CN105956715A - Soil moisture status prediction method and device - Google Patents

Abstract

The invention provides a soil moisture status prediction method and device. The method comprises that prediction sample data of H influential factors which influence the soil moisture status of a soil sample to be predicted is extracted; characteristic vectors formed by the prediction sample data are input to a soil moisture status prediction model, the model is pre-established according to the weights and thresholds calculated by utilizing adaptive adjusting intersection and variation rules of an improved heredity diversity function on the basis of a heredity algorithm, and the soil moisture status prediction model calculates the characteristic vectors by utilizing the preset weights and thresholds; and a soil moisture status prediction result of the soil sample to be predicted is output. The weights and thresholds of the soil moisture status prediction model are calculated by utilizing the adaptive probability intersection and variation rules under the effect of the improved heredity diversity function and fitness function, it can be ensured that offspring performance is better than parent performance as possible, and thus, the prediction precision for the soil moisture status is improved.

Description

A kind of Forecast of Soil Moisture Content method and device

Technical field

The present invention relates to genetic neural network algorithm optimisation technique field, particularly relate to a kind of soil moisture content pre- Survey method and device.

Background technology

Soil moisture content, refers to the humidity condition of soil.Nowadays, water resource is the key affecting plant growth One of factor, reasonable prediction soil moisture content can calculate the actual water requirement of crop, in the premise ensureing yield Under, improve the utilization rate of irrigation water, provide scientific basis for grain-production safety and field irrigation water-saving plan, Drought Condition can be predicted in advance by trend prediction simultaneously, provide safeguard for crops disaster alarm in time. Therefore, Forecast of Soil Moisture Content monitors important in inhibiting to plant growth.

Prior art provides a kind of Forecast of Soil Moisture Content side based on traditional genetic neural network algorithm Method.Generally, the genetic algorithm part in genetic neural network algorithm can include selecting, intersecting and make a variation three Individual basic operation.At present, the genetic algorithm part in conventional genetic neural network algorithm uses roulette Selection strategy, this selection strategy makes each individuality all have selected probability (i.e. to select to have random Property), and cross and variation carries out based on constant probability, therefore it cannot be guaranteed that filial generation performance is always better than father Generation so that the maximum that the final fitness value evolved may not necessarily obtain the overall situation (feelings of Premature Convergence i.e. occurs Condition), namely global optimizing ability is poor, and the precision of prediction causing soil moisture content is low, it is difficult to meet soil moisture in the soil The precision of prediction requirement of feelings.

Summary of the invention

The purpose of the embodiment of the present invention is to provide a kind of Forecast of Soil Moisture Content method and device, to improve soil The precision of prediction of soil moisture content.

For reaching above-mentioned purpose, the embodiment of the invention discloses a kind of Forecast of Soil Moisture Content method, described method Including:

S1) the forecast sample data of H factor of influence of the soil moisture content affecting soil sample to be predicted are extracted, Wherein, H is positive integer；

S2) characteristic vector described forecast sample data constituted, input is to according to based on genetic algorithm and profit Intersect with the genetic diversity function Automatic adjusument improved and variation rule calculates the weights and threshold value obtained And in the Forecast of Soil Moisture Content model pre-build, described Forecast of Soil Moisture Content model utilize preset weights and Described characteristic vector is calculated by threshold value；

S3) output is calculated for described soil sample to be predicted through described Forecast of Soil Moisture Content model Forecast of Soil Moisture Content result.

Optionally, described step S2) in the method for building up of Forecast of Soil Moisture Content model include:

A1) network topology structure is determined, including input layer, hidden layer and output layer；

A2) weights and the threshold value of described network topology structure are obtained；

A3) training sample of H factor of influence of the soil moisture content of soil sample to be trained described in impact is extracted Data and actual soil moisture content value；

A4) the input vector input training sample data of each training sample constituted successively is to described input Layer, the weights corresponding with output layer according to described input layer, hidden layer and threshold value, calculate each layer respectively Output valve, updates current weights and threshold value, until all training sample input is complete；

A5) relatively described actual soil moisture content value and described step A4) predictive value of output layer that obtains, Obtain the error e predicted the outcome；

A6) judge that whether described error e is less than the minimum allowable error e preset_T；If being less than, then output is worked as Front weights and threshold value.

Optionally, described step A2), including:

A21) determining the population P that body quantity one by one is N, wherein, N is positive integer；

A22) according to the weights preset and threshold value create-rule, each individuality in described population P is generated right The weights answered and threshold value；

A23) according to default chromosome coding rule, by each individual corresponding weights and threshold coding For item chromosome；

A24) the ideal adaptation angle value that the chromosome of each individuality is corresponding, and described population are calculated respectively The genetic diversity degree of P；

A25) according to each ideal adaptation angle value described and described genetic diversity degree, to described population P In each individuality carry out evolution process, it is thus achieved that after evolution, individual amount is the population P' of N', and wherein, N' is Positive integer；

A26) the ideal adaptation angle value that in described population P', the chromosome of each individuality is corresponding is calculated respectively, And the genetic diversity degree of described population P'；

A27) determine that the population that ideal adaptation angle value is described population P' that in described population P', numerical value is maximum is fitted Answer angle value；

A28) judge whether the Population adaptation angle value of described population P' meets the condition stopping evolving preset, If meeting, then export the individual weights corresponding to chromosome and the threshold value of the ideal adaptation angle value of described maximum, Weights and threshold value as described network topology structure.

Optionally, the ideal adaptation angle value that the chromosome of each individuality is corresponding is calculated in the following manner:

According to following formula, calculate the ideal adaptation angle value that the chromosome of each individuality is corresponding:

$f\left(S_i\right)={\left(\underset{r=1}{\overset{m}{\Σ}}\right|T_r-O_r\left|\right)}^{-1},r=1,2,\mathrm{...},m$

Wherein, m is the number of output layer neuron, T_rIt it is the preferable output i.e. actual monitoring of r neuron Value, O_rBe r neuron prediction output i.e. calculate output valve.

Optionally, the genetic diversity degree of described population P is calculated in the following manner:

According to following formula, calculate the genetic diversity degree of described population P:

${\mathrm{\μ}}_t=\frac{\underset{i=1}{\overset{N-1}{\Σ}}\underset{j=i+1}{\overset{N}{\Σ}}\underset{k=1}{\overset{L}{\Σ}}|{x_{ik}}^{\left(t\right)}-{x_{jk}}^{\left(t\right)}|}{\underset{i=1}{\overset{N-1}{\Σ}}(N-i)},N>2$

Wherein, N is the population scale of population, and t is the algebraically of Evolution of Population, and L is each individuality in population The code length of chromosome.

Optionally, described step A25) including:

A251) utilize the selection strategy of roulette, from described population P, determine individuality to be evolved；

A252) intersect based on default adaptive probability and variation is regular, obtain the new population after cross and variation；

A253) utilize elitist selection strategy, optimize the new population after described cross and variation, generate after evolving Population P'.

Optionally, described step A252) in the adaptive probability preset intersect and variation rule, including:

According to following crossover probability expression formula, calculate crossover probability P_c:

$P_c=\left\{\begin{array}{cc}{P}_{c0}& f^{\′}>f_{avg}\\ P_{c0}-a_c\frac{f_{avg}-f^{\′}}{f_{avg}-f_{\min }}& f^{\′}\≤f_{avg}\end{array}\right.$

Wherein, k_cAnd a_cFor constant, f_minFor fitness value minimum in current population, f_avgPutting down for current population All fitness values, f' is fitness value bigger in two individualities intersected,Genetic diversity for current population Property degree.

According to following mutation probability P_m, calculate mutation probability P_m:

$P_m=\left\{\begin{array}{cc}{P}_{m0}& f\≤f_{avg}\\ P_{m0}-a_m\frac{f-f_{avg}}{f_{\max }-f_{avg}}& f>f_{avg}\end{array}\right.$

Wherein, k_mAnd a_mFor constant, f_maxFor fitness value maximum in current population, f_avgFor current population Average fitness value, f is the fitness value that variation is individual,Genetic diversity degree for current population.

Optionally, described step S2) in Forecast of Soil Moisture Content model at least include the submodel that is exemplified below In one or more:

A: for same depth of soil, and the prediction submodel of the soil moisture content after preset time period；

B: for same predicted time point, and the prediction submodel of the soil moisture content of the different soils degree of depth；

C: for different depth of soil, and the prediction submodel of the soil moisture content after preset time period.

For reaching above-mentioned purpose, the embodiment of the invention discloses a kind of Forecast of Soil Moisture Content device, described device Including:

Prediction data extraction module, individual for extracting the H of the soil moisture content affecting described soil sample to be predicted The forecast sample data of factor of influence, wherein, H is positive integer；

Prediction data computing module, for the characteristic vector described forecast sample data constituted, input is to root According to the genetic diversity function Automatic adjusument intersection based on genetic algorithm and utilizing improvement and variation rule meter In the Forecast of Soil Moisture Content model calculating the weights and threshold value that obtain and pre-build, described Forecast of Soil Moisture Content mould Type utilizes the weights preset and threshold value to calculate described characteristic vector；

Forecast of Soil Moisture Content module, calculated for institute through described Forecast of Soil Moisture Content model for output State the Forecast of Soil Moisture Content result of soil sample to be predicted.

Optionally, also include Forecast of Soil Moisture Content model building module, including:

Network structure determines submodule, is used for determining network topology structure, including input layer, hidden layer and defeated Go out layer；

Initial parameter obtains submodule, for obtaining weights and the threshold value of described network topology structure；

Training data extracts submodule, for extracting the H of the soil moisture content of soil sample to be trained described in impact The number of training of individual factor of influence is according to this and actual soil moisture content value；

Training sample calculating sub module, defeated for successively the training sample data of each training sample are constituted Incoming vector input to described input layer, the weights corresponding with output layer according to described input layer, hidden layer and threshold Value, calculates the output valve of each layer respectively, updates current weights and threshold value, until all training sample is defeated Enter complete；

Error obtains submodule, for relatively described actual soil moisture content value and described step A4) obtain defeated Go out the predictive value of layer, it is thus achieved that the error e predicted the outcome；

Target component obtains submodule, for judging that whether described error e is less than the minimum allowable error preset e_T；If being less than, then export current weights and threshold value.

A kind of Forecast of Soil Moisture Content method and device that the embodiment of the present invention provides.When predicting soil moisture content, First the forecast sample data of H factor of influence of the soil moisture content affecting this soil sample to be predicted are extracted, Then the characteristic vector forecast sample data extracted constituted, input is to according to based on genetic algorithm and profit Intersect with the genetic diversity function Automatic adjusument improved and variation rule calculates the weights and threshold value obtained And in the Forecast of Soil Moisture Content model pre-build, this Forecast of Soil Moisture Content model utilizes the weights and threshold preset Characteristic vector is calculated by value, finally exports through Forecast of Soil Moisture Content model calculated for be predicted The Forecast of Soil Moisture Content result of soil sample.As known from the above, the soil inputted due to forecast sample data Weights in the network topology structure of soil moisture content forecast model and threshold value, be the cross and variation utilizing adaptive probability Rule calculates acquisition, different from the mode that constant probability used in the prior art carries out cross and variation, The scheme that the embodiment of the present invention provides can ensure that filial generation performance is better than parent, namely improves as much as possible The ability of global optimizing, therefore, by forecast sample data input to utilize improve genetic diversity function and Fitness function jointly act under adaptive probability intersection and variation rule calculate obtain weights and threshold The Forecast of Soil Moisture Content model that value is trained calculates, and obtained predicting the outcome is more accurate, it is seen then that Improve the precision of prediction of soil moisture content.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to enforcement In example or description of the prior art, the required accompanying drawing used is briefly described, it should be apparent that, describe below In accompanying drawing be only some embodiments of the present invention, for those of ordinary skill in the art, do not paying On the premise of going out creative work, it is also possible to obtain other accompanying drawing according to these accompanying drawings.

The schematic flow sheet of a kind of Forecast of Soil Moisture Content method that Fig. 1 provides for the embodiment of the present invention；

The flow process signal of the method for building up of a kind of Forecast of Soil Moisture Content model that Fig. 2 provides for the embodiment of the present invention Figure；

A kind of weights obtaining network topology structure that Fig. 3 provides for the embodiment of the present invention and the method for threshold value Schematic flow sheet；

The schematic flow sheet of a kind of Evolution of Population method that Fig. 4 provides for the embodiment of the present invention；

The structural representation of a kind of Forecast of Soil Moisture Content device that Fig. 5 provides for the embodiment of the present invention.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clearly Chu, be fully described by, it is clear that described embodiment be only a part of embodiment of the present invention rather than Whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art are not making creation The every other embodiment obtained under property work premise, broadly falls into the scope of protection of the invention.

Soil moisture content, refers to the humidity condition of soil.In prior art, it is provided that a kind of utilization is based on heredity The model that neural network algorithm is set up carries out the method for Forecast of Soil Moisture Content.General, genetic neural network Genetic algorithm part in algorithm can include selecting, intersecting and three basic operations that make a variation.At present, existing The genetic algorithm part in conventional genetic neural network algorithm employed in technology uses roulette Selection strategy, this selection strategy makes each individuality all have selected probability (i.e. selecting have randomness), And cross and variation is carried out based on constant probability, therefore it cannot be guaranteed that filial generation performance is always better than parent, make Obtain the fitness value finally evolved and may not necessarily obtain the maximum (situation of Premature Convergence i.e. occurs) of the overall situation, Namely global optimizing ability is poor, and the precision of prediction causing soil moisture content is low, it is difficult to meet the pre-of soil moisture content Survey required precision.

The invention provides one based on conventional genetic neural network algorithm, improve algorithm, the present invention Using adaptive probability heredity reverse transmittance nerve network algorithm, this algorithm weighs population at accurate definition The genetic diversity function of gene diversityOn the basis of, utilize logistic letter in conjunction with fitness function f The crossover probability P of number matching Automatic adjusument_cWith mutation probability P_m, and introduce elitist selection strategy in the overall situation In the range of find optimal solution, no matter under Forecast of Soil Moisture Content or standard test functions, global optimizing performance There is raising greatly.

Below in conjunction with specific embodiment, a kind of Forecast of Soil Moisture Content method being provided the embodiment of the present invention is entered Row describes in detail.

Embodiment one:

As it is shown in figure 1, the schematic flow sheet of a kind of Forecast of Soil Moisture Content method provided for the embodiment of the present invention, The method may comprise steps of:

S1) the forecast sample data of H factor of influence of the soil moisture content affecting soil sample to be predicted are extracted, Wherein, H is positive integer.

Field test data shows, soil moisture content often with air themperature, air humidity, light radiation, wind Speed, effective rainfall irrigation volume, preset the first soil layer the soil moisture and preset the second soil layer soil moisture etc. because of Have pass.

It should be noted that above-mentioned " preset the first soil layer " mentioned and " default second soil layer " can be by The concrete condition that technical staff in ability applies according to reality is reasonably arranged, it addition, preset first Soil layer can be identical with the soil depth presetting the second soil layer, it is also possible to the depth of soil presetting the second soil layer Difference, the embodiment of the present invention need not above-mentioned " presetting the first soil layer " and the tool of " presetting the second soil layer " Body numerical value is defined.

It is emphasized that the described herein soil moisture presetting the first soil layer and preset the second soil layer soil Humidity is only used to obtain the factor of the soil moisture content affecting soil sample to be predicted, soil to be predicted The soil depth of sample can preset the first soil layer with above-mentioned " or " preset the second soil layer " identical, certainly, The first soil layer can also be preset from above-mentioned " or " preset the second soil layer " different.

In a kind of implementation, can be by air themperature, air humidity, light radiation, wind speed, effectively drop Rain irrigation volume, the soil moisture presetting the first soil layer and default these 7 factors of second soil layer soil moisture determine For affecting the influence factor of the soil moisture content of soil sample to be predicted.And then, H the factor of influence extracted Forecast sample data, can be that the numerical value of above-mentioned 7 factors of influence is as relative with each factor of influence The forecast sample data answered.

It should be noted that the above-mentioned implementation enumerated is only the one in numerous implementation, impact H factor of influence of the soil moisture content of soil sample to be predicted is also not limited in above-mentioned implementation listed These 7 kinds lifted, the embodiment of the present invention is not required to be defined the particular content of H factor of influence, ability Technical staff in territory needs the concrete condition in applying according to reality reasonably to arrange.

S2) characteristic vector forecast sample data constituted, input is to according to changing based on genetic algorithm and utilization The genetic diversity function Automatic adjusument that enters intersects and variation rule calculates the weights and threshold value obtained and pre- In the Forecast of Soil Moisture Content model first set up, this Forecast of Soil Moisture Content model utilizes the weights and threshold value pair preset Described characteristic vector calculates.

Concrete, can by extracted with the forecast sample data corresponding to each factor of influence according to specific Rule generate a characteristic vector, further, data process for convenience, it is also possible to this feature Vector is normalized.It should be noted that the generation of characteristic vector and the normalization of characteristic vector It is processed as technological means relatively common in prior art, can be found in correlation technique of the prior art, at this In be not described in detail.

In a kind of implementation, as in figure 2 it is shown, a kind of Forecast of Soil Moisture Content provided for the embodiment of the present invention The schematic flow sheet of the method for building up of model, step S2) in the method for building up bag of Forecast of Soil Moisture Content model Include:

A1) network topology structure is determined, including input layer, hidden layer and output layer.

Concrete, this implementation can use BP neural network algorithm (Back Propagation, reversely Propagation algorithm) the Forecast of Soil Moisture Content model set up.

In this implementation, it may be determined that this network topology structure is: input layer 1 layer, hidden layer 1 layer and Output layer 1 layer, wherein, the characteristic vector of input layer be by air themperature, air humidity, light radiation, Wind speed, effective rainfall irrigation volume, preset the first soil layer the soil moisture and preset the second soil layer soil moisture this The characteristic vector that the forecast sample data of 7 factors are constituted, accordingly, input layer number of nodes It it is 7；Hidden layer neuron number of nodes is 9；The result of output layer output is soil sample to be predicted Soil moisture content, accordingly, it is 1 that the neuron of output layer examines quantity.As known from the above, determined by Network topology structure is (7,9,1).

The above-mentioned network topology structure enumerated is a kind of concrete form in this implementation, certainly, also Can there be other feasible network topology structures, such as: input layer 1 layer, hidden layer 2 layers and output layer 1 layer. According to same factor of influence, then may determine that this network topology structure is: (7,9,6,1).Its In, ground floor hidden layer neuron number of nodes is 9, and second layer hidden layer neuron number of nodes is 6.Need It is noted that the method determining BP neutral net hidden layer neuron number of nodes is the most public Open, may refer to prior art determines rule about the relevant of hidden layer neuron, the most no longer carry out Describe in detail.

A2) weights and the threshold value of this network topology structure are obtained.

After network topology structure determines, this Forecast of Soil Moisture Content model to be set up, in addition it is also necessary to obtain and belong to Adjacent layer (such as, for the network topology structure of single hidden layer, input layer and hidden layer are adjacent layer, Hidden layer and output layer are adjacent layer) two neuron nodes between weights, and each layer of neuron The threshold value (except input layer) of node self, after these weights and threshold value determine, also implies that this Completing of BP Establishment of Neural Model.

The BP neutral net citing being still (7,9,1) with the network topology structure of single hidden layer, including: 7*9+9*1=72 weights, 9+1=10 threshold value.

Experimental data shows, the weights of network topology structure and the quality of threshold value are directly connected to this BP nerve net Network model can be set up and the precision of prediction of model.Therefore, it is thus achieved that one group preferably weights and threshold value for The foundation of this BP neural network model is just particularly important.

Concrete, as it is shown on figure 3, a kind of power obtaining network topology structure provided for the embodiment of the present invention The schematic flow sheet of the method for value and threshold value, may comprise steps of:

A21) determining the population P that body quantity one by one is N, wherein, N is positive integer.

Concrete, in order to ensure the accuracy that population P evolves, the numerical value of individual amount N is difficult to the least, examines again Considering to amount of calculation and the time of process, individual amount N tries one's best the most not too big, it is preferred that N can be 300 Numerical value between 500.Certainly, the embodiment of the present invention is not required to the individual amount N included by population P Numerical value be defined, those skilled in the art need according to reality apply in concrete condition close The setting of reason.

A22) according to the weights preset and threshold value create-rule, each individual correspondence in population P is generated Weights and threshold value.

Concrete, it is possible to use random algorithm, for weights and the threshold value of each individuality in population P, Generate a random number respectively and produced random number is defined as the initial value of each weights and threshold value.Need Being noted that and the most only list a kind of concrete mode generating weights and threshold value, the present invention is not required to The concrete mode generating each individual corresponding weights and threshold value is defined, any feasible realization Mode all can apply to the present invention.

A23) according to default chromosome coding rule, by each individual corresponding weights and threshold coding For item chromosome.

Concrete, the chromosome coding rule preset can be to be the chromosome coding rule using real coding mode Then.It should be noted that the present invention need not be defined the specific coding rule of chromosome, Ren Heke The implementation of energy all can apply to the present invention.

A24) the ideal adaptation angle value that the chromosome of each individuality is corresponding is calculated respectively, and population P Genetic diversity degree.

In a kind of implementation, calculate the ideal adaptation that the chromosome of each individuality is corresponding in the following manner Angle value:

According to following formula, calculate the ideal adaptation angle value that the chromosome of each individuality is corresponding:

$f\left(S_i\right)={\left(\underset{r=1}{\overset{m}{\Σ}}\right|T_r-O_r\left|\right)}^{-1},r=1,2,\mathrm{...},m$

Wherein, m is the number of output layer neuron, T_rIt it is the preferable output i.e. actual monitoring of r neuron Value, O_rBe r neuron prediction output i.e. calculate output valve.

In a kind of implementation, the genetic diversity degree of calculating population P in the following manner:

According to following formula, calculate the genetic diversity degree of population P:

${\mathrm{\μ}}_t=\frac{\underset{i=1}{\overset{N-1}{\Σ}}\underset{j=i+1}{\overset{N}{\Σ}}\underset{k=1}{\overset{L}{\Σ}}|{x_{ik}}^{\left(t\right)}-{x_{jk}}^{\left(t\right)}|}{\underset{i=1}{\overset{N-1}{\Σ}}(N-i)},N>2$

Wherein, N is the population scale of population, and t is the algebraically of Evolution of Population, and L is each individuality in population The code length of chromosome.

Unlike the conventional genetic neural network algorithm that it is emphasized that and mention in prior art, this Inventive embodiments provide scheme, the basis of the genetic algorithm part in conventional genetic neural network algorithm it On, further contemplate the genetic multifarious factor that can weigh in population chromosome, i.e. lost Passing diversity level, this genetic diversity degree can be by above-mentioned expression formula (referred to as " genetic diversity letter Number ") calculate acquisition, will appreciate that in Advanced group species, each chromosome is wrapped by this genetic diversity degree The diversity level of the gene contained.

A25) according to each ideal adaptation angle value and genetic diversity degree, to each individuality in population P Carrying out evolution process, it is thus achieved that after evolution, individual amount is the population P' of N', wherein, N' is positive integer.

As shown in Figure 4, the schematic flow sheet of a kind of Evolution of Population method provided for the embodiment of the present invention, can To comprise the following steps:

A251) utilize the selection strategy of roulette, from population P, determine individuality to be evolved.

Concrete, for the population P={S that individual amount is N₁, S₂..., S_NFor }, i-th is individual Chromosome S_iThe fitness function of ∈ P is f (S_i), then individual probability selected for i is:

$P_i=\frac{f\left(S_i\right)}{\underset{i=1}{\overset{N}{\Σ}}f\left(S_i\right)},i=1,2,......,N;$

A252) intersect based on default adaptive probability and variation is regular, obtain the new population after cross and variation.

Concrete, step A252) in the adaptive probability preset intersect and variation rule, including:

According to following crossover probability expression formula, calculate crossover probability P_c:

$P_c=\left\{\begin{array}{cc}{P}_{c0}& f^{\′}>f_{avg}\\ P_{c0}-a_c\frac{f_{avg}-f^{\′}}{f_{avg}-f_{\min }}& f^{\′}\≤f_{avg}\end{array}\right.$

Wherein, k_cAnd a_cFor constant, f_minFor fitness value minimum in current population, f_avgPutting down for current population All fitness values, f' is fitness value bigger in two individualities intersected,Genetic diversity for current population Property degree.

More specifically, intersect and use real number interior extrapolation method, for the population P={S that individual amount is N₁, S₂..., S_NFor }, choose i-th chromosome S_iWith jth chromosome S_jThe public affairs operated of intersecting at kth gene location Formula is:

$\left\{\begin{array}{c}{S}_{ik}=S_{ik}(1-b)+S_{jk}b\\ S_{jk}=S_{jk}(1-b)+S_{ik}b\end{array}\right.$

Wherein, b is the random number between [0,1].

According to following mutation probability P_m, calculate mutation probability P_m:

$P_m=\left\{\begin{array}{cc}{P}_{m0}& f\≤f_{avg}\\ P_{m0}-a_m\frac{f-f_{avg}}{f_{\max }-f_{avg}}& f>f_{avg}\end{array}\right.$

Wherein, k_mAnd a_mFor constant, f_maxFor fitness value maximum in current population, f_avgFor current population Average fitness value, f is the fitness value that variation is individual,Genetic diversity degree for current population.

More specifically, variation uses non-uniform mutation method, for the population P={S that individual amount is N₁, S₂..., S_NFor }, choose i-th chromosome S_iKth gene make a variation, the public affairs of mutation operation Formula is:

$S_{ik}=\left\{\begin{array}{cc}{S}_{ik}+(S_{ik}-S_{\max })*f\left(g\right),& r_1>0.5\\ S_{ik}+(S_{\min }-S_{ik})*f\left(g\right),& r_1\≤0.5\end{array}\right.$

Wherein, r₁For the random number between [0,1], S_maxFor gene S_ikThe upper bound, S_minFor gene S_ikLower bound, F (g)=r₂*(1-g/G_max)², r₂For the random number between [-1,1], g is current iteration number of times, G_maxFor maximum Evolution number of times.

Unlike the conventional genetic neural network algorithm that it is emphasized that and mention in prior art, this The scheme that inventive embodiments provides, for selected individuality to be evolved, intersects and makes a variation carrying out The when of operation, intersect according to adaptive probability and variation rule is carried out, it is, fit for individuality For answering the individuality that angle value is different, namely for the individuality that performance in population is good and bad, cross and variation Probability is discrepant, this ensure that global optimization degree individual in population.

A253) utilize elitist selection strategy, optimize the new population after cross and variation, generate the population after evolving P'。

Concrete, elitist selection strategy design considerations in: when being inferior to parent when filial generation performance, introduce elite plan Slightly replace the worst individuality of filial generation；When filial generation performance is better than parent, the most do not introduce elitism strategy.It can be seen that Elitist selection strategy further ensures global optimization degree individual in population.

A26) the ideal adaptation angle value that in population P', the chromosome of each individuality is corresponding is calculated respectively, and The genetic diversity degree of population P'.

A27) the Population adaptation angle value that ideal adaptation angle value is population P' that in population P', numerical value is maximum is determined.

A28) judge whether the Population adaptation angle value of population P' meets the condition stopping evolving preset, if full Foot, then export the individual weights corresponding to chromosome and the threshold value of the ideal adaptation angle value of maximum, as this The weights of network topology structure and threshold value.

Concrete, the condition stopping evolving preset may is that the Evolution of Population judging whether to reach default Iterations, if reached, then judges to meet the condition stopping evolving preset, is otherwise unsatisfactory for.Certainly, Can also have the condition stopping evolving that other are feasible, this is not defined by the present invention.

A3) the training sample data of H factor of influence of the soil moisture content affecting soil sample to be trained are extracted And actual soil moisture content value.

A4) successively the input vector that the training sample data of each training sample are constituted is inputted to input layer, The weights corresponding with output layer according to this input layer, hidden layer and threshold value, calculate the output valve of each layer respectively, Update current weights and threshold value, until all training sample input is complete.

It should be noted that prior art has been disclosed for by the training sample data composition of training sample Input vector input is to input layer, and according to this input layer, hidden layer weights corresponding with output layer and threshold Value is weighted the concrete calculating process of summation operation, does not repeats them here, can be found in phase of the prior art Close calculating process.

A5) relatively actual soil moisture content value and step A4) predictive value of output layer that obtains, it is thus achieved that prediction The error e of result.

A6) whether error in judgement e is less than the minimum allowable error e preset_T；If being less than, then export current Weights and threshold value.

It addition, step S2) in Forecast of Soil Moisture Content model at least include in the submodel being exemplified below Individual or multiple:

A: for same depth of soil, and the prediction submodel of the soil moisture content after preset time period；

B: for same predicted time point, and the prediction submodel of the soil moisture content of the different soils degree of depth；

C: for different depth of soil, and the prediction submodel of the soil moisture content after preset time period.

In actual application, in order to meet different Forecast of Soil Moisture Content needs, it is possible to use above-mentioned soil moisture content The mode of setting up of model sets up prediction submodel based on different forecast demands.Above-mentioned A, B and C enumerated The difference of prediction submodel is, the particular content of selected actual soil moisture content value is different, citing For, for A, due to for same depth of soil, such as 10cm soil layer soil, train sample Notebook data may is that air themperature, air humidity, light radiation, wind speed, effective rainfall irrigation volume, 10cm The soil layer soil moisture and 10cm soil layer soil moisture, preset time period is set to 24 hours, accordingly, selected The actual soil moisture content value taken is: the 10cm soil layer soil moisture after 24 hours.For B, owing to being For same predicted time point, such as 14:00, training sample data may is that air themperature, air are wet Degree, light radiation, wind speed, effective rainfall irrigation volume, the 10cm soil layer soil moisture and 10cm soil layer soil Humidity, the default different soils degree of depth is: 20cm soil layer soil moisture, accordingly, selected actual soil Earth soil moisture content value is: the 20cm soil layer soil moisture of 14:00.Also being similar for C, here is omitted. Sum it up, for different Forecast of Soil Moisture Contents needs, it is possible to use mode set up by same model, Difference is that the particular content of selected training sample data and/or actual soil moisture content value is different.

S3) output is through the calculated soil moisture content for soil sample to be predicted of Forecast of Soil Moisture Content model Predict the outcome.

As known from the above, the network topology knot of the Forecast of Soil Moisture Content model inputted due to forecast sample data Weights in structure and threshold value, be to utilize the cross and variation rule of adaptive probability to calculate to obtain, with existing skill The mode that constant probability employed in art carries out cross and variation is different, the scheme energy that the embodiment of the present invention provides Reach and ensure that filial generation performance is better than parent the most as much as possible, namely improve the ability of global optimizing, therefore, incite somebody to action Under the input of forecast sample data acts on jointly to the genetic diversity function and fitness function utilizing improvement The Forecast of Soil Moisture Content that the intersection of adaptive probability and the weights of variation rule calculating acquisition and threshold value are trained Model calculates, and obtained predicting the outcome is more accurate, it is seen then that improve the prediction essence of soil moisture content Degree.

Embodiment two:

As it is shown in figure 5, the structural representation of a kind of Forecast of Soil Moisture Content device provided for the embodiment of the present invention, This device can include with lower module:

Prediction data extraction module 210, for extracting H shadow of the soil moisture content affecting soil sample to be predicted Ringing the forecast sample data of the factor, wherein, H is positive integer；

Prediction data computing module 220, for characteristic vector forecast sample data constituted, input is to root According to the genetic diversity function Automatic adjusument intersection based on genetic algorithm and utilizing improvement and variation rule meter In the Forecast of Soil Moisture Content model calculating the weights and threshold value that obtain and pre-build, this Forecast of Soil Moisture Content model Utilize the weights preset and threshold value that this feature vector is calculated；

Forecast of Soil Moisture Content module 230, for output through this Forecast of Soil Moisture Content model calculated for The Forecast of Soil Moisture Content result of this soil sample to be predicted.

Concrete, also include Forecast of Soil Moisture Content model building module, including:

Network structure determines submodule, is used for determining network topology structure, including input layer, hidden layer and defeated Go out layer；

Initial parameter obtains submodule, for obtaining weights and the threshold value of this network topology structure；

Training data extracts submodule, for extracting H shadow of the soil moisture content affecting soil sample to be trained Ring the number of training of the factor according to this and actual soil moisture content value；

Training sample calculating sub module, defeated for successively the training sample data of each training sample are constituted Incoming vector input to input layer, the weights corresponding with output layer according to input layer, hidden layer and threshold value, respectively Calculate the output valve of each layer, update current weights and threshold value, until all training sample input is complete；

Error obtains submodule, for relatively actual soil moisture content value and step A4) output layer that obtains pre- Measured value, it is thus achieved that the error e predicted the outcome；

Target component obtains submodule, for judging that whether this error e is less than the minimum allowable error e preset_T； If being less than, then export current weights and threshold value.

Concrete, this initial parameter obtains submodule, including:

Population determines unit, and for determining the population P that body quantity one by one is N, wherein, N is positive integer；

Weights and threshold value signal generating unit, for according to the weights preset and threshold value create-rule, generating this population Each individual corresponding weights and threshold value in P；

Chromosome coding unit, for according to default chromosome coding rule, by each individual correspondence Weights and threshold coding are item chromosome；

First computing unit, for calculating the ideal adaptation angle value that the chromosome of each individuality is corresponding respectively, And the genetic diversity degree of this population P；

Evolution of Population unit, for according to each ideal adaptation angle value and genetic diversity degree, to population P In each individuality carry out evolution process, it is thus achieved that after evolution, individual amount is the population P' of N', and wherein, N' is Positive integer；

Second computing unit, for calculating the individuality that in this population P', the chromosome of each individuality is corresponding respectively Fitness value, and the genetic diversity degree of this population P'；

Population's fitness determines unit, for determining that the ideal adaptation angle value that in this population P', numerical value is maximum is to plant The Population adaptation angle value of group P'；

Stop Evolution of Population judging unit, preset for judging whether the Population adaptation angle value of this population P' meets The condition stopping evolving, if meeting, then the individual chromosome institute of the ideal adaptation angle value exporting maximum is right The weights answered and threshold value, as weights and the threshold value of this network topology structure.

Concrete, this first computing unit or the second computing unit, specifically for:

According to following formula, calculate the ideal adaptation angle value that the chromosome of each individuality is corresponding:

$f\left(S_i\right)={\left(\underset{r=1}{\overset{m}{\Σ}}\right|T_r-O_r\left|\right)}^{-1},r=1,2,\mathrm{...},m$

Wherein, m is the number of output layer neuron, T_rIt it is the preferable output i.e. actual monitoring of r neuron Value, O_rBe r neuron prediction output i.e. calculate output valve.

Concrete, this first computing unit or the second computing unit, specifically for:

According to following formula, calculate the genetic diversity degree of this population P:

${\mathrm{\μ}}_t=\frac{\underset{i=1}{\overset{N-1}{\Σ}}\underset{j=i+1}{\overset{N}{\Σ}}\underset{k=1}{\overset{L}{\Σ}}|{x_{ik}}^{\left(t\right)}-{x_{jk}}^{\left(t\right)}|}{\underset{i=1}{\overset{N-1}{\Σ}}(N-i)},N>2$

Wherein, N is the population scale of population, and t is the algebraically of Evolution of Population, and L is each individuality in population The code length of chromosome.

Concrete, this Evolution of Population unit, including:

Evolution individual selection subelement, for utilizing the selection strategy of roulette, determine from population P treat into The individuality changed；

Cross and variation subelement, for intersecting based on default adaptive probability and variation rule, is intersected New population after variation；

New population generates subelement, is used for utilizing elitist selection strategy, optimizes the new population after cross and variation, Generate the population P' after evolving.

Concrete, this cross and variation subelement, specifically for:

According to following crossover probability expression formula, calculate crossover probability P_c:

$P_c=\left\{\begin{array}{cc}{P}_{c0}& f^{\′}>f_{avg}\\ P_{c0}-a_c\frac{f_{avg}-f^{\′}}{f_{avg}-f_{\min }}& f^{\′}\≤f_{avg}\end{array}\right.$

Wherein, k_cAnd a_cFor constant, f_minFor fitness value minimum in current population, f_avgPutting down for current population All fitness values, f' is fitness value bigger in two individualities intersected,Genetic diversity for current population Property degree.

According to following mutation probability P_m, calculate mutation probability P_m:

$P_m=\left\{\begin{array}{cc}{P}_{m0}& f\≤f_{avg}\\ P_{m0}-a_m\frac{f-f_{avg}}{f_{\max }-f_{avg}}& f>f_{avg}\end{array}\right.$

Wherein, k_mAnd a_mFor constant, f_maxFor fitness value maximum in current population, f_avgFor current population Average fitness value, f is the fitness value that variation is individual,Genetic diversity degree for current population.

Concrete, it is one or more that this Forecast of Soil Moisture Content model at least includes in the submodel being exemplified below:

A: for same depth of soil, and the prediction submodel of the soil moisture content after preset time period；

B: for same predicted time point, and the prediction submodel of the soil moisture content of the different soils degree of depth；

C: for different depth of soil, and the prediction submodel of the soil moisture content after preset time period.

As known from the above, the network topology knot of the Forecast of Soil Moisture Content model inputted due to forecast sample data Weights in structure and threshold value, be to utilize the cross and variation rule of adaptive probability to calculate to obtain, with existing skill The mode that constant probability employed in art carries out cross and variation is different, the scheme energy that the embodiment of the present invention provides Reach and ensure that filial generation performance is better than parent the most as much as possible, namely improve the ability of global optimizing, therefore, incite somebody to action Under the input of forecast sample data acts on jointly to the genetic diversity function and fitness function utilizing improvement The Forecast of Soil Moisture Content that the intersection of adaptive probability and the weights of variation rule calculating acquisition and threshold value are trained Model calculates, and obtained predicting the outcome is more accurate, it is seen then that improve the prediction essence of soil moisture content Degree.

For device embodiment, owing to it is substantially similar to embodiment of the method, so describing the simplest Single, relevant part sees the part of embodiment of the method and illustrates.

It should be noted that in this article, the relational terms of such as first and second or the like be used merely to by One entity or operation separate with another entity or operating space, and not necessarily require or imply these Relation or the order of any this reality is there is between entity or operation.And, term " includes ", " bag Contain " or its any other variant be intended to comprising of nonexcludability, so that include a series of key element Process, method, article or equipment not only include those key elements, but also include being not expressly set out Other key elements, or also include the key element intrinsic for this process, method, article or equipment.? In the case of there is no more restriction, statement " including ... " key element limited, it is not excluded that including The process of described key element, method, article or equipment there is also other identical element.

One of ordinary skill in the art will appreciate that all or part of step realizing in said method embodiment Suddenly the program that can be by completes to instruct relevant hardware, and described program can be stored in computer can Read in storage medium, storage medium designated herein, such as: ROM/RAM, magnetic disc, CD etc..

The foregoing is only presently preferred embodiments of the present invention, be not intended to limit protection scope of the present invention. All any modification, equivalent substitution and improvement etc. made within the spirit and principles in the present invention, are all contained in In protection scope of the present invention.

Claims (10)

1. a Forecast of Soil Moisture Content method, it is characterised in that described method includes:

S1) the forecast sample data of H factor of influence of the soil moisture content affecting soil sample to be predicted are extracted, Wherein, H is positive integer；

S2) characteristic vector described forecast sample data constituted, input is to according to based on genetic algorithm and profit Intersect with the genetic diversity function Automatic adjusument improved and variation rule calculates the weights and threshold value obtained And in the Forecast of Soil Moisture Content model pre-build, described Forecast of Soil Moisture Content model utilize preset weights and Described characteristic vector is calculated by threshold value；

S3) output is calculated for described soil sample to be predicted through described Forecast of Soil Moisture Content model Forecast of Soil Moisture Content result.

Method the most according to claim 1, it is characterised in that described step S2) in soil moisture content The method for building up of forecast model includes:

A1) network topology structure is determined, including input layer, hidden layer and output layer；

A2) weights and the threshold value of described network topology structure are obtained；

A3) training sample of H factor of influence of the soil moisture content of soil sample to be trained described in impact is extracted Data and actual soil moisture content value；

A4) the input vector input training sample data of each training sample constituted successively is to described input Layer, the weights corresponding with output layer according to described input layer, hidden layer and threshold value, calculate each layer respectively Output valve, updates current weights and threshold value, until all training sample input is complete；

A5) relatively described actual soil moisture content value and described step A4) predictive value of output layer that obtains, Obtain the error e predicted the outcome；

A6) judge that whether described error e is less than the minimum allowable error e preset_T；If being less than, then output is worked as Front weights and threshold value.

Method the most according to claim 2, it is characterised in that described step A2), including:

A21) determining the population P that body quantity one by one is N, wherein, N is positive integer；

A22) according to the weights preset and threshold value create-rule, each individuality in described population P is generated right The weights answered and threshold value；

A23) according to default chromosome coding rule, by each individual corresponding weights and threshold coding For item chromosome；

A24) the ideal adaptation angle value that the chromosome of each individuality is corresponding, and described population are calculated respectively The genetic diversity degree of P；

A25) according to each ideal adaptation angle value described and described genetic diversity degree, to described population P In each individuality carry out evolution process, it is thus achieved that after evolution, individual amount is the population P' of N', and wherein, N' is Positive integer；

A26) the ideal adaptation angle value that in described population P', the chromosome of each individuality is corresponding is calculated respectively, And the genetic diversity degree of described population P'；

A27) determine that the population that ideal adaptation angle value is described population P' that in described population P', numerical value is maximum is fitted Answer angle value；

A28) judge whether the Population adaptation angle value of described population P' meets the condition stopping evolving preset, If meeting, then export the individual weights corresponding to chromosome and the threshold value of the ideal adaptation angle value of described maximum, Weights and threshold value as described network topology structure.

Method the most according to claim 3, it is characterised in that calculate in the following manner each each and every one The ideal adaptation angle value that the chromosome of body is corresponding:

According to following formula, calculate the ideal adaptation angle value that the chromosome of each individuality is corresponding:

$f\left(S_i\right)={\left(\underset{r=1}{\overset{m}{\Σ}}\right|T_r-O_r\left|\right)}^{-1},r=1,2,...,m$

Wherein, m is the number of output layer neuron, T_rIt it is the preferable output i.e. actual monitoring of r neuron Value, O_rBe r neuron prediction output i.e. calculate output valve.

Method the most according to claim 3, it is characterised in that calculate described population in the following manner The genetic diversity degree of P:

According to following formula, calculate the genetic diversity degree of described population P:

${\mathrm{\μ}}_t=\frac{\underset{i=1}{\overset{N-1}{\Σ}}\underset{j=i+1}{\overset{N}{\Σ}}\underset{k=1}{\overset{L}{\Σ}}|{x_{ik}}^{\left(t\right)}-{x_{jk}}^{\left(t\right)}|}{\underset{i=1}{\overset{N-1}{\Σ}}(N-i)},N>2$

Wherein, N is the population scale of population, and t is the algebraically of Evolution of Population, and L is each individuality in population The code length of chromosome.

Method the most according to claim 3, it is characterised in that described step A25) including:

A251) utilize the selection strategy of roulette, from described population P, determine individuality to be evolved；

A252) intersect based on default adaptive probability and variation is regular, obtain the new population after cross and variation；

A253) utilize elitist selection strategy, optimize the new population after described cross and variation, generate after evolving Population P'.

Method the most according to claim 6, it is characterised in that described step A252) in preset from Adapt to probability intersect and variation rule, including:

According to following crossover probability expression formula, calculate crossover probability P_c:

$P_c=\left\{\begin{array}{cc}{P}_{c0}& f^{\′}>f_{avg}\\ P_{c0}-a_c\frac{f_{avg}-f^{\′}}{f_{avg}-f_{\min }}& f^{\′}\≤f_{avg}\end{array}\right.$

Wherein, k_cAnd a_cFor constant, f_minFor fitness value minimum in current population, f_avgPutting down for current population All fitness values, f' is fitness value bigger in two individualities intersected,Genetic diversity for current population Property degree.

According to following mutation probability P_m, calculate mutation probability P_m:

$P_m=\left\{\begin{array}{cc}{P}_{m0}& f\≤f_{avg}\\ P_{m0}-a_m\frac{f-f_{avg}}{f_{\max }-f_{avg}}& f>f_{avg}\end{array}\right.$

Wherein, k_mAnd a_mFor constant, f_maxFor fitness value maximum in current population, f_avgFor current population Average fitness value, f is the fitness value that variation is individual,Genetic diversity degree for current population.

Method the most according to claim 1, it is characterised in that described step S2) in soil moisture content It is one or more that forecast model at least includes in the submodel being exemplified below:

A: for same depth of soil, and the prediction submodel of the soil moisture content after preset time period；

B: for same predicted time point, and the prediction submodel of the soil moisture content of the different soils degree of depth；

C: for different depth of soil, and the prediction submodel of the soil moisture content after preset time period.

9. a Forecast of Soil Moisture Content device, it is characterised in that described device includes:

Prediction data extraction module, individual for extracting the H of the soil moisture content affecting described soil sample to be predicted The forecast sample data of factor of influence, wherein, H is positive integer；

Prediction data computing module, for the characteristic vector described forecast sample data constituted, input is to root According to the genetic diversity function Automatic adjusument intersection based on genetic algorithm and utilizing improvement and variation rule meter In the Forecast of Soil Moisture Content model calculating the weights and threshold value that obtain and pre-build, described Forecast of Soil Moisture Content mould Type utilizes the weights preset and threshold value to calculate described characteristic vector；

Forecast of Soil Moisture Content module, calculated for institute through described Forecast of Soil Moisture Content model for output State the Forecast of Soil Moisture Content result of soil sample to be predicted.

Method the most according to claim 9, it is characterised in that also include Forecast of Soil Moisture Content model Set up module, including:

Network structure determines submodule, is used for determining network topology structure, including input layer, hidden layer and defeated Go out layer；

Initial parameter obtains submodule, for obtaining weights and the threshold value of described network topology structure；

Training data extracts submodule, for extracting the H of the soil moisture content of soil sample to be trained described in impact The number of training of individual factor of influence is according to this and actual soil moisture content value；

Training sample calculating sub module, defeated for successively the training sample data of each training sample are constituted Incoming vector input to described input layer, the weights corresponding with output layer according to described input layer, hidden layer and threshold Value, calculates the output valve of each layer respectively, updates current weights and threshold value, until all training sample is defeated Enter complete；

Error obtains submodule, for relatively described actual soil moisture content value and described step A4) obtain defeated Go out the predictive value of layer, it is thus achieved that the error e predicted the outcome；

Target component obtains submodule, for judging that whether described error e is less than the minimum allowable error preset e_T；If being less than, then export current weights and threshold value.