KR20220065556A - Method for building an optimal linear model of production and environment of smart farm - Google Patents

Abstract

Disclosed is a method for building an optimal linear model of production volume and environment of a smart farm. According to the present invention, the method for building the optimal linear model of production volume and environment of the smart farm comprises: a step of collecting production volume data and environment data of the smart farm; a step of using the collected production volume data and the environment data, and calculating the correlation coefficient of the production volume of the current week and the environment factors of the previous week, and calculating the time difference by using the calculated correlation coefficient; a step of using the calculated time difference and calculating the initial corresponding week (CW); a step of improving the calculated initial CW; and a step of using the improved CW generated through the improvement and calculating a multiple linear regression model. The present invention aims to provide a method capable of building an optimal multiple linear regression model.

Description

{Method for building an optimal linear model of production and environment of smart farm}

The present invention relates to a method of constructing an optimal linear model for the production and environment of a smart farm.

Recently, by constructing a smart farm through the convergence of agriculture and ICT, they are focusing on improving the productivity and quality of crops using big data and Internet of Things technology. By using IoT technology to measure and monitor the environmental information of crops in real time from sensors, and to automatically manage crops through the establishment of an optimal growth management system, productivity and quality have dramatically increased.

In Japan, using greenhouse information collected through sensors such as temperature, humidity, and illuminance at plant factories, the tomato growth environment is controlled with scheduling software, and the industrialization of agriculture is achieved through the optimal cultivation environment with artificial light applied.

In the case of the Netherlands, since the establishment of facility horticulture in 1945, productivity has been improved through the development of a precise and systematic environmental control system and automation technology. Although the export of vegetables from the Netherlands amounted to 4.2 billion euros in 2010, among the cultivated area of horticultural crops in the Netherlands, the cultivated area of facility vegetables accounted for only 3.4% (5,041 ha) in 2011. In the Netherlands, despite poor weather conditions such as insufficient sunlight and insufficient labor, there are glass greenhouse farms with a tomato production of 70 kg/m ² , and has been ranked second in the world tomato export market since the late 2000s.

On the other hand, in the case of Korea, the supply and spread of smart farms are being led by the government. Currently, smart farms are in the early stages of introduction, so research on optimal environment setting models for crop growth using large amounts of environmental information automatically collected through sensors in smart farms is insignificant.

Research on the growth of crops related to the existing meteorological climate is being conducted using the experimental method and modeling method on samples. The experimental method is limited to results within an appropriate reference range for various environmental factors such as temperature, humidity, light, irrigation, and nutrient solution. Since the modeling method does not lead to the derivation of the optimal environment setting model for crop growth, research on the growth environment model focusing on smart facility horticulture is required. Recently, research is being conducted to analyze data through statistical techniques that derive correlations between variables using automatically measured environmental data and production data in smart farm farms, and to derive optimal tomato growth environmental factors for smart farms. there is.

However, there is a limit to analyzing environmental factors of crops such as paprika and tomatoes through the lateral data analysis method. This is because the yield and environmental information are matched without time difference. If the data are analyzed without considering the appropriate time difference, the influence of environmental factors affecting crop growth cannot be analyzed. Therefore, in order to closely analyze the change in yield according to environmental factors, it is necessary to consider the time difference in which the effect of the environmental setting occurs.

Republic of Korea Patent Publication No. 10-1811640 (2017.12.18)

The present invention is a smart farm that creates an optimal multiple linear regression model between the production of crops and environmental factors of the smart farm in consideration of the effect of environmental settings on crop growth being fermented after a certain period of time has elapsed from the time of setting the environment. This is to provide a method of constructing an optimal linear model for the environment.

According to one aspect of the present invention, a method for constructing an optimal linear model for the production and environment of a smart farm is disclosed.

The method of constructing an optimal linear model for the production volume and environment of the smart farm according to an embodiment of the present invention includes the steps of collecting the production data and environment data of the smart farm, and using the collected production data and environment data, the current parking ( Week) and calculating a correlation coefficient for the environmental factor of the previous week, calculating a time difference using the calculated correlation coefficient, and using the calculated time difference to calculate the initial corresponding week (CW) : calculating Corresponding Week), improving the calculated initial corresponding week, and calculating a multiple linear regression model using the improved corresponding week generated through the improvement.

The time difference is a difference value between the previous parking and the current parking, which has the greatest influence on the production capacity of the current parking.

The step of calculating the time difference using the calculated correlation coefficient includes: generating a correlation coefficient table for the production of the current parking and environmental factors of the previous parking; calculating a maximum correlation coefficient value, calculating a time difference value of the calculated maximum correlation coefficient value, and a time difference value and a maximum correlation coefficient value in which the maximum correlation coefficient value is less than a preset threshold in the generated maximum correlation coefficient table and generating a final maximum correlation coefficient table by removing it.

The correlation coefficient ρ is calculated using the following equation.

이고, l은 시간차이로서, l=1, 2, …, L이고, T는 전치(transpose) 행렬이고, M개의 환경인자에 각각 적용되고, D는 미리 설정된 생산구간의 주수이고, p는 생산량 데이터이고, e는 환경 데이터이고, w_c는 현재 주차이다.here,

and l is the time difference, l = 1, 2, ... , L, T is a transpose matrix, each applied to M environmental factors, D is the number of weeks of a preset production section, p is production data, e is environmental data, and w _c is the current week .

In the calculating of the time difference value of the calculated maximum correlation coefficient value, the maximum correlation coefficient value and the time difference value of each column of the correlation coefficient table are calculated using the following equation.

Here, l _m ^* is a time difference value calculated for each column of the correlation coefficient table, and m is an environmental factor.

The initial corresponding parking is calculated using the following equation.

Here, u is the corresponding parking, w _c is the current parking, and l is the time difference.

In the step of improving the calculated initial corresponding parking, the correlation coefficient extracts the parking value of the production quantity equal to or greater than a preset threshold and the corresponding parking value of the environmental factor, and the corresponding parking value of the previous parking for the corresponding parking where the correlation coefficient is less than the threshold , calculating a regression line of the initial corresponding week and a mean squared error (MSE) of the corresponding week using the extracted and assigned corresponding parking values, and using the mean squared error and calculating the improvement corresponding parking.

The mean square error is calculated using the following equation.

는 상기 회귀직선으로서, 회귀 방정식

로부터

를 구하여

로 산출되고, u는 대응주차이고, w는 주차이고, α는 회귀 파라미터이다.here,

is the regression line, the regression equation

from

to get

, where u is the corresponding parking, w is the parking, and α is the regression parameter.

Calculating the improvement corresponding parking may include calculating a regression equation of the following equation by using a corresponding parking value in which the mean square error is equal to or less than a threshold value, and using the following equation to obtain the improvement corresponding parking (u) from the regression equation ^r ).

Here, u is the corresponding parking, w is the parking, and γ is the regression parameter.

The multiple linear regression model is calculated using the following equation.

는 β의 해 값이다.where y is the production amount of parking section D, x is D×M environmental data, β is a regression parameter,

is the solution value of β.

The method of constructing an optimal linear model for the production and environment of the smart farm according to an embodiment of the present invention considers that the effect of the environment setting on the crop growth is fermented after a certain period of time from the time of setting the environment. It is possible to create an optimal multiple linear regression model between environmental factors.

1 is a flowchart illustrating a method of constructing an optimal linear model for the production and environment of a smart farm according to an embodiment of the present invention.
2 to 10 are diagrams for explaining a method of constructing an optimal linear model for the production volume and environment of a smart farm according to an embodiment of the present invention.

As used herein, the singular expression includes the plural expression unless the context clearly dictates otherwise. In this specification, terms such as "consisting of" or "comprising" should not be construed as necessarily including all of the various components or various steps described in the specification, some of which components or some steps are It should be construed that it may not include, or may further include additional components or steps. In addition, terms such as "...unit" and "module" described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software, or a combination of hardware and software. .

Hereinafter, various embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating a method for constructing an optimal linear model for production and environment of a smart farm according to an embodiment of the present invention, and FIGS. 2 to 10 are optimal for production and environment of a smart farm according to an embodiment of the present invention It is a diagram for explaining a method of constructing a linear model. Hereinafter, a method for constructing an optimal linear model for the production volume and environment of a smart farm according to an embodiment of the present invention will be described with reference to FIG. 1 , with reference to FIGS. 2 to 10 .

In step S100, production data and environmental data are collected.

2 shows the relationship between production data and environmental data during the week. Referring to FIG. 2 , if the week in which planting of crops starts is _one week, M pieces of environmental data may be acquired until week W. Crop production can be measured on a weekly basis starting from the W _s week and up to the W ₂ week.

For example, environmental data includes internal temperature, external temperature, internal humidity, cumulative solar radiation, electrical conductivity (EC: Electrica; Conductivity), amount of light, carbon dioxide (CO ₂ ), upper limit of appropriate temperature, lower limit of appropriate temperature, and temperature difference between day and night (DIF) may include.

3 shows examples of environmental data measured on a weekly basis from the 1st week to the 20th week. 3, the environmental factors corresponding to each of the heat numbers from 1 to 10 are internal temperature, external temperature, internal humidity, cumulative solar radiation, electrical conductivity (EC: Electrica; Conductivity), light quantity, carbon dioxide (CO ₂ ), upper limit of appropriate temperature , the lower limit of the appropriate temperature and the temperature difference between day and night (DIF).

In this specification, the production data and environmental data used are about paprika of a smart farm farm located in Gangwon-do collected by the Gangwon-do Agricultural Research and Extension Services.

이고, 환경 데이터 e는 w_c의 이전 주차에서 모두 획득된다.

이고,

이다.For each parking, production data p and environment data e of section D are given. D can usually be set around 8 to 12 weeks. If the current parking is w _c ,

, and the environmental data e are all obtained from the previous week of w _c .

ego,

am.

Since the production data is a Dx1 vector and the environment data is M, it can be expressed as a matrix of DxM.

In step S200, a correlation coefficient value is calculated for the production amount of the current parking and the environmental factor of the previous parking, and a time difference value is calculated using the calculated correlation coefficient value.

4 is a flowchart illustrating detailed steps of step S200. Hereinafter, step S200 will be described with reference to FIG. 4 .

In step S210, a correlation coefficient table is generated for the production amount of the current parking and the environmental factors of the previous parking.

의 생산량이 영향을 가장 많이 받는 환경인자값의 주차를 조사하기 위하여 수행되는 것이다. 도 2를 참조하면, 현재 주차

이며, M개의 환경인자가 존재하고, 각각의 환경인자는 다른 주차의 생산량에 영향을 줄 수 있다. 그래서, 이전 주차와 현재 주차와의 시간차이(lag) 개념을 도입하여 통계적인 자료분석을 위한 상관계수가 정의될 수 있다.Step S200 is a section

This is carried out to investigate the parking of environmental factor values that are most affected by the production of 2 , the current parking

, and there are M environmental factors, and each environmental factor can affect the yield of different parking lots. Therefore, a correlation coefficient for statistical data analysis can be defined by introducing the concept of a lag between the previous parking and the current parking.

In order to derive the correlation coefficient, the response variable y and the explanatory variable x may be defined by the following equation.

[Equation 1]

Here, l = 1, 2, ... , L, and T are transpose matrices, and can be applied to M environment factors, respectively.

That is, the production data p may be substituted into the response variable y, and the delayed environmental data e by the time difference l may be substituted into the explanatory variable x.

In y, the production data of the current week w _c and the production data during the previous D-1 week may be substituted. For example, the mth environmental factor e _m 7 weeks ago is w _c -7, and the time difference value may be 7.

The correlation coefficient ρ may be defined by the following equation.

[Equation 2]

That is, it is possible to generate a correlation coefficient table of LxM by calculating the correlation coefficient ρ value for each parking.

In step S220, the maximum correlation coefficient value of the correlation coefficient values of each column is calculated from the generated correlation coefficient table, and a time difference value of the calculated maximum correlation coefficient value is calculated.

That is, the value l having the largest correlation coefficient value can be calculated using the following equation.

[Equation 3]

Here, m is an environmental factor.

Then, the maximum correlation coefficient table is generated in which the time difference value l _m ^* and the maximum correlation coefficient value ρ _max calculated for each column are composed of a 2×M matrix. That is, time difference values may be arranged in the first row, and maximum correlation coefficient values may be arranged in the second row.

In step S230, the maximum correlation coefficient value ρ _max is less than a preset threshold (eg, 0.3 to 0.4) from the generated maximum correlation coefficient table, the time difference value and the maximum correlation coefficient value are removed to generate the final maximum correlation coefficient table . The final maximum correlation coefficient table generated may be composed of a 2×N matrix, where N ≤ M.

By generating the final maximum correlation coefficient table for all the parkings in which the production data exists, the time difference value and the maximum correlation coefficient value for each week as shown in FIGS. 5 and 6 can be calculated.

5 and 6 show examples of calculation of the time difference value and the maximum correlation coefficient value between the production amount for each week and the environmental factor, respectively. For example, referring to FIG. 5 , in the case of the 22nd week, the production corresponds to 5, 6, 5, 8, 4, 5, 1, 5, 4 previous weeks for each environmental factor.

Referring back to FIG. 1 , in step S300 , an initial corresponding week (cw) is calculated using the calculated time difference value.

The time difference is expressed as a difference value between the current parking and the previous parking, and the corresponding parking u is calculated by the following equation.

[Equation 4]

For example, if the current parking w _c is 23 weeks and the time difference l is 7 weeks ago, the corresponding parking lot becomes u=23-7=16.

7 (a) to (c) are graphs showing corresponding parking values calculated when environmental factors are internal temperature, carbon dioxide, and internal humidity, respectively.

In step S400, the calculated initial corresponding parking is improved.

The initial corresponding parking calculated using the correlation coefficient above may cause an error in the value for each parking. That is, it may be ideal that the corresponding parking is calculated in the form of monotonically increasing in proportion to the parking.

For example, referring to (a) of FIG. 7 , since there are sections in which the 18th, 20th, and 26th weeks are matched to the 18th, 18th, and 14th weeks, respectively, and decreasing sections exist, a corresponding parking error may occur. .

8 is a flowchart illustrating detailed steps of step S400. Hereinafter, step S400 will be described with reference to FIG. 8 .

In step S410, the parking and corresponding parking data is corrected.

[W _s , W ₂ ] Production P parkings w ₁ , w ₂ , … , w _p , and the initial response parking for each environmental factor u ₁ , u ₂ , … , u _p . w _i matches u _i , and (w _i , u _i ) has a correlation coefficient ρ _i .

The initial correlation coefficient value calculated in step S200 above extracts the parking of production and the corresponding parking (w _i , u _i ) of the environmental factor above a preset threshold (eg, 0.3), and the initial correlation coefficient value is less than the threshold For parking, the corresponding parking value of the previous parking is assigned (u _i = u _i-1 ).

As a result, P data [(w ₁ , u ₁ ), (w ₂ , u ₂ ), ... , (w _p , u _p )] can be obtained.

In step S420 , a regression line of the corresponding week and a mean squared error (MSE) of the corresponding week are calculated.

The regression equation of the corresponding parking can be expressed by the following equation.

[Equation 5]

Here, the corresponding parking u is a P×1 vector, the parking w is a P×1 matrix, and the regression parameter α is a 2×1 vector.

를 구하면, 회귀 직선은 하기 수학식으로 나타낼 수 있다.From the regression equation of Equation 5

, the regression line can be expressed by the following equation.

[Equation 6]

And, the mean square error (MSE) of the regression straight line and the corresponding week may be calculated by the following equation.

[Equation 7]

In step S430, the improved corresponding parking is calculated using the mean square error (MSE _i ) of the regression straight line and the corresponding parking.

First, as shown in the following Equation, a regression equation is generated using corresponding weeks having a mean square error (MSE _i ) equal to or less than a preset threshold T _c .

[Equation 8]

Here, assuming that the number of corresponding parking values whose mean square error is less than or equal to the threshold value is G, corresponding parking u is a G×1 vector, parking w is a G×1 vector, and the regression parameter γ is a 2×1 vector.

A regression straight line is calculated from the regression equation of Equation 8 as shown in the following Equation.

[Equation 9]

Through this, finally an improved corresponding parking u ^r can be obtained.

9 (a) to (c) are graphs showing improved corresponding parking values calculated when environmental factors are internal temperature, carbon dioxide, and internal humidity, respectively. As shown in FIG. 9 , the improved corresponding parking values appear to increase monotonically.

Referring back to FIG. 1 , in step S500 , a multiple linear regression model is calculated using the improved corresponding parking values.

The multiple linear regression model may be defined by the following equation.

[Equation 10]

Here, y is the production amount of the parking section D, and each environmental factor has a different corresponding parking, so X is an environmental data value having a corresponding parking. And, β is a regression parameter.

In multiple linear regression analysis, M independent variables x ₁ , x ₂ , … , x _M is introduced, the production amount y can be expressed by the following equation.

[Equation 11]

And, the independent variable x can be expressed by the following equation.

[Equation 12]

x is M environment factors, and each column is D environment data. Each data is a value having a corresponding parking lot.

And, the regression parameter β can be expressed by the following equation.

[Equation 13]

Here, β ₀ is an offset value.

A solution β value may be determined using the following equation.

[Equation 14]

[Equation 15]

10 shows an example of a multiple linear regression model calculated using the regression parameter β value. That is, FIG. 10 shows an example of a relational expression between environmental factors affecting production and production per pyeong in weeks 17 to 35.

For example, the multiple linear regression model of week 23 in FIG. 10 can be expressed as follows.

[Equation 16]

[Equation 17]

In other words, paprika production at the 23rd week was affected by the cumulative solar radiation 5 weeks before harvest, the amount of light 4 weeks before the harvest, the external temperature 3 weeks before the harvest, the carbon dioxide 1 week before the harvest, and the temperature difference between day and night (DIF) 4 weeks before the harvest.

The above-described embodiments of the present invention have been disclosed for purposes of illustration, and various modifications, changes, and additions will be possible within the spirit and scope of the present invention by those skilled in the art having ordinary knowledge of the present invention, and such modifications, changes and additions should be regarded as belonging to the following claims.

Claims (10)

In the method of constructing an optimal linear model for the production and environment of the smart farm,
collecting production data and environmental data of the smart farm;
Calculating a correlation coefficient between the production of the current week and the environmental factor of the previous week using the collected production data and environmental data, and calculating a time difference using the calculated correlation coefficient ;
calculating an initial correspondence week (CW) using the calculated time difference;
improving the calculated initial corresponding parking; and
A method of constructing an optimal linear model for production and environment of a smart farm, comprising calculating a multiple linear regression model using the improvement corresponding parking generated through the improvement.
According to claim 1,
The time difference is the optimal linear model construction method for the production and environment of the smart farm, characterized in that it is the difference value between the previous parking and the current parking that has the most influence on the production of the current parking.
According to claim 1,
Calculating the time difference using the calculated correlation coefficient comprises:
generating a correlation coefficient table for the production amount of the current parking and the environmental factors of the previous parking;
calculating a maximum correlation coefficient value of the correlation coefficient values of each column from the generated correlation coefficient table, and calculating a time difference value of the calculated maximum correlation coefficient value; and
The production and environment of the smart farm, characterized in that it includes the step of generating a final maximum correlation coefficient table by removing the time difference value and the maximum correlation coefficient value in which the maximum correlation coefficient value is less than a preset threshold from the generated maximum correlation coefficient table How to build an optimal linear model for
4. The method of claim 3,
The correlation coefficient (ρ) is an optimal linear model construction method for the production and environment of the smart farm, characterized in that calculated using the following equation.

here,

and l is the time difference, l = 1, 2, ... , L, T is a transpose matrix, each is applied to M environmental factors, D is the number of weeks of a preset production section, p is production data, e is environmental data, w _c is current parking
4. The method of claim 3,
Calculating the time difference value of the calculated maximum correlation coefficient value comprises:
An optimal linear model construction method for the production and environment of the smart farm, characterized in that the maximum correlation coefficient value and the time difference value of each column of the correlation coefficient table are calculated using the following equation.

Here, l _m ^* is a time difference value calculated for each column of the correlation coefficient table, and m is an environmental factor.
According to claim 1,
The method of constructing an optimal linear model for the production and environment of the smart farm, characterized in that the initial corresponding parking is calculated using the following equation.

where u is the corresponding parking, w _c is the current parking, and l is the time difference
According to claim 1,
The step of improving the calculated initial corresponding parking is,
extracting a parking value of a production quantity equal to or greater than a preset threshold and a corresponding parking value of an environmental factor, and assigning a corresponding parking value of the previous parking to the corresponding parking having the correlation coefficient less than a threshold;
calculating a regression line of the initial corresponding parking and a mean squared error (MSE) of the corresponding parking using the extracted and assigned corresponding parking values; and
The method of constructing an optimal linear model for the production and environment of the smart farm, characterized in that it comprises the step of calculating the improvement corresponding parking by using the mean square error.
8. The method of claim 7,
The method for constructing an optimal linear model for the production and environment of the smart farm, characterized in that the mean square error is calculated using the following equation.

here,

is the regression line, the regression equation

from

to get

, where u is the corresponding parking, w is the parking, and α is the regression parameter.
8. The method of claim 7,
The step of calculating the improvement corresponding parking is,
calculating a regression equation of the following equation by using a corresponding parking value in which the mean square error is equal to or less than a threshold value; and

where u is the corresponding parking, w is the parking, and γ is the regression parameter
An optimal linear model construction method for the production and environment of the smart farm, characterized in that it comprises the step of calculating the improvement corresponding parking (u ^r ) from the regression equation using the following equation.

The method of claim 1,
The multiple linear regression model is an optimal linear model construction method for the production and environment of the smart farm, characterized in that calculated using the following equation.

where y is the production amount of parking section D, x is D×M environmental data, β is a regression parameter,

is the solution value of β

Images