CN110276482A

CN110276482A - A kind of forecasting procedure of plant flowers and fruits beginning and ending time breeding time

Info

Publication number: CN110276482A
Application number: CN201910474904.0A
Authority: CN
Inventors: 姚日升; 黄鹤楼; 杨栋; 顾思南; 涂小萍
Original assignee: Ning Bo Shiqixiangtai
Current assignee: Ning Bo Shiqixiangtai
Priority date: 2019-06-03
Filing date: 2019-06-03
Publication date: 2019-09-24
Anticipated expiration: 2039-06-03
Also published as: CN110276482B

Abstract

The invention discloses the forecasting procedures of plant flowers and fruits beginning and ending time breeding time a kind of, its flowers and fruits breeding time for collecting plant starts or the historical summary of end time, the flowers and fruits breeding time including plant over the years start or the end time day sequence, start over the years to the flowers and fruits breeding time of plant or end time relevant multiple meteorological factors and the phenology factor；Available predictors are determined according to meteorological factor and the phenology factor；According to available predictors, select a kind of types of models as forecasting model constructed type from three kinds of linear model, nonlinear model and mixed model typess of models；Model construction type according to weather report constructs best forecasting model；The beginning of the flowers and fruits breeding time of the plant of forecast is treated according to best forecasting model or the end time is forecast；Advantage is their ability to make full use of the prediction of various weather constituents of long timeliness and starts carrying out plant flowers and fruits breeding time or the forecast of end time, and forecasts that accuracy is high.

Description

A kind of forecasting procedure of plant flowers and fruits beginning and ending time breeding time

Technical field

The present invention relates to a kind of plant flowers and fruits breeding time forecasting techniques, more particularly, to a kind of plant flowers and fruits breeding time start-stop The forecasting procedure of time.

Background technique

With the rapid development of social economy with the continuous improvement of living standards of the people, to admire the beauty of flowers, picking activity is the theme All kinds of local red-letter days and tour itineraries be surging forward throughout the country, carry out plant flowers and fruits breeding time (early flowering season, full-bloom stage, Fructescence etc.) forecast one of the work for also becoming Meteorological Services society.Japan is known as " state of oriental cherry ", annual spring cherry The season that the flowers are in blossom puts, the florescence of oriental cherry just become the topic that japanese people is paid special attention to.Plant growth temperature dependent, humidity, The factors such as precipitation and illumination, agricultural and worker in meteorology carry out the chilling requirement and calorific requirement of deciduous fruit tree specific breeding time Research, obtains effective accumulated temperature required for different plant varieties breeding time.For the ability for improving Meteorological Services society, various regions meteorology Also the florescence to a variety of flowers and trees and fructescence forecast have carried out a large amount of research to worker, establish various forecasting models, These forecasting models are substantially all using linear regression model, substantially there is two class predictors: one is based on phenology because Son, such as a certain phenological period of other plants or opposite early-maturing variety, Du Yaodong etc. (ecological magazine, 2006) also studied previous The relationship of year peach priming leaf phase and current year peach full-bloom stage；Another kind is meteorological factor, and main includes the moon, the mean of dekan gas of early period Temperature is stablized the date by a certain temperature, accumulated temperature, ground temperature of certain time etc., Wei Xiulan (meteorological, 2001), Kong Fanzhong (China Agricultural weather, 2011) meteorological factor is extracted also from upper air circulation early period to make the Medium-long Term Prediction of peony forescence.These gas As the factor mainly for plant flowers and fruits breeding time early period a certain meteorological element accumulation, wherein most importantly accumulated temperature, this In accumulated temperature be positive accumulated temperature, i.e., from a certain day, such as the accumulated temperature counted on January 1, and in fact, plant flowers and fruits give birth to Phase is often higher with reverse accumulated temperature correlation, and reverse accumulated temperature referred to from plant flowers and fruits breeding time, the special time period of time back Accumulated temperature.For the forecast of plant flowers and fruits breeding time, if the meteorological element of time early period section is used only, it cannot embody and face The influence of the weather conditions of nearly flowers and fruits breeding time, to make more accurate forecast, it is necessary to utilize the gas of following a period of time As the forecast of element, weather forecast timeliness can be up to 10 days or more, with the development and weather forecast of Numerical Forecast Technology at present The raising of technology can also obtain finer, the longer forecast of timeliness from now on, it is therefore desirable to more reasonable forecasting model is established, with The prediction of various weather constituents of long timeliness is made full use of to carry out the forecast of plant flowers and fruits breeding time.

Summary of the invention

Technical problem to be solved by the invention is to provide the forecasting procedure of plant flowers and fruits beginning and ending time breeding time a kind of, The prediction of various weather constituents of long timeliness be can make full use of to carry out plant flowers and fruits breeding time and start or the forecast of end time, and pre- Report accuracy high.

The technical scheme of the invention to solve the technical problem is: a kind of plant flowers and fruits beginning and ending time breeding time Forecasting procedure, it is characterised in that the following steps are included:

Step 1: the flowers and fruits breeding time for collecting plant starts or the historical summary of end time, the flowers and fruits including plant over the years Breeding time start or the end time day sequence, start over the years to the flowers and fruits breeding time of plant or end time relevant multiple meteorologies The factor starts to the flowers and fruits breeding time of plant or the end time relevant phenology factor over the years, and meteorological factor is linear prediction The factor is Nonlinear predictor, and the phenology factor is the linear prediction factor or is Nonlinear predictor；If linear pre- The report factor and Nonlinear predictor are not present, then without forecast, otherwise, execute step 2；

Above-mentioned, range over the years is 10~30 years, has forward cumulative temperature, reverse for the meteorological factor of the linear prediction factor It accumulates temperature, forward cumulative precipitation, reverse accumulative rainfall amount, forward cumulative humidity, inversely accumulate humidity, forward cumulative illumination When number, inversely accumulate lighting delay number, be Nonlinear predictor meteorological factor be to be obtained according to historical experience；

Step 2: being started according to the flowers and fruits breeding time of plant over the years or the day sequence of end time, the flowers and fruits for estimating plant are raw The phase of educating starts or end time possible most sequence and possible day sequence at the latest early, correspondence are denoted as D₁And D₂；Wherein, D₁< D₂；

Step 3: only in the presence of the linear prediction factor, the flowers and fruits breeding time for calculating each year plant in over the years starts Or the day sequence of end time starts with the flowers and fruits breeding time of corresponding time plant or each linear prediction factor of end time Linearly dependent coefficient；Then related significance verification is carried out to all linearly dependent coefficients；It will be verified again by related significance All linearly dependent coefficients corresponding to the linear prediction factor as available predictors；

Only in the presence of Nonlinear predictor, the flowers and fruits breeding time of plant over the years is started or the institute of end time There is Nonlinear predictor as available predictors；

In the presence of the linear prediction factor and Nonlinear predictor are equal, the flowers and fruits of each year plant in over the years are calculated Breeding time starts or the day sequence of end time starts with the flowers and fruits breeding time of corresponding time plant or each of end time is linear The linearly dependent coefficient of predictor；Then related significance verification is carried out to all linearly dependent coefficients；Correlation will be passed through again The flowers and fruits breeding time of the linear prediction factor corresponding to all linearly dependent coefficients of checking validity and plant over the years start or All Nonlinear predictors of end time are as available predictors；

Step 4: according to available predictors, from three kinds of linear model, nonlinear model and mixed model typess of models Selecting a kind of types of models, wherein mixed model is made of linear block and nonlinear block as forecasting model constructed type, If available predictors are the linear prediction factor, any one of linear model, nonlinear model and mixed model are selected Types of models is as forecasting model constructed type；If available predictors are Nonlinear predictor, nonlinear model is selected Type is as forecasting model constructed type；If the existing linear prediction factor of available predictors has Nonlinear predictor again, select Any types of models in nonlinear model and mixed model is selected as forecasting model constructed type, when selection mixed model is made Linear block when for forecasting model constructed type in mixed model can only select linear predictor to construct；Then according to pre- Model construction type is reported, best forecasting model is constructed；

If forecasting model constructed type is linear model, the building process of best forecasting model are as follows:

A1, by the flowers and fruits breeding time of plant over the years start or the end time day sequence and flowers and fruits breeding time of plant over the years open Begin or all available predictors of end time constitute historical sample set；

A2, historical sample set is by year divided into two subsets, respectively as modeling sample set and test samples collection It closes；

A3, using regression equation method, and utilize modeling sample set, establish initial forecasting model；

A4, by year all available predictors in each time in test samples set are input to and initially forecast mould In type, the day sequence predicted value in corresponding time is obtained；

A5, by year calculate each time day sequence predicted value with the corresponding time in test samples set day sequence difference Value；Then the average value of corresponding difference of all times is calculated；Then judge whether average value is less than or equal to setting value D_s1, such as Fruit is, then initial forecasting model is determined as best forecasting model, and by all available predictors in historical sample set It is determined as selecting predictor eventually, then executes step 5；Otherwise, step A6 is executed；Wherein, D_s1∈ [1,3] day；

A6, the corresponding available predictors of minimal linear related coefficient in historical sample set are deleted, then returns to step Rapid A2 is continued to execute, if all available predictors in historical sample set are disposed and do not determine best forecasting model, Then without forecast；

If forecasting model constructed type is nonlinear model, the building process of best forecasting model are as follows:

B1, it is over the years in the currently pending time is by year defined as current year；

B2, by D in current year₁~D₂Currently pending day sequence is defined as when day before yesterday sequence in day sequence；

B3, ought day before yesterday sequence be denoted as D_cur；It ought day before yesterday sequence D_curAs additional predictor, and ought day before yesterday sequence D_curIt is It is no to be greater than or equal to D₀Return value as Forecasting Object；Wherein, D_cur∈[D₁,D₂], D₀Indicate the flowers and fruits of current year plant Breeding time starts or the day sequence of end time, D₁< D₀< D₂If D_cur≥D₀Then Forecasting Object value is 1, if D_cur< D₀It is then pre- Reporting object value is 0；

B4, the flowers and fruits breeding time of Forecasting Object, additional predictor, plant over the years is started or all times of end time Predictor is selected to constitute a historical sample；

B5, by D in current year₁~D₂Next day sequence to be processed is used as when day before yesterday sequence in day sequence, then return step B3 is continued to execute, until D in current year₁~D₂An each of day sequence day sequence is disposed, and executes step B6；

B6, will it is over the years in next time to be processed as current year, then return step B2 is continued to execute, up to Each time over the years is disposed, will it is over the years in D in all times₁~D₂The corresponding all historical samples of day sequence constitute history Sample set；

B7, historical sample set is by year divided into two subsets, respectively as modeling sample set and test samples collection It closes；

B8, using artificial neural network or support vector machine method, and utilize modeling sample set, establish initial forecast Model；

B9, by D in each time in test samples set₁~D₂Additional forecast in the historical sample day by day of day sequence The factor and its corresponding all available predictors are input in initial forecasting model, obtain the additional corresponding forecast of predictor Value, is set to 1 for predicted value if predicted value >=0.5；Predicted value is set to 0 if predicted value < 0.5；The additional forecast of judgement Whether the corresponding predicted value of factor Forecasting Object corresponding with the additional predictor in test samples set is identical, if identical It then indicates that forecast is correct, forecast mistake is indicated if not identical；If all historical samples in test samples set are corresponding Overall forecast accuracy reaches 80%, then initial forecasting model is determined as best forecasting model, and will be in historical sample set All available predictors be determined as selecting predictor eventually, then execute step 5；Otherwise, step B10 is executed；

If the available predictors in B10, historical sample set are the linear prediction factor, will be in historical sample set The corresponding linear prediction factor of minimal linear related coefficient is deleted；If the available predictors in historical sample set are non-thread Property predictor, then by historical sample set any one Nonlinear predictor delete；If in historical sample set The existing linear prediction factor of available predictors has Nonlinear predictor again, then minimal linear in historical sample set is related The corresponding linear prediction factor of coefficient or any one Nonlinear predictor are deleted；

B11, return step B7 are continued to execute, if all available predictors in historical sample set be disposed and not Determine best forecasting model, it is determined that without forecast；

If forecasting model constructed type is mixed model, the building process of forecasting model are as follows:

If C1, available predictors are the linear prediction factor, the flowers and fruits breeding time of plant over the years is started or terminated Time day sequence and plant over the years flowers and fruits breeding time start or the part available predictors of end time constitute linear block Historical sample set；If the existing linear prediction factor of available predictors has Nonlinear predictor, by plant over the years again Flowers and fruits breeding time start or the end time day sequence and flowers and fruits breeding time of plant over the years start or all or portion of end time Heterogeneous linear predictor constitutes linear block historical sample set；

C2, linear block historical sample set is by year divided into two subsets, respectively as linear block modeling sample Set and linear block test samples set；

C3, using regression equation method, and utilize linear block modeling sample set, establish initial linear module；

C4, by year by all available predictors in each time in linear block test samples set be input to just In beginning linear block, the day sequence predicted value in corresponding time is obtained；

C5, by year calculate each time day sequence predicted value with the corresponding time in linear block test samples set The difference of day sequence；Then the average value of corresponding difference of all times is calculated；Then judge whether average value is less than or equal to set Definite value D_s1, if it is, initial linear module is determined as optimum linear module, and will be in linear block historical sample set The end that all available predictors are determined as optimum linear module selects predictor, then will be in linear block historical sample set All ends in each time select predictor to be input in optimum linear module, obtain the day sequence predicted value in corresponding time, execute Step C7；Otherwise, step C6 is executed；Wherein, D_s1∈ [1,3] day；

C6, the corresponding available predictors of minimal linear related coefficient in linear block historical sample set are deleted, so Return step C2 is continued to execute afterwards, if all available predictors in linear block historical sample set be disposed and not really Determine optimum linear module, then linear block is invalid, and mixed model will only include nonlinear block, and forecasting model constructed type is changed For nonlinear model；

C7, it is over the years in the currently pending time is by year defined as current year；

C8, by D in current year₁~D₂Currently pending day sequence is defined as when day before yesterday sequence in day sequence；

C9, ought day before yesterday sequence be denoted as D_cur；It ought day before yesterday sequence D_curAs the first additional predictor, and ought day before yesterday sequence D_curMore than or equal to D₀Return value as Forecasting Object；The current year that step C5 is obtained by optimum linear module Day, sequence predicted value was as the second additional predictor；Wherein, D_cur∈[D₁,D₂], D₀Indicate the flowers and fruits fertility of current year plant Phase starts or the day sequence of end time, D₁< D₀< D₂If D_cur≥D₀Then Forecasting Object value is 1, if D_cur< D₀Then forecast pair As value is 0；

C10, by Forecasting Object, the first additional predictor, the second additional predictor, plant over the years flowers and fruits breeding time Start or the end time day sequence and plant over the years flowers and fruits breeding time start or the end time except constitute linear block history Remaining available predictors outside sample set constitute a nonlinear block historical sample；

C11, by D in current year₁~D₂Next day sequence to be processed is used as when day before yesterday sequence in day sequence, then returns to step Rapid C9 is continued to execute, until D in current year₁~D₂An each of day sequence day sequence is disposed, and executes step C12；

C12, will it is over the years in next time to be processed as current year, then return step C8 is continued to execute, directly Be disposed to each time over the years, will it is over the years in D in all times₁~D₂The corresponding all historical samples of day sequence constitute non- Linear block historical sample set；

C13, nonlinear block historical sample set is by year divided into two subsets, modeled respectively as nonlinear block Sample set and nonlinear block test samples set；

C14, it is built jointly using artificial neural network or support vector machine method, and using nonlinear block modeling sample collection Vertical initial non-linearities module；

C15, by D in each time in nonlinear block test samples set₁~D₂The nonlinear block day by day of day sequence The first additional predictor and its corresponding all available predictors in historical sample are input in initial non-linearities module, The corresponding predicted value of the first additional predictor is obtained, predicted value is set to 1 if predicted value >=0.5；If predicted value < Predicted value is set to 0 by 0.5；Judge in the corresponding predicted value of the first additional predictor and nonlinear block test samples set The corresponding Forecasting Object of the first additional predictor it is whether identical, if the same indicate that forecast is correct, if not identical Indicate forecast mistake；If the corresponding overall forecast of all nonlinear block historical samples in nonlinear block test samples set Accuracy reaches 80%, then initial non-linearities module is determined as Optimal Nonlinear module, and by nonlinear block historical sample The end that all available predictors in set are determined as Optimal Nonlinear module selects predictor, will by optimum linear module and The mixed model of Optimal Nonlinear module composition is as best forecasting model, then executes step 5；Otherwise, step C16 is executed；

If the available predictors in C16, nonlinear block historical sample set are the linear prediction factor, will be non-thread Property module history sample set in the corresponding linear prediction factor of minimal linear related coefficient delete；If nonlinear block history sample Available predictors in this set are Nonlinear predictor, then will be any one in nonlinear block historical sample set A Nonlinear predictor is deleted；If the existing linear prediction factor of available predictors in nonlinear block historical sample set There is Nonlinear predictor again, then by the corresponding linear prediction of minimal linear related coefficient in nonlinear block historical sample set The factor or any one Nonlinear predictor are deleted；

C17, return step C13 are continued to execute, if all available predictors in nonlinear block historical sample set It is disposed and does not determine Optimal Nonlinear module, it is determined that without forecast；

Step 5: the beginning of flowers and fruits breeding time of plant to be forecast or the time of end time are defined as the forecast time； The day sequence for starting from the date of forecast has been defined as report day sequence, has been denoted as D_s；Meteorology can be provided to the longest of weather forecast The number of days of effective timeliness is denoted as D_max；

When the type of best forecasting model is linear model, with being selected needed for predictor eventually for calculating for forecast time The best forecasting model of material computation corresponding each end select the factor values of predictor, best forecasting model is corresponding all It selects the factor values of predictor to be input in best forecasting model eventually, a day sequence value is calculated；If the day sequence value being calculated In the time range D of the flowers and fruits breeding time of plant to be forecast₁~D₂It is interior, then it is assumed that the day sequence value being calculated is effective, is plant Flowers and fruits breeding time beginning or the end time day sequence forecast result, then execute step 6；Otherwise it is assumed that being calculated Day sequence value it is invalid, without forecast；

When the type of best forecasting model is nonlinear model, if D_s< D₁-D_maxOr D_s> D₂, then without pre- Report；Otherwise, following steps are executed:

Step 5_1a, by the day sequence D on the date currently calculated_jIt is defined as calculating day sequence, wherein D_jInitial value be D_s；

Step 5_1b, according to D_jThe factor values that best forecasting model corresponding each end selects predictor are calculated, it will be best Forecasting model corresponding all ends select the factor values of predictor to be input in best forecasting model, and predicted value is calculated；

If predicted value >=0.5 step 5_1c, being calculated, works as D_jIn flowers and fruits breeding time of plant to be forecast Time range D₁~D₂When interior, it is believed that D_jEffectively, be plant to be forecast flowers and fruits breeding time beginning or the day sequence of end time Forecast result, then execute step 6；Work as D_jNot in the time range D of the flowers and fruits breeding time of plant to be forecast₁~D₂When interior, Think D_jIn vain, without forecast；If predicted value < 0.5 being calculated, thens follow the steps 5_1d；

If step 5_1d, D_j> D₂Or D_j> D_s+D_max, then without forecast；Otherwise, D is enabled_j=D_j+ 1, by it is next into The day sequence on the date that row calculates returns again to step 5_1b and continues to execute as day sequence is calculated；Wherein, D_j=D_j"=" in+1 is Assignment；

When the type of best forecasting model is mixed model, with being selected needed for predictor eventually for calculating for forecast time Linear block corresponding each end of the best forecasting model of material computation select the factor values of predictor, by best forecasting model Linear block corresponding all ends select the factor values of predictor to be input in the linear block of best forecasting model, calculate The day sequence value forecast to linear block；If be calculated linear block forecast day sequence value not in the flower of plant to be forecast Fruit breeding time time range D₁~D₂It is interior, it is believed that the day sequence value for the linear block forecast being calculated is invalid, without forecast；Such as The forecast of linear block that fruit is calculated day sequence value plant to be forecast flowers and fruits breeding time time range D₁~D₂It is interior, then Think that the day sequence value of the linear block being calculated forecast is effective, then execute following steps:

If step 5_2a, D_s< D₁-D_maxOr D_s> D₂, then without forecast；Otherwise, step 5_2b is executed；

Step 5_2b, by the day sequence D on the date currently calculated_jIt is defined as calculating day sequence, wherein D_jInitial value be D_s；

Step 5_2c, according to D_jThe factor values that best forecasting model corresponding each end selects predictor are calculated, it will be best Forecasting model corresponding all ends select the factor values of predictor to be input in best forecasting model, and predicted value is calculated；

If predicted value >=0.5 step 5_2d, being calculated, works as D_jIn flowers and fruits breeding time of plant to be forecast Time range D₁~D₂When interior, it is believed that D_jEffectively, be plant to be forecast flowers and fruits breeding time beginning or the day sequence of end time Forecast result, then execute step 6；Work as D_jNot in the time range D of the flowers and fruits breeding time of plant to be forecast₁~D₂When interior, Think D_jIn vain, without forecast；If predicted value < 0.5 being calculated, thens follow the steps 5_2e；

If step 5_2e, D_j> D₂Or D_j> D_s+D_max, then without forecast；Otherwise, D is enabled_j=D_j+ 1, by it is next into The day sequence on the date that row calculates returns again to step 5_2c and continues to execute as day sequence is calculated；Wherein, D_j=D_j"=" in+1 is Assignment；

Step 6: obtained forecast result being converted into the date, the beginning of the flowers and fruits breeding time of the plant as the forecast time Or the date terminated, forecast terminate.

In the step 1, the phenology factor is the phenological period of other plants or opposite early-maturing variety based on phenology.

In the step A2 and the step B7, the quantity of the sample in modeling sample set is historical sample set In sample quantity 70%, the quantity of the sample in test samples set is the quantity of the sample in historical sample set 30%.

In the step C2, the quantity of the sample in linear block modeling sample set is linear block historical sample collection The 70% of the quantity of sample in conjunction, the quantity of the sample in linear block test samples set are linear block historical sample collection The 30% of the quantity of sample in conjunction；In the step C13, the quantity of the sample in nonlinear block modeling sample set is The 70% of the quantity of sample in nonlinear block historical sample set, the number of the sample in nonlinear block test samples set Amount is the 30% of the quantity of the sample in nonlinear block historical sample set.

Compared with the prior art, the advantages of the present invention are as follows:

1) the method for the present invention can select linear model, nonlinear model or mixed model, according to available predictors Type and test result finally determine suitable forecasting model, improve forecast accuracy.

2) prognostic equation of linear regression building can only select the linear relevant predictor i.e. linear prediction factor, this hair Nonlinear model in bright method can choose Nonlinear predictor, increase the range of choice of predictor.

3) primary to plant using day sequence as Forecasting Object when linear regression method building plant flowers and fruits breeding time forecasting model Object flowers and fruits breeding time data can be only formed a historical sample, the nonlinear block of nonlinear model of the invention and mixed model Whether to reach flowers and fruits breeding time as Forecasting Object, from possible most sequence daily shape into possible day sequence at the latest early in 1 year At a historical sample, therefore a plant flowers and fruits breeding time data can form many historical samples, improve statistics and calculate The stability of method, while also improving the value of forecasting.

4) weather forecast product can be used in available predictors used in the method for the present invention, something also can be used The reverse accumulated value of reason amount can sufficiently reflect weather conditions when closing on plant flowers and fruits breeding time as available predictors in this way Influence, and in fact it is this influence it is bigger than the influence of weather conditions early period, improve forecast accuracy.

5) there are florescence, fructescence etc. in typical plant flowers and fruits breeding time, and the method for the present invention can individually forecast flowers and fruits The time that breeding time starts or terminates, can also simultaneously be forecast respectively, and then forecast plant flowers and fruits breeding time process when Between, for the nursing of plant flowers and fruits breeding time and to admire the beauty of flowers, all kinds of social activitieies that picking activity is the theme technical support is provided.

Detailed description of the invention

Fig. 1 is the overall implementation process block diagram of the method for the present invention.

Specific embodiment

The present invention will be described in further detail below with reference to the embodiments of the drawings.

A kind of forecasting procedure of plant flowers and fruits beginning and ending time breeding time proposed by the present invention comprising following steps:

Step 1: the flowers and fruits breeding time for collecting plant starts or the historical summary of end time, the flowers and fruits including plant over the years Breeding time start or the end time day sequence, start over the years to the flowers and fruits breeding time of plant or end time relevant multiple meteorologies The factor starts to the flowers and fruits breeding time of plant or the end time relevant phenology factor over the years, and meteorological factor is linear prediction The factor is Nonlinear predictor, and the phenology factor is the linear prediction factor or is Nonlinear predictor；If linear pre- The report factor and Nonlinear predictor are not present, then without forecast, otherwise, execute step 2.

Above-mentioned, range over the years is 10~30 years, has forward cumulative temperature, reverse for the meteorological factor of the linear prediction factor It accumulates temperature, forward cumulative precipitation, reverse accumulative rainfall amount, forward cumulative humidity, inversely accumulate humidity, forward cumulative illumination When number, inversely accumulate lighting delay number, the forward direction of these meteorological elements or the beginning and ending time inversely accumulated by mass-election, according to plant Flowers and fruits breeding time of object start or the end time day sequence correlativity principle of selecting the best qualified determine that the beginning and ending time of accumulation is a kind of Be it is fixed, it is over the years in each year accumulation beginning and ending time it is all identical, referred to herein as forward cumulative；Another kind is with plant Flowers and fruits breeding time start or the Close Date for accumulation from date, back accumulation arrive preceding D_cIt, such as takes D_c=30, it is referred to as herein Inversely to accumulate, it is over the years in each year plant flowers and fruits breeding time beginning or the end time it is all different, therefore inversely accumulate Beginning and ending time is all different because of the beginning of flowers and fruits breeding time or the difference of end time of plant, is Nonlinear predictor Meteorological factor is obtained according to historical experience, is that the composition of each meteorological factor of Nonlinear predictor and expression way need Want depending on the circumstances；Day sequence in a few days serial number of the phase relative to January 1, if the day sequence on January 1 is 1, the day sequence on January 2 It is 2, and so on；The phenology factor is the phenological period of other plants or opposite early-maturing variety based on phenology.

Step 2: being started according to the flowers and fruits breeding time of plant over the years or the day sequence of end time, the flowers and fruits for estimating plant are raw The phase of educating starts or end time possible most sequence and possible day sequence at the latest early, correspondence are denoted as D₁And D₂；Wherein, D₁< D₂。

Step 3: only in the presence of the linear prediction factor, the flowers and fruits breeding time for calculating each year plant in over the years starts Or the day sequence of end time starts with the flowers and fruits breeding time of corresponding time plant or each linear prediction factor of end time Linearly dependent coefficient；Then related significance verification is carried out to all linearly dependent coefficients；It will be verified again by related significance All linearly dependent coefficients corresponding to the linear prediction factor as available predictors.

Only in the presence of Nonlinear predictor, the flowers and fruits breeding time of plant over the years is started or the institute of end time There is Nonlinear predictor as available predictors.

In the presence of the linear prediction factor and Nonlinear predictor are equal, the flowers and fruits of each year plant in over the years are calculated Breeding time starts or the day sequence of end time starts with the flowers and fruits breeding time of corresponding time plant or each of end time is linear The linearly dependent coefficient of predictor；Then related significance verification is carried out to all linearly dependent coefficients；Correlation will be passed through again The flowers and fruits breeding time of the linear prediction factor corresponding to all linearly dependent coefficients of checking validity and plant over the years start or All Nonlinear predictors of end time are as available predictors.

Step 4: according to available predictors, from three kinds of linear model, nonlinear model and mixed model typess of models Selecting a kind of types of models, wherein mixed model is made of linear block and nonlinear block as forecasting model constructed type, If available predictors are the linear prediction factor, any one of linear model, nonlinear model and mixed model are selected Types of models is as forecasting model constructed type；If available predictors are Nonlinear predictor, nonlinear model is selected Type is as forecasting model constructed type；If the existing linear prediction factor of available predictors has Nonlinear predictor again, select Any types of models in nonlinear model and mixed model is selected as forecasting model constructed type, when selection mixed model is made Linear block when for forecasting model constructed type in mixed model can only select linear predictor to construct；Then according to pre- Model construction type is reported, best forecasting model is constructed.

A1, by the flowers and fruits breeding time of plant over the years start or the end time day sequence and flowers and fruits breeding time of plant over the years open Begin or all available predictors of end time constitute historical sample set.

A2, historical sample set is by year divided into two subsets, respectively as modeling sample set and test samples collection It closes.

A3, using regression equation method, and utilize modeling sample set, establish initial forecasting model.

A4, by year all available predictors in each time in test samples set are input to and initially forecast mould In type, the day sequence predicted value in corresponding time is obtained.

A5, by year calculate each time day sequence predicted value with the corresponding time in test samples set day sequence difference Value；Then the average value of corresponding difference of all times is calculated；Then judge whether average value is less than or equal to setting value D_s1, such as Fruit is, then initial forecasting model is determined as best forecasting model, and by all available predictors in historical sample set It is determined as selecting predictor eventually, then executes step 5；Otherwise, step A6 is executed；Wherein, D_s1∈ [1,3] day, such as takes D_s1=2 days.

A6, the corresponding available predictors of minimal linear related coefficient in historical sample set are deleted, then returns to step Rapid A2 is continued to execute, if all available predictors in historical sample set are disposed and do not determine best forecasting model, Then without forecast.

B1, it is over the years in the currently pending time is by year defined as current year.

B2, by D in current year₁~D₂Currently pending day sequence is defined as when day before yesterday sequence in day sequence.

B3, ought day before yesterday sequence be denoted as D_cur；It ought day before yesterday sequence D_curAs additional predictor, and ought day before yesterday sequence D_curIt is It is no to be greater than or equal to D₀Return value as Forecasting Object；Wherein, D_cur∈[D₁,D₂], D₀Indicate the flowers and fruits of current year plant Breeding time starts or the day sequence of end time, D₁< D₀< D₂If D_cur≥D₀Then Forecasting Object value is 1, if D_cur< D₀It is then pre- Reporting object value is 0.

B4, the flowers and fruits breeding time of Forecasting Object, additional predictor, plant over the years is started or all times of end time Predictor is selected to constitute a historical sample；The method of the present invention is to judge whether handled forecast day sequence is big actually giving the correct time in advance Start or the day sequence of Close Date in or equal to flowers and fruits breeding time, in history of forming sample, Forecasting Object is as day before yesterday sequence D_cur Whether D is greater than or equal to₀Return value, as day before yesterday sequence D_curBeing pre flowers and fruits breeding time first starts or the day of Close Date Sequence, then being started using flowers and fruits breeding time or the Close Date is back accumulated as from date to preceding D_cIt the forecast inversely accumulated because Son, it is necessary to for as day before yesterday sequence D_cur, calculate the factor values corresponding to it.

B5, by D in current year₁~D₂Next day sequence to be processed is used as when day before yesterday sequence in day sequence, then return step B3 is continued to execute, until D in current year₁~D₂An each of day sequence day sequence is disposed, and executes step B6.

B6, will it is over the years in next time to be processed as current year, then return step B2 is continued to execute, up to Each time over the years is disposed, will it is over the years in D in all times₁~D₂The corresponding all historical samples of day sequence constitute history Sample set.

B7, historical sample set is by year divided into two subsets, respectively as modeling sample set and test samples collection It closes.

B8, using artificial neural network or support vector machine method, and utilize modeling sample set, establish initial forecast Model.

B9, by D in each time in test samples set₁~D₂Additional forecast in the historical sample day by day of day sequence The factor and its corresponding all available predictors are input in initial forecasting model, obtain the additional corresponding forecast of predictor Value, is set to 1 for predicted value if predicted value >=0.5；Predicted value is set to 0 if predicted value < 0.5；The additional forecast of judgement Whether the corresponding predicted value of factor Forecasting Object corresponding with the additional predictor in test samples set is identical, if identical It then indicates that forecast is correct, forecast mistake is indicated if not identical；If all historical samples in test samples set are corresponding Overall forecast accuracy reaches 80%, then initial forecasting model is determined as best forecasting model, and will be in historical sample set All available predictors be determined as selecting predictor eventually, then execute step 5；Otherwise, step B10 is executed.

If the available predictors in B10, historical sample set are the linear prediction factor, will be in historical sample set The corresponding linear prediction factor of minimal linear related coefficient is deleted；If the available predictors in historical sample set are non-thread Property predictor, then by historical sample set any one Nonlinear predictor delete；If in historical sample set The existing linear prediction factor of available predictors has Nonlinear predictor again, then minimal linear in historical sample set is related The corresponding linear prediction factor of coefficient or any one Nonlinear predictor are deleted.

B11, return step B7 are continued to execute, if all available predictors in historical sample set be disposed and not Determine best forecasting model, it is determined that without forecast.

If C1, available predictors are the linear prediction factor, the flowers and fruits breeding time of plant over the years is started or terminated Time day sequence and plant over the years flowers and fruits breeding time start or the part available predictors of end time constitute linear block Historical sample set；If the existing linear prediction factor of available predictors has Nonlinear predictor, by plant over the years again Flowers and fruits breeding time start or the end time day sequence and flowers and fruits breeding time of plant over the years start or all or portion of end time Heterogeneous linear predictor constitutes linear block historical sample set.

C2, linear block historical sample set is by year divided into two subsets, respectively as linear block modeling sample Set and linear block test samples set.

C3, using regression equation method, and utilize linear block modeling sample set, establish initial linear module.

C4, by year by all available predictors in each time in linear block test samples set be input to just In beginning linear block, the day sequence predicted value in corresponding time is obtained.

C5, by year calculate each time day sequence predicted value with the corresponding time in linear block test samples set The difference of day sequence；Then the average value of corresponding difference of all times is calculated；Then judge whether average value is less than or equal to set Definite value D_s1, if it is, initial linear module is determined as optimum linear module, and will be in linear block historical sample set The end that all available predictors are determined as optimum linear module selects predictor, then will be in linear block historical sample set All ends in each time select predictor to be input in optimum linear module, obtain the day sequence predicted value in corresponding time, execute Step C7；Otherwise, step C6 is executed；Wherein, D_s1∈ [1,3] day, such as takes D_s1=2 days.

C6, the corresponding available predictors of minimal linear related coefficient in linear block historical sample set are deleted, so Return step C2 is continued to execute afterwards, if all available predictors in linear block historical sample set be disposed and not really Determine optimum linear module, then linear block is invalid, and mixed model will only include nonlinear block, and forecasting model constructed type is changed For nonlinear model, i.e. execution step B1 to step B11.

C7, it is over the years in the currently pending time is by year defined as current year.

C8, by D in current year₁~D₂Currently pending day sequence is defined as when day before yesterday sequence in day sequence.

C9, ought day before yesterday sequence be denoted as D_cur；It ought day before yesterday sequence D_curAs the first additional predictor, and ought day before yesterday sequence D_curMore than or equal to D₀Return value as Forecasting Object；The current year that step C5 is obtained by optimum linear module Day, sequence predicted value was as the second additional predictor；Wherein, D_cur∈[D₁,D₂], D₀Indicate the flowers and fruits fertility of current year plant Phase starts or the day sequence of end time, D₁< D₀< D₂If D_cur≥D₀Then Forecasting Object value is 1, if D_cur< D₀Then forecast pair As value is 0.

C10, by Forecasting Object, the first additional predictor, the second additional predictor, plant over the years flowers and fruits breeding time Start or the end time day sequence and plant over the years flowers and fruits breeding time start or the end time except constitute linear block history Remaining available predictors outside sample set constitute a nonlinear block historical sample；The method of the present invention is given the correct time in advance actually It is to judge whether handled forecast day sequence is greater than or equal to flowers and fruits breeding time and starts or the day sequence of Close Date, it is non-thread being formed Property module history sample when Forecasting Object be as day before yesterday sequence D_curWhether D is greater than or equal to₀Return value, as day before yesterday sequence D_curFirst Being pre flowers and fruits breeding time starts or the day sequence of Close Date, then being started using flowers and fruits breeding time or the Close Date is starting date Phase is back accumulated to preceding D_cIt the predictor inversely accumulated, it is necessary to for as day before yesterday sequence D_cur, calculate corresponding to it because Subvalue.

C11, by D in current year₁~D₂Next day sequence to be processed is used as when day before yesterday sequence in day sequence, then returns to step Rapid C9 is continued to execute, until D in current year₁~D₂An each of day sequence day sequence is disposed, and executes step C12.

C12, will it is over the years in next time to be processed as current year, then return step C8 is continued to execute, directly Be disposed to each time over the years, will it is over the years in D in all times₁~D₂The corresponding all historical samples of day sequence constitute non- Linear block historical sample set.

C13, nonlinear block historical sample set is by year divided into two subsets, modeled respectively as nonlinear block Sample set and nonlinear block test samples set.

C14, it is built jointly using artificial neural network or support vector machine method, and using nonlinear block modeling sample collection Vertical initial non-linearities module.

C15, by D in each time in nonlinear block test samples set₁~D₂The nonlinear block day by day of day sequence The first additional predictor and its corresponding all available predictors in historical sample are input in initial non-linearities module, The corresponding predicted value of the first additional predictor is obtained, predicted value is set to 1 if predicted value >=0.5；If predicted value < Predicted value is set to 0 by 0.5；Judge in the corresponding predicted value of the first additional predictor and nonlinear block test samples set The corresponding Forecasting Object of the first additional predictor it is whether identical, if the same indicate that forecast is correct, if not identical Indicate forecast mistake；If the corresponding overall forecast of all nonlinear block historical samples in nonlinear block test samples set Accuracy reaches 80%, then initial non-linearities module is determined as Optimal Nonlinear module, and by nonlinear block historical sample The end that all available predictors in set are determined as Optimal Nonlinear module selects predictor, will by optimum linear module and The mixed model of Optimal Nonlinear module composition is as best forecasting model, then executes step 5；Otherwise, step C16 is executed.

If the available predictors in C16, nonlinear block historical sample set are the linear prediction factor, will be non-thread Property module history sample set in the corresponding linear prediction factor of minimal linear related coefficient delete；If nonlinear block history sample Available predictors in this set are Nonlinear predictor, then will be any one in nonlinear block historical sample set A Nonlinear predictor is deleted；If the existing linear prediction factor of available predictors in nonlinear block historical sample set There is Nonlinear predictor again, then by the corresponding linear prediction of minimal linear related coefficient in nonlinear block historical sample set The factor or any one Nonlinear predictor are deleted.

C17, return step C13 are continued to execute, if all available predictors in nonlinear block historical sample set It is disposed and does not determine Optimal Nonlinear module, it is determined that without forecast.

Step 5: the beginning of flowers and fruits breeding time of plant to be forecast or the time of end time are defined as the forecast time； The day sequence for starting from the date of forecast has been defined as report day sequence, has been denoted as D_s；Meteorology can be provided to the longest of weather forecast The number of days of effective timeliness is denoted as D_max。

When the type of best forecasting model is linear model, with being selected needed for predictor eventually for calculating for forecast time The best forecasting model of material computation corresponding each end select the factor values of predictor, best forecasting model is corresponding all It selects the factor values of predictor to be input in best forecasting model eventually, a day sequence value is calculated；If the day sequence value being calculated In the time range D of the flowers and fruits breeding time of plant to be forecast₁~D₂It is interior, then it is assumed that the day sequence value being calculated is effective, is plant Flowers and fruits breeding time beginning or the end time day sequence forecast result, then execute step 6；Otherwise it is assumed that being calculated Day sequence value it is invalid, without forecast.

Step 5_1a, by the day sequence D on the date currently calculated_jIt is defined as calculating day sequence, wherein D_jInitial value be D_s。

Step 5_1b, according to D_jThe factor values that best forecasting model corresponding each end selects predictor are calculated, it will be best Forecasting model corresponding all ends select the factor values of predictor to be input in best forecasting model, and predicted value is calculated.

If predicted value >=0.5 step 5_1c, being calculated, works as D_jIn flowers and fruits breeding time of plant to be forecast Time range D₁~D₂When interior, it is believed that D_jEffectively, be plant to be forecast flowers and fruits breeding time beginning or the day sequence of end time Forecast result, then execute step 6；Work as D_jNot in the time range D of the flowers and fruits breeding time of plant to be forecast₁~D₂When interior, Think D_jIn vain, without forecast；If predicted value < 0.5 being calculated, thens follow the steps 5_1d.

If step 5_1d, D_j> D₂Or D_j> D_s+D_max, then without forecast；Otherwise, D is enabled_j=D_j+ 1, by it is next into The day sequence on the date that row calculates returns again to step 5_1b and continues to execute as day sequence is calculated；Wherein, D_j=D_j"=" in+1 is Assignment.

If step 5_2a, D_s< D₁-D_maxOr D_s> D₂, then without forecast；Otherwise, step 5_2b is executed.

Step 5_2b, by the day sequence D on the date currently calculated_jIt is defined as calculating day sequence, wherein D_jInitial value be D_s。

Step 5_2c, according to D_jThe factor values that best forecasting model corresponding each end selects predictor are calculated, it will be best Forecasting model corresponding all ends select the factor values of predictor to be input in best forecasting model, and predicted value is calculated.

If predicted value >=0.5 step 5_2d, being calculated, works as D_jIn flowers and fruits breeding time of plant to be forecast Time range D₁~D₂When interior, it is believed that D_jEffectively, be plant to be forecast flowers and fruits breeding time beginning or the day sequence of end time Forecast result, then execute step 6；Work as D_jNot in the time range D of the flowers and fruits breeding time of plant to be forecast₁~D₂When interior, Think D_jIn vain, without forecast；If predicted value < 0.5 being calculated, thens follow the steps 5_2e.

If step 5_2e, D_j> D₂Or D_j> D_s+D_max, then without forecast；Otherwise, D is enabled_j=D_j+ 1, by it is next into The day sequence on the date that row calculates returns again to step 5_2c and continues to execute as day sequence is calculated；Wherein, D_j=D_j"=" in+1 is Assignment.

In the present embodiment, in step A2 and the step B7, the quantity of the sample in modeling sample set is history The 70% of the quantity of sample in sample set, the quantity of the sample in test samples set are the sample in historical sample set Quantity 30%.

In the present embodiment, in step C2, the quantity of the sample in linear block modeling sample set is gone through for linear block The quantity of the 70% of the quantity of sample in history sample set, the sample in linear block test samples set is gone through for linear block The 30% of the quantity of sample in history sample set；Sample in the step C13, in nonlinear block modeling sample set Quantity be sample in nonlinear block historical sample set quantity 70%, in nonlinear block test samples set The quantity of sample is the 30% of the quantity of the sample in nonlinear block historical sample set.

Claims

1. a kind of forecasting procedure of plant flowers and fruits beginning and ending time breeding time, it is characterised in that the following steps are included:

Step 1: the flowers and fruits breeding time for collecting plant starts or the historical summary of end time, the flowers and fruits fertility including plant over the years Phase start or the end time day sequence, start over the years to the flowers and fruits breeding time of plant or the end time it is relevant it is multiple it is meteorological because Son starts to the flowers and fruits breeding time of plant or the end time relevant phenology factor over the years, meteorological factor be linear prediction because Son is Nonlinear predictor, and the phenology factor is the linear prediction factor or is Nonlinear predictor；If linear prediction The factor and Nonlinear predictor are not present, then without forecast, otherwise, execute step 2；

Above-mentioned, range over the years is 10~30 years, has forward cumulative temperature, reverse accumulation for the meteorological factor of the linear prediction factor Temperature, forward cumulative precipitation, reverse accumulative rainfall amount, forward cumulative humidity, inversely accumulate humidity, forward cumulative illumination when Number inversely accumulates lighting delay number, for the meteorological factor of Nonlinear predictor is obtained according to historical experience；

Step 2: being started according to the flowers and fruits breeding time of plant over the years or the day sequence of end time, estimate the flowers and fruits breeding time of plant Start or end time possible most sequence and possible day sequence at the latest early, correspondence are denoted as D₁And D₂；Wherein, D₁< D₂；

Step 3: only in the presence of the linear prediction factor, the flowers and fruits breeding time for calculating each year plant in over the years starts or ties The day sequence of beam time start with the flowers and fruits breeding time of corresponding time plant or each linear prediction factor of end time it is linear Related coefficient；Then related significance verification is carried out to all linearly dependent coefficients；The institute that will be verified again by related significance The linear prediction factor is as available predictors corresponding to linear related coefficient；

Only in the presence of Nonlinear predictor, by the flowers and fruits breeding time of plant over the years start or the end time it is all non- The linear prediction factor is as available predictors；

In the presence of the linear prediction factor and Nonlinear predictor are equal, the flowers and fruits fertility of each year plant in over the years is calculated Phase starts or the day sequence of end time starts with the flowers and fruits breeding time of corresponding time plant or each linear prediction of end time The linearly dependent coefficient of the factor；Then related significance verification is carried out to all linearly dependent coefficients；It again will be by related significant Property verification all linearly dependent coefficients corresponding to flowers and fruits breeding time of the linear prediction factor and plant over the years start or terminate All Nonlinear predictors of time are as available predictors；

Step 4: according to available predictors, being selected from three kinds of linear model, nonlinear model and mixed model typess of models A kind of types of models is as forecasting model constructed type, and wherein mixed model is made of linear block and nonlinear block, if waiting Selecting predictor is the linear prediction factor, then selects any model in linear model, nonlinear model and mixed model Type is as forecasting model constructed type；If available predictors are Nonlinear predictor, nonlinear model is selected to make For forecasting model constructed type；If the existing linear prediction factor of available predictors has Nonlinear predictor again, select non- Any types of models in linear model and mixed model as forecasting model constructed type, when select mixed model as pre- Linear block when reporting model construction type in mixed model can only select linear predictor to construct；Then mould according to weather report Type constructed type constructs best forecasting model；

A1, by the flowers and fruits breeding time of plant over the years start or the end time day sequence and plant over the years flowers and fruits breeding time start or All available predictors of end time constitute historical sample set；

A2, historical sample set is by year divided into two subsets, respectively as modeling sample set and test samples set；

A4, all available predictors in each time in test samples set are by year input to initial forecasting model In, obtain the day sequence predicted value in corresponding time；

A5, by year calculate each time day sequence predicted value with the corresponding time in test samples set day sequence difference； Then the average value of corresponding difference of all times is calculated；Then judge whether average value is less than or equal to setting value D_s1If It is initial forecasting model to be then determined as best forecasting model, and all available predictors in historical sample set are true It is set to and selects predictor eventually, then executes step 5；Otherwise, step A6 is executed；Wherein, D_s1∈ [1,3] day；

A6, the corresponding available predictors of minimal linear related coefficient in historical sample set are deleted, then return step A2 It continues to execute, if all available predictors in historical sample set are disposed and do not determine best forecasting model, no It is forecast；

B3, ought day before yesterday sequence be denoted as D_cur；It ought day before yesterday sequence D_curAs additional predictor, and ought day before yesterday sequence D_curIt is whether big In or equal to D₀Return value as Forecasting Object；Wherein, D_cur∈[D₁,D₂], D₀Indicate the flowers and fruits fertility of current year plant Phase starts or the day sequence of end time, D₁< D₀< D₂If D_cur≥D₀Then Forecasting Object value is 1, if D_cur< D₀Then forecast pair As value is 0；

B4, by Forecasting Object, additional predictor, plant over the years flowers and fruits breeding time start or the end time it is all candidate pre- The factor is reported to constitute a historical sample；

B5, by D in current year₁~D₂Next day sequence to be processed is used as when day before yesterday sequence in day sequence, then return step B3 after It is continuous to execute, until D in current year₁~D₂An each of day sequence day sequence is disposed, and executes step B6；

B6, will it is over the years in next time to be processed as current year, then return step B2 is continued to execute, until over the years Each time be disposed, will it is over the years in D in all times₁~D₂The corresponding all historical samples of day sequence constitute historical samples Set；

B7, historical sample set is by year divided into two subsets, respectively as modeling sample set and test samples set；

B8, using artificial neural network or support vector machine method, and utilize modeling sample set, establish initial forecast mould Type；

B9, by D in each time in test samples set₁~D₂Additional predictor in the historical sample day by day of day sequence And its corresponding all available predictors are input in initial forecasting model, obtain the additional corresponding predicted value of predictor, Predicted value is set to 1 if predicted value >=0.5；Predicted value is set to 0 if predicted value < 0.5；Judge additional predictor Whether corresponding predicted value Forecasting Object corresponding with the additional predictor in test samples set is identical, if the same table Show that forecast is correct, forecast mistake is indicated if not identical；If the corresponding totality of all historical samples in test samples set Forecast accuracy reaches 80%, then initial forecasting model is determined as best forecasting model, and by the institute in historical sample set There are available predictors to be determined as selecting predictor eventually, then executes step 5；Otherwise, step B10 is executed；

It, will be minimum in historical sample set if the available predictors in B10, historical sample set are the linear prediction factor The corresponding linear prediction factor of linearly dependent coefficient is deleted；If the available predictors in historical sample set are non-linear pre- The factor is reported, then is deleted any one Nonlinear predictor in historical sample set；If the candidate in historical sample set The existing linear prediction factor of predictor has Nonlinear predictor again, then by minimal linear related coefficient in historical sample set The corresponding linear prediction factor or any one Nonlinear predictor are deleted；

B11, return step B7 are continued to execute, if all available predictors in historical sample set are disposed and do not determine Best forecasting model, it is determined that without forecast；

If C1, available predictors are the linear prediction factor, the flowers and fruits breeding time of plant over the years is started or the end time Day sequence and plant over the years flowers and fruits breeding time start or the part available predictors of end time constitute linear block history Sample set；If the existing linear prediction factor of available predictors has Nonlinear predictor, by the flowers and fruits of plant over the years again Breeding time start or the end time day sequence and flowers and fruits breeding time of plant over the years start or all or part of lines of end time Property predictor constitute linear block historical sample set；

C2, linear block historical sample set is by year divided into two subsets, respectively as linear block modeling sample set With linear block test samples set；

C4, all available predictors in each time in linear block test samples set are by year input to initial line Property module in, obtain the day sequence predicted value in corresponding time；

C5, by year calculate each time day sequence predicted value and the corresponding time in linear block test samples set day sequence Difference；Then the average value of corresponding difference of all times is calculated；Then judge whether average value is less than or equal to setting value D_s1, if it is, initial linear module is determined as optimum linear module, and will be all in linear block historical sample set The end that available predictors are determined as optimum linear module selects predictor, then by each of linear block historical sample set All ends in time select predictor to be input in optimum linear module, obtain the day sequence predicted value in corresponding time, execute step C7；Otherwise, step C6 is executed；Wherein, D_s1∈ [1,3] day；

C6, the corresponding available predictors of minimal linear related coefficient in linear block historical sample set are deleted, is then returned It returns step C2 to continue to execute, if all available predictors in linear block historical sample set are disposed and do not determine most Good linear block, then linear block is invalid, and mixed model will only include nonlinear block, forecasting model constructed type is changed to non- Linear model；

C9, ought day before yesterday sequence be denoted as D_cur；It ought day before yesterday sequence D_curAs the first additional predictor, and ought day before yesterday sequence D_curGreatly In or equal to D₀Return value as Forecasting Object；The day sequence for the current year that step C5 is obtained by optimum linear module is pre- Report value is as the second additional predictor；Wherein, D_cur∈[D₁,D₂], D₀Indicate that the flowers and fruits breeding time of current year plant starts Or the day sequence of end time, D₁< D₀< D₂If D_cur≥D₀Then Forecasting Object value is 1, if D_cur< D₀Then Forecasting Object value It is 0；

C10, the flowers and fruits breeding time of Forecasting Object, the first additional predictor, the second additional predictor, plant over the years is started Or the end time day sequence and plant over the years flowers and fruits breeding time start or the end time except constitute linear block historical sample Remaining available predictors outside set constitute a nonlinear block historical sample；

C11, by D in current year₁~D₂Next day sequence to be processed is used as when day before yesterday sequence in day sequence, then return step C9 It continues to execute, until D in current year₁~D₂An each of day sequence day sequence is disposed, and executes step C12；

C12, will it is over the years in next time to be processed as current year, then return step C8 is continued to execute, up to going through Year each time be disposed, will it is over the years in D in all times₁~D₂The corresponding all historical samples of day sequence constitute non-linear Module history sample set；

C13, nonlinear block historical sample set is by year divided into two subsets, respectively as nonlinear block modeling sample Set and nonlinear block test samples set；

C14, it is established just using artificial neural network or support vector machine method, and using nonlinear block modeling sample set Beginning nonlinear block；

C15, by D in each time in nonlinear block test samples set₁~D₂The nonlinear block history day by day of day sequence The first additional predictor and its corresponding all available predictors in sample are input in initial non-linearities module, are obtained The corresponding predicted value of first additional predictor, is set to 1 for predicted value if predicted value >=0.5；If predicted value < 0.5 Predicted value is set to 0；Judge the in the corresponding predicted value of the first additional predictor and nonlinear block test samples set Whether the corresponding Forecasting Object of one additional predictor is identical, if the same indicates that forecast is correct, indicates if not identical Forecast mistake；If the corresponding overall forecast of all nonlinear block historical samples in nonlinear block test samples set is correct Rate reaches 80%, then initial non-linearities module is determined as Optimal Nonlinear module, and by nonlinear block historical sample set In all available predictors be determined as end of Optimal Nonlinear module and select predictor, will be by optimum linear module and best The mixed model that nonlinear block is constituted is as best forecasting model, then executes step 5；Otherwise, step C16 is executed；

If the available predictors in C16, nonlinear block historical sample set are the linear prediction factor, by nonlinear model The corresponding linear prediction factor of minimal linear related coefficient is deleted in block historical sample set；If nonlinear block historical sample collection Available predictors in conjunction are Nonlinear predictor, then any one in nonlinear block historical sample set is non- The linear prediction factor is deleted；If the existing linear prediction factor of available predictors in nonlinear block historical sample set has again Nonlinear predictor, then by the corresponding linear prediction factor of minimal linear related coefficient in nonlinear block historical sample set Or any one Nonlinear predictor is deleted；

C17, return step C13 are continued to execute, if all available predictors in nonlinear block historical sample set are handled It finishes and does not determine Optimal Nonlinear module, it is determined that without forecast；

Step 5: the beginning of flowers and fruits breeding time of plant to be forecast or the time of end time are defined as the forecast time；It will open The day sequence for beginning to make the date of forecast has been defined as report day sequence, is denoted as D_s；The longest that meteorology can be provided weather forecast is effective The number of days of timeliness is denoted as D_max；

When the type of best forecasting model be linear model when, with forecast the time for calculate eventually select predictor needed for money Material calculates the factor values for selecting predictor at best forecasting model corresponding each end, by the corresponding all choosings eventually of best forecasting model The factor values of predictor are input in best forecasting model, and a day sequence value is calculated；If be calculated day sequence value to The time range D of the flowers and fruits breeding time of the plant of forecast₁~D₂It is interior, then it is assumed that the day sequence value being calculated is effective, is the flower of plant The beginning of fruit breeding time or end time day sequence forecast result, then execute step 6；Otherwise it is assumed that the day being calculated Sequence value is invalid, without forecast；

When the type of best forecasting model is nonlinear model, if D_s< D₁-D_maxOr D_s> D₂, then without forecast；It is no Then, following steps are executed:

Step 5_1b, according to D_jThe factor values that best forecasting model corresponding each end selects predictor are calculated, will most preferably forecast mould Type corresponding all ends select the factor values of predictor to be input in best forecasting model, and predicted value is calculated；

If predicted value >=0.5 step 5_1c, being calculated, works as D_jIn the time of the flowers and fruits breeding time of plant to be forecast Range D₁~D₂When interior, it is believed that D_jEffectively, be plant to be forecast flowers and fruits breeding time beginning or the end time day sequence it is pre- Then report is as a result, execute step 6；Work as D_jNot in the time range D of the flowers and fruits breeding time of plant to be forecast₁~D₂When interior, it is believed that D_jIn vain, without forecast；If predicted value < 0.5 being calculated, thens follow the steps 5_1d；

If step 5_1d, D_j> D₂Or D_j> D_s+D_max, then without forecast；Otherwise, D is enabled_j=D_j+ 1, it is counted next The day sequence on the date of calculation returns again to step 5_1b and continues to execute as day sequence is calculated；Wherein, D_j=D_j"=" in+1 is assignment Symbol；

When the type of best forecasting model be mixed model when, with forecast the time for calculate eventually select predictor needed for money Linear block corresponding each end that material calculates best forecasting model selects the factor values of predictor, by the line of best forecasting model Property module corresponding all ends select the factor values of predictor to be input in the linear block of best forecasting model, and line is calculated Property module forecast day sequence value；If the day sequence value for the linear block forecast being calculated is not raw in the flowers and fruits of plant to be forecast Educate phase time range D₁~D₂It is interior, it is believed that the day sequence value for the linear block forecast being calculated is invalid, without forecast；If meter Obtain linear block forecast day sequence value plant to be forecast flowers and fruits breeding time time range D₁~D₂It is interior, then it is assumed that The day sequence value for the linear block forecast being calculated is effective, then executes following steps:

Step 5_2c, according to D_jThe factor values that best forecasting model corresponding each end selects predictor are calculated, will most preferably forecast mould Type corresponding all ends select the factor values of predictor to be input in best forecasting model, and predicted value is calculated；

If predicted value >=0.5 step 5_2d, being calculated, works as D_jIn the time of the flowers and fruits breeding time of plant to be forecast Range D₁~D₂When interior, it is believed that D_jEffectively, be plant to be forecast flowers and fruits breeding time beginning or the end time day sequence it is pre- Then report is as a result, execute step 6；Work as D_jNot in the time range D of the flowers and fruits breeding time of plant to be forecast₁~D₂When interior, it is believed that D_jIn vain, without forecast；If predicted value < 0.5 being calculated, thens follow the steps 5_2e；

If step 5_2e, D_j> D₂Or D_j> D_s+D_max, then without forecast；Otherwise, D is enabled_j=D_j+ 1, it is counted next The day sequence on the date of calculation returns again to step 5_2c and continues to execute as day sequence is calculated；Wherein, D_j=D_j"=" in+1 is assignment Symbol；

Step 6: obtained forecast result being converted into the date, the beginning of the flowers and fruits breeding time of the plant as the forecast time or knot The date of beam, forecast terminate.

2. the forecasting procedure of plant flowers and fruits beginning and ending time breeding time according to claim 1 a kind of, it is characterised in that described Step 1 in, the phenology factor is the phenological period of other plants based on phenology or opposite early-maturing variety.

3. the forecasting procedure of plant flowers and fruits beginning and ending time breeding time according to claim 1 or 2 a kind of, it is characterised in that institute In the step A2 and the step B7 stated, the quantity of the sample in modeling sample set is the sample in historical sample set The 70% of quantity, the quantity of the sample in test samples set are the 30% of the quantity of the sample in historical sample set.

4. the forecasting procedure of plant flowers and fruits beginning and ending time breeding time according to claim 3 a kind of, it is characterised in that described Step C2 in, the quantity of the sample in linear block modeling sample set is the sample in linear block historical sample set The 70% of quantity, the quantity of the sample in linear block test samples set are the sample in linear block historical sample set The 30% of quantity；In the step C13, the quantity of the sample in nonlinear block modeling sample set is gone through for nonlinear block The 70% of the quantity of sample in history sample set, the quantity of the sample in nonlinear block test samples set are nonlinear model The 30% of the quantity of sample in block historical sample set.