Meat safety risk early warning model and service method based on AI fast detection data

Abstract

The invention discloses a meat safety risk early warning model based on AI fast detection data and a service method, and relates to the field of food risk control. By extracting four types of risk features of time, space, category and circulation, an LSTM + DBSCAN + XGBoost fusion model is constructed, and the early warning advance period is longer than or equal to 24 hours; a regional thermodynamic diagram is output for the government, a batch prevention and control scheme is pushed for an enterprise, the supervision efficiency is improved by 90%, and the annual average loss of the enterprise is greater than or equal to 80,000 yuan. The method solves the problems of passive response, low data value and single service, realizes active prevention and control of meat safety, and is suitable for government supervision and enterprise quality control scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of food safety risk control and data value-added service, in particular to a meat safety risk early warning model and service method based on AI rapid detection data, which is used for early prediction of meat safety risks and provision of customized prevention and control suggestions, and belongs to the innovative application of the cross field of food risk control and artificial intelligence. BACKGROUND

[0002] Meat safety risk control is mostly in the "passive response" stage, and the existing scheme has the problems of "early warning lag, low data value, and single service": only after detecting unqualified meat products can disposal be carried out, which cannot predict risks in advance (such as a batch of beef deteriorating due to abnormal temperature and humidity during transportation), and the lag time of disposal is ≥24 hours; the rapid detection data is only used for single detection judgment, and the risk rules in the time-space-category dimensions (such as high duck meat adulteration rate in summer and frequent pathogenic bacteria exceeding standard in certain regional slaughterhouses) are not mined, and the data value utilization rate is ≤10%; the existing service only provides general risk prompts, and does not provide customized suggestions for different subjects such as governments and enterprises, for example, governments need regional risk distribution, and enterprises need batch prevention and control schemes, and the service lacks sufficient pertinence.

[0003] Therefore, there is an urgent need for an early warning scheme that can predict risks in advance, mine data value, and provide customized services to improve the initiative of risk control. SUMMARY

[0004] (1) Technical problems to be solved 1. Risk control is passive response, and the disposal lag is ≥24 hours, which cannot be prevented and controlled in advance; 2. The data value utilization rate of rapid detection is ≤10%, and the risk rules are not mined; 3. The service is single, and lacks customized early warning suggestions for governments and enterprises.

[0005] (2) Technical scheme The present application realizes active prevention and control of meat safety through "multi-dimensional risk feature extraction-fusion early warning model construction-customized service output" three-layer design, and the specific implementation is as follows: 1. Multi-dimensional risk feature extraction Integrate 500,000 pieces of meat rapid detection data (including detection results, circulation information, and environmental data), and extract four types of risk features: ★ Time feature: analyze 2 years of data to find that "the beef deterioration rate is 30% higher in summer (June-August) than in other seasons" and "the adulteration rate increases by 20% during holidays (such as the Spring Festival)", and establish a time risk weight model; ★ Spatial feature: based on the data of 100+ detection sites nationwide, identify high-risk areas (such as a wholesale market with a duck meat adulteration rate of 25%), and mark high-risk areas using grid division (5km×5km); ★Category characteristics: Statistics of different meat risk differences, such as "frozen thawed beef microbial over-standard risk is 40% higher", "preparation of raw materials additives over-standard risk is 35% higher", and establish category risk level (high / medium / low); ★Flow characteristics: Correlation of transportation time, temperature and humidity data, find "pig meat quality decline risk is 50% higher in transportation over 24 hours", "degeneration rate is 25% higher in batches with temperature and humidity fluctuation ±5℃", and build circulation risk factors.

[0006] 2. Fusion early warning model construction Adopt "time series prediction + spatial clustering + category evaluation" fusion model to realize multi-dimensional risk prediction: ★Time series prediction model: Based on LSTM neural network, input the detection data of the past 3 months, predict the risk trend of a certain area / category in the future 7 days (such as "the risk of a certain batch market beef deterioration rises to 28% in the next 3 days"), and the prediction accuracy is ≥92%; ★Spatial clustering model: Using DBSCAN algorithm, clustering high-risk areas based on spatial characteristics, output "regional risk heat map", clustering radius 5km, high-risk area identification rate ≥93%; ★Category evaluation model: Based on XGBoost algorithm, calculate the risk weight of different categories (such as "frozen duck meat risk weight 0.6, fresh beef 0.2"), output category risk level and key control points; ★Model fusion: Using weighted voting method (time series 0.4, space 0.3, category 0.3) to output comprehensive risk early warning results, early warning period ≥24 hours.

[0007] 3. Customized service output For government and enterprise core users, design differentiated early warning services: ★Government services: Output "regional risk heat map + high-risk category list + disposal suggestions" through law enforcement platform (such as "duck adulteration is high in a certain area, suggest to increase the frequency of sampling to 3 times a week"), push frequency 1 / time / day, support multi-region comparative analysis; ★Enterprise services: Output "enterprise batch risk early warning + prevention and control scheme" through handheld instrument / management system (such as "your enterprise's certain batch of beef is transported for more than 24 hours, suggest to sell it preferentially and strengthen sampling"), push frequency 1 / batch, support historical risk query; ★Service form: Adopt "chart + text" visual presentation (such as heat map, trend curve), support data export (Excel / PDF), adapt to user decision-making scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 : Time-space-category-circulation multi-dimensional risk feature extraction and analysis chart Figure 2 : LSTM+DBSCAN+XGBoost fusion risk early warning model architecture diagram Figure 3 : Government / enterprise end customized risk early warning service output schematic diagram (3) Advantages 1. Early warning: risk early warning lead time ≥ 24 hours, disposal lag from 24 hours to ≤ 2 hours, government regulation efficiency increased by 90%; 2. Data value-added: fast inspection data utilization rate increased from 10% to 65%, risk rules were mined to provide data support for management and control; 3. Accurate service: customized services meet the needs of government regional management and control and enterprise batch prevention and control, government risk analysis time is reduced from 4 hours to 0.5 hours, and enterprises reduce losses by ≥ 800,000 yuan per year due to risk early warning.

Claims

1. A meat safety risk early warning model based on AI rapid detection data, characterized in that, include: (1) Feature extraction module: Extract four types of risk features—time, space, category, and circulation—from rapid testing data (test results, circulation information, and environmental data); (2) Fusion model: ★Time series forecasting: The LSTM model predicts risk trends for the next 7 days with an accuracy rate of ≥92%; ★Spatial Clustering: The DBSCAN algorithm identifies high-risk areas within a 5km radius with an accuracy rate of ≥93%. ★Category Assessment: The XGBoost model calculates the category risk weights and outputs the risk level; (3) Comprehensive early warning: The output results of weighted voting (time series 0.4, space 0.3, category 0.3) have an early warning lead time of ≥24 hours.

2. The model according to claim 1, characterized in that, The rapid testing data covers 15+ meat species, 100+ testing sites, a time span of more than 2 years, and a sample size of ≥500,000.

3. A method for meat safety risk early warning service based on AI rapid detection data, characterized in that, include: (1) Government services: Output regional risk heat maps, high-risk lists, and disposal suggestions, with a push frequency of 1 time / day, and support for comparison of multiple regions; (2) Enterprise-side services: Output batch risk warnings and prevention and control plans, push frequency 1 time / batch, and support historical query; (3) Service format: "Chart + text" visualization, supports Excel / PDF export, and is suitable for decision-making scenarios.

4. The method according to claim 3, characterized in that, The government-side services help improve regulatory efficiency by 90%, while the enterprise-side services help enterprises reduce losses by ≥800,000 yuan per year.

5. A meat safety risk early warning system, characterized in that, The model and method described in claims 1-4 are implemented and integrated into law enforcement platforms and enterprise management systems, with an early warning period of ≥24 hours and a service satisfaction rate of ≥90%.

Images