be understood only in accordance with the following claims, and all of the equivalent or equivalent variations will fall within the scope of the present invention.

Further, the apparatus according to the present invention can be implemented as a computer-readable code on a computer-readable recording medium. A computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer system is stored. Examples of the recording medium include a ROM, a RAM, an optical disk, a magnetic tape, a floppy disk, a hard disk, a nonvolatile memory, and the like, and a carrier wave (for example, transmission via the Internet). The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

50 Insolation Prediction Devices 51 Case Database
52 Cluster Division 53 Insolation Hybrid Prediction
531 Quantitative Predictive Analytics 532 Case-based Reasoning Model
533 Filtering section 534 Parameter optimization section

Claims (13)

Provides test cases with independent variable test values, each with a case database represented as a plurality of independent variables containing geographic and seasonal observed meteorological properties, and a dependent variable containing seasonally observed solar radiation. Doing;
Dividing the seasonal meteorological attributes and the seasonal insolation of the cases in the case database into seasonal clusters; And
And estimating a solar radiation forecasting season based on a seasonal search result retrieved from a seasonal cluster of the case database with respect to the given independent variable test values. The method of claim 1, wherein the search of the case database,
Determining a filtering range based on respective dependent variable predictions and error rates based on at least two or more quantitative prediction methodologies;
Optimizing at least parameters for Case Based Reasoning using a Genetic Algorithm; And
And outputting, as similar cases, search cases whose dependent variable values are among the search cases obtained from the case-based reasoning based on the optimized parameters as similar cases. The method of claim 2, wherein the at least two quantitative prediction methodologies include MRA (multiple regression analysis) and ANN (artificial neural network). The method of claim 2, wherein the filtering range is
Characterized in that it is a cross-range between a maximum of lower limits of prediction ranges obtained from the at least two prediction methodologies and a minimum of upper limits of prediction ranges obtained from the at least two prediction methodologies. Insolation forecasting method. The method of claim 2, wherein the filtering range is
An enlarged crossover that extends an intersection range between a maximum of lower limits of prediction ranges obtained from the at least two prediction methodologies and a minimum of upper limits of prediction ranges obtained from the at least two prediction methodologies by a tolerance range. Seasonal solar radiation prediction method characterized in that the range. The method of claim 5, wherein the at least parameters for case-based reasoning are:
Minimum criterion for scoring attribute similarity (MCAS), range of nth attribute weight (RAC), range of case selection (RCS) as parameters for case-based reasoning, and tolerance range of cross range as parameters for determination of filtering range. Seasonal solar radiation forecasting method comprising between the predicted values of MRA and ANN models. The method according to claim 1, wherein the seasons are each one of the month, season, semi-annual, solar radiation forecasting method characterized in that one. A case database in which each case is represented as a plurality of independent variables including geographic attributes and meteorological attributes observed per season and dependent variables including solar radiation observed per season;
A cluster division unit for dividing the seasonal meteorological attributes and the solar radiation in each season into the clusters by season; And
If the test case is provided with the independent variable test values, the solar radiation prediction unit including a solar radiation prediction unit for estimating the solar radiation dependent variable prediction value for each season based on the results retrieved from the case database with respect to the given independent variable test values . The method according to claim 8, wherein the solar radiation prediction unit
A quantitative prediction analysis unit that calculates dependent variable prediction values, error rates, and prediction ranges with respect to the cases of the case database based on at least two or more quantitative prediction methodologies;
A case-based reasoning model for searching for cases similar to test cases using case-based reasoning among the cases of the case database;
A parameter optimizer which optimizes parameters for the case-based reasoning model using a genetic algorithm; And
And a filtering unit configured to output similar cases by filtering the cases found in the case-based reasoning model with a filtering range determined based on the dependent variable prediction values, error rates, and prediction ranges. The apparatus of claim 9, wherein the at least two quantitative prediction methodologies include MRA (multiple regression analysis) and ANN (artificial neural network). The method of claim 9, wherein the filtering unit
Determine a cross-range between the maximum of the lower bounds of the prediction ranges obtained from the at least two prediction methodologies and the minimum of the upper bounds of the prediction ranges obtained from the at least two prediction methodologies as the filtering range. Seasonal solar radiation prediction apparatus characterized in that the operation. The method of claim 9, wherein the filtering unit
Filtering the enlarged crossover range by enlarging the crossover range between a maximum of the lower bounds of the prediction ranges obtained from the at least two prediction methodologies and a minimum of the upper bounds of the prediction ranges obtained from the at least two prediction methodologies by an acceptable threshold rate. Seasonal solar radiation prediction apparatus characterized in that the operation to determine the range. The apparatus of claim 8, wherein the seasons are any one of a month, a season, and a half year.

Images