An intelligent prediction method for corn yield

Abstract

The application discloses an intelligent prediction method for corn yield, a multilayer feedforward neural network model is obtained through training, the multilayer feedforward neural network model is trained through a genetic method, and the obtained prediction model has higher prediction accuracy when predicting the corn yield; the method can obtain more accurate prediction data of the corn growth state by assimilating vegetation net primary productivity data which has a greater influence on the corn yield, the network model is updated based on the prediction data, the model is better adapted to the current corn growth data, and the accuracy of the corn yield prediction is improved; the method can more scientifically and reasonably predict the corn yield by improving the prediction operation speed of the multilayer feedforward neural network model, so that the corn production and management measures can be accurately adjusted in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of corn yield prediction, and particularly relates to an intelligent prediction method for corn yield. BACKGROUND

[0002] Corn is widely concerned due to its rich nutrition, wide application and large market potential, at present, the corn planting area in China is gradually expanded, and the efficient prediction of corn yield is of great significance for making precise management decisions during its growth period. The WOFOST prediction model is a dynamic interpretation model, which is driven by daily meteorological data to simulate the growth process of crops from sowing to maturity, including the simulation of crop basic physiological processes such as photosynthesis, respiration, transpiration and dry matter distribution. Although the model has strong adaptability, the data and operation process used by the model have large operation amount, and the final yield value cannot be quickly and accurately obtained when predicting the corn yield, and the running flexibility is poor. The existing BP (backpropagation) multilayer feedforward neural network model can quickly obtain the prediction result, but has the problems of low test accuracy and poor robustness in prediction.

[0003] Therefore, it is necessary to develop an intelligent prediction method for corn yield aiming at the above-mentioned defects. SUMMARY

[0004] The purpose of the present application is to provide an intelligent prediction method for corn yield, which can solve the problem that the existing prediction model cannot consider accuracy and reaction speed, and has poor flexibility.

[0005] To solve the above technical problems, the technical scheme adopted by the present application is as follows:

[0006] The intelligent prediction method for corn yield comprises the following steps:

[0007] Step 1, collecting remote sensing observation data and growth related data of corn planting state in the to-be-tested area, and processing the remote sensing observation data by using the CASA model to obtain vegetation net primary productivity data;

[0008] Step 2, selecting corn growth related data from the database, and based on the corresponding vegetation net primary productivity data, using the ensemble Kalman filter assimilation method to assimilate the WOFOST model, and using the assimilated WOFOST model to obtain the prediction data of multiple groups of growth related data after the current state of the corn in the to-be-tested area;

[0009] Step 3, selecting corn growth-related and corresponding corn yield data from the database, training the multi-layer feedforward neural network model based on the genetic method, inputting the multiple sets of prediction data obtained by running the assimilated WOFOST model into the trained multi-layer feedforward neural network model, and obtaining the primary corn yield prediction result;

[0010] Step 4, combining the growth-related data of the previous stage with the primary corn yield prediction result of the adjacent next stage to obtain an updated data set, and updating and training the multi-layer feedforward neural network model based on the genetic method;

[0011] Step 5, using the updated neural network model to obtain the corn yield prediction result corresponding to the corn in the test area.

[0012] Further, the specific steps of the WOFOST model assimilation in step 2 include:

[0013] Step 2.1, obtaining the relevant parameter values of the WOFOST model using the existing growth data in the database, and creating the WOFOST model based on the relevant parameters;

[0014] Step 2.2, using the WOFOST model to obtain further prediction values of corn growth data in the test area;

[0015] Step 2.3, using the ensemble Kalman filter assimilation method to assimilate the obtained net primary productivity data into the further prediction values to obtain the assimilated prediction data.

[0016] Further, the training steps of the multi-layer feedforward neural network model in step 3 include:

[0017] Step 3.1, obtaining corn growth-related and corresponding corn yield data in the database, preprocessing the obtained data, and the corn growth-related data including meteorological data, field environment data, plant characteristics and fertilization information;

[0018] Step 3.2, generating an initialization population based on the preprocessed data and calculating individual fitness values;

[0019] Step 3.3, based on the genetic method, the generated population is iteratively selected based on the genetic method;

[0020] Step 3.4, obtaining the optimal position that meets the condition after iteration and calculating the optimal weight threshold value;

[0021] Step 3.5, inputting the obtained optimal solution into the multi-layer feedforward neural network model for training and prediction analysis, and obtaining the trained multi-layer feedforward neural network model.

[0022] Further, the fitness function F of the individual fitness value in step 3.2 is represented as:

[0023] Wherein, f i is the expected output value, f(x i ) is the actual output value, i is the input layer corresponding code, and n is the number of input layers.

[0024] Further, the specific steps of step 3.3 include:

[0025] Step 3.3.1, select excellent samples from the original population by selection operation, and the probability function of individual selection in the selection operation is represented as Wherein, F i is the population fitness value of individual i, and N is the population quantity.

[0026] Step 3.3.2, generate new individual samples with strong fitness from excellent samples by crossover operation;

[0027] Step 3.3.3, generate variation individuals from new individual samples by mutation operation to obtain a new population.

[0028] Further, the specific steps of step 3.4 include obtaining the fitness value of the individual in the newly generated population, repeating the optimization iteration until the fitness value meets the requirement or the evolution times reach the maximum value, and obtaining the weight and threshold of the multi-layer feedforward neural network model optimized by the genetic method.

[0029] Further, the function of the assimilated prediction data in step 2.3 is represented as: X t = X t，t-1 + P t，t-1 H T (HP t，t-1 H T + R T ) -1 (Y t -HX t，t-1 ), wherein X t represents the assimilated prediction data at time t, Y t represents the vegetation net primary productivity data at time t, X t，t-1 represents the further prediction value at time t, P t，t-1 represents the covariance of further prediction, R t is the mean of observation equation error, and H represents the observation operator.

[0030] Further, the function of the further prediction value in step 2.2 is represented as X t，t-1 =f(X t-1 , U t , θ), wherein Xt-1 represents the estimated value at time t-1, X t，t-1 represents X t-1 represents the further predicted value at time t obtained after the prediction by the WOFOST model, U t is the inputted growth-related data for running the WOFOST model; and θ is the relevant parameter of the WOFOST model.

[0031] Further, the function of the crossover operation in step 3.3.2 is represented as a kj = a kj (1-b) + a ij wherein a kj is two individuals a k and a i in the excellent sample, b is a random number.

[0032] Further, the function of the mutation operation in step 3.3.3 is represented as:

[0033] wherein a max , a min are the boundary conditions of a ij , r2 is a random number, g is the iteration number, G max is the maximum evolution number, a ij is the individual a i in the new individual sample which occurs mutation at j and forms a new individual.

[0034] Compared with the prior art, the present application has the beneficial technical effects that:

[0035] The intelligent prediction method for corn yield disclosed by the present application trains a multi-layer feedforward neural network model, trains the multi-layer feedforward neural network model by using a genetic method, and overcomes the blindness and uncertainty of parameter selection of a traditional neural network model when predicting the corn yield, so that the prediction model has higher prediction accuracy. The method can obtain more accurate prediction data of the corn growth state by assimilating the vegetation net primary productivity data which has a great influence on the corn yield, and can make the model better adapt to the current corn growth data by updating the network model based on the prediction data, thereby further improving the accuracy of the corn yield prediction. The method can more scientifically and reasonably predict the corn yield by improving the prediction operation speed of the multi-layer feedforward neural network model, so that the corn production and management measures can be accurately adjusted in time. BRIEF DESCRIPTION OF DRAWINGS

[0036] The present application will be further described below in combination with the drawings.

[0037] Figure 1 Process block diagram of the intelligent prediction method for corn yield of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the protection scope of the present application.

[0039] Please refer to Figure 1 , a specific embodiment of the intelligent prediction method for corn yield provided by the present application will be described. The intelligent prediction method for corn yield comprises:

[0040] S1, collecting remote sensing observation data and growth related data of corn planting state in a to-be-measured area, and processing the remote sensing observation data by using a CASA (Carnegie-Ames-Stanford approach) model to obtain vegetation net primary productivity data;

[0041] S2, selecting corn growth related data from a database, and based on the corresponding vegetation net primary productivity data, using a set Kalman filter assimilation method to assimilate a WOFOST model, and using the assimilated WOFOST model to obtain prediction data of multiple sets of growth related data after the current state of the corn in the to-be-measured area;

[0042] The related parameter values of the WOFOST model are determined by using the historical corn key growth period observation data in the database, including effective accumulated temperature from sowing to emergence, effective accumulated temperature from emergence to flowering, and effective accumulated temperature from flowering to maturity. The set Kalman filter assimilation method can be used for data assimilation of a nonlinear system, and the calculation amount is small. The specific steps of the WOFOST model assimilation in S2 include:

[0043] S2.1, obtaining the related parameter values of the WOFOST model by using the existing growth data in the database, and creating the WOFOST model based on the related parameters;

[0044] S2.2, obtaining further prediction values of the corn growth data in the to-be-measured area by using the WOFOST model;

[0045] The function expression of the further prediction values in S2.2 is X t，t-1 =f(X t-1 , U t , θ), wherein X t-1 represents the estimated value at t-1 time, Xt，t-1 X t-1 Further prediction value of t time obtained after WOFOST model prediction, U t is inputted growth related data of running WOFOST model; θ is related parameter of WOFOST model.

[0046] S2.3, adopt set Kalman filter assimilation method, assimilate obtained vegetation net primary productivity data into further prediction value, obtain assimilated prediction data.

[0047] Function expression of assimilated prediction data in S2.3 is as follows:

[0048] X t = X t，t-1 + P t，t-1 H T (HP t，t-1 H T + R T ) -1 (Y t -HX t，t-1 ), P t，t-1 = (X t，t-1 -X t )(X t，t-1 -X t ) T >Wherein, X t indicate assimilated prediction data of t time, Y t indicate vegetation net primary productivity data of t time, X t，t-1 indicate further prediction value of t time, P t，t-1 indicate covariance of further prediction, R t it is mean of observation equation error, H indicates observation operator.

[0049] S3, select corn growth related and corresponding corn yield data from database, train multilayer feedforward neural network model based on genetic method, input multiple groups of prediction data obtained by running assimilated WOFOST model to trained multilayer feedforward neural network model, obtain primary corn yield prediction result;

[0050] Training steps of multilayer feedforward neural network model in S3 include:

[0051] S3.1, obtain corn growth related and corresponding corn yield data in database, pretreat obtained corn growth related data, and corn growth related data includes meteorological data, field environment data, plant characteristics and fertilization information;

[0052] Before building the model in S3.1, the topology of the neural network model needs to be determined. The meteorological data in S3.1 includes atmospheric humidity, atmospheric temperature, and rainfall. The database consists of field IoT data or corn growth data collected in the field. Corn yield shows a significant correlation with monthly minimum soil temperature, monthly average soil temperature, monthly maximum atmospheric temperature, monthly average atmospheric humidity, monthly maximum temperature, monthly average soil moisture content, monthly average atmospheric temperature, monthly rainfall, and monthly average atmospheric humidity. The model is trained using these four parameters as the primary data. A multi-layer feedforward neural network model with multiple input nodes and one output node is constructed using multiple selected indicators as input and corn yield as output. This model is then trained and used for prediction.

[0053] S3.2 Based on the preprocessed data, set the population size to generate an initial population and calculate the individual fitness value. The population mainly includes the weights and thresholds of the neural network model.

[0054] The fitness function F for individual fitness values ​​in S3.2 is expressed as:

[0055] Among them, f i f(x) is the expected output value. i ) represents the actual output value, i represents the corresponding code of the input layer, and n represents the number of input layers.

[0056] S3.3. Based on genetic methods, the generated population is subjected to optimal iteration based on genetic methods, including population selection, population crossover, and population variation;

[0057] The specific steps in S3.3 include:

[0058] S3.3.1. Selection operations are used to select superior samples from the original population. Selection operation refers to selecting superior samples from the original population with a certain probability, and then generating the next generation of sample data through reproduction. The probability function of individual selection in the selection operation is expressed as: Among them, F i Let be the population fitness value of individual i, and N be the population size;

[0059] S3.3.2. Generate new, highly adaptable individual samples from excellent samples through crossover operations;

[0060] Crossover operation refers to randomly selecting two samples from the population and generating new, highly adaptable individuals through exchange and combination. An example of two individuals a... k a i The crossover operation is performed at point j. The function for the crossover operation in S3.3.2 is represented as a. kj =akj (1-b)+a ij , a ij = a ij (1-b)+a kj , where a kj is a new individual obtained by crossover operation of two individuals a k and a i in the new individual sample, and b is a random number between 0 and 1.

[0061] S3.3.3, generating a mutated individual from the new individual sample through mutation operation to obtain a new population.

[0062] The population diversity can be maintained through population mutation, and the mutation operation is to randomly select an individual from the population, select a part of the individual for mutation operation, and thus generate a better individual. The function of the mutation operation in S3.3.3 is represented as: where a max , a min are boundary conditions of a ij , r2 is a random number, g is the number of iterations, G max is the maximum number of evolutions, a ij is a new individual formed by mutation of a i in the new individual sample at j .

[0063] S3.4, obtaining the optimal position satisfying the condition after iteration and calculating the optimal weight threshold value;

[0064] The specific steps of S3.4 include obtaining the fitness value of the individual in the newly generated population, repeatedly performing optimal iteration until the fitness value meets the requirement or the number of evolutions reaches the maximum value, and obtaining the weight and threshold value of the multilayer feedforward neural network model optimized by the genetic method.

[0065] S3.5, inputting the obtained optimal solution into the multilayer feedforward neural network model for training and prediction analysis to obtain the trained multilayer feedforward neural network model.

[0066] The multilayer feedforward neural network model optimized by the genetic algorithm for the number of hidden layer nodes and the learning rate of the multilayer feedforward neural network model overcomes the blindness and uncertainty of parameter selection of the traditional neural network model, and effectively improves the prediction accuracy of the model.

[0067] S4, combining the growth-related data of the previous stage with the primary corn yield prediction results of the adjacent next stage to obtain an updated data set, and updating and training the multilayer feedforward neural network model based on the genetic method;

[0068] Genetic method is the most widely used adaptive heuristic search algorithm for solving optimization problems.

[0069] S5, the collected growth-related data is used to obtain the corn yield prediction result of the to-be-tested area of corn by using the updated neural network model.

[0070] Compared with the prior art, the intelligent prediction method for corn yield trains a multi-layer feedforward neural network model, trains the multi-layer feedforward neural network model by using a genetic method, and the obtained prediction model overcomes the blindness and uncertainty of parameter selection of a traditional neural network model when predicting corn yield, and has higher prediction accuracy; the method can obtain more accurate prediction data of the corn growth state by assimilating the vegetation net primary productivity data which has a greater impact on the corn yield, and can make the model better adapt to the current corn growth data by updating the network model based on the prediction data, thereby further improving the accuracy of corn yield prediction; the method improves the prediction operation speed by using the multi-layer feedforward neural network model, and has better flexibility, so that the corn yield can be more scientifically and reasonably predicted, thereby facilitating accurate and timely adjustment of corn production and management measures, and solving the problem that the existing prediction model cannot balance accuracy and reaction speed and has poor flexibility.

[0071] The above-described embodiments are only used to describe the preferred modes of the present application, and do not limit the scope of the present application, and various modifications and improvements to the technical solutions of the present application made by those skilled in the art without departing from the design spirit of the present application shall fall within the protection scope of the present application.

Claims

1. A method for intelligent prediction of corn yield characterized by: The application relates to a method for predicting the yield of corn in a test area. The method comprises the following steps: Step 1: collecting remote sensing observation data and growth-related data of corn planting state in a test area, and processing the remote sensing observation data by using a CASA model to obtain vegetation net primary productivity data; Step 2: selecting corn growth-related data from a database, and assimilating a WOFOST model based on corresponding vegetation net primary productivity data, so as to obtain prediction data of multiple groups of growth-related data after the current state of corn in the test area by using the assimilated WOFOST model and the collected current vegetation net primary productivity data and growth-related data; Step 3: selecting corn growth-related data and corresponding corn yield data from the database, training a multilayer feedforward neural network model based on a genetic method, inputting the multiple groups of prediction data obtained by running the assimilated WOFOST model into the trained multilayer feedforward neural network model, and obtaining a primary corn yield prediction result; The training step of the multilayer feedforward neural network model in step 3 comprises the following steps: Step 3.1: obtaining corn growth-related data and corresponding corn yield data in the database, and preprocessing the obtained data, wherein the corn growth-related data comprises meteorological data, field environment data, plant characteristics and fertilization information; The fitness function F of the individual fitness value in step 3.2 is expressed as: Wherein, f i is the expected output value, f(x i ) is the actual output value, i is the input layer corresponding code, and n is the number of input layers. Step 3.2: generating an initial population based on the preprocessed data and calculating individual fitness values; Step 3.3: iteratively selecting excellent samples from the generated population based on a genetic method; Step 3.3.1, selecting excellent samples from the original population by a selection operation, in which a probability function of individual selection is expressed as where F i is the population fitness value of individual i, and N is the population number; The specific steps of step 3.3 comprise the following steps: The function of the cross operation in step 3.3.2 is represented as a kj = a kj (1-b) + a ij , wherein a kj is a new individual obtained by cross operation of two individuals a k and a i in the excellent sample at j, and b is a random number Step 3.3.2: generating new individual samples with strong adaptability from the excellent samples through crossover operation; Step 3.3.3: generating variation individuals from the new individual samples through mutation operation, and obtaining a new population; wherein a max , a min is the boundary condition of a ij , r2 is a random number, g is the number of iterations, G max is the maximum number of evolution, a ij is the individual a i at j is mutated to form a new individual; The function of the mutation operation in step 3.3.3 is represented as follows: Step 3.4: obtaining an optimal position meeting a condition after iteration and calculating an optimal weight threshold value; Step 3.4 specifically comprises the following steps: obtaining the fitness values of individuals in the newly generated population, repeatedly performing the iteration of selecting excellent samples, until the fitness values meet the requirements or the evolution times reach a maximum value, and obtaining the weight and threshold value of the multilayer feedforward neural network model optimized by the genetic method; Step 3.5: inputting the obtained optimal solution into the multilayer feedforward neural network model to perform training and prediction analysis, and obtaining the trained multilayer feedforward neural network model; Step 4: combining the growth-related data of the previous stage and the primary corn yield prediction result of the adjacent next stage to obtain an updated data set, and updating and training the multilayer feedforward neural network model based on the genetic method; 2. The method for intelligent prediction of corn yield as claimed in claim 1 wherein: Step 5: obtaining the corresponding corn yield prediction result of corn in the test area by using the updated neural network model. The specific steps of the WOFOST model assimilation in step 2 comprise the following steps: Step 2.1: obtaining the relevant parameter values of the WOFOST model by using existing growth data in the database, and creating the WOFOST model based on the relevant parameters; Step 2.2: obtaining further prediction values of the corn growth data in the test area by using the WOFOST model; Step 2.

3. The obtained vegetation net primary productivity data is assimilated into the further predicted values by using the ensemble Kalman filter assimilation method to obtain the assimilated predicted data.

3. The method for intelligent prediction of corn yield as claimed in claim 2 wherein: The function of the assimilated predicted data in step 2.3 is represented as: X t = X t，t-1 + P t，t-1 H T (HP t，t-1 H T + R T ) -1 (Y t - HX t，t-1 ), wherein X t represents the assimilated predicted data at time t, Y t represents the vegetation net primary productivity data at time t, X t，t-1 represents a further predicted value at time t, P t，t-1 represents a covariance of the further prediction, R T is a mean of observation equation error, and H represents an observation operator.

4. The method for intelligent prediction of corn yield as claimed in claim 2, wherein: The function representing the further predicted value in step 2.2 is X. t，t-1 =f(X) t-1 U t ,θ), where X t-1 X represents the estimated value at time t-1. t，t-1 X represents t-1 The further predicted value U at time t obtained after the WOFOST model prediction t θ represents the growth-related data for running the WOFOST model; θ represents the relevant parameters of the WOFOST model.

Images