Method and equipment for predicting the decomposition cycle of agricultural mulch film based on the RF-Meta model

Abstract

The present invention discloses a method for predicting the decomposition cycle of an agricultural multi-film based on an RF-Meta model. This method includes the steps of collecting the original dataset of the decomposition of the biodegradable multi-film through Meta-analysis; ranking, through Meta-analysis, the influence QM of different factors in the original dataset on the degree of decomposition; screening to leave the factors that meet the requirements in the database and filling in the missing values in the original database using the random forest algorithm to obtain a complete database; splitting the complete database into a modeling set and a verification set and constructing an initial RF-Meta model; training the initial RF-Meta model and optimizing the hyperparameters to obtain an optimized RF-Meta model; and introducing the physical and chemical data of the initial experiment regarding the decomposition of the target biodegradable multi-film into the optimized RF-Meta model to obtain a prediction curve of the complete decomposition cycle. The model proposed by the present invention can provide, for the first time, a specific numerical prediction of the degree of decomposition in the decomposition process of BDM and has broad applicability.

Description

[Technical Field]

[0001] The present invention relates to the field of agricultural informatization, and in particular to a method and apparatus for predicting the soil burial decomposition cycle of biodegradable mulch films through a random forest algorithm combined with a Meta analysis method. [Background technology]

[0002] Mulch film is an important agricultural production tool. It can effectively suppress weed growth, increase crop yields, maintain soil moisture, reduce water evaporation, and conserve water resources. It also increases soil temperature, promotes crop growth, reduces soil erosion, and maintains soil fertility, thereby extending crop life cycles, reducing the risk of pests and diseases, reducing the use of pesticides and chemical fertilizers, and improving grain quality. China uses a high volume of mulch film annually, at 137.9 million tons. However, due to UV radiation and external forces, mulch film easily fragments and forms plastic fragments, causing serious plastic pollution in farmland and posing a potential threat to ecosystems and human health. The impact of plastic pollution is currently attracting increasing attention from the scientific community and the general public. Biodegradable plastics are plastics that can be decomposed and utilized by microorganisms, ultimately converting them into microbial biomass, carbon dioxide, and water. Therefore, replacing PE mulch film with biodegradable plastic mulch film (BDM) is one of the key strategies for preventing plastic pollution in agriculture. However, the degradation cycle of BDM in the environment is still unclear, mainly because the degradation of BDM is a complex process and is closely related to the material components and natural conditions, and the degradation effects of different formulations of BDM in different regions often vary greatly.

[0003] Currently, research on the decomposition cycle of specific BDM mulch film products mainly involves empirical estimation after one to two years of use in local farmland. While there have been studies examining the effects of material, soil, and regional factors on BDM decomposition, these studies primarily rely on experimental data and can only determine general trends in decomposition. The relatively simple research methods and the limited number of control groups in experiments make it difficult to measure the impact of relevant factors on decomposition, making it impossible to predict the decomposition cycle. Existing studies have used simple fitting models to evaluate the photodecomposition cycle of mulch film exposed to ultraviolet (UV) light. However, these studies have the following shortcomings: (1) UV is not the only relevant variable affecting BDM; (2) BDM decomposition is complex, often involving nonlinear interactions between material, soil, climate, and human factors. Analysis of a small number of factors fails to comprehensively analyze these interactions, making it impossible to specifically predict the decomposition process. (3) Existing studies have been unable to address the regional characteristics of mulch film decomposition. Therefore, it is urgent to develop a nonlinear BDM decomposition prediction model that covers a wider range of variables and can be widely applied to mulch film use regions. Summary of the Invention [Problem to be solved by the invention]

[0004] The present invention aims to provide an RF-Meta model that combines the Meta analysis method and the Random Forest algorithm to predict the decomposition cycle of agricultural mulch films, achieve accurate estimation of the decomposition cycle, and improve the versatility of the model results. [Means for solving the problem]

[0005] A method for predicting the decomposition cycle of agricultural mulch films based on the RF-Meta model, (1) Collecting an original data set of biodegradable mulch film decomposition data through a meta-analysis method, constructing an original database based on the original data set, and verifying the reliability of the original data through a meta-analysis method; (2) Through meta-analysis, the influence of different factors in the original dataset on the degree of decomposition Q M and (3) According to the results of the Meta analysis, M Selecting factors with a value of >1 to remain in the original database, and filling in the missing values ​​in the original database using a random forest algorithm to obtain a complete database; (4) splitting the complete database into a modeling set and a validation set and building an initial RF-Meta model; (5) training the initial RF-Meta model and optimizing the hyperparameters to obtain an optimized RF-Meta model; (6) The physical and chemical data from the initial experiments on the degradation of the target biodegradable mulch film are introduced into the optimized RF-Meta model to obtain a prediction curve for the complete degradation cycle.

[0006] Furthermore, in the step (1), the specific steps for constructing the original database are as follows: (1.1) The steps for developing the search strategy, inclusion and exclusion criteria according to the PICOS principles are: P: The research subject is to determine biodegradable mulch films in agricultural production situations, I: Determine intervention measures, select the range of basic material components that make up degradable mulch films, and select different decomposition measures, decomposition environments, and pre-treatments before decomposition; C: To determine the properties of the mulch film before and after decomposition as a control measure; O: Determining the resulting direction as a decomposition rate parameter of the mulch film; S: Includes limiting research designs to laboratory or field experiments and excluding reviews and opinion articles.

[0007] (1.2) Select an article database, set search terms based on the principles of (1.1), and manually select further articles that meet the research criteria and can be repeated. (1.3) The structure of the feature dataset G1 to be collected for each article is 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 represents the a-th item of basic information of the article to be collected, and C e represents the e-th item of the climatic conditions to be collected, and SP p represents the pth item of the physical condition of the soil to be collected, and SB b represents the bth item of biological and chemical conditions of the soil to be collected, and M m represents the m-th item of the characteristics of the multi-film material to be collected, and E e represents the e-th item of the experimental conditions to be collected.

[0008] (1.4) The composition of the decomposition result dataset G2 to be collected for each article is G2 = {E m ,E v ,C m ,C v}, where E m represents the mean value of the experimental group results, and E v represents the variance of the experimental group results, and C m represents the mean value of the results for the control group, and C v represents the variance of the results for the control group, and the formulas for the mean and standard deviation are as follows:

number

[0009] (1.5) By observing the relationship between the magnitude of the decomposition rate and its variance, we can detect whether there is publication bias in the collected articles, that is, whether there is an obvious trend in the data source. We then rank the decomposition rates of biodegradable mulch films obtained from each group of experimental data according to their magnitude, and assign each data position number u i , and the data for each group are ordered according to the magnitude of the experimental variance, followed by the position number v j and calculate the probability of occurrence P of the hypothesis represented by D and the correlation coefficient D between the magnitude of the degradation rate and its experimental variance.

number

[0010] (1.6) If publication bias is found, first use the R language to draw a funnel chart with the decomposition rate as the X-axis data and the experimental variance as the Y-axis data. The position vector of each data in the chart is d i =(xi ,s i ) and

number

number

[0011] (1.7) Evaluate whether there is a general rule in the data in the database, and the significance level of the data trend is calculated as follows:

number

[0012] Furthermore, step (2) is specifically as follows. (2.1) Scan the decomposition rate data and variance data to check whether there are any missing or null values. If 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 nearest neighbor linear interpolation.

[0013] (2.2) For the database, a multivariate random effect model in Meta analysis is constructed. The basic random effect part of the model is generated by different literature sources extracted from the data. The formula of the multivariate random effect model is as follows:

number

[0014] (2.3) Calculate the weight value of the influence of each feature on the result using the following formula:

number

number

[0015] Furthermore, the step (3) specifically includes: 1) The initial state is as follows: some data values ​​in the original database constructed using the Meta analysis method are empty, and the jth feature data of the i-th data in the database is A ij Record it as 2) Form a random forest architecture and organize the data according to the type of feature data collected in step (1.3) (A, C, SP, SB, M, E). Each feature category has different feature data. For each feature category, build an independent decision tree model for each feature and simultaneously extract features from different feature categories.

number

[0016] Furthermore, in step (4), the complete dataset is divided into a modeling set and a validation set using a random sampling method.

[0017] Furthermore, we optimize the RF-Meta model by 5-fold cross-validation.

[0018] Furthermore, in the hyperparameter optimization, GridSearchCV is used to perform high-speed grid search using multiple cores to find the optimal hyperparameters.

[0019] An apparatus for predicting the decomposition cycle of agricultural mulch film based on the RF-Meta model, comprising a processor and a program stored in memory and executable on the processor, wherein when the processor executes the executable program, the apparatus realizes a method for predicting the decomposition cycle of agricultural mulch film based on the RF-Meta model.

[0020] A storage medium containing a computer-executable program, characterized in that the computer-executable program, when executed by a computer processor, performs a method for predicting the decomposition cycle of agricultural mulch film based on the RF-Meta model. [Effects of the Invention]

[0021] The beneficial effects of the present invention are as follows: 1. Accurately estimate the decomposition cycle. There are many parameters that need to be considered in the decomposition of mulch film, including temperature, humidity, soil type, and microbial activity. The model can examine the complex interactions of these parameters and achieve accurate estimation of the decomposition cycle. 2. Improve the versatility of the model results. By collecting experimental data from major mulch film-using regions around the world, the constructed model can be used in global BDM decomposition scenarios. 3. Accelerate decomposition. Using predictive models, farmers can gain a better understanding of the relationship between key parameters, such as soil moisture and total soil nitrogen, and the rate of mulch film decomposition. This allows them to take steps to adjust these parameters to accelerate mulch film decomposition, thereby promoting environmental protection and reducing plastic residues on the land. 4. Scientific selection of mulch film: The predictive model can also provide farmers with a scientific basis for selecting the type of mulch film that is most suitable for their local soil and climate conditions, thereby improving the effectiveness and benefits of mulch film, reducing waste and resource consumption, and ensuring healthy crop growth. 5. Reduce labor costs. After the decomposition cycle of mulch film is understood, farmers no longer need large-scale manpower to clean up mulch debris from their fields. Farmers can plan their farms more flexibly, better plan mulch replacement and crop rotation, and achieve sustainable and environmentally friendly agricultural production. [Brief explanation of the drawings]

[0022] [Figure 1] FIG. 1 is a flowchart of the model construction in the present invention. [Figure 2] Figure 2 shows the results of the predictive accuracy of the independent validation set as assessed by the coefficient of determination. [Figure 3] Figure 3 shows a funnel chart drawn according to the Egger regression test results. DETAILED DESCRIPTION OF THE INVENTION

[0023] The technical solutions of the present invention are further described below in conjunction with the accompanying drawings.

[0024] As shown in FIG. 1, the model construction flowchart of the present invention is as follows: Collecting an original dataset of biodegradable mulch film degradation through a meta-analysis method; (1) filling in missing values ​​in the original dataset using a random forest model to obtain a complete database; (2) splitting the complete database into a modeling set and a validation set and building an initial RF-Meta model; (3) training the initial RF-Meta model and optimizing the hyperparameters to obtain an optimized RF-Meta model; (4) optimizing the hyperparameters of the optimized random forest predictive model to obtain an optimal RF-Meta model; (5) verifying the accuracy of the random forest prediction model. [Example]

[0025] 1. Obtain a BDM decomposition dataset. The original data for building the model was collected and integrated using the PRISMA workflow of the Meta-analysis method. The Web of Science website was used to search peer-reviewed publications investigating the effects of climate, materials, and soil environment on the decomposition of biodegradable films over a 10-year period, as well as the China National Knowledge Infrastructure Database. To maintain quality control, data collection criteria were established. (1) Experimental conditions should mimic natural decomposition conditions, and no additional bacterial strains should be added to the soil environment. The decomposition temperature of the soil (excluding compost) should not exceed 35°C. (2) The main components of the BDM material used in the experiment must be reported, with the proportion of PLA and PBAT exceeding 30%. (3) Experimental data could be repeated to obtain information on the degraded soil during the experiment.

[0026] Variables collected from these documents included the title of the document, publication year, study location, meteorological variables (including cumulative rainfall, average temperature, and solar radiation during the decomposition cycle), soil properties (including initial soil moisture content, total soil nitrogen content, soil organic matter, pH, clay content, mud content, sand content, volumetric weight, and soil coarse particle content), material variables (including PLA content, PBAT content, material thickness, and material surface area), decomposition time, burial depth, and soil ecological variables (Chao's index and Shannon index). We also considered whether the BDM had experienced outdoor exposure for more than 7 days before decomposition and whether it had been tilled during the decomposition process.

[0027] Some missing values ​​were filled in through known databases, and the database sources were the China Global Land Reanalysis 40-year Product (CRA / Land) - Monthly Product (LandProduct) and the SoilGrids database. The calculation formula for the BDM decomposition rate is as follows:

number

number

[0028] The funnel chart drawn according to the Egger regression test results of the model is shown in Figure 3. The left and right sides are symmetrical, and the data points are distributed at the bottom of the funnel, showing no obvious asymmetry or deviation from a straight line, indicating that there is no obvious bias in the data and that the results are highly reliable. The fail-safe number calculated using the Rosenthal method is 87,778, which means that approximately 87,778 insignificant studies would be required to offset the currently observed statistical significance, indicating that the data in question is clearly significant.

[0029] 2. A multivariate random effects model (meta-regression) was fitted using the rma.mv function in the metafor package in R. The collected factors were ranked according to the heterogeneity QM caused by each factor, and the factors with a QM > 1 remaining were cumulative rainfall, mean temperature, initial soil moisture content, total soil nitrogen content, soil organic matter, pH, clay content, mud content, sand content, volumetric weight, soil coarse particle content, PLA content, PBAT content, burial depth, Chao1 index, Shannon index, and decomposition time.

[0030] 3. Filling the database data: Because the data source is relatively complex and 8% of the data is null, before formal modeling begins, the same method as for final modeling is used to fill in the small amount of null data in the database without changing the final result. An intelligent step-by-step method is used, and the gist of this method is to traverse all features and start filling in the features with the least missing data, thereby minimizing changes to the features in the original data. The filling process is as follows: (1) The initial state is as follows: The database contains multiple features, some of which may have missing values. Before starting the filling process, we first select the feature with the least missing values ​​and fill the missing values ​​of other features with 0. (2) Regression prediction: Random forest regression prediction is performed using known data to estimate possible values ​​of missing features through data relationships with other complete data. The values ​​estimated in this way are consistent with the features of the entire data. This predicted value is then added back to the original feature matrix to gradually complete the data. (3) Progressive filling: After one feature is successfully filled, continue to select the next feature with the least number of missing values, and repeat the above steps until all features have been traversed. Finally, based on existing information, all null values ​​in the database were filled with reliable values.

[0031] 4. Construct a modeling set and an independent validation set. All the collected datasets above are used as the sample set, with the decomposition rate as the fitting target and the remainder as feature data. The sample set is divided into a modeling set and an independent validation set. By random sampling, 70% of the data in the dataset is introduced into the modeling set, and the remaining 30% of the data is introduced into the independent validation set. The total dataset in this example contains 732 data samples.

[0032] 5. Train the initial random forest prediction model. The BDM degradation rate data and many selected variables in the modeling set are used as training data to train the initial random forest prediction model, and the parameters of the initial random forest prediction model are optimized by 5-fold cross-validation to obtain the random forest prediction model.

[0033] The random forest prediction model provided by the present invention is a data mining algorithm that uses the bootstrap method (sampling with replacement) to generate n datasets (usually n is set to 500) of the same size as the training sample set, and then constructs n decision trees. Environmental variables are randomly divided into multiple environment variable subsets, and in each decision tree, nodes are branched by randomly dividing the multiple environment variable subsets. The final prediction result of the model is the average of the prediction results of all the decision trees.

[0034] Optimize the hyperparameters. Optimize the hyperparameters in the model to obtain the best fitting effect. GridSearchCV is used to perform a fast grid search using multiple cores to find the optimal hyperparameters. The final hyperparameter values ​​obtained are n_estimators=186, random_state=42, and max_depth=23.

[0035] 6. Validation of the Random Forest Prediction Model. The constructed Random Forest prediction model was used to predict an independent validation set, and the true decomposition rate within the independent validation set was compared to evaluate the predictive accuracy of the Random Forest transformation function. In this example, the predictive accuracy of the independent validation set was evaluated using the coefficient of determination (R2). The evaluation results are shown in Figure 2. As can be seen from Figure 2, the R2 of the independent validation set was 0.965, demonstrating good predictive effectiveness. The root mean square error (RMSE) of the data was also verified, yielding a value of 6.332, representing an average prediction error of the model of 6.3%. Considering the complexity of agricultural scenarios, the constructed model can predict the decomposition rate of BDM very well and is considered to have practical prospects.

[0036] Meta-analysis is a statistical method for integrating the results of multiple independent studies to arrive at stronger, more comprehensive conclusions. Meta-analysis can identify the mode, effect size, and reliability, helping decision makers reach more reliable conclusions. It can clarify general rules across studies and simultaneously consider the weight and variance of each study, improving the comprehensiveness and reliability of the data.

[0037] Random forests are a tree-based machine learning method that can handle nonlinear relationships well and exhibits good robustness for most regression problems. Furthermore, regression forests are suitable for building models with small amounts of data because they can better avoid overfitting. Therefore, combining Meta analysis with random forest models is expected to build a predictive model for BDM decomposition.

[0038] The model proposed by this invention is the first to provide specific numerical predictions of the degree of degradation in the BDM degradation process. It has wide applicability and can be used to predict BDM degradation in different soil, climate, and material environments. Furthermore, by filtering original information through meta-analysis, it is possible to prevent data sets from providing incorrect information to the model. In new scenarios, future degradation potential can be predicted simply by providing the initial physical and chemical information for the degradation scenario. Compared to traditional testing methods that rely on field experiments to obtain BDM degradation characteristics in local agricultural fields, this model offers a new approach to predicting BDM degradation cycles without requiring extensive human and material resources or long experimental periods.

Claims

1. A method for predicting the decomposition cycle of agricultural mulch film based on an RF-Meta model by a computer having a processor and a memory executing a program stored in the memory, comprising: (1) collecting original data sets of biodegradable mulch film decomposition data through Meta analysis, constructing an original database based on the original data sets, and verifying the reliability of the original data through Meta analysis means; (2) Through Meta analysis, the influence of different factors in the original dataset on the degree of decomposition, Q Ｍ and (3) According to the Meta analysis results, Q Ｍ selecting factors that satisfy a set value range so as to remain in the original database, and imputing missing values ​​in the original database using a random forest algorithm to obtain a database in which the missing values ​​have been imputed; (4) dividing the database with the imputed missing values ​​into a modeling set and a validation set, and constructing an initial RF-Meta model; (5) training the initial RF-Meta model and optimizing the hyperparameters to obtain an optimized RF-Meta model; (6) A method for predicting the decomposition cycle of agricultural mulch films based on the RF-Meta model, comprising the steps of inputting physical and chemical data for the initial state of the biodegradable mulch film to be predicted into the optimized RF-Meta model to obtain a prediction curve for the entire decomposition cycle.

2. In the step (1), the specific steps for constructing the original database are as follows: (1.1) The steps for developing the search strategy, inclusion and exclusion criteria according to the PICOS principles are: P: The research subject is to determine the biodegradable mulch film in agricultural production. I: Identify the factors that can affect the decomposition of biodegradable mulch films, select the range of basic material components that make up biodegradable mulch films, and select different decomposition means, decomposition environments, and pre-treatments before decomposition; C: Determining the properties of the mulch film before and after decomposition as a control means; O: Determining the evaluation index to be a decomposition rate parameter of the mulch film; S: Including limiting research designs to laboratory or field experiments and excluding reviews and opinion articles; (1.2) Select an article database, set search terms based on the principles of (1.1), and manually select further articles that meet the research criteria and can be repeated; (1.3) Feature dataset G to be collected for each article １ The composition of G １ = {A １ , A ２ ...., A ａ ; C １ , C ２ ,...,C ｅ ;SP １ ,SP ２ ,...,SP ｐ ;SB １ ,SB ２ ...,SB ｂ ;M １ ,M ２ ...,M ｍ ;E １ ,E ２ ...., E ｅ }, where A ａ represents the a-th item of basic information of the article to be collected, and C ｅ represents the e-th item of the climatic conditions to be collected, and SP ｐ represents the p-th item of the physical conditions of the soil to be collected, and SB ｂ represents the b-th item of biological and chemical conditions of the soil to be collected, and M ｍ represents the m-th item of the characteristics of the multi-film material to be collected, and E ｅ represents the e-th item of the experimental conditions to be collected, (1.4) Dataset G of the decomposition results collected for each article ２ The composition of G ２ = {E ｍ , E ｖ , C ｍ , C ｖ }, where E ｍ represents the mean value of the experimental group results, and E ｖ represents the variance of the experimental group results, and C ｍ represents the mean value of the results for the control group, and C ｖ represents the variance of the results of the control group, and the formulas for the mean and standard deviation are: [0012] is the average value of the decomposition rate, s is the standard deviation of the decomposition rate, and x ｉ The method for predicting the decomposition cycle of agricultural mulch film based on the RF-Meta model according to claim 1, characterized in that: is the decomposition rate of the i-th group, where i belongs to 1, 2, 3...n.

3. Specifically, step (2) is (2.1) Scan the decomposition rate data and variance data to check whether there are missing or null values. If 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 nearest neighbor linear interpolation; (2.2) A multivariate random effects model in Meta analysis was constructed for the database. The basic random effects part of the model was generated by different literature sources extracted from the data. The formula of the multivariate random effects model was as follows: [Equation 17] where yi is the effect size of the ith study, Xi is the predictor variable associated with the ith study, β is the regression coefficient of the model, ui is the random effect of the ith study, representing heterogeneity between studies, and εi is the error term of the model. (2.3) Calculate the weight value of the influence of each feature on the result using the following formula: [Equation 18] Q ｍ is a weight value for measuring the influence of each feature on the result, j represents calculating Qm for the jth feature, k represents the number of data containing the feature in the database, [Equation 19] is the weight of the jth feature factor, and x .ｊ The method for predicting the decomposition cycle of agricultural mulch film based on the RF-Meta model according to claim 1, characterized in that j is the magnitude of the decomposition rate of the j-th characteristic factor.

4. The method for predicting the decomposition cycle of agricultural mulch film based on the RF-Meta model of claim 1, characterized in that the RF-Meta model is optimized by five-fold cross-validation.

5. A prediction device for the decomposition cycle of agricultural mulch film based on the RF-Meta model, comprising: a processor; and a memory storing a program including instructions that, when executed by the processor, cause the processor to perform each of the steps (1) to (6) of the method of claim 1.

6. A storage medium readable by a computer, having recorded thereon a program for causing a computer to execute each of steps (1) to (6) of the method according to claim 1.

Images