Crop yield forecasting method

Abstract

The invention discloses a crop yield forecasting method, and relates to the technical field of agricultural crop yield forecasting. Comprising the steps of collecting meteorological observation data and crop data; carrying out related meteorological element analysis on the meteorological observation data to determine meteorological factors, and carrying out stepwise regression analysis on the meteorological factors to screen out key meteorological factors; crop growth period and yield analysis is carried out on the crop data, and trend yield is calculated; and inputting the key meteorological factors and the trend yield into the BP neural network to forecast the crop yield, thereby improving the crop yield forecasting efficiency and accuracy. According to the invention, the number of input parameters is reduced by screening key meteorological factors, and the operation speed is improved; and inputting the key meteorological factors and the trend yield into the BP neural network to forecast the crop yield, thereby improving the crop yield forecasting efficiency and accuracy.

Description

Technical Field

[0001] The invention relates to the technical field of agricultural crop yield forecasting, and in particular to a crop yield forecasting method. Background Art

[0002] In recent years, agricultural meteorological disasters have shown a trend of high frequency, great intensity and serious harm.

[0003] In the existing technology, the crop yield meteorological forecast model based on the regression statistical model has been studied for remote sensing yield estimation, and a national, provincial and municipal (county / district) agricultural meteorological yield forecast business system has been established, using dBASEⅢ and BASIC language to establish a forecast service system.

[0004] The defects of the above-mentioned prior art are: data transmission between dBASEⅢ and high-level BASIC language is mostly transmitted by compressing data into files. In practical processes, the transmission of a small number of parameters is often encountered. The method of using files to transmit data will affect the running speed of the program, thereby affecting the efficiency of crop yield forecasting. Summary of the invention

[0005] Based on this, it is necessary to provide a crop yield forecasting method to address the above technical issues.

[0006] An embodiment of the present invention provides a crop yield forecasting method, comprising:

[0007] Collect meteorological observations and crop data;

[0008] Analyze the meteorological observation data for relevant meteorological elements to determine the meteorological factors, and perform stepwise regression analysis on the meteorological factors to screen out the key meteorological factors;

[0009] Analyze crop growth period and yield of crop data and calculate trend yield;

[0010] Key meteorological factors and trend yields are input into the BP neural network to predict crop yields.

[0011] In addition, the meteorological observation data includes temperature, precipitation, and sunshine;

[0012] The crop data include growth period, yield, and yield factors.

[0013] In addition, the analysis of relevant meteorological elements includes: according to the impact of temperature, precipitation, sunshine hours and meteorological disasters on yield, principal component analysis and hierarchical analysis are performed on the factors of temperature, precipitation, sunshine hours, meteorological disasters and yield to determine the meteorological factors that affect crop yield.

[0014] In addition, the calculation of trend yield includes: performing a one-dimensional linear fitting on the results obtained from the crop growth period and yield analysis to obtain a fitting equation of yield and crop growth period, and calculating the trend yield through the fitting equation.

[0015] In addition, the BP neural network is:

[0016] Select S-type transfer function

[0017]

[0018] Back propagation error function

[0019]

[0020] Among them, t i is the expected output, O i For the network's computational output, the weights and thresholds are continuously adjusted to minimize the error function E.

[0021] Compared with the prior art, the crop yield forecasting method provided by the embodiment of the present invention has the following beneficial effects:

[0022] The present invention adopts stepwise regression analysis to screen key meteorological factors, and adopts one-dimensional linear fitting regression to calculate trend yield. By screening key meteorological factors, the number of input parameters is reduced and the running speed is improved. The key meteorological factors and trend yield are input into the BP neural network to predict crop yield, thereby improving the efficiency and accuracy of crop yield forecasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of the structure of a crop yield forecasting method provided in one embodiment;

[0024] Figure 2 A graph showing the change trend of spring and summer corn yields from 2010 to 2020 provided in an embodiment;

[0025] Figure 3 A stepwise regression analysis result diagram of meteorological elements and yield of winter wheat during the whole growth period and at different time periods in a crop yield forecasting method provided in an embodiment;

[0026] Figure 4 A stepwise regression analysis result diagram of meteorological elements and meteorological yields during the whole growth period and at different time periods of a crop yield forecasting method provided in an embodiment;

[0027] Figure 5 A stepwise regression analysis result diagram of meteorological elements and meteorological yield during the whole growth period and different periods of summer corn in a crop yield forecasting method provided in an embodiment;

[0028] Figure 6 A comparison chart of the fitted yield and actual yield of winter wheat calculated by the GRNN, BP and RBF neural network methods of a crop yield forecasting method provided in an embodiment;

[0029] Figure 7 A comparison chart of the fitted yield and actual yield of spring corn calculated by GRNN, BP and RBF neural network methods of a crop yield forecasting method provided in an embodiment;

[0030] Figure 8 This is a comparison chart between the fitted yield and actual yield of summer corn calculated by GRNN, BP and RBF neural network modeling of a crop yield forecasting method provided in an embodiment. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0032] In one embodiment, a crop yield forecasting method is provided, such as Figure 1 As shown, the method includes:

[0033] 1. Data collection and organization

[0034] According to the research content, the growth period and yield data of the main winter wheat, spring and summer corn planting areas in Beijing (Changping, Daxing, Fangshan, Miyun, Pinggu, Shunyi, Tongzhou and other districts) from 2009 to 2020 were collected and organized, and the corresponding planting areas' daily average temperature, maximum temperature, minimum temperature, precipitation, sunshine hours and other data from 2009 to 2020 were collected to provide data support for the research content.

[0035] 2. Analysis of spring and summer corn yield trends and meteorological yields

[0036] In the past decade, due to social factors such as variety improvement and technological progress, the yield of spring and summer corn in Beijing has increased significantly. Therefore, this study first used a linear regression to calculate the trend yield and meteorological yield. Figure 2 As shown in Table 1, the meteorological output and meteorological forecast factors without social factors are used to build the forecast model.

[0037] Table 1 Trend yield and meteorological yield of spring and summer corn from 2010 to 2020 (kg / hectare)

[0038]

[0039] 3. Calculation of key meteorological factors in different growth periods

[0040] (1)Winter wheat

[0041] According to the effects of temperature, precipitation, sunshine hours and meteorological disasters on yield, the accumulated temperature greater than 0℃, precipitation, sunshine hours, days of frost damage in early spring, precipitation after greening, accumulated temperature after greening, sunshine hours after greening, days of cold damage, accumulated temperature before wintering, average temperature, average maximum temperature, average minimum temperature, precipitation, sunshine, etc. in five growth stages (sowing-overwintering, overwintering-greening, greening-rising, rising-jointing, jointing-heading) were calculated, totaling 34 meteorological factors, as shown in Table 2. Stepwise regression analysis was used to screen key predictors.

[0042] Table 2 Data on meteorological forecast factors for winter wheat yield from 2009 to 2021

[0043]

[0044]

[0045]

[0046] (2) Spring corn

[0047] According to the influence of temperature, precipitation, sunshine hours and meteorological disasters on spring corn yield, the accumulated temperature above 0℃, precipitation, sunshine hours and high temperature heat damage (the index is the number of days above 35℃) during the sowing-silking period, the precipitation, accumulated temperature and sunshine hours during the sowing-jointing period, the precipitation, accumulated temperature and sunshine hours during the tasseling-maturity period, the average temperature, the extreme maximum temperature, the extreme minimum temperature, precipitation and sunshine hours during the sowing-emergence period, the average temperature, the extreme maximum temperature, the extreme minimum temperature, precipitation and sunshine hours during the germination-jointing period, the average temperature, the extreme maximum temperature, the extreme minimum temperature, precipitation and sunshine hours during the jointing-tasseling period, the average temperature, the extreme maximum temperature, the extreme minimum temperature, precipitation and sunshine hours during the tasseling-silking period, a total of 30 meteorological factors were calculated, and the factor data were predicted year by year, as shown in Table 3. Stepwise regression analysis was used to screen key prediction factors.

[0048] Table 3 Data on meteorological forecast factors for spring corn yield from 2009 to 2020

[0049]

[0050]

[0051] (3) Summer corn

[0052] According to the influence of temperature, precipitation, sunshine hours and meteorological disasters on spring corn yield, the accumulated temperature above 0℃, precipitation, sunshine hours and days of high temperature heat damage during the sowing-silking period (the indicator is the number of days above 35℃ during the whole growth period), precipitation, accumulated temperature and sunshine hours during the sowing-jointing period, precipitation, accumulated temperature and sunshine hours during the tasseling-maturity period, average temperature, extreme maximum temperature, extreme minimum temperature, precipitation and sunshine hours during the sowing-seedling period, average temperature, extreme maximum temperature, extreme minimum temperature, precipitation and sunshine hours during the seedling-jointing period, average temperature, extreme maximum temperature, extreme minimum temperature, precipitation and sunshine hours during the jointing-tasseling period, average temperature, extreme maximum temperature, extreme minimum temperature, precipitation and sunshine hours during the tasseling-silking period, etc., a total of 30 meteorological factors were calculated, and the annual forecast factor data are shown in Table 4. Stepwise regression analysis was used to screen key forecast factors.

[0053] Table 4 Data on meteorological forecast factors for summer corn yield from 2009 to 2020

[0054]

[0055]

[0056]

[0057] 4. Introduction to Neural Network Algorithm

[0058] (1) Introduction to GRNN Neural Network

[0059] GRNN is structurally composed of four layers, namely input layer, pattern layer, summation layer and output layer, and has strong nonlinear mapping ability and learning speed.

[0060] The theoretical basis of GRNN network calculation is to assume that f(x, y) is the joint probability density function of random variables x and y. If the observed value of x is known to be X, then the regression of y relative to X, that is, the conditional mean, is:

[0061]

[0062] This formula is the predicted output of Y when the input is X.

[0063] The final network output can be expressed as:

[0064]

[0065] Among them, X i and Y i are the sample observations of the random variables x and y, m is the sample size, p is the dimension of the random variable x, and σ is the width coefficient of the Gaussian function.

[0066] This algorithm is suitable for intelligent models with fewer training samples, and the network finally converges to the optimal regression with more sample size. In recent years, the frequency of using GRNN neural network in meteorological element forecasting and analysis has increased. Even when the sample data is small, the forecast effect is very good, and the network can also handle unstable data.

[0067] (2) BP Neural Network

[0068] BP neural network, i.e., multilayer perceptron network error back propagation algorithm of nonlinear continuous transformation function, is composed of two processes: the forward propagation of signal and the back propagation of error.

[0069] The network uses an S-type transfer function By back-propagating the error function (t i is the expected output, O i is the calculated output of the network), and the weights and thresholds are continuously adjusted to minimize the error function E.

[0070] The application of BP neural network in the field of meteorology can prompt the parameters in the neural network to be continuously adjusted, autonomously adapt and learn at the same time, so as to achieve the most desired output, and the nonlinear mapping ability is considerable.

[0071] (3) RBF neural network

[0072] The RBF neural network is divided into three layers. The first layer is the input layer, which is composed of signal source nodes; the second layer is the hidden layer. The transformation function of the neurons in the hidden layer, namely the radial basis function, is a non-negative linear function that is radially symmetric and decays about the center point. This function is a local response function; the third layer is the output layer, which responds to the input pattern and adopts a linear optimization strategy, so the learning speed is faster.

[0073] The activation function of this neural network can be expressed as:

[0074]

[0075] In the fields of agricultural products and meteorology, RBF networks can approximate arbitrary nonlinear functions, can handle difficult-to-analyze regularities within the system, have good generalization capabilities, and have a fast learning convergence speed.

[0076] 5. Stepwise regression factor screening

[0077] (1)Winter wheat

[0078] A stepwise regression analysis was conducted on the meteorological factors and yield during the entire growth period (sowing-heading period) and different periods of winter wheat from 2009 to 2021. The results showed that the winter wheat yield was extremely significantly correlated with the accumulated temperature after returning to green (p<0.01), followed by precipitation during the growth period, the average temperature from jointing to heading, and the minimum temperature from jointing to heading, all of which reached a significant level (p<0.05). Figure 3 shown.

[0079] (2) Spring corn

[0080] A stepwise regression analysis was conducted on the meteorological elements and meteorological yield during the entire growth period (sowing-silking period) of spring corn from 2010 to 2020 and at different times. The results showed that the meteorological yield of spring corn was highly correlated with the accumulated temperature during the sowing-jointing period, the average temperature from sowing to emergence, the average temperature from tasseling to silking, and the extreme minimum temperature from emergence to jointing. Figure 4 shown.

[0081] (3) Summer corn

[0082] A stepwise regression analysis was conducted on the meteorological elements and meteorological yield during the entire growth period (sowing-silking period) of summer corn from 2010 to 2020 and at different times. The results showed that the meteorological yield of summer corn was highly correlated with the average temperature from jointing to tasseling, the average temperature from tasseling to silking, the precipitation from sowing to seedling emergence, and the precipitation from jointing to tasseling, all of which reached an extremely significant level (p<0.01). Figure 5 shown.

[0083] 6. Predicting crop yields

[0084] (1)Winter wheat

[0085] Based on the key meteorological factors and winter wheat yield data selected by stepwise regression analysis, MATLAB programming algorithm was used to build models and forecasts based on GRNN, BP and RBF neural network methods respectively. The winter wheat yield and precipitation during the growth period, accumulated temperature after greening, average temperature from jointing to heading, and minimum temperature from jointing to heading from 2009 to 2020 were used as training modeling factors. The fitting yield calculated by the fitting regression equation was compared with the actual winter wheat yield. Figure 6 As shown, the root mean square errors (RMSE) were 159.25, 0.27, and 127.58 kg / hectare, respectively, and the modeling passed the extremely significant test.

[0086] Based on the established neural network model, the precipitation during the growing period of winter wheat in 2021, the accumulated temperature after greening, the average temperature from jointing to heading, and the minimum temperature from jointing to heading are used as prediction factors to predict the yield results: 5493 kg / hectare for the GRNN method, 5203 kg / hectare for the BP method, and 5274 kg / hectare for the RBF method.

[0087] (2) Spring corn

[0088] Based on the key meteorological factors and spring corn yield data selected by stepwise regression analysis, MATLAB programming algorithm was used to model and forecast based on GRNN, BP and RBF neural network methods respectively. The meteorological yield of spring corn and the average temperature from jointing to tasseling, the average temperature from tasseling to silking, the precipitation from sowing to seedling emergence, and the precipitation from jointing to tasseling from 2010 to 2020 were used as training modeling factors. The simulated meteorological yield calculated by the neural network model was added to the year-on-year trend yield to obtain the year-on-year fitting yield. The fitting yield was compared with the actual spring corn yield. Figure 7 As shown, the root mean square errors (RMSE) were 241.74 kg / hectare, 45.45 kg / hectare, and 189.23 kg / hectare, respectively, and the modeling passed the extremely significant test.

[0089] Based on the established neural network model, the average temperature of jointing-tasking, the average temperature of tasking-silking, the precipitation from sowing to emergence, and the precipitation from jointing to tasking of spring corn in 2021 were used as prediction factors, and the yield results were predicted: 9730 kg / hectare for the GRNN method, 9652 kg / hectare for the BP method, and 9788 kg / hectare for the RBF method.

[0090] (3) Summer corn

[0091] Based on the key meteorological factors and summer corn yield data selected by stepwise regression analysis, MATLAB programming algorithm was used to model and forecast based on GRNN, BP and RBF neural network methods respectively. The meteorological yield of summer corn and the average temperature from jointing to tasseling, the average temperature from tasseling to silking, the precipitation from sowing to seedling emergence, and the precipitation from jointing to tasseling from 2010 to 2020 were used as training modeling factors. The simulated meteorological yield calculated by the neural network model was added to the year-on-year trend yield to obtain the year-on-year fitting yield. The fitting yield was compared with the actual summer corn yield. Figure 8 As shown, the root mean square error (RMSE) was 162.9 kg / hectare, 44.8 kg / hectare, and 130.7 kg / hectare, respectively, and the modeling passed the extremely significant test.

[0092] Based on the established neural network model, the average temperature of jointing-tasking, the average temperature of tasking-silking, the precipitation from sowing to emergence, and the precipitation from jointing to tasking of summer corn in 2021 were used as prediction factors, and the yield results were predicted: 9024 kg / hectare for the GRNN method, 9309 kg / hectare for the BP method, and 9042 kg / hectare for the RBF method.

[0093] (4) Analysis of prediction results using stepwise regression and neural network methods

[0094] Based on the selected key meteorological factor data affecting winter wheat yield and actual yield data, the winter wheat yield data was forecasted year by year through the stepwise regression method, GRNN, BP and RBF neural network methods, and the RMSE (root mean square error) of various forecasting methods was calculated. It was found that the forecast result of the BP neural network method was closest to the actual yield (RMSE was 11.65 kg / hectare), followed by the stepwise regression method, and the GRNN neural network method had the worst forecast result, as shown in Table 5.

[0095] Table 5 2009-2020 winter wheat annual yield forecast results (unit: kg / hectare)

[0096]

[0097] Based on the selected key meteorological factor data affecting spring corn yield and actual yield data, the spring corn yield data was predicted year by year through the stepwise regression method, GRNN, BP and RBF neural network methods, and the RMSE (root mean square error) of various forecasting methods was calculated. It was found that the forecast result of the BP neural network method was closest to the actual yield (RMSE was 45.45 kg / hectare), followed by the stepwise regression method, and the GRNN neural network method had the worst forecast result, as shown in Table 6.

[0098] Table 6 2010-2020 spring corn annual yield forecast results (unit: kg / hectare)

[0099]

[0100] Based on the selected key meteorological factor data affecting summer corn yield and actual yield data, the summer corn yield data was predicted year by year through the stepwise regression method, GRNN, BP and RBF neural network methods, and the RMSE (root mean square error) of various forecasting methods was calculated. It was found that the forecast result of the BP neural network method was closest to the actual yield (RMSE was 44.77 kg / hectare), followed by the stepwise regression method, and the GRNN neural network method had the worst forecast result, as shown in Table 7.

[0101] Table 7 2010-2020 summer corn yield forecast results (unit: kg / hectare)

[0102]

[0103] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A crop yield forecasting method, characterized in that: include: Collect meteorological observations and crop data; Analyze the meteorological observation data for relevant meteorological elements to determine the meteorological factors, and perform stepwise regression analysis on the meteorological factors to screen out the key meteorological factors; The crop growth period and yield were analyzed on the crop data, and the trend yield was calculated using one-dimensional linear regression fitting; Key meteorological factors and trend yields are input into the BP neural network to predict crop yields.

2. A crop yield forecasting method according to claim 1, characterized in that: The meteorological observation data include temperature, precipitation, and sunshine; The crop data include growth period, yield, and yield factors.

3. A crop yield forecasting method according to claim 1, characterized in that: The analysis of relevant meteorological elements includes: according to the impact of temperature, precipitation, sunshine hours and meteorological disasters on yield, principal component analysis and hierarchical analysis are carried out on the factors of temperature, precipitation, sunshine hours, meteorological disasters and yield to determine the meteorological factors that affect crop yield.

4. A crop yield forecasting method according to claim 1, characterized in that: The calculation of trend yield includes: performing a one-dimensional linear fitting on the results obtained from the crop growth period and yield analysis to obtain a fitting equation of yield and crop growth period, and calculating the trend yield through the fitting equation.

5. A crop yield forecasting method as claimed in claim 1, characterized in that: The BP neural network is: Select S-type transfer function Back propagation error function Among them, t i is the expected output, O i For the network's computational output, the weights and thresholds are continuously adjusted to minimize the error function E.

Images