Intelligent detection system and method for food-borne disease monitoring platform

Abstract

The invention discloses an intelligent detection system and method for a food-borne disease monitoring platform. The method comprises the following steps: S1, constructing a food sample data set; s2, preprocessing the food sample data set to generate a standardized food sample data set; s3, performing fuzzification processing on the microbial indicator, pollutant concentration and pathogen detection rate data in the standardized food sample data set; s4, converting the fuzzy risk score into a specific numerical food sample risk index; s5, performing risk assessment on the defuzzified food sample risk indexes by using an asynchronous dominant actor commentator algorithm; s6, after the asynchronous learning process is completed, the weight is updated to the global network; and S7, when the defuzzification risk index of the food sample reaches or exceeds a risk early warning trigger threshold, the system automatically generates food-borne disease risk early warning information, and sends the food-borne disease risk early warning information to a related department or a food supply chain management user through an information system. According to the invention, real-time prediction and dynamic adjustment of food-borne disease risks are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of foodborne diseases, and particularly to an intelligent detection system and method for a foodborne disease monitoring platform. Background Art

[0002] With the development of intelligent detection and artificial intelligence technologies, intelligent monitoring and early warning systems have gradually been introduced in the field of food safety to detect pathogens and contaminants in food samples, thereby preventing and controlling the spread of foodborne diseases. Foodborne disease monitoring is of great significance to public health, capable of identifying food contamination, providing early warnings, and supporting public health departments and the food production chain. In traditional monitoring methods, the detection of food samples usually relies on manual or semi-automated methods, which are not only time-consuming and laborious but also suffer from problems such as untimely detection, missed detection, and false alarms, making it difficult to achieve rapid and comprehensive detection of diverse food samples.

[0003] In the prior art, food safety detection mostly relies on laboratory means and manual analysis, using professional equipment to detect pathogens or chemical contaminants in food. Although the above methods have certain advantages in terms of accuracy, due to the high cost of manual operation and equipment use, the detection efficiency is greatly limited. In the face of complex food ingredients and different sample sources, traditional methods are difficult to complete the detection in a short time. In addition, existing food detection systems mostly adopt a single static detection method and cannot be adaptively adjusted according to environmental changes and detection data, lacking intelligent risk identification and prediction functions, and it is difficult to prevent foodborne disease outbreaks in a timely and effective manner.

[0004] Some automated detection systems have also begun to be applied in the field of foodborne disease monitoring, using sensors and data processing modules to achieve a certain degree of automation. However, current automatic detection systems mainly rely on rule matching and simple data processing, lacking advanced risk assessment and dynamic adjustment functions. Existing automatic detection systems cannot effectively integrate multi-dimensional data and conduct complex risk analysis, making it difficult to achieve timely early warning for sudden risks. In the face of diverse and large-scale food samples, the limitations of the prior art are gradually emerging: on the one hand, the detection process lacks flexibility and is difficult to adapt to the risk characteristics of different food samples; on the other hand, when the system faces changing detection data and environmental states, it is difficult to optimize strategies through real-time learning, resulting in low accuracy of risk identification and early warning. In addition, existing automatic detection technologies have a high latency at the data processing and decision-making levels, which is not conducive to the rapid prevention and control of foodborne diseases.

[0005] In summary, the existing technologies have significant deficiencies in terms of detection efficiency, warning accuracy, and system self - adaptability, and cannot meet the requirements of modern food safety monitoring. To address these issues, there is an urgent need for a new method and system to achieve efficient, accurate, and dynamically adjustable intelligent detection, so as to improve the reliability and timeliness of foodborne disease monitoring. Summary of the Invention

[0006] An object of the present invention is to propose an intelligent detection system and method for a foodborne disease monitoring platform, and the present invention realizes real - time prediction and dynamic adjustment of foodborne disease risks.

[0007] An intelligent detection method for a foodborne disease monitoring platform according to an embodiment of the present invention includes the following steps:

[0008] S1. Real - time collect data on microbial indicators, pollutant concentrations, and pathogen detection rates of food samples through multiple distributed sensors, and construct a food sample data set;

[0009] S2. Pre - process the food sample data set to generate a standardized food sample data set;

[0010] S3. Perform fuzzification processing on the microbial indicators, pollutant concentrations, and pathogen detection rates in the standardized food sample data set, and convert the variables into fuzzy values represented in the form of low, medium, and high risk levels;

[0011] S4. Reason about the fuzzified food sample data based on a preset fuzzy rule base. The fuzzy rule base contains a set of logical rules for risk assessment. Generate a risk score through the fuzzy reasoning process, and perform defuzzification processing on the risk score to convert the fuzzy risk score into a specific numerical food sample risk indicator;

[0012] S5. Use the asynchronous advantage actor - critic algorithm to perform risk assessment on the defuzzified food sample risk indicators;

[0013] S6. Dynamically adjust the risk warning trigger threshold to the optimal threshold of the current environmental state based on the risk assessment strategy. The optimal threshold adjustment is achieved through asynchronous update. Each thread independently processes different sensor or food sample data, and updates the weights to the global network after completing the asynchronous learning process;

[0014] S7. When the defuzzified risk indicator of the food sample reaches or exceeds the risk warning trigger threshold, the system automatically generates a foodborne disease risk warning message and sends it to relevant departments or food supply chain management users through the information system.

[0015] Optionally, the S1 specifically includes:

[0016] S11. Use multiple distributed sensors to collect the microbial index M i , pollutant concentration C i , and pathogen detection rate data P i in real time. Among them, the collected data of the microbial index is represented by M i , and the collected data of the pollutant concentration is represented by C i .

[0017] S12. Integrate the microbial index data, pollutant concentration data, and pathogen detection rate data collected by the distributed sensors to generate a preliminary food sample dataset:

[0018] D = {(M i , C i , P i )|i = 1, 2,..., n};

[0019] Among them, n is the total number of samples.

[0020] Optionally, the S2 specifically includes:

[0021] S21. Perform data cleaning on the food sample dataset D, including filtering missing values, removing outliers, and filling in missing data;

[0022] S22. Denoise the food sample dataset after data cleaning, use a noise filtering algorithm to smooth data fluctuations, remove abnormal fluctuations and measurement errors, and obtain a denoised food sample dataset D';

[0023] S23. Standardize the denoised food sample dataset D' to generate a standardized food sample dataset:

[0024]

[0025] Optionally, the S3 specifically includes:

[0026] Perform fuzzification on the microbial index , pollutant concentration , and pathogen detection rate in the standardized food sample dataset, set the fuzzy membership function for each index, convert each data value into a fuzzy risk level, and define a triangular fuzzy membership function for each index:

[0027]

[0028]

[0029]

[0030] Among them, Xi Data values representing microbial indicators, pollutant concentrations, and pathogen detection rates. a, b, and c are the fuzzy classification thresholds for the corresponding indicators, set according to food safety standards and used to divide low-risk, medium-risk, and high-risk levels.

[0031] Optionally, step S4 specifically includes:

[0032] S41. Based on a preset fuzzy rule base, perform fuzzy inference on the fuzzified data of the standardized food sample dataset, where the fuzzy rule base includes multiple logical rules:

[0033] If is high risk, is high risk, is high risk, then the overall risk of the food sample is high;

[0034] If is medium risk, is high risk, is medium risk, then the overall risk of the food sample is medium;

[0035] If is low risk, is low risk, is low risk, then the overall risk of the food sample is low;

[0036] S42. Use fuzzy logic inference to perform combined calculations on the fuzzified indicators to generate a comprehensive fuzzy risk score μ Risk (i):

[0037]

[0038] where w M , w C , w P are the weight coefficients of the microbial indicator, pollutant concentration, and pathogen detection rate respectively, satisfying w M +w C +w P =1;

[0039] S43. Perform defuzzification on the comprehensive fuzzy risk membership degree μ Risk (i) and convert it into a specific food sample risk index R i , and the food sample risk score is used to characterize the potential foodborne disease risk level of the food sample:

[0040]

[0041] where r is a continuous variable of the risk score, r min and r maxare the lowest and highest risk score values; μ Risk (r) is the comprehensive fuzzy membership function corresponding to the risk score r.

[0042] Optionally, the S5 specifically includes:

[0043] S51. Define the environmental state S of the detection platform t , and use the defuzzified risk score R of the food sample i and the historical risk level data H t as the environmental state input to form the current environmental state S t = [R i , H t ;

[0044] S52. Use the multi-layer neural network f θ (S t ) to extract features from the environmental state S t and generate the information φ representing the current features and potential risks of the food sample t = f θ (S t ), where θ is the parameter set of the neural network;

[0045] S53. Construct an asynchronous advantage actor-critic algorithm model and adopt an actor-critic architecture:

[0046] The actor network π θ (a t |φ t ) takes the feature information φ t as the input and outputs the risk assessment strategy a t , and the risk assessment strategy a t includes the setting of the risk warning trigger threshold T t and the selection of the detection method m t ;

[0047] The critic network evaluates the value of the current strategy, takes the risk assessment accuracy as the value index, and calculates the state value function

[0048] S54. Execute the strategy a t at time step t and obtain the immediate reward r t , and the immediate reward is defined according to the deviation between the actual risk assessment result and the expected result as:

[0049]

[0050] where is the actually observed risk level, and R iis the risk score predicted by the asynchronous advantage actor-critic algorithm model;

[0051] S55. Calculate the advantage function A t , which is used to guide policy optimization:

[0052]

[0053] where γ is the discount factor and φ t+1 is the feature information at the next time step;

[0054] S56. Use the advantage function A t to guide the dynamic optimization of the risk assessment policy and update the parameters of the actor network and the critic network:

[0055] Update of actor network parameters:

[0056]

[0057] Update of critic network parameters:

[0058]

[0059] where α θ and are the learning rates of the actor network and the critic network respectively;

[0060] S57. Through multi-threaded asynchronous update, each thread independently processes different food sample data or sensor data, and after completing the asynchronous learning process, updates the parameters of each thread to the global network.

[0061] Optionally, the specific steps of S6 include:

[0062] S61. Based on the risk assessment policy of the asynchronous advantage actor-critic algorithm model, dynamically adjust the risk warning trigger threshold to the optimal threshold of the current environmental state

[0063] S62. Through each independent thread to process sensor or food sample data, at time step t, the current risk assessment policy a t =[T t ,m t is adjusted in real time, so that the adjustment target of the risk warning trigger threshold meets the following optimization conditions:

[0064]

[0065] where T t is the risk warning trigger threshold at the current time step, R t+1,i is the risk score prediction of the i-th food sample at the next time step, R target,iis the target risk level corresponding to the food sample, ω i are the weight factors of each sample, λ is the regularization parameter, used to balance the deviation of the risk score and the adjustment range of the threshold, T base is the basic reference threshold;

[0066] S63. Adopt an asynchronous update mechanism, and each thread independently executes policy evaluation and threshold update for the risk warning trigger threshold T t Update according to the risk prediction results of each thread, so that each thread in the asynchronous learning process executes in parallel;

[0067] S64. After the thread completes the calculation, update the optimal weight calculated by each thread to the global network, and the weight update of the global network is as follows:

[0068]

[0069] Among them, represents the global weight vector before update, represents the global weight vector after update, represents the optimal weight obtained by the k-th thread at time step t, K is the total number of threads, α w is the learning rate of global weight update, and β is the adjustment factor.

[0070] An intelligent detection system for a foodborne disease monitoring platform, including the following modules:

[0071] A distributed data acquisition module, used to collect data on microbial indicators, pollutant concentrations, and pathogen detection rates of food samples in real time through multiple sensors, and construct a food sample data set;

[0072] A data preprocessing module, which performs data cleaning, denoising, and standardization processing on the collected food sample data set to generate a standardized food sample data set;

[0073] A fuzzification processing module, which fuzzifies the data on microbial indicators, pollutant concentrations, and pathogen detection rates of the standardized food sample data set, and uses a preset triangular membership function to generate a fuzzy risk level;

[0074] A fuzzy inference module, which performs logical inference on the fuzzified data based on a preset fuzzy rule base, combines the fuzzy membership degree, generates a comprehensive risk membership degree through the inference process, and further generates a specific risk score of the food sample through defuzzification processing;

[0075] The asynchronous risk assessment module inputs the defuzzified risk indicators of food samples into the environmental state of the detection platform based on the asynchronous advantage actor-critic algorithm, extracts feature information using a multi-layer neural network, outputs a risk assessment strategy through an actor-critic architecture and evaluates the value of the current strategy, and calculates the advantage function based on the immediate reward.

[0076] The risk threshold dynamic adjustment module, through an asynchronous update mechanism, independently processes food sample data and sensor data for each thread, dynamically adjusts the risk warning trigger threshold to the optimal threshold of the current environmental state, and updates the optimal weights of the thread to the global network after the asynchronous learning process is completed.

[0077] The warning generation and information transmission module, when the risk score of a food sample reaches or exceeds the optimal threshold for triggering a risk warning, automatically generates a risk warning message for foodborne diseases and sends the warning message to relevant departments or food supply chain management users through an information system.

[0078] The beneficial effects of the present invention are:

[0079] (1) The present invention uses fuzzy logic to perform defuzzification processing on the microbial indicators, pollutant concentrations, and pathogen detection rates in food samples at low, medium, and high risk levels, and generates the risk membership degree of food samples through an adaptive fuzzy rule base, improving the flexibility of the system in processing different types of complex food samples. The present invention performs combined reasoning on different risk factors through fuzzy inference and multi-dimensional analysis of membership degrees, significantly improving the accuracy of risk assessment and enabling the system to flexibly respond to complex detection environments.

[0080] (2) By introducing the asynchronous advantage actor-critic algorithm, the present invention can adaptively adjust the strategy during the risk assessment process. The system automatically optimizes the warning threshold and detection strategy according to the real-time environmental state. The A3C algorithm performs multi-threaded asynchronous learning, and each thread independently processes the data of different food samples, enabling the system to quickly iterate and learn in diverse data streams, realizing real-time prediction and dynamic adjustment of foodborne disease risks. Compared with the defects of traditional detection systems that cannot be flexibly adaptive, the present invention has significant advantages in terms of dynamics and adaptability, effectively improving the accuracy of risk identification and response speed.

[0081] (3) The present invention utilizes the asynchronous multi-threaded architecture of the A3C algorithm, enabling each thread to independently process data from different sensors or food samples. After asynchronous learning, the optimal policy parameters are synchronized to the global network. Compared with the single-threaded or synchronous learning process of traditional methods, the present invention significantly improves the system's parallel processing ability and data stream processing efficiency, enabling the system to still operate efficiently in the face of a large amount of food sample data and ensuring the real-time and reliability of the data. In large-scale monitoring applications, the multi-threaded asynchronous optimization design of the present invention reduces the delay in the detection process, making the release of risk assessment and early warning information more timely, thereby better supporting the prevention and control of foodborne diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used in conjunction with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0083] Figure 1 is a flowchart of an intelligent detection system and method for a foodborne disease monitoring platform proposed by the present invention;

[0084] Figure 2 is a flowchart of the application of the asynchronous advantage actor-critic algorithm in the risk assessment module in the flowchart of an intelligent detection system and method for a foodborne disease monitoring platform proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0085] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0086] Refer to Figure 1 - Figure 2 , an intelligent detection method for a foodborne disease monitoring platform, comprising the following steps:

[0087] S1. Real-time collect data on microbial indicators, pollutant concentrations, and pathogen detection rates of food samples through multiple distributed sensors, and construct a food sample dataset;

[0088] S2. Preprocess the food sample dataset to generate a standardized food sample dataset;

[0089] S3. Fuzzify the data on microbial indicators, pollutant concentrations, and pathogen detection rates in the standardized food sample dataset, and convert the variables into fuzzy values represented in the form of low, medium, and high risk levels;

[0090] S4. Infer the fuzzified food sample data based on a preset fuzzy rule base. The fuzzy rule base contains a set of logical rules for evaluating risks. Through the fuzzy inference process, a risk score is generated, and the risk score is defuzzified to convert the fuzzy risk score into a specific numerical food sample risk indicator.

[0091] S5. Use the asynchronous advantage actor-critic algorithm to evaluate the risk of the defuzzified food sample risk indicators.

[0092] S6. Dynamically adjust the risk warning trigger threshold to the optimal threshold of the current environmental state based on the risk assessment strategy. The optimal threshold adjustment is achieved through asynchronous updates. Each thread independently processes different sensor or food sample data, and after completing the asynchronous learning process, the weights are updated to the global network.

[0093] S7. When the defuzzified risk indicator of the food sample reaches or exceeds the risk warning trigger threshold, the system automatically generates a risk warning message for foodborne diseases and sends it to the relevant departments or food supply chain management users through the information system.

[0094] In this embodiment, S1 specifically includes:

[0095] S11. Use multiple distributed sensors to collect the microbial index M i , pollutant concentration C i , and pathogen detection rate data P i of the food sample in real time. Among them, the collected data of the microbial index is represented by M i , and the collected data of the pollutant concentration is represented by C i .

[0096] S12. Integrate the microbial index data, pollutant concentration data, and pathogen detection rate data collected by the distributed sensors to generate a preliminary food sample data set:

[0097] D = {(M i , C i , P i ) | i = 1, 2,..., n};

[0098] Among them, n is the total number of samples.

[0099] In this embodiment, S2 specifically includes:

[0100] S21. Clean the data of the food sample data set D, and perform operations such as filtering missing values, removing outliers, and filling in missing data.

[0101] S22. Denoise the food sample dataset after data cleaning. Use a noise filtering algorithm to smooth data fluctuations, remove abnormal fluctuations and measurement errors, and obtain a denoised food sample dataset D'.

[0102] S23. Standardize the denoised food sample dataset D' to generate a standardized food sample dataset:

[0103]

[0104] In this embodiment, S3 specifically includes:

[0105] For the microbial indicators pollutant concentrations and pathogen detection rates in the standardized food sample dataset, perform fuzzification processing. Set the fuzzy membership function for each indicator, convert each data value into a fuzzy risk level, and define a triangular fuzzy membership function for each indicator:

[0106]

[0107]

[0108]

[0109] where X i represents the data values of microbial indicators, pollutant concentrations, and pathogen detection rates. a, b, and c are the fuzzy classification thresholds for the corresponding indicators, set according to food safety standards, and are used to divide low-risk, medium-risk, and high-risk levels.

[0110] In this embodiment, S4 specifically includes:

[0111] S41. Based on a preset fuzzy rule base, perform fuzzy reasoning on the fuzzified data of the standardized food sample dataset. The fuzzy rule base includes multiple logical rules:

[0112] If is high risk, is high risk, is high risk, then the overall risk of the food sample is high;

[0113] If is medium risk, is high risk, is medium risk, then the overall risk of the food sample is medium;

[0114] If is low risk, is low risk, is low risk, then the overall risk of the food sample is low;

[0115] S42. Use fuzzy logic inference to perform combined calculations on the fuzzified indicators to generate a comprehensive fuzzy risk score μ Risk (i):

[0116]

[0117] where w M , w C , w P are the weight coefficients of the microbial index, pollutant concentration, and pathogen detection rate, respectively, satisfying w M + w C + w P = 1;

[0118] S43. Defuzzify the comprehensive fuzzy risk membership degree μ Risk (i) and convert it into a specific food sample risk index R i . The food sample risk score is used to characterize the potential foodborne disease risk level of the food sample:

[0119]

[0120] where r is a continuous variable of the risk score, r min and r max are the lowest and highest risk score values, respectively; μ Risk (r) is the comprehensive fuzzy membership degree function corresponding to the risk score r.

[0121] In this embodiment, S5 specifically includes:

[0122] S51. Define the environmental state S t of the detection platform, and use the defuzzified food sample risk score R i and the historical risk level data H t as the environmental state input to form the current environmental state S t = [R i , H t ;

[0123] S52. Use a multi-layer neural network f θ (S t ) to extract features from the environmental state S t and generate information φ t = f θ (S t ) representing the current features and potential risks of the food sample, where θ is the parameter set of the neural network;

[0124] S53. Construct an asynchronous advantage actor-critic algorithm model and adopt an actor-critic architecture:

[0125] Actor network π θ (a t |φ t ) takes the feature information φ t as input and outputs a risk assessment strategy a t , and the risk assessment strategy a t includes the setting of the risk warning trigger threshold T t and the selection of the detection method m t ;

[0126] Critic network Evaluates the value of the current strategy, uses the risk assessment accuracy as the value metric, and calculates the state value function

[0127] S54. Execute the strategy a at time step t t , and obtain an immediate reward r t . The immediate reward is defined according to the deviation between the actual risk assessment result and the expected result as:

[0128]

[0129] where is the actually observed risk level, and R i is the risk score predicted by the asynchronous advantage actor-critic algorithm model;

[0130] S55. Calculate the advantage function A t , which is used to guide the strategy optimization:

[0131]

[0132] where γ is the discount factor and φ t+1 is the feature information at the next time step;

[0133] S56. Use the advantage function A t to guide the dynamic optimization of the risk assessment strategy and update the parameters of the actor network and the critic network:

[0134] Update of actor network parameters:

[0135]

[0136] Update of critic network parameters:

[0137]

[0138] where α θ and are the learning rates of the actor network and the critic network respectively;

[0139] S57. Through multi-threaded asynchronous update, each thread independently processes different food sample data or sensor data, and updates the parameters of each thread to the global network after completing the asynchronous learning process.

[0140] In this embodiment, S6 specifically includes:

[0141] S61. Based on the risk assessment strategy of the asynchronous advantage actor-critic algorithm model, dynamically adjust the risk warning trigger threshold to the optimal threshold of the current environmental state

[0142] S62. Process sensor or food sample data through each independent thread, and adjust the current risk assessment strategy a t =[[T t ,m t in real time at time step t, so that the adjustment target of the risk warning trigger threshold meets the following optimization conditions:

[0143]

[0144] where T t is the risk warning trigger threshold at the current time step, R t+1,i is the predicted risk score of the i-th food sample at the next time step, R target,i is the target risk level corresponding to the food sample, ω i is the weight factor of each sample, λ is a regularization parameter used to balance the risk score deviation and the amplitude of threshold adjustment, T base is the basic reference threshold;

[0145] S63. Adopt an asynchronous update mechanism, and each thread independently executes policy evaluation and threshold update to update the risk warning trigger threshold T t according to the risk prediction results of each thread, so that each thread in the asynchronous learning process executes in parallel;

[0146] S64. After the thread completes the calculation, update the optimal weight calculated by each thread to the global network, and the weight update of the global network is as follows:

[0147]

[0148] where, represents the global weight vector before update, represents the global weight vector after update, represents the optimal weight obtained by the k-th thread at time step t, K is the total number of threads, and α w is the learning rate of global weight update, and β is a regulation factor.

[0149] An intelligent detection system for a foodborne disease monitoring platform, comprising the following modules:

[0150] A distributed data acquisition module, used to collect data on microbial indicators, pollutant concentrations, and pathogen detection rates of food samples in real time through multiple sensors, and construct a food sample data set;

[0151] A data preprocessing module, which performs data cleaning, denoising, and standardization processing on the collected food sample data set to generate a standardized food sample data set;

[0152] A fuzzification processing module, which performs fuzzification processing on the microbial indicators, pollutant concentrations, and pathogen detection rates of the standardized food sample data set, and uses a preset triangular membership function to generate a fuzzy risk level;

[0153] A fuzzy inference module, which performs logical inference on the fuzzified data based on a preset fuzzy rule base, combines the fuzzy membership degree, generates a comprehensive risk membership degree through the inference process, and further generates a specific risk score of the food sample through defuzzification processing;

[0154] An asynchronous risk assessment module, which inputs the defuzzified food sample risk indicators into the environmental state of the detection platform based on the asynchronous advantage actor-critic algorithm, extracts feature information using a multi-layer neural network, outputs a risk assessment strategy through an actor-critic architecture and evaluates the value of the current strategy, and calculates an advantage function based on the immediate reward;

[0155] A risk threshold dynamic adjustment module, through an asynchronous update mechanism, each thread independently processes food sample data and sensor data, dynamically adjusts the risk warning trigger threshold to the optimal threshold of the current environmental state, and updates the optimal weight of the thread to the global network after the asynchronous learning process is completed;

[0156] A warning generation and information transmission module, when the risk score of the food sample reaches or exceeds the optimal threshold for triggering a risk warning, the system automatically generates a foodborne disease risk warning message, and sends the warning message to relevant departments or food supply chain management users through an information system.

[0157] Example 1:

[0158] In a certain food production base, the intelligent detection system of the present invention is applied to the safety monitoring of meat products and fresh foods. The products produced by the base are diverse, and different batches of foods have different microbial loads, pollutant concentrations, and pathogen detection rates. The system collects microbial and pollutant data of various food samples in real time during the food production, storage, and transportation links to quickly identify potential foodborne disease risks.

[0159] At 14:30 on the afternoon of June 12, 2024, the system detected a batch of pork samples in the raw material processing area of the base. The data of the microbial indicators, pollutant concentrations, and pathogen detection rates of the samples showed an abnormal increase. Through the real-time data collected by distributed sensors, the system found that the concentration of Escherichia coli in the samples reached 210 CFU / g, far higher than the standard safety value (100 CFU / g), and the pollutant concentration was detected to be 0.3 mg / kg. The risk assessment module based on fuzzy logic rated this sample as a high risk.

[0160] After the detection data was fuzzified and input into the asynchronous advantage actor-critic algorithm model of the system, when the system evaluated that the risk of this sample exceeded the warning threshold, the intelligent detection system generated a warning signal within less than 3 seconds, alerting relevant operators to pay attention to the potential risk of foodborne diseases. The system further detailedly recorded the source information of this batch of samples, including the sampling time of 14:30 on June 12 and the sampling location of Cold Storage Warehouse No. 2 in the raw material processing area, and generated a risk assessment report.

[0161] In the next half hour, the system conducted real-time monitoring on the transportation route and storage environment of this batch of pork. At 15:00 on June 12, when the system detected the storage conditions of this batch of pork in the finished product area, it found that the temperature reached 10°C, while the ideal storage temperature for this type of fresh product should be below 4°C. The system recorded this abnormal temperature through the environmental sensor and judged it as a potential risk factor. After being evaluated by the risk algorithm, the situation of the temperature exceeding the standard was also regarded as a high-risk item, and immediately transmitted a new round of risk report to the management center.

[0162] Meanwhile, to further ensure risk control, the system initiated a multi-threaded parallel analysis mode to deeply analyze the data of this batch and adjacent batch samples. Each thread processed the historical data, environmental temperature, and transportation duration characteristics of a food sample, comprehensively evaluating potential risks. During the analysis process, the system used the asynchronous advantage actor-critic algorithm to dynamically adjust the real-time data and lowered the warning trigger threshold to 100 CFU / g to ensure that risks could be captured in a timely manner.

[0163] The system completed the analysis of dozens of samples in a short time, completed the risk assessment at 15:45 on June 12, and transmitted the results to the management personnel in real time. The report automatically generated by the system showed that the risk assessment result of this batch of pork was 96%, reaching the risk warning standard. The warning report detailedly listed the source information of the samples, the abnormal pollutant indicators and temperature data monitored, and gave the risk level (high risk), prompting the management center to conduct isolation treatment. The management center immediately instructed relevant staff to seal and isolate this batch of pork according to the report information, stop its circulation path, and arrange for subsequent processing.

[0164] To verify the effectiveness of this system, the base introduced a traditional manual detection method to detect the same batch of samples and compared the results with those of the intelligent detection system. The traditional method uses laboratory means for analysis, and the detection time is about 48 hours. The detection results show that the concentration of Escherichia coli is 200 CFU / g and the concentration of pollutants is 0.3 mg / kg, which is basically consistent with the data of the intelligent detection system. However, due to the long detection time, the batch of pork has entered the subsequent production process, resulting in a delay in isolation measures.

[0165] In terms of the comparison of detection efficiency, the detection time of the intelligent detection system is only 30 minutes, while the traditional method takes 48 hours. In terms of early warning response, the real-time and dynamic adjustment mechanism of the intelligent detection system enables the system to generate early warning information in the first time, accurately identify the risks of this batch of food, and avoid the subsequent risk of pollution spread.

[0166] This invention uses fuzzy logic to perform fuzzy processing on the data of microbial indicators, pollutant concentrations, and pathogen detection rates in food samples at low, medium, and high risk levels. Through an adaptive fuzzy rule base, the risk membership degree of food samples is generated, which improves the flexibility of the system in dealing with different types of complex food samples. This invention combines and reasons different risk factors through fuzzy inference and multi-dimensional analysis of membership degrees, significantly improving the accuracy of risk assessment and enabling the system to flexibly respond to complex detection environments.

[0167] This invention can adaptively adjust the strategy during the risk assessment process by introducing the Asynchronous Advantage Actor-Critic (A3C) algorithm. The system automatically optimizes the early warning threshold and detection strategy according to the real-time environmental state. Through multi-threaded asynchronous learning, each thread independently processes the data of different food samples, enabling the system to quickly iterate and learn in diverse data streams, realizing real-time prediction and dynamic adjustment of foodborne disease risks.

[0168] This invention uses the asynchronous multi-threaded architecture of the A3C algorithm to enable each thread to independently process the data of different sensors or food samples, and synchronize the optimal strategy parameters to the global network after asynchronous learning. Compared with the single-threaded or synchronous learning process of traditional methods, this invention significantly improves the parallel processing ability and data stream processing efficiency of the system, enabling the system to still operate efficiently in the face of a large amount of food sample data and ensuring the real-time and reliability of the data. In large-scale monitoring applications, the multi-threaded asynchronous optimization design of this invention reduces the delay in the detection process, making the release of risk assessment and early warning information more timely, thus better supporting the prevention and control of foodborne diseases.

[0169] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, shall be covered by the protection scope of the present invention.

Claims

1. An intelligent detection method for a foodborne disease monitoring platform, characterized in that: The steps include: S1. Collect microbial indicators, pollutant concentrations and pathogen detection rate data of food samples in real time through multiple distributed sensors to build a food sample data set; S2, preprocessing the food sample data set to generate a standardized food sample data set; S3. Perform fuzzification on the microbial indicators, contaminant concentrations, and pathogen detection rate data in the standardized food sample data set, converting the variables into fuzzy values ​​to represent the risk levels in the form of low, medium, and high; S4, reasoning the fuzzy food sample data based on a preset fuzzy rule base, wherein the fuzzy rule base includes a set of logical rules for assessing risk, generating a risk score through a fuzzy reasoning process, defuzzifying the risk score, and converting the fuzzy risk score into a specific numerical food sample risk index; S5, using asynchronous advantage actor critic algorithm to conduct risk assessment on the risk indicators of food samples after defuzzification; S6. Based on the risk assessment strategy, the risk warning trigger threshold is dynamically adjusted to the optimal threshold for the current environmental state. The optimal threshold adjustment is achieved through asynchronous update. Each thread independently processes different sensor or food sample data. After completing the asynchronous learning process, the weight is updated to the global network. S7. When the defuzzified risk index of a food sample reaches or exceeds the risk warning trigger threshold, the system automatically generates foodborne disease risk warning information and sends it to relevant departments or food supply chain management users through the information system.

2. The intelligent detection method for foodborne disease monitoring platform according to claim 1, characterized in that: The S1 specifically includes: S11. Microbiological indicators of food samples using multiple distributed sensors M i , pollutant concentration C i and pathogen detection rate data P i Real-time data collection is performed, among which the microbial index data is collected using M i Indicates that pollutant concentration data is collected using C i express; S12. Integrate the microbial index data, pollutant concentration data, and pathogen detection rate data collected by distributed sensors to generate a preliminary food sample data set: D={(M i ,C i ,P i )|i=1,2,...,n}; Where n is the total number of samples.

3. The intelligent detection method for foodborne disease monitoring platform according to claim 1, characterized in that: The S2 specifically includes: S21, performing data cleaning on the food sample data set D, filtering missing values, removing outliers, and filling in missing data; S22, performing denoising on the food sample data set after data cleaning, using a noise filtering algorithm to smooth data fluctuations, remove abnormal fluctuations and measurement errors, and obtain a denoised food sample data set D'; S23, standardize the food sample data set D' after denoising to generate a standardized food sample data set:

4. The intelligent detection method for foodborne disease monitoring platform according to claim 1, characterized in that: The S3 specifically includes: Microbiological indicators in standardized food sample datasets Pollutant concentration and pathogen detection rate Perform fuzzy processing, set the fuzzy membership function of each indicator, convert each data value into a fuzzy risk level, and define a triangular fuzzy membership function for each indicator: Among them, X i The data values ​​representing microbial indicators, pollutant concentrations and pathogen detection rates. a, b, and c are the fuzzy grading thresholds of the corresponding indicators, which are set according to food safety standards and are used to divide low-risk, medium-risk, and high-risk levels.

5. The intelligent detection method for foodborne disease monitoring platform according to claim 1, characterized in that: The S4 specifically includes: S41. Based on a preset fuzzy rule base, fuzzy reasoning is performed on the fuzzified data of the standardized food sample data set, wherein the fuzzy rule base includes a plurality of logical rules: if For high risk, For high risk, is high risk, then the overall risk of the food sample is high; if For medium risk, For high risk, If the risk is medium, the overall risk of the food sample is medium; if For low risk, For low risk, is low risk, then the overall risk of the food sample is low; S42, using fuzzy logic reasoning, perform combined calculations on the fuzzified indicators to generate a comprehensive fuzzy risk score μ Risk (i) Among them, w M 、w C 、w P are the weight coefficients of microbial indicators, pollutant concentrations and pathogen detection rates, respectively, satisfying w M +w C +w P =1; S43, the comprehensive fuzzy risk membership μ Risk (i) Perform defuzzification and convert it into a specific food sample risk index R i , the food sample risk score is used to characterize the potential foodborne disease risk level of a food sample: Among them, r is a continuous variable of risk score, r min and r max are the lowest and highest risk score values ​​respectively; μ Risk (r) is the comprehensive fuzzy membership function corresponding to the risk score r.

6. The intelligent detection method for foodborne disease monitoring platform according to claim 1, characterized in that: The S5 specifically includes: S51. Define the environmental status S of the detection platform t , the defuzzified food sample risk score R i and historical risk level data H t As the environment state input, form the current environment state S t =[R i ,H t ]; S52, using multi-layer neural network f θ (S t ) for the environmental state S t Perform feature extraction to generate information φ representing the current characteristics and potential risks of food samples t =f θ (S t ), where θ is the parameter set of the neural network; S53. Construct an asynchronous advantage actor-critic algorithm model using the actor-critic architecture: Actor Network π θ (a t |φ t ) with feature information φ t As input, the output risk assessment strategy a t , risk assessment strategy a t Including setting of risk warning trigger threshold T t and the choice of detection method t ; Critics Network Evaluate the value of the current strategy, use risk assessment accuracy as a value indicator, and calculate the state value function S54. Execute strategy a at time step t t , get instant rewards t , the immediate reward is defined according to the deviation between the actual risk assessment result and the expected result: in, is the actual observed risk level, R i Risk scores predicted for the asynchronous advantage actor-critic algorithm model; S55. Calculate advantage function A t , used to guide strategy optimization: Among them, γ is the discount factor, φ t+1 is the feature information of the next time step; S56. Using advantage function A t Guide the dynamic optimization of the risk assessment strategy and update the parameters of the actor network and the critic network: Actor network parameter update: Critic network parameter update: Among them, α θ and are the learning rates of the actor network and the critic network, respectively; S57. Through multi-threaded asynchronous updating, each thread independently processes different food sample data or sensor data, and after completing the asynchronous learning process, the parameters of each thread are updated to the global network.

7. The intelligent detection method for foodborne disease monitoring platform according to claim 1, characterized in that: The S6 specifically includes: S61. Risk assessment strategy based on asynchronous advantage actor-critic algorithm model, dynamically adjusting the risk warning trigger threshold to the optimal threshold for the current environment state S62, through each independent thread processing sensor or food sample data, at time step t, the current risk assessment strategy a t =[T t ,m t ] to make real-time adjustments so that the adjustment target of the risk warning trigger threshold meets the following optimization conditions: Among them, T t is the risk warning trigger threshold for the current time step, R t+1,i is the risk score prediction of the i-th food sample in the next time step, R target,i is the target risk level corresponding to the food sample, ω i is the weight factor of each sample, λ is the regularization parameter used to balance the risk score deviation and the amplitude of threshold adjustment, T base is the basic reference threshold; S63, using an asynchronous update mechanism, each thread independently performs strategy evaluation and threshold update to trigger the risk warning threshold T t Update the risk prediction results of each thread so that each thread in the asynchronous learning process can be executed in parallel; S64, after the thread completes the calculation, the optimal weight calculated by each thread is Update to the global network, the weight of the global network is updated as follows: in, represents the global weight vector before update, represents the updated global weight vector, represents the optimal weight obtained by the kth thread at time step t, K is the total number of threads, α w is the learning rate for global weight update, and β is the adjustment factor.

8. An intelligent detection system for a foodborne disease monitoring platform, characterized in that: Includes the following modules: A distributed data collection module is used to collect microbial indicators, pollutant concentrations, and pathogen detection rate data of food samples in real time through multiple sensors to build a food sample data set; The data preprocessing module performs data cleaning, denoising and standardization on the collected food sample data set to generate a standardized food sample data set; The fuzzy processing module performs fuzzy processing on the microbial indicators, pollutant concentrations and pathogen detection rate data of the standardized food sample data set, and generates fuzzy risk levels using a preset triangular membership function; The fuzzy reasoning module performs logical reasoning on the fuzzified data based on the preset fuzzy rule base, combines the fuzzy membership, generates a comprehensive risk membership through the reasoning process, and further generates a specific risk score for the food sample through defuzzification processing; The asynchronous risk assessment module uses the asynchronous advantage actor-critic algorithm to input the defuzzified food sample risk indicators into the environmental state of the detection platform, extracts feature information using a multi-layer neural network, outputs the risk assessment strategy through the actor-critic architecture, evaluates the value of the current strategy, and calculates the advantage function based on the immediate reward; The risk threshold dynamic adjustment module uses an asynchronous update mechanism to allow each thread to independently process food sample data and sensor data, dynamically adjust the risk warning trigger threshold to the optimal threshold for the current environmental state, and update the optimal weight of the thread to the global network after the asynchronous learning process is completed; Warning generation and information transmission module: when the risk score of a food sample reaches or exceeds the optimal threshold for triggering a risk warning, the system automatically generates foodborne disease risk warning information and sends the warning information to relevant departments or food supply chain management users through the information system.

Images