CN101480143A - Method for predicating single yield of crops in irrigated area - Google Patents

Abstract

The invention provides a method for forecasting the crop yield of per unit area in an irrigation district. Respective forecasting models of economic yield and environmental yield are respectively built, and then a combined forecasting model of crop field is obtained through the weighted sum of weight coefficients. The method weakens the respective defects of different models to enable the two models to mutually and self-adaptively complement each other, and also weakens the system analysis limitations brought by unilaterally forecasting the crop yield according to the socio-economic factors or natural environment factors simultaneously.

Description

A kind of method of predicting irrigated area crop yield amount

Technical field

The invention belongs to irrigated area analysis of ecological system technical field, be specifically related to a kind of irrigated area crop yield forecast method based on gray theory and evolution algorithm.

Background technology

Along with the continuous development of agricultural commodities marketization, great changes have taken place for crop varieties and pattern of farming, and irrigated area agricultural production administrative department and local agricultural producer and agriculture-related enterprise are more and more higher to the requirement of crop production forecast information.The variation of prediction crop yield all has crucial meaning to local policy macro adjustments and controls and regional structure adjustment.Set up suitable forecast model, strengthen crop production forecast research, can strengthen the specific aim of irrigated area agricultural weather service, increase accuracy for predicting, satisfy higher agricultural weather service needs.By carrying out yield prediction comparatively accurately, can also improve the productivity ratio of agricultural and related industry, strengthen directive function to agricultural production.

The factor that influences the irrigated area crop yield is intricate, and existing social human factor has factor of natural environment again, and different functional departments are often single to go out to send to carry out analyses and prediction from self professional angle.For factor of natural environment, some crop growth phase is longer, the meteorological condition influence is complicated, such as rainfall, air pressure, soil fertility, solar radiation quantity, temperature, sunshine, atmospheric circulation, soil moisture content, air drying power or the like, and quantifiable irrigated area man-made environment factor (such as irrigation quantity, fertilizing amount etc.), use methods such as traditional climatic statistics analysis, output factor analysis, agronomy analysis, predictablity rate is not high.For socio-economic factor, often be difficult to quantitative analysis especially, such as local policies and measures, local irrigation technique, peasant planting level, fertilizer quality or the like.Along with the progressively development of soft computing technique for many years, agriculture and meteorological scientific and technical personnel are constantly attempting introducing multiple forecasting procedure based on soft computational science, as methods such as neutral net, gray prediction, markoff chain predictions, but in actual applications, certain static models of single use often are difficult to obtain gratifying effect, because every kind of method is not omnipotent after all, the many factors of influence that not necessarily are fit to phylogeny, each comfortable its mathematical theory is according to last defective or the limitation that himself is all arranged in addition.

Summary of the invention

The object of the present invention is to provide a kind of method of predicting the irrigated area crop yield, this method is taken into account environmental factor and socio-economic factor, has overcome the defective of single model, and is simple, has higher reasonability and accuracy.

A kind of method of predicting irrigated area crop yield amount, the single rate of known preceding n, following single rate is predicted in n 〉=5, this method is specific as follows:

Steps A adopts grey modeling method according to the single rate of 1～n, makes up the economic flow rate predictor formula Y in irrigated area _e(k '), k ' are to be predicted year;

Step B behind the employing grey correlation selection environment main gene, uses neural network modeling approach to make up the environment yield prediction formula Y in irrigated area according to the single rate of 1～n _i(k '), k ' are to be predicted year;

Step C determines economic flow rate weight coefficient α and environment output weight coefficient β, alpha+beta=1, and α and economic flow rate predicated error are inversely proportional to, and β and environment yield prediction error are inversely proportional to;

Described economic flow rate predicated error is defined as: according to the actual single rate of 1～v, make up economic flow rate correction formula according to the mode of steps A

Adopt formula Prediction f=v+1 ..., the economic flow rate during n will predict the outcome and v+1 ..., the actual single rate of n relatively obtains the economic flow rate predicated error;

Described environment yield prediction error is defined as: according to the actual single rate of 1～v, according to the mode constructing environment output correction formula of step (B)

Use formula

Prediction f=v+1 ..., the environment output during n will predict the outcome and v+1 ..., the actual single rate of n relatively obtains environment yield prediction error;

Wherein, 1 " v＜n;

Step D adopts single rate formula Y (k ')=α Y _e(k ')+β Y _i(k ') prediction irrigated area is at the single rate in k ' year,

Described step C is specially:

Step C1 extracts the actual single rate of 1～v, predicts the irrigated area at f=v+1 according to the mode of steps A ..., the economic flow rate predicted value Y of n _e(f); Extract the actual single rate of 1～v, and environment main gene measured data, predict the irrigated area at f=v+1 according to the mode of step B ..., the environment yield prediction value Y of n _i(f);

Step C2 calculates f=v+1 ..., the economic flow rate predicted value Y of n _e(f) with f=v+1 ..., the difference between the single rate of n reality is made sum of squares to these differences and is calculated, and asks for its S1 reciprocal again; Calculate f=v+1 ..., the environment yield prediction value Y of n _i(f) with f=v+1 ..., the difference between the single rate of n reality is made sum of squares to these differences and is calculated, and asks for its S2 reciprocal again;

Step C3 calculates the weight coefficient of economic flow rate $\mathrm{\&alpha;}=\frac{S1}{\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="A200910060634E00201.png"\; state="\{\{state\}\}">S</figure-callout>}1+S2},$ The weight coefficient of environment output $\mathrm{\&beta;}=\frac{S2}{\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="A200910060634E00201.png"\; state="\{\{state\}\}">S</figure-callout>}1+S2}.$

Technique effect of the present invention is embodied in:

(1) the present invention is from social economy's productivity and two angles of irrigated area environment (meteorological naturally and artificial planting environment), analyze influence factor separately, thus the irrigated area crop yield is decomposed into economic flow rate and environment output, they are respectively by social human factor, factor of natural environment decision.To economic flow rate and the independent forecast model of setting up separately of environment output, obtain the combination forecasting of crop yield then by the weight coefficient weighted sum respectively.This method different models defective separately that weakened makes that two models are adaptive mutually to learn from other's strong points to offset one's weaknesses, the network analysis limitation that also weakened simultaneously socio-economic factor and factor of natural environment are unilaterally predicted crop yield and brought.

(2) determine by socio-economic factor (local policies and measures, local irrigation technique, peasant planting level, fertilizer quality etc.) owing to economic flow rate, it is difficult to quantitative expression to the influence of per unit area yield, and it is quite complicated, what its reflected is the influence of regular period social productive forces development level to crop yield, therefore also be metastable factor, in the regular period class is a process stably, so it is carried out Grey Prediction Modeling.Grey modeling is a kind of the system that contains uncertain factor to be carried out forecast method, and it is unified between the darky system between white color system.GM (1,1) the gray model Forecasting Methodology is by making to generate the rule of handling the change of searching system to crop annual production initial data, generation has strong regular data sequence, sets up the corresponding differential equation model then, thus the situation of the following output development trend of prediction things.This method has following several respects advantage: 1. can keep higher forecast precision and stronger stability in medium-and long-term forecasting; 2. overcritical large sample amount, even select 5 to 8 small samples can carry out modeling is particularly suitable for predicting under the situation that data are difficult to try to achieve; 3. modeling flexible and convenient, amount of calculation is little, and artificial calculating can realize, and be convenient to debugging and control.

(3) when carrying out the crop economy yield prediction,, strengthened the feasibility and the stability of crop economy yield prediction modeling for traditional Grey Prediction Modeling has added perfect data preanalysis processing, model testing and the supporting flow process of model correction of a cover.Be that simultaneously this gray model implements new breathization processing, up-to-date information is replaced the sequence sliding window of the continuous mobility model of old information, make model have more actual effect and prediction accuracy.Owing to require less to the sample data amount when utilizing gray model prediction crop economy output, recent more data are effective more, and initial data has not required good statistical law, way is that initial data is done corresponding preliminary treatment according to development coefficient area requirement, and prediction data done residual test and correction, passed through these a series of preliminary treatment, check and makeover process, the medium-and long-term forecasting effect of institute's established model is better, but its last-period forecast effect specific diameter is poor to base neural net.So when predicting, at first carry out grey correlation analysis to making substance environment output, choose suitable natural environment main gene, next it is done the radial base neural net modeling.The advantage of radial base neural net is that the short-term forecast precision is very high, it has the advantage that network is approached in the part, can be similar to any nonlinear function, it uses the simulation and the prediction of data in the near future, precision is quite high, but do not do certain check to predicting the outcome, the medium-and long-term forecasting effect is poor than grey forecasting model.So, in order to make full use of the effective information that single model reflects, overcome the defective of single model, reduce the randomness of prediction, improve precision of prediction, the present invention has set up combination forecasting.

The rational crop production forecast of the present invention provides important decision references foundation for local irrigated area agricultural sector arranges procurement of agricultural product, storage, import, outlet and consumption etc.; Be that local government judges agricultural form, carrying out agricultural decision making etc. provides scientific basis, very helpful to the job guide and the policy making of agricultural production and irrigated area economic development.

Description of drawings

Fig. 1 is overall flow figure of the present invention;

Fig. 2 is that the grey that the present invention relates to newly ceases modeling and model testing flow chart.

Embodiment

Carry out this particular problem of crop production forecast for the irrigated area ecosystem, principle based on basic per unit area yield predictor formula, if can carry out compositional modeling targetedly, and make built-up pattern possess the dynamic self-adapting updating ability of new city metabolism, just can reflect intuitively from social productive forces factor and environmental factor and take all factors into consideration influence this problem, therefore can improve the accuracy rate of prediction, thereby better instruct the irrigated area agricultural production.

According to the characteristic of the many factors of influence of crop yield, the irrigated area crop yield is reduced two classes, i.e. economic flow rate and environment output.The crop yield amount is obtained by economic flow rate and the weighted sum of environment output.Computing formula is: Y=α Y _e+ β Y _i(wherein Y represents crop yield amount, Y _e, Y _iBe respectively economic flow rate and environment output, α, β are respectively both weight coefficients separately);

Economic flow rate is determined by socio-economic factor (local policies and measures, local irrigation technique, peasant planting level, fertilizer quality etc.), it is difficult to quantitative expression to the influence of per unit area yield, and it is quite complicated, what its reflected is the influence of regular period social productive forces development level to crop yield, therefore also be metastable factor, in the regular period class is a process stably, therefore is fit to it is set up GM (1,1) grey forecasting model;

Environment output is influenced by a series of natural environments in irrigated area and man-made environment, the continuous variation of many envirment factors such as precipitation, air pressure, solar radiation, temperature, sunshine, atmospheric circulation, soil moisture content, irrigation quantity, fertilizing amount, make environment output constantly fluctuate, therefore environment output is non-stationary stochastic process, has reflected the influence of environmental factor to the irrigated area crop yield;

The present invention is based on dynamic self-adapting compositional modeling Forecasting Methodology such as Fig. 1 of the irrigated area crop yield prediction of gray theory and evolution algorithm, concrete steps are as follows:

1. historical yield data is set up new breath GM (1,1) grey forecasting model.Concrete route following (flow process is as shown in Figure 2):

(1) the historical single rate data of getting before the irrigated area n (n 〉=5) year are formed initial time sequence x ⁽⁰⁾, x ⁽⁰⁾={ x ⁽⁰⁾(1), x ⁽⁰⁾(2) ..., x ⁽⁰⁾(n) }, to x ⁽⁰⁾Generation adds up $x^{\left(1\right)}\left(i\right)=\underset{k=1}{\overset{i}{\mathrm{\&Sigma;}}}{x}^{\left(0\right)}\left(k\right),$ I=1,2 ..., n, i.e. x ⁽¹⁾={ x ⁽¹⁾(1), x ⁽¹⁾(2) ..., x ⁽¹⁾(n) }.Make z ⁽¹⁾Be x ⁽¹⁾Next-door neighbour's average formation sequence, z ⁽¹⁾={ z ⁽¹⁾(2), z ⁽¹⁾(3) ..., z ⁽¹⁾(n) }, wherein $z^{\left(1\right)}k=\frac{1}{2}{x}^{\left(1\right)}(k-1)+\frac{1}{2}{x}^{\left(1\right)}\left(k\right),$ k＝2，3，…，n；

(2) can set up the albefaction differential equation of following form according to gray system theory: $\frac{{\mathrm{dx}}^{\left(1\right)}}{\mathrm{dt}}+{\mathrm{ax}}^{\left(1\right)}=u,$ A: the development coefficient, u: grey actuating quantity, t represents the time, d represents differential.

Equation separate for: $x^{\left(1\right)}\left(t\right)=(x^{\left(1\right)}\left(t_0\right)-\frac{u}{a})e^{-a(t-t_0)}+\frac{u}{a}$ (t ₀Be initial time)

Getting uniformly-spaced to it, centrifugal pump obtains: $x^{\left(1\right)}(k\&prime;+1)=(x^{\left(1\right)}\left(1\right)-\frac{u}{a})e^{-\mathrm{ak}\&prime;}+\frac{u}{a}$ (time response function), k ' gets positive integer value, e=2.71828 for the sampling moment successively since 1;

(3) with method of least squares sequential value is calculated the estimated value that develops coefficient and grey actuating quantity, promptly

Expression formula is: $\left[\frac{\hat{a}}{\hat{u}}\right]={\left(B^{T}B\right)}^{-1}{B}^T{Y}_n$ (T represents matrix is carried out transposition, and-1 expression is to matrix inversion), wherein $B=\left(\begin{array}{cc}{-z}^{\left(1\right)}\left(2\right),& 1\\ {-z}^{\left(1\right)}\left(3\right),& 1\\ \&CenterDot;\&CenterDot;\&CenterDot;& \&CenterDot;\&CenterDot;\&CenterDot;\\ {-z}^{\left(1\right)}\left(n\right),& 1\end{array}\right),$ Y _n=(x ⁽⁰⁾(2), x ⁽⁰⁾(3) ..., x ⁽⁰⁾(n)) ^T, at this moment,

If ●

With current x ⁽¹⁾Return to x ⁽⁰⁾Sequence is done the generation that adds up again, obtains new x ⁽¹⁾Sequence is got back to step (2) then;

If ● $\left|\hat{a}\right|\&GreaterEqual;0.8,$ Then upgrade original yield data, update mode can adopt makes smoothing processing to original yield data, gets back to step (1);

● if $\left|\hat{a}\right|\&le;0.3,$ Can do medium-and long-term forecasting when entering next step;

● if

Preferably only do short-term forecast when entering next step;

(4) will

Separating of the substitution differential equation obtains forecast model: ${\hat{x}}^{\left(1\right)}(k\&prime;+1)=(x^{\left(1\right)}\left(1\right)-\frac{\hat{u}}{\hat{a}})e^{-\hat{a}k\&prime;}+\frac{\hat{u}}{\hat{a}},$ ${\hat{x}}^{\left(1\right)}\left(1\right)=x^{\left(0\right)}\left(1\right),$ When k '=1,2 ..., during n-1, the sequence estimation value is the match value of model, when k ' 〉=n, the sequence estimation value is the predicted value of model;

(5) computation model precision.At first define relative error ε (k '), average relative error ε and precision τ are as follows: $\mathrm{\&epsiv;}(k\&prime;)=\frac{x^{\left(1\right)}(k\&prime;)-{\hat{x}}^{\left(1\right)}(k\&prime;)}{x^{\left(1\right)}(k\&prime;)}\&times;100\%,$ $\stackrel{\&OverBar;}{\mathrm{\&epsiv;}}=\frac{1}{n-1}\underset{k\&prime;=2}{\overset{n}{\mathrm{\&Sigma;}}}\left|\mathrm{\&epsiv;}(k\&prime;)\right|,$ τ＝1-ε；

(6) τ is compared with artificial predetermined precision required value, if less than (promptly not meeting required precision), can near substantial deviation point, set up GM (1 with residual sequence, 1) residual error gray model, original model is revised, and promptly the sequence of forming with the difference of real value and estimated value is made residual sequence ${\mathrm{\&epsiv;}\&prime;}^{\left(1\right)}(k\&prime;)=x^{\left(1\right)}(k\&prime;)-{\hat{x}}^{\left(1\right)}(k\&prime;),$ Set up new Differential Equation Model with GM (1,1) grey modeling method, again with the estimated value of this residual error model

Be added to the sequence estimation value

Get on to time response equation do correction.Revise repeatedly, up to satisfying required precision.

(7) the model prediction sequence of revising after qualified is done the tired generation that subtracts, the tired number of times that subtracts equates with the accumulative frequency of doing before, obtains year crop economy yield prediction value time series to be predicted: ${\hat{x}}^{\left(1\right)}(k\&prime;)={\hat{x}}^{\left(1\right)}(k\&prime;)-{\hat{x}}^{\left(1\right)}(k\&prime;-1).$

With the breath processing newly of above-mentioned Forecasting Methodology, promptly insert fresh information x ⁽⁰⁾(n+1), remove the oldest information x simultaneously ⁽⁰⁾(1), with x ⁽⁰⁾={ x ⁽⁰⁾(2) ..., x ⁽⁰⁾(n), x ⁽⁰⁾(n+1) } compose to x ⁽⁰⁾={ x ⁽⁰⁾(1), x ⁽⁰⁾(2) ..., x ⁽⁰⁾(n) }, rebulid GM (1,1) model, so As time goes on, following disturbance factor has constantly joined in the model goes, and makes model have self metabolic ability, has increased the robustness of forecast model.

2. be that time granularity is done grey correlation analysis to the irrigated area environmental impact factor of making substance environment output and come the selection environment main gene with the year.Envirment factor comprises radiation, temperature, sunshine, atmospheric circulation, soil moisture content, irrigation quantity, fertilizing amount etc., and the purpose of this step is the several environments factor of finding out from envirment factor the yield effect maximum, is called the environment main gene.Concrete steps are as follows:

(1) (be x with crop annual production historical data ⁽⁰⁾={ x ⁽⁰⁾(1), x ⁽⁰⁾(2) ..., x ⁽⁰⁾(n) } sequence) do value just and handle generation auxiliary sequence X ₍₀₎, the monocycle border factor measured data sequence of 1～n is done value just handles generation monocycle border factor sequence X _{(i ')}, i '=1,2 ..., m considers m envirment factor, X altogether _{(i ')}In element number be n, wherein to be the irrigated area make data after value is just handled in the individual envirment factor of the i ' of j actual measurement for j element representation, obtained whole nondimensionalization grey correlation system like this.

(2) calculate the incidence coefficient of each envirment factor: ${\mathrm{\&eta;}}_{(i\&prime;)}\left(j\right)=\frac{\mathrm{\&rho;}\max \max \left|{X_{(i\&prime;)}\left(j\right)-X}_{\left(0\right)}\left(j\right)\right|}{|X_{(i\&prime;)}\left(j\right)-X_{\left(0\right)}\left(j\right)|+\mathrm{\&rho;}\max \max |X_{(i\&prime;)}\left(j\right)-X_{\left(0\right)}\left(j\right)|},$ Max max|X in the formula _{(i ')}(j)-X ₍₀₎(j) | be two-stage maximum difference, | X _{(i ')}(j)-X ₍₀₎(j) | be X ₍₀₎In j element and X _{(i ')}J element between Error Absolute Value, j=1,2 ... n, the value in resolution ρ value in 0 to 1 open interval;

(3) calculate X _(.)With X _{(i ')}The degree of association: $r_{(i\&prime;)}=\frac{1}{n}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}\mathrm{\&eta;}\left(j\right);$

(4) from big to small m sub-sequence permutation formed related preface according to subsequence with the degree of association of auxiliary sequence, related preface has directly reflected each subsequence " primary and secondary " relation to same auxiliary sequence, before choosing

The individual factor has just obtained the main gene of environment output very easily.

3. from the envirment factor measured data of each year, extract the envirment factor measured data of main gene correspondence, be configured to

Single year environment main gene measured data sequence of dimension, as an input sample, then input layer comprises the individual neuron of m ' with it, will be corresponding to the actual year single rate of this year as the target output sample, be that output layer comprises 1 neuron, it is right so just to have constituted a sample.The sample of selecting 1～n makes up radially basic artificial neural network to as training sample.Predict as network simulation input with the single year environment main gene measured data sequence in to be predicted year, obtain to be predicted year environment YIELD PREDICTION of crop value.

4. use said method, to the historical data prediction v+1 of n annual control 1～v ..., the economic flow rate of n and environment output promptly are used for the prediction effect of testing model, 1 " v＜n.

5. with the v of step 4 prediction, v+1 ..., the economic flow rate of n and actual v+1 ..., the single rate of n is compared and is obtained error, calculates the sum of squares of n-v+1 error, asks for its S1 reciprocal again.Ask for v+1 in the same way ..., the sum of squares of the environment output error of n S2 reciprocal, then $\mathrm{\&alpha;}=\frac{S1}{\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="A200910060634E00201.png"\; state="\{\{state\}\}">S</figure-callout>}1+S2},$ $\mathrm{\&beta;}=\frac{S2}{\mathrm{<figure-callout\; id="1"\; label="S"\; filenames="A200910060634E00201.png"\; state="\{\{state\}\}">S</figure-callout>}1+S2}.$ The every prediction of built-up pattern 1 year, just up-to-date economic flow rate, environment yield prediction value and actual production value are inserted model, the test samples of the oldest grey forecasting model, radial base neural net model is rejected from model, recomputate S1, S2, α, β, so, the weight coefficient of two models is brought in constant renewal in built-up pattern is predicted year by year.

. above-mentioned the 1st step, the 3rd is gone on foot to be predicted year economic flow rate (Y of the crop that obtains respectively _e), environment output (Y _i) predict the outcome, and both weight coefficients (α, β) that the 5th step obtained are brought built-up pattern single rate computing formula (Y=α Y per year into _e+ β Y _i), the Y value that obtains is the corresponding year final predicted value of crop yield amount.

When doing built-up pattern single step prediction (only predicting single rate in the coming year), final predicted value of crop yield amount in the coming year that said method can be obtained and real value are brought the historical data of existing n into as up-to-date sample, n value accumulated counts (n=n+1) is carried out the prediction in farther future with new historical sample simultaneously.

Embodiment

Initial data is the historical data (as shown in table 1) of 1988 to 2005 the cotton single rates in ecological irrigated area, Gansu Province and the component environment factor.

Table 1

Time	Cotton single rate (kg/hm^2)	Air pressure (hundred handkerchiefs)	Vegetative period daily mean temperature (℃)	Humidity (%)	Wind speed (0.1m/s)	Precipitation (mm)
							1988	1035	852.7	19.1	38.8	20.3	47.9
1989	1080	853.2	19.9	40.5	20.8	48.2

1990	1170	853.8	21.7	41.2	23.1	49.0
							1991	1335	852.4	22.2	46.6	23.7	51.3
1992	1545	853.6	22.8	48.3	22.1	53.0
							1993	1392	852.7	20.6	42.9	24.4	52.5
1994	1060	851.0	19.6	40.4	22.5	51.3
							1995	1173	852.6	20.9	45.7	19.8	52.7
1996	1278	851.9	21.7	46.8	20.6	53.1
							1997	1290	850.6	21.5	46.9	21.6	53.6
1998	1580	854.7	22.0	49.3	20.8	56.2
							1999	1820	852.8	23.2	53.3	24.5	58.8
2000	1391	853.1	21.7	48.5	20.1	54.8
							2001	1657	854.5	23.9	53.0	19.8	59.4
2002	1740	853.9	23.7	53.1	19.6	59.2
							2003	1722	855.2	23.6	53.9	20.1	58.9
2004	1659	854.8	23.8	53.7	21.3	59.2
							2005	1610	853.2	24.2	52.8	22.8	58.4

Detailed process is as follows:

To 1988 to 2005 historical output of cotton data (n=13) set up GM (1,1) grey forecasting model.Fitting precision is 87.7135% (the fitting precision requirement is satisfied in supposition, can predict economic flow rate Ye), and obtaining economic flow rate predicted value in 2006 is 1788.0 (kg/hm^2), and economic flow rate predicted value in 2007 is 1832.9 (kg/hm^2);

2. put 1988 to 2005 historical yield datas and historical meteorological environment data (n=13) in order, the meteorological environment data contain air pressure (being numbered 1), daily mean temperature in vegetative period (being numbered 2), humidity (being numbered 3), wind speed (being numbered 4), precipitation (being numbered 5) totally 5 envirment factors (m=5), make auxiliary sequence X with yield data ₍₀₎, each meteorological environment data are made subsequence X respectively ₍₁₎, X ₍₂₎, X ₍₃₎, X ₍₄₎, X ₍₅₎, carry out grey correlation analysis, obtain each subsequence degree of association relative and auxiliary sequence and be followed successively by: 0.56920.69020.75000.61360.6554, grey related preface is Here choose the main gene of the 3rd, 2,5 totally three (m '=3) factors (be humidity, vegetative period daily mean temperature, annual precipitation) as environment output;

With top 1988 historical datas to three environment main genes in 2005 as input, the cotton actual production data in corresponding year are as training objective, set up radial base neural net model (is that example experimentizes with radial base neural net parameter s pread=0.6), to the model training.Then with three environment main gene data (as shown in table 2) of year (2006-2007) to be predicted as input, emulation is exported and is obtained 2006,2007 environment yield prediction value (Y _i) be respectively 1568.9 (kg/hm^2), 1672.5 (kg/hm^2);

Table 2

Time	Vegetative period daily mean temperature (℃)	Humidity (%)	Annual precipitation (mm)
				2006	24.1	52.3	59.1
2007	23.8	53.6	58.4

4. use said method, preceding 13 years (1988-2000) historical datas in 1988 to 2005 are predicted the economic flow rate and the environment output of back 5 years (2001-2005), the result is as shown in table 3;

Table 3

Time	2001	2002	2003	2004	2005
						Economic flow rate (kg/hm^2)	1574.2	1614.2	1655.2	1697.2	1740.2
Environment output (kg/hm^2)	1620.2	1681.5	1667.2	1645.6	1596.9

5. for the test samples of calendar year 2001 to 2005 year, calculate the weight coefficient of grey forecasting model and neural network model with squared prediction error and counting backward technique, i.e. weight, the β of economic flow rate and environment output, the result is as shown in table 4;

Table 4

Time, weight

2001

2002

2003

2004

2005

α、β

Gray model prediction relative error	0.0500	0.0723	0.0388	0.0230	0.0809	α＝0.1515
							The neural network prediction relative error	0.0222	0.0336	0.0318	0.0081	0.0081	β＝0.8485

6. the to be predicted year crop economy output, the environment output that said method are obtained, and both weight coefficients are brought per unit area yield predictor formula Y=α Y into _e+ β Y _i, obtaining the final predicted value of cotton single rate in 2006 is 1602.1 (kg/hm^2), the final predicted value of cotton single rate in 2007 is 1696.8 (kg/hm^2).

Claims (4)

1, a kind of method of predicting irrigated area crop yield amount, the single rate of known n before the prediction, n 〉=5, this method is specific as follows:

Steps A adopts grey modeling method to make up the economic flow rate predictor formula Y in irrigated area according to the single rate of 1～n _e(k '), k ' are to be predicted year;

Step B behind the employing grey correlation selection environment main gene, uses neural network modeling approach to make up the environment yield prediction formula Y in irrigated area according to the single rate of 1～n _i(k '), k ' are to be predicted year;

Step C determines economic flow rate weight coefficient α and environment output weight coefficient β, alpha+beta=1, and α and economic flow rate predicated error are inversely proportional to, and β and environment yield prediction error are inversely proportional to;

Described economic flow rate predicated error is defined as: according to the actual single rate of 1～v, make up economic flow rate correction formula according to the mode of steps A

Adopt formula Prediction f=v+1 ..., the economic flow rate during n will predict the outcome and v+1 ..., the actual single rate of n relatively obtains the economic flow rate predicated error;

Described environment yield prediction error is defined as: according to the actual single rate of 1～v, according to the mode constructing environment output correction formula of step B

Adopt formula Prediction f=v+1 ..., the environment output during n will predict the outcome and v+1 ..., the actual single rate of n relatively obtains environment yield prediction error;

Wherein, 1＜＜v＜n;

Step D adopts single rate formula Y (k ')=α Y _e(k ')+β Y _i(k ') prediction irrigated area is at the single rate in k ' year.

2, a kind of method of predicting irrigated area crop yield amount according to claim 1 is characterized in that described steps A is specially: the single rate data of 1～n are formed original output sequence x ⁽⁰⁾={ x ⁽⁰⁾(1), x ⁽⁰⁾(2) ..., x ⁽⁰⁾(n) },

Steps A 1 is calculated $x^{\left(1\right)}\left(k\right)=\underset{i=1}{\overset{k}{\mathrm{\&Sigma;}}}{x}^{\left(0\right)}\left(i\right),k=\mathrm{1,2}\&CenterDot;\&CenterDot;\&CenterDot;,n;$

Steps A 2 is according to formula $\left[\frac{\hat{a}}{\hat{u}}\right]={\left(B^{T}B\right)}^{-1}{B}^T{Y}_n$ Determine development coefficient estimated value

With grey actuating quantity estimated value

T representing matrix transposition wherein, Y _n=(x ⁽⁰⁾(2), x ⁽⁰⁾(3) ..., x ⁽⁰⁾(n)) ^T, $B=\left(\begin{array}{cc}{-z}^{\left(1\right)}\left(2\right),& 1\\ {-z}^{\left(1\right)}\left(3\right),& 1\\ \&CenterDot;\&CenterDot;\&CenterDot;& \&CenterDot;\&CenterDot;\&CenterDot;\\ {-z}^{\left(1\right)}\left(n\right),& 1\end{array}\right),$ $z^{\left(1\right)}\left(k\right)=\frac{1}{2}{x}^{\left(1\right)}(k-1)+\frac{1}{2}{x}^{\left(1\right)}\left(k\right),k=\mathrm{2,3},\&CenterDot;\&CenterDot;\&CenterDot;,n;$

Steps A 3 if $0.5<\left|\hat{a}\right|<0.8,$ Then original output sequence x ⁽⁰⁾=x ⁽¹⁾(k), return steps A 1; If $\left|\hat{a}\right|\&GreaterEqual;0.8,$ Then to original output sequence x ⁽⁰⁾In data do smoothing processing, return steps A 1; If $\left|\hat{a}\right|\&le;0.5,$ Enter steps A 4;

Steps A 4 makes up formula ${\hat{x}}^{\left(1\right)}(k\&prime;+1)=(x^{\left(1\right)}\left(1\right)-\frac{\hat{u}}{\hat{a}})e^{-\hat{a}k\&prime;}+\frac{\hat{u}}{\hat{a}},$ ${\hat{x}}^{\left(1\right)}\left(1\right)=x^{\left(0\right)}\left(1\right),$ K ' is a positive integer;

Steps A 5 computational accuracies τ=1-ε, average relative error $\stackrel{\&OverBar;}{\mathrm{\&epsiv;}}=\frac{1}{n-1}\underset{k\&prime;=2}{\overset{n}{\mathrm{\&Sigma;}}}\left|\mathrm{\&epsiv;}(k\&prime;)\right|,$ Relative error $\mathrm{\&epsiv;}(k\&prime;)=\frac{x^{\left(1\right)}(k\&prime;)-{\hat{x}}^{\left(1\right)}(k\&prime;)}{x^{\left(1\right)}(k\&prime;)};$

Steps A 6 is if precision τ less than precision threshold, then makes up residual sequence near substantial deviation point ${\mathrm{\&epsiv;}\&prime;}^{\left(1\right)}(k\&prime;)=x^{\left(1\right)}(k\&prime;)-{\hat{x}}^{\left(1\right)}(k\&prime;),$ Adopt GM (1,1) grey modeling method to obtain residual sequence ε ' ⁽¹⁾The residual error estimated value sequence that (k ') is corresponding Upgrade ${\hat{x}}^{\left(1\right)}(k\&prime;)={\hat{x}}^{\left(1\right)}(k\&prime;)+{\mathrm{\&epsiv;}\&prime;}^{\left(1\right)}(k\&prime;),$ Return steps A 5; Otherwise, enter steps A 7;

Steps A 7 makes up the economic flow rate predictor formula of irrigated area in k ' year $Y_e(k\&prime;)={\hat{x}}^{\left(1\right)}(k\&prime;)-{\hat{x}}^{\left(1\right)}(k\&prime;-1).$

3, a kind of method of predicting irrigated area crop yield amount according to claim 1 is characterized in that described step B is specially: the single rate data of 1～n are formed original output sequence x ⁽⁰⁾={ x ⁽⁰⁾(1), x ⁽⁰⁾(2) ..., x ⁽⁰⁾(n) },

Step B1 is to original output sequence x ⁽⁰⁾={ x ⁽⁰⁾(1), x ⁽⁰⁾(2) ..., x ⁽⁰⁾(n) } do value just and handle generation auxiliary sequence X ₍₀₎M the monocycle border factor measured data sequence of 1～n done value just handle generation subsequence X _{(i ')}, i '=1,2 ..., m, m are the envirment factor number;

Step B2 calculates the incidence coefficient of each envirment factor ${\mathrm{\&eta;}}_{(i\&prime;)}\left(j\right)=\frac{\mathrm{\&rho;}\max \max |X_{(i\&prime;)}\left(j\right)-X_{\left(0\right)}\left(j\right)|}{|X_{(i\&prime;)}\left(j\right)-X_{\left(0\right)}\left(j\right)|+\mathrm{\&rho;}\max \max |X_{(i\&prime;)}\left(j\right)-X_{\left(0\right)}\left(j\right)|},$ J=1,2 ... n,

, | X _{(i ')}(j)-X ₍₀₎(j) | be X ₍₀₎In j element and X _{(i ')}In j element between Error Absolute Value;

Step B3 calculates X ₍₀₎With X _{(i ')}The degree of association $r_{(i\&prime;)}=\frac{1}{n}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&eta;}}_{(i\&prime;)}\left(j\right);$

Step B4 sorts the degree of association of m envirment factor according to subsequence and auxiliary sequence from big to small, selects the preceding individual envirment factor of m ' as environment output main gene,

It is right that step B5 makes up the sample of j, and the input sample of sample centering is the single year environment main gene measured data sequence of j, and output sample is the actual single rate of j, j=1, and 2 ... n;

Step B6 to as training sample, makes up radially basic artificial neural network with the sample of 1～n;

Step B7 makes up the environment yield prediction formula Y of irrigated area in k ' year _i(k '): with the irrigated area in the single year environment main gene measured data sequence in k ' year as the radially input of basic artificial neural network, output is the environment output Y of irrigated area in k ' year _i(k ').

4, a kind of method of predicting irrigated area crop yield amount according to claim 1 is characterized in that described step C is specially:

Step C1 extracts the actual single rate of 1～v, predicts the irrigated area at f=v+1 according to the mode of steps A ..., the economic flow rate predicted value Y of n _e(f); Extract the actual single rate of 1～v, and environment main gene measured data, predict the irrigated area at f=v+1 according to the mode of step B ..., the environment yield prediction value Y of n _i(f);

Step C2 calculates f=v+1 ..., the economic flow rate predicted value Y of n _e(f) with f=v+1 ..., the difference between the single rate of n reality is made sum of squares to these differences and is calculated, and asks for its S1 reciprocal again; Calculate f=v+1 ..., the environment yield prediction value Y of n _i(f) with f=v+1 ..., the difference between the single rate of n reality is made sum of squares to these differences and is calculated, and asks for its S2 reciprocal again;

Step C3 calculates the weight coefficient of economic flow rate $\mathrm{\&alpha;}=\frac{S1}{S1+S2},$ The weight coefficient of environment output $\mathrm{\&beta;}=\frac{S2}{S1+S2}.$

Images