CN105184400A - Tobacco field soil moisture prediction method - Google Patents

Abstract

The invention provides a tobacco field soil moisture prediction method. The method provides a tobacco field soil moisture prediction model based on principal component analysis and radial basis function (RBF) neural network. The method is characterized by, to begin with, carrying analysis on influencing factors of tobacco field soil moisture and determining influencing parameters; then, eliminating correlation of original input layer data by utilizing a principal component analysis method to obtain a group of uncorrelated new input variables; and finally, carrying out soil moisture prediction with the reconstructed training sample space being as the input of the RBF neural network, and the method is proved to be high in predication precision through instance simulation and comparative analysis.

Description

A kind of vega soil moisture Forecasting Methodology

Technical field

The present invention relates to monitoring field, tobacco leaf planting land for growing field crops, more specifically, relate to a kind of vega soil moisture Forecasting Methodology.

Background technology

The change of vega soil moisture all has a great impact the growth of tobacco leaf, output and tobacco aroma quality, amount and utilization rate of fertilizer, cigarette strain root growth, native transmissibility disease and pest etc.Therefore, vega soil moisture prediction is the important evidence of tabacco water geotechnique journey production management, especially determines that fertilization type, quantity and tobacco soil-borne disease prediction have great importance to the planning of cigarette district plantation.

Soil moisture is a complicated System with Nonlinear Coupling, affects very large, under normal conditions, only take into full account each factor affecting its prediction, could set up the forecast model meeting actual needs by factors such as external environments.Conventional Forecast of Soil Moisture Content model has experience equation, water balance method, Soil Moisture Dynamics method, time series anylysis model etc.These methods are when required boundary condition all possesses, great majority can obtain satisfied result, but there is following problem in actual applications: 1) need to obtain qualified various parameter by test determination or by statistical study, because natural conditions vary, need to drop into a large amount of manpower and materials, and the model set up is too complicated, hampers the actual of model and applies; 2) conventional aqueous forecast model needs the parameter of input substantially to determine, if lack part input quantity wherein, will have a strong impact on predicting the outcome of model, and practical application, most parameter is difficult to obtain, and makes model be difficult to application.

In recent years, in order to theorize, basis is solid, form is relatively simple, parameter is easy to obtain, and can meet the Forecast of Soil Moisture Content model of actual needs, many scholars adopt neural net method to be studied Prediction of Soil Water Content, and obtain certain achievement.Because neural network does not need to set up accurate mathematical model, just can realize the Nonlinear Mapping from input end to output terminal, therefore be widely applied in Prediction of Soil Water Content.At present, feature and the Problems existing of these researchs have: 1) major part is all adopt error back propagation (backpropagation, BP) neural network carries out prediction modeling, and the speed of convergence that BP neural network has learning algorithm is slow, the training airplane duration of model, global error is easily absorbed in the deficiencies such as local minimum in the training process; 2) affect the factor huge number of soil moisture and complicated and changeable, the Forecasting Methodology of traditional neural network is simply all factors or artificially using the input of part principal element as neural network.But, owing to there is stronger correlativity between the element of each ridge, if all factors artificially simplified simply or merge, loss or the overlap of a large amount of useful information can be caused, thus impact prediction precision.

Summary of the invention

The invention provides the vega soil moisture Forecasting Methodology that a kind of precision of prediction is higher.

In order to reach above-mentioned technique effect, technical scheme of the present invention is as follows:

A kind of vega soil moisture Forecasting Methodology, comprises the following steps:

S1: determine the parameter affecting vega soil moisture;

S2: set up vega soil Water Prediction Models;

S3: collect sample data, namely starts to carry out soil moisture analyses and prediction after training to the vega soil Water Prediction Models set up.

Further, the detailed process of described step S1 is as follows:

Utilize soil moisture empirical formula Q ₂=P'+I-S-ET-L+Q ₁, in conjunction with tobacco leaf production practical experience, the parametric variable obtaining affecting vega soil moisture main has: the storage capacity-Q of the first unit area of each growth phase of vega tobacco leaf ₁, the storage capacity-Q of vega tobacco leaf each growth phase end unit area ₂the each growth phase evapotranspiration amount-ET of vega tobacco leaf, the each growth phase temperature on average-T of vega tobacco leaf, the each growth phase precipitation-P' of vega tobacco leaf, vega tobacco leaf each growth phase sunshine-duration-t, the each growth phase irrigation quantity-I of vega tobacco leaf, each growth phase run-off-S of vega tobacco leaf, each growth phase leakage-L of vega tobacco leaf.

Further, the detailed process of described step S2 is as follows:

1) the sample data matrix X standardization, to n × m, each row of X correspond to a variable, and every a line corresponds to a sample:

$X_s=\[X-{\left(11...1\right)}^{T}M\]diag\left(\frac{1}{s_1},\frac{1}{s_2},\mathrm{...},\frac{1}{s_m}\right),X\∈R^{n\×m}$

M=[m in formula ₁m ₂m _m] be the average of X, s=[s ₁s ₂s _m] be the standard deviation of X;

2), to X _scarry out pivot analysis to obtain $X^{\′}=t_1{p}_1^T+t_2{p}_2^T+...+t_k{p}_k^T+...+t_m{p}_m^T,$ T in formula ₁, t ₂, t _m∈ R ⁿfor score vector, p ₁, p ₂, p _m∈ R ^mfor load vector, and replace Main change in data and major component by a front k pivot, then X' is expressed as approx

3), make $X^{\′}=X_p^{\′}=t_1{p}_1^T+t_2{p}_2^T+...+t_k{p}_k^T=\{X_{p_1}^{\′},X_{p_2}^{\′},...,X_{p_k}^{\′}\},$ Radial basis function neural network is utilized to come X' _panalyzing and processing obtains forecast model: r in formula _i(X') be the radial basis function in hidden layer, i is the interstitial content of hidden layer, i=1, and 2, l, s are the number of output neuron, s=1,2, and, h, W _isbe the weights of i-th Hidden unit to output unit.

Further, described step 3) middle R _i(X') computation process is as follows:

$R_i\left(X^{\′}\right)=\exp \[-\frac{1}{2}{\[\frac{\left|\right|X^{\′}-c_i\left|\right|}{{\mathrm{\σ}}_i}\]}^2\]$

C in formula _ibe the center of the radial basis function that i-th Hidden unit is corresponding, σ _ibe the perceived device central point width that i-th Hidden unit is corresponding, be used for the sensitivity of regulating networks, || X'-c _i|| be vectorial X'-c _inorm, represent X' and c _ieuclidean distance.

Further, determine that the detailed process at the center of the radial basis function that i-th Hidden unit is corresponding is as follows:

Be provided with A group sample data, choose jth group input vector, calculate euclideam norm: σ _i(j)=|| X'(j)-c _i(j-1) || ², find out with input amendment apart from minimum center c _min, adjustment center is:

c _min(j)＝c _min(j-1)+a(X'(j)-c _min(j-1))

In formula, a is learning rate, often learns once, turns an a down, and other each center vectors remain unchanged, then calculates the euclideam norm σ after adjustment _min(j)=|| X'(j)-c _min(j) || ², repeat the deterministic process of center and width, make training sample minimum apart from the distance at this center, c (j) now namely centered by.

Compared with prior art, the beneficial effect of technical solution of the present invention is:

The present invention proposes a kind of vega soil Water Prediction Models based on pivot analysis and radial basis function (radialbasisfunction, RBF) neural network, first the influence factor of vega soil moisture is analyzed, determine affecting parameters; Then utilize principle component analysis to eliminate the correlativity of original input layer data, obtain one group of incoherent new input variable each other; Finally using reconstruct training sample space as the input of RBF neural, carry out Prediction of Soil Water Content, and show that the method precision of prediction is higher by Case Simulation and comparative analysis.

Accompanying drawing explanation

Fig. 1 is the inventive method process flow diagram.

Embodiment

Accompanying drawing, only for exemplary illustration, can not be interpreted as the restriction to this patent;

In order to better the present embodiment is described, some parts of accompanying drawing have omission, zoom in or out, and do not represent the size of actual product;

To those skilled in the art, in accompanying drawing, some known features and explanation thereof may be omitted is understandable.

Below in conjunction with drawings and Examples, technical scheme of the present invention is described further.

Embodiment 1

As shown in Figure 1, a kind of vega soil moisture Forecasting Methodology, comprises the following steps:

S1: determine the parameter affecting vega soil moisture;

S2: set up vega soil Water Prediction Models;

S3: collect sample data, namely starts to carry out soil moisture analyses and prediction after training to the vega soil Water Prediction Models set up.

Further, the detailed process of step S1 is as follows:

Utilize soil moisture empirical formula Q ₂=P'+I-S-ET-L+Q ₁, in conjunction with tobacco leaf production practical experience, the parametric variable obtaining affecting vega soil moisture main has: the storage capacity-Q of the first unit area of each growth phase of vega tobacco leaf ₁, the storage capacity-Q of vega tobacco leaf each growth phase end unit area ₂the each growth phase evapotranspiration amount-ET of vega tobacco leaf, the each growth phase temperature on average-T of vega tobacco leaf, the each growth phase precipitation-P' of vega tobacco leaf, vega tobacco leaf each growth phase sunshine-duration-t, the each growth phase irrigation quantity-I of vega tobacco leaf, each growth phase run-off-S of vega tobacco leaf, each growth phase leakage-L of vega tobacco leaf.

Further, the detailed process of step S2 is as follows:

1) the sample data matrix X standardization, to n × m, each row of X correspond to a variable, and every a line corresponds to a sample:

$X_s=\[X-{\left(11...1\right)}^{T}M\]diag\left(\frac{1}{s_1},\frac{1}{s_2},\mathrm{...},\frac{1}{s_m}\right),X\∈R^{n\×m}$

M=[m in formula ₁m ₂m _m] be the average of X, s=[s ₁s ₂s _m] be the standard deviation of X;

2), to X _scarry out pivot analysis to obtain $X^{\′}=t_1{p}_1^T+t_2{p}_2^T+...+t_k{p}_k^T+...+t_m{p}_m^T,$ T in formula ₁, t ₂, t _m∈ R ⁿfor score vector, p ₁, p ₂, p _m∈ R ^mfor load vector, and replace Main change in data and major component by a front k pivot, then X' is expressed as approx

3), make $X^{\′}=X_p^{\′}=t_1{p}_1^T+t_2{p}_2^T+...+t_k{p}_k^T=\{X_{p_1}^{\′},X_{p_2}^{\′},...,X_{p_k}^{\′}\},$ Radial basis function neural network is utilized to come X' _panalyzing and processing obtains forecast model: r in formula _i(X') be the radial basis function in hidden layer, i is the interstitial content of hidden layer, i=1, and 2, l, s are the number of output neuron, s=1,2, and, h, W _isbe the weights of i-th Hidden unit to output unit.

Further, described step 3) middle R _i(X') computation process is as follows:

$R_i\left(X^{\′}\right)=\exp \[-\frac{1}{2}{\[\frac{\left|\right|X^{\′}-c_i\left|\right|}{{\mathrm{\σ}}_i}\]}^2\]$

C in formula _ibe the center of the radial basis function that i-th Hidden unit is corresponding, σ _ibe the perceived device central point width that i-th Hidden unit is corresponding, be used for the sensitivity of regulating networks, || X'-c _i|| be vectorial X'-c _inorm, represent X' and c _ieuclidean distance.

Further, determine that the detailed process at the center of the radial basis function that i-th Hidden unit is corresponding is as follows:

Be provided with A group sample data, choose jth group input vector, calculate euclideam norm: σ _i(j)=|| X'(j)-c _i(j-1) || ², find out with input amendment apart from minimum center c _min, adjustment center is:

c _min(j)＝c _min(j-1)+a(X'(j)-c _min(j-1))

In formula, a is learning rate, often learns once, turns an a down, and other each center vectors remain unchanged, then calculates the euclideam norm σ after adjustment _min(j)=|| X'(j)-c _min(j) || ², repeat the deterministic process of center and width, make training sample minimum apart from the distance at this center, c (j) now namely centered by.

Prosperous long-term for In Guangdong Province tobacco growing, the variable of selection is: the storage capacity-Q of the first unit area of each growth phase of vega tobacco leaf ₁the each growth phase evapotranspiration amount-ET of vega tobacco leaf, the each growth phase temperature on average-T of vega tobacco leaf, the each growth phase precipitation-P' of vega tobacco leaf, vega tobacco leaf each growth phase sunshine-duration-t, collect 2010 to 2014 38 sample datas, will wherein 33 groups of data as training sample, using last 5 groups as test sample book.

33 groups of training sample data of collecting are carried out pivot analysis.Table 1 gives the major component coefficient that raw sample data obtains after pivot analysis, and table 2 gives principal component contributor rate, contribution rate of accumulative total, and table 3 is model prediction result and predicts with traditional BP comparing of carrying out.

The major component coefficient that table 1 raw sample data obtains after pivot analysis

Table 2

Major component	Eigenwert	Contribution rate	Contribution rate of accumulative total
				PC1	2.198	55.9	55.9
PC2	0.9546	19.3	75.3
				PC3	0.5216	11.0	86.2
PC4	0.4762	9.1	95.4
				PC5	0.2451	4.5	100

Table 3 predicts the outcome

First the present invention analyzes the influence factor of vega soil moisture, determines affecting parameters; Then utilize principle component analysis to eliminate the correlativity of original input layer data, obtain one group of incoherent new input variable each other; Finally using the input of the training sample space of reconstruct as RBF neural, carry out Prediction of Soil Water Content, can be found out by table 3, the method precision of prediction is apparently higher than total factor traditional BP nerve net, the method can reflect the change of vega soil moisture exactly, especially determines that fertilization type, quantity and tobacco soil-borne disease prediction serve certain directive function to the planning of cigarette district plantation.

The corresponding same or analogous parts of same or analogous label;

Describe in accompanying drawing position relationship for only for exemplary illustration, the restriction to this patent can not be interpreted as;

Obviously, the above embodiment of the present invention is only for example of the present invention is clearly described, and is not the restriction to embodiments of the present invention.For those of ordinary skill in the field, can also make other changes in different forms on the basis of the above description.Here exhaustive without the need to also giving all embodiments.All any amendments done within the spirit and principles in the present invention, equivalent to replace and improvement etc., within the protection domain that all should be included in the claims in the present invention.

Claims (5)

1. a vega soil moisture Forecasting Methodology, is characterized in that, comprises the following steps:

S1: determine the parameter affecting vega soil moisture;

S2: set up vega soil Water Prediction Models;

S3: collect sample data, namely starts to carry out soil moisture analyses and prediction after training to the vega soil Water Prediction Models set up.

2. vega soil moisture Forecasting Methodology according to claim 1, it is characterized in that, the detailed process of described step S1 is as follows:

Utilize soil moisture empirical formula Q ₂=P'+I-S-ET-L+Q ₁, in conjunction with tobacco leaf production practical experience, the parametric variable obtaining affecting vega soil moisture main has: the storage capacity-Q of the first unit area of each growth phase of vega tobacco leaf ₁, the storage capacity-Q of vega tobacco leaf each growth phase end unit area ₂the each growth phase evapotranspiration amount-ET of vega tobacco leaf, the each growth phase temperature on average-T of vega tobacco leaf, the each growth phase precipitation-P' of vega tobacco leaf, vega tobacco leaf each growth phase sunshine-duration-t, the each growth phase irrigation quantity-I of vega tobacco leaf, each growth phase run-off-S of vega tobacco leaf, each growth phase leakage-L of vega tobacco leaf.

3. vega soil moisture Forecasting Methodology according to claim 1, it is characterized in that, the detailed process of described step S2 is as follows:

1) the sample data matrix X standardization, to n × m, each row of X correspond to a variable, and every a line corresponds to a sample:

$X_s=\[X-{\left(11...1\right)}^{T}M\]diag\left(\frac{1}{s_1},\frac{1}{s_2},\mathrm{...},\frac{1}{s_m}\right),X\∈R^{n\×m}$

M=[m in formula ₁m ₂m _m] be the average of X, s=[s ₁s ₂s _m] be the standard deviation of X;

2), to X _scarry out pivot analysis to obtain $X^{\′}=t_1{p}_1^T+t_2{p}_2^T+...+t_k{p}_k^T+...+t_m{p}_m^T,$ T in formula ₁, t ₂, t _m∈ R ⁿfor score vector, p ₁, p ₂, p _m∈ R ^mfor load vector, and replace Main change in data and major component by a front k pivot, then X' is expressed as approx

3), make $X^{\′}=X_p^{\′}=t_1{p}_1^T+t_2{p}_2^T+...+t_k{p}_k^T=\{X_{p_1}^{\′},X_{p_2}^{\′},...,X_{p_k}^{\′}\},$ Radial basis function neural network is utilized to come X' _panalyzing and processing obtains forecast model: r in formula _i(X') be the radial basis function in hidden layer, i is the interstitial content of hidden layer, i=1, and 2, l, s are the number of output neuron, s=1,2, and, h, W _isbe the weights of i-th Hidden unit to output unit.

4. vega soil moisture Forecasting Methodology according to claim 3, is characterized in that, described step 3) middle R _i(X') computation process is as follows:

$R_i\left(X^{\′}\right)=\exp \[-\frac{1}{2}{\[\frac{\left|\right|X^{\′}-c_i\left|\right|}{{\mathrm{\σ}}_i}\]}^2\]$

C in formula _ibe the center of the radial basis function that i-th Hidden unit is corresponding, σ _ibe the perceived device central point width that i-th Hidden unit is corresponding, be used for the sensitivity of regulating networks, ‖ X'-c _i‖ is vectorial X'-c _inorm, represent X' and c _ieuclidean distance.

5. vega soil moisture Forecasting Methodology according to claim 4, is characterized in that, determines that the detailed process at the center of the radial basis function that i-th Hidden unit is corresponding is as follows:

Be provided with A group sample data, choose jth group input vector, calculate euclideam norm: σ _i(j)=‖ X'(j)-c _i(j-1) ‖ ², find out with input amendment apart from minimum center c _min, adjustment center is:

c _min(j)＝c _min(j-1)+a(X'(j)-c _min(j-1))

In formula, a is learning rate, often learns once, turns an a down, and other each center vectors remain unchanged, then calculates the euclideam norm σ after adjustment _min(j)=‖ X'(j)-c _min(j) ‖ ², repeat the deterministic process of center and width, make training sample minimum apart from the distance at this center, c (j) now namely centered by.