A system and method for building a large regulatory model based on artificial intelligence

Abstract

A supervision large model construction system based on artificial intelligence comprises: an original data module, a model training module, a commodity sampling module, a supervision positioning module and an intelligent training module. The original data module is used to obtain initial training data of the model. The model training module is used to simulate commodity diffusion and build a commodity distribution model. The commodity sampling module is used to sample problematic commodities and reference commodities. The supervision positioning module is used to determine the diffusion direction and commodity source center of problematic commodities. The intelligent training module is used to perform directional clustering of diffusion bias and iterative training of the distribution model. The present invention can simplify the construction process of the commodity diffusion model, reduce the degree of dependence of the supervision model on the original data, improve the accuracy and reliability of the model prediction, and help the supervision department to locate the sellers of problematic commodities more accurately and quickly, thereby improving the speed and accuracy of market supervision.

Description

Technical Field

[0001] The present invention relates to the field of large model construction, and specifically to a supervision large model construction system and method based on artificial intelligence. Background Art

[0002] Regulatory big models refer to large data models used by regulatory agencies or supervisory departments to monitor, evaluate and predict market activities, risks and compliance, thereby identifying abnormal transactions or problematic commodities in the market. The establishment of regulatory big models generally requires steps such as model selection, data training and model deployment, so that the model is applicable to the areas that need to be regulated and the problems that need to be solved.

[0003] When detecting the source of problematic goods in the market, merchants selling problematic goods are generally located by inquiring or retrieving transaction records. However, problematic goods often take a long time to be discovered, and customers have numerous transaction records within a period of time, making it impossible to accurately find the source of problematic goods. The existing large-scale model supervision system locates the source of goods by analyzing the distribution of problematic goods, but is limited by factors such as urban terrain and regional purchasing preferences, and cannot be accurately applied to the traceability process of specific locations.

[0004] In addition, the large regulatory model is obtained by matching the model with training data, which is prone to become dependent on the original training data. When faced with market changes, the model's prediction error is large and it is difficult to adapt to the complex market environment and changing regulatory needs. Summary of the invention

[0005] The purpose of the present invention is to provide a system and method for building a large regulatory model based on artificial intelligence to solve the problems raised in the above-mentioned background technology.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a supervision large model construction system based on artificial intelligence, comprising: an original data module, a model training module, a commodity sampling module, a supervision positioning module and an intelligent training module;

[0007] The raw data module is used to calculate the probability distribution function of the commodity distribution distance according to the usage characteristics of each commodity, mark the business locations of all merchants and the sales ratio of each commodity on the electronic map, and predict the commodity sales volume according to the flow of people and historical transactions at the business location, and store the historical commodity sampling data as the initial training data of the model in the database;

[0008] The model training module is used to use the distribution distance of the commodity as the scatter radius and the sales volume as the scatter density, use the intelligent classification model to simulate the commodity diffusion, build a global commodity scatter model according to the simulation results, input the initial training data into the commodity scatter model, reconstruct the attention function of the commodity scatter model, and output a multi-dimensional expansion matrix of each commodity monitoring point;

[0009] The commodity sampling inspection module is used to perform sampling inspection in an area with a preset monitoring length as a radius with the discovery location as the center after discovering the problematic commodity, mark the discovery locations of all problematic commodities, and randomly select a commodity with a scatter probability higher than a threshold as a reference commodity, input the sampling inspection result into the commodity scatter model, and obtain the density stream function of the problematic commodity within the sampling inspection radius after processing with the multi-dimensional expansion matrix at the commodity discovery location;

[0010] The supervision and positioning module is used to calculate the partial derivative values ​​of the density stream function in each direction, and the direction pointed to when the partial derivative value is the largest is taken as the first diffusion direction of the problematic product. The maximum distribution distance is moved forward along the first diffusion direction to conduct another random inspection. The maximum distribution distance is the product distribution distance corresponding to the maximum probability. The results of the two random inspections are input into the product distribution model to obtain the distribution center area of ​​the product, and the probability that the product comes from each merchant is given;

[0011] The intelligent training module is used to detect the source of the problematic goods, and then adjust the original diffusion model according to the deviation distance between the inferred source and the actual source, the actual sampling results within the sampling radius, and the sampling results of the reference goods during the sampling process, and perform targeted clustering on the diffusion bias area to output the iterative dispersion model.

[0012] Further, the original data module includes: a commodity feature unit and a business map unit;

[0013] The commodity feature unit is used to calculate the probability distribution of the distance between the commodity and the purchase source according to the usage characteristics of the commodity, wherein the usage characteristics include: the weight, volume, shelf life and usage location of the commodity;

[0014] The business map unit is used to mark business operators and predict the sales volume of various commodities of the merchants.

[0015] Further, the model training module includes: a pattern recognition unit and a diffusion modeling unit;

[0016] The pattern recognition unit is used to perform commodity diffusion simulation according to a preset diffusion distance, diffusion center and diffusion density;

[0017] The diffusion modeling unit is used to train a feedforward neural network using initial training data, and input the diffusion simulation results into the trained feedforward neural network to obtain a commodity distribution model.

[0018] Furthermore, the commodity sampling module includes: a density simulation unit, a reference sampling unit and a model adjustment unit;

[0019] The density simulation unit is used to generate a density stream function of the problematic commodity within the sampling radius according to the sampling inspection result;

[0020] The reference sampling unit is used to sample the problematic goods and the reference goods within a sampling radius of the problematic goods center;

[0021] The model adjustment unit is used to analyze the distribution density of the problem commodity and the reference commodity in the commodity distribution model and adjust the model.

[0022] Further, the supervision and positioning module includes: a supplementary sampling unit and a central positioning unit;

[0023] The supplementary sampling unit is used to determine the first diffusion direction of the problematic commodity and conduct another sampling in the first diffusion direction;

[0024] The central positioning unit is used to speculate the source of the problematic product and provide the probability of the source being each merchant.

[0025] Furthermore, the intelligent training module includes: a diffusion distortion unit, an AI simulation unit and a model iteration unit;

[0026] The diffusion distortion unit is used to calculate the deviation distance between the inferred source center and the actual source center, and the statistical deviation of the sampling result;

[0027] The AI ​​simulation unit is used to use the AI ​​clustering model to perform directional clustering on the diffusion bias area to obtain the distortion direction and distortion rate of the dispersion model;

[0028] The model iteration unit is used to iteratively reconstruct the model according to the distortion of the dispersion model in this tracing, and update the training data in the database.

[0029] A method for constructing a large regulatory model based on artificial intelligence, comprising the following steps:

[0030] Step S1. According to the usage characteristics of each type of goods, the distribution of the goods is simulated to obtain the probability distribution function of the distribution distance, the business operators are marked, and the sales volume of each type of goods by the business operators is predicted based on the historical sales records;

[0031] Step S2. Taking the business operator as the center point, the distribution distance of the goods as the scatter radius, and the sales volume of the goods as the scatter density, a global goods scatter model is constructed, and the initial training data is input into the goods scatter model to reconstruct the attention function of the goods scatter model;

[0032] Step S3. After the problematic product is found, a random inspection is conducted within the inspection radius with the discovery location as the center, and products with a distribution probability higher than the threshold are randomly selected as reference products for simultaneous inspection, marking the type, inspection location and inspection results of all products;

[0033] Step S4. Use the AI ​​classification model to classify the sources of the problematic products and reference products according to their distribution density. Output the density stream function of the problematic products from a single source within the sampling radius based on the classification result, and cluster the diffusion directions of each source based on the direction pointed when the partial derivative of the function is the largest. Determine the source of the problematic products based on the distance between each merchant and the clustering result.

[0034] Step S5. After the source merchant is detected, the deviation direction and distortion rate of the distribution simulation result are fitted based on the difference between the actual distribution state and the simulated distribution state of the problematic product, and the model is iteratively reconstructed to update the training data in the database.

[0035] Further, step S1 includes:

[0036] Step S11. Simulate the distribution of the goods according to the usage characteristics of the goods to obtain a probability distribution function f(r) of the distribution distance of the goods, where r represents the distribution distance of the goods. The usage characteristics include: weight, volume, shelf life and usage location of the goods. Tools for performing distribution simulation include: Monte Carlo simulation, Copula model simulation and Bootstrap simulation;

[0037] Step S12. Mark the business on the electronic map, obtain the business's historical commodity sales function H(d), where d represents the date, and predict the business's sales on the current date:

[0038]

[0039] Among them, Q represents the sales volume of the merchant on the current date, D0 represents the maximum number of days of historical sales records, v represents the preset traffic flow change index, and H(i) represents the sales volume of the i-th day in the historical records.

[0040] Further, step S2 includes:

[0041] Step S21. Make a radial commodity distribution model with the business as the center point. The distribution distance between the distribution points and the center point is randomly generated according to the probability distribution function f(r). The total distribution amount is set to Q. The commodity distribution models of all businesses are marked in space with different dimensions to generate a global commodity distribution model.

[0042] Step S22. Track some of the goods sold by the business to obtain initial training data, use the initial training data to train the feedforward neural network to obtain the distribution trend of the goods, weight the attention coefficient of the product distribution model in the direction of the distribution trend, and reconstruct the attention function of the product distribution model.

[0043] Further, step S3 includes:

[0044] Step S31. After the problem product is detected for the first time, random product sampling is performed with the location where the problem product is found as the center and the sampling distance as the radius until w problem products are sampled. The sampling distance is determined by the simulation accuracy of the product distribution model, w is a preset value, and W>1;

[0045] Step S32. Select reference commodities for random inspection. The reference commodities are other types of commodities randomly selected by the commodity distribution model in the random inspection area. The type of the reference commodities selected is greater than or equal to 1. Mark the types, random inspection locations and random inspection results of all random inspection commodities in the random inspection area.

[0046] Further, step S4 includes:

[0047] Step S41. Using the simulation results of the distribution of problem commodities and reference commodities in the commodity distribution model, the commodity operation ratio of the business operators and the commodity sampling results as the classification basis, the AI ​​classification model is used to classify the sources of the sampled problem commodities. The AI ​​classification models used include: decision tree model, ensemble learning model, logistic regression model and K nearest neighbor algorithm model;

[0048] Step S42. For each source of problematic goods, generate a density flow function F(x,y) of the goods distribution, where F(x,y) satisfies F(x0,y0)=h, Where x0 and y0 represent the horizontal and vertical coordinates of the location of the problematic product, h is the preset height coefficient, and g() is the preset fading function;

[0049] Step S43. Calculate the partial derivatives of the density stream function F(x,y) in each direction:

[0050]

[0051] Where T lx represents the partial derivative of F(x,y) in the direction of a·x+b·y=0. a and b are the directional coefficients of the horizontal and vertical coordinates respectively. Adjust the directional coefficients so that T lx The maximum value is obtained. At this time, the direction pointed by a·x+b·y=0 is recorded as the diffusion direction of the problematic product;

[0052] Step S44. Cluster the diffusion directions of problematic commodities from all sources to obtain clustering directions. Use the commodity distribution model to detect the points in the clustering directions where the distribution density of problematic commodities converges to 0. Use the obtained points as clustering centers and conduct market investigations in order of the distances between each business operator and the clustering center.

[0053] Further, step S5 includes:

[0054] Step S51. After the source merchant of the problematic product is detected, the radial distribution in the product distribution model is compared with the location of the source merchant and the scatter angle and scatter density of the problematic product in the sampling radius, and the deviation of the scatter direction and directional distortion rate between the radial distribution model and the actual product distribution is calculated;

[0055] Step S52: Reconstruct the commodity distribution model according to the deviation, and store the sampling results in the training database of the commodity distribution model.

[0056] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0057] 1. The present invention can calculate the distribution distance of commodities according to commodity characteristics, evaluate the distribution density of commodities according to the sales characteristics of merchants, simulate commodity diffusion using artificial intelligence models, and construct a commodity diffusion model according to the simulation results. This can simplify the construction process of the commodity diffusion model, increase the transparency of market information, and improve the accuracy and reliability of model predictions.

[0058] 2. When problematic commodities are detected, the present invention enables the commodity diffusion model to output the density flow function of the problematic commodities within a fixed sampling radius, compares it with the actual sampling results, corrects the density flow function, and determines the diffusion center of the problematic commodities based on the partial derivatives of the density flow function in each direction, which helps to locate the sellers of the problematic commodities more accurately and quickly, and improve the speed and accuracy of market supervision.

[0059] 3. After the present invention is able to detect the source of problematic goods, it adjusts the original diffusion model based on the deviation distance between the inferred source and the actual source, the actual sampling results within the sampling radius, and the sampling results of the reference goods during the sampling process, and conducts targeted clustering of diffusion-biased areas to form a diffusion model suitable for specific areas, which helps to reduce the degree of dependence of the regulatory model on the original data, improve the efficiency of market supervision, and improve the quality of regional goods. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0061] Figure 1 It is a structural schematic diagram of a supervision large model construction system based on artificial intelligence of the present invention;

[0062] Figure 2 It is a schematic diagram of the steps of a method for building a large supervision model based on artificial intelligence in the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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.

[0064] See also Figure 1 ,The present invention provides a technical solution: a supervision large model construction system based on artificial intelligence, including: an original data module, a model training module, a commodity sampling module, a supervision positioning module and an intelligent training module;

[0065] The raw data module is used to calculate the probability distribution function of the commodity distribution distance according to the usage characteristics of each commodity, mark the business locations of all merchants and the sales ratio of each commodity on the electronic map, and predict the commodity sales volume according to the flow of people and historical transactions at the business location, and store the historical commodity sampling data as the initial training data of the model in the database;

[0066] The original data module includes: a commodity feature unit and a business map unit;

[0067] The commodity feature unit is used to calculate the probability distribution of the distance between the commodity and the purchase source according to the usage characteristics of the commodity, wherein the usage characteristics include: the weight, volume, shelf life and usage location of the commodity;

[0068] The business map unit is used to mark business operators and predict the sales volume of various commodities of the merchants.

[0069] The model training module is used to use the distribution distance of the commodity as the scatter radius and the sales volume as the scatter density, use the intelligent classification model to simulate the commodity diffusion, build a global commodity scatter model according to the simulation results, input the initial training data into the commodity scatter model, reconstruct the attention function of the commodity scatter model, and output a multi-dimensional expansion matrix of each commodity monitoring point;

[0070] The model training module includes: a pattern recognition unit and a diffusion modeling unit;

[0071] The pattern recognition unit is used to perform commodity diffusion simulation according to a preset diffusion distance, diffusion center and diffusion density;

[0072] The diffusion modeling unit is used to train a feedforward neural network using initial training data, and input the diffusion simulation results into the trained feedforward neural network to obtain a commodity distribution model.

[0073] The commodity sampling inspection module is used to perform sampling inspection in an area with a preset monitoring length as a radius with the discovery location as the center after discovering the problematic commodity, mark the discovery locations of all problematic commodities, and randomly select a commodity with a scatter probability higher than a threshold as a reference commodity, input the sampling inspection result into the commodity scatter model, and obtain the density stream function of the problematic commodity within the sampling inspection radius after processing with the multi-dimensional expansion matrix at the commodity discovery location;

[0074] The commodity sampling module includes: a density simulation unit, a reference sampling unit and a model adjustment unit;

[0075] The density simulation unit is used to generate a density stream function of the problematic commodity within the sampling radius according to the sampling inspection result;

[0076] The reference sampling unit is used to sample the problematic goods and the reference goods within a sampling radius of the problematic goods center;

[0077] The model adjustment unit is used to analyze the distribution density of the problem commodity and the reference commodity in the commodity distribution model and adjust the model.

[0078] The supervision and positioning module is used to calculate the partial derivative values ​​of the density stream function in each direction, and the direction pointed to when the partial derivative value is the largest is taken as the first diffusion direction of the problematic product. The maximum distribution distance is moved forward along the first diffusion direction to conduct another random inspection. The maximum distribution distance is the product distribution distance corresponding to the maximum probability. The results of the two random inspections are input into the product distribution model to obtain the distribution center area of ​​the product, and the probability that the product comes from each merchant is given;

[0079] The supervision and positioning module includes: a supplementary sampling unit and a central positioning unit;

[0080] The supplementary sampling unit is used to determine the first diffusion direction of the problematic commodity and conduct another sampling in the first diffusion direction;

[0081] The central positioning unit is used to speculate the source of the problematic product and provide the probability of the source being each merchant.

[0082] The intelligent training module is used to detect the source of the problematic goods, and then adjust the original diffusion model according to the deviation distance between the inferred source and the actual source, the actual sampling results within the sampling radius, and the sampling results of the reference goods during the sampling process, and perform targeted clustering on the diffusion bias area to output the iterative dispersion model.

[0083] The intelligent training module includes: a diffusion distortion unit, an AI simulation unit and a model iteration unit;

[0084] The diffusion distortion unit is used to calculate the deviation distance between the inferred source center and the actual source center, and the statistical deviation of the sampling result;

[0085] The AI ​​simulation unit is used to use the AI ​​clustering model to perform directional clustering on the diffusion bias area to obtain the distortion direction and distortion rate of the dispersion model;

[0086] The model iteration unit is used to iteratively reconstruct the model according to the distortion of the dispersion model in this tracing, and update the training data in the database.

[0087] like Figure 2 As shown, a method for constructing a large supervision model based on artificial intelligence includes the following steps:

[0088] Step S1. According to the usage characteristics of each type of goods, the distribution of the goods is simulated to obtain the probability distribution function of the distribution distance, the business operators are marked, and the sales volume of each type of goods by the business operators is predicted based on the historical sales records;

[0089] Step S1 includes:

[0090] Step S11. Simulate the distribution of the goods according to the usage characteristics of the goods to obtain a probability distribution function f(r) of the distribution distance of the goods, where r represents the distribution distance of the goods. The usage characteristics include: weight, volume, shelf life and usage location of the goods. Tools for performing distribution simulation include: Monte Carlo simulation, Copula model simulation and Bootstrap simulation;

[0091] Step S12. Mark the business on the electronic map, obtain the business's historical commodity sales function H(d), where d represents the date, and predict the business's sales on the current date:

[0092]

[0093] Among them, Q represents the sales volume of the merchant on the current date, D0 represents the maximum number of days of historical sales records, v represents the preset traffic flow change index, and H(i) represents the sales volume of the i-th day in the historical records.

[0094] Step S2. Taking the business operator as the center point, the distribution distance of the goods as the scatter radius, and the sales volume of the goods as the scatter density, a global goods scatter model is constructed, and the initial training data is input into the goods scatter model to reconstruct the attention function of the goods scatter model;

[0095] Step S2 includes:

[0096] Step S21. Make a radial commodity distribution model with the business as the center point. The distribution distance between the distribution points and the center point is randomly generated according to the probability distribution function f(r). The total distribution amount is set to Q. The commodity distribution models of all businesses are marked in space with different dimensions to generate a global commodity distribution model.

[0097] Step S22. Track some of the goods sold by the business to obtain initial training data, use the initial training data to train the feedforward neural network to obtain the distribution trend of the goods, weight the attention coefficient of the product distribution model in the direction of the distribution trend, and reconstruct the attention function of the product distribution model.

[0098] Step S3. After the problematic product is found, a sampling inspection is conducted in an area within a sampling inspection radius centered on the location where the problem is found, and products with a distribution probability higher than a threshold are randomly selected as reference products for simultaneous sampling inspection, and the types, sampling inspection locations, and sampling inspection results of all products are marked;

[0099] Step S3 includes:

[0100] Step S31. After the problem product is detected for the first time, random product sampling is performed with the location where the problem product is found as the center and the sampling distance as the radius until w problem products are sampled. The sampling distance is determined by the simulation accuracy of the product distribution model, w is a preset value, and W>1;

[0101] Step S32. Select reference commodities for random inspection. The reference commodities are other types of commodities randomly selected by the commodity distribution model in the random inspection area. The type of the reference commodities selected is greater than or equal to 1. Mark the types, random inspection locations and random inspection results of all random inspection commodities in the random inspection area.

[0102] Step S4. Use the AI ​​classification model to classify the sources of the problematic products and reference products according to their distribution density. Output the density stream function of the problematic products from a single source within the sampling radius based on the classification result, and cluster the diffusion directions of each source based on the direction pointed when the partial derivative of the function is the largest. Determine the source of the problematic products based on the distance between each merchant and the clustering result.

[0103] Step S4 includes:

[0104] Step S41. Using the simulation results of the distribution of problem commodities and reference commodities in the commodity distribution model, the commodity operation ratio of the business operators and the commodity sampling results as the classification basis, the AI ​​classification model is used to classify the sources of the sampled problem commodities. The AI ​​classification models used include: decision tree model, ensemble learning model, logistic regression model and K nearest neighbor algorithm model;

[0105] Step S42. For each source of problematic goods, generate a density flow function F(x,y) of the goods distribution, where F(x,y) satisfies F(x0,y0)=h, Where x0 and y0 represent the horizontal and vertical coordinates of the location of the problematic product, h is the preset height coefficient, and g() is the preset fading function;

[0106] Step S43. Calculate the partial derivatives of the density stream function F(x,y) in each direction:

[0107]

[0108] Where T lx represents the partial derivative of F(x,y) in the direction of a·x+b·y=0. a and b are the directional coefficients of the horizontal and vertical coordinates respectively. Adjust the directional coefficients so that T lx The maximum value is obtained. At this time, the direction pointed by a·x+b·y=0 is recorded as the diffusion direction of the problematic product;

[0109] Step S44. Cluster the diffusion directions of problematic commodities from all sources to obtain clustering directions. Use the commodity distribution model to detect the points in the clustering directions where the distribution density of problematic commodities converges to 0. Use the obtained points as clustering centers and conduct market investigations in order of the distances between each business operator and the clustering center.

[0110] Step S5. After the source merchant is detected, the deviation direction and distortion rate of the distribution simulation result are fitted based on the difference between the actual distribution state and the simulated distribution state of the problematic product, and the model is iteratively reconstructed to update the training data in the database.

[0111] Step S5 includes:

[0112] Step S51. After the source merchant of the problematic product is detected, the radial distribution in the product distribution model is compared with the location of the source merchant and the scatter angle and scatter density of the problematic product in the sampling radius, and the deviation of the scatter direction and directional distortion rate between the radial distribution model and the actual product distribution is calculated;

[0113] Step S52: Reconstruct the commodity distribution model according to the deviation, and store the sampling results in the training database of the commodity distribution model.

[0114] Embodiment: There are 3 sales merchants in the monitoring area, with coordinates of (1, 0), (0, 1) and (0, 0), and sales volumes of 10, 15 and 20 respectively. There are 3 kinds of goods sold in total, among which the scatter distance function of product 1 is f(r). When the distance is 1, the scatter distance probability is the largest. A scatter model is established based on the scatter distance, sales volume and sales merchant of the goods. When a problematic product of product 1 is monitored, product 1 in the area is sampled. The coordinates of the problematic products detected by the sampling are (1.5, 0.8), (1.2, 1.1) and (0.9, 0.9) respectively. According to the results of density flow analysis, the probability that the problematic product comes from sales merchant 3 is the highest.

[0115] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0116] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for constructing a large supervision model based on artificial intelligence, characterized in that: The method comprises the following steps: Step S1. According to the usage characteristics of each type of goods, the distribution of the goods is simulated to obtain the probability distribution function of the distribution distance, the business operators are marked, and the sales volume of each type of goods by the business operators is predicted based on the historical sales records; Step S2. Taking the business operator as the center point, the distribution distance of the goods as the scatter radius, and the sales volume of the goods as the scatter density, a global goods scatter model is constructed, and the initial training data is input into the goods scatter model to reconstruct the attention function of the goods scatter model; Step S3. After the problematic product is found, a random inspection is conducted within the inspection radius with the discovery location as the center, and products with a distribution probability higher than the threshold are randomly selected as reference products for simultaneous inspection, marking the type, inspection location and inspection results of all products; Step S4. Use the AI ​​classification model to classify the sources of the problematic products and reference products according to their distribution density. Output the density stream function of the problematic products from a single source within the sampling radius based on the classification result, and cluster the diffusion directions of each source based on the direction pointed when the partial derivative of the function is the largest. Determine the source of the problematic products based on the distance between each merchant and the clustering result. Step S5. After the source merchant is detected, the deviation direction and distortion rate of the dispersion simulation result are fitted based on the difference between the actual dispersion state and the simulated dispersion state of the problematic product, and the model is iteratively reconstructed to update the training data in the database; Step S4 includes: Step S41. Using the simulation results of the distribution of problem commodities and reference commodities in the commodity distribution model, the commodity operation ratio of the business operators and the commodity sampling results as the classification basis, the AI ​​classification model is used to classify the sources of the sampled problem commodities. The AI ​​classification models used include: decision tree model, ensemble learning model, logistic regression model and K nearest neighbor algorithm model; Step S42. For each source of problematic goods, generate a density flow function F(x,y) of the goods distribution, where F(x,y) satisfies F(x0,y0)=h, Where x0 and y0 represent the horizontal and vertical coordinates of the location of the problematic product, h is the preset height coefficient, and g() is the preset fading function; Step S43. Calculate the partial derivatives of the density flow function F(x,y) in each direction: Where T lx represents the partial derivative of F(x,y) in the direction of a·x+b·y=0. a and b are the directional coefficients of the horizontal and vertical coordinates respectively. Adjust the directional coefficients so that T lx The maximum value is obtained. At this time, the direction pointed by a·x+b·y=0 is recorded as the diffusion direction of the problematic product; Step S44. Cluster the diffusion directions of problematic commodities from all sources to obtain clustering directions. Use the commodity distribution model to detect the points in the clustering directions where the distribution density of problematic commodities converges to 0. Use the obtained points as clustering centers and conduct market investigations in order of the distances between each business operator and the clustering center.

2. The method for constructing a large supervision model based on artificial intelligence according to claim 1, characterized in that: Step S1 includes: Step S11. Simulate the distribution of the goods according to the usage characteristics of the goods to obtain a probability distribution function f(r) of the distribution distance of the goods, where r represents the distribution distance of the goods. The usage characteristics include: weight, volume, shelf life and usage location of the goods. Tools for performing distribution simulation include: Monte Carlo simulation, Copula model simulation and Bootstrap simulation; Step S12. Mark the business on the electronic map, obtain the business's historical commodity sales function H(d), where d represents the date, and predict the business's sales on the current date: Among them, Q represents the sales volume of the merchant on the current date, D0 represents the maximum number of days of historical sales records, v represents the preset traffic flow change index, and H(i) represents the sales volume of the i-th day in the historical records.

3. The method for constructing a large supervision model based on artificial intelligence according to claim 2, characterized in that: Step S2 includes: Step S21. Make a radial commodity distribution model with the business as the center point. The distribution distance between the distribution points and the center point is randomly generated according to the probability distribution function f(r). The total distribution amount is set to Q. The commodity distribution models of all businesses are marked in space with different dimensions to generate a global commodity distribution model. Step S22. Track some of the goods sold by the business to obtain initial training data, use the initial training data to train the feedforward neural network to obtain the distribution trend of the goods, weight the attention coefficient of the product distribution model in the direction of the distribution trend, and reconstruct the attention function of the product distribution model.

4. The method for constructing a large supervision model based on artificial intelligence according to claim 3, characterized in that: Step S3 includes: Step S31. After the problem product is detected for the first time, random product sampling is performed with the location where the problem product is found as the center and the sampling distance as the radius until w problem products are sampled. The sampling distance is determined by the simulation accuracy of the product distribution model, w is a preset value, and W>1; Step S32. Select reference commodities for random inspection. The reference commodities are other types of commodities randomly selected by the commodity distribution model in the random inspection area. The type of the reference commodities selected is greater than or equal to 1. Mark the types, random inspection locations and random inspection results of all random inspection commodities in the random inspection area.

5. The method for constructing a large supervision model based on artificial intelligence according to claim 4, characterized in that: Step S5 includes: Step S51. After the source merchant of the problematic product is detected, the radial distribution in the product distribution model is compared with the location of the source merchant and the scatter angle and scatter density of the problematic product in the sampling radius, and the deviation of the scatter direction and directional distortion rate between the radial distribution model and the actual product distribution is calculated; Step S52: Reconstruct the commodity distribution model according to the deviation, and store the sampling results in the training database of the commodity distribution model.

6. A supervision model construction system based on artificial intelligence, characterized in that: The system includes the following modules: original data module, model training module, commodity sampling module, supervision and positioning module and intelligent training module; The raw data module is used to calculate the probability distribution function of the commodity distribution distance according to the usage characteristics of each commodity, mark the business locations of all merchants and the sales ratio of each commodity on the electronic map, and predict the commodity sales volume according to the flow of people and historical transactions at the business location, and store the historical commodity sampling data as the initial training data of the model in the database; The model training module is used to use the distribution distance of the commodity as the scatter radius and the sales volume as the scatter density, perform commodity diffusion simulation according to the randomly generated probability distribution function, build a global commodity scatter model according to the simulation results, input the initial training data into the commodity scatter model, reconstruct the attention function of the commodity scatter model, and output a multi-dimensional expansion matrix of each commodity monitoring point; The commodity sampling inspection module is used to perform sampling inspection in an area with a preset monitoring length as a radius with the discovery location as the center after discovering the problematic commodity, mark the discovery locations of all problematic commodities, and randomly select a commodity with a scatter probability higher than a threshold as a reference commodity, input the sampling inspection result into the commodity scatter model, and obtain the density stream function of the problematic commodity within the sampling inspection radius after processing with the multi-dimensional expansion matrix at the commodity discovery location; The supervision and positioning module is used to calculate the partial derivative values ​​of the density stream function in each direction, and the direction pointed to when the partial derivative value is the largest is taken as the first diffusion direction of the problematic product. The maximum distribution distance is moved forward along the first diffusion direction to conduct another random inspection. The maximum distribution distance is the product distribution distance corresponding to the maximum probability. The results of the two random inspections are input into the product distribution model to obtain the distribution center area of ​​the product, and the probability that the product comes from each merchant is given; The intelligent training module is used to adjust the original diffusion model after detecting the source of the problematic product, based on the deviation distance between the inferred source and the actual source, the actual sampling results within the sampling radius, and the sampling results of the reference products during the sampling process, and to perform directional clustering on the diffusion bias area, and output the iterative dispersion model; The implementation steps of the supervision and positioning module are as follows: Step S41. Using the simulation results of the distribution of problem commodities and reference commodities in the commodity distribution model, the commodity operation ratio of the business operators and the commodity sampling results as the classification basis, the AI ​​classification model is used to classify the sources of the sampled problem commodities. The AI ​​classification models used include: decision tree model, ensemble learning model, logistic regression model and K nearest neighbor algorithm model; Step S42. For each source of problematic goods, generate a density stream function F(x,y) of the goods distribution, where F(x,y) satisfies Where x0 and y0 represent the horizontal and vertical coordinates of the location of the problematic product, h is the preset height coefficient, and g() is the preset fading function; Step S43. Calculate the partial derivatives of the density flow function F(x,y) in each direction: Where T lx represents the partial derivative of F(x,y) in the direction of a·x+b·y=0. a and b are the directional coefficients of the horizontal and vertical coordinates respectively. Adjust the directional coefficients so that T lx The maximum value is obtained. At this time, the direction pointed by a·x+b·y=0 is recorded as the diffusion direction of the problematic product; Step S44. Cluster the diffusion directions of problematic commodities from all sources to obtain clustering directions. Use the commodity distribution model to detect the points in the clustering directions where the distribution density of problematic commodities converges to 0. Use the obtained points as clustering centers and conduct market investigations in order of the distances between each business operator and the clustering center.

7. The artificial intelligence-based supervision large model construction system according to claim 6 is characterized by: The original data module includes: a commodity feature unit and a business map unit; The commodity feature unit is used to calculate the probability distribution of the distance between the commodity and the purchase source according to the usage characteristics of the commodity, wherein the usage characteristics include: the weight, volume, shelf life and usage location of the commodity; The business map unit is used to mark the business operators and predict the sales volume of various commodities of the merchants; The model training module includes: a pattern recognition unit and a diffusion modeling unit; The pattern recognition unit is used to perform commodity diffusion simulation according to a preset diffusion distance, diffusion center and diffusion density; The diffusion modeling unit is used to train a feedforward neural network using initial training data, and input the diffusion simulation results into the trained feedforward neural network to obtain a commodity distribution model.

8. The artificial intelligence-based supervision large model construction system according to claim 7 is characterized by: The commodity sampling module includes: a density simulation unit, a reference sampling unit and a model adjustment unit; The density simulation unit is used to generate a density stream function of the problematic commodity within the sampling radius according to the sampling inspection result; The reference sampling unit is used to sample the problematic goods and the reference goods within a sampling radius of the problematic goods center; The model adjustment unit is used to analyze the distribution density of the problem commodity and the reference commodity in the commodity distribution model and adjust the model.

9. The artificial intelligence-based supervision large model construction system according to claim 8 is characterized by: The supervision and positioning module includes: a supplementary sampling unit and a central positioning unit; The supplementary sampling unit is used to determine the first diffusion direction of the problematic commodity and conduct another sampling in the first diffusion direction; The central positioning unit is used to speculate the source of the problematic product and provide the probability of the source being each merchant.

10. The artificial intelligence-based supervision large model construction system according to claim 9, characterized in that: The intelligent training module includes: a diffusion distortion unit, an AI simulation unit and a model iteration unit; The diffusion distortion unit is used to calculate the deviation distance between the inferred source center and the actual source center, and the statistical deviation of the sampling result; The AI ​​simulation unit is used to use the AI ​​clustering model to perform directional clustering on the diffusion bias area to obtain the distortion direction and distortion rate of the dispersion model; The model iteration unit is used to iteratively reconstruct the model according to the distortion of the dispersion model in this tracing, and update the training data in the database.

Images