Agricultural mulching film degradation period prediction method and device based on RF-meta model

Abstract

This invention discloses a method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model, comprising the following steps: collecting raw datasets of biodegradable mulch film degradation using meta-analysis; and analyzing the influence of different factors in the raw dataset on the degree of degradation using meta-analysis. M Sort; Filter Q M The factors meeting the requirements are retained in the database. Missing values ​​in the original dataset are imputed using the Random Forest algorithm to obtain a complete database. This complete database is then divided into a modeling set and a validation set to establish an initial RF-Meta model. The initial RF-Meta model is trained, and its hyperparameters are optimized to obtain an optimized RF-Meta model. The initial experimental physicochemical data of the target biodegradable mulch film's degradation are imported into the optimized RF-Meta model to obtain a prediction curve for the entire degradation cycle. This invention's proposed model is the first to provide specific numerical predictions of the degradation degree of BDM during its degradation process and has broad applicability.

Description

Technical Field

[0001] This invention relates to the field of agricultural informatization, and in particular to a method and equipment for predicting the degradation cycle of biodegradable mulch film buried in soil by combining random forest algorithm with meta-analysis. Background Technology

[0002] Plastic mulch is an important agricultural production material. Its use effectively suppresses weed growth, increases crop yields, maintains soil moisture, reduces water evaporation, and conserves water resources. Furthermore, it can raise soil temperature, promote crop growth, reduce soil erosion, maintain soil fertility, thereby extending crop lifespan, reducing pest and disease risks, decreasing pesticide and fertilizer use, and improving grain quality. However, due to ultraviolet radiation and external forces, plastic mulch is prone to breakage, forming plastic fragments that cause serious plastic pollution in farmland, posing a potential threat to ecosystems and human health. Biodegradable plastics are plastics that can be degraded and utilized by microorganisms, ultimately transforming into microbial biomass, carbon dioxide, and water. Therefore, replacing PE mulch with biodegradable plastic film (BDM) is one of the main strategies for preventing plastic pollution in agriculture. However, the decomposition cycle of BDM in the environment is currently unclear, mainly because BDM degradation is a complex process highly dependent on material composition and natural conditions; different formulations of BDM often exhibit vastly different degradation effects in different regions.

[0003] Currently, research on the degradation cycle of specific BDM mulch film products mainly relies on experience to infer the degradation cycle after 1-2 years of use in local farmland. While there are reports on the effects of material factors, soil factors, and regional factors on BDM degradation, these studies primarily rely on experimental data and can only provide a general trend of degradation. The research methods are relatively limited, and the small number of control groups makes it difficult to measure the influence of relevant factors on degradation and to predict the degradation cycle. Existing research has established a simple fitting model for the degradation cycle of mulch film under ultraviolet (UV) irradiation to assess the photodegradation cycle. However, the shortcomings of this research are: (1) UV is not the only relevant variable affecting BDM; (2) The degradation process of BDM is complex, and material factors, soil factors, climate factors, and human factors often exhibit nonlinear interactions. Analyzing a small number of factors cannot fully analyze this interaction characteristic, making it impossible to predict the specific degradation process; (3) Current research cannot solve the regional characteristics of mulch film degradation. Therefore, it is urgent to develop a nonlinear BDM degradation prediction model with richer variables and wider applicability for mulch film usage areas. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide an RF-Meta model that combines Meta-analysis and Random Forest algorithm to predict the degradation cycle of agricultural mulch film, thereby achieving accurate estimation of the degradation cycle and improving the universality of the model results.

[0005] Technical solution: A method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model, comprising the following steps:

[0006] (1) Collect the original dataset of biodegradable mulch film degradation by meta-analysis, establish an original database based on the original dataset, and verify the reliability of the original data by meta-analysis.

[0007] (2) The impact of different factors on the degree of degradation in the original dataset through meta-analysis Q M Sort;

[0008] (3) Based on the meta-analysis results, Q was selected. M Factors with a value greater than 1 are left in the original database. The random forest algorithm is used to fill in the missing values ​​in the original database to obtain a complete database.

[0009] (4) Divide the complete database into a modeling set and a validation set, and establish an initial RF-Meta model;

[0010] (5) Train the initial RF-Meta model and optimize the hyperparameters to obtain the optimized RF-Meta model;

[0011] (6) The initial experimental physicochemical data of the degradation of the target biodegradable mulch film were imported into the optimized RF-Meta model to obtain the prediction curve of the entire degradation cycle.

[0012] Furthermore, the specific steps for constructing the original database in step (1) are as follows:

[0013] (1.1) Develop search strategies and inclusion / exclusion criteria based on the PICOS guidelines, including:

[0014] P: The research subject is determined to be biodegradable mulch film in agricultural production scenarios;

[0015] I: Determine intervention measures and select the range of components that constitute the basic materials of biodegradable mulch film; and screen different degradation methods, degradation environments, and pretreatment before degradation.

[0016] C: The control measure was determined to be the characteristics of the mulch film before and after degradation;

[0017] O: The result is determined to be guided by the mulch film degradation rate parameter;

[0018] S: The research design is limited to laboratory or field trials, excluding reviews and opinion articles;

[0019] (1.2) Select an article database, establish search terms according to the criteria in (1.1), and manually further screen articles that meet the research conditions and can be repeated;

[0020] (1.3) The composition of the feature data set G1 to be collected in each article is as follows:

[0021] G1 = {A1, A2, ..., A} a C1,C2,...,C e SP1, SP2, ..., SP p ;SB1,SB2...,SB b M1, M2..., M m E1, E2..., E e}; where A a C represents item a in the basic information of the article to be collected. e SP represents the e-th item in the climate conditions to be collected. p SB represents the p-th term in the soil physical conditions to be collected. b M represents the bth item in the biological and chemical conditions of the soil to be collected. m E represents the m-th term in the properties of the mulch film material to be collected. e This represents the e-th item in the experimental conditions to be collected;

[0022] (1.4) The composition of the degradation result dataset G2 to be collected in each article is as follows:

[0023] G2={E m E v C m C v}; where E m E represents the mean of the experimental group results. v C represents the variance of the experimental group results. m C represents the mean of the control group results. v The variance of the control group results is represented by the following formulas for the mean and standard deviation:

[0024]

[0025]

[0026] in, is the mean of the degradation rate, and s is the standard deviation of the degradation rate; x i It is the degradation rate of the i-th group, where i belongs to 1, 2, 3...n.

[0027] Furthermore, the verification of the reliability of the original data through meta-analysis includes:

[0028] (1.5) The publication bias of the collected articles was examined by observing the relationship between the degradation rate and its variance, i.e., the data source showed obvious bias; the degradation rate of the biodegradable mulch film obtained from each set of experimental data was ranked according to the degradation rate, and the location number u of each data point was obtained. i Then, sort each group of data according to the size of its experimental variance, and obtain the position number v after sorting. j Then, the correlation coefficient D between the degradation rate and its experimental variance, and the probability P of the hypothesis represented by D, were calculated:

[0029]

[0030] P = 2 × (1 - cdf(D))

[0031] Where n is the total amount of data in the database, and C n D represents the number of concordant pairs in pairwise comparisons. n The number of disharmony pairs in pairwise comparisons; if the u of the two variables is... i and v j If both variables increase or decrease simultaneously, they form a concordant pair, where u represents the sum of the two variables. i and v j If one variable increases while the other decreases, they are a disharmony pair; cdf(D) is the cumulative distribution function value of D under the standard normal distribution; by referring to the D value and p value, we can know whether there is publication bias. If the absolute value of D is greater than 1.96 and the p value is less than 0.05, the data is considered to have a significant bias, and data sources with significant bias need to be excluded; if there is no significant bias, we can continue to the next step.

[0032] (1.6) When publication bias is detected, first use R language to plot a funnel plot with degradation rate as the x-axis and experimental variance as the y-axis. The position vector of each data point in the plot is d. i =(x i s i );by If the numerical vector d is the axis, then... t =(x t s t If there is clearly no data near the corresponding point on the other side of the axis, resulting in an asymmetrical funnel plot, then further determine whether the data vector corresponds to a well-designed experimental environment that meets the requirements and whose journal has an impact factor greater than 6. If so, add the vector to the database. To eliminate bias; if the data vector corresponds to an experiment with special experimental conditions or low reliability, delete the data in the database; after filtering out the data that cause asymmetry in the funnel plot, repeat (1.6) to re-check publication bias;

[0033] (1.7) Assess whether there are general patterns in the data in the database; the significance level of the data is calculated as follows:

[0034]

[0035] Data significance propensity level F n When F is greater than zero, the data in the database is considered to have a significant general pattern. n If the value is less than 0, then we need to go back to step (1.1) to add intervention measures to the selection of articles again, so as to ensure that the data will have significant regularity.

[0036] Furthermore, step (2) specifically involves:

[0037] (2.1) Scan the degradation rate data and variance data to check for missing or null values. Once a null value is found, record the location information of the null value, including the row m and column index n where the null value is located, and fill it using the nearest neighbor linear interpolation method;

[0038] (2.2) Establish a multivariate random effects model for the database in the meta-analysis; the basic random effects component of the model arises from the different sources from which the data were extracted; the formula for the multivariate random effects model is:

[0039] yi=Xiβ+ui+∈i

[0040] Where: yi is the effect size of the i-th study, Xi is the covariate / predictor variable associated with the i-th study, β is the regression coefficient of the model, ui is the random effect of the i-th study, representing the heterogeneity among studies, and ∈i is the error term of the model;

[0041] (2.3) The weight values ​​of each feature's influence on the result are calculated using the following formula:

[0042]

[0043]

[0044] Qm is a weight value used to measure the influence of each feature on the result, j represents the Qm value for the j-th feature, and k represents the number of data points in the database containing that feature. It is the weighted sum of the j-th feature factors, x ·j It represents the degradation rate of the j-th characteristic factor.

[0045] Furthermore, the characteristic is that step (3) specifically comprises:

[0046] 1) Initial state: The original database established using the Meta-analysis method contains some empty data values. Let A be the j-th feature data of the i-th data in the database. ij ;

[0047] 2) Random Forest Architecture: The data is organized according to the different feature types (A, C, SP, SB, M, E) to be collected in step (1.3). Each feature category has different feature data. For each feature category, an independent decision tree model is built for each feature. At the same time, different feature categories are randomly sampled. For each feature, d independent decision tree models are built to simulate the mutual influence of features under different feature classifications. Finally, a total of d+6 decision tree models are built and integrated into a random forest model.

[0048] 3) Extracting data relationship features: Each decision tree will judge and classify each feature j, and learn the data relationship under the specific feature classification from the data; the data relationship features include the value range of the feature, the association between features and the data pattern;

[0049] 4) Progressive filling: Iterate through all missing values ​​in the database and record the number of missing values ​​under each feature column j as θ. j Establish the matrix [θ1,θ2,…θ] j The matrix [θ1, θ2, ... θ] is first processed, and then checked for any values ​​less than or equal to 0. If any are found, the data is removed. Next, bubble sort is used to sort the θ values ​​in the matrix from smallest to largest. Then, based on the sorted results, all data under feature j are filled in sequentially from smallest to largest. Finally, the matrix [θ1, θ2, ... θ] is traversed. j After checking all features, iterate through the database again to ensure there are no missing values.

[0050] 5) Regression prediction: Based on the experimental group data with complete feature values ​​in the original database, the random forest algorithm is used. Using the d+6 decision trees established in 2), the values ​​judged by the d+6 decision trees are calculated respectively, and the average value is taken to fill in the missing values ​​in turn.

[0051] Furthermore, in step (4), the complete dataset is divided into a modeling set and a validation set by random sampling.

[0052] Furthermore, the RF-Meta model was optimized through five-fold cross-validation.

[0053] Furthermore, the hyperparameter optimization uses GridSearchCV for multi-core grid-based acceleration to find the optimal hyperparameters.

[0054] An agricultural mulch film degradation cycle prediction device based on the RF-Meta model includes a processor and an executable program stored in a memory and capable of running on the processor. The device is characterized in that: when the processor executes the executable program, it implements an agricultural mulch film degradation cycle prediction method based on the RF-Meta model.

[0055] A storage medium containing a computer-executable program, characterized in that, when executed by a computer processor, the computer-executable program is used to perform a method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model.

[0056] Beneficial Effects: 1. Accurate Estimation of Degradation Cycle: Numerous parameters need to be considered in the degradation of plastic film (BDM), such as temperature, humidity, soil type, and microbial activity. The model examines the complex interactions of these parameters, thus obtaining an accurate estimate of the degradation cycle. 2. Improved Universality of Model Results: By collecting experimental data from major BDM-using areas worldwide, the established model can be applied to global BDM degradation scenarios. 3. Accelerated Degradation: The application of the predictive model allows farmers to better understand the relationship between key parameters such as soil moisture and total soil nitrogen and the rate of BDM degradation. This allows them to take measures to adjust these parameters to accelerate BDM degradation, thereby promoting environmental protection and reducing plastic residue on the land. 4. Scientific Selection of BDM: The predictive model can also provide farmers with a scientific basis to select the type of BDM best suited to local soil and climate conditions. This helps improve the efficiency of BDM, reduce waste and resource consumption, and ensure healthy crop growth. 5. Reduced Labor Costs: With the degradation cycle of BDM known, farmers no longer need large-scale manpower to clean up BDM fragments in farmland. Farmers can plan their farmland more flexibly, better schedule the replacement of plastic film and crop rotation, and achieve sustainable ecological agricultural production. Attached Figure Description

[0057] Figure 1 A flowchart is established for the model of this invention;

[0058] Figure 2 A graph showing the prediction accuracy of the independent validation set evaluated by the coefficient of determination;

[0059] Figure 3 A funnel plot of the results of the Egger regression test. Detailed Implementation

[0060] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0061] like Figure 1 As shown, the model building process of this invention includes the following steps:

[0062] (1) Collect the original dataset of biodegradable mulch film degradation using meta-analysis;

[0063] (2) Use a random forest model to fill in the missing values ​​in the original dataset to obtain a complete database;

[0064] (3) Divide the complete database into a modeling set and a validation set, and establish an initial RF-Meta model.

[0065] (4) Train the initial RF-Meta model and optimize the hyperparameters to obtain the optimized RF-Meta model;

[0066] (5) Optimize the hyperparameters of the optimized random forest prediction model to obtain the optimal RF-Meta model;

[0067] (6) Verify the accuracy of the random forest prediction model.

[0068] Example:

[0069] 1. Obtaining the BDM Degradation Dataset: The original data for model building were collected and integrated using the PRISMA workflow of the meta-analysis method. Peer-reviewed publications investigating the effects of climate, materials, and soil environments on the degradation of biodegradable films over 10 years and the China National Knowledge Infrastructure Database were searched using the Web of Science website. To maintain quality control, the following criteria were established for data inclusion: (1) Experimental conditions should be similar to the natural degradation environment, and no additional strains should be added to the soil environment for cultivation. The degradation temperature of the soil (except for compost) should not exceed 35°C. (2) The main components of the BDM materials used in the experiment need to be reported, with PLA and PBAT accounting for more than 30%. (3) Experimental data are replicated, and information on the degraded soil in the experiment can be obtained.

[0070] Variables collected from these literatures included: literature titles; publication year; study location; meteorological variables, including cumulative rainfall, average temperature, and solar radiation during the degradation period; soil properties, including initial soil moisture content, total soil nitrogen content, soil organic matter, pH, clay content, mud content, sand content, bulk density, and coarse soil content; material variables, including PLA content, PBAT content, material thickness, and material surface area; degradation time; burial depth; and the Chao1 and Shannon indices among soil ecological variables. Additionally, we considered whether the BDM had undergone more than seven days of outdoor exposure before interpretation and whether it had been tilled during degradation.

[0071] We supplemented some missing values ​​using known databases: the China Global Land Reanalysis 40-Year Product (CRA / Land) - Monthly Product and the SoilGrids database. The formula for calculating the degradation rate of BDM is as follows:

[0072]

[0073] This formula represents the degradation rate obtained by calculating the mass decrease before and after degradation. In the formula, W... b and W t This represents the weight of the sample before it was buried in the soil and after it was extracted from the soil at the end of degradation.

[0074]

[0075] This equation calculates the percentage of current degradation by dividing the amount of CO2 released by the theoretical amount of CO2 released when BDM is fully biodegraded. t The value represents all CO2 emissions generated during the biodegradation of BDM, while (CO2) b Corresponding to the blank experiment without BDM, it represents CO2 emissions in the soil environment where BDM is absent. t It is the total dry weight of biodegradable plastic materials, C t 44 is the relative weight of total carbon in the sample, and 12 is the molecular mass of CO2 and the atomic mass of carbon.

[0076] The funnel plot of the model's Egger regression test results is shown below. Figure 3 As shown, the left and right sides are symmetrical, and the data points are distributed at the bottom of the funnel, rather than exhibiting obvious asymmetry or deviation from a straight line, indicating that the data is not significantly biased and the results are highly reliable. The Rosenthal method was used to calculate a safety factor of 87778, meaning that approximately 87778 insignificant studies are needed to offset the currently observed statistical significance, indicating that the data is highly significant.

[0077] 2. A multivariate random-effects model (meta-regression) was fitted using the `rma.mv` function from the `metafor` package in R. The collected factors were ranked according to their heterogeneity (QM). Factors with QM > 1 were retained as follows: cumulative rainfall, average temperature, initial soil moisture content, total soil nitrogen content, soil organic matter, pH, clay content, mud content, sand content, bulk density, soil coarse grain content, PLA content, PBAT content, burial depth, Chao1 index, Shannon index, and degradation time.

[0078] 3. Database Data Population: Given the diverse sources of our data, resulting in 8% of the data being null values, we populated this small amount of null data in the database before formally starting modeling, using the same method as the final modeling process, without altering the final result. We adopted an intelligent, stepwise approach. The core idea of ​​this method is to traverse all features and start filling in the null values ​​from the features with the fewest missing data, thereby ensuring minimal modification to the original data features. The population process is as follows:

[0079] (1) Initial state: Our database contains multiple features, some of which may have missing values. Before the imputation process begins, we first select the feature with the fewest missing values, and we fill the missing values ​​of the other features with 0.

[0080] (2) Regression Prediction: We use known data to perform random forest regression prediction. By estimating the possible values ​​of missing features through the data relationships of other complete data, the estimated values ​​are consistent with the characteristics of the overall data. This predicted value will be put back into the original feature matrix to gradually improve the data.

[0081] (3) Progressive imputation: After we successfully imput a feature, we continue to select the next feature with the fewest missing values ​​and repeat the above steps until all features have been traversed.

[0082] Ultimately, we filled all the empty values ​​in the database with highly reliable values ​​based on the existing information.

[0083] 4. Constructing the Modeling Set and Independent Validation Set: The entire collected dataset is used as the sample set, with the degradation rate representing the fitting target and the remainder as feature data. The sample set is divided into a modeling set and an independent validation set. Through random sampling, 70% of the data is assigned to the modeling set, and the remaining 30% is assigned to the independent validation set. In this embodiment, the total dataset contains 732 data samples.

[0084] 5. Training the initial random forest prediction model: Using the BDM degradation rate data and selected variables from the modeling set as training data, the initial random forest prediction model is trained. The parameters of the initial random forest prediction model are optimized through five-fold cross-validation to obtain the random forest prediction model.

[0085] The random forest prediction model provided in this invention is a data mining algorithm that generates n datasets (n is usually set to 500) of the same size as the training dataset using the bootstrap method (sampling with replacement), and these datasets are used to construct n decision trees. Environmental variables are randomly divided into multiple subsets, and within each decision tree, branches are generated by randomly dividing these subsets. The final prediction result of the model is the average of the prediction results from all decision trees.

[0086] Hyperparameter optimization: The hyperparameters in the model are optimized to obtain the best fitting effect. GridSearchCV is used for multi-kernel grid search to find the optimal hyperparameters. The final hyperparameter values ​​obtained are: n_estimators = 186, random_state = 42, max_depth = 23.

[0087] 6. Validate the Random Forest Prediction Model: The established random forest prediction model is used to predict the independent validation set, and the prediction accuracy of the random forest transformation function is evaluated by comparing it with the actual degradation rate on the independent validation set. In this embodiment, the prediction accuracy of the independent validation set is evaluated using the coefficient of determination (R²). The evaluation results are as follows: Figure 2 As shown. By Figure 2 As shown, the R² of the independent validation set is 0.965, indicating good predictive performance. We also validated the model with the root mean square error (RMSE) of 6.332, indicating that the mean prediction error of our model is 6.3%. Given the complexity of agricultural scenarios, it can be considered that our established model can effectively predict the degradation rate of BDM and has practical application prospects.

[0088] Meta-analysis is a statistical method used to synthesize the results of multiple independent studies to obtain more robust and comprehensive conclusions. Through meta-analysis, patterns, effect sizes, and confidence levels can be identified, helping decision-makers make more reliable conclusions. It can reveal general patterns in research while taking into account the weights and biases of individual studies, thus improving the comprehensiveness and reliability of the data.

[0089] Random forest is a tree-based machine learning method that effectively handles nonlinear relationships and exhibits good robustness to most regression problems. Furthermore, regression forests can effectively avoid model overfitting, making them suitable for building models with smaller datasets. Therefore, combining meta-analysis and random forest models holds promise for establishing a degradation prediction model for BDM.

[0090] The model proposed in this invention is the first to provide specific numerical predictions of the degradation degree of BDM during its degradation process, and it has broad applicability, applicable to BDM prediction in different soil, climate, and material environments. Furthermore, through meta-analysis to filter and process the original information, it ensures that the dataset will not mislead the model. In new scenarios, only the initial physicochemical information of the degradation scenario is needed to predict its future degradation capacity. Compared to conventional testing methods that obtain BDM degradation characteristics from local farmland through field experiments, this approach requires less manpower and resources, and eliminates the need for lengthy experimental periods, providing a new approach to predicting BDM degradation cycles.

Claims

1. A method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model, characterized in that, Includes the following steps: (1) Collect the original dataset of biodegradable mulch film degradation by meta-analysis, establish an original database based on the original dataset, and verify the reliability of the original data by meta-analysis. (2) The impact of different factors on the degree of degradation in the original dataset through meta-analysis Q M Sort; (3) Based on the meta-analysis results, screen those that meet the Q criteria. M Factors with defined value ranges are left in the original database. The random forest algorithm is used to fill in the missing values ​​in the original database to obtain a complete database. (4) Divide the complete database into a modeling set and a validation set, and establish an initial RF-Meta model; (5) Train the initial RF-Meta model and optimize the hyperparameters to obtain the optimized RF-Meta model; (6) The initial experimental physicochemical data of the degradation of the target biodegradable mulch film were imported into the optimized RF-Meta model to obtain the prediction curve of the entire degradation cycle. The specific steps for constructing the original database in step (1) are as follows: (1.1) Develop search strategies and inclusion / exclusion criteria based on the PICOS guidelines; (1.2) Select an article database, establish search terms according to the criteria in (1.1), and manually further screen articles that meet the research conditions and can be repeated; (1.3) The feature data set G1 to be collected in each article is composed as follows: G1={A1,A2...,A...} a ;C1,C2,...,C e ;SP1,SP2,...,SP p ;SB1,SB2..., SB b M1, M2..., M m E1, E2..., E e }; Among them, A a C represents item a in the basic information of the article to be collected. e SP represents the e-th item in the climate conditions to be collected. p SB represents the p-th term in the soil physical conditions to be collected. b M represents the bth item in the biological and chemical conditions of the soil to be collected. m E represents the m-th term in the properties of the mulch film material to be collected. e This represents the e-th item in the experimental conditions to be collected; (1.4) The composition of the degradation result dataset G2 to be collected in each article is as follows: G2={E m E v C m C v }; where E m E represents the mean of the experimental group results. v C represents the variance of the experimental group results. m C represents the mean of the control group results. v The variance of the control group results is represented by the following formulas for the mean and standard deviation: in, is the mean of the degradation rate, and s is the standard deviation of the degradation rate; It is the degradation rate of the i-th group, where i belongs to 1, 2, 3...n, and n is the total amount of data in the database; Step (2) is as follows: (2.1) Scan the degradation rate data and variance data to check for missing or null values. Once a null value is found, record the location information of the null value, including the row and column index where the null value is located, and fill it using the nearest neighbor linear interpolation method. (2.2) Establish a multivariate random effects model for the database in the meta-analysis; the basic random effects component of the model arises from the different sources of the extracted data; the formula for the multivariate random effects model is: yi=Xiβ+ui+ϵi Where: yi is the effect size of the i-th study, Xi is the covariate / predictor variable associated with the i-th study, β is the regression coefficient of the model, ui is the random effect of the i-th study, representing the heterogeneity among studies, and ϵi is the error term of the model; (2.3) The weight of each feature's influence on the result is calculated using the following formula: Qm is a weight value used to measure the influence of each feature on the result, where j represents the Qm value for the j-th feature, and k represents the number of data points in the database containing that feature. It is the weighted sum of the j-th feature factors. It represents the degradation rate of the j-th characteristic factor.

2. The method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model according to claim 1, characterized in that, The verification of the reliability of the original data through meta-analysis includes: (1.5) The publication bias of the collected articles was examined by observing the relationship between the degradation rate and its variance, i.e., the data source had obvious bias; the degradation rate of the biodegradable mulch film obtained from each experimental data set was ranked according to the degradation rate, and the location number u of each data point was obtained. i Then, sort each group of data according to the size of its experimental variance, and obtain the position number v after sorting. j Then, the correlation coefficient D between the degradation rate and its experimental variance, and the probability P of the hypothesis represented by D, were calculated: in, This represents the number of concordance pairs in pairwise comparisons. The number of disharmony pairs in pairwise comparisons; if the u of the two variables is... i and v j If both variables increase or decrease simultaneously, they form a concordant pair, where u represents the sum of the two variables. i and v j If one variable increases while the other decreases, they form a disharmony pair. The D value is the cumulative distribution function value under the standard normal distribution. By referring to the D value and the P value, we can determine whether there is publication bias. If the absolute value of the D value is greater than 1.96 and the p value is less than 0.05, the data is considered to have a significant bias, and data sources with significant bias need to be excluded. If there is no significant bias, we can continue to the next step. (1.6) When publication bias is detected, first use R language to plot a funnel plot with degradation rate as the x-axis and experimental variance as the y-axis. The position vector of each data point in the plot is d. i =( , );by If the numerical vector d is the axis, then... t =(x t ,s t If there is clearly no data near the corresponding point on the other side of the axis, resulting in an asymmetrical funnel plot, then further determine whether the data vector corresponds to a sound experimental design, meets the experimental requirements, and the journal whose paper is published has an impact factor (IF) > 6. If so, add the vector to the database. To eliminate bias; if the data vector corresponds to an experiment with special experimental conditions or low reliability, delete the data in the database; after filtering out the data that cause asymmetry in the funnel plot, repeat (1.6) and recheck the publication bias; (1.7) Assess whether there are general patterns in the data in the database; the significance level of the data is calculated as follows: Data significance propensity level When F is greater than zero, the data in the database is considered to have a significant general pattern. n If the value is less than 0, then we need to go back to step (1.1) to add intervention measures to the selection of articles again.

3. The method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model according to claim 1, characterized in that, Step (3) specifically involves: 1) Initial state: The original database established using the Meta-analysis method contains some empty data values. Let A be the j-th feature data of the i-th data in the database. ij ; 2) Random Forest Architecture: The data is organized according to the different feature data to be collected in step (1.3). There are different feature data under each feature category. For each feature category, an independent decision tree model is built for each feature. At the same time, s features from different feature categories are randomly selected to build d independent decision tree models to simulate the mutual influence of features under different feature categories. Finally, a total of d+6 decision tree models are built and integrated into a random forest model. 3) Extracting data relationship features: Each decision tree will judge and classify each feature j, and learn the data relationship under the specific feature classification from the data; the data relationship features include the value range of the feature, the association between features and the data pattern; 4) Progressive Filling: Iterate through all missing values ​​in the database and record the number of missing values ​​under each feature column j as... Establish a matrix ], and check if there are any values ​​less than or equal to 0. If so, remove the data. Then, bubble sort is used to sort the matrix. The values ​​are sorted in ascending order; then, based on the sorted results, all data under feature j are filled in sequentially from smallest to largest; finally, the matrix is ​​traversed. After checking all features, iterate through the database again to ensure there are no missing values. 5) Regression prediction: Based on the experimental group data with complete feature values ​​in the original database, the random forest algorithm is used. Using the d+6 decision trees established in 2), the values ​​judged by the d+6 decision trees are calculated respectively, and the average value is taken to fill in the missing values ​​in turn.

4. The method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model according to claim 1, characterized in that, In step (4), the complete dataset is divided into a modeling set and a validation set by random sampling.

5. The method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model according to claim 1, characterized in that, The RF-Meta model was optimized using five-fold cross-validation.

6. The method for predicting the degradation cycle of agricultural mulch film based on the RF-Meta model according to claim 1, characterized in that, The hyperparameter optimization uses GridSearchCV for multi-core grid-based acceleration to find the optimal hyperparameters.

7. A device for predicting the degradation cycle of agricultural mulch film based on an RF-Meta model, comprising a processor and an executable program stored in a memory and capable of running on the processor, characterized in that: When the processor executes the executable program, it implements the method as described in any one of claims 1-6.

8. A storage medium containing a computer-executable program, characterized in that, When executed by a computer processor, the computer executable program is used to perform the method as described in any one of claims 1-6.

Images