CN103234916A - Prediction method for net photosynthetic rate of population - Google Patents
Prediction method for net photosynthetic rate of population Download PDFInfo
- Publication number
- CN103234916A CN103234916A CN2013101153870A CN201310115387A CN103234916A CN 103234916 A CN103234916 A CN 103234916A CN 2013101153870 A CN2013101153870 A CN 2013101153870A CN 201310115387 A CN201310115387 A CN 201310115387A CN 103234916 A CN103234916 A CN 103234916A
- Authority
- CN
- China
- Prior art keywords
- photosynthetic rate
- net photosynthetic
- data
- model
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Investigating Or Analysing Biological Materials (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a prediction method for the net photosynthetic rate of a population and aims to overcome the problems that it is difficult to establish a prediction regression equation for the net photosynthetic rate of a population, considerable related data are needed in establishment of a prediction model for the net photosynthetic rate of the population and prediction accuracy is not high. The method comprises the following steps: 1, acquiring proportioning data of spectral radiance of every wave band of visible light in a region, wherein a portable multispectral radiometer is installed at a height H higher than the canopy of a plant in the test region with an area of S, the proportioning data M_D of spectral radiance of every wave band of visible light in the test region in different periods is obtained, and [0,1] normalization processing is carried out; 2, acquiring data of the net photosynthetic rate of the population; 3, constructing a bionic kernel function; 4, establishing an SVM training set and an SVM prediction set; 5, carrying out tool selection and parameter optimization on a prediction model; and 6, predicting the net photosynthetic rate of the population, which comprises obtainment of the prediction model Model, obtainment of a predicted value Predict of the net photosynthetic rate of the population and determination of reliability of the prediction model Model.
Description
Technical field
The present invention relates to a kind of plant population Net Photosynthetic Rate Forecasting Methodology, more particularly, the present invention relates to a kind of colony's Net Photosynthetic Rate Forecasting Methodology of biomimetic type kernel function.
Background technology
Light is as the most essential resource of plant, it is the important factor that influences its form and function, individual Net Photosynthetic Rate has embodied the organic accumulation of individual plants, it is the important factor that influences individual plants form and function, colony's Net Photosynthetic Rate then is the summation of such plant individual Net Photosynthetic Rate in the zone, reflected should the gross photosynthesis synthesis of organic substance accumulation in a period of time of zone plant situation, for analyzed area plant configuration and function important value is arranged, therefore predict that colony's Net Photosynthetic Rate has very strong realistic meaning in agricultural production.Because the leaf area size has reflected the Net Photosynthetic Rate size, we can predict Net Photosynthetic Rate by setting up the method for regression equation between leaf area and the Net Photosynthetic Rate, but the method for traditional test leaf area has limitation such as complicacy, apparatus expensive.If by the regression equation method between leaf area and the Net Photosynthetic Rate, guarantee the accuracy of Net Photosynthetic Rate predicted value, then need the great amount of samples data, owing to will predict colony's Net Photosynthetic Rate, need to measure the leaf area of a large amount of plants, workload is heavier.
Support vector machine (Support Vector Machine, SVM) be a kind of machine learning method based on Statistical Learning Theory, it is solving small sample, show many distinctive advantages in the non-linear and higher-dimension pattern-recognition, in SVM, kernel function is its core, the inner product operation of higher dimensional space is converted into the kernel function calculating of the low-dimensional input space, solved the problems of in high-dimensional feature space, calculating such as " dimension disasters ", the variation of its form and parameter can implicitly change from the input space to the feature space shines upon, and then feature space character is exerted an influence.Kernel function is an inner product in essence, basic role is the vector of accepting in two lower dimensional spaces, can calculate through the inner product of vectors value in higher dimensional space after certain conversion, namely determine lower dimensional space to the higher dimensional space mapping relations, become the linear inseparable key of solution.
By seven color chameleons researchs to title that incomparable magnificent rainbow is arranged in the organic sphere, find that its skin color not only changes along with the integrated environment change color, and the local skin color is all close with the environmental colors that it approaches, also can change with the local environment color change thereupon, embody the very strong adaptive faculty to overall surrounding environment color and local ambient color.To the spontaneous phenomenon that surrounding environment adapts to, it is significant to construct a kind of biomimetic type kernel function that can adjust separately global data and local data from chameleon.
Summary of the invention
Technical matters to be solved by this invention be overcome the test zone plant leaf area workload that prior art exists big, set up colony's Net Photosynthetic Rate prediction regression equation difficulty, set up the problem that colony's required related data of Net Photosynthetic Rate forecast model is more and forecasting accuracy is not high, a kind of groups Net Photosynthetic Rate Forecasting Methodology is provided.
For solving the problems of the technologies described above, the present invention adopts following technical scheme to realize: the step of a described kind of groups Net Photosynthetic Rate Forecasting Methodology is as follows:
1) obtain each band spectrum radiation proportion relation data of regional visible light:
(1) in area is the pilot region of S, be higher than influences of plant crown be the height of H model is installed is the portable multi-spectrum radiacmeter of MSR-16, wherein: the S value is 6m
2To 55m
2, the H value is 1.2m or 2m, and model is that portable multi-spectrum radiacmeter effective measurement diameter on ground under it of MSR-16 is H/2, and the observation area is S
1, S
1=[π * (H/4)
2], the number that calculates the test cell of pilot region division is M, its numerical value is got S/S
1Integral part;
(2) in the Measuring Time section of appointment, each hour measured once, each picked at random N test cell, N<M wherein, the test unit that requires at every turn to choose is different from the test unit that other Measuring Time have been chosen, during each the measurement, H place, test cell top when placing this time to measure the portable multi-spectrum radiacmeter of MSR-16, the H value is 1.2m or 2m, get 5 one-point measurements in each test cell, every point measurement is averaged for 2 times, and 5 fixed point mean values are as the spectral composition of this test cell, and the mean value of 3 test cells is as each band spectrum radiation proportion relation data of visible light in this period pilot region;
(3) by the method for (2) step, in the time span of test, obtain each band spectrum radiation proportion relation data M _ D of visible light in the different period pilot regions, carry out [0,1] normalized, obtain the data M _ D after the normalization
1
2) obtain colony's Net Photosynthetic Rate data:
(1) model that adopts U.S. CID company to produce is the individual Net Photosynthetic Rate of portable photosynthesis measurement systematic survey of CI-310;
(2) the same kind plant of random choose 3 strains in each selected test cell, if these plant lazy weight 3 strains, measure by actual plant quantity, 5 blades of every strain plant random choose, if 5 of blade quantity less thaies, measure by actual blade quantity, it is Net Photosynthetic Rate of portable photosynthesis measurement systematic survey of CI-310 that every blade adopts model, average as this strain plant Net Photosynthetic Rate, 3 test unit mean values are as this strain plant population Net Photosynthetic Rate in this period pilot region;
(3) by the method for (2) step, at the trial between in the span, obtain the Net Photosynthetic Rate C_D of this kind plant colony in the different period pilot regions, carry out [0,1] normalized, obtain the data C_D after the normalization
1
3) make up the biomimetic type kernel function;
4) set up SVM training set and forecast set:
With data M _ D
1Be divided into two parts data M _ D by different time sections
11And M_D
12, according to M_D
1Be divided into two parts data M _ D
11And M_D
12Time period, with data C_D
1Be divided into two parts data C_D by the corresponding time period
11And C_D
12With the data M _ D of first
11And C_D
11As the training set of SVM, with second portion data M _ D
12And C_D
12Forecast set as SVM.
5) instrument of setting up forecast model is selected and parameter optimization.
6) prediction colony Net Photosynthetic Rate.
The step of the structure biomimetic type kernel function described in the technical scheme is as follows:
1) with Gaussian kernelK
1(x, x
i)=exp (γ || x-x
i||
2) and polynomial kernelK
2(x, x
i)=k (<x, x
i〉+c) is the benchmark kernel function, and wherein, γ and c are parameter, x and x
iBe the lower dimensional space multi-C vector.With K
1(x, x
i)=exp (γ || x-x
i||
2) import among the libsvm, as the SVM kernel function, training set data is trained, adopt grid-search search optimal parameter, determine the γ value the highest to the training set predictablity rate, be defined as the Γ value.
2) seek out Gaussian kernel characteristic curve tangent slope maximal value diff (K under the Γ value
1(x
m, x
i)) and minimum value diff (K
1(x
n, x
i)), and corresponding point of contact coordinate (x
m, K
1(x
m, x
i)), (x
n, K
1(x
n, x
i)), such point of contact is called break.
3) according to break coordinate (x
m, K
1(x
m, x
i)), (x
n, K
1(x
n, x
i)) and slope diff (K
1(x
m, x
i))/ξ, diff (K
1(x
n, x
i))/ξ, wherein ξ is real variable, determines polynomial kernel expression formula, expression formula is by K
2' (x, x
i)=diff (K
1(x
m, x
i))/ξ/x
i* (<x, x
i〉+c) and K
2" (x, x
i)=diff (K
1(x
n, x
i))/ξ/x
i* (<x, x
iThe two parts of 〉+c) are formed.
4) make up the biomimetic type kernel function:
K
bsf(x,x
i)=exp(-Γ||x-x
i||
2)+diff(exp(-Γ||x
B-x
i||
2)/ξ/x
i×(<x,x
i>+c)
Wherein: || x||>|| x
i|| the time, x
BBe taken as x
n, || x||<|| x
i|| the time, x
BBe taken as x
m, || x||=||x
i|| the time, with x and x
iBe classified as similarly, need not to use kernel function that it is calculated.
Biomimetic type kernel function K
Bsf(x, x
i) can realize the adjustment to global data and local data by the adjusting of parameter Γ and ξ, to adapt to the requirement that different pieces of information is sorted out.
The instrument of setting up forecast model described in the technical scheme selects and parameter optimization refers to:
Adopt MATLAB with biomimetic type kernel function K
Bsf(x, x
i) put into the libsvm tool box, the position that replaces original RBF kernel function, realize grid-search method optimization parameter by function S VMcgForRegress (), utilize grid-search method initial optimization parametric procedure, in function S VMcgForRegress (), adopt Gaussian kernelK
1(x, x
i), by [bestmse, bestc, bestg]=SVMcgForRegress (C_D
11, M_D
11,-8,8 ,-8,8), can obtain parameter b estc and bestg, bestc is best penalty parameter c value, bestg is best parameter γ value, i.e. Γ value;
In function S VMcgForRegress (), adopt biomimetic type kernel function K
Bsf(x, x
i), by [bestmse, bestc, bestCMG]=SVMcgForRegress (C_D
11, M_D
11, bestc, bestc ,-8,8), can obtain optimal parameter bestCMG, i.e. the ξ value.
The step of the prediction colony Net Photosynthetic Rate described in the technical scheme is as follows:
1) obtains forecast model model
Utilize the svmtrain () in the libsvm tool box, namely by model=svmtrain (C_D
11, M_D
11Cmd), can obtain forecast model model, cmd=['-c' wherein, num2str (bestc), '-g', num2str (bestg), '-s3-p0.01-t2'], s is made as 3 representatives and adopts e – SVR formula, p is made as 0.01 representative the value of loss function among the e-SVR is set, and t is made as 2, and to represent the kernel function that adopts among the SVM be K
Bsf(x, x
i); The information that in model, has comprised kernel function type, support vector number and support vector coefficient scope in decision function.
2) obtain the predicted value predict of colony's Net Photosynthetic Rate
Utilize the svmpredict () in the libsvm tool box, by [predict, mse]=svmpredict (C_D
12, M_D
12, model) realize utilizing M_D
12To C_D
12Predict, obtain the predicted value predict of colony's Net Photosynthetic Rate.
3) reliability of judgement forecast model model
Simultaneously, pass through C_D
12Calculate between the two coefficient R with predict, judge the accuracy of predicted value predict, i.e. the reliability of forecast model model.
Compared with prior art the invention has the beneficial effects as follows:
1. a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention is compared with traditional colony's Net Photosynthetic Rate prediction regression equation method of setting up, and has model and sets up the characteristics that process is simple, the required input data are few, predictablity rate is high, be easy to realize.
2. a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention is compared with the method for the individual Net Photosynthetic Rate of existing prediction, prediction colony Net Photosynthetic Rate can reflect a regional implants Net Photosynthetic Rate situation, is conducive to the situation of reflecting regional plant configuration and function.
3. a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention is compared with the method for the existing SVM of utilization prediction colony Net Photosynthetic Rate, adopted novel kernel function, realized that the better classification capacity of SVM input data is had the advantages that the forecast model generalization ability is strong, predictablity rate is high.
4. a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention, only can realize predicting colony's Net Photosynthetic Rate by each band spectrum radiation proportion relation of regional visible light, have the few characteristics of required input data, be conducive to the influence relation of each band spectrum radiation proportion relation of reflecting regional visible light and Net Photosynthetic Rate.
5. a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention, adopted novel kernel function, overcome the deficiency that traditional core function overall situation learning ability and local learning ability can not independently be adjusted, can adjust the performance of kernel function according to each band spectrum radiation proportion relation data characteristic distributions of visible light, improved forecasting process efficient and accuracy, the method can directly be used in other predicted application.
6. a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention, its method is simple, convenient, and speed is fast, and step is clear, saves time, and is laborsaving, and the predictablity rate height of Forecasting Methodology.
Description of drawings
The present invention is further illustrated below in conjunction with accompanying drawing:
MSR-16 type portable multi-spectrum radiacmeter placement location and the Validity Test area schematic of Fig. 1 for adopting in the kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention;
Fig. 2 is the synoptic diagram that a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention adopts CI-310 instrument test Net Photosynthetic Rate;
Fig. 3 carries out the process block diagram of initial optimization to parameter for the grid-search method in the kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention;
Fig. 4 is the process block diagram that the biomimetic type kernel function parameter in the kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention is determined;
Fig. 5 is interior each the band spectrum radiation proportion relation data profile of visible light in the zone after the employing kind of groups Net Photosynthetic Rate Forecasting Methodology normalization of the present invention;
Fig. 6 is the colony's Net Photosynthetic Rate data profile after the employing kind of groups Net Photosynthetic Rate Forecasting Methodology normalization of the present invention;
Fig. 7 is the training set colony Net Photosynthetic Rate prediction curve (article one) of a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention;
Fig. 8 is the forecast set colony Net Photosynthetic Rate prediction curve (second) of a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention;
Fig. 9 is the FB(flow block) of a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention.
Embodiment
Below in conjunction with accompanying drawing the present invention is explained in detail:
Consult Fig. 9, the present invention has overcome the problem that prior art exists, one kind of groups Net Photosynthetic Rate Forecasting Methodology is provided, namely only utilize each band spectrum radiation proportion relation of visible light realize having high accuracy to colony's Net Photosynthetic Rate forecast method, its step is as follows:
1. obtain each band spectrum radiation proportion relation data of regional visible light
The step that obtains each band spectrum radiation proportion relation data of regional visible light is as follows:
Adopting model produced in USA is the portable multi-spectrum radiacmeter of MSR-16, and 460-710nm wave band visible light composition, the ratio of analyzing portable multi-spectrum radiacmeter top incident change.
1) consult Fig. 1, establishing the pilot region area is S, and the S value is 6m
2To 55m
2, model is that the height that the portable multi-spectrum radiacmeter of MSR-16 is higher than influences of plant crown is H, and the H value is 1.2m or 2m, and model is that portable multi-spectrum radiacmeter effective measurement diameter on ground under it of MSR-16 is H/2, and the observation area is S
1, S
1=[π * (H/4)
2], wherein: π is circular constant.According to the pilot region S that divides in the process of the test and MSR-16 instrument test height H, the number that calculates the test cell of being divided by zone to be tested is M, and its numerical value is got S/S
1Integral part.
2) in the Measuring Time section of appointment, each hour measured once, each picked at random N test cell, N<M wherein, the test unit that requires at every turn to choose is different from the test unit that other Measuring Time have been chosen, during each the measurement, H place, test cell top when placing this time to measure the portable multi-spectrum radiacmeter of MSR-16, the H value is 1.2m or 2m, get 5 one-point measurements in each test cell, every point measurement is averaged for 2 times, and 5 fixed point mean values are as the spectral composition of this test cell, and the mean value of 3 test cells is as each band spectrum radiation proportion relation data of visible light in this period pilot region.
3) by the 2nd) method of step, in the time span of test, obtain each band spectrum radiation proportion relation data M _ D of visible light in the different period pilot regions, carry out [0,1] normalized, obtain the data M _ D after the normalization
1
2. obtain colony's Net Photosynthetic Rate data
Consult Fig. 2, the step that obtains colony's Net Photosynthetic Rate data method is as follows:
1) model that adopts U.S. CID company to produce is the individual Net Photosynthetic Rate of portable photosynthesis measurement systematic survey of CI-310:
Select 25 * 25(cm
2) square leaf chamber insert main frame, it is indoor that the probe of sensor front end is pressed into leaf, the plug of sensor rear end inserts the corresponding jack of main frame, one end of data line is connected with main frame, the other end of data line links to each other with the PC card, the PC cartoon is crossed the PC draw-in groove and is connected computer, draws in a pipe to the 4L surge flask from " INTAKE " mouth of CI-310 sensor, and this surge flask is placed in the outside air to obtain not breathed the CO that influences
2Air carries out CO with the soda-lime pipe that is equipped with at random before each the measurement
2Zeroing.Close the leaf chamber after blade put into the leaf chamber, press the main frame measuring switch, the measurement of first blade of beginning, measurement finishes, and computer generation warning is opened the leaf chamber, closes the leaf chamber after another sheet is put into the leaf chamber, begins the measurement of next blade.
2) the same kind plant of random choose 3 strains (if these plant lazy weight 3 strains in each selected test cell, measure by actual plant quantity), 5 blades of every strain plant random choose are (as if 5 of blade quantity less thaies, measure by actual blade quantity), it is Net Photosynthetic Rate of portable photosynthesis measurement systematic survey of CI-310 that every blade adopts model, average as this strain plant Net Photosynthetic Rate, 3 test unit mean values are as this strain plant population Net Photosynthetic Rate in this period pilot region.
3) by the 2nd) method of step, at the trial between in the span, obtain the Net Photosynthetic Rate C_D of this kind plant colony in the different period pilot regions, carry out [0,1] normalized, obtain the data C_D after the normalization
1
3. make up the biomimetic type kernel function
1) with Gaussian kernelK
1(x, x
i)=exp (γ || x-x
i||
2) and polynomial kernelK
2(x, x
i)=k (<x, x
i〉+c) is the benchmark kernel function, and wherein, γ and c are parameter, x and x
iBe the lower dimensional space multi-C vector.With K
1(x, x
i)=exp (γ || x-x
i||
2) import among the libsvm, as the SVM kernel function, training set data is trained, adopt grid-search search optimal parameter, determine the γ value the highest to the training set predictablity rate, be defined as the Γ value.
2) seek out Gaussian kernel characteristic curve tangent slope maximal value diff (K under the Γ value
1(x
m, x
i)) and minimum value diff (K
1(x
n, x
i)), and corresponding point of contact coordinate (x
m, K
1(x
m, x
i)), (x
n, K
1(x
n, x
i)), such point of contact is called break.
3) according to break coordinate (x
m, K
1(x
m, x
i)), (x
n, K
1(x
n, x
i)) and slope diff (K
1(x
m, x
i))/ξ, diff (K
1(x
n, x
i))/ξ, wherein ξ is real variable, determines polynomial kernel expression formula, expression formula is by K
2' (x, x
i)=diff (K
1(x
m, x
i))/ξ/x
i* (<x, x
i〉+c) and K
2" (x, x
i)=diff (K
1(x
n, x
i))/ξ/x
i* (<x, x
iThe two parts of 〉+c) are formed.
4) make up the biomimetic type kernel function
K
bsf(x,x
i)=exp(-Γ||x-x
i||
2)+diff(exp(-Γ||x
B-x
i||
2)/ξ/x
i×(<x,x
i>+c)
Wherein: || x||>|| x
i|| the time, x
BBe taken as x
n, || x||<|| x
i|| the time, x
BBe taken as x
m, || x||=||x
i|| the time, with x and x
iBe classified as similarly, need not to use kernel function that it is calculated.
Similar according to the process that the surrounding environment change color changes with the skin of chameleon, biomimetic type kernel function K
Bsf(x, x
i) can realize the adjustment to global data and local data by the adjusting of parameter Γ and ξ, to adapt to the requirement that different pieces of information is sorted out.
In fact, the biomimetic type kernel function form of above-mentioned structure can be expressed as:
K
bsf(x,x
i)=K
1(x,x
i)+K
2(x,x
i)。
Wherein: K
1(x, x
i) be Gaussian kernel, K
2(x, x
i) be polynomial kernel, according to kernel function character: and if only if, and function K is negative definite, and then K:X * X → R is kernel function,
Can get following proof procedure: to any a ∈ R
l, have
Because K
1And K
2Be kernel function, so
Can push away
Namely
Be positive semi-definite, so K
1+ K
2Be kernel function, i.e. the K of Gou Jianing
Bsf(x, x
i) can be used as kernel function.
4. set up SVM training set and forecast set
With data M _ D
1Be divided into two parts data M _ D by different time sections
11And M_D
12, according to M_D
1Be divided into two parts data M _ D
11And M_D
12Time period, with data C_D
1Be divided into two parts data C_D by the corresponding time period
11And C_D
12With the data M _ D of first
11And C_D
11As the training set of SVM, with second portion data M _ D
12And C_D
12Forecast set as SVM.
5. setting up the instrument of forecast model selects and parameter optimization
Consult Fig. 3 and Fig. 4, utilize the SVM that small sample is had peculiar advantage, the biomimetic type kernel function that the present invention makes up is put into the libsvm tool box, with M_D
11As SVM input quantity, C_D
11As the SVM output quantity, for parameter Γ, parameter ξ in the penalty parameter c among the SVM, parameter γ and the biomimetic type kernel function, by web search method (grid-search) optimizing.Adopt MATLAB, with biomimetic type kernel function K
Bsf(x, x
i) put into the libsvm tool box, the position that replaces original RBF kernel function, realize grid-search method optimization parameter by function S VMcgForRegress (), utilize grid-search method initial optimization parametric procedure, as shown in Figure 3, in function S VMcgForRegress (), adopt Gaussian kernelK
1(x, x
i), by [bestmse, bestc, bestg]=SVMcgForRegress (C_D
11, M_D
11,-8,8 ,-8,8), can obtain parameter b estc and bestg, bestc is best penalty parameter c value, bestg is best parameter γ value, i.e. Γ value.As shown in Figure 4, in function S VMcgForRegress (), adopt biomimetic type kernel function K
Bsf(x, x
i), by [bestmse, bestc, bestCMG]=SVMcgForRegress (C_D
11, M_D
11, bestc, bestc ,-8,8), can obtain optimal parameter bestCMG, i.e. the ξ value.
6. predict colony's Net Photosynthetic Rate
1) obtains forecast model model
Utilize the svmtrain () in the libsvm tool box, namely by model=svmtrain (C_D
11, M_D
11Cmd), can obtain forecast model model, cmd=['-c' wherein, num2str (bestc), '-g', num2str (bestg), '-s3-p0.01-t2'], s is made as 3 representatives and adopts e – SVR formula, p is made as 0.01 representative the value of loss function among the e-SVR is set, and t is made as 2, and to represent the kernel function that adopts among the SVM be K
Bsf(x, x
i); The information that in model, has comprised kernel function type, support vector number and support vector coefficient scope in decision function;
2) obtain the predicted value predict of colony's Net Photosynthetic Rate
Utilize the svmpredict () in the libsvm tool box, by [predict, mse]=svmpredict (C_D
12, M_D
12, model) realize utilizing M_D
12To C_D
12Predict, obtain the predicted value predict of colony's Net Photosynthetic Rate,
3) reliability of judgement forecast model model
Simultaneously, pass through C_D
12Calculate between the two coefficient R with predict, judge the accuracy of predicted value predict, i.e. the reliability of forecast model model.
Embodiment
1. obtain each band spectrum radiation proportion relation data of regional visible light
Consult Fig. 5, select S=17m arbitrarily in sylvan life ginseng planting base
2Pilot region, the portable multi-spectrum radiacmeter of MSR-16 type is placed from sylvan life ginseng influences of plant crown height H=2m place, then the ground plant to observe diameter be H/2=1m, the observation area is S
1=0.785, S/S then
1=21.656, test cell quantity M=21, Measuring Time section span be July 1 in 2010 to August 31, Measuring Time has 7 test duration sections for test 9:00-16:00 on the same day, so test unit quantity N=3 that chooses, each the band spectrum radiation proportion relation data preparation method of regional visible light that adopts the present invention to introduce obtains each band spectrum radiation proportion relation data M _ D of regional visible light, carries out [0,1] normalized obtains the data M _ D after the normalization
1, as shown in FIG..
2. obtain colony's Net Photosynthetic Rate data
Consult Fig. 6, random choose 3 strain sylvan lifes ginseng in each test unit in 21 test cells, 5 blades of every strain plant random choose, the colony's Net Photosynthetic Rate data preparation method that adopts the present invention to introduce, obtain the sylvan life ginseng Net Photosynthetic Rate C_D of colony, carry out [0,1] normalized, obtain the data C_D after the normalization
1, as shown in FIG..
3. the parameter optimization of forecast model
With the M_D during 1 day July in 2010 to August 10
11As the SVM input quantity, with the C_D during this
11As the SVM output quantity.Utilize that the present invention introduces to penalty parameter c and parameter γ optimization method, obtain best penalty parameter c=2.51, optimal parameter γ=1, i.e. Γ=1, MSE is 0.016; Utilize that the present invention introduces to biomimetic type kernel function parameter optimization method, determine that finally the optimal parameter ξ of biomimetic type kernel function is 1.68, parameter Γ is 2.83.
4. predict colony's Net Photosynthetic Rate
Consult Fig. 7 and Fig. 8, the prediction colony Net Photosynthetic Rate method that adopts the present invention to introduce, penalty parameter c=2.51, Γ=2.83, ξ=1.68 are by model=svmtrain (C_D
11, M_D
11, cmd), obtain forecast model model, wherein model.totalSV is 146, and namely support vector has 146, and minimum value is-0.625 among the model.sv_coef, and maximal value is 0.625, namely the coefficient scope of support vector in decision function is [0.625,0.625].This model makes to training set C_D
11The best accuracy of prediction reaches 89%, and prediction effect as shown in Figure 7.With the M_D during August 11 to August 31 in 2010
12As the SVM input quantity, with the C_D during this
12As the SVM output quantity, adopt this model, by [predict, mse]=svmpredict (C_D
12, M_D
12, model), obtain forecast set C_D
12Prediction accuracy reaches 81%, and prediction effect as shown in Figure 8.
5. judge the accuracy that predicts the outcome
Get different S and H, repeat above-mentioned colony Net Photosynthetic Rate forecasting process, obtain colony's Net Photosynthetic Rate forecasting accuracy.Measuring Time section span be July 1 in 2010 to August 31, Measuring Time is for test 9:00-16:00 on the same day, with the M_D during 1 day July in 2010 to August 10
11And C_D
11As the training set of SVM, with the M_D during August 11 to August 31 in 2010
12And C_D
12Forecast set as SVM.When H=2m, (1) gets S=6m
2, S then
1=0.785, S/S
1=7.643, test cell quantity M=7, test unit quantity N=1; (2) get S=11m
2, S then
1=0.785, S/S
1=14.006, test cell quantity M=14, test unit quantity N=2; (3) get S=22m
2, S then
1=0.785, S/S
1=28.025, test cell quantity M=28, test unit quantity N=4; (4) get S=28m
2, S then
1=0.785, S/S
1=35.669, test cell quantity M=35, test unit quantity N=5; (5) get S=33m
2, S then
1=0.785, S/S
1=42.038, test cell quantity M=42, test unit quantity N=6; (6) get S=39m
2, S then
1=0.785, S/S
1=49.681, test cell quantity M=49, test unit quantity N=7; (7) get S=44m
2, S then
1=0.785, S/S
1=56.051, test cell quantity M=56, test unit quantity N=8; (8) get S=50m
2, S then
1=0.785, S/S
1=63.694, test cell quantity M=63, test unit quantity N=9; (9) get S=55m
2, S then
1=0.785, S/S
1=70.064, test cell quantity M=70, test unit quantity N=10.To the Net Photosynthetic Rate forecast set C_D of colony
12Forecasting accuracy as shown in table 1.
Table 1H=2m, the Net Photosynthetic Rate forecast set C_D of colony under the different S value situations
12Forecasting accuracy
Measuring Time section span be July 1 in 2011 to August 31, Measuring Time is for test 9:00-16:00 on the same day, with the M_D during 1 day July in 2011 to August 10
11And C_D
11As the training set of SVM, with the M_D during August 11 to August 31 in 2011
12And C_D
12Forecast set as SVM.When H=1.2m, (1) gets S=2m
2, S1=0.283 then, S/S
1=7.067, test cell quantity M=7, test unit quantity N=1; (2) get S=4m
2, S then
1=0.283, S/S
1=14.134, test cell quantity M=14, test unit quantity N=2; (3) get S=6m
2, S then
1=0.283, S/S
1=21.201, test cell quantity M=21, test unit quantity N=3; (4) get S=8m
2, S then
1=0.283, S/S
1=28.269, test cell quantity M=28, test unit quantity N=4; (5) get S=10m
2, S then
1=0.283, S/S
1=35.336, test cell quantity M=35, test unit quantity N=5; (6) get S=12m
2, S then
1=0.283, S/S
1=42.403, test cell quantity M=42, test unit quantity N=6; (7) get S=14m
2, S then
1=0.283, S/S
1=49.470, test cell quantity M=49, test unit quantity N=7; (8) get S=16m
2, S then
1=0.283, S/S
1=56.537, test cell quantity M=56, test unit quantity N=8; (9) get S=18m
2, S then
1=0.283, S/S
1=63.604, test cell quantity M=63, test unit quantity N=9; (10) get S=20m
2, S then
1=0.283, S/S1=70.671, test cell quantity M=70, test unit quantity N=10; As shown in table 2 to colony's Net Photosynthetic Rate forecasting accuracy.
Table 2H=1.2m, the colony's Net Photosynthetic Rate forecasting accuracy under the different S value situations
From table 1 and table 2 as seen, adopt biomimetic type kernel function K
Bsf(x, x
i), utilize each band spectrum radiation proportion relation data M _ D of visible light, the Net Photosynthetic Rate C_D of colony is predicted, its forecasting accuracy reaches more than 80%, effect is satisfactory, utilize a kind of groups Net Photosynthetic Rate Forecasting Methodology of the present invention, can realize other plant colony Net Photosynthetic Rate in the agricultural are predicted.
Claims (4)
1. a kind of groups Net Photosynthetic Rate Forecasting Methodology is characterized in that, the step of a described kind of groups Net Photosynthetic Rate Forecasting Methodology is as follows:
1) obtain each band spectrum radiation proportion relation data of regional visible light:
(1) in area is the pilot region of S, be higher than influences of plant crown be the height of H model is installed is the portable multi-spectrum radiacmeter of MSR-16, wherein: the S value is 6m
2To 55m
2, the H value is 1.2m or 2m, and model is that portable multi-spectrum radiacmeter effective measurement diameter on ground under it of MSR-16 is H/2, and the observation area is S
1, S
1=[π * (H/4)
2], the number that calculates the test cell of pilot region division is M, its numerical value is got S/S
1Integral part;
(2) in the Measuring Time section of appointment, each hour measured once, each picked at random N test cell, N<M wherein, the test unit that requires at every turn to choose is different from the test unit that other Measuring Time have been chosen, during each the measurement, H place, test cell top when placing this time to measure the portable multi-spectrum radiacmeter of MSR-16, the H value is 1.2m or 2m, get 5 one-point measurements in each test cell, every point measurement is averaged for 2 times, and 5 fixed point mean values are as the spectral composition of this test cell, and the mean value of 3 test cells is as each band spectrum radiation proportion relation data of visible light in this period pilot region;
(3) by the method for (2) step, in the time span of test, obtain each band spectrum radiation proportion relation data M _ D of visible light in the different period pilot regions, carry out [0,1] normalized, obtain the data M _ D after the normalization
1
2) obtain colony's Net Photosynthetic Rate data:
(1) model that adopts U.S. CID company to produce is the individual Net Photosynthetic Rate of portable photosynthesis measurement systematic survey of CI-310;
(2) the same kind plant of random choose 3 strains in each selected test cell, if these plant lazy weight 3 strains, measure by actual plant quantity, 5 blades of every strain plant random choose, if 5 of blade quantity less thaies, measure by actual blade quantity, it is Net Photosynthetic Rate of portable photosynthesis measurement systematic survey of CI-310 that every blade adopts model, average as this strain plant Net Photosynthetic Rate, 3 test unit mean values are as this strain plant population Net Photosynthetic Rate in this period pilot region;
(3) by the method for (2) step, at the trial between in the span, obtain the Net Photosynthetic Rate C_D of this kind plant colony in the different period pilot regions, carry out [0,1] normalized, obtain the data C_D after the normalization
1
3) make up the biomimetic type kernel function;
4) set up SVM training set and forecast set:
With data M _ D
1Be divided into two parts data M _ D by different time sections
11And M_D
12, according to M_D
1Be divided into two parts data M _ D
11And M_D
12Time period, with data C_D
1Be divided into two parts data C_D by the corresponding time period
11And C_D
12With the data M _ D of first
11And C_D
11As the training set of SVM, with second portion data M _ D
12And C_D
12Forecast set as SVM;
5) instrument of setting up forecast model is selected and parameter optimization;
6) prediction colony Net Photosynthetic Rate.
2. according to the described kind of groups Net Photosynthetic Rate Forecasting Methodology of claim 1, it is characterized in that the step of described structure biomimetic type kernel function is as follows:
1) with Gaussian kernelK
1(x, x
i)=exp (γ || x-x
i||
2) and polynomial kernelK
2(x, x
i)=k (<x, x
i〉+c) is the benchmark kernel function, and wherein, γ and c are parameter, x and x
iBe the lower dimensional space multi-C vector.With K
1(x, x
i)=exp (γ || x-x
i||
2) import among the libsvm, as the SVM kernel function, training set data is trained, adopt grid-search search optimal parameter, determine the γ value the highest to the training set predictablity rate, be defined as the Γ value;
2) seek out Gaussian kernel characteristic curve tangent slope maximal value diff (K under the Γ value
1(x
m, x
i)) and minimum value diff (K
1(x
n, x
i)), and corresponding point of contact coordinate (x
m, K
1(x
m, x
i)), (x
n, K
1(x
n, x
i)), such point of contact is called break;
3) according to break coordinate (x
m, K
1(x
m, x
i)), (x
n, K
1(x
n, x
i)) and slope diff (K
1(x
m, x
i))/ξ, diff (K
1(x
n, x
i))/ξ, wherein ξ is real variable, determines polynomial kernel expression formula, expression formula is by K
2' (x, x
i)=diff (K
1(x
m, x
i))/ξ/x
i* (x, x
i+ c) and K
2" (x, x
i)=diff (K
1(x
n, x
i))/ξ/x
i* (<x, x
iThe two parts of 〉+c) are formed;
4) make up the biomimetic type kernel function:
K
bsf(x,x
i)=exp(-Γ||x-x
i||
2)+diff(exp(-Γ||x
B-x
i||
2)/ξ/x
i×(<x,x
i>+c)
Wherein: || x||>|| x
i|| the time, x
BBe taken as x
n, || x||<|| x
i|| the time, x
BBe taken as x
m, || x||=||x
i|| the time, with x and x
iBe classified as similarly, need not to use kernel function that it is calculated;
Biomimetic type kernel function K
Bsf(x, x
i) can realize the adjustment to global data and local data by the adjusting of parameter Γ and ξ, to adapt to the requirement that different pieces of information is sorted out.
3. according to the described kind of groups Net Photosynthetic Rate Forecasting Methodology of claim 1, it is characterized in that the described instrument of setting up forecast model selects and parameter optimization refers to:
Adopt MATLAB with biomimetic type kernel function K
Bsf(x, x
i) put into the libsvm tool box, the position that replaces original RBF kernel function, realize grid-search method optimization parameter by function S VMcgForRegress (), utilize grid-search method initial optimization parametric procedure, in function S VMcgForRegress (), adopt Gaussian kernelK
1(x, x
i), by [bestmse, bestc, bestg]=SVMcgForRegress (C_D
11, M_D
11,-8,8 ,-8,8), can obtain parameter b estc and bestg, bestc is best penalty parameter c value, bestg is best parameter γ value, i.e. Γ value;
In function S VMcgForRegress (), adopt biomimetic type kernel function K
Bsf(x, x
i), by [bestmse, bestc, bestCMG]=SVMcgForRegress (C_D
11, M_D
11, bestc, bestc ,-8,8), can obtain optimal parameter bestCMG, i.e. the ξ value.
4. according to the described kind of groups Net Photosynthetic Rate Forecasting Methodology of claim 1, it is characterized in that the step of described prediction colony Net Photosynthetic Rate is as follows:
1) obtains forecast model model
Utilize the svmtrain () in the libsvm tool box, namely by model=svmtrain (C_D
11, M_D
11Cmd), can obtain forecast model model, cmd=['-c' wherein, num2str (bestc), '-g', num2str (bestg), '-s3-p0.01-t2'], s is made as 3 representatives and adopts e – SVR formula, p is made as 0.01 representative the value of loss function among the e-SVR is set, and t is made as 2, and to represent the kernel function that adopts among the SVM be K
Bsf(x, x
i); The information that in model, has comprised kernel function type, support vector number and support vector coefficient scope in decision function;
2) obtain the predicted value predict of colony's Net Photosynthetic Rate
Utilize the svmpredict () in the libsvm tool box, by [predict, mse]=svmpredict (C_D
12, M_D
12, model) realize utilizing M_D
12To C_D
12Predict, obtain the predicted value predict of colony's Net Photosynthetic Rate;
3) reliability of judgement forecast model model
Simultaneously, pass through C_D
12Calculate between the two coefficient R with predict, judge the accuracy of predicted value predict, i.e. the reliability of forecast model model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310115387.0A CN103234916B (en) | 2013-04-06 | 2013-04-06 | Prediction method for net photosynthetic rate of population |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310115387.0A CN103234916B (en) | 2013-04-06 | 2013-04-06 | Prediction method for net photosynthetic rate of population |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103234916A true CN103234916A (en) | 2013-08-07 |
CN103234916B CN103234916B (en) | 2015-03-18 |
Family
ID=48882967
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310115387.0A Expired - Fee Related CN103234916B (en) | 2013-04-06 | 2013-04-06 | Prediction method for net photosynthetic rate of population |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103234916B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104914053A (en) * | 2015-05-15 | 2015-09-16 | 中国农业大学 | Tree canopy photosynthetic character prediction model construction and photosynthetic character detection |
CN105678405A (en) * | 2015-12-31 | 2016-06-15 | 西北农林科技大学 | Modeling method for cucumber GA-RBF photosynthetic rate prediction model integrated with stomatal conductance |
CN105675551A (en) * | 2016-01-14 | 2016-06-15 | 中国矿业大学(北京) | Estimation method of plant photosynthetic fixed carbon content |
WO2016180245A1 (en) * | 2015-05-14 | 2016-11-17 | 中国科学院上海生命科学研究院 | Method for calculating photosynthetic rate of crown |
CN108229065A (en) * | 2018-02-06 | 2018-06-29 | 北京林业大学 | A kind of construction method and Forecasting Methodology of the Net Photosynthetic Rate prediction model based on xylophyta leaf morphology feature |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN2204418Y (en) * | 1994-06-03 | 1995-08-02 | 北京农业大学 | Photosynthetic transpiration instrument |
CN1387034A (en) * | 2002-06-18 | 2002-12-25 | 华南师范大学 | Method and device for quickly measuring photosynthesizing speed of delay light generated by visiblelight inducing |
US20030174328A1 (en) * | 2002-03-11 | 2003-09-18 | Edgar Russell | Calibration system and method for calibration of various types of polarimeters |
CN1687772A (en) * | 2005-04-19 | 2005-10-26 | 华南师范大学 | Quick detector for photosynthetic capacity of irradiation from light emitting diode |
CN1837787A (en) * | 2006-04-19 | 2006-09-27 | 南京大学 | Non-destructive precise determination method for biophysical parameters of cotton |
CN102313699A (en) * | 2011-05-26 | 2012-01-11 | 北京农业信息技术研究中心 | Estimation method of total nitrogen content in crop canopy leaf |
-
2013
- 2013-04-06 CN CN201310115387.0A patent/CN103234916B/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN2204418Y (en) * | 1994-06-03 | 1995-08-02 | 北京农业大学 | Photosynthetic transpiration instrument |
US20030174328A1 (en) * | 2002-03-11 | 2003-09-18 | Edgar Russell | Calibration system and method for calibration of various types of polarimeters |
CN1387034A (en) * | 2002-06-18 | 2002-12-25 | 华南师范大学 | Method and device for quickly measuring photosynthesizing speed of delay light generated by visiblelight inducing |
CN1687772A (en) * | 2005-04-19 | 2005-10-26 | 华南师范大学 | Quick detector for photosynthetic capacity of irradiation from light emitting diode |
CN1837787A (en) * | 2006-04-19 | 2006-09-27 | 南京大学 | Non-destructive precise determination method for biophysical parameters of cotton |
CN102313699A (en) * | 2011-05-26 | 2012-01-11 | 北京农业信息技术研究中心 | Estimation method of total nitrogen content in crop canopy leaf |
Non-Patent Citations (2)
Title |
---|
武海巍: "光合有效辐射预测模型的核函数组合优化", 《农业机械学报》 * |
武海巍等: "基于参数优化支持向量机的林下参净光合速率预测模型", 《光谱学与光谱分析》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2016180245A1 (en) * | 2015-05-14 | 2016-11-17 | 中国科学院上海生命科学研究院 | Method for calculating photosynthetic rate of crown |
CN106295093A (en) * | 2015-05-14 | 2017-01-04 | 中国科学院上海生命科学研究院 | A kind of method calculating canopy photosynthesis speed |
CN104914053A (en) * | 2015-05-15 | 2015-09-16 | 中国农业大学 | Tree canopy photosynthetic character prediction model construction and photosynthetic character detection |
CN104914053B (en) * | 2015-05-15 | 2017-07-28 | 中国农业大学 | Build trees canopy photosynthesis trait predictive model and detection Photosynthetic Characters |
CN105678405A (en) * | 2015-12-31 | 2016-06-15 | 西北农林科技大学 | Modeling method for cucumber GA-RBF photosynthetic rate prediction model integrated with stomatal conductance |
CN105678405B (en) * | 2015-12-31 | 2017-02-22 | 西北农林科技大学 | Modeling method for cucumber GA-RBF photosynthetic rate prediction model integrated with stomatal conductance |
CN105675551A (en) * | 2016-01-14 | 2016-06-15 | 中国矿业大学(北京) | Estimation method of plant photosynthetic fixed carbon content |
CN105675551B (en) * | 2016-01-14 | 2018-12-21 | 中国矿业大学(北京) | A method of estimation plant photosynthesis carbon capacity |
CN108229065A (en) * | 2018-02-06 | 2018-06-29 | 北京林业大学 | A kind of construction method and Forecasting Methodology of the Net Photosynthetic Rate prediction model based on xylophyta leaf morphology feature |
CN108229065B (en) * | 2018-02-06 | 2019-03-26 | 北京林业大学 | A kind of construction method and prediction technique of the Net Photosynthetic Rate prediction model based on xylophyta leaf morphology feature |
Also Published As
Publication number | Publication date |
---|---|
CN103234916B (en) | 2015-03-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103234916B (en) | Prediction method for net photosynthetic rate of population | |
CN107169258B (en) | A kind of multi-source weather information data assimilation method and its application in rainfall forecast | |
CN108921351A (en) | Crop production forecast method based on trend yield and Meteorological Output | |
CN111898922B (en) | Multi-scale crop yield assessment method and system | |
CN107316116A (en) | A kind of leafy vegetable production prediction method | |
CN111855589A (en) | Remote sensing inversion model and method for rice leaf nitrogen accumulation | |
CN111855591A (en) | Rice overground part carbon-nitrogen ratio remote sensing inversion model and method | |
CN102524024A (en) | Crop irrigation system based on computer vision | |
CN107423850A (en) | Region corn maturity period Forecasting Methodology based on time series LAI curve integral areas | |
CN104778349B (en) | One kind is used for rice table soil nitrogen application Classified Protection | |
CN109325633A (en) | A kind of weather forecast set member choosing method and system | |
CN108520127A (en) | A kind of EO-1 hyperion inversion method of seeds leaf area index | |
CN108228538A (en) | SO in a kind of city2The Forecasting Methodology of concentration value | |
CN109214591B (en) | Method and system for predicting aboveground biomass of woody plant | |
CN103278467A (en) | Rapid nondestructive high-accuracy method with for identifying abundance degree of nitrogen element in plant leaf | |
CN111044516A (en) | Remote sensing estimation method for chlorophyll content of rice | |
CN112270131A (en) | Remote sensing inversion model and method for rice leaf area index based on ARD regression algorithm | |
CN205262963U (en) | A remote sensing system that is used for wheat blade chlorophyll content estimation | |
Omar et al. | Seasonal clustering forecasting technique for intelligent hourly solar irradiance systems | |
CN114460015A (en) | Method and model for predicting chlorophyll content of canopy of rice in cold region | |
Nie et al. | Spatial prediction of soil moisture content in winter wheat based on machine learning model | |
CN112362812A (en) | Remote sensing inversion model and method for rice leaf chlorophyll carotenoid content ratio based on Lars algorithm | |
Küçükdermenci | Design of A Cost-Effective Weather Station with Rain Forecast | |
Ruiz-Ramos et al. | ALAMEDA, a structural–functional model for faba bean crops: morphological parameterization and verification | |
Luo et al. | Study on wind-induced fatigue of heliostat based on artificial neural network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150318 Termination date: 20170406 |
|
CF01 | Termination of patent right due to non-payment of annual fee |