Virtual simulation and real-time decision-making system for production quality risk of canned fish

Abstract

The invention relates to the technical field of quality monitoring, in particular to a virtual simulation and real-time decision-making system for canned fish production quality risks, which is characterized in that data stability and jump accuracy are improved by adopting a filtering mode, a key path area is screened, a conduction trunk is formed, synchronous construction of a mapping and adjusting strategy is realized, and a real-time decision-making result is obtained. By introducing action time, a control time consumption difference value and a node jump amplitude trend multi-dimensional quantitative judgment standard into an execution structure, a feedback recognition mechanism of index fluctuation is enhanced, the abnormal trend response granularity is improved in cooperation with multi-dimensional nested calculation of temperature and pressure parameters, synchronous association of key control parameters and historical operation tracks in a fault chain is complemented, and the reliability of the fault chain is improved. Through starting from a beat logic structure, data cleaning, trend grouping and traceability return are performed alternately, so that a risk propagation chain obtains collaborative modeling among a pressure difference, a temperature difference and a control instruction, and risk path response consistency and intervention precision under a beat driving condition are enhanced.

Description

Technical Field

[0001] This invention relates to the field of quality monitoring technology, and in particular to a virtual simulation and real-time decision-making system for quality risks in canned fish production. Background Technology

[0002] In the field of quality monitoring technology, specifically in the food processing industry, intelligent systems for process control and risk management focus on the continuous collection, analysis, and feedback of key parameters in the production process. The aim is to ensure product quality stability, improve production efficiency, and reduce safety risks. In food processing scenarios, especially for canned fish products involving multiple key control points such as heat sterilization, sealing packaging, and shelf life, quality monitoring not only meets food safety standards but also has batch consistency and full-process traceability capabilities.

[0003] The core functions of the virtual simulation and real-time decision-making system for quality risks in canned fish production include: establishing a digital model of the process flow to implement virtual simulation; simulating the impact of parameter fluctuations in raw material pretreatment, packaging, and sterilization on quality; and combining real-time collected environmental and process data to predict and dynamically adjust potential quality risks. The aim is to build a closed-loop system from prediction to intervention to feedback, enhance quality assurance capabilities, reduce intervention errors, achieve early warning and precise intervention of quality problems in the canned fish production process, and improve product qualification rate and production line flexibility.

[0004] Current technologies for quality control still primarily rely on continuous monitoring of single sensor parameters and single-point outliers. In complex collaborative processes, the lack of a holistic, sequential cycle logic makes it impossible to accurately express the dynamic response relationships between parameters. Quality fluctuations are often corrected through local feedback, neglecting the hierarchical transmission characteristics of interference between upstream and downstream processes and cycle synchronization errors. This leads to issues such as time delays and repeated interventions in control behavior. Furthermore, the lack of distributed identification of abnormal behaviors driven by multiple variables results in unstable trigger locations and response mechanisms for the same fault in different batches, making it difficult to achieve unified adjustment logic. In terms of data processing, the lack of filtering and fusion mechanisms makes the data susceptible to instantaneous fluctuations and sensor drift interference, leading to inconsistent responses in quality control processes, inaccurate anomaly predictions, and lagging adjustment mechanisms. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a virtual simulation and real-time decision-making system for quality risks in canned fish production.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a virtual simulation and real-time decision-making system for quality risk in canned fish production, comprising: Process construction module: Based on the raw material handling section, preheating and packaging section, heat sterilization channel, cooling area and whole box packaging section in the fish canning production line, extract the flow rate set value and action cycle of the packaging tank, screen the nodes with peak difference in cycle time between sections, connect the cycle time structure, and establish a thermal cycle time connection diagram. Risk propagation module: Based on the aforementioned thermal cycle connection diagram, extract the pressure and cycle deviation data of the heat sterilization and packaging sections, use Kalman filtering to optimize the pressure sequence smoothing results, select the high-cycle pressure difference overlap section as the critical path, and establish the main structure for wave propagation; Control strategy module: Based on the wave conduction backbone structure, extract the beat setpoint and temperature difference control term, use graph neural network to enhance the identification ability of control association between nodes, determine the number of control actions of the sealing section and the beat difference range, extract matching parameters and construct mapping path, and establish a beat linkage adjustment parameter set; Dynamic evaluation module: Based on the aforementioned beat linkage adjustment parameter set, extract the encapsulation segment action time and outlet temperature difference change, locate the peak segment of jump amplitude, screen the segment combination that simultaneously satisfies the action trough value and jump peak value, and establish a sealing conflict frequency band map; Anomaly tracking module: Based on the sealing conflict frequency band spectrum, extract the inlet pressure, processing time and outlet temperature difference of the heat sterilization section, extract the parameter group and operation index within the abnormal control period, organize the inter-segment amplitude information, and establish an abnormal parameter tracing index set.

[0007] As a further aspect of the present invention, the thermal cycle timing diagram includes a sealing flow rate setting sequence, an inter-segment cycle peak index set, and a cycle connection path matrix; the wave conduction backbone structure includes a pressure direction consistency segment group, a jump node continuous index chain, and a cycle pressure difference overlapping segment set; the cycle linkage adjustment parameter set includes a cycle control mapping table, node action difference segments, and temperature difference joint control parameter pairs; the sealing conflict frequency band spectrum includes a jump amplitude node index, an action cycle valley segment, and a cycle temperature difference conflict control group; and the abnormal parameter tracing index set includes an abnormal operation time index group, a temperature and pressure parameter offset set, and an inter-segment amplitude change label table.

[0008] As a further aspect of the present invention, the process construction module includes: The beat positioning submodule: Based on the raw material handling section, preheating and packaging section, heat sterilization channel, cooling area and whole box packaging section in the fish canning production line, it extracts the flow rate set value of each packaging tank and matches it with the time index sequence, obtains the duration of each action cycle and binds it with the sequence number, then calculates the flow rate difference and action cycle difference between adjacent sections, filters the top two node sections in the beat peak difference value and locates the original index position, and generates a set of inter-segment beat mutation nodes; The structural splicing submodule: Based on the set of inter-segment beat mutation nodes, it extracts the order number of the preceding and following segments of the mutation nodes and generates an inter-segment connection index table, locates the insertion position of the sorting index within the beat difference segment, generates the corresponding segment processing order connection chain and updates the nodes synchronously, extracts the beat value of each segment in the sorted node sequence, splices them to generate a path diagram, and establishes a thermal beat connection diagram.

[0009] As a further aspect of the present invention, the risk propagation module includes: The differential pressure extraction submodule extracts pressure data from the heat sterilization and packaging sections based on the thermal beat connection diagram, aligns the beat index positions, uses Kalman filtering to perform state estimation and noise suppression on the pressure sequence, calculates the difference between the beat value and the corresponding pressure value, extracts continuously changing sections within the segment and judges the consistency of the difference direction, compares the difference sequence with the mean jump multiple and filters out abnormal intervals, marks the difference exceeding the segment to complete the segment classification and grouping, and establishes a pressure beat offset group. Jump positioning submodule: Based on the pressure beat offset group, extract jump index points within the segment and construct a lag time sequence, determine the continuity of the lag difference between adjacent indices and count the number of cross-segment connections, extract the beat value within the segment with a jump amplitude greater than a set threshold and map it to the jump index, construct an overlap rate matrix for jump points and select high-matching node segments, and establish a lag jump node set; Path construction submodule: Based on the set of lagging jump nodes, count the number of connection jumps between all node indices and determine whether they are continuously reachable. Extract node segments with consistent pressure direction and compare the range of jump amplitude changes. Filter segment pairs that meet the requirements of pressure difference direction synchronization and similar jump trend and establish connection chains. Concatenate and merge the segments according to the order of segment index to form a path graph and establish the main structure of wave transmission.

[0010] As a further aspect of the present invention, the execution process of the Kalman filter is as follows: based on the thermal cycle connection diagram, pressure data in the heat sterilization section and the packaging section are extracted and the cycle index positions are aligned. The start and end range of the pressure observation sequence is set. The difference between the observed value corresponding to each cycle index and the state value of the previous index is calculated. The state update step size is constructed in combination with the pressure change rate. It is determined whether the current pressure value falls within the covariance range of the estimated mean of the historical sequence state. If the deviation exceeds the set threshold, the observed value is corrected to the predicted value of the previous state value plus the residual adjustment value. The node state value and the estimation error covariate are updated. Then, the state of the entire pressure sequence is smoothed and noise is removed.

[0011] As a further aspect of the present invention, the screening of abnormal intervals specifically involves calculating the difference sequence between the beat value and the pressure value, extracting the absolute value of the difference points in the sequence and calculating the average difference amplitude, then setting a jump amplitude multiple threshold, calculating the ratio of the absolute value of each difference point to the average difference amplitude, screening out the beat index of the difference point where the ratio is greater than the set multiple threshold, using the continuously occurring beat indices exceeding the threshold to form candidate abnormal segments, and then judging the length valley value and removing segments that do not meet the segment length requirements, marking them as abnormal intervals.

[0012] As a further aspect of the present invention, the control strategy module includes: Parameter filtering submodule: Based on the wave conduction backbone structure, set the beat value of each beat node and extract the corresponding temperature difference control item and then match the node index. Use graph neural network to mine the structural features of temperature difference correlation between beat nodes, compare the node index with the temperature difference interval value and filter the peak node of temperature difference change and the adjacent beat range to generate a set of control parameter combinations. Action Judgment Submodule: Based on the control parameter combination set, it counts the number of action executions and the number of beat indices corresponding to each combination within the sealed segment, compares the difference range between the number of action executions and the beat, excludes combinations with insufficient difference, sorts the remaining combinations according to the time consumption, filters the combinations with the lowest time consumption, obtains the corresponding node and beat difference range within the combination, and generates the control action filtering result. Path generation submodule: Based on the control action filtering results, filter node indexes and sort them according to the beat order, then connect adjacent jump point indexes, judge the consistency of continuous jump amplitude between connected indexes and remove abnormal jump points, map the beat value to the temperature difference control item and form a node sequence mapping table, and establish a beat linkage adjustment parameter set.

[0013] As a further aspect of the present invention, the execution process of the graph neural network is as follows: based on the wave propagation backbone structure, the beat value and temperature difference control term of each beat node are extracted as initial node features; directed edge relationships between nodes in terms of physical connection order and beat similarity are constructed, and a beat node graph structure is generated; the state of each node is updated using a graph neural network; the weighted aggregation of node features and information between adjacent nodes are transmitted; in the update, the current node's own features and the weighted features of adjacent nodes are combined and a new node is output; after repeating multiple rounds of propagation, the node embedding vector is extracted; temperature difference gradient clustering is performed on the node embedding vector; the set of nodes whose temperature difference mutation amplitude is located outside the upper boundary of the average gradient interval of adjacent nodes is selected to obtain the node index and structural position.

[0014] As a further aspect of the present invention, the dynamic evaluation module includes: Action recognition submodule: Based on the beat linkage adjustment parameter set, extract the action time of the encapsulation segment and the change of the outlet temperature difference, locate the time interval of each beat index in the action sequence and extract the action duration, filter the segments with lower action time, map the corresponding temperature difference curve and extract the interval with obvious temperature difference jump, and generate a set of temperature difference jump nodes. Conflict screening stator module: Based on the set of temperature difference jump nodes, compare the corresponding action time with the temperature difference jump interval, count the beat synchronization segment and establish a cross-index table, filter the interval where the action valley value and the jump peak value coincide, output the beat matching segment index set, and establish the sealing conflict frequency band map.

[0015] As a further aspect of the present invention, the anomaly tracking module includes: Anomaly Extraction Submodule: Based on the sealing conflict frequency band spectrum, the inlet pressure curve of the heat sterilization section is sampled and a time index group is generated. The time segment and the outlet temperature difference curve are mapped and an aligned interval set is constructed. The index intersection of pressure peak, processing time valley and temperature difference jump occurs simultaneously. It is determined whether the three data co-occurs in a continuous segment. The time index and physical location of the co-occurrence segment are extracted and a synchronous anomaly segment index set is established. The parameter collection submodule, based on the synchronous abnormal segment index set, retrieves the control parameter group within the corresponding time period and extracts the adjustment frequency and operation range of each parameter. It statistically summarizes the frequency and generates a parameter jump interval index sequence. It sorts the amplitude increment changes of each control parameter between adjacent segments according to the time axis, extracts the peak segment of the increment difference and the corresponding control type, records the operation position of the control type and generates an attribution list, and establishes an abnormal parameter tracing index set.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: 1. In this invention, by using filtering to improve data stability and abrupt change accuracy, key path regions are screened and a transmission backbone is formed, high-value control regions are identified, high-frequency control indexes and beat mutation combinations are matched, and a node adjustment parameter system is constructed to achieve the synchronous construction of mapping and adjustment strategies; 2. In this invention, by introducing multi-dimensional quantitative judgment criteria such as action time, control time difference, and node jump trend into the execution structure, the feedback identification mechanism of index fluctuation is strengthened. Combined with multi-dimensional nested calculation of temperature and pressure parameters, the granularity of abnormal trend response is improved, and the synchronous association between key control parameters and historical operation trajectory in the fault chain is completed. 3. In this invention, by starting from the rhythm logic structure and interspersing data cleaning, jump screening, control identification, trend grouping and source tracing, the risk propagation chain is modeled collaboratively between pressure difference, temperature difference and control commands, thereby enhancing the consistency of risk path response and intervention accuracy under rhythm-driven conditions. Attached Figure Description

[0017] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] Example Please see Figure 1 This invention provides a technical solution: a virtual simulation and real-time decision-making system for quality risk in canned fish production, comprising: Process construction module: Based on the raw material handling section, preheating and packaging section, heat sterilization channel, cooling area and whole box packaging section in the fish canning production line, extract the flow rate set value and action cycle of the packaging tank, screen the nodes with peak difference in cycle time between sections, connect the cycle time structure, and establish a thermal cycle time connection diagram. Risk propagation module: Based on the thermal cycle connection diagram, extract the pressure and cycle deviation data of the heat sterilization and packaging sections, use Kalman filtering to optimize the pressure sequence smoothing results, select the high cycle pressure difference overlap rate as the critical path, and establish the main structure of wave propagation; Control strategy module: Based on the wave transmission backbone structure, extract the beat setpoint and temperature difference control term, use graph neural network to enhance the identification ability of control association between nodes, determine the number of control actions and beat difference range of the sealing section, extract matching parameters and construct mapping path, and establish a beat linkage adjustment parameter set; Dynamic evaluation module: Based on the beat linkage adjustment parameter set, extract the action time of the encapsulation section and the change of the outlet temperature difference, locate the peak segment of the jump amplitude, screen the segment combination that simultaneously satisfies the action trough value and the jump peak value, and establish the sealing conflict frequency band map; Anomaly tracking module: Based on the sealing conflict frequency band spectrum, extract the inlet pressure, processing time and outlet temperature difference of the heat sterilization section, extract the parameter group and operation index within the abnormal control period, organize the inter-segment amplitude information, and establish an abnormal parameter tracing index set.

[0021] The thermal cycle timing diagram includes the encapsulation flow rate setting sequence, the inter-segment cycle peak index set, and the cycle connection path matrix. The main structure of wave propagation includes pressure direction consistency segment groups, jump node continuous index chains, and cycle pressure difference overlapping segment sets. The cycle linkage adjustment parameter set includes the cycle control mapping table, node action difference segments, and temperature difference joint control parameter pairs. The sealing conflict frequency band spectrum includes jump amplitude node indexes, action cycle valley segments, and cycle temperature difference conflict control groups. The abnormal parameter tracing index set includes abnormal operation time index groups, temperature and pressure parameter offset sets, and inter-segment amplitude change label tables.

[0022] Please see Figure 2 The process construction module includes: The beat positioning submodule: Based on the raw material handling section, preheating and packaging section, heat sterilization channel, cooling area and whole box packaging section in the fish canning production line, it extracts the flow rate set value of each packaging tank and matches it with the time index sequence, obtains the duration of each action cycle and binds it with the sequence number, then calculates the flow rate difference and action cycle difference between adjacent sections, filters the top two node sections in the beat peak difference value and locates the original index position, and generates a set of inter-segment beat mutation nodes; The structural splicing submodule: Based on the set of inter-segment beat mutation nodes, it extracts the order number of the preceding and following segments of the mutation nodes and generates an inter-segment connection index table, locates the insertion position of the sorting index within the beat difference segment, generates the corresponding segment processing order connection chain and updates the nodes synchronously, extracts the beat value of each segment in the sorted node sequence, splices them to generate a path diagram, and establishes a thermal beat connection diagram. The beat positioning submodule extracts the flow rate setpoints of the packaging tanks in the raw material handling section, preheating and packaging section, heat sterilization channel, cooling area and whole case packaging section of the fish canning production line. It sets the time span and step length of each window, divides the time series data of each segment according to the window and constructs a continuous beat index sequence. It obtains the duration of each segment's action cycle by combining the original beat time axis and binds it to the corresponding sequence number. It calculates the average flow rate difference and action cycle difference between adjacent segments, establishes a beat difference list through the difference, sorts it according to the difference value, extracts the node segments with the top two difference magnitudes, tracks the original index position in the time series, and generates a set of inter-segment beat mutation nodes. The structural splicing submodule, based on the inter-segment beat change node set, extracts the upstream and downstream segments corresponding to the change nodes, determines their number positions in the processing sequence, initializes inter-segment connection relationships, establishes connection mappings between each group of node pairs and updates the connection status at corresponding positions, forms an inter-segment connection index table, marks the insertion order position in the beat difference segments, inserts the node information corresponding to each segment into the processing order list and updates the sorting relationship, extracts the beat values ​​of each segment in the sorted node sequence, aggregates and reorganizes the beat values ​​according to the node sequence order, connects the inter-segment paths of the updated node order and generates a path diagram, and establishes a thermal beat connection diagram. Please see Figure 2 The risk communication module includes: The differential pressure extraction submodule extracts pressure data from the heat sterilization and packaging sections based on the thermal beat diagram, aligns the beat index positions, uses Kalman filtering to perform state estimation and noise suppression on the pressure sequence, calculates the difference between the beat value and the corresponding pressure value, extracts continuously changing sections within the segment and judges the consistency of the difference direction, compares the difference sequence with the mean jump multiple and filters out abnormal intervals, marks the difference exceeding the segment to complete the segment classification and grouping, and establishes a pressure beat offset group. Jump Positioning Submodule: Based on the pressure beat offset group, it extracts jump index points within a segment and constructs a lag time sequence, determines the continuity of lag difference between adjacent indices and counts the number of cross-segment connections, extracts beat values ​​within a segment with jump amplitude greater than a set threshold and maps them to jump indices, constructs an overlap rate matrix for jump points and selects high-matching node segments, and establishes a lag jump node set. The path construction submodule: Based on the set of lagging jump nodes, it counts the number of connection jumps between all node indices and determines whether they are continuous and reachable. It extracts node segments with consistent pressure direction and compares the range of jump amplitude changes. It selects segments that meet the requirements of pressure difference direction synchronization and similar jump trends and establishes connection chains. It splices and merges the segments according to the order of the segment indices to form a path diagram and establishes the main structure of wave propagation. The differential pressure extraction submodule extracts pressure data from the heat sterilization and packaging sections based on the thermal cycle diagram, aligns the cycle index positions, uses Kalman filtering to estimate the state and suppress noise in the pressure sequence, sets the prediction matrix of the state transition model to a first-order linear time step matrix, and the observation matrix to an identity matrix, sets the initial covariance value to a diagonal matrix, defines the process noise covariance as a random disturbance within a fixed value range, executes the prediction step to obtain the current pressure estimation state, and performs collaborative correction using the pressure observation values ​​in the update step. After generating the state estimation sequence, it calculates the difference between the cycle value and the corresponding pressure value, extracts continuously changing segments within the segment and judges the consistency of the difference direction, compares the difference sequence with the mean jump multiple obtained by the moving average method, sets the jump threshold interval value and filters abnormal intervals, marks the segment exceeding the difference to complete the segment classification and grouping, and establishes a pressure cycle offset group. Jump Positioning Submodule: Based on the pressure beat offset group, it extracts jump index points within a segment and constructs a lag time sequence. It sets a lag interval threshold range, calculates the difference in the interval between the beat positions of adjacent consecutive index points, determines whether the lag difference is continuous and distributed within the threshold, and counts the number of index points in the cross-segment connection state. It extracts the beat values ​​within the segment with jump amplitude greater than the set threshold and maps them to the jump index. The amplitude threshold is set to twice the average pressure jump value within the segment. It uses the jump amplitude and the beat difference to construct a jump point index overlap rate matrix, filters node segments based on the overlap rate, and establishes a lag jump node set. The path construction submodule calculates the number of hops between each node index and counts the number of connected paths based on the hysteresis jump node set. It determines whether there are continuous reachable paths, sets the upper limit of the number of hops to the distance of three jumps, extracts node segments with consistent pressure direction signs, uses sign matching logic to compare the pressure change direction between nodes, uses an amplitude difference list to compare the jump amplitude change range, filters segment pairs with the same jump amplitude direction and amplitude within a set ratio range, establishes segment pair connection chains and splices each segment index, sorts the nodes and completes index reorganization, generates a path graph and establishes the main structure of wave propagation.

[0023] The Kalman filter process is as follows: Based on the thermal cycle connection diagram, pressure data in the heat sterilization section and the packaging section are extracted and the cycle index position is aligned. The start and end range of the pressure observation sequence is set. The difference between the observed value corresponding to each cycle index and the state value of the previous index is calculated. The state update step size is constructed in combination with the pressure change rate. It is determined whether the current pressure value falls within the covariance range of the estimated mean of the historical sequence state. If the deviation exceeds the set threshold, the observed value is corrected to the predicted value of the previous state value plus the residual adjustment value. The node state value and the estimation error covariate are updated. Then, the state of the entire pressure sequence is smoothed and noise is removed. Kalman filtering, according to the formula: ; in: Indicates time The pressure state update value, Indicates time The predicted value of the pressure state, Indicates time Kalman gain, Indicates time The pressure observation value, Represents the observation matrix. This represents the inlet pressure correction weighting coefficient. This indicates the increment of the pressure jump in the encapsulation section. This represents the weighting factor for the correction of the temperature difference during the beat. This indicates the increment of the cycle temperature difference between the heat sterilization section and the packaging section.

[0024] Execution process: First, extract the index time of the previous cycle based on the thermal cycle time connection diagram of the fish canning production line. The pressure state update value, combined with the state transition matrix With control input matrix And input the control action value of the encapsulation section. Generate beat index pressure state prediction values Then, using the observation matrix Match the pressure measurement points of the encapsulation section and calculate the observed predicted values. Pressure observations were collected in the heat sterilization and packaging sections. Obtain the observation residuals Extract the prediction error covariance matrix And combined with the observation noise covariance matrix Generating Kalman Gain The pressure sequence of the encapsulated segment is obtained by continuously switching segments. The difference between the average value of the switching segment and the average value of the adjacent segment is calculated to form the pressure jump amplitude increment. Then, extract the temperature difference between the heat sterilization section and the packaging section under the same cycle index, and calculate the temperature difference increment by calculating the temperature difference between adjacent sections. The ratio of the inlet pressure sequence variance to the overall pressure sequence variance within the sliding window is normalized to generate the inlet pressure correction weighting coefficient. The local variance of the temperature difference sequence between the packaging section and the heat sterilization section is normalized to the ratio of the mean to the variance to generate the beat temperature difference correction weighting coefficient. The Kalman gain correction term, pressure jump correction term, and temperature difference increment correction term are combined and superimposed to output the pressure state update value at the beat index. .

[0025] The specific steps for filtering abnormal intervals are as follows: calculate the difference sequence between the beat value and the pressure value, extract the absolute value of the difference points in the sequence and calculate the average difference amplitude, then set a jump amplitude multiple threshold, calculate the ratio of the absolute value of each difference point to the average difference amplitude, filter out the beat index of the difference point where the ratio is greater than the set multiple threshold, use the consecutively occurring beat indices exceeding the threshold to form candidate abnormal segments, and then judge the length valley value and remove segments that do not meet the segment length requirements, and mark them as abnormal intervals.

[0026] Please see Figure 2 The control strategy module includes: Parameter filtering submodule: Based on the wave conduction trunk structure, the module sets the beat value of each beat node and extracts the corresponding temperature difference control item, then matches the node index. The graph neural network is used to mine the structural features of the temperature difference correlation between beat nodes. The module compares the node index with the temperature difference interval value and filters the peak node of temperature difference change and the adjacent beat range to generate a set of control parameter combinations. Action Judgment Submodule: Based on the control parameter combination set, it counts the number of action executions and the number of beat indices corresponding to each combination within the closed segment, compares the difference range between the number of action executions and the beat, excludes combinations with insufficient difference, sorts the remaining combinations according to the time consumption, filters the combinations with the lowest time consumption, obtains the corresponding nodes and beat difference range within the combination, and generates the control action filtering results. Path generation submodule: Based on the control action filtering results, filter node indexes and sort them according to the beat order, then connect adjacent jump point indexes, judge the consistency of continuous jump amplitude between connected indexes and remove abnormal jump points, map the beat value to the temperature difference control item and form a node sequence mapping table, and establish a beat linkage adjustment parameter set; The parameter filtering submodule, based on the wave conduction backbone structure, sets the beat value of each beat node and extracts the corresponding temperature difference control item, then matches the node index. It uses a graph neural network to mine the structural features of the temperature difference correlation between beat nodes, constructs a directed graph structure with beat nodes as vertices and beat flow direction as edges. The node features include the current beat value, temperature difference value, and index position encoding. The edge weight is set as the linear normalized value of the absolute difference of the beat jump amplitude. The adjacency matrix is ​​weighted and aggregated. After the first layer aggregates the node adjacency features, it generates the intermediate layer by linear transformation and activation function mapping. After the second layer fuses the information of the whole graph, it outputs the temperature difference between nodes. The temperature difference mutation threshold is set as the upper limit of the dynamic standard deviation, and the peak nodes of temperature difference change and the adjacent beat range are filtered to generate a set of control parameter combinations. Action Judgment Submodule: Based on the control parameter combination set, it counts the number of times the corresponding action is executed and the number of beat indices for each combination within the closed segment, counts the execution records of actions within each group of nodes, sets the action execution threshold to directly exclude combinations with less than three executions, sets the upper limit based on the beat difference range to 75% of the peak value of the corresponding combination jump value, marks and removes combinations below the threshold range, retains the remaining combinations and sorts them by the average time of the beat segment within the combination from smallest to largest, sets the time valley judgment logic to sort the combinations with the average time in the top 20%, filters the combinations with time valleys, extracts the corresponding nodes and beat difference ranges within the combination, and generates the control action filtering results; The path generation submodule filters node indices based on the control action filtering results and sorts them according to the beat order. Then, it connects adjacent jump point indices and sets the peak value of the jump point connection index to twice the standard deviation of the jump amplitude of the historical paragraph. Nodes with a jump amplitude greater than the set value between two indices are marked as abnormal and removed. After retaining paragraphs with consistent jump amplitudes, the beat values ​​in the remaining nodes are mapped to the temperature difference control items in a one-to-one correspondence. The mapping logic is to establish a sequence table by binding the beat timestamp and the corresponding temperature difference value by matching the node index, establish a node sequence mapping table based on the mapping table, and splice the mapped paragraphs to complete the beat linkage adjustment parameter set.

[0027] The execution process of the graph neural network is as follows: Based on the wave propagation backbone structure, the beat value and temperature difference control term of each beat node are extracted as the initial node features. The directed edge relationship between nodes in terms of physical connection order and beat similarity is constructed, and the beat node graph structure is generated. The graph neural network is used to update the state of each node, and the weighted aggregation of node features and information between neighboring nodes are transmitted. In the update, the current node's own features and the weighted features of neighboring nodes are combined and a new node is output. After repeating multiple rounds of propagation, the node embedding vector is extracted. Temperature difference gradient clustering is performed on the node embedding vectors, and the set of nodes whose temperature difference mutation amplitude is located outside the upper boundary of the average gradient interval of adjacent nodes is selected to obtain the node index and structural position. Graph neural networks, according to the formula: ; in: For the first Layer nodes The state vector, For activation function, For the first Layer weight matrix, Represents a node The set of adjacent nodes, For nodes With adjacent nodes The temperature difference correlation attention coefficient For the first Layer Adjacency Nodes The state vector, For nodes Local temperature difference correction weighting coefficient For nodes The temperature gradient increment, For nodes Beat span correction weighting coefficient For nodes The increment of the difference between the beat value and that of adjacent nodes For the first Layer bias vector.

[0028] Execution process: First, extract the first wave propagation backbone structure... Layer beat node state vector Traverse the set of adjacent nodes of a node Inner adjacent nodes Extract the state vector of adjacent nodes Calculate the temperature difference correlation attention coefficient , For nodes With adjacent nodes The ratio of the mean to the local variance of the temperature difference interval is used to extract nodes. Local temperature gradient increment Local temperature difference correction weighting coefficient Based on nodes The ratio of local temperature difference peak value to global temperature difference standard deviation is normalized and nodes are extracted. Increment of the difference between the beat value and the beat values ​​of adjacent nodes Beat span correction weight coefficient For nodes The ratio of the peak beat span to the difference between the average beat span of upstream and downstream nodes will be used to transform the state vector of adjacent nodes. according to After weighted aggregation, the first Layer weight matrix Linear transformation, superimposed with local temperature gradient correction term With beat span correction term and bias Global input activation function Then output the updated node state vector. .

[0029] Please see Figure 2 The dynamic evaluation module includes: Action recognition submodule: Based on the beat linkage adjustment parameter set, extract the action time of the encapsulation segment and the temperature difference change at the outlet, locate the beat index of each segment in the time interval of the action sequence and extract the action duration, filter the segments with lower action time, map the corresponding temperature difference curve and extract the interval with obvious temperature difference jump, and generate a set of temperature difference jump nodes. The collision screening stator module: Based on the temperature difference jump node set, it compares the corresponding action time with the temperature difference jump interval, counts the beat synchronization segment and establishes a cross-index table, filters the interval where the action valley value and the jump peak value coincide, outputs the beat matching segment index set, and establishes the sealing collision frequency band map. Action recognition submodule: Based on the beat linkage adjustment parameter set, it performs interval overlap analysis on the encapsulated segment action time series and beat index, sets the window step size to 1 second, and the window width to 80% of the peak range of the action duration, calibrates the action segment corresponding to each beat point and extracts the temperature difference change value simultaneously, uses the first-order difference calculation function to extract the jump amplitude value of the temperature difference sequence, filters fluctuation points with amplitude greater than three times the standard deviation of the benchmark mean and matches the range of adjacent beat nodes to generate a set of temperature difference jump nodes; The collision screening stator module, based on the temperature difference jump node set, compares the start and end boundaries of the action time segment and the jump interval. It performs Boolean filtering on records where the overlapping start point is less than the start point of the jump interval and the end point is greater than the end point of the jump interval. It establishes a cross-tabulation of the beat and temperature difference jump segment index, compares the execution position of the segment where the action valley value and the jump peak value are located, and extracts the segments in the sequence where the number of consecutive overlaps exceeds a set threshold. It outputs a set of synchronous matching beat segment indexes and establishes a sealing collision frequency band map.

[0030] Please see Figure 2 The anomaly tracking module includes: Anomaly Extraction Submodule: Based on the sealing conflict frequency band spectrum, the inlet pressure curve of the heat sterilization section is sampled and a time index group is generated. The time segment and the outlet temperature difference curve are mapped and an aligned interval set is constructed. The index intersection of pressure peak, processing time valley and temperature difference jump occurs simultaneously. It is determined whether the three data co-occurs in a continuous segment. The time index and physical location of the co-occurrence segment are extracted and a synchronous anomaly segment index set is established. The parameter collection submodule, based on the synchronous abnormal segment index set, retrieves the control parameter group within the corresponding time period and extracts the adjustment frequency and operation range of each parameter. It summarizes the frequency and generates a parameter jump interval index sequence, sorts the amplitude increment changes of each control parameter between adjacent segments according to the time axis, extracts the peak segment of the increment difference and the corresponding control type, records the operation position of the control type and generates an attribution list, and establishes an abnormal parameter tracing index set. Anomaly Extraction Submodule: Based on the sealing conflict frequency band spectrum, the inlet pressure curve of the heat sterilization section is sampled for 1 second and extreme value index points are extracted to generate a complete time index group. This is mapped to the outlet temperature difference curve to construct an alignment interval set between the temperature difference curve and the pressure time segment. A three-dimensional intersection-exchange verification function is used to sequentially detect local peak points in the pressure curve, local valley points that appear in the heat sterilization treatment time, and inflection point index sets in the temperature difference jump curve where the difference exceeds the specified fluctuation threshold. The intersection of the three sets is determined and the index segments that appear synchronously within the continuous time window are marked. The time index of each co-occurrence point is located and matched with the physical segment position to generate a synchronous anomaly segment index set. The parameter collection submodule, based on the synchronous abnormal segment index set, retrieves the control parameter group within the corresponding time period of the abnormal segment, extracts the pressure regulation value, temperature difference setpoint, and key control quantities for executing valve opening and closing commands, calculates the number of numerical changes of each control quantity per unit time, counts and sorts the adjustment frequency, extracts the value range change interval of each control parameter on the time axis, calculates the incremental amplitude of the same parameter in adjacent segments with a window length of 4 time nodes, extracts the local extreme values ​​of the amplitude difference of the incremental sequence, obtains the peak segment position and the corresponding control type index, merges the control type and operation position and establishes a classification index table, and generates an abnormal parameter tracing index set.

[0031] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A virtual simulation and real-time decision-making system for quality risks in canned fish production, characterized in that, The system includes: Process construction module: Based on the raw material handling section, preheating and packaging section, heat sterilization channel, cooling area and whole box packaging section in the fish canning production line, extract the flow rate set value and action cycle of the packaging tank, screen the nodes with peak difference in cycle time between sections, connect the cycle time structure, and establish a thermal cycle time connection diagram. Risk propagation module: Based on the aforementioned thermal cycle connection diagram, extract the pressure and cycle deviation data of the heat sterilization and packaging sections, use Kalman filtering to optimize the pressure sequence smoothing results, select the high-cycle pressure difference overlap section as the critical path, and establish the main structure for wave propagation; Control strategy module: Based on the wave conduction backbone structure, extract the beat setpoint and temperature difference control term, use graph neural network to enhance the identification ability of control association between nodes, determine the number of control actions of the sealing section and the beat difference range, extract matching parameters and construct mapping path, and establish a beat linkage adjustment parameter set; Dynamic evaluation module: Based on the aforementioned beat linkage adjustment parameter set, extract the encapsulation segment action time and outlet temperature difference change, locate the peak segment of jump amplitude, screen the segment combination that simultaneously satisfies the action trough value and jump peak value, and establish a sealing conflict frequency band map; Anomaly tracking module: Based on the sealing conflict frequency band spectrum, extract the inlet pressure, processing time and outlet temperature difference of the heat sterilization section, extract the parameter group and operation index within the abnormal control period, organize the inter-segment amplitude information, and establish an abnormal parameter tracing index set.

2. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 1, characterized in that, The thermal cycle timing diagram includes a sealing flow rate setting sequence, a set of inter-segment cycle peak indexes, and a cycle connection path matrix. The wave propagation backbone structure includes pressure direction consistency segment groups, jump node continuous index chains, and a set of cycle pressure difference overlapping segments. The cycle linkage adjustment parameter set includes a cycle control mapping table, node action difference segments, and temperature difference joint control parameter pairs. The sealing conflict frequency band spectrum includes jump amplitude node indexes, action cycle valley segments, and cycle temperature difference conflict control groups. The abnormal parameter tracing index set includes an abnormal operation time index group, a set of temperature and pressure parameter offsets, and an inter-segment amplitude change label table.

3. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 1, characterized in that, The process construction module includes: The beat positioning submodule: Based on the raw material handling section, preheating and packaging section, heat sterilization channel, cooling area and whole box packaging section in the fish canning production line, it extracts the flow rate set value of each packaging tank and matches it with the time index sequence, obtains the duration of each action cycle and binds it with the sequence number, then calculates the flow rate difference and action cycle difference between adjacent sections, filters the top two node sections in the beat peak difference value and locates the original index position, and generates a set of inter-segment beat mutation nodes; The structural splicing submodule: Based on the set of inter-segment beat mutation nodes, it extracts the order number of the preceding and following segments of the mutation nodes and generates an inter-segment connection index table, locates the insertion position of the sorting index within the beat difference segment, generates the corresponding segment processing order connection chain and updates the nodes synchronously, extracts the beat value of each segment in the sorted node sequence, splices them to generate a path diagram, and establishes a thermal beat connection diagram.

4. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 1, characterized in that, The risk propagation module includes: The differential pressure extraction submodule extracts pressure data from the heat sterilization and packaging sections based on the thermal beat connection diagram, aligns the beat index positions, uses Kalman filtering to perform state estimation and noise suppression on the pressure sequence, calculates the difference between the beat value and the corresponding pressure value, extracts continuously changing sections within the segment and judges the consistency of the difference direction, compares the difference sequence with the mean jump multiple and filters out abnormal intervals, marks the difference exceeding the segment to complete the segment classification and grouping, and establishes a pressure beat offset group. Jump positioning submodule: Based on the pressure beat offset group, extract jump index points within the segment and construct a lag time sequence, determine the continuity of the lag difference between adjacent indices and count the number of cross-segment connections, extract the beat value within the segment with a jump amplitude greater than a set threshold and map it to the jump index, construct an overlap rate matrix for jump points and select high-matching node segments, and establish a lag jump node set; Path construction submodule: Based on the set of lagging jump nodes, count the number of connection jumps between all node indices and determine whether they are continuously reachable. Extract node segments with consistent pressure direction and compare the range of jump amplitude changes. Filter segment pairs that meet the requirements of pressure difference direction synchronization and similar jump trend and establish connection chains. Concatenate and merge the segments according to the order of segment index to form a path graph and establish the main structure of wave transmission.

5. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 4, characterized in that, The Kalman filter execution process is as follows: based on the thermal cycle connection diagram, pressure data in the heat sterilization section and the packaging section are extracted and the cycle index positions are aligned. The start and end range of the pressure observation sequence is set. The difference between the observed value corresponding to each cycle index and the state value of the previous index is calculated. The state update step size is constructed in combination with the pressure change rate. It is determined whether the current pressure value falls within the covariance range of the estimated mean of the historical sequence state. If the deviation exceeds the set threshold, the observed value is corrected to the predicted value of the previous state value plus the residual adjustment value. The node state value and the estimation error covariate are updated. Then, the state of the entire pressure sequence is smoothed and noise is removed.

6. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 4, characterized in that, The specific steps for screening abnormal intervals are as follows: calculate the difference sequence between the beat value and the pressure value, extract the absolute value of the difference points in the sequence and calculate the average difference amplitude, then set a jump amplitude multiple threshold, calculate the ratio of the absolute value of each difference point to the average difference amplitude, filter out the beat index of the difference point where the ratio is greater than the set multiple threshold, use the consecutively occurring beat indices exceeding the threshold to form candidate abnormal segments, and then judge the length valley value and remove segments that do not meet the segment length requirements, and mark them as abnormal intervals.

7. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 1, characterized in that, The control strategy module includes: Parameter filtering submodule: Based on the wave conduction backbone structure, set the beat value of each beat node and extract the corresponding temperature difference control item and then match the node index. Use graph neural network to mine the structural features of temperature difference correlation between beat nodes, compare the node index with the temperature difference interval value and filter the peak node of temperature difference change and the adjacent beat range to generate a set of control parameter combinations. Action Judgment Submodule: Based on the control parameter combination set, it counts the number of action executions and the number of beat indices corresponding to each combination within the sealed segment, compares the difference range between the number of action executions and the beat, excludes combinations with insufficient difference, sorts the remaining combinations according to the time consumption, filters the combinations with the lowest time consumption, obtains the corresponding node and beat difference range within the combination, and generates the control action filtering result. Path generation submodule: Based on the control action filtering results, filter node indexes and sort them according to the beat order, then connect adjacent jump point indexes, judge the consistency of continuous jump amplitude between connected indexes and remove abnormal jump points, map the beat value to the temperature difference control item and form a node sequence mapping table, and establish a beat linkage adjustment parameter set.

8. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 1, characterized in that, The execution process of the graph neural network is as follows: based on the wave propagation backbone structure, the beat value and temperature difference control term of each beat node are extracted as initial node features. Directed edge relationships between nodes in terms of physical connection order and beat similarity are constructed, and a beat node graph structure is generated. The graph neural network is used to update the state of each node, and the weighted aggregation of node features and information between adjacent nodes are transmitted. In the update, the current node's own features and the weighted features of adjacent nodes are combined and a new node is output. After repeating multiple rounds of propagation, the node embedding vector is extracted. Temperature difference gradient clustering is performed on the node embedding vectors, and the set of nodes whose temperature difference mutation amplitude is located outside the upper boundary of the average gradient interval of adjacent nodes is selected to obtain the node index and structural position.

9. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 1, characterized in that, The dynamic evaluation module includes: Action recognition submodule: Based on the beat linkage adjustment parameter set, extract the action time of the encapsulation segment and the change of the outlet temperature difference, locate the time interval of each beat index in the action sequence and extract the action duration, filter the segments with lower action time, map the corresponding temperature difference curve and extract the interval with obvious temperature difference jump, and generate a set of temperature difference jump nodes. Conflict screening stator module: Based on the set of temperature difference jump nodes, compare the corresponding action time with the temperature difference jump interval, count the beat synchronization segment and establish a cross-index table, filter the interval where the action valley value and the jump peak value coincide, output the beat matching segment index set, and establish the sealing conflict frequency band map.

10. The virtual simulation and real-time decision-making system for quality risk in canned fish production according to claim 1, characterized in that, The anomaly tracking module includes: Anomaly Extraction Submodule: Based on the sealing conflict frequency band spectrum, the inlet pressure curve of the heat sterilization section is sampled and a time index group is generated. The time segment and the outlet temperature difference curve are mapped and an aligned interval set is constructed. The index intersection of pressure peak, processing time valley and temperature difference jump occurs simultaneously. It is determined whether the three data co-occurs in a continuous segment. The time index and physical location of the co-occurrence segment are extracted and a synchronous anomaly segment index set is established. The parameter collection submodule, based on the synchronous abnormal segment index set, retrieves the control parameter group within the corresponding time period and extracts the adjustment frequency and operation range of each parameter. It statistically summarizes the frequency and generates a parameter jump interval index sequence. It sorts the amplitude increment changes of each control parameter between adjacent segments according to the time axis, extracts the peak segment of the increment difference and the corresponding control type, records the operation position of the control type and generates an attribution list, and establishes an abnormal parameter tracing index set.

Images