Tea garden drought prediction method based on climate environment monitoring data

Abstract

The invention provides a tea garden drought prediction method based on climate environment monitoring data in the technical field of tea garden management. The tea garden drought prediction method comprises the following steps: S1, acquiring historical climate environment monitoring data of a tea garden and calculating monitoring indexes; s2, performing difference analysis and COX single-factor regression analysis on the monitoring indexes, and acquiring the monitoring indexes related to tea garden drought as modeling factors; step S3, screening each modeling factor; s4, performing COX multi-factor regression analysis on each modeling factor to construct a column graph model; s5, optimizing the column graph model to obtain a tea garden drought prediction model; s6, verifying the tea garden drought prediction model through the ROC curve and the calibration curve; s7, building a user interface for calling the tea garden drought prediction model based on Shiny and DynNOM; and S8, calling the tea garden drought prediction model through a user interface to perform tea garden drought prediction. The method has the advantage that the precision and efficiency of tea garden drought short-term prediction are greatly improved.

Description

Technical Field

[0001] The invention relates to the technical field of tea garden management, and in particular to a tea garden drought prediction method based on climate environment monitoring data. Background Art

[0002] Tea is a special economic crop, and drought has a very serious impact on its yield. With the rapid development of agricultural facilities, driven by the rapid development of cloud computing, the Internet of Things, big data, machine learning and other technologies, smart irrigation has become one of the important means to solve the drought problem. However, the current smart irrigation still has problems such as low precision management level, poor environmental control ability, and low intelligence. Especially in areas with diverse weather changes, rainfall often occurs after irrigation, which not only causes a waste of funds and water resources, but also easily damages the growth of tea trees. Therefore, predicting tea garden drought and providing guidance for agricultural production is of great significance to improving the yield and quality of tea.

[0003] With the rapid development of machine learning, many scholars have begun to work on developing better drought prediction models. Although traditional models have shown high accuracy in long-term drought prediction, they have problems such as large amount of calculation, complex calculation, and high difficulty of operation, resulting in low prediction efficiency. In addition, it is still impossible to accurately predict short-term and small-scale drought changes, which is extremely important for the growth of tea trees.

[0004] Therefore, how to provide a tea garden drought prediction method based on climate and environmental monitoring data to improve the accuracy and efficiency of short-term prediction of tea garden drought has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a tea garden drought prediction method based on climate and environmental monitoring data, so as to improve the accuracy and efficiency of short-term prediction of tea garden drought.

[0006] The present invention is implemented as follows: a tea garden drought prediction method based on climate environment monitoring data comprises the following steps:

[0007] Step S1, obtaining a large amount of historical climate and environmental monitoring data of the tea garden, expanding the samples of each of the historical climate and environmental monitoring data by Bootstrap, and calculating the monitoring indicators of each of the historical climate and environmental monitoring data;

[0008] Step S2, performing difference analysis and COX univariate regression analysis on each of the monitoring indicators, and obtaining the monitoring indicators related to tea garden drought as modeling factors;

[0009] Step S3, screening each of the modeling factors by using the LASSO regression algorithm;

[0010] Step S4, performing COX multivariate regression analysis on each of the modeling factors after screening to construct a nomogram model;

[0011] Step S5, optimizing the nomogram model by a 5-fold cross validation method to obtain a tea garden drought prediction model;

[0012] Step S6, verifying the tea garden drought prediction model through ROC curve and calibration curve;

[0013] Step S7: Building a user interface for calling the tea garden drought prediction model based on Shiny technology and DynNOM technology;

[0014] Step S8: Calling the tea garden drought prediction model through the user interface to perform tea garden drought prediction.

[0015] Further, in step S1, the historical climate environment monitoring data at least includes soil temperature, soil moisture, light intensity, light duration, air temperature, air humidity, carbon dioxide concentration and TBD radiation value;

[0016] The monitoring indicators at least include the average value per unit time, the average value per day, the maximum value per day, the minimum value per day, the floating value of daily changes, the average value per cycle, the starting difference value of cycle changes and the maximum difference value of cycle changes.

[0017] Further, in step S2, the difference analysis is based on Limma;

[0018] The formula of the COX univariate regression analysis is:

[0019] h(t)=h0(t)exp(α1x1+...+α j x j );

[0020] Among them, h(t) represents the risk coefficient of tea garden drought; h0(t) represents the intercept of the regression equation; α j represents the jth regression coefficient; x j Represents the jth monitoring indicator.

[0021] Furthermore, in step S3, the formula of the LASSO regression algorithm is:

[0022]

[0023] Among them, L represents the modeling factor after screening; m represents the number of samples; h θ represents the design matrix; x (i) represents the model parameter vector; y (i)represents the response vector; λ represents the regularization parameter; represents the L1 regularization term; k represents the number of parameters; w j Represents the sum of the absolute values ​​of all elements in the parameter vector.

[0024] Furthermore, the step S5 is specifically as follows:

[0025] Construct a data set based on each of the monitoring indicators, divide the data set into five parts on average, randomly select four of them as training sets and the other as a test set, and construct five different combinations of training sets and test sets;

[0026] The nomogram model is trained by the training sets of each combination respectively, the trained nomogram model is tested by the test set, and the nomogram model with the highest prediction accuracy is selected as the tea garden drought prediction model.

[0027] Furthermore, the step S7 is specifically as follows:

[0028] A framework is constructed based on Shiny technology, and a user interface for calling the tea garden drought prediction model to perform tea garden drought prediction is built in combination with DynNOM technology, and access rights of the user interface are set.

[0029] Furthermore, the step S8 is specifically as follows:

[0030] After the access authority is verified through the user interface, the user interface obtains the input real-time climate and environmental monitoring data, calculates the monitoring index of the real-time climate and environmental monitoring data, inputs it into the tea garden drought prediction model, and outputs the tea garden drought prediction result.

[0031] The advantages of the present invention are:

[0032] By obtaining a large amount of historical climate and environmental monitoring data of the tea garden, the samples of each historical climate and environmental monitoring data were expanded through Bootstrap, and the monitoring indicators of each historical climate and environmental monitoring data were calculated. Then, the difference analysis and COX univariate regression analysis of each monitoring indicator were performed, and the monitoring indicators related to tea garden drought were obtained as modeling factors. The modeling factors were screened by LASSO regression algorithm, and COX multivariate regression analysis was performed on the screened modeling factors to construct a nomogram model; then the nomogram model was optimized by 5-fold cross-validation method to obtain the tea garden drought prediction model, which was verified by ROC curve and calibration curve. Based on Shiny technology and DynNOM technology, a calibration curve was built. Use the user interface of the tea garden drought prediction model, and finally call the tea garden drought prediction model through the user interface to predict the tea garden drought; that is, through difference analysis and COX univariate regression analysis, select the factors related to tea garden drought from various monitoring indicators as modeling factors, and then further screen the modeling factors through the LASSO regression algorithm, effectively reduce the complexity of the tea garden drought prediction model, reduce the risk of overfitting, and improve the calculation speed; optimize the nomogram model through the 5-fold cross-validation method, select the nomogram model with the highest prediction accuracy as the tea garden drought prediction model, combined with the verification of the ROC curve and the calibration curve, to ensure the tea garden drought prediction model for the short-term prediction of tea garden drought, and ultimately greatly improve the accuracy and efficiency of short-term prediction of tea garden drought. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.

[0034] Figure 1 The present invention is a flow chart of a tea garden drought prediction method based on climate and environment monitoring data. DETAILED DESCRIPTION

[0035] The technical solution in the embodiment of the present application has the following overall idea: through difference analysis and COX univariate regression analysis, factors related to tea garden drought are screened from various monitoring indicators as modeling factors, and then the modeling factors are further screened through the LASSO regression algorithm, effectively reducing the complexity of the tea garden drought prediction model, reducing the risk of overfitting, and improving the calculation speed; the nomogram model is optimized through the 5-fold cross-validation method, and the nomogram model with the highest prediction accuracy is selected as the tea garden drought prediction model, combined with the verification of the ROC curve and the calibration curve, to ensure the tea garden drought prediction model for the short-term prediction of tea garden drought, thereby improving the accuracy and efficiency of short-term prediction of tea garden drought.

[0036] Please refer to Figure 1As shown, a preferred embodiment of a tea garden drought prediction method based on climate environment monitoring data of the present invention comprises the following steps:

[0037] Step S1, obtaining a large amount of historical climate and environmental monitoring data of the tea garden, and calculating the monitoring indicators of each of the historical climate and environmental monitoring data after sample expansion of each of the historical climate and environmental monitoring data through Bootstrap; in specific implementation, each of the monitoring indicators can be classified according to a cycle of 3 days, 5 days and 7 days;

[0038] Bootstrap is a resampling technique in statistics. The core idea is to sample with replacement from the original data to form a new "virtual" data set. Due to the randomness of sampling, most "virtual" data sets are different to a certain extent, which reduces the uncertainty caused by the size limitation of the data set and improves the reliability of model prediction.

[0039] Step S2, performing difference analysis and COX univariate regression analysis on each of the monitoring indicators, and obtaining the monitoring indicators related to tea garden drought as modeling factors;

[0040] COX univariate regression analysis is an analysis method based on the COX proportional hazard model (COX Proportional Hazards Model);

[0041] Step S3, screening each of the modeling factors by LASSO (Least Absolute Shrinkage and Selection Operator) regression algorithm;

[0042] For example, the following 11 modeling factors are selected: average light intensity (average value, five-day cycle), maximum temperature (average value, three-day cycle), average night temperature (average value, three-day cycle), maximum air humidity (variation difference, seven-day cycle), maximum air humidity (variation floating value, five-day cycle), minimum air humidity (variation floating value, five-day cycle), average night carbon dioxide concentration (variation floating value, seven-day cycle), average soil moisture (variation difference, seven-day cycle), daily average soil moisture difference (average value, seven-day cycle), daily average soil moisture difference (variation difference, three-day cycle), daily average soil moisture difference (variation floating value, seven-day cycle).

[0043] The LASSO regression algorithm is generally used for model feature selection and parameter sparsification. Its main feature is to add the L1 regularization term to the loss function to achieve sparse processing of model parameters. By constraining the parameters, the complexity of the model is reduced and the risk of overfitting is reduced.

[0044] Step S4, performing COX multivariate regression analysis on each of the modeling factors after screening to construct a nomogram model (NOMOGRAM), integrating each of the modeling factors, and converting the risk function and probability function of the nomogram model, so as to express the influence of the modeling factors;

[0045] As a visualization operation based on multi-factor regression analysis, the nomogram model can intuitively display the relationship between the variables. By drawing scale lines according to the proportion on the same plane, the prediction weights of each modeling factor can be visualized. The core idea is to assign points to each value of each modeling factor based on the predictive effect of each modeling factor on the drought change in the tea garden (determined by the regression coefficient), and finally realize the prediction of drought change in the tea garden through the functional transformation between the total score and the probability of drought change.

[0046] Step S5, optimizing the nomogram model by 5-fold cross validation to obtain a tea garden drought prediction model;

[0047] Step S6, verifying the tea garden drought prediction model through ROC (Receiver Operating Characteristic) curve and calibration curve; the calibration curve is a standard curve of climate environment monitoring data and prediction results;

[0048] Step S7: Building a user interface for calling the tea garden drought prediction model based on Shiny technology and DynNOM technology;

[0049] Step S8: Calling the tea garden drought prediction model through the user interface to perform tea garden drought prediction.

[0050] The invention proposes a LASSO-COX-NOMOGRAM machine learning algorithm to accurately predict short-term and small-scale drought changes in tea gardens. The invention performs climate expression analysis on drought changes of different degrees through Limma difference analysis, combines COX single factor regression analysis to jointly study the influence mechanism of climate and environmental changes on tea garden drought, screens modeling factors through LASSO regression algorithm, adopts 5-fold cross validation method and COX multi-factor regression analysis to establish a tea garden drought prediction model, and builds a visual prediction system through Shiny and DynNOM.

[0051] In step S1, the historical climate environment monitoring data at least includes soil temperature, soil moisture, light intensity, light duration, air temperature, air humidity, carbon dioxide concentration and TBD radiation value;

[0052] The monitoring indicators at least include the average value per unit time, the average value per day, the maximum value per day, the minimum value per day, the floating value of daily changes, the average value per cycle, the starting difference value of cycle changes and the maximum difference value of cycle changes.

[0053] The formulas for each of the monitoring indicators are as follows:

[0054]

[0055]

[0056] MPD1(t)=MAX{CV β ,CV β+1 ,...,CV t};

[0057] MPD2(t)=MIN{CV β ,CV β+1 ,...,CV t};

[0058] DV(t)=MPD1(t)-MPD2(t);

[0059]

[0060] VD(T)=D T -D1;

[0061] FR(T)=MAX{D1,D2,...,D T}-MIN{D1,D2,...,D T};

[0062] Among them, CV(h) represents the average value per unit time; h represents the length of the unit time; α represents the number of packet losses of the sensor group per unit time; V i represents the original value of the output of the i-th sensor; APD(t) represents the daily average value; t represents the end time of the calculation; β represents the start time of the calculation; CV j represents the jth CV(h); MPD1(t) represents the maximum value of a single day; MPD2(t) represents the minimum value of a single day; CV t represents the t-th CV(h); DV(t) represents the daily change floating value; AV(T) represents the cycle average; T represents the cycle length; D k represents the test value on the kth day; VD(T) represents the starting difference of the cycle change; DT It represents the test value on the Tth day; D1 represents the test value on the 1st day; FR(T) represents the maximum difference of the cycle change.

[0063] In step S2, the difference analysis is based on Limma, and the screening is performed with significant correlation factors P < 0.05 and logFC| ≥ 1 as the criteria;

[0064] The formula of the COX univariate regression analysis is:

[0065] h(t)=h0(t)exp(α1x1+...+α j x j );

[0066] Among them, h(t) represents the risk coefficient of tea garden drought; h0(t) represents the intercept of the regression equation; α j represents the jth regression coefficient; x j Represents the jth monitoring indicator.

[0067] In step S3, the formula of the LASSO regression algorithm is:

[0068]

[0069] Among them, L represents the modeling factor after screening; m represents the number of samples; h θ represents the design matrix; x (i) represents the model parameter vector; y (i) represents the response vector; λ represents the regularization parameter; represents the L1 regularization term; k represents the number of parameters; w j Represents the sum of the absolute values ​​of all elements in the parameter vector.

[0070] The sparsity constraint of features is implemented by introducing the L1 regularization term. The complexity is mainly controlled by λ. As λ increases, the penalty of LASSO will also increase. This process not only helps to eliminate features that contribute less to the model and reduce the dimension of the model, but also avoids the introduction of too much noise and redundant information, thereby improving the generalization performance of the model.

[0071] The step S5 is specifically as follows:

[0072] Construct a data set based on each of the monitoring indicators, divide the data set into five parts on average, randomly select four of them as training sets and the other as a test set, and construct five different combinations of training sets and test sets;

[0073] The nomogram model is trained by the training sets of each combination respectively, the trained nomogram model is tested by the test set, and the nomogram model with the highest prediction accuracy is selected as the tea garden drought prediction model.

[0074] The step S7 is specifically as follows:

[0075] A framework is constructed based on Shiny technology, and a user interface for calling the tea garden drought prediction model to perform tea garden drought prediction is built in combination with DynNOM technology, and access rights of the user interface are set.

[0076] The step S8 is specifically as follows:

[0077] After the access authority is verified through the user interface, the user interface obtains the input real-time climate and environmental monitoring data, calculates the monitoring index of the real-time climate and environmental monitoring data, inputs it into the tea garden drought prediction model, and outputs the tea garden drought prediction result. The real-time climate and environmental monitoring data and the historical climate and environmental monitoring data are collected by a sensor group including a humidity sensor, a temperature sensor, a light intensity sensor, a carbon dioxide sensor, and a TBQ total radiation sensor.

[0078] In summary, the advantages of the present invention are:

[0079] By obtaining a large amount of historical climate and environmental monitoring data of the tea garden, the samples of each historical climate and environmental monitoring data were expanded through Bootstrap, and the monitoring indicators of each historical climate and environmental monitoring data were calculated. Then, the difference analysis and COX univariate regression analysis of each monitoring indicator were performed, and the monitoring indicators related to tea garden drought were obtained as modeling factors. The modeling factors were screened by LASSO regression algorithm, and COX multivariate regression analysis was performed on the screened modeling factors to construct a nomogram model; then the nomogram model was optimized by 5-fold cross-validation method to obtain the tea garden drought prediction model, which was verified by ROC curve and calibration curve. Based on Shiny technology and DynNOM technology, a calibration curve was built. Use the user interface of the tea garden drought prediction model, and finally call the tea garden drought prediction model through the user interface to predict the tea garden drought; that is, through difference analysis and COX univariate regression analysis, select the factors related to tea garden drought from various monitoring indicators as modeling factors, and then further screen the modeling factors through the LASSO regression algorithm, effectively reduce the complexity of the tea garden drought prediction model, reduce the risk of overfitting, and improve the calculation speed; optimize the nomogram model through the 5-fold cross-validation method, select the nomogram model with the highest prediction accuracy as the tea garden drought prediction model, combined with the verification of the ROC curve and the calibration curve, to ensure the tea garden drought prediction model for the short-term prediction of tea garden drought, and ultimately greatly improve the accuracy and efficiency of short-term prediction of tea garden drought.

[0080] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A tea garden drought prediction method based on climate and environmental monitoring data, characterized in that: The steps include: Step S1, obtaining a large amount of historical climate and environmental monitoring data of the tea garden, expanding the samples of each of the historical climate and environmental monitoring data by Bootstrap, and calculating the monitoring indicators of each of the historical climate and environmental monitoring data; Step S2, performing difference analysis and COX univariate regression analysis on each of the monitoring indicators, and obtaining the monitoring indicators related to tea garden drought as modeling factors; Step S3, screening each of the modeling factors by using the LASSO regression algorithm; Step S4, performing COX multivariate regression analysis on each of the modeling factors after screening to construct a nomogram model; Step S5, optimizing the nomogram model by a 5-fold cross validation method to obtain a tea garden drought prediction model; Step S6, verifying the tea garden drought prediction model through ROC curve and calibration curve; Step S7: Building a user interface for calling the tea garden drought prediction model based on Shiny technology and DynNOM technology; Step S8: Calling the tea garden drought prediction model through the user interface to perform tea garden drought prediction.

2. A tea garden drought prediction method based on climate and environment monitoring data as claimed in claim 1, characterized in that: In step S1, the historical climate environment monitoring data at least includes soil temperature, soil moisture, light intensity, light duration, air temperature, air humidity, carbon dioxide concentration and TBD radiation value; The monitoring indicators at least include the average value per unit time, the average value per day, the maximum value per day, the minimum value per day, the floating value of daily changes, the average value per cycle, the starting difference value of cycle changes and the maximum difference value of cycle changes.

3. A tea garden drought prediction method based on climate environment monitoring data as claimed in claim 1, characterized in that: In the step S2, the difference analysis is based on Limma; The formula of the COX univariate regression analysis is: h(t)=h0(t)exp(α1x1+...+α j x j ); Among them, h(t) represents the risk coefficient of tea garden drought; h0(t) represents the intercept of the regression equation; α j represents the jth regression coefficient; x j Represents the jth monitoring indicator.

4. A tea garden drought prediction method based on climate and environment monitoring data as claimed in claim 1, characterized in that: In step S3, the formula of the LASSO regression algorithm is: Among them, L represents the modeling factor after screening; m represents the number of samples; h θ represents the design matrix; x (i) represents the model parameter vector; y (i) represents the response vector; λ represents the regularization parameter; represents the L1 regularization term; k represents the number of parameters; w j Represents the sum of the absolute values ​​of all elements in the parameter vector.

5. A tea garden drought prediction method based on climate environment monitoring data as claimed in claim 1, characterized in that: The step S5 is specifically as follows: Construct a data set based on each of the monitoring indicators, divide the data set into five parts on average, randomly select four of them as training sets and the other as a test set, and construct five different combinations of training sets and test sets; The nomogram model is trained by the training sets of each combination respectively, the trained nomogram model is tested by the test set, and the nomogram model with the highest prediction accuracy is selected as the tea garden drought prediction model.

6. A tea garden drought prediction method based on climate and environment monitoring data as claimed in claim 1, characterized in that: The step S7 is specifically as follows: A framework is constructed based on Shiny technology, and a user interface for calling the tea garden drought prediction model to perform tea garden drought prediction is built in combination with DynNOM technology, and access rights of the user interface are set.

7. A tea garden drought prediction method based on climate and environment monitoring data as claimed in claim 1, characterized in that: The step S8 is specifically as follows: After the access authority is verified through the user interface, the user interface obtains the input real-time climate and environmental monitoring data, calculates the monitoring index of the real-time climate and environmental monitoring data, inputs it into the tea garden drought prediction model, and outputs the tea garden drought prediction result.

Images