Intelligent central kitchen food safety monitoring method based on multi-dimensional sensing data

Abstract

The invention provides an intelligent central kitchen food safety monitoring method based on multi-dimensional sensing data, and relates to the technical field of food safety monitoring. The method comprises the steps that multifunctional partitions are divided according to the technological process, and key control points and control limit values are determined to generate a partition control framework; generating batch identifiers for the raw materials and the products and collecting multi-dimensional sensing data; time-space correction is completed through an anchor event and a synchronization signal, a material flow graph is constructed, and a batch boundary is deduced; extracting hazard feature data based on batch boundaries and performing fusion calculation on risk indexes; and finally, generating a treatment plan and verifying an execution state to obtain a monitoring result reflecting batch risk and treatment information. The problem that in the prior art, multi-dimensional sensing data fusion of the whole process of a central kitchen and dynamic evaluation and closed-loop verification of food safety risks cannot be achieved is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of food safety monitoring, in particular to an intelligent central kitchen food safety monitoring method based on multi-dimensional sensing data. BACKGROUND

[0002] The central kitchen covers multiple sub-zones and processes such as receipt inspection, primary processing, thermal processing, rapid cooling, sub-packaging, cold storage and freezing, turnover distribution, etc. Food safety risks come not only from the microbial load of raw materials and environmental exposure, but also from equipment status, personnel operation specifications and cross-zone cross-contamination. Existing digital monitoring focuses on temperature and humidity, cold chain history or single-process quality fluctuations. Patent document CN 103238161 B proposes collecting environmental information through mobile and fixed sensor tags during transportation and sales and calculating a quality index for pricing and remaining shelf life judgment, emphasizing quality perception at the circulation end. Patent document CN 120181290 A evaluates the shelf life of finished products through multi-sensor fusion and model prediction, and links storage and processing parameters to improve taste and stability, with the core goal focusing on quality and shelf life. Patent document CN 116933210 A focuses on filling equipment, builds a comprehensive quality deviation and abnormal confidence coefficient based on weight, liquid level and color, and is used for equipment anomaly early warning. It can be seen that the industry has the basis for multi-source data collection and quality evaluation, but it is mostly single-link or single-attribute optimization, and has not yet formed a food safety risk closed loop for the whole process and batch of central kitchens.

[0003] Regulators and enterprises are promoting the digitization of HACCP and the online of the five elements of man-machine-material-method-environment. The trend is reflected in the fusion and coverage of multi-modal sensing, including physical, chemical, biological, visual and process operation data. The governance approach moves from post-detection to predictive microbiology model-driven forward-looking control, from fixed thresholds to dynamic thresholds considering uncertainty, from simple alarms to executable disposal recommendations and process interlocking, and from scattered records to full-link traceability and evidence. At the same time, models and data drift due to seasonality and menus require a safe and auditable adaptive update mechanism. Although the technical route is clear, it still needs to unify hazard-specific dynamics modeling, evidence fusion, disposal closed loop and uncertainty management into a set of auditable system framework to truly land in the complex sub-zones and batch flow of central kitchens.

[0004] In view of the above trend, three patent documents make improvements, but still have key defects. The patent document with publication number CN120181290 A improves the fusion and prediction ability of multi-source data, and can link storage and processing parameters, but its optimization target is mainly shelf life and taste, lacks hazard dynamics modeling targeting Salmonella and Listeria monocytogenes, does not include food safety signals such as ATP swab, air biological particles and allergen rapid detection into a unified risk index, does not establish visual compliance detection and process interlocking of personnel dress and cross-zone behavior, and lacks a closed loop of post-disposal re-verification and uncertainty gating. The patent document with publication number CN103238161 B strengthens environmental perception and quality index quantification of the flow process, and supports dynamic sampling interval, but mainly covers transportation and sales terminals, does not face real-time hazard assessment of multiple critical control points in central kitchens, does not form a risk intervention and verification closed loop, and does not handle the uncertainty fusion of multi-source evidence. The patent document with publication number CN116933210 A realizes multi-dimensional monitoring and abnormal correction at the filling station, but the indicators focus on quality consistency and equipment performance, do not establish models and batch-level risk assessment targeting food safety hazards, and lack cross-partition linkage, batch traceability and post-disposal verification mechanisms. SUMMARY

[0005] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide an intelligent central kitchen food safety monitoring method based on multi-dimensional sensing data, which solves the problem that the prior art cannot realize multi-dimensional sensing data fusion and dynamic assessment and closed loop verification of food safety risks in the whole process of central kitchens.

[0006] To achieve the above purpose, the present application provides the following solutions: The application discloses a kind of intelligent central kitchen food safety monitoring methods based on multidimensional sensing data, comprising: according to process flow division multifunctional partition and determine the key control point and control limit value of the multifunctional partition, obtain partition control framework;Based on the partition control framework, for each batch of raw materials and intermediate products entering production generates unique batch identification and initial space-time reference, obtains batch reference model;Based on the batch modeling and reference, multidimensional sensing information is collected at key control point, and after the multidimensional sensing information is attached time stamp and source identification, centralized convergence is carried out, and after the multidimensional sensing information after convergence is associated with batch identification, multidimensional sensing data set is obtained;According to the multidimensional sensing data set, anchor event with time sequence characteristics is identified;Wherein, the anchor event includes: process start-stop, access opening and closing, weighing mutation, temperature platform change and alarm recovery;According to the anchor event and weak synchronization signal, time axis drift modeling and correction are carried out to the data source of key control point, and space-time alignment model is obtained;Based on the space-time alignment model, job record, equipment state, access and weighing signal are combined to construct material flow graph with time constraint;Based on the material flow graph and using sequence inference algorithm, the entering, exit node and residence time of batch in multi-partition are determined, and boundary confidence is quantified simultaneously, to obtain batch boundary model;Based on the batch boundary model, space-time constraint feature extraction is carried out to the multidimensional sensing data set, and hazard characteristic data is obtained;According to the hazard characteristic data, evidence fusion model is constructed, to calculate and obtain food safety risk index;Based on the food safety risk index, corresponding disposal plan is generated, and the execution state of the disposal plan is verified, to obtain food safety monitoring result comprehensively reflecting batch risk state and disposal information.

[0007] The application discloses the following technical effects: The application provides a kind of intelligent central kitchen food safety monitoring method based on multidimensional sensing data, comprising: according to process flow, multiple function subareas are divided and the key control points and control limits of the multiple function subareas are determined, to obtain subarea control framework;Based on the subarea control framework, a unique batch identification and initial space-time reference are generated for each batch of raw materials and intermediate products entering production, to obtain batch reference model;Based on the batch modeling and reference, multidimensional sensing information is collected at the key control points, and after the multidimensional sensing information is attached with time stamp and source identification, it is concentrated and converged, and the converged multidimensional sensing information is associated with batch identification, to obtain multidimensional sensing data set;According to the multidimensional sensing data set, anchor events with time sequence characteristics are identified;Wherein, the anchor events include: process start-stop, access control opening and closing, weighing mutation, temperature platform change and alarm recovery;According to the anchor events and weak synchronization signals, time axis drift modeling and correction are carried out on the data sources of key control points, to obtain space-time alignment model;Based on the space-time alignment model, job records, equipment status, access control and weighing signals are combined to construct material flow diagram with time constraints;Based on the material flow diagram and using sequence inference algorithm, the entering, exiting nodes and residence time of batch in multiple subareas are determined, and the boundary confidence is quantified, to obtain batch boundary model;Based on the batch boundary model, space-time constraint feature extraction is carried out on the multidimensional sensing data set, to obtain hazard feature data;According to the hazard feature data, evidence fusion model is constructed to calculate food safety risk index;Based on the food safety risk index, corresponding disposal plan is generated, and the execution state of the disposal plan is verified, to obtain food safety monitoring result comprehensively reflecting batch risk state and disposal information, the application corrects multi-source signal drift through space-time alignment model, constructs material flow diagram with time constraints, realizes accurate tracking of batch in multiple function subareas;Using sequence inference algorithm and batch boundary model, the spatial and temporal range of production process is quantified, to ensure the positioning accuracy of risk identification;And on this basis, through hazard feature extraction and evidence fusion model, dynamic calculation and quantitative evaluation of risk index are realized.Compared with the existing methods relying on artificial sampling or post-tracing, the application can realize real-time perception, accurate traceability and disposal closed-loop verification of food safety risk, improve the automation degree and response speed of monitoring system, and significantly enhance the food safety guarantee capability of central kitchen in complex production environment. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0009] Figure 1A flowchart of a smart central kitchen food safety monitoring method based on multidimensional sensor data is provided for an embodiment of the present invention. Figure 2 A flowchart illustrating the construction process of the partition control framework provided in this embodiment of the invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] like Figure 1 As shown, this invention provides an intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data, comprising: Step 100: Divide the multi-functional zones according to the process flow and determine the key control points and control limits of the multi-functional zones to obtain the zone control framework; Step 200: Based on the partition control framework, generate a unique batch identifier and initial spatiotemporal reference for each batch of raw materials and intermediate products entering production, and obtain a batch reference model; Step 300: Based on the batch modeling and benchmark, collect multi-dimensional sensing information at key control points, add timestamps and source identifiers to the multi-dimensional sensing information, and then aggregate it. The aggregated multi-dimensional sensing information is then associated with the batch identifier to obtain a multi-dimensional sensing dataset. Step 400: Identify anchor events with temporal characteristics based on the multidimensional sensor dataset; wherein, the anchor events include: process start / stop, access control opening / closing, sudden changes in weighing, temperature platform changes, and alarm recovery; Step 500: Based on the anchor event and weak synchronization signal, perform time axis drift modeling and correction on the data source of the key control point to obtain the spatiotemporal alignment model; Step 600: Based on the spatiotemporal alignment model, construct a time-constrained material flow diagram by combining work records, equipment status, access control, and weighing signals; Step 700: Based on the material flow diagram and using a sequence inference algorithm, determine the entry and exit nodes and dwell time of the batch in the multiple partitions, and quantify the boundary confidence to obtain the batch boundary model; Step 800: based on the batch boundary model, spatio-temporal constraint feature extraction is performed on the multi-dimensional sensing data set to obtain hazard feature data; Step 900: an evidence fusion model is constructed according to the hazard feature data to calculate a food safety risk index; Step 1000: a corresponding disposal plan is generated based on the food safety risk index, and the execution state of the disposal plan is verified to obtain a food safety monitoring result comprehensively reflecting the batch risk state and disposal information.

[0013] Further, as shown in Figure 1 The specific implementation process of step 100 is as follows: First, the process flow model and supplementary data are obtained, the spatial clustering and functional classification of each process node are performed, and the initial partition structure is obtained. Then, the historical production data and real-time monitoring samples in each initial partition are collected, and if the data is not complete, it is waited for collection to be completed; when the data meets the conditions, the correlation analysis of temperature, humidity, time and equipment running state is performed, and the key control point set is identified and extracted. Then, according to the key control point set, the upper and lower limits of the parameters are calculated to form the actual control interval, and then the control limit value data is obtained. Finally, the initial partition structure, key control point set and control limit value data are mapped in space and structure coded to realize the unified mapping output of process partition, control point and limit value constraint, complete the process parameter boundary construction and data structured result generation.

[0014] Specifically, the embodiment first acquires process flow topology data covering raw material receiving, cleaning processing, heat processing, sub-packaging, cooling and logistics out-of-warehouse, etc. for the central kitchen food material processing process. By clustering calculation on the spatial coordinates and equipment function information of each process node, the node group with physical adjacency and similar function is identified, and the initial multi-functional partition structure is established to ensure that the material flow path and process logic in the same region have high consistency in space, providing a spatial boundary basis for subsequent key control point extraction.

[0015] After obtaining the initial partition structure, the historical production data and current real-time monitoring samples of each partition are called, and the parameters such as temperature, humidity, processing time, equipment running state and energy consumption are extracted, and the correlation and deviation degree analysis model is used to identify the process variables that have a significant impact on food quality and safety as key control points. The identification process is completed by automatic feature screening, which can dynamically adjust the key control point set under different batches and seasonal working conditions, thereby ensuring the representativeness and stability of the control points.

[0016] The embodiment is further based on the historical process data distribution of the key control points and the corresponding regulatory requirements and equipment technical parameters, calculates the allowable fluctuation range of each control point parameter, and constructs the control limit data formed by the upper and lower limits of the indicators. Then, the initial partition structure, the key control point set and the control limit data are mapped into a unified data structure format, the structure coding is completed, and the partition control framework described in the embodiment is formed, realizing the standardized and batched generation of process constraints for multi-functional partitions.

[0017] Further, the specific implementation process of step 200 is as follows: The embodiment first extracts the multi-functional partition structure and the key control point set in the established partition control framework, analyzes the material flow transfer law and the corresponding process constraint relationship between different partitions, and determines the time and space scales and process connection conditions that affect batch identification. Through joint modeling of process start and stop signals, device access mode and partition boundary data, the spatial and process constraint condition set of batch identification is formed as the batch modeling input data. This process ensures that batch identification not only depends on material information, but also is subject to the time and space constraints of the actual production path, thereby providing data basis for subsequent dynamic batch modeling.

[0018] After determining the batch modeling input data, the embodiment collects the identity information of the raw materials and intermediate products flowing into each multi-functional partition at the entrance, including the raw material number, the entry time stamp, the device access signal and the material state identifier. Through matching operation of the above collected information and batch modeling input data, the effective batch entry event of the current partition is identified, and batch entry event data is generated. Then, according to the event trigger sequence, a combined unique batch identifier is formed using the time stamp, the raw material number and the partition code. This identifier can realize cross-partition tracing without relying on centralized database numbering, and effectively avoid batch confusion caused by data intersection or overlap between different partitions.

[0019] The embodiment further extracts the temperature, humidity, processing state and time characteristics of the batch at the initial stage of production according to the key control point parameters of the partition corresponding to the unique batch identifier, generates initial time and space reference data. Then, the unique batch identifier and the initial time and space reference data are bound one by one, and an index relationship structure is established, realizing unified retrieval within and outside the partition in the data storage layer. Through structured storage, a batch-level data entity is formed, and finally a batch reference model is obtained. This model can support the correlation mapping of subsequent multi-dimensional sensing data and the traceability analysis of the whole process of the batch, and provide an expandable batch modeling basis for intelligent food safety monitoring.

[0020] Further, the specific implementation process of step 300 is as follows: After the batch reference model is established, the embodiment first calls the model to determine the key control point positions in the multi-functional partition and the corresponding spatio-temporal coordinate ranges of each batch, forming a reference data for sensing collection. Through the reference data, the embodiment can realize collection boundary limitation by batch in the sensing layer, that is, only trigger the collection task within the control point range where the batch passes or stays, thereby avoiding the interference of irrelevant environmental data. This step ensures that the collected data is highly consistent with the spatio-temporal path of the batch, laying a foundation for subsequent multi-dimensional sensing information aggregation and batch mapping.

[0021] The embodiment secondly calls temperature, humidity, gas concentration, access control opening and closing, weighing change and alarm state and other multi-type sensing devices at each key control point to read signals in real time according to the reference data for sensing collection, forming original multi-dimensional sensing signals. For each sensing channel, the real-time collection module synchronously records the signal generation time and the physical identifier of the sensor, generating labeled sensing data containing the collection timestamp and the source identifier. The embodiment ensures the consistency of the sampling reference of various types of signals within the same time window through a device-level time synchronization mechanism, ensuring the comparability and fusion accuracy of multi-source sensing signals on the time axis.

[0022] The embodiment further aggregates the labeled multi-dimensional sensing data according to the spatial encoding and key control point number provided by the batch reference model, forming an aggregation data structure by partition, by control point and by time block. Then, through index mapping, the aggregated multi-dimensional sensing information is one-to-one corresponding to the corresponding unique batch identifier, realizing batch-level data aggregation. This process logically completes the mapping from device-level signals to batch entities, and finally obtains a structured multi-dimensional sensing data set, providing a unified data input basis for subsequent anchor event recognition and spatio-temporal alignment modeling.

[0023] Further, the specific implementation process of step 400 is: The embodiment first extracts the time series segments of temperature, humidity, load, gas concentration, access state and other multi-type signals by batch and key control point dimensions based on the aforementioned multi-dimensional sensing data set, and obtains continuous time series feature data through interpolation and time window sliding processing. This process ensures that each control point has comparability under the same time reference, so that the time series change can truly reflect the dynamic behavior in the process running process, providing complete time sequence input for subsequent anchor event recognition.

[0024] After obtaining the time series feature data, the embodiment uses mutation detection and trend analysis algorithms to calculate the instantaneous gradient, change rate and periodic response mode of the signal, identifies significant fluctuation segments and forms a candidate event set. The candidate events include possible process start-stop, load mutation, temperature control platform transition, access control action and alarm recovery, etc. Typical process behaviors. Through multi-scale detection, fast response events and slow state transitions can be captured at the same time, ensuring the comprehensiveness and robustness of event recognition.

[0025] The embodiment further performs multi-source feature matching on the candidate event set, jointly analyzes the time synchronization features, change trends and position encodings of different sensing signals, and determines the event type. For events that meet the key process node, state switching and abnormal recovery features, the embodiment is screened, labeled and classified to generate time series anchor point data. Finally, the time series anchor point data is aggregated and encoded according to the process stage, partition and time index to form an anchor event data set. The anchor event data set logically defines standard reference nodes for space-time calibration and batch alignment in the production process, providing accurate event-driven basis for subsequent space-time alignment models.

[0026] Further, the specific implementation process of step 500 is: The embodiment first takes the aforementioned anchor event data set as a time reference benchmark, extracts multi-dimensional sensing signal segments corresponding to anchor events at each key control point, and calculates the time offset and response delay characteristics between different data sources. By comparing the rising edge, peak value or steady state change point of each signal when the anchor event occurs, the embodiment establishes an initial time deviation estimation matrix for quantifying the time drift distribution of each data source within the same process stage. The estimation process simultaneously introduces a weak synchronization signal as an auxiliary correction basis, which utilizes its relative continuity and time transfer characteristics among multiple points to improve the stability and accuracy of time drift evaluation.

[0027] After obtaining the time offset estimation matrix, the embodiment uses a drift modeling function to describe the time drift evolution relationship of each sensing channel, and performs fitting and correction on data channels with nonlinear drift trends. Through a time series modeling method combining local linear regression and constrained smooth interpolation, the drift rate and relative offset of different time periods are gradually approximated to build a time axis drift correction model. This model can adaptively reflect the dynamic time deviation caused by sampling delay or communication differences of different sensors, and provide a functional expression for the unified alignment of multi-source signals.

[0028] The embodiment further applies the drift correction model to the multi-dimensional sensing data corresponding to the key control points, performs time mapping and compensation operations on the original time axis, outputs the corrected timestamp sequence, and jointly forms a unified space-time coordinate with the spatial index of each data source. The space-time alignment model generated thereby can realize the synchronized expression of multi-sensing signals in a unified reference space-time framework, and provide high-consistency raw data support for subsequent material flow tracking, batch boundary inference, and risk feature calculation.

[0029] Specifically, in establishing the space-time alignment model, the embodiment first extracts the signal response differences at different times for the multi-dimensional sensing signals at each key control point according to the comparison results of the anchor event data set and the weak synchronization signal, calculates the time offset of each sensing channel, and records the distribution law thereof at the spatial coordinate position. By fitting the time offset with time, a corrected time mapping relationship of the sensing data is formed, thereby realizing the resynchronization of the signals under the unified physical time reference. The core of the embodiment lies in that the original sensing data field is converted into a corrected synchronized sensing data field through the synergistic action of the three mechanisms of time drift compensation, spatial gradient correction, and event trigger correction, thereby ensuring the time consistency and spatial continuity of the signals of different partitions and different devices.

[0030] In the model implementation process, the time drift compensation amount is derived from the comparison results between the anchor event activation times. For example, when there is a delay of 0.8 seconds between the access opening signal and the temperature rise signal, the delay value is taken as the time correction input; the spatial gradient response coefficient is used to reflect the amplitude change of the signal attenuation or enhancement between adjacent sensing nodes, for example, the spatial gradient value is higher in the heating area, and the system linearly amplifies the signal difference by setting the response coefficient to restore the balanced distribution; the event correction coefficient is used to adjust the numerical offset under the trigger of a specific event, for example, when the alarm recovery event causes the temperature sensor to record abnormally high for a short time, the event correction coefficient can weaken the influence of the abnormal mutation.

[0031] In addition, the spatial gradient field is used here to describe the continuous change of the sensing data in the plane or functional partition direction, for example, on the same processing line, the temperature change collected by different sensors along the material flow direction can form a differentiable spatial gradient distribution; the anchor event activation function is used to represent the trigger state of a key event at a specific time or spatial position, and the value is positive when the event is triggered and zero when it is not, thereby realizing the dynamic control of the local correction term in the model. The initial values of the parameters can be obtained by fitting the historical data, for example, the time drift compensation amount can be averaged by ten alignment results; the spatial gradient response coefficient can be set by calculating the average gradient of the spatial temperature difference of the same type of device in the stable running stage; the event correction coefficient can be determined by data playback according to the alarm or human intervention frequency.

[0032] Further, the specific implementation process of step 600 is as follows: The embodiment first extracts the time series information of each key control point after time drift correction and its spatial coordinates in the partition layout based on the aforementioned space-time alignment model to form the aligned space-time data. Through this process, the embodiment ensures that all sensing signals are recorded under the same space-time reference framework, eliminating the time error and spatial displacement deviation between multi-source signals. For equipment nodes in different production partitions, such as cleaning tables, hot processing equipment, weighing tables, and cooling zone conveying ports, the embodiment establishes a time sequence sampling table for each node to record its start, pause, and completion times during batch flow to facilitate subsequent accurate calculation of the material flow path.

[0033] The embodiment further collects multi-source operation information corresponding to the aligned space-time data in time, including operator operation records, equipment running state switching logs, access control opening and closing signals, and weighing change sequences. Through time stamp matching, the above multi-source information is re-formatted according to a unified time sequence to generate multi-source operation correlation data. This data can reflect the comprehensive state of production activities within a specific time window, for example, when raw materials enter the cleaning area, the access control signal triggers the door opening record, the weighing signal causes a weight mutation, and the equipment state changes from standby to running. The embodiment enables each material event node to have a clear operation background through this multi-source information fusion, improving the accuracy of subsequent event recognition and path inference.

[0034] After obtaining the multi-source operation correlation data, the embodiment performs fusion analysis combined with the aligned space-time data, identifies continuous time change patterns and spatial displacement trajectories, determines the entry, processing, and exit nodes of materials in each partition, and thus constructs a material node set. For the time relationship and spatial continuity of events in the node set, the embodiment establishes flow transfer connections using the time difference and position sequence of adjacent nodes, and adds constraint parameters such as operation state, duration, and equipment load on each connection edge to generate material relationship data with time constraints. Finally, the embodiment encodes the relationship data into a graph structure to form a material flow graph with time constraints, realizing dynamic flow modeling of materials among multiple partitions and multiple equipment. This material flow graph can be used for subsequent batch boundary inference and risk propagation analysis, and also provides visual data support for operation scheduling and food safety monitoring of the central kitchen.

[0035] Further, the specific implementation process of step 700 is as follows: The present embodiment first extracts the time attribute information of the corresponding material flow transfer path and node of each batch in the multi-partition production process based on the aforementioned time-constrained material flow diagram to form batch flow transfer sequence data. This data records the entry time, device access, job completion and exit time of each batch in each partition, and maintains the continuity of time sequence. The present embodiment converts the complex graph structure information into time sequence data sequence with analyzability through this step, so that the subsequent path identification and stay analysis can be carried out in the time sequence dimension, thereby ensuring the traceability and quantifiability of the batch flow transfer characteristics.

[0036] After obtaining the batch flow transfer sequence data, the present embodiment identifies the batch cross-partition path by using a dynamic sequence inference-based analysis algorithm. This algorithm takes time continuity and spatial adjacency as the main characteristic criterion, extracts and pairs the signal transfer segments of the same batch between adjacent partitions to obtain a batch path segment set. For example, when the weighing signal and the access opening and closing signal of the same batch are continuous in time and the corresponding spatial coordinates are adjacent, the system determines it as a "partition transfer event". Subsequently, by calculating the time difference and spatial transfer direction of adjacent nodes, the present embodiment confirms the path integrity and flow transfer direction of each partition segment, providing basic data for the determination of batch segment entry and exit nodes.

[0037] The present embodiment further performs segment aggregation analysis on the batch path segment set to determine the entry node, exit node and corresponding stay duration of each batch in different partitions. For each partition segment, the present embodiment calculates the processing or waiting duration in that partition by time difference calculation, and integrates the results into batch segment stay data. At the same time, according to the batch segment stay data and the verification results of the sensing signals, the correct rate of entry node and exit node identification is calculated to obtain boundary confidence data reflecting the identification reliability. Finally, the stay data and confidence data are fused to form a batch boundary structure containing time boundary, spatial boundary and confidence index, and a batch boundary model is established. This model can accurately define the actual flow transfer range and duration of each batch in the multi-partition process, providing accurate data support for subsequent batch tracing, process tracing and risk propagation analysis.

[0038] Further, the specific implementation process of step 800 is: The embodiment first calls the aforementioned batch boundary model to extract the time boundary, spatial boundary and confidence parameter corresponding to each batch, forming batch spatio-temporal constraint data. The time boundary is used to limit the actual start and end time interval of the batch in the production process, the spatial boundary is used to define the activity range of the batch in the multi-partition layout, and the confidence parameter is used to represent the credibility of boundary identification. The embodiment filters and segments the multi-dimensional sensor data by taking these boundary constraints as filtering conditions, so that the subsequent feature extraction is strictly limited to the spatio-temporal range consistent with the real production track of the batch, thereby ensuring that the feature calculation result has good traceability accuracy and batch correspondence.

[0039] After obtaining the batch spatio-temporal constraint data, the embodiment extracts a subset of sensor signals within the range according to the time and space coordinates in the range from the multi-dimensional sensor data set, forming spatio-temporal constraint sensor data. This data only contains time series signals such as temperature, humidity, volatile gas concentration, equipment running state, access action collected during the existence of the batch and in the activity area. For the extracted data, the embodiment performs data cleaning and statistical standardization processing to eliminate abnormal points and communication loss segments, and eliminates the dimension difference between different sensors through the sliding mean and variance standardization method to obtain normalized sensor feature data. This process ensures that the subsequent feature statistics have comparability under different signal types and sampling frequencies, providing a stable input structure for hazard pattern recognition.

[0040] The embodiment further calculates a candidate hazard feature set based on the normalized sensor feature data, extracts indexes including temperature abnormal fluctuation amplitude, humidity sustained rising rate, volatile gas concentration sudden increase amplitude, equipment running state variation rate and access abnormal opening frequency. Then, these candidate features are combined with the batch spatio-temporal constraint data to filter and aggregate the feature items by analyzing their association with the boundary confidence parameter. For feature items with obvious abnormal association strength and spatio-temporal aggregation, the embodiment determines them as hazard feature data. The hazard feature data reflects the multi-dimensional change pattern highly related to safety risk within the effective spatio-temporal window of a specific batch, providing a reliable data basis for subsequent hazard event modeling and intelligent early warning.

[0041] Further, the specific implementation process of step 900 is: In calculating the food safety risk index, this embodiment constructs an evidence fusion model based on hazard characteristic data to comprehensively describe the joint impact of multidimensional spatiotemporal characteristic variables on the overall food safety status. This model obtains a continuous risk value reflecting the global risk level by weighted integral of hazard characteristic intensity over defined spatial coordinates and time domains. The core idea of ​​this embodiment is to map multi-source indicators from different spatial locations and time segments into measurable risk energy density, followed by weighted fusion using an evidence coupling function and exponential decay processing to form a comprehensive risk index.

[0042] In this embodiment, the hazard characteristic intensity function is used to represent the comprehensive intensity of hazard characteristics within a specified spatial location and time step. Its value can be calculated based on indicators such as temperature fluctuation amplitude, duration of abnormal humidity, peak concentration of volatile gases, equipment operating status variation rate, and access control frequency.

[0043] The integration domain defines the computational boundary of the model and can be determined based on the partition range and time interval obtained from the batch boundary model. For example, integration can be performed within the entire spatiotemporal range from the start of the processing line to the cooling zone, or from the start to the end of the batch. Through the above integration operation, this embodiment sums the hazard characteristic intensities corresponding to all spatial points and time steps after coupling and attenuation weighting, ultimately obtaining the food safety risk index. The value of this risk index generally ranges from 0 to 1, with higher values ​​indicating greater potential safety risks.

[0044] Specifically, during the night shift production process in the central kitchen, when conducting a risk assessment for a particular batch, based on factors such as increased temperature in the processing area, persistently high humidity, and abnormal odor sensor detection, the final risk index was 0.82, indicating a significant safety hazard. In the calculation, the hazard characteristic intensity can be represented as a risk intensity of approximately 0.7 when the temperature in the hot processing area exceeds the threshold by 4 degrees Celsius and the gas concentration increases by 18% after 15 minutes. Strong interactions exist between different characteristics, with an evidence coupling value close to 0.9, indicating that multiple anomalies are superimposed. The time decay coefficient is 0.15, indicating that the impact of an abnormal event 30 minutes ago on the current risk has decreased by approximately 80%. If the physical distance between the hot processing area and the adjacent cooling area is 12 meters, then the spatial distance is 12, used to describe the correlation of risk propagation. The integration range covers the production period from 8:00 AM to 10:00 AM and four zones: the cleaning area, the cutting and preparation area, the hot processing area, and the cooling area. The horizontal coordinates of the planar location are 0, 10, 20, and 30, respectively, and the functional zone labels are 1, 2, 3, and 4. The minimum granularity of time resolution is set to 10 seconds to continuously record changes in each sensor signal. By superimposing and weighting these data, a batch-level comprehensive risk result can be obtained, enabling the quantitative identification of food safety hazards.

[0045] Furthermore, the specific implementation process of step 1000 is as follows: After obtaining the food safety risk index, the embodiment first generates a corresponding disposal plan according to the numerical classification of the risk index. When the risk index is higher than 0.75, the embodiment immediately generates an emergency disposal plan, which includes closed operation instructions of controlled partition, on-site batch shutdown inspection and abnormal sample retention detection process; when the risk index is between 0.45 and 0.75, a preventive disposal plan is generated, which instructs to include restrictive operation, intensified cleaning, environmental sampling or local equipment maintenance; when the risk index is lower than 0.45, only a regular risk patrol plan is generated to maintain normal production state. The disposal plan is composed of four dimensions of risk type, affected partition, time window and responsible post, each of which is based on the model output result to ensure that the disposal content corresponds to the spatial and temporal characteristics of the risk source, so as to realize targeted response.

[0046] The embodiment further dynamically verifies the execution process of the plan after the disposal plan is generated. By reading the device running state signal, access control action record and cleaning and disinfection data, the embodiment judges whether the key instructions in the plan are executed within the specified time, such as whether the closed valve instruction is landed, whether the sampling device is running, and whether the abnormal batch is isolated. When it is found that the key instructions are not executed on time or the execution result does not meet the set verification condition, the embodiment will recalculate the disposal execution confidence to correct the state assessment of the current batch. At the same time, the execution steps and the original risk index are also compared back in the verification process, and the disposal effect is judged by the abnormal elimination rate and the residual risk ratio to provide parameter basis for subsequent plan adjustment.

[0047] The embodiment finally fuses the verified disposal result data with the risk index, batch identifier and space-time information to form a food safety monitoring result which comprehensively reflects the risk state of the batch and the disposal execution. The monitoring result includes risk level, execution accuracy, residual risk value and response time, which can intuitively show the whole process from risk identification to disposal implementation. When the monitoring result shows that the residual risk of the high-risk batch is reduced to below 0.3 after effective disposal, the system automatically confirms that the risk is removed; if the residual risk is still higher than 0.5, the secondary review is triggered and the production operation is temporarily suspended. Through the above steps, the embodiment realizes the closed-loop management of risk assessment, disposal decision and execution feedback, so that the food safety monitoring result not only has the early warning function, but also dynamically reflects the risk control effect, providing quantifiable basis for traceability and management decision.

[0048] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts between the embodiments can be referred to each other.

[0049] The principles and implementation manners of the present application are described by using specific examples in the present application, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, the specific implementation manners and application ranges will be changed according to the idea of the present application. In conclusion, the content of the present specification should not be understood as the limitation of the present application.

Claims

1. A method for monitoring food safety in an intelligent central kitchen based on multidimensional sensor data, characterized in that, include: Based on the process flow, a multi-functional zone is divided, and the key control points and control limits of the multi-functional zone are determined to obtain the zone control framework. Based on the aforementioned partition control framework, a unique batch identifier and initial spatiotemporal reference are generated for each batch of raw materials and intermediate products entering production, thus obtaining a batch reference model. Based on the batch modeling and benchmark, multidimensional sensing information is collected at key control points, and after adding timestamps and source identifiers to the multidimensional sensing information, it is centrally aggregated. The aggregated multidimensional sensing information is then associated with the batch identifier to obtain a multidimensional sensing dataset. Anchor events with temporal characteristics are identified based on the multidimensional sensor dataset; wherein, the anchor events include: process start / stop, access control opening / closing, sudden changes in weighing, temperature platform changes, and alarm recovery; Based on the anchor events and weak synchronization signals, time axis drift modeling and correction are performed on the data sources of key control points to obtain a spatiotemporal alignment model. Based on the aforementioned spatiotemporal alignment model, a material flow diagram with time constraints is constructed by combining work records, equipment status, access control, and weighing signals. Based on the material flow diagram and using a sequence inference algorithm, the entry and exit nodes and dwell time of the batch in multiple partitions are determined, and the boundary confidence is quantified to obtain the batch boundary model. Based on the batch boundary model, spatiotemporal constraint features are extracted from the multidimensional sensor dataset to obtain hazard feature data; An evidence fusion model is constructed based on the aforementioned hazard characteristic data to calculate the food safety risk index; Based on the food safety risk index, a corresponding disposal plan is generated, and the execution status of the disposal plan is verified to obtain a food safety monitoring result that comprehensively reflects the batch risk status and disposal information.

2. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, The process of dividing multi-functional zones according to the process flow and determining the key control points and control limits of the multi-functional zones to obtain the zone control framework includes: Obtain process flow topology data, wherein the process flow topology data includes: raw material receiving process node, cleaning and processing process node, heat processing process node, packaging process node, cooling process node, and logistics outbound process node; Spatial clustering and functional classification are performed on each process node in the process flow topology data to obtain an initial partition structure; Historical production data and real-time monitoring samples are collected in each initial partition, and correlation analysis is performed on temperature, humidity, time and equipment operating status to obtain a set of key control points; Based on the distribution of historical process data, regulatory limits, and equipment operating characteristics corresponding to the set of key control points, the upper and lower limits of parameters are calculated and a constraint interval is formed to obtain control limit data. The initial partition structure, the set of key control points, and the control limit data are mapped and structurally encoded to obtain the partition control framework.

3. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, Based on the partition control framework, a unique batch identifier and initial spatiotemporal reference are generated for each batch of raw materials and intermediate products entering production, resulting in a batch reference model, including: Based on the multi-functional partition information and key control points in the partition control framework, determine the spatial and technological constraints required for batch identification, and obtain the batch modeling input data. Collect the identity information, timestamps, and equipment access signals of raw materials and intermediate products entering each zone, and match them using the batch modeling input data to obtain batch entry event data; Based on the batch entry event data, a unique batch identifier is assigned to the raw materials and intermediate products, consisting of a combination of timestamp, raw material number and partition code, to obtain a unique batch identifier. Based on the key control point parameters corresponding to the unique batch identifier, the initial spatiotemporal reference data is obtained; The unique batch identifier is bound to the initial spatiotemporal reference data, indexed, and structured for storage to establish a batch-level data entity, thus obtaining the batch reference model.

4. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, Based on the batch modeling and benchmark, multidimensional sensor information is collected at key control points. This multidimensional sensor information is then aggregated after being appended with timestamps and source identifiers. The aggregated multidimensional sensor information is then associated with batch identifiers to obtain a multidimensional sensor dataset, including: By calling the batch baseline model, the key control points and spatiotemporal coordinate ranges corresponding to each batch are determined to obtain the sensor acquisition baseline data. Within the set of key control points, temperature, humidity, gas concentration, access control, weighing, and alarm status signals are acquired to obtain the original multidimensional sensing signals; Add an acquisition timestamp and corresponding sensor source identifier to the original multidimensional sensing signal to obtain labeled multidimensional sensing data; The labeled multidimensional sensing data are aggregated according to the spatial coding in the key control points and batch benchmark model to obtain aggregated multidimensional sensing information. The aggregated multidimensional sensing information is associated and mapped with the corresponding unique batch identifier to obtain a multidimensional sensing dataset.

5. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, The step of identifying anchor events with temporal characteristics based on the multidimensional sensor dataset includes: Based on the multidimensional sensor dataset, the time series signals corresponding to each key control point are extracted to obtain time series feature data; The instantaneous abrupt changes, gradual trends, and periodic patterns of the time series feature data are detected and analyzed to obtain a set of candidate events; Multi-source feature matching and event type determination are performed on the candidate event set to obtain the event classification results; The key process nodes, state transitions, and anomaly recovery events in the event classification results are filtered and labeled to obtain time-series anchor data; The time-series anchor data is aggregated and encoded, and the corresponding process stage, partition, and time index are identified to obtain the anchor event dataset.

6. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, The expression for the spatiotemporal alignment model is: ； in, The target sensing data field after spatiotemporal alignment correction; The original multidimensional sensing signal in two-dimensional space coordinates With time The distribution function under; The horizontal coordinates of the process zones or equipment in the plan layout; For spatial vertical or functional partition indexes; The physical time for signal acquisition; The time drift is obtained through anchor event identification and comparison with weak synchronization signals; The spatial gradient response coefficient; For spatial gradient fields; This is the event correction factor; The activation function for the anchor event in spatial coordinates and time The value is taken at that location.

7. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, The construction of a time-constrained material flow diagram based on the spatiotemporal alignment model, combined with work records, equipment status, access control, and weighing signals, includes: Based on the spatiotemporal alignment model, the time series information and spatial coordinates of key control points after time drift correction are obtained to obtain aligned spatiotemporal data. Collect work records, equipment operating status, access control opening and closing signals, and weighing change signals corresponding to the spatiotemporal data, and format them uniformly in chronological order to obtain multi-source work association data; The multi-source operation association data and the aligned spatiotemporal data are fused and analyzed to identify material entry, processing and exit event nodes in each partition, and a set of material nodes is obtained. Based on the temporal sequence and spatial continuity of events in the material node set, flow connections between nodes are established, and time differences and operation status constraints are added to each connection edge to obtain material relationship data with time constraints. The material relationship data is encoded in a graph structure to obtain a time-constrained material flow graph.

8. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, The process of determining the entry and exit nodes and dwell time of batches across multiple partitions based on the material flow diagram and using a sequence inference algorithm, while quantifying boundary confidence, to obtain a batch boundary model includes: Based on the time-constrained material flow diagram, extract the material flow path and node time attributes corresponding to each batch to obtain batch flow sequence data; Based on the batch flow sequence data, a sequence inference algorithm is used to identify the continuous path segments of each batch in different partitions, and the time difference and spatial transfer direction of adjacent nodes are calculated to obtain a set of batch path segments. The batch path segment set is aggregated to determine the entry and exit nodes of the batch in each partition, and the corresponding dwell time is calculated to obtain the batch segment dwell data. The entry and exit node identification accuracy is calculated based on the batch segment dwell data to obtain boundary confidence data. By integrating the batch segment dwell data with the boundary confidence data, a batch boundary structure including time boundary, spatial boundary and confidence index is constructed to obtain the batch boundary model.

9. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, Based on the batch boundary model, spatiotemporal constraint features are extracted from the multidimensional sensor dataset to obtain hazard feature data, including: Based on the batch boundary model, the temporal boundary, spatial boundary, and corresponding confidence parameters of each batch are extracted to obtain batch spatiotemporal constraint data. Based on the batch of spatiotemporal constraint data and the multidimensional sensing data, a subset of sensing signals within the current time and space range is selected to obtain spatiotemporal constraint sensing data. The spatiotemporal constraint sensing data is cleaned and statistically standardized to obtain normalized sensing feature data. Temperature, humidity, volatile gas concentration, equipment operating status variation rate, and access control frequency are calculated from the normalized sensor feature data to obtain a set of candidate hazard features; Based on the candidate hazard feature set, and combined with the boundary confidence parameters in the batch spatiotemporal constraint data, feature items with high anomaly correlation and high spatiotemporal clustering are selected to obtain hazard feature data.

10. The intelligent central kitchen food safety monitoring method based on multi-dimensional sensor data according to claim 1, characterized in that, The expression for the evidence fusion model is: ； in, This is the final calculated food safety risk index; In spatial coordinates With time The hazard characteristic intensity function is as follows; For evidence coupling function; This is the time decay coefficient; It is a spatial distance function; For the integration domain; The horizontal spatial coordinates of the partition on the production plane; For vertical or functional partition index coordinates; This is a micro time step.

Images