A regional agricultural product over-standard probability prediction and error analysis method

Abstract

The application relates to a regional agricultural product over-standard probability prediction and error analysis method. The research region is divided into different sub-regions by using existing soil monitoring information and agricultural product monitoring information of the research region, and iterative accurate calculation is carried out in the sub-regions, so that the accuracy and applicability of an agricultural product risk prediction model are improved, the accuracy is greatly improved, the on-site sampling workload can be greatly reduced, and data support is provided for production site environment early warning work.

Description

Technical Field

[0001] This invention belongs to the field of agricultural environmental technology, specifically relating to a method for predicting the probability of regional agricultural products exceeding standards and for error analysis. Background Technology

[0002] Agricultural product pollution is becoming increasingly serious year by year. Analyzing the main causes, identifying key links, and proposing corresponding countermeasures are of great significance for controlling the risk of heavy metal contamination in agricultural products. Heavy metal contamination in agricultural products is characterized by high risk uncertainty, complex influencing factors, and varying physiological mechanisms of accumulation. Currently, risk prediction methods are relatively simplistic and have low accuracy. Furthermore, agricultural product safety is closely related to soil quality, but current soil and agricultural product monitoring is poorly systematic and data utilization is low, with limited research on the response relationship between agricultural products and soil heavy metal content.

[0003] CN113020232A, based on the results of intensive sampling and detection, preliminarily divides the target area into units, further partitions each unit according to its pollution characteristics, selects corresponding remediation and treatment measures, and tracks, evaluates, and dynamically adjusts the treatment effect. CN112883137A discloses a method for intensive investigation and early warning of agricultural product origin safety, including multi-stage and multi-process processes such as calculation of the probability of agricultural products exceeding standards, intensive investigation, classification and early warning, and early warning response. All of the above technologies involve methods of intensive sampling and detection. However, these methods do not involve calculating the area represented by the intensive sampling points based on the detection data of the main influencing factors, thus having a low effectiveness in representing the characteristics of the study area.

[0004] The main problems with the above technical methods are: (1) low comprehensive utilization of soil and agricultural product monitoring data; (2) limited implementation of agricultural product risk prediction work with simple technology, low accuracy and poor applicability; and (3) high workload and high cost in agricultural product risk prediction work. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention discloses a method for predicting the probability of regional agricultural product exceeding standards and for error analysis. The method includes: (1) Main factor screening (1.1) Establish a database of production environment Collect environmental information, historical soil monitoring information, and agricultural product monitoring information of the study area to establish a production area environment database; (1.2) Main factor screening Using agricultural product monitoring index j as the research index, the relative entropy Ei method was used to screen soil monitoring indicators in the study area. The specific formula for calculating the relative entropy Ei is as follows:

[0006] Where: xj is the monitoring value of agricultural product research indicator j; p(xi) is the matrix composed of normalized values ​​of soil monitoring indicators; q(xj) is the matrix composed of normalized values ​​of agricultural product research indicators. The normalization calculation formula is as follows: xi = (Xi - MINi) / (MAXi - MINi) Where Xi is the monitored value of the monitoring indicator, MAXi is the maximum monitored value of the monitoring indicator, and MINi is the minimum monitored value of the monitoring indicator; Based on the relative entropy of each soil monitoring indicator, the relative entropy weight Wi of each indicator is calculated. After sorting the relative entropy weights Wi in descending order, the monitoring indicators with the top Y weights and proportions are selected as the main influencing factors. The specific calculation formula is as follows:

[0007] The value of Y ranges from 0 to 100%; preferably, the value of Y ranges from 55% to 85%. (2) Interpolation error analysis (2.1) Kriging interpolation A suitable Kriging interpolation method was selected to perform interpolation calculations on various influencing factors, and the standard error data of index prediction within each grid was obtained; at the same time, the number of grids in the study area was determined. (2.2) Interpolation error analysis (2.2.1) Determine the standard error range of the prediction of each main influencing factor. Based on whether the single main influencing factor exceeds the standard error S, the study area is divided into two categories: single index conforming area and single index non-conforming area. The single index non-conforming areas of each main influencing factor are spatially merged to obtain the comprehensive index non-conforming area, and the historical soil monitoring points in the area are extracted. The value of S ranges from 0 to 100%; preferably, the value of S ranges from 50% to 85%. (2.2.2) Based on the extracted historical soil point influencing factor monitoring data, the range of each index is calculated using GS+ and ArcGIS software, and the area where the comprehensive index does not match is divided into grids using the minimum range as the side length unit; if there are no historical soil monitoring points in the grid, the grid center point is obtained, and the grid center point is the densified point, and the latitude and longitude information of the densified point is extracted. (2.2.3) Based on the latitude and longitude information of the encrypted points, all soil monitoring information of each point was obtained on site; all soil monitoring information obtained on site was consistent with the content of historical soil monitoring information; (2.3) For areas where the comprehensive indicators do not match, interpolation and error analysis are performed iteratively by combining the monitoring data of various main influencing factors of historical soil monitoring points and densified points in the area. If the research requirements are met, proceed to the next step. The research requirements are that the area of ​​all areas where the comprehensive indicators do not match is less than 30 * the representative area of ​​the largest point in the research area. (3) Error analysis of agricultural product content prediction Error analysis of agricultural product content prediction based on all grid cells in the study area. (3.1) Error Calculation (3.1.1) Historical agricultural product monitoring points exist within the grid. The distribution characteristics of historical agricultural product research indicator monitoring data in the study area were analyzed. Regression analysis was selected to obtain the prediction model corresponding to agricultural product research indicator j. Then, the predicted content values ​​for each historical agricultural product monitoring point were calculated, and the prediction error (Dj) for each point was calculated using relative error. The specific calculation formula is as follows:

[0008] Where cj is the predicted content value of agricultural product research indicator j, and Xj is the monitored value of agricultural product research indicator j. (3.1.2) There are no historical agricultural product monitoring points within the grid. The K-nearest neighbor method is used to select the K historical agricultural product monitoring points closest to the target grid, obtain the agricultural product prediction error of each point and the distance Lj between each point and the target grid, and obtain the corresponding error weight (Wj) for each point based on Lj, and then calculate the agricultural product content prediction error (Cj) of the target grid. The specific calculation formula is as follows:

[0009] Where L1, L2…LK are the distances between the 1st, 2nd…Kth points and the target grid, respectively; Cj=W1* C1 +W2 * C2+W3 * C3+....+WK * CK (3.2) Error Analysis (3.2.1) Determine the content prediction error range of agricultural product research index j. Based on whether it exceeds the content prediction error M, divide the study area into two categories: error conformity area and error non-conformity area. Extract historical soil monitoring points in the error non-conformity area. The value of M ranges from 0 to 100%; preferably, the value of M ranges from 15% to 50%. (3.2.2) Based on the extracted historical soil monitoring point impact main factor monitoring data, variable range calculation is performed and collaborative densification points are obtained. The calculation process is the same as (2.2.2). (3.2.3) Based on the information from the collaborative encryption points, collect and test soil and agricultural product samples, and record the monitoring information; The soil and agricultural product monitoring information is consistent with the monitoring information in the database; (3.3) For areas with discrepancies in error, the main causes are screened and error analysis is performed iteratively by combining the monitoring data of historical soil monitoring points and collaborative densification points in the area. If the research requirements are met, the next step is carried out. The research requirements are that the prediction error of agricultural products in the areas with discrepancies in error is less than 15%. (4) Identification of potential areas with excessive levels of agricultural products Based on the agricultural product content prediction model, the agricultural product content in each grid is calculated. According to the limit value of research indicator j in the national standard GB2762, the agricultural products are divided into three categories: exceeding the limit, not exceeding the limit, and potentially exceeding the limit. The evaluation result distribution map is output.

[0010] Furthermore, the environmental information includes, but is not limited to, regional area, surrounding environment, pollution source information, etc.; historical soil monitoring information includes: soil monitoring indicators, monitoring date, fertilizer application amount, inputs, representative area of ​​the monitoring point, etc.; agricultural product monitoring information includes: agricultural product monitoring indicators, monitoring date, planting system, agricultural product type, annual yield, etc. Furthermore, the agricultural product research indicator j is one or more of the agricultural product monitoring indicators; Furthermore, the soil monitoring indicators include one or more of Cd, Hg, As, Pb, Cr, pH, CEC, SOM, Cu, Zn, Ni, Mg, Fe, Mn, Ca, Si, Se, K, N, and P; Furthermore, the monitoring indicators for agricultural products include one or more of Cd, Hg, As, Pb, and Cr; Furthermore, the Kriging interpolation methods include, but are not limited to: ordinary Kriging, simple Kriging, generalized Kriging, classical Bayesian Kriging, indicator Kriging, and probabilistic Kriging; Furthermore, the kriging interpolation is performed using ArcGIS software; Furthermore, the regression methods include, but are not limited to: linear regression, ridge regression, logarithmic regression, multiple regression, and stepwise regression. Furthermore, K is set to 3-5; Furthermore, the category of exceeding the limit is defined as the content value of research indicator j being greater than 1.1 times the limit value, the category of not exceeding the limit is defined as the content value of research indicator j being less than or equal to 0.9 times the limit value, and the category of potential exceeding the limit is defined as the content value of research indicator j being between 0.9 times the limit value and 1.1 times the limit value. Furthermore, the number of soil influence principal grids is consistent across all interpolation methods.

[0011] The present invention provides a method for predicting and analyzing the probability of regional agricultural products exceeding standards, which has the following advantages: 1. This invention utilizes existing soil monitoring information and agricultural product monitoring information in the study area to divide the study area into different sub-regions, and performs iterative and precise calculations within the sub-regions to improve the accuracy and applicability of the agricultural product risk prediction model. While significantly improving the accuracy rate, it can greatly reduce the workload of on-site sampling and provide data support for early warning work of the production area environment. 2. This invention can use a variety of prediction methods, and the accuracy of the prediction results is relatively similar, which reduces the professional quality requirements of the staff; 3. This invention considers both the morphological and spatial distribution of soil data, increasing the accuracy of predicting agricultural products and soil data within each grid, and saving a significant amount of working time and financial investment compared to normal monitoring work. 4. Based on existing monitoring data, this invention selects the most suitable method for denser monitoring to complete the denser monitoring work. Combined with steps such as predicting the error of agricultural product content and identifying potential areas of agricultural product exceeding standards, it can quickly identify potential areas of exceeding standards. On this basis, it provides guidance for regional planning and ecological restoration, greatly reducing monitoring costs while improving the accuracy of environmental early warning work. Attached Figure Description

[0012] Figure 1 A technical roadmap for predicting the probability of agricultural products exceeding standards in a region and analyzing errors. Figure 2 Map showing the distribution of historical monitoring points in County T; Figure 3 This is a graph showing the relative entropy results of soil monitoring indicators; Figure 4 This is a statistical distribution chart of the range; Figure 5 Distribution map of encryption points; Figure 6 Interpolation error analysis results; Figure 7 Distribution map of areas with discrepancies in Cd content in rice; Figure 8 Risk distribution map of Cd content in rice; Detailed Implementation

[0013] The present invention will be further described in detail below through embodiments, so that those skilled in the art can implement it based on the description.

[0014] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0015] Example 1 1. Main Factor Screening ① Establishment of a production environment database: Farmland within County T was selected as the research object ( Figure 2 The study collected environmental information, historical soil and agricultural product (all rice) monitoring data for the area. Soil monitoring indicators included 28 items: Cd, Hg, As, Pb, Cr, pH, CEC, SOM, P, K, N, Cu, Zn, Ni, Al, Si, Fe, Mn, Ca, Mg, Ce, La, Mo, Sr, Se, Co, Ti, and Na. Rice monitoring indicators included 5 items: Cd, Hg, As, Pb, and Cr. The representative area range for each monitoring point was defined as 25-150 mu / point. ② Using rice cadmium (Cd) as the research indicator, the relative entropy of various soil monitoring indicators was calculated. After sorting the indicators by weight from largest to smallest, the indicators with the highest relative entropy information content were selected as the main factors. These included 13 items: Pb, Cd, SOx, N, Hg, Ni, Zn, pH, As, CEC, Cr, Cu, and K. (See below) Figure 3 ; 2. Interpolation Error Analysis ① Using ArcGIS software, ordinary Kriging interpolation was selected to interpolate the 13 main influencing factors, and a total of 67,017 grids were determined for County T; the prediction standard error range of the main influencing factors for each grid was obtained. ② Using 75% of the standard error range of each main influencing factor as the limit, the area where the standard error of the prediction is greater than the limit is defined as the single index mismatch area; the mismatch areas of the 13 main influencing factors are spatially superimposed, and the area is the comprehensive index mismatch area. ③ 187 historical soil monitoring points in areas where the comprehensive indicators did not match were extracted, and GS was used to analyze them. + The software calculated the range of each major influencing factor, with Cd having the smallest range at 2400 meters. Figure 4 ); ④ Using ArcGIS software with a side length of 2400 meters, the areas where the comprehensive indicators do not meet the requirements are divided into grids, and 131 grid center points are extracted as densification points. Figure 5 The latitude and longitude information of each point was obtained, and the X-ray fluorescence method was used to quickly obtain all soil monitoring information of each point on site. ⑤ Combining the main influencing factors of the densified monitoring points and historical soil monitoring points in areas where the comprehensive indicators do not meet the requirements, repeat steps ①-④ until all areas in T County where the comprehensive indicators do not meet the requirements are less than 4,500 mu. After repeating the error analysis three times, a final 305 soil monitoring points were obtained. Figure 6 ); 3. Error Analysis of Agricultural Product Content Prediction ① Based on the soil impact data and rice Cd data from all historical monitoring points, linear regression analysis was performed using SPSS software, and the relevant regression model is as follows: Cd (水稻) =0.847+0.007Pb+0.320Cd+0.007SOM-0.002N+0.160Hg+0.008Ni- 0.002Zn-0.112pH-0.004As+0.004CEC-0.001Cr-0.001Cu-0.001K ② The Cd content corresponding to the historical rice monitoring points was calculated based on the prediction model, and the corresponding rice content prediction error was calculated according to the formula; ③ Select the content prediction error of the three nearest historical rice sites, calculate the rice content prediction error in each grid, and obtain the rice content prediction error distribution map. ④ Based on the rice content prediction error distribution map, obtain the monitoring values ​​of the main influencing factors of all soil monitoring points in the region, calculate the range of each index using SPSS (minimum value is 1600 meters), and obtain 45 collaborative densification points accordingly, and simultaneously obtain the monitoring information of rice and soil in the current season (the process is the same as step 2). ⑤ Based on the rice monitoring data and soil monitoring data obtained above, repeat steps ①-④ until the predicted error of rice Cd content in all areas with discrepancies in T County is less than 15%; after one repetition, the maximum predicted error of rice Cd content in areas with discrepancies is 10.4%. Figure 7 ); 4. Identification of potential areas with excessive levels of pesticide residues in agricultural products. The latest national standard GB2762-2017 (National Food Safety Standard: Limits of Contaminants in Food) specifies a limit of 0.2 mg / kg for Cd in rice. Based on a rice content prediction model, the Cd content in each grid cell is calculated to have limits of 0.18 mg / kg and 0.22 mg / kg. The Cd content in rice is then divided into three categories, and a distribution map is drawn. Figure 8 Among them, the potential rice exceeding the standard risk area is the potential exceeding standard area.

[0016] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and embodiments shown and described herein.

Claims

1. A method for predicting the probability of regional agricultural products exceeding standards and analyzing errors, characterized in that, The method includes: The method includes: (1) Main factor screening (1.1) Establish a database of production environment (1.2) Main factor screening Using agricultural product monitoring index j as the research index, the relative entropy Ei method was used to screen soil monitoring indicators in the study area. The specific formula for calculating the relative entropy Ei is as follows: ； Where: xj is the monitoring value of agricultural product research indicator j; p(xi) is the matrix composed of normalized values ​​of soil monitoring indicators; q(xj) is the matrix composed of normalized values ​​of agricultural product monitoring indicators. The normalization calculation formula is as follows: xi = (Xi - MINi) / (MAXi - MINi); Where Xi is the monitoring value of the soil monitoring index, MAXi is the maximum monitoring value of the monitoring index, and MINi is the minimum monitoring value of the monitoring index; Based on the relative entropy of each soil monitoring indicator, the relative entropy weight Wi of each indicator is calculated. After sorting the relative entropy weights Wi in descending order, the monitoring indicators with the top Y weights and proportions are selected as the main influencing factors. The specific calculation formula is as follows: ； The value of Y ranges from 0 to 100%; (2) Interpolation error analysis (2.1) Kriging interpolation A suitable Kriging interpolation method was selected to perform interpolation calculations on various influencing factors, and the standard error data of index prediction within each grid was obtained; at the same time, the number of grids in the study area was determined. (2.2) Interpolation error analysis (2.2.1) Determine the standard error range of the prediction of each main influencing factor. Based on whether the single main influencing factor exceeds the standard error S, the study area is divided into two categories: single index conforming area and single index non-conforming area. The single index non-conforming areas of each main influencing factor are spatially merged to obtain the comprehensive index non-conforming area, and the historical soil monitoring points in the area are extracted. The value of S ranges from 0 to 100%; (2.2.2) Based on the extracted historical soil point influencing factor monitoring data, the range of each index is calculated using GS+ and ArcGIS software, and the area where the comprehensive index does not match is divided into grids using the minimum range as the side length unit; if there are no historical soil monitoring points in the grid, the grid center point is obtained, and the grid center point is the densified point, and the latitude and longitude information of the densified point is extracted. (2.2.3) Based on the latitude and longitude information of the encrypted points, all soil monitoring information of each point was obtained on site; all soil monitoring information obtained on site was consistent with the content of historical soil monitoring information; (2.3) For areas where the comprehensive indicators do not match, interpolation and error analysis are performed iteratively by combining the monitoring data of various main influencing factors of historical soil monitoring points and densified points in the area. If the research requirements are met, proceed to the next step. The research requirements are that the area of ​​all areas where the comprehensive indicators do not match is less than 30 * the representative area of ​​the largest point in the research area. (3) Error analysis of agricultural product content prediction Error analysis of agricultural product content prediction based on all grid cells in the study area. (3.1) Error Calculation (3.1.1) Historical agricultural product monitoring points exist within the grid. The distribution characteristics of historical agricultural product research indicator monitoring data in the study area were analyzed. Regression analysis was selected to obtain the prediction model corresponding to agricultural product research indicator j. Then, the predicted content values ​​for each historical agricultural product monitoring point were calculated, and the prediction error Dj for each point was calculated using relative error. The specific calculation formula is as follows: ； Where cj is the predicted content value of agricultural product research indicator j, and Xj is the monitored value of agricultural product research indicator j. (3.1.2) There are no historical agricultural product monitoring points within the grid. The K-nearest neighbor method is used to select the K historical agricultural product monitoring points closest to the target grid, obtain the agricultural product prediction error of each point and the distance Lj between each point and the target grid, and obtain the corresponding error weight Wj for each point based on Lj. Then, the prediction error Cj of agricultural product content of the target grid is calculated. The specific calculation formula is as follows: ； L1, L2…LK are the distances between the 1st, 2nd…Kth points and the target grid, respectively; Cj = W1 * C1 + W2 * C2 + W3 * C3 + ... + WK * CK (3.2) Error Analysis (3.2.1) Determine the content prediction error range of agricultural product research index j. Based on whether it exceeds M% of the content prediction error, divide the study area into two categories: error conformity area and error non-conformity area. Extract historical soil monitoring points in the error non-conformity area. The value of M ranges from 0 to 100%; (3.2.2) Based on the extracted historical soil monitoring point impact main factor monitoring data, variable range calculation is performed and collaborative densification points are obtained. The calculation process is the same as (2.2.2). (3.2.3) Based on the information from the collaborative encryption points, collect and test soil and agricultural product samples, and record the monitoring information; The soil and agricultural product monitoring information is consistent with the monitoring information in the database; (3.3) For areas with discrepancies in error, the main causes are screened and error analysis is performed iteratively by combining the monitoring data of historical soil monitoring points and collaborative densification points in the area. If the research requirements are met, the next step is carried out. The research requirements are that the prediction error of agricultural products in the areas with discrepancies in error is less than 15%. (4) Identification of potential areas with excessive levels of agricultural products Based on the agricultural product content prediction model, the agricultural product content in each grid is calculated. According to the limit value of research indicator j in the national standard GB2762, the agricultural products are divided into three categories: exceeding the limit, not exceeding the limit, and potentially exceeding the limit. The evaluation result distribution map is output. The number of K values ​​is set to 3-5. The categories of exceeding the limit are those where the content value of research indicator j is greater than 1.1 times the limit value, those not exceeding the limit are those where the content value of research indicator j is less than or equal to 0.9 times the limit value, and those potentially exceeding the limit are those where the content value of research indicator j is between 0.9 times and 1.1 times the limit value.

2. The method for predicting and analyzing the probability of regional agricultural product exceeding standards as described in claim 1, characterized in that, The origin environment database collects environmental information, historical soil monitoring information, and agricultural product monitoring information for the study area.

3. The method for predicting the probability of regional agricultural products exceeding standards and analyzing errors as described in claim 2, characterized in that, The environmental information includes the area, surrounding environment, and pollution source information.

4. The method for predicting and analyzing the probability of regional agricultural product exceeding standards as described in claim 2, characterized in that, Historical soil monitoring information includes: soil monitoring indicators, monitoring date, fertilizer application amount, inputs, and representative area of ​​the monitoring point; agricultural product monitoring information includes: agricultural product monitoring indicators, monitoring date, planting system, agricultural product type, and annual yield.

5. The method for predicting and analyzing the probability of regional agricultural product exceeding standards as described in claim 4, characterized in that, The soil monitoring indicators include one or more of the following: Cd, Hg, As, Pb, Cr, pH, CEC, SOM, Cu, Zn, Ni, Mg, Fe, Mn, Ca, Si, Se, K, N, and P.

6. The method for predicting and analyzing the probability of regional agricultural product exceeding standards as described in claim 4, characterized in that, The monitoring indicators for agricultural products include one or more of Cd, Hg, As, Pb, and Cr.

7. The method for predicting and analyzing the probability of regional agricultural product exceeding standards as described in claim 1, characterized in that, The Kriging interpolation methods include: ordinary Kriging, simple Kriging, generalized Kriging, classical Bayesian Kriging, indicator Kriging, and probabilistic Kriging.

8. The method for predicting and analyzing the probability of regional agricultural product exceeding standards as described in claim 1, characterized in that, The regression analysis methods are: linear regression, ridge regression, logarithmic regression, multiple regression, and stepwise regression.

Images