A tobacco benchmark system intelligent early warning auxiliary decision-making method and system

Abstract

The application discloses a tobacco pair benchmark system intelligent early warning auxiliary decision method and system, and particularly relates to the technical field of industrial internet intelligent manufacturing, and is used for solving the problems of chain deviation caused by single index optimization of existing tobacco production and low efficiency of artificial decision, identifying over-standard indexes and their associated index sets based on an antagonistic effect rule library, constructing a directed influence conduction path diagram to depict the constraint relationship between indexes, generating a non-dominated solution set satisfying multi-index benchmarks by analyzing historical big data, screening a Pareto frontier solution set in combination with disturbance propagation depth and equipment eigenmode stability, calculating risk entropy values through fluctuation variance and path complexity, outputting minimum entropy solution to a production execution system, fusing process knowledge and equipment state data, realizing multi-index collaborative optimization decision, effectively avoiding optimization deadlock, and improving production stability and resource utilization rate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial internet intelligent manufacturing, more specifically, the present application relates to a tobacco benchmarking system intelligent early warning auxiliary decision-making method and system. BACKGROUND

[0002] In the intelligent transformation process of tobacco manufacturing industry, enterprises generally establish multi-dimensional benchmarking index system (such as single box leaf consumption, cigarette filling value, energy consumption efficiency, etc.) to realize production state monitoring. The existing technology usually collects real-time data of production line through industrial internet platform, triggers single index early warning based on preset threshold, and relies on manual experience to adjust equipment parameters to respond to early warning. This mode relies on independent analysis of the deviation of each index from the preset benchmark, and does not fully consider the objective nonlinear constraint relationship between indexes.

[0003] Due to the complex nonlinear antagonistic effect between tobacco production parameters (such as reducing leaf consumption may lead to deterioration of filling value), the existing early warning system often triggers a chain of other key indicators deviating from the benchmark when triggering single index optimization decision, which causes optimization deadlock problem caused by implicit conflict between indexes, making it difficult for manual or conventional automatic system to generate a globally optimal decision scheme, resulting in continuous loss of production resources. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a tobacco benchmarking system intelligent early warning auxiliary decision-making method and system to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] A tobacco benchmarking system intelligent early warning auxiliary decision-making method, comprising:

[0007] S1, acquiring multi-dimensional benchmarking index data of tobacco production line in real time through industrial internet platform;

[0008] S2, when any dimension of benchmarking index data exceeds the preset threshold, marking it as an out-of-standard benchmarking index, and identifying a set of associated benchmarking indexes that have antagonistic effect with the corresponding out-of-standard benchmarking index based on the antagonistic effect rule base;

[0009] S3, constructing a directed influence conduction path graph of out-of-standard benchmarking index to associated benchmarking index set based on tobacco process knowledge base;

[0010] S4, based on historical big data analysis of nonlinear constraint relationship between benchmarking indexes in the directed influence conduction path graph, generating a non-dominated solution set that satisfies all associated benchmarking index benchmark values at the same time;

[0011] S5, filtering out a Pareto front solution set from the non-dominated solution set according to the disturbance propagation depth of the directed influence conduction path graph and the intrinsic modal stability of the equipment vibration signal;

[0012] S6, generating a risk entropy value of each solution in the Pareto front solution set by analyzing the fluctuation variance of the corresponding benchmark index and the complexity of the directed influence conduction path;

[0013] S7, selecting the solution with the minimum risk entropy value from the Pareto front solution set as the decision scheme and outputting it to the production execution system.

[0014] Further, the multi-dimensional benchmark index data of the tobacco production line is obtained in real time through the industrial internet platform, including:

[0015] The single-box leaf consumption data is collected from the silk making process, the cigarette filling value data is collected from the rolling process, and the energy consumption efficiency data is collected from the power plant;

[0016] The single-box leaf consumption data, cigarette filling value data and energy consumption efficiency data are integrated into multi-dimensional benchmark index data, which is transmitted in real time through the OPC UA protocol built in the industrial internet platform.

[0017] Further, when any dimension of the benchmark index data exceeds the preset threshold, it is marked as an over-standard benchmark index, and an associated benchmark index set that has an antagonistic effect with the corresponding over-standard benchmark index is identified based on an antagonistic effect rule library, including:

[0018] Each index value in the multi-dimensional benchmark index data is compared with the preset threshold;

[0019] When the single-box leaf consumption index value exceeds the single-box leaf consumption preset threshold, it is marked as an over-standard benchmark index;

[0020] When the cigarette filling value index value exceeds the cigarette filling value preset threshold, it is marked as an over-standard benchmark index;

[0021] When the energy consumption efficiency index value exceeds the energy consumption efficiency preset threshold, it is marked as an over-standard benchmark index;

[0022] Based on the antagonistic relationship mapping table between benchmark indexes stored in the antagonistic effect rule library, an associated benchmark index set that has an antagonistic effect with the over-standard benchmark index is identified.

[0023] Further, the antagonistic effect rule library is derived from the benchmark index conflict records obtained by analyzing the historical production data in the tobacco process knowledge base.

[0024] Further, a directed influence conduction path graph of the over-standard benchmark index to the associated benchmark index set is constructed based on the tobacco process knowledge base, including:

[0025] extracting process influence relationship data between the over-standard benchmark index and each benchmark index in the associated benchmark index set from the tobacco process knowledge base;

[0026] taking the over-standard benchmark index as a starting node of the directed influence conduction path graph;

[0027] taking the benchmark index in the associated benchmark index set as a terminal node of the directed influence conduction path graph;

[0028] establishing a directed edge between the starting node and the terminal node according to the process influence relationship data;

[0029] wherein the process influence relationship data includes direct influence relationship and indirect influence relationship of the benchmark index through intermediate processes;

[0030] the direction of all directed edges is from the over-standard benchmark index to the benchmark index in the associated benchmark index set.

[0031] Further, based on historical big data analysis of the nonlinear constraint relationship between the benchmark indexes in the directed influence conduction path graph, a non-dominated solution set satisfying all associated benchmark index reference values is generated, including:

[0032] extracting historical production data of each benchmark index in the directed influence conduction path graph from historical big data;

[0033] for each two benchmark indexes connected by a directed edge in the directed influence conduction path graph, analyzing the numerical change correlation between the corresponding two benchmark indexes in the historical production data;

[0034] establishing a nonlinear constraint function expression between the benchmark indexes according to the numerical change correlation;

[0035] taking the simultaneous achievement of reference values of all benchmark indexes in the associated benchmark index set as the goal, generating a feasible solution space under the constraint condition of the nonlinear constraint function expression;

[0036] selecting a solution set in which there is no other feasible solution that is superior to the corresponding solution on all benchmark indexes as a non-dominated solution set in the feasible solution space;

[0037] wherein the nonlinear constraint function expression contains a combination of polynomial terms and logarithmic terms.

[0038] Further, according to the disturbance propagation depth of the directed influence conduction path graph and the intrinsic modal stability of the equipment vibration signal, a Pareto frontier solution set is selected from the non-dominated solution set, including:

[0039] for each non-dominated solution in the non-dominated solution set, determining the corresponding conduction path of the corresponding non-dominated solution in the directed influence conduction path graph;

[0040] a path length of a corresponding conducting path from the over-standard indicator node to a farthest associated over-standard indicator termination node is calculated as a disturbance propagation depth;

[0041] An equipment vibration signal corresponding to the corresponding non-dominated solution is obtained;

[0042] The equipment vibration signal is subjected to an intrinsic modal decomposition to obtain a dominant modal component;

[0043] An energy concentration degree of the dominant modal component is calculated as an intrinsic modal stability;

[0044] When the disturbance propagation depth does not exceed a preset propagation depth threshold and the intrinsic modal stability is not lower than a preset stability threshold, the corresponding non-dominated solution is retained;

[0045] All the retained non-dominated solutions constitute a Pareto front solution set.

[0046] Further, by analyzing the over-standard indicator fluctuation variance corresponding to each solution and the directed influence conducting path complexity, a risk entropy value of each solution in the Pareto front solution set is generated, including:

[0047] For each Pareto front solution in the Pareto front solution set, historical production data corresponding to the corresponding Pareto front solution is obtained;

[0048] Based on the historical production data, a fluctuation variance of each over-standard indicator corresponding to the corresponding Pareto front solution within a preset time window is calculated;

[0049] A conducting path corresponding to the corresponding Pareto front solution in the directed influence conducting path graph is determined;

[0050] A product of a total number of all nodes in the corresponding conducting path and a total number of all directed edges is calculated as the directed influence conducting path complexity;

[0051] The fluctuation variances of each over-standard indicator are summed to obtain a total fluctuation variance;

[0052] The total fluctuation variance is multiplied by the directed influence conducting path complexity to obtain a risk entropy value of the corresponding Pareto front solution;

[0053] The preset time window is determined according to a tobacco production batch cycle.

[0054] Further, the solution with the minimum risk entropy value is selected from the Pareto front solution set as a decision scheme and output to a production execution system, including:

[0055] The risk entropy values of all the Pareto front solutions in the Pareto front solution set are compared;

[0056] The Pareto front solution with the minimum risk entropy value is determined;

[0057] Convert the benchmark index value in the Pareto frontier solution with the minimum risk entropy value into the equipment control parameter;

[0058] Send the equipment control parameter to the production execution system through the industrial internet platform;

[0059] The equipment control parameter includes the cigarette machine cutter head rotating speed parameter, the drying machine temperature parameter and the flavoring machine flow parameter;

[0060] The equipment control parameter and the benchmark index value in the decision scheme are mapped through the conversion rule of the tobacco process knowledge base.

[0061] In another aspect, the present application provides a tobacco benchmark system intelligent early warning auxiliary decision system, comprising:

[0062] The data acquisition module is used for acquiring the multi-dimensional benchmark index data of the tobacco production line in real time through the industrial internet platform;

[0063] The antagonistic recognition module is used for marking the benchmark index as an over-standard benchmark index when the benchmark index data of any dimension exceeds the preset threshold value, and identifying the associated benchmark index set having an antagonistic effect with the corresponding over-standard benchmark index based on the antagonistic effect rule base;

[0064] The conduction construction module is used for constructing a directed influence conduction path graph of the over-standard benchmark index to the associated benchmark index set based on the tobacco process knowledge base;

[0065] The solution set generation module is used for generating a non-dominated solution set satisfying the benchmark values of all associated benchmark indexes based on the nonlinear constraint relationship between the benchmark indexes in the directed influence conduction path graph through historical big data analysis;

[0066] The frontier screening module is used for screening the Pareto frontier solution set from the non-dominated solution set according to the disturbance propagation depth of the directed influence conduction path graph and the intrinsic modal stability of the equipment vibration signal;

[0067] The entropy value generation module is used for generating the risk entropy value of each solution in the Pareto frontier solution set by analyzing the benchmark index fluctuation variance and the directed influence conduction path complexity corresponding to each solution;

[0068] The scheme output module is used for selecting the solution with the minimum risk entropy value from the Pareto frontier solution set as the decision scheme and outputting the decision scheme to the production execution system.

[0069] Compared with the prior art, the present application has the following beneficial effects:

[0070] 1. By constructing a directed influence conduction path diagram between multi-dimensional indicators, the explicit modeling and quantitative analysis of antagonistic effect in the field of tobacco production are realized, the non-dominated solution set is generated based on the industrial big data mining of the nonlinear constraint relationship between indicators, the limitation of traditional single indicator optimization is broken through, the optimization deadlock problem is eliminated from the root, the scheme with the strongest process robustness is accurately identified in the Pareto frontier solution set through the double-dimensional screening mechanism of disturbance propagation depth and equipment intrinsic modal stability, and the chain fluctuation risk caused by decision-making is significantly reduced.

[0071] 2. By fusing a dynamic risk entropy evaluation model, the historical indicator fluctuation variance and conduction path complexity are included in the decision factor, the quantitative prediction of production risk is realized, the equipment control parameters output based on the risk entropy minimization criterion ensure the executability through the rule mapping of the tobacco process knowledge base, the multi-objective dynamic collaborative optimization is realized through big data analysis technology under the industrial internet architecture, the resource utilization efficiency is effectively improved, and the quantifiable decision basis is provided for tobacco intelligent manufacturing. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 A flowchart of a tobacco benchmarking system intelligent early warning auxiliary decision-making method of the present application;

[0073] Figure 2 A structural schematic diagram of a tobacco benchmarking system intelligent early warning auxiliary decision-making system of the present application. DETAILED DESCRIPTION

[0074] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0075] Embodiment 1: Figure 1 A tobacco benchmarking system intelligent early warning auxiliary decision-making method of the present application is given, which comprises:

[0076] S1, acquiring multi-dimensional benchmarking indicator data of a tobacco production line in real time through an industrial internet platform;

[0077] S2, when any-dimensional benchmarking indicator data exceeds a preset threshold value, marking the benchmarking indicator as an out-of-standard benchmarking indicator, and identifying a set of associated benchmarking indicators that exist antagonistic effect with the corresponding out-of-standard benchmarking indicator based on an antagonistic effect rule base;

[0078] S3, constructing a directed influence conduction path diagram from the out-of-standard benchmarking indicator to the set of associated benchmarking indicators based on a tobacco process knowledge base;

[0079] S4, generating a non-dominated solution set satisfying all the benchmark indexes based on the nonlinear constraint relationship between the benchmark indexes in the directed influence conduction path diagram;

[0080] S5, selecting a Pareto front solution set from the non-dominated solution set according to the disturbance propagation depth of the directed influence conduction path diagram and the intrinsic modal stability of the equipment vibration signal;

[0081] S6, generating the risk entropy value of each solution in the Pareto front solution set by analyzing the benchmark index fluctuation variance and the directed influence conduction path complexity corresponding to each solution;

[0082] S7, selecting the solution with the minimum risk entropy value from the Pareto front solution set as the decision scheme and outputting it to the production execution system.

[0083] S1, acquiring multi-dimensional benchmark index data of the tobacco production line in real time through the industrial internet platform, which specifically includes:

[0084] When acquiring multi-dimensional benchmark index data of the tobacco production line in real time through the industrial internet platform, first, a dynamic weighing sensor is installed on the outlet conveyor belt of the cutting machine in the primary processing procedure to collect single-box leaf consumption data, for example, a general industrial dynamic weighing sensor is used to record the total mass of tobacco passing through the conveyor belt at a fixed time interval and divide it by the number of cigarette boxes produced in that period, obtaining single-box leaf consumption data in units of kilograms per box. A laser volume scanner is installed on the cigarette forming section of the cigarette maker in the rolling procedure to collect cigarette filling value data, for example, a general industrial laser scanning device is used to calculate the cigarette filling value data in units of cubic centimeters per gram by scanning the volume and mass of the cigarette. Intelligent electric meters are installed on the steam main pipe and air compressor main pipe in the power workshop to collect energy consumption efficiency data, for example, general industrial intelligent electric meters are used to record energy consumption and calculate energy consumption efficiency data in units of kilograms of standard coal per ten thousand based on production data.

[0085] When the single-box consumption data, the cigarette filling value data and the energy consumption efficiency data are integrated into multi-dimensional benchmarking index data, a data alignment operation is performed on an edge computing node of the industrial internet platform, specifically including timestamp alignment processing and data unit standardization processing. The timestamp alignment processing refers to that when the recording time of the single-box consumption data is a specific moment, the cigarette filling value data takes the average value of the data within a specific time period before and after the moment, and the energy consumption efficiency data takes the data of a specific period of time including the moment, so as to ensure that the three index data have the same time reference. The data unit standardization processing refers to that the single-box consumption data is kept to a specific number of decimal places and expressed in kilograms per box, the cigarette filling value data is kept to a specific number of decimal places and expressed in cubic centimeters per gram, and the energy consumption efficiency data is kept to a specific number of decimal places and expressed in kilograms of standard coal per ten thousand, to form a structured data table including a timestamp field, a single-box consumption field, a cigarette filling value field and an energy consumption efficiency field.

[0086] When the multi-dimensional benchmarking index data is transmitted in real time through the OPC UA protocol built in the industrial internet platform, the following communication process is specifically performed: the edge computing node encapsulates the structured data table into a specific format message conforming to the OPC UA data model, which includes a field storing index values, a field storing data quality codes and a field storing collection timestamps. The encapsulated message is transmitted to the platform data center through a specific communication tunnel, an encryption protocol is enabled during the transmission process, the certificate validity period is set to a specific number of days, and the transmission frequency is a specific hertz value. When the network delay exceeds a specific millisecond value, a data compression algorithm is automatically enabled, which processes the data fields in a specific encoding manner. After receiving the message, the data center performs validity verification, specifically including value range verification and time continuity verification. The value range verification refers to that the single-box consumption value is in a specific kilogram per box interval, the cigarette filling value is in a specific cubic centimeter per gram interval, and the energy consumption efficiency value is in a specific kilogram of standard coal per ten thousand interval, which are marked as valid. The time continuity verification refers to that the difference between the current recording timestamp and the last recording timestamp is within a specific second value range, which is marked as valid. The data passing the verification is stored in a specific data partition of the time series database, which is indexed according to the cigarette brand and production shift.

[0087] A redundant safeguard mechanism is established during data transmission, and when a network interruption exceeding a certain second value is detected, it automatically switches to a local cache queue, which uses a specific storage structure and has a maximum capacity of a certain number of records. When the network is restored, the data is transmitted according to the first-in, first-out principle. At the same time, transmission priority policies are set, such as single-box leaf consumption data marked as high priority, cigarette filling value data marked as medium priority, and energy efficiency data marked as low priority. When the bandwidth is lower than a certain megabit per second value, high-priority data is transmitted first. The communication status of the industrial internet platform is monitored through a heartbeat package mechanism. The edge computing node sends a heartbeat package containing device identification information and resource usage status to the platform every certain number of seconds. If the platform does not receive a heartbeat package for more than a certain number of seconds, it triggers a device offline alarm.

[0088] The data acquisition device performs calibration operations regularly. The dynamic weighing sensor performs full-scale calibration using standard weights at a certain period. The calibration process records the zero drift value and the linear error of the range. When the zero drift exceeds a certain percentage of the full-scale range or the linear error exceeds a certain percentage, a calibration anomaly alarm is triggered. The laser volume scanner performs spatial resolution verification using standard geometric bodies at a certain period. The verification standard is to identify volume deviations above a certain millimeter value. The smart meter is verified by the measurement agency at a certain period. If the verification result does not meet the measurement accuracy requirements, the data acquisition function is stopped. All calibration records are stored in the device management log and associated with the production data of the corresponding period.

[0089] The storage structure of multi-dimensional benchmarking index data uses a hierarchical design. The raw acquisition layer retains the unprocessed data output directly by the sensor. The business integration layer stores the timestamp-aligned structured data table. The application service layer stores the unit-standardized transmission data. The data retention period for each layer is a certain number of days for the raw acquisition layer, a certain number of days for the business integration layer, and a certain number of years for the application service layer. Data exceeding the retention period is automatically archived to cold storage media. Data access permissions are allocated according to production roles, such as process engineers who can read and write all fields, device maintenance personnel who can only read single-box leaf consumption fields, and energy administrators who can only read energy efficiency fields. Each data access generates an audit log, which includes the operation time, operator identification information, and accessed field name.

[0090] The hardware deployment of the industrial internet platform adopts a distributed architecture, the edge computing node is deployed in the workshop site using an industrial control computer, the data center is deployed in the enterprise data center using a server, and the two are connected through an industrial network. The platform software system is deployed based on container technology, the data collection service runs in a container, the container resources are configured with a specific number of central processors and a specific capacity of memory, and when the resource usage rate exceeds a specific percentage, a new container instance is automatically started. The complete acquisition cycle of multi-dimensional benchmarking index data is controlled within a specific millisecond value, and the total delay from sensor acquisition to storage in the time series database does not exceed a specific millisecond value. When a single device fails, the platform automatically migrates the collection task to a standby device in the same workshop, and the data loss rate during the migration process is less than a specific thousandth value.

[0091] S2, when any dimension of the benchmarking index data exceeds the preset threshold, it is marked as an out-of-benchmark benchmarking index, and based on the antagonistic effect rule library, a set of associated benchmarking indexes that have antagonistic effects with the corresponding out-of-benchmark benchmarking index are identified, and the specific implementation includes:

[0092] After acquiring the multi-dimensional benchmarking index data in real time, each index value in the multi-dimensional benchmarking index data is compared with the preset threshold. First, load the preset threshold configuration table from the tobacco process knowledge base. The configuration table includes three fields: single-box leaf consumption preset threshold, cigarette filling value preset threshold, and energy consumption efficiency preset threshold. The single-box leaf consumption preset threshold is set to different values according to the cigarette brand, for example, some brands are set to a specific kilogram per box value. The threshold is determined based on the statistical distribution characteristics of the single-box leaf consumption data of the brand in the historical production period. Specifically, a reasonable boundary value is set by analyzing the historical data fluctuation range. The cigarette filling value preset threshold is set according to the cigarette specification, for example, some specifications are set to a specific cubic centimeter per gram value. The threshold determination method is based on the technical lower limit value determined by the physical property test of the cigarette. The energy consumption efficiency preset threshold is dynamically adjusted according to the production season, for example, different seasons are set to different kilogram of standard coal per ten thousand values. The adjustment basis is the influence law of environmental parameter changes on energy consumption.

[0093] The comparison process is performed on the real-time analysis node of the industrial internet platform, and a time window mechanism is used to process the data stream. When the single-box leaf consumption index value in the multi-dimensional benchmarking index data exceeds the single-box leaf consumption preset threshold, it is marked as an out-of-benchmark benchmarking index. The marking operation includes generating a record containing the out-of-benchmark index name, the out-of-benchmark time point, the measured value and the threshold, and triggering a warning signal. The warning signal is broadcast through the platform message channel, and the signal code contains an event level identifier, in which the single-box leaf consumption out-of-benchmark corresponds to a specific level identifier. After the marking is completed, the current out-of-benchmark data is implemented with write protection.

[0094] When the cigarette filling value index value exceeds the cigarette filling value preset threshold, it is marked as an over-standard index, and a differentiated processing mechanism is adopted. The cigarette filling value over-standard record additionally contains the cigarette specification and machine information fields, the warning signal corresponds to a specific level identifier and sets a delay confirmation mechanism. During the marking process, data snapshots are saved, and the original sampling data within a specific time period before and after the over-standard moment are stored in a temporary storage area.

[0095] When the energy consumption efficiency index value exceeds the energy consumption efficiency preset threshold, it is marked as an over-standard index, and the processing flow increases the energy consumption type differentiation link. According to the proportion of different energy consumption in the over-standard period, the over-standard event is classified into a specific type, the warning signal corresponds to a specific level identifier and is attached with an energy consumption type label, and the event record increases the continuous over-standard time interval field. After the marking operation is completed, the associated analysis process is automatically started.

[0096] When identifying the associated index set based on the antagonistic effect rule library stored in the antagonistic relationship mapping table between the index, first load the antagonistic effect rule library. The rule library is stored in a specific partition of the tobacco process knowledge base and contains an antagonistic relationship mapping table structure. The mapping table is a two-dimensional matrix with rows representing over-standard index types and columns representing potential associated index types. The matrix elements take specific numerical values to indicate the existence or absence of antagonistic relationships. For example, when the single-box leaf consumption exceeds the standard, the element value of a specific associated index column in the corresponding row is a specific numerical value. The mapping table update mechanism is to periodically recalculate the matrix element value based on the newly added conflict record.

[0097] The identification process is executed using a rule engine, and the input parameters are the current over-standard index name and event level. The rule engine queries the antagonistic relationship mapping table to obtain the basic associated index set, and then filters the results according to the event level. For example, high-level events retain all associated indexes, and medium-level events filter low-relevance indexes. The identification algorithm sets a timeout control mechanism, which returns intermediate results when the processing time exceeds a specific millisecond value. The identification result is output as a data structure containing a list of index names and a list of antagonistic strengths.

[0098] The antagonistic effect rule library is derived from the over-standard index conflict records obtained by analyzing historical production data in the tobacco process knowledge base. The specific construction process is as follows: periodically extract historical production anomaly records from the knowledge base, and filter conflict events with multiple indexes simultaneously abnormal. Perform root cause analysis on conflict events to determine the antagonistic direction between indexes, such as when one index is abnormal accompanied by another index being abnormal in the opposite direction. Conflict record analysis uses association rule mining methods, sets specific support and confidence thresholds to generate rule items. New rule items are updated to the antagonistic relationship mapping table after being reviewed and confirmed by a group of process experts. The review standard requires a specific number of independent events to support the rule.

[0099] Quality monitoring is implemented in the identification process, and integrity verification is performed on the returned associated pair of index sets. The verification content includes whether the number of set elements is within a certain interval, whether the index name belongs to a predefined type, and whether the antagonistic strength value is within a reasonable range. When the verification fails, the standby rule base is used to re-identify and record diagnostic information. The final output result is attached with version identification, timestamp, and verification status data.

[0100] The industrial internet platform allocates dedicated resources for the identification operation, and the rule engine is deployed in a separate container with a specific number of cores and a specific amount of memory. The system sets the upper limit of concurrent processing to a certain number of events per second, and when exceeded, it enters a queue waiting state. The identification intermediate data is retained for a certain period of time, and the final result is written to the knowledge base database and indexed. The platform provides a rule base update interface, and modifications need to go through a specific approval process to take effect.

[0101] After the associated pair of index sets is output, an antagonistic effect analysis report is generated, including the details of the over-standard index, the list of associated indexes, historical cases, and processing suggestions. For example, when a specific index is over-standard, the report lists the associated indexes and suggests historical solutions. The report is stored in a specific format and pushed to the business system through the message mechanism, and is also updated to the tobacco process knowledge base case library.

[0102] S3, based on the tobacco process knowledge base, a directed influence transmission path graph of the over-standard benchmark index to the associated benchmark index set is constructed, including:

[0103] When constructing the directed influence transmission path graph of the over-standard benchmark index to the associated benchmark index set based on the tobacco process knowledge base, first, extract the process influence relationship data from the tobacco process knowledge base. The tobacco process knowledge base is stored in a specific database partition of the industrial internet platform, including process parameter relationship data table, device linkage rule data table, and process influence coefficient data table. The extraction operation extracts the process influence relationship data between the over-standard benchmark index and the associated benchmark index set based on the output of step S2. The process influence relationship data includes direct influence relationship and indirect influence relationship: the direct influence relationship refers to the physical change relationship between the over-standard benchmark index and the associated benchmark index, such as the process relationship that the single-box leaf consumption directly affects the cigarette filling value; the indirect influence relationship refers to the cascading influence relationship through intermediate process index, such as the transmission relationship that the single-box leaf consumption first affects the cut tobacco width, and then the cut tobacco width affects the cigarette filling value. The extraction process uses a hierarchical query mechanism, first retrieves direct relationship entries, and when no direct relationship is found, uses a path derivation mechanism to find indirect relationship paths.

[0104] When the over-standard-to-target indicator is taken as the starting node of the directed influence conduction path diagram, a starting node data structure is created in the memory workspace. The data structure includes a node identifier field, an indicator name field, a current value field, and a node type identifier field. For example, when the single-box consumption of tobacco leaves exceeds the standard, a starting node with a unique identifier is created, and the node type identifier is set to a specific type identifier. The spatial position of the starting node is dynamically calculated according to the over-standard degree. For example, when the over-standard degree exceeds a certain proportion, the position coordinates are offset by a certain distance value in a certain direction. The node visual attribute configuration rule is: when the over-standard degree exceeds a certain threshold, a graphical identifier with a specific color and shape is displayed, otherwise a secondary color identifier is displayed. After the starting node is generated, a connection channel is established with the real-time data source, and when the over-standard value changes, the node display attribute is automatically updated.

[0105] When the associated-to-target indicator set is taken as the terminal node of the directed influence conduction path diagram, an independent node entity is created for each indicator in the set. The terminal node data structure includes an indicator name field, a benchmark value range field, and a node level identifier field. For example, when the cigarette filling value is taken as the terminal node, the benchmark value range data is recorded. The node level is determined according to the process association distance. For example, directly affected indicators are set to a specific level identifier, and indirectly affected indicators are set to a secondary level identifier. The arrangement of the terminal nodes in the display area uses a spatial layout algorithm. Nodes of the same level are distributed in a certain direction, and the level spacing is set to a certain pixel value. Each terminal node is associated with a data detail interface, through which the historical data curve of the indicator in a certain period can be viewed.

[0106] When the directed edge is established between the starting node and the terminal node according to the process influence relationship data, the influence type attribute and the influence intensity attribute in the process influence relationship data are analyzed. The direct influence relationship establishes a single-segment directed edge, and the directed edge attribute includes a direction identifier and an influence intensity value. For example, the directed edge from the single-box consumption of tobacco leaves node to the cigarette filling value node records a specific influence intensity value. The indirect influence relationship establishes multiple-segment directed edges, and intermediate process-to-target indicator nodes are created as transfer points in the path. For example, a specific parameter node of a specific process is created as an intermediate node. The visual features of the directed edge are dynamically configured according to the influence intensity: when the influence intensity exceeds a certain threshold, the edge with a specific line width and line type is displayed, and when it is below a certain threshold, the edge with a secondary line width and line type is displayed. A positive correlation relationship is represented by a specific color arrow, and a negative correlation relationship is represented by a contrasting color arrow.

[0107] The direction of the directed edge strictly follows the principle of pointing to the pair of indicators in the associated pair of indicator set by the super-label pair of indicator, and the direction determination mechanism is based on the process causal relationship chain. For example, when the single-box leaf consumption exceeds the standard and causes the cigarette filling value to change, the direction is set from the single-box leaf consumption node to the cigarette filling value node; when there is a two-way influence, the main influence direction is preferentially retained, and the secondary direction is marked as an auxiliary edge. Direction confirmation needs to be verified by process rules: search historical event cases in the tobacco process knowledge base, and if more than a certain proportion of historical cases support the direction, it is confirmed to be valid. After the direction is set, loop detection is performed, and when a circular path is found, the weakest edge is automatically disconnected.

[0108] The indirect influence relationship in the process influence relationship data is processed by using a graph search method. The starting node is taken as the search starting point, and the terminal node is taken as the search target. The transmission path is searched in the process parameter relationship data table. The search depth is limited to a certain number of levels, for example, the maximum search depth is set to a certain number of level values. The comprehensive influence strength of each path is calculated as the product of the influence strength of each segment on the path, for example, when the path passes through two intermediate nodes, the comprehensive influence strength is equal to the product of the influence strength of the two segments. Finally, the optimal path with a comprehensive influence strength exceeding a certain threshold is retained, and the remaining paths are marked as reference paths. During the search process, intermediate process pair of indicator nodes are dynamically created, and the intermediate nodes are named in the combined format of process name and parameter name, for example, a certain process name plus a certain parameter name.

[0109] The constructed directed influence transmission path graph performs topological structure verification. The verification content includes: the terminal node set is completely matched with the associated pair of indicator set; there is no unconnected isolated node; the reverse path between any two nodes does not exceed a certain number. After the verification passes, the path graph metadata is generated, including node number statistics, directed edge number statistics, maximum path depth and other attribute data. The path graph is stored as a structured data file, and a visual description file is also generated, and the two files are associated through a unique identifier.

[0110] The industrial internet platform allocates dedicated computing resources for path graph construction, and the graph computing service is deployed in an independent container instance, with a central processor with a certain number of cores and a memory with a certain capacity. The construction process sets the timeout control to a certain number of seconds, and after the timeout, the current progress is saved and the breakpoint information is recorded. The construction result is stored in the path graph storage partition of the process knowledge base, and is associated with the original super-label event index. The platform provides a path graph interactive editing tool, and the process engineer can adjust the node position or modify the edge attribute, and all modification operations record the version history.

[0111] After the directed influence conduction path graph is output, a post-trigger analysis report generation process is triggered, and the platform automatically creates an influence conduction analysis report. The report contains a path graph summary view, a key path list, an intermediate node list, and a monitoring suggestion field. For example, when a specific indicator exceeds the standard, the report lists the main path to the associated indicator and the influence intensity data. The report is pushed to the relevant business system through the message service and is updated to the case library partition of the tobacco process knowledge base. The path graph data is synchronized in real time to the mobile monitoring device, supporting process personnel to view the complete conduction relationship network.

[0112] S4, based on historical big data analysis of the nonlinear constraint relationship between the benchmark indicators in the directed influence conduction path graph, generate a set of non-dominated solutions that satisfy all associated benchmark indicator reference values, including:

[0113] When analyzing the nonlinear constraint relationship between the benchmark indicators in the directed influence conduction path graph based on historical big data, first extract the historical production data of each benchmark indicator in the directed influence conduction path graph from the time series database of the industrial internet platform. The extraction range of historical production data is set to the data of the last specific number of consecutive production cycles, for example, the data records of the last specific month are extracted. The extraction operation acquires the corresponding data sequence in the time series database according to the timestamp alignment method based on the indicator name marked in the directed influence conduction path graph, and the data time resolution is consistent with the collection frequency of step S1. For time periods with missing data, use the adjacent data point interpolation method to complete the interpolation, and the interpolation execution condition is that the missing time length does not exceed a specific time threshold. After extraction, perform data normalization to convert different dimension indicators such as single-box leaf consumption data, cigarette filling value data, and energy efficiency data into dimensionless relative values. The conversion method is to divide the current value by the reference value, and the reference value is derived from the standard value of each indicator defined in the tobacco process knowledge base.

[0114] For each two benchmark indicators connected by a directed edge in the directed influence conduction path graph, analyze the numerical change correlation in their historical production data using a time window sliding analysis method. Set a fixed length time window, for example, the window length is the data of a specific production batch length, and the window sliding step is a specific time interval. Calculate the correlation characteristics of the two indicator data in each window: first calculate the covariance value to represent the consistency of the change direction, then calculate the correlation coefficient value to represent the linear correlation strength, and finally calculate the mutual information value to represent the nonlinear dependence degree. During the analysis process, exclude data from abnormal working conditions, and the abnormal working condition determination standard is the existence of equipment failure records or process parameter fluctuations exceeding a specific percentage threshold. The correlation analysis results are stored as a relationship matrix, with the matrix rows and columns corresponding to the path graph nodes, and the matrix elements containing three fields of covariance value, correlation coefficient value, and mutual information value.

[0115] When establishing the nonlinear constraint function expression between the pairs of benchmark indicators according to the relevance of numerical changes, a piecewise function construction strategy is adopted. For the pairs of indicators with nonlinear correlation, the function expression contains a combination of polynomial terms and logarithmic terms: the polynomial terms are used to describe the change trend of the indicator values in a certain interval, such as terms containing quadratic operations; the logarithmic terms are used to describe the change characteristics of the indicator values in another interval, such as terms containing natural logarithm operations. The combination weights of specific terms are determined according to the fitting effect of historical data: different function forms are fitted by regression analysis method, and the function combination with a decision coefficient exceeding a certain threshold is selected. The function coefficients are calculated by an optimization algorithm, and the calculation samples are generated by a resampling method to generate a specific number of sample sets. The finally established constraint function expression is in the form of: the target value of the correlated indicators is equal to a certain coefficient multiplied by the current indicator raised to a certain power, plus a certain coefficient multiplied by the natural logarithm value of the current indicator, plus a certain constant term.

[0116] When generating the feasible solution space under the restriction of the nonlinear constraint function expression, a multi-objective optimization method is adopted to achieve the goal of simultaneously reaching the benchmark values of all pairs of benchmark indicators in the set. The optimization objective function is defined as the minimization of the sum of the absolute values of the deviations of the actual values of each indicator from the benchmark values, such as the minimization of the sum of the deviations of the single-box leaf consumption, the cigarette filling value, and the energy consumption efficiency. The constraint conditions include: the inequality relationship between the indicators defined by all nonlinear constraint function expressions; the upper and lower limit values of the process feasible range of each indicator, such as the single-box leaf consumption between a certain lower limit value and a certain upper limit value. The optimization process adopts an evolutionary algorithm framework: the initial solution set size is set to a certain number; the fitness function is the inverse of the above-mentioned objective function; the selection operation adopts a probability selection strategy; the crossover operation probability is set to a certain value; the mutation operation probability is set to a certain value. The iteration stopping condition is that the fitness improvement rate of a certain number of consecutive generations is lower than a certain threshold.

[0117] When screening the non-dominated solution set in the feasible solution space, the Pareto front screening of the multi-objective optimization solution set is performed. The generated feasible solution set is compared with each other: for solution A and solution B, if the numerical values of solution A on all pairs of benchmark indicators are better than those of solution B, then solution A dominates solution B; if there is no solution dominating solution A, then solution A is a Pareto optimal solution, i.e. a non-dominated solution. The screening process adopts a hierarchical sorting algorithm: in the first round of screening, all solutions that are not dominated by any solution are marked as the first front layer; after removing the first front layer solutions, the process is repeated to obtain the second front layer; the cycle continues until all solutions are layered. Finally, the first front layer solutions are selected to constitute the non-dominated solution set, and each non-dominated solution in the solution set records the corresponding numerical combination of each pair of benchmark indicators.

[0118] The entire processing process sets quality control measures. When the historical data sample size is insufficient for a specific number, automatically expand the data extraction period; when the fitting effect of the nonlinear constraint function is lower than a certain standard, increase the complexity of the function term; when the optimization algorithm does not converge, adjust the constraint boundary by a certain percentage to recalculate; when the non-dominated solution set is empty, reduce the benchmark value requirement by a certain percentage. All abnormal processing operations record audit logs, which include exception type, processing measures and result verification data.

[0119] After the non-dominated solution set is output, a visualization conversion is performed, and a solution set distribution graph is displayed on the industrial internet platform. A multi-dimensional space coordinate system is established with each pair of benchmark indicators as the coordinate axis, each non-dominated solution represents a point in the space, and the visual feature of the point represents the size of the comprehensive deviation value. At the same time, a solution set feature analysis report is generated, including solution quantity statistics, each indicator value range and recommended solution identifier field. For example, when the associated benchmark indicator set contains three indicators, the report displays a projection graph of the Pareto front in a three-dimensional space, and marks the solutions at specific positions as recommended candidate solutions. The report is pushed to subsequent processing through a data interface, and is also backed up to the optimization case storage area of the tobacco process knowledge base.

[0120] The industrial internet platform allocates dedicated computing resources for the calculation process, and the optimization calculation service is deployed in an independent container with a specific number of central processing units and a specific capacity of memory. The calculation task sets a timeout control threshold, and if it is not completed within a certain time, it is interrupted and the intermediate result is returned. The input and output data establish a complete traceability relationship, and the original directed influence conduction path graph, historical data set and final non-dominated solution set are associated through a unique transaction identifier. The platform provides a parameter adjustment interface, and the process engineer can adjust the benchmark value range or the constraint strength coefficient. The recalculated task enters the asynchronous processing queue for execution.

[0121] S5, according to the disturbance propagation depth of the directed influence conduction path graph and the intrinsic modal stability of the equipment vibration signal, the Pareto front solution set is selected from the non-dominated solution set, which includes:

[0122] According to the disturbance propagation depth of the directed influence conduction path diagram and the intrinsic modal stability of the equipment vibration signal, when screening the Pareto front solution set from the non-dominated solution set, first, for each non-dominated solution in the non-dominated solution set, determine the corresponding conduction path of the non-dominated solution in the directed influence conduction path diagram. The determination method of the conduction path is: locating the index node set contained in the non-dominated solution in the path diagram, starting from the superordinate-to-subordinate index node, traversing all directed edges pointing to the associated subordinate index termination node, and extracting the minimum connected subgraph containing all related nodes as the conduction path. For example, when the non-dominated solution contains single-box consumption leaf quantity index, cigarette filling value index and energy consumption efficiency index, extract the path branch connecting the three nodes. The data structure of the conduction path is stored as an ordered node list, and the starting element of the list is the superordinate-to-subordinate index node, and the subsequent elements are arranged in order according to the direction of the directed edge.

[0123] When calculating the disturbance propagation depth of the corresponding conduction path, starting from the superordinate-to-subordinate index node, the path length to each associated subordinate index termination node is calculated along the direction of the directed edge. The path length measurement method is to count the number of intermediate nodes passed, that is, the number of node hops. For example, the path length from the superordinate node to the termination node through a certain number of intermediate nodes is equal to the number of intermediate nodes. After traversing all termination nodes, the maximum path length value is taken as the disturbance propagation depth of the conduction path. The calculation process uses a graph traversal method, the search range is limited within the conduction path subgraph, and the upper limit of the search depth is set to a certain number of levels. The propagation depth result is represented by an integer value, and the associated conduction path identifier and non-dominated solution identifier are stored.

[0124] When obtaining the corresponding equipment vibration signal corresponding to the non-dominated solution, according to the process equipment type involved in the non-dominated solution, the vibration signal data is extracted from the equipment monitoring database of the industrial internet platform. The device type and index association relationship is predefined in the mapping table: for example, the single-box consumption leaf quantity index corresponds to a specific device of the silk making process, the cigarette filling value index corresponds to a specific device of the rolling process, and the energy consumption efficiency index corresponds to a specific device of the power plant. The vibration signal acquisition frequency is a certain hertz value, and the signal length covers a certain time period before and after the production period corresponding to the non-dominated solution. After the signal is obtained, preprocessing is performed, a filter with a certain cutoff frequency is used to remove interference components, and the amplitude of the filtered signal is normalized to a certain value range.

[0125] When performing intrinsic modal decomposition on the equipment vibration signal to obtain the dominant modal component, a signal energy analysis method is used. The time domain vibration signal is converted into a frequency domain energy distribution, and a certain frequency range with the largest energy proportion is identified as the dominant frequency band. The signal component in this frequency band is extracted by a frequency domain filtering method, and then converted into a time domain modal component. For example, in a certain equipment vibration signal, the signal component in a certain frequency range is identified as the dominant modal component. The decomposition process sets an upper limit on the number of components, and at most a certain number of dominant modal components are retained.

[0126] When calculating the energy concentration of the dominant modal component as the intrinsic modal stability, the energy distribution of the modal component is analyzed within a specific time window. The proportion of the component energy to the total energy of the signal is calculated, for example, a specific percentage of the total energy of the dominant modal component energy. The energy calculation uses the square integral method, and the square value of the component signal is integrated within the time window. The stability result is expressed as a dimensionless value, the value range is between zero and one, the higher the value, the more concentrated the energy, the better the stability.

[0127] When the disturbance propagation depth does not exceed the preset propagation depth threshold and the intrinsic modal stability is not lower than the preset stability threshold, the corresponding non-dominated solution is retained. The preset propagation depth threshold is obtained according to the historical stable production data statistics analysis of the factory, for example, a specific percentage of the propagation depth under normal working conditions is taken as the threshold. The preset stability threshold is calibrated by normal operation data, for example, a specific percentage of the minimum energy concentration under normal working conditions is taken as the threshold. The decision process uses logical and operation: set the retention flag to true when the propagation depth meets the condition and the stability meets the condition, otherwise, set it to false.

[0128] All retained non-dominated solutions constitute the Pareto front solution set, and the solution set construction process performs repetitive verification. The verification rule is: if the index value difference of two non-dominated solutions is less than a certain tolerance threshold, it is considered as a repeated solution and only one of them is retained. The solution set data structure includes a solution identifier list, a set of each solution index values, and a screening judgment record field. For example, when a certain number of solutions are retained, a numerical set of a certain dimension is formed. The solution set metadata record screening parameter information, including the propagation depth threshold and the stability threshold used.

[0129] The industrial internet platform allocates computing resources for the screening process, and the processing engine is deployed in an independent container. The resource configuration is a certain number of central processing units and a certain capacity of memory, and the data processing capacity is set to a certain number of solutions per second. The calculation task sets the priority rule: the propagation depth calculation task has higher priority than the stability analysis task. When the system resources are insufficient, the low-priority tasks are delayed and the delay information is recorded.

[0130] The screening process implements full-process monitoring, and generates a processing record for each non-dominated solution. The record includes: conduction path extraction state, propagation depth calculation result, vibration signal quality, energy concentration calculation result and final judgment result field. When the propagation depth calculation is abnormal or the vibration signal is missing, the backup processing method is automatically enabled: the propagation depth uses the path estimation method, and the vibration signal uses the adjacent period replacement signal. All abnormal processing records have detailed diagnosis data, including abnormal type, processing measures and result reliability evaluation.

[0131] After the Pareto front solution set is output, a visualization labeling process is performed, marking the retained solutions with specific visual features in the industrial internet platform interface. Simultaneously, a screening analysis report is generated, including statistics on the number of solutions, a propagation depth distribution chart, a stability value distribution chart, and detailed fields for screening parameters. For example, the report displays a graph showing the relationship between propagation depth and stability, marking the threshold boundary positions and the locations of retained solutions. The report is pushed to subsequent processing via the data channel and stored in the screening case library of the tobacco process knowledge base.

[0132] The platform provides a threshold adjustment interface, allowing process engineers to modify propagation depth or stability threshold parameters. After modification, the system automatically re-executes the screening process, compares the new solution set with historical versions, and generates a change analysis report. Threshold modifications require a specific approval process to take effect, and historical threshold configurations are retained for at least a specific time period.

[0133] S6. By analyzing the volatility variance of the benchmark index and the complexity of the directed influence transmission path for each solution, the risk entropy value of each solution in the Pareto front solution set is generated. Specific implementation includes:

[0134] For each Pareto front solution in the Pareto front solution set, when acquiring the historical production data corresponding to that Pareto front solution, all benchmark indicator data related to that solution are extracted from the time-series database of the industrial internet platform. The extraction range of historical production data is determined by a preset time window based on the tobacco production batch cycle. For example, if the standard production batch cycle of a certain brand of cigarettes is a specific number of hours, then the preset time window is set to that specific number of hours. The extraction operation takes the production period corresponding to the solution as the center point and extends forward and backward by a specific proportion of time periods to ensure coverage of the complete production cycle. During data extraction, precise timestamp alignment processing is performed, with the alignment accuracy consistent with the data acquisition frequency. Missing data is filled in by interpolation of data from adjacent time periods at the same position.

[0135] When calculating the variance of each benchmark indicator corresponding to the Pareto front solution within a preset time window based on historical production data, a piecewise variance calculation method is used. The preset time window is divided into a specific number of equal-length time intervals. Within each time interval, the variance of the indicator is calculated: first, the average value of the data points within the time interval is calculated; then, the squared deviation of each data point from the average value is calculated; finally, the sum of squared deviations is divided by the number of data points to obtain the variance of the time interval. The maximum value of the variance across all time intervals is taken as the final variance of the indicator. For example, for the single-box blade consumption indicator, the maximum variance across a specific number of time intervals is taken as the variance of the indicator. Data from periods of abnormal operating conditions is excluded during the calculation process; the criterion for determining abnormal periods is a change in equipment status records.

[0136] When determining the corresponding Pareto frontier solution in the directed influence conduction path graph, the minimum connected subgraph is located in the directed influence conduction path graph generated in step S3 according to the index node set contained in the solution. The locating method is as follows: taking the super-index-to-sub-index node as the starting point and the associated sub-index set node as the end point, the minimum directed path set connecting all starting points and end points is extracted. The conduction path data structure is stored in the form of a combination of node sequence and directed edge sequence, the node sequence is arranged in the order of path direction, and the directed edge sequence records the starting point and end point identifiers of each edge.

[0137] When calculating the directed influence conduction path complexity of the corresponding conduction path, the total number of all independent nodes in the conduction path and the total number of all effective directed edges are counted, and the two values are multiplied to obtain the complexity value. The total number of nodes includes all unique nodes in the path, and repeated nodes are not counted; the total number of directed edges includes all effective connection edges. For example, if a certain path contains a certain number of independent nodes and a certain number of directed edges, the complexity is the product of the total number of nodes and the total number of directed edges. The calculation result is stored as an integer value and is associated with the conduction path.

[0138] When summing up the fluctuation variance of each sub-index, the fluctuation variances of all sub-indexes involved in the Pareto frontier solution are accumulated. Before accumulation, dimensionless processing is performed: all variance values need to be converted to the same dimension unit, and if there is a dimensionless variance, it is multiplied by the square of the reference value to convert it to a physical dimension unit. The total fluctuation variance result is kept to a certain number of decimal places, and the number of indexes involved in the calculation is recorded.

[0139] When multiplying the total fluctuation variance and the directed influence conduction path complexity to obtain the risk entropy value of the corresponding Pareto frontier solution, the two parameters are first standardized. The total fluctuation variance is divided by its maximum observed value to convert it to a relative value; the complexity is divided by its theoretical maximum value to convert it to a relative value. After standardization, the multiplication operation is performed: the risk entropy value is equal to the standardized total fluctuation variance multiplied by the standardized complexity, and then multiplied by a certain adjustment coefficient. The adjustment coefficient is set according to the historical risk event frequency of the production line, for example, a certain value is taken for a high-frequency risk production line and a certain value is taken for a low-frequency risk production line. The final risk entropy value is stored in floating point format, and the value range is within a certain interval.

[0140] Multiple verification mechanisms are implemented in the risk entropy value calculation process. Input data verification: confirm whether the historical data time window matches the batch cycle; path structure verification: check whether there are isolated nodes in the conduction path; dimension consistency verification: ensure that all variance units are uniform. When verification fails, automatic correction is started: if the time window does not match, the window is redefined; if there are isolated nodes, necessary connection edges are supplemented; if the dimensions are not uniform, unit standardization conversion is performed. All correction operations are recorded with detailed logs, including original data, correction method and correction result fields.

[0141] The industrial internet platform allocates batch processing resources for the calculation process, and the calculation service is deployed in a dedicated container instance. The resource configuration is a specific number of central processing units and a specific capacity of memory, and the parallel processing capability is set to a specific number of solutions per second. The task scheduling adopts a priority strategy: the current production batch solution is processed first. When the system resources are insufficient, queue management is started, and the timeout threshold is set to a specific number of minutes. The timeout task is transferred to the background asynchronous processing.

[0142] After the risk entropy value is output, a visual encoding is performed, and the entropy value is represented by color depth in the Pareto frontier solution distribution graph. For example, a chromatography mapping scheme is adopted: low entropy value solutions are marked as cool tones, and high entropy value solutions are marked as warm tones. A risk analysis report is also generated, including a solution entropy value sorting list, key influence factor analysis, and risk prevention and control suggestion fields. The report is pushed to the decision support system through the data interface and archived to the risk case library partition of the tobacco process knowledge base.

[0143] The platform provides a parameter configuration interface, and process engineers can adjust the adjustment coefficient or reference value parameters. After adjustment, the system automatically recalculates the risk entropy value, and the new result is compared with the historical version to generate a change analysis report. Parameter modification records need to be approved through a specific approval process to take effect, and historical parameter configurations are saved for a specific length of time.

[0144] S7, select the solution with the minimum risk entropy value from the Pareto frontier solution set as the decision scheme output to the production execution system, which includes:

[0145] When comparing the risk entropy values of all Pareto frontier solutions in the Pareto frontier solution set, the risk analysis service of the industrial internet platform performs sorting operations. First, load the risk entropy value data set generated in step S6, which is stored as a correspondence table of solution identifiers and entropy values. The comparison process uses a sorting algorithm: initialize an empty solution sequence container, and insert each Pareto frontier solution into the appropriate position in the container in turn, so that the container always maintains an ascending order of entropy values. For example, when processing a new solution, compare the entropy values from the end of the container forward until a solution with a smaller entropy value is found, and then insert it after the solution. During the comparison process, data validity checks are performed: confirm that the entropy value data type is numeric and within a reasonable range; verify the matching relationship between the solution identifier and the Pareto frontier solution set. If an abnormal entropy value is found, automatically trigger the recalculation process and record the abnormal event.

[0146] When determining the Pareto frontier solution with the minimum risk entropy value, the first element is extracted from the sorted solution sequence container. The extraction operation performs a double verification: first, it verifies that the container is not empty and contains a valid solution; then it verifies that the entropy value of the first element is less than or equal to the entropy values of all other solutions in the container. For example, by traversing and comparing to ensure that the entropy value of the first element is not greater than the entropy value of any subsequent element. After determination, a solution selection record is generated, containing a solution identifier field, an entropy value field, and a selection timestamp field. When there are multiple solutions with the same minimum entropy value, a secondary selection mechanism is activated: compare the path complexity values of these solutions in the directed influence propagation path graph, and select the solution with the lowest complexity value; if the complexity values are still the same, select the most recently generated solution.

[0147] When converting the benchmark indicator values in the Pareto frontier solution with the minimum risk entropy value into device control parameters, the conversion operation is performed through the conversion rule mapping table of the tobacco process knowledge base. The conversion rule mapping table is stored in the process conversion data partition of the knowledge base, containing a benchmark indicator name field, a device parameter type field, and a conversion function identifier field. The conversion process processes each indicator one by one: first, query the mapping table to obtain the device parameter type and conversion function identifier corresponding to the indicator, then call the predefined conversion function to calculate the device control parameter value. For example, the single-box leaf consumption indicator is mapped to the cigarette machine cutter speed parameter through a specific conversion function, the cigarette filling value indicator is mapped to the cut tobacco drying machine temperature parameter through a specific conversion function, and the energy efficiency indicator is mapped to the flavoring machine flow parameter through a specific conversion function. The conversion function is implemented using a linear conversion method, and the function parameters are dynamically loaded according to the device model.

[0148] When mapping the benchmark indicator values in the decision scheme and the device control parameters through the conversion rule mapping of the tobacco process knowledge base, the mapping relationship maintenance mechanism is as follows: the mapping rules are regularly reviewed by the process engineer team, and the conversion function parameter table is adjusted according to the device update situation. The rule update process is as follows: the device maintenance department provides new parameter ranges, the process laboratory conducts multi-batch production testing, the test data are analyzed to determine new conversion coefficients, and finally the knowledge base is updated through the review process. The mapping exception handling mechanism includes: when the indicator value exceeds the definition range of the conversion function, automatically switch to the device safety boundary value and issue a prompt message; when the conversion result violates the process constraints, an artificial review process is started.

[0149] When sending device control parameters to the production execution system through the industrial internet platform, a hierarchical transmission mechanism is adopted. First, the device control parameters are packaged into a standard control instruction data structure, which includes a device number field, a parameter type field, a parameter value field, and an effective time window field. For example, the cigarette machine cutter head speed parameter is packaged into a speed control instruction for a specific device number. Then, the transmission is carried out through the platform's message service, and the transmission protocol uses the industrial communication protocol standard, with the data packet attached to the verification information. The sending process implements a confirmation mechanism: the platform sends the instruction and waits for the receiving response from the production execution system; if no response is received, it is re-sent at a specific interval; if it fails more than a certain number of times, it records a communication exception.

[0150] The industrial internet platform provides reliability guarantee measures for the sending process. Before sending the control parameters, a device state check is performed: the real-time monitoring data of the device is queried to confirm that the device is in a state that can receive instructions; and it is checked whether the parameter value is within the current allowed operating range of the device. Data protection is implemented during the sending process: an encryption algorithm is used to process the control instruction, and the key is changed periodically. After the sending is completed, execution monitoring is started: within the parameter effective time window, the actual parameter value of the device is compared with the to-be-executed parameter value in real time, and if the deviation exceeds a certain threshold, the parameter adjustment process is triggered.

[0151] The device control parameters include three types of core parameters: cigarette machine cutter head speed parameters, drying machine temperature parameters, and flavoring machine flow parameters. An independent safe operating range is set for each type of parameter: the cigarette machine cutter head speed parameter has a specific lower limit and a specific upper limit; the drying machine temperature parameter has a specific lower limit and a specific upper limit; and the flavoring machine flow parameter has a specific lower limit and a specific upper limit. When converting parameters, range checking is automatically performed: when the conversion result exceeds the safe range, it is automatically adjusted to the nearest boundary value and a range out-of-bounds record is generated. The range values are dynamically updated according to the device technical documents, and after the update, they are automatically synchronized to all related conversion functions.

[0152] The platform provides an emergency handling mechanism for decision scheme output. When the production execution system returns a parameter execution exception, the recovery process is automatically started: first, the parameter settings of the device in the previous specific period are rolled back; then, a new parameter is generated by reselecting the Pareto frontier solution with the second smallest risk entropy value; finally, the abnormal event is recorded and the local recalculation process is triggered. All output operations generate operation logs, including the original decision scheme, the sent parameter value, the device feedback, and the execution result fields. The operation logs are saved for a certain period of time.

[0153] After the decision scheme is output, an execution summary report is generated, including selected solution details, a conversion parameter list, a sending state, and a risk analysis field. For example, the report shows the risk entropy value distribution and marks the selected solution position, and lists all converted device control parameter values. The report is synchronously pushed to the process management system and archived in the decision case storage area of the tobacco process knowledge base. The platform provides a parameter manual adjustment function, and process engineers can modify specific device parameter values, and adjustments need to be accompanied by explanations and go through an approval process.

[0154] The tobacco benchmarking system intelligent early warning auxiliary decision method of the embodiment realizes full-link collaborative optimization of the tobacco benchmarking system by constructing an antagonistic effect driven conduction path network and a multi-target risk quantification mechanism. Specifically, when it is monitored that a benchmarking index exceeds a standard, an associated index set is identified based on an antagonistic rule library; a directed influence conduction path diagram is constructed using a process knowledge base; a non-dominated solution set satisfying multi-index constraints is generated in combination with historical data; a Pareto frontier solution set is screened according to a disturbance depth and a device stability; a risk entropy value is calculated through fluctuation variance and path complexity; and finally, a decision scheme with the minimum entropy value is output to an execution system, so as to generate a globally optimal device parameter regulation instruction under index conflict constraints.

[0155] Embodiment 2 Figure 2 A structure diagram of a tobacco benchmarking system intelligent early warning auxiliary decision system is given, and the tobacco benchmarking system intelligent early warning auxiliary decision system comprises:

[0156] A data acquisition module is configured to acquire multi-dimensional benchmarking index data of a tobacco production line in real time through an industrial internet platform;

[0157] An antagonistic identification module is configured to mark any-dimensional benchmarking index data exceeding a preset threshold as an exceeding benchmarking index, and identify an associated benchmarking index set having an antagonistic effect with the corresponding exceeding benchmarking index based on an antagonistic effect rule library;

[0158] A conduction construction module is configured to construct a directed influence conduction path diagram from the exceeding benchmarking index to the associated benchmarking index set based on a tobacco process knowledge base;

[0159] A solution set generation module is configured to analyze a nonlinear constraint relationship between benchmarking indexes in the directed influence conduction path diagram based on historical big data, and generate a non-dominated solution set satisfying benchmark values of all associated benchmarking indexes;

[0160] A frontier screening module is configured to screen a Pareto frontier solution set from the non-dominated solution set according to a disturbance propagation depth of the directed influence conduction path diagram and an intrinsic modal stability of a device vibration signal;

[0161] An entropy value generation module is configured to generate a risk entropy value of each solution in the Pareto frontier solution set by analyzing the variance of the target index fluctuation corresponding to each solution and the complexity of the directed influence conduction path.

[0162] A scheme output module is configured to select a solution with the minimum risk entropy value from the Pareto frontier solution set as a decision scheme and output the decision scheme to a production execution system.

[0163] In the embodiments, all the calculations are dimensionless numerical calculations, and the preset parameters and threshold values in the calculations are set by a person skilled in the art according to actual conditions.

[0164] It should be noted that the present application can be deployed on a device itself to realize embedded applications, or can be run on a PC or other terminal with a user interface, thereby meeting various hardware environments and use requirements.

[0165] The above embodiments can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, the above embodiments can be realized in the form of a computer program product in whole or in part. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another through wireless or wired transmission. The wired transmission includes optical fiber, twisted pair, coaxial cable, etc. The wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0166] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and module can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0167] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the division of the above-described device embodiment is merely an example, and the division of the modules can be different, for example, a plurality of modules or a component can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection can be indirect coupling or communication connection through some interfaces, devices or modules, and can be electrical, mechanical or other forms.

[0168] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical modules, and can be located in one place or distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0169] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can be physically present alone, or two or more modules can be integrated in one module.

[0170] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0171] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0172] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application, should be included in the scope of protection of the present application.

Claims

1. A smart early warning and decision-making auxiliary method for a tobacco benchmarking system, characterized in that, include: S1. Obtain multi-dimensional benchmark data of tobacco production line in real time through industrial internet platform; S2. When the benchmark index data of any dimension exceeds the preset threshold, it is marked as an out-of-benchmark benchmark index, and the set of associated benchmark indices that have an antagonistic effect with the corresponding out-of-benchmark benchmark index is identified based on the antagonistic effect rule base. S3. Construct a directed influence transmission path diagram from the exceeding benchmark indicators to the set of related benchmark indicators based on the tobacco process knowledge base, including: Data on the process influence relationship between the exceeding benchmark indicators and each benchmark indicator in the set of related benchmark indicators were extracted from the tobacco process knowledge base. The exceeding benchmark indicator is used as the starting node of the directional impact transmission path diagram; The benchmarking indicators in the associated benchmarking indicator set are used as the terminal nodes of the directed influence transmission path diagram; Based on the process influence relationship data, a directed edge is established between the start node and the end node; The data on process influence relationships include direct influence relationships and indirect influence relationships through benchmarking indicators in intermediate processes; All directed edges point from the super-benchmarked index to the benchmarked index in the associated benchmarked index set; S4. Based on historical big data analysis, analyze the nonlinear constraint relationship between benchmark indicators in the directed influence transmission path diagram, and generate a nondominated solution set that simultaneously satisfies the benchmark values ​​of all associated benchmark indicators. S5. Based on the disturbance propagation depth of the directed influence propagation path diagram and the intrinsic modal stability of the equipment vibration signal, select the Pareto front solution set from the non-dominated solution set, including: For each non-dominated solution in the non-dominated solution set, determine the corresponding propagation path in the directed influence propagation path graph. The path length of the corresponding propagation path from the node of the out-of-standard benchmark to the terminal node of the farthest associated benchmark is calculated as the perturbation propagation depth; Obtain the equipment vibration signal corresponding to the non-dominated solution; The dominant modal components are obtained by performing intrinsic mode decomposition on the equipment vibration signal; The energy concentration of the dominant modal component is used as the intrinsic modal stability. When the perturbation propagation depth does not exceed the preset propagation depth threshold and the intrinsic mode stability is not lower than the preset stability threshold, the corresponding non-dominated solution is retained. All retained nondominated solutions constitute the Pareto front solution set; S6. By analyzing the variance of the benchmark index and the complexity of the directed influence transmission path for each solution, the risk entropy value of each solution in the Pareto front solution set is generated, including: For each Pareto front solution in the Pareto front solution set, obtain the historical production data corresponding to the Pareto front solution; Calculate the variance of each benchmarking index corresponding to the Pareto front solution within a preset time window based on historical production data. Determine the corresponding propagation path of the Pareto front solution in the directed influence propagation path diagram; The product of the total number of nodes in the corresponding propagation path and the total number of directed edges is used as the complexity of the directed influence propagation path. The total variance is obtained by summing the variances of each benchmark indicator. Multiplying the total volatility variance by the complexity of the directed influence transmission path yields the risk entropy value of the corresponding Pareto front solution; The preset time window is determined based on the tobacco production batch cycle; S7. Select the solution with the minimum risk entropy from the Pareto front solution set as the decision scheme and output it to the production execution system.

2. The intelligent early warning and auxiliary decision-making method for a tobacco benchmarking system according to claim 1, characterized in that, Real-time acquisition of multi-dimensional benchmarking data from the tobacco production line through an industrial internet platform, including: Data on single-box leaf consumption was collected from the tobacco processing stage; data on cigarette filling values ​​was collected from the cigarette rolling and splicing stage; and energy efficiency data was collected from the power plant. Data on single-box leaf consumption, cigarette filling value, and energy efficiency are integrated into multi-dimensional benchmarking data, which is then transmitted in real time via the OPC UA protocol built into the industrial internet platform.

3. The intelligent early warning and auxiliary decision-making method for a tobacco benchmarking system according to claim 2, characterized in that, When the data of any dimension of the benchmark index exceeds a preset threshold, it is marked as an out-of-benchmark benchmark index. Based on the antagonistic effect rule base, a set of associated benchmark indices that have an antagonistic effect with the corresponding out-of-benchmark benchmark index is identified, including: Each indicator value in the multidimensional benchmarking data is compared with a preset threshold; When the single-box leaf consumption index value exceeds the preset threshold for single-box leaf consumption, it is marked as an out-of-standard benchmark index. When the cigarette filling value index exceeds the preset threshold for cigarette filling value, it is marked as an out-of-standard benchmark index. When the energy efficiency index value exceeds the preset threshold for energy efficiency, it is marked as an out-of-standard benchmark index. Based on the antagonistic relationship mapping table between benchmark indicators stored in the antagonistic effect rule base, the set of associated benchmark indicators that have an antagonistic effect with the benchmark indicators that exceed the standard is identified.

4. The intelligent early warning and auxiliary decision-making method for a tobacco benchmarking system according to claim 3, characterized in that, The antagonistic effect rule base is derived from the benchmarking indicator conflict records obtained by analyzing historical production data in the tobacco technology knowledge base.

5. The intelligent early warning and auxiliary decision-making method for a tobacco benchmarking system according to claim 1, characterized in that, Based on historical big data analysis of the nonlinear constraint relationships between benchmark indicators in the directed influence transmission path diagram, a nondominated solution set that simultaneously satisfies the benchmark values ​​of all associated benchmark indicators is generated, including: Historical production data for each benchmark indicator in the directed influence transmission path diagram is extracted from historical big data. For each pair of benchmark indicators connected by directed edges in the directed influence transmission path graph, analyze the correlation of numerical changes between the corresponding two benchmark indicators in historical production data; Establish nonlinear constraint function expressions between benchmarking indicators based on the correlation of numerical changes; With the goal of all benchmarking indicators in the associated benchmarking indicator set simultaneously reaching the benchmark value, a feasible solution space is generated under the constraint of a nonlinear constraint function expression. In the feasible solution space, select the set of solutions that have no other feasible solutions and are superior to the corresponding solutions on all benchmarking indices as the non-dominated solution set; The expression for the nonlinear constraint function includes a combination of polynomial and logarithmic terms.

6. The intelligent early warning and auxiliary decision-making method for a tobacco benchmarking system according to claim 1, characterized in that, The solution with the minimum risk entropy value is selected from the Pareto front solution set and output as the decision scheme to the production execution system, including: Compare the risk entropy values ​​of all Pareto front solutions in the Pareto front solution set; Determine the Pareto front solution that minimizes the risk entropy; Convert the benchmark index value in the Pareto front solution with the minimum risk entropy into equipment control parameters; The equipment control parameters are sent to the production execution system through the industrial internet platform; The equipment control parameters include the speed parameters of the cigarette machine cutter head, the temperature parameters of the drying machine, and the flow parameters of the flavoring machine; The equipment control parameters and the benchmark values ​​in the decision-making scheme are mapped through the conversion rules of the tobacco process knowledge base.

7. A tobacco benchmarking system intelligent early warning and auxiliary decision-making system, used to implement the tobacco benchmarking system intelligent early warning and auxiliary decision-making method according to any one of claims 1-6, characterized in that, include: The data acquisition module is used to acquire multi-dimensional benchmark data of the tobacco production line in real time through the industrial internet platform; The antagonism identification module is used to mark any benchmark indicator data in any dimension as an out-of-benchmark benchmark indicator, and to identify the set of associated benchmark indicators that have an antagonistic effect with the corresponding out-of-benchmark benchmark indicator based on the antagonism effect rule base. The transmission construction module is used to construct a directed influence transmission path diagram from the exceeding benchmark indicators to the set of related benchmark indicators based on the tobacco process knowledge base; The solution set generation module is used to analyze the nonlinear constraint relationship between benchmark indicators in the directed influence transmission path diagram based on historical big data analysis, and generate a nondominated solution set that simultaneously satisfies the benchmark values ​​of all associated benchmark indicators. The front screening module is used to screen out the Pareto front solution set from the non-dominated solution set based on the disturbance propagation depth of the directed influence propagation path diagram and the intrinsic mode stability of the equipment vibration signal. The entropy generation module is used to generate the risk entropy value of each solution in the Pareto front solution set by analyzing the fluctuation variance of the benchmark index and the complexity of the directed influence transmission path for each solution. The solution output module is used to select the solution with the minimum risk entropy value from the Pareto front solution set as the decision solution and output it to the production execution system.

Images