Compost maturity evaluation method and application

Abstract

The invention provides a compost maturity evaluation method and application. According to the method, an optimized data set is provided for establishing an optimal prediction model for machine learning by performing Meta analysis data preprocessing. Determining thresholds of four internal maturity indexes suitable for the data set by adopting a statistical method; and calculating the maturity day number meeting the statistical maturity threshold value by using the optimal prediction model. And carrying out importance grade sorting on the four internal maturity indexes in the compost heap by using a K-means unsupervised clustering algorithm based on the statistical maturity threshold. And according to the importance grade sequence, an analytic hierarchy process (AHP) is used to establish weight endowing, maturity days of different indexes are multiplied by corresponding weights and summed, and the unified compost maturity evaluation days are obtained. According to the method, key support is provided for machine learning modeling by means of Meta analysis, the machine learning model and the analytic hierarchy process based on Kmeans unsupervised clustering are combined, a comprehensive evaluation framework of compost maturity is established, and a new thought is provided for maturing judgment of charcoal compost.

Description

Technical Field

[0001] The invention belongs to the technical field of compost evaluation methods, and in particular relates to a compost maturity comprehensive evaluation framework design method and application. Background Art

[0003] Composting is widely recognized as an efficient method for recycling organic waste. It has gained significant recognition as a promising sustainable and eco-friendly biochemical technology due to its cost-effectiveness, simple operation, and adaptability to effectively manage the increasing amount of solid organic waste. The composting process utilizes inherent microorganisms to convert degradable organic matter into humus-like substances, thereby achieving harmless treatment of organic solid waste. The resulting products can be used as soil conditioners to improve soil physical and chemical properties. However, traditional aerobic composting, which relies solely on degradation and fermentation by native bacteria, has its limitations, such as low composting temperature during the thermophilic period, long maturation process, unstable product quality, greenhouse gas emissions, and nitrogen loss with ammonia. Therefore, it is imperative to solve these problems and realize efficient organic waste composting technology. Previous studies have adopted various measures to optimize the composting environment and accelerate the composting process. One of the most common methods is the incorporation of exogenous additives. Additives can be divided into three categories: chemical, physical, and microbial. Compared with chemical additives and microbial additives, biochar, zeolite, and MnO 2 Physical additives such as 3 showed greater efficacy in terms of emissions, thus demonstrating a positive potential for enhanced composting.

[0004] The effects of biochar on conventional solid waste composting have been extensively studied over the past three decades. The incorporation of biochar, which has multiple functional groups, large surface area, high porous structure, and cation exchange capacity, has emerged as a promising strategy to enhance composting by extending the thermotolerance period, promoting the humification process, accelerating maturity, and reducing greenhouse gas emissions. Co-composting with biochar overcomes the limitations associated with conventional composting. However, previous studies on the optimal biochar application conditions have been limited to single-additive designs with one or two indicators. Meanwhile, most previous studies have focused on a single aspect of reducing gas emissions or improving compost characteristics (e.g., pH, moisture content, and carbon-nitrogen ratio) and mitigating ecological risks. In addition, given the high heterogeneity of biochar characteristics and composting conditions that affect compost maturity (e.g., variations in biochar feedstock, pyrolysis temperature, particle size, total porosity), as well as initial composting conditions (including initial C / N, initial moisture content, and initial pH), the results of these experiments present substantial differences or even contradictions. These differences limit our comprehensive understanding of the main factors affecting the compost maturity process. Therefore, quantitative evaluation of the effects of biochar additives for enhancing composting is necessary.

[0005] Meta-analysis is a valuable statistical method that uses published studies to comprehensively evaluate the effects of biochar addition on target composting indicators. Previous studies have used meta-analysis to examine the effects of biochar properties and initial compost properties on internal maturity indicators (such as pH, conductivity, carbon and nitrogen losses, ammonia nitrogen, nitrate nitrogen) and external maturity indicators (such as N 2 O、CH 4 NH 3 ). Therefore, it is crucial to integrate internal and external maturity indicators and comprehensively evaluate the impact of high heterogeneity of biochar characteristics and composting conditions on compost maturity. Most of the existing meta-analyses do not verify the rationality of subgroups through heterogeneity tests, publication bias tests, and minimal significance analysis based on one-way analysis of variance (ANOVA), and the reliability of meta-analysis results cannot be guaranteed.

[0006] However, composting is a complex biochemical process that is easily affected by various conditions, which poses a challenge to accurately adjust parameters in achieving rapid composting. Therefore, developing a convenient, rapid, and user-friendly method to predict compost maturity indicators and achieve rapid composting is key to achieving efficient compost production.

[0007] Machine learning (ML), as a powerful tool for processing and analyzing large-scale datasets, provides a promising approach for predicting results quickly and accurately. Recent effective studies have used ML to predict the maturity stage of a specific composting system involving food waste and green waste based on one or two maturity indicators. These indicators include C / N loss rate, ammonium nitrogen to nitrate nitrogen ratio, germination index GI, etc. However, these predictions come from relatively small datasets, ranging from 242 to 480 samples. Existing machine learning work on predicting compost maturity does not perform a reliable data preprocessing process when building the model, which leads to incomplete input features of the machine learning model or even poor input features. At the same time, the dataset is not grouped and divided during modeling, which poses the risk of data leakage and model overfitting during the modeling process, and the true prediction performance of the machine learning model cannot be guaranteed. At the same time, by adding composting process indicators to the machine learning modeling process, the machine learning algorithm can be used to accurately predict the indicator changes throughout the compost maturity process, which can avoid the errors of manual experiments to determine maturity indicators and reduce the time and labor input required for experiments.

[0008] Furthermore, it is noteworthy that the prediction results of maturity indicators used to assess the maturity stage of compost in these studies were not consistent. Therefore, the lack of standardized criteria for evaluating maturity hinders the standardization of compost production. This is crucial because only mature compost can be considered as an organic fertilizer suitable for agricultural production. The utilization of immature compost products may lead to the production of plant toxic substances and contribute to the occurrence of pests and diseases that are detrimental to plant growth. Therefore, it is necessary to establish a compost maturity evaluation standard to optimize process performance and accurately predict the maturity stage. Summary of the invention

[0009] The purpose of the present invention is to solve the defects of the above-mentioned prior art and provide a compost maturity evaluation method and application.

[0010] Biochar is often used as an additive in the composting process due to its rich surface functional groups, unique porous structure and large specific surface area, and plays a very important role in improving the maturity of compost and improving the humification of compost. The maturity of the biochar composting system is judged by scoring various indicators of the compost. However, due to differences in raw material ratios, composting methods, environmental conditions, climate and other factors, there are differences in the performance of various indicators in the unripe and mature periods, and the lack of a unified evaluation standard has hindered the standardization and large-scale application of the composting process. Therefore, it is urgent to clarify the optimal properties that affect composting and to unify the threshold value of the maturity standard for composting. The present invention uses Meta-analysis to screen out the best biochar and compost properties, which provide key properties for machine learning modeling. At the same time, with the help of machine learning ML, a model is established to achieve accurate prediction of compost maturity indicators. On the basis of the model, a standardized maturity evaluation standard is established with composting days as a bridge, which significantly improves the operating efficiency, thereby improving decision-making and support for the engineering application of composting technology.

[0011] The present invention adopts the following technical solution:

[0012] A compost maturity evaluation method comprises the following steps:

[0013] Step 1. Perform a meta-analysis to comprehensively evaluate eight maturity indicators inside and outside the compost, screen the key biochar properties and initial compost properties that affect the final composting, and provide an optimized data set for machine learning to establish the optimal prediction model. The indicators inside the compost are C / N, GI, NO 3 - -N, NH 4 + -N, the in vitro indicator of the pile is NH 3 、N 2 O、CO 2 , CH 4 .

[0014] Step 2. Determine the statistical maturity threshold suitable for the data set, and use statistical methods to define the statistical maturity thresholds of the four internal maturity indicators; and when the four internal maturity indicators meet the statistical maturity thresholds, use the four indicators in the optimal prediction model of step 1 to calculate their respective maturity days.

[0015] Step 3. Based on the statistical maturity thresholds of the four internal maturity indicators, the K-means unsupervised clustering algorithm is used to rank the importance of the four internal maturity indicators in the compost pile. The K-means clustering algorithm is used to perform binary unsupervised clustering on the two-by-two combinations of the four compost indicators. In this process, the data points are automatically divided into mature and immature clusters. By calculating the distribution accuracy changes of cluster points in mature clusters under the statistical maturity threshold, the axial data are shuffled and re-clustered in a random shuffle manner. Compare the relative changes in the accuracy of the data points in the original cluster, the randomly shuffled X-axis cluster, and the randomly shuffled Y-axis cluster in the mature cluster to determine the importance ranking of the four internal indicators.

[0016] Step 4. According to the importance ranking results of the four internal maturity indicators in step 3, the importance evaluation matrix is ​​established using the hierarchical analysis method, and an evaluation is performed to determine whether the results are passed and tested to obtain the weights of the four internal maturity indicators.

[0017] Step 5. Based on the weights of the four internal maturity indicators in step 4, multiply the respective maturity days in step 2 to obtain a weighted unified maturity day.

[0018] Furthermore, the meta-analysis in step 1 includes:

[0019] S101. By setting keywords, literature was collected and screened in the database, and the subgroup combination was rationally verified through heterogeneity test, publication bias test, one-way analysis of variance combined with post hoc least significant analysis;

[0020] S102. Meta win 2.1 software was used to perform forest plots to explore the specific effects of biochar properties after subgroup classification and initial compost properties on eight compost maturity indicators, and the key biochar properties and initial compost properties that affect maturity were evaluated and proposed. The average effect size of each classification subgroup was calculated through the experimental and control groups of each indicator and the corresponding standard deviation and sample size. The average effect size of the subgroup was compared with the calculated overall effect size considering only whether biochar was added or not, and the indicators with better effects in the classification subgroups were determined and recommended;

[0021] S103. Based on the results of the forest plot, the influence relationship between biochar properties and compost properties and maturity indicators was analyzed through the structural equation model. The influence of property indicators on maturity indicators was judged and ranked through the standardized total effect of the structural equation model. The reliability of the forest plot results was verified and the importance of key biochar properties and key compost properties was revealed, providing optimization indicators for the machine learning model to establish a data set;

[0022] S104. Add composting process indicators when building the machine learning model, and establish a data set including 8 maturity indicators; select basic machine learning algorithms extra trees, random forest, gradient boosting, eXtremeGradient Boosting, and build an aggregate model (FUSION) of the five models through Python editing, group the data records of the same experiment, and then stratify them into training sets and test sets in a ratio of 8:2; use Bayesian optimization to perform hyperparameter tuning in R 2 (coefficient of determination), RMSE (root mean square error), and MAE (mean absolute error) were used as the criteria for evaluating the hyperparameter optimization algorithm, and the algorithm with the best performance in predicting the maturity index inside the pile and the gas index outside the pile was selected; 10 different random states were set to take the average to evaluate the performance of the five models, and R 2 , RMSE and MAE to compare the prediction accuracy of the five models, and select the best model whose performance is close to the average model for output as the best prediction model.

[0023] Furthermore, the data set establishment includes: using cosine similarity to establish the application domain of the model, when determining the application domain, using the training set as a reference data set, using the test set as an exploration set, using cosine similarity to calculate the similarity between the data in the test set and the data modeled in the training set, based on the similarity measure, forming different parts of the test set, data below a certain threshold is data outside the application domain, and data above a certain threshold is data within the application domain, for each pre-set cosine similarity, setting a gradient to discard data outside the application domain from the test data set and recalculating R 2 , the root mean square error and the amount of data discarded, discarding the lowest R 2 After the threshold data with the highest RMS error value is determined, ensure that the data outside the applicable range is small, and the data greater than the threshold will be identified as the optimal threshold.

[0024] Further, the screening rules are as follows: (1). The inclusion criteria require original research, and the quantitative research results show that the application of biochar leads to changes in compost maturity. (When collecting literature, the need for "original research" means that the literature must be data obtained by the author's experiment, and the topic must be (or include) literature on the effect of biochar as an additive on the maturity of compost. It is not a review or machine learning or meta-analysis article that requires the collection of data sets.) (2). Each study needs to include a control group and an experimental group, consisting of multiple groups for comparison. (3). The selected studies are expected to include biochar characteristics and initial compost. (In some studies, although biochar as an additive affects the maturity of compost, only the effect of adding or not adding biochar on the compost is analyzed. However, the properties of biochar or initial compost are not provided, so this type of article is outside our screening scope.)

[0025] Additionally, obtain the mean, standard deviation, and number of control and experimental groups or digitize from the chart using the GetDate image digitizer. For studies where only standard errors are available, use the formula SD was calculated, where n represented the sample size and for studies that did not provide SD and SE, SD was calculated as 1 / 10 of the mean.

[0026] Furthermore, the process indicators are: process moisture: MP_P, process temperature: TEMP_P, process pH: pH_P, process time: Day_P.

[0027] Furthermore, step 5 also includes taking the unified maturity days as input of the optimal prediction model and calculating the thresholds of the four internal maturity days indicators.

[0028] Another aspect of the present invention provides composting using the above compost maturity evaluation method.

[0029] Beneficial effects of the present invention:

[0030] The heterogeneity test is used to check the heterogeneity between data and evaluate the robustness of the results. The publication bias test is used to evaluate whether the results have potential bias affecting the final results in order to ensure the reliability of the Meta-analysis results. The purpose of the one-way analysis of variance (ANOVA) combined with the post hoc least significant difference analysis (LSD) for subgroup grouping significance verification is to ensure whether the subgroup grouping is reasonable and whether it conforms to the data distribution of its own data set. All three analyses are used to ensure the reliability of the Meta-analysis results.

[0031] The present invention uses machine learning to use compost maturity days as a bridge and proposes a compost maturity evaluation method and application, which can provide maturity days for a specific biochar composting system, thereby optimizing process performance, accurately predicting the maturity stage, simplifying the operating steps, and providing decision support for judging compost maturity in production practice.

[0032] The present invention adds composting process indicators (process moisture: MP_P, process temperature: TEMP_P, process pH: pH_P, process time: Day_P) when establishing a machine learning model, and can use a machine learning algorithm to accurately predict changes in indicators during the entire maturation process of the compost, thereby avoiding errors in manual experiments to determine maturity indicators and reducing the time and labor input required for the experiment; at the same time, the final composting time can be reduced by adjusting the composting process indicators during the composting process to achieve rapid composting.

[0033] The data records of the same experiment are grouped and then stratified into training set and test set in a ratio of 8:2, so that the same group of experimental data can be used for training set modeling or for test set model verification at the same time, avoiding the modeling data (training set) and test data from the same group of experimental data, greatly reducing the risk of data leakage and preventing overfitting of machine learning modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 Recommended characteristics of biochar and initial compost are proposed for the forest plot analysis based on the influence of compost maturity indicators. The degree of recommendation: green circle with white font indicates strong recommendation; light green circle with black font indicates moderate recommendation; red circle with yellow font indicates opposite recommendation.

[0035] Figure 2 To rank the importance of key characteristics affecting compost maturity based on structural equation modeling.

[0036] Figure 3 Comparison and establishment of the optimal prediction model for maturity prediction. (a)-(d) are comparisons of RF, XGB, ET, GB and FUSION models. (a) is C / N, (b) is GI, and (c) is NO 3 - -N, (d) is NH 4 + -N; (e)-(h) are the establishment of the optimal prediction model, (e) is C / N, (f) is GI, and (g) is NO 3 - -N, (h) is NH 4 + -N.

[0037] Figure 4Comparison and establishment of the optimal prediction model for maturity prediction. (a)-(d) are comparisons of RF, XGB, ET, GB and FUSION models; (a) is CO 2 , (b) is CH 4 , (c) is N 2 O, (d) is NH 3 ; (e)-(h) are the establishment of the optimal prediction model, (e) is CO 2 , (f) is CH 4 , (g) is N 2 O, (h) is NH 3 .

[0038] Figure 5 is a statistical maturity threshold diagram of the present invention;

[0039] Figure 6 It is the binary clustering of K-means unsupervised clustering algorithm; A-1 to F-1 are the original data; A-2 to F-2 are the shuffled X-axis data; A-3 to F-3 are the shuffled Y-axis data.

[0040] Figure 7 Binary comparison of the importance hierarchy of compost internal maturity indicators based on the K-means unsupervised clustering algorithm.

[0041] Figure 8 Flowchart for uniform assessment of days for compost maturity based on weight distribution.

[0042] Fig. 9 The calculation results of compost maturity days at different biochar addition rates. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] like Figure 8 As shown, a compost maturity evaluation method of the present invention comprises:

[0045] Step 1. First, conduct a Meta-analysis to comprehensively evaluate the intra-heap (C / N, GI, NO 3 - -N, NH 4 + -N) and in vitro (NH 3 , N 2 O,CO 2, CH 4 )Eight maturity indicators were used to screen the key biochar properties and initial compost properties that affect the final composting, providing an optimized data set for machine learning model building.

[0046] Included: By setting the keywords (biochar or black carbon or char) and (maturity or mature) and (composting or compost), excluding review and meta-analysis. The search time was set from January 1, 1990 to January 31, 2024 to collect literature on databases such as Web of Science, Google Scholar, Elsevier Science Direct, and PubMed and screened according to certain screening rules. The screening rules are as follows: First, the inclusion criteria require original research, and quantitative research results show that biochar application leads to changes in compost maturity. Second, each study needs to include a control group (not containing biochar) and an experimental group (containing biochar), consisting of multiple groups for comparison. Third, the selected studies are expected to include biochar characteristics and initial composting. In addition, the mean, standard deviation (SD), and number of control and experimental groups were obtained from the table, or digitized from the chart using the GetDate image digitizer. For studies that only provide standard errors (SE), the formula is used. SD was calculated, where n represents the sample size. For studies that did not provide SD and SE, SD was calculated as 1 / 10 of the mean. Finally, 125 relevant papers were screened, containing a total of 269 sets of valid data for subsequent meta-analysis.

[0047] The heterogeneity test, publication bias test, one-way analysis of variance (ANOVA) combined with post hoc least significant difference analysis (LSD) were used to verify the significance of subgroup grouping to ensure the reliability of the meta-analysis results. The biochar properties and initial compost properties were finally screened out as shown in Table 1, the classification of biochar properties is shown in Table 2, and the classification of initial compost properties is shown in Table 3.

[0048] Table 1 Nomenclature and abbreviations of complete biochar characteristics, initial compost characteristics, and compost maturity indicators

[0049]

[0050]

[0051] Table 2 Subgroup classification and threshold values ​​of biochar properties

[0052]

[0053]

[0054] Table 3 Subgroup classification and threshold values ​​of initial compost properties

[0055]

[0056] Meta Win2.1 software was used to perform forest plots to explore in detail the specific effects of the classified biochar properties and initial compost properties on the eight final compost maturity indicators, and a comprehensive evaluation was conducted to propose the key biochar properties and initial compost properties that affect the final maturity. The average effect size of each classified subgroup was calculated by comparing the experimental groups and experimental groups of each indicator, as well as the corresponding standard deviation and sample size. By comparing the average effect size of the subgroup with the calculated overall effect size, the indicators with better effects in the classified subgroups were judged in turn and recommended, such as Figure 1 The results showed that specific biochar characteristics such as FT (straw), PT (>400), C / N (100-200), PV (<0.02), SA (<100) and AR (>12) had significant positive effects on compost maturity and could be used as recommended indicators for improving compost maturity. In terms of initial compost characteristics, ICN (24-28), WT (sludge) and IMC (55-60) showed good signs of improving compost maturity.

[0057] Recommendation level: A green circle with white font indicates a strong recommendation; a light green circle with black font indicates a general recommendation; a red circle with yellow font indicates the opposite recommendation.

[0058] Based on the results of the forest plot, the influence relationship between biochar properties and compost properties and maturity indicators was analyzed through the structural equation model. The influence of property indicators on maturity indicators was judged and ranked by the standardized total effect of the structural equation model. This verifies the reliability of the forest plot results and reveals the importance of key biochar properties and key compost properties, providing optimization indicators for machine learning models to establish data sets, such as Figure 2 As shown. The key traits affecting compost maturity were ranked. The results showed that biochar property PV had the most significant effect on compost maturity, followed by FT, AR and IMC, WT, IC / N, C / N, IpH, PT and SA.

[0059] Based on the key properties of Meta-analysis, in order to accurately control and achieve rapid composting, composting process indicators (process moisture: MP_P, process temperature: TEMP_P, process pH: pH_P, process time: Day_P) were added when the machine learning model was established, and a data set including 8 maturity indicators was established, including 949 groups of data for predicting C / N, 909 groups of data for predicting GI, and NO 3 - -N predicts 1210 sets of data, NH4 +-N predicts 1204 sets of data, CO 2 Predict 734 sets of data, CH 4 Predict 493 sets of data, N 2 O predicts 451 sets of data, NH 3 Predict 974 sets of data. Select the basic machine learning algorithms extra trees (ET), random forest (RF), gradient boosting (GB), eXtreme Gradient Boosting (XGB), and build an aggregate model (Fusion) of the five models through Python editing. In order to reduce the risk of data leakage, the data records of the same experiment are grouped and then stratified into training sets and test sets in a ratio of 8:2; use Bayesian optimization to perform hyperparameter tuning in R 2 , RMSE, MAE as the criteria for evaluating the algorithm, and screen out the maturity indicators (C / N, GI, NO 3 - -N, NH 4 + -N) and gas index outside the reactor (NH 3 , N 2 O,CO 2 , CH 4 ) algorithm with the best performance; set 10 different random states to evaluate the performance of the ML model and use the R square score (R 2 ), root mean square error (RMSE) and mean absolute error (MAE) to compare the prediction accuracy of the prediction model; output an optimal prediction model with a prediction performance close to the average as an engineering application model, such as Figure 3 , Figure 4 The results show that: in the model comparison process, Figure 3 In (a)-(d), to enhance the robustness of the model, the average performance of the models (across 10 random states that affect the training and test splits) is compared. The best model with a prediction performance close to the average is selected, such as Figure 3 The results show that the GB model performs better in predicting the carbon-nitrogen ratio, while the ET model performs poorly in predicting GI and NO 3 - -N is 0.77, NH 4 + -N is 0.76. In addition, when predicting the four indicators other than composting, such as Figure 4 As shown, the ET model has a significant impact on CH 4 NH 3 and N 2 O emissions; the RF model was poor at predicting CO 2Emissions are better than other models, R 2 They are 0.52, 0.58, 0.33 and 0.48 respectively.

[0060] To ensure the reasonable application of the compost maturity prediction model, the application domain (AD) of the model is established using cosine similarity. When determining the AD, the training set is used as the reference data set and the test set is used as the exploration set. The cosine similarity is used to calculate the similarity between the data in the test set and the data modeled in the training set. Based on this similarity measure, different parts of the test set are formed, and the data below a certain threshold is the data outside the AD, and the data above a certain threshold is the data inside the AD. For each pre-set cosine similarity, a gradient is set to discard the data outside the AD from the test data set and recalculate R 2 , RMSE and the amount of data discarded. The lowest R 2 After the threshold data with the highest RMSE value is obtained, the data outside the applicable range is ensured to be as small as possible. The data greater than the threshold will be identified as the optimal threshold; the Shapley additive interpretation method (SHAP) is used to quantitatively evaluate the contribution of the input feature values ​​to the compost maturity index and the prediction accuracy of the optimal prediction model is verified through experiments.

[0061] Step 2. Determine a statistical maturity threshold that is appropriate for this dataset. Given the urgent need to establish a standardized and applicable maturity evaluation system, the maturity threshold of the compost must first be ensured. Greenhouse gas emission patterns, which are external maturity indicators of compost maturity, generally show consistent trends rather than clear thresholds. Therefore, they are called "alternative indicators of compost maturity" and are not within the scope of our unified threshold analysis. For compost internal maturity indicators, there are large differences in the empirical thresholds of previous studies. For example, C / N ≤ 12, ≤ 20, or ≤ 23 are used to indicate the maturity of the compost. Similarly, GI ≥ 50%, or ≥ 70%, ≥ 80%, NH 4 + -N content ≤ 0.4g / kg or ≤ 0.5g / kg, NO 3 -— Benchmarks such as N content ≥ 0.16 g / kg have also been used to assess the maturity level of compost. Although these empirical thresholds have certain values, they often ignore the different conditions and materials in the composting process in different experimental systems. 3 -- N, NH 4 + The threshold values ​​of internal indicators such as -N are a challenge for the comprehensive evaluation of compost maturity.

[0062] Therefore, in this study, a statistical method was used to define the statistical maturity thresholds of the four internal maturity indicators. If the threshold standard can explain at least 80% of the data, it is considered that the threshold standard replaces the mature empirical threshold in this study. The results show that C / N ≤ 19.10, GI ≥ 83.34%, NH4 + -N≤0.5g / kg, NO 3 - -N≥0.02g / kg is the best statistical maturity threshold representative of this data set, e.g. Figure 5 shown.

[0063] The data sets are sourced from different documents, so it is impossible to guarantee that all data are mature. Therefore, the present invention groups the data records of the same experiment, takes out the data of the last day in different data groups (because the data of the last day of composting in the published articles must be the data of mature composting), and sorts the data sets from small to large (C / N, NH4 + -N) or from large to small (GI, NO 3 - -N) in order. From the four data sets, a value that can cover 80% of the points can be found, which is the statistical maturity threshold in the present invention.

[0064] Step 3. Use the K-means unsupervised clustering algorithm to rank the importance of the four indicators in the compost pile based on the statistical maturity threshold. The K-means clustering algorithm is used to perform binary unsupervised clustering on the two-by-two combinations of the four compost indicators. In this process, the data points are automatically divided into mature and immature clusters. Figure 6 and Figure 7 As shown in Figure 2, the four maturity indicators are ranked by a series of steps. For example, when comparing the ranking between C / N ratio and GI, it is found that the data points with GI>83.34% are mainly clustered in the mature cluster 0, regardless of the random shuffling of C / N ratio or GI values, as shown in Figure 2. Figure 6 In addition, when the GI value is randomly transformed instead of the C / N ratio, the distribution accuracy of the data points with a C / N ratio of less than 19.10 in the mature cluster in cluster 0 drops from 57.58% to 45.02%, and the impact is more obvious. Figure 7 These observations indicate that GI is more representative than C / N ratio in assessing compost maturity and is of vital importance. By carefully comparing the six sets of data, the ranking of the four compost maturity indicators was obtained, and six sets of non-conflicting inequalities were obtained, such as Figure 8 As shown in B: GI (%) > C / N > NO 3 - -N(g / Kg)>NH 4+ -N(g / Kg).

[0065] In addition, the rationality of the ranking (considering whether to put it in or not) was verified by 3D K-means unsupervised clustering through a four-way comparison of compost maturity indicators.

[0066] Step 4. Based on the ranking results according to the importance level, an importance evaluation matrix was established and evaluated using the analytic hierarchy process (AHP). As shown in Table 4, the consistency test results passed, and CR < 0.1. The results showed that among the total indicator weight thresholds, GI dominated, accounting for 47.62%, followed by C / N, accounting for 28.88%, and NO 3 - -N, 17.59% and NH4 + -N, 5.91%.

[0067] Table 4 Using the analytic hierarchy process (AHP) to establish and test the importance evaluation matrix

[0068]

[0069] Step 5. Unify the days for compost maturity evaluation based on the weight distribution of the four indicators in the compost pile and its optimal prediction model. By assigning appropriate weights to the four compost maturity indicators, these indicators are integrated to accurately predict the days for compost maturity evaluation. The first step is to evaluate the input feature values. When the four indicators meet the maturity statistical thresholds (C / N ≤ 19.10, GI ≥ 83.34%, NH 4 + -N≤0.5g / kg, NO 3 - -N≥0.02g / kg), the four indicators in its optimal prediction model are used to calculate the respective maturity days. Then, the allocation weight is multiplied by the respective maturity days to obtain an accurately weighted unified compost maturity day. Then, this unified evaluation day is used as the input of its corresponding optimal prediction model to calculate the thresholds of the four indicators, such as Figure 8 shown.

[0070] Step 6. Apply the theory of uniform maturity to the actual composting process, set up composts with different biochar addition rates, and calculate the uniform maturity, such as Fig. 9 As shown ( Fig. 9Chinese: The entire composting process lasts for 51 days. By adding biochar with different addition rates (CK: 0%; T1: 8%; T2: 15%) to the compost, the model prediction is performed with zero-day data as input and the compost maturity assessment days are calculated. The results show that the unified compost maturity days calculated by the process of the invention are 48 days, 38 days and 44 days, respectively. Adding 8% biochar can significantly shorten the compost maturity time).

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A compost maturity evaluation method, characterized in that , including the following steps: Step 1. Perform a meta-analysis to comprehensively evaluate eight maturity indicators inside and outside the compost, screen the key biochar properties and initial compost properties that affect the final composting, and provide an optimized data set for machine learning to establish the optimal prediction model. The indicators inside the compost are C / N, GI, NO3 - -N, NH4 + -N, the in vitro indicators of the pile are NH3, N2O, CO2, and CH4; Step 2. Determine the statistical maturity threshold suitable for the data set, and use statistical methods to define the statistical maturity thresholds of the four internal maturity indicators; and when the four internal maturity indicators meet the statistical maturity thresholds, use the four indicators in the optimal prediction model of step 1 to calculate their respective maturity days; Step 3. Based on the statistical maturity thresholds of the four internal maturity indicators, the K-means unsupervised clustering algorithm is used to rank the importance of the four internal maturity indicators in the compost pile. The K-means clustering algorithm is used to perform binary unsupervised clustering on the two-by-two combinations of the four compost indicators. In this process, the data points are automatically divided into mature and immature clusters. By calculating the distribution accuracy changes of the cluster points in the mature clusters under the statistical maturity threshold, the axial data are shuffled and re-clustered in a random shuffle manner, and the relative changes in the accuracy of the data points in the mature clusters of the original clusters, the randomly shuffled X-axis clusters, and the randomly shuffled Y-axis clusters are compared to determine the importance ranking of the four internal indicators. Step 4. According to the importance ranking results of the four internal maturity indicators in step 3, the importance evaluation matrix is ​​established by using the hierarchical analysis method, and the evaluation is carried out to determine whether the results are passed and whether they are passed, and the weights of the four internal maturity indicators are obtained; Step 5. Based on the weights of the four internal maturity indicators in step 4, multiply the respective maturity days in step 2 to obtain a weighted unified maturity day.

2. The method according to claim 1, characterized in that Meta-analysis in step 1 includes: S101. By setting keywords, literature was collected and screened in the database, and the subgroup combination was rationally verified through heterogeneity test, publication bias test, one-way analysis of variance combined with post hoc least significant analysis; S102. Use Metawin 2.1 software to perform forest plots to explore the specific effects of biochar properties after subgroup classification and initial compost properties on eight compost maturity indicators, propose key biochar properties and initial compost properties that affect maturity, calculate the average effect size of each classification subgroup through the experimental and control groups of each indicator and the corresponding standard deviation and sample size, compare the subgroup average effect size with the calculated overall effect size considering only whether biochar is added or not, determine the indicators with better effects in the classification subgroups and make recommendations; S103. Based on the results of the forest plot, the influence relationship between biochar properties and compost properties and maturity indicators was analyzed through the structural equation model. The influence of the property indicators on the maturity indicators was determined and ranked through the standardized total effect of the structural equation model, providing optimization indicators for the machine learning model to establish a data set; S104. Add composting process indicators when building the machine learning model, and establish a data set including 8 maturity indicators; select basic machine learning algorithms extra trees, random forest, gradient boosting, and eXtremeGradient Boosting, and build an aggregate model of the five models through Python editing, group the data records of the same experiment, and then stratify them into training sets and test sets in a ratio of 8:2; use Bayesian optimization to perform hyperparameter tuning with R 2 , RMSE, and MAE were used as the criteria for evaluating the hyperparameter optimization algorithm, and the algorithm with the best performance in predicting the maturity index inside the pile and the gas index outside the pile was selected; 10 different random states were set to take the average to evaluate the performance of the five models, and R 2 , RMSE and MAE to compare the prediction accuracy of the five models, and select the best model whose performance is close to the average model for output as the best prediction model.

3. The method according to claim 2, characterized in that The data set establishment includes: using cosine similarity to establish the application domain of the model. When determining the application domain, the training set is used as the reference data set and the test set is used as the exploration set. The cosine similarity is used to calculate the similarity between the data in the test set and the data modeled in the training set. Based on the similarity metric, different parts of the test set are formed. Data below a certain threshold is data outside the application domain, and data above a certain threshold is data within the application domain. For each pre-set cosine similarity, a gradient is set to discard data outside the application domain from the test data set and recalculate R. 2 , root mean square error and the amount of discarded data.

4. The method according to claim 1, characterized in that: The screening rules are as follows: (1) The inclusion criteria require original studies with quantitative research results showing that biochar application leads to changes in compost maturity; (2) Each study needs to include a control group and an experimental group, and multiple groups are required for comparison; (3) The selected studies will include biochar characteristics and initial compost; (4) Obtain the mean, standard deviation, and number of control and experimental groups, or digitize them from the chart using the GetDate image digitizer; (5) For studies that only provide standard errors, use the formula SD was calculated, where n represented the sample size and for studies that did not provide SD and SE, SD was calculated as 1 / 10 of the mean.

5. The method according to claim 1, characterized in that The process indicators are: process moisture: MP_P, process temperature: TEMP_P, process pH: pH_P, process time: Day_P.

6. The method according to claim 1, characterized in that Step 5 also includes using the unified maturity days as input to the optimal prediction model and calculating the thresholds of the four internal maturity days indicators.

7. Use the compost maturity evaluation method as described in claims 1-6 to guide biochar composting.

Images