Production control method, system and storage medium applied to environmental online monitoring

Abstract

The application discloses a production control method and system applied to online environmental monitoring and a storage medium, wherein the method generates a directed transmission edge set according to process correlation information of equipment entity nodes in a production process, generates a process cause-effect topological atlas based on the equipment entity nodes and the directed transmission edge set, performs abnormal fluctuation mode identification on emission index data streams collected by an environmental monitoring sensor to generate an abnormal trigger signal, marks the abnormal trigger signal on a corresponding equipment entity node, performs link reverse propagation processing and edge constraint relationship checking in a reverse direction along the directed transmission edge, obtains a cause-effect bifurcation identification node, traces a disturbance conduction path, extracts running state information, performs semantic synthesis to generate disturbance conduction explanation semantics, calls a root cause matching logic tree to perform root cause logic matching to obtain an operation primitive sequence, and converts the operation primitive sequence into a control instruction to regulate production equipment, so that source tracing and automatic regulation of emission abnormalities are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, specifically to a production control method, system, and storage medium for online environmental monitoring. Background Technology

[0002] In industrial settings with continuous production processes, such as chemical, power, and pharmaceutical industries, real-time monitoring of emissions during production and adjustment of production equipment based on monitoring results are important means to ensure environmental compliance and production safety.

[0003] In existing technologies, environmental monitoring sensors are typically deployed at various emission points or monitoring locations within the production process to collect emission data. Fixed thresholds or simple statistical control charts are then used to determine if emissions exceed standards. When an anomaly is detected, operators rely on experience to identify the source of the anomaly and manually adjust the operating parameters of relevant equipment. However, existing technologies can only detect whether emission indicators are abnormal; they cannot trace the transmission path of the anomaly within the production process, making it difficult to pinpoint the root cause of the anomaly. Furthermore, they cannot automatically generate targeted control commands, resulting in significant delays in anomaly response and a control effect that depends heavily on the operator's experience. Summary of the Invention

[0004] The purpose of this invention is to provide a production control method, system, and storage medium for online environmental monitoring, in order to solve the problems mentioned in the background art.

[0005] This invention provides a production control method for online environmental monitoring, comprising: Based on the process association information between equipment entity nodes in the production process, a set of directed transit edges from upstream equipment entity nodes to downstream equipment entity nodes is generated, and a process causal topology graph is generated based on the equipment entity nodes and the set of directed transit edges. The emission index data stream collected by environmental monitoring sensors is processed for abnormal fluctuation pattern recognition, an abnormal trigger signal is generated, and the abnormal trigger signal is marked as a causal injection event on the corresponding equipment entity node in the process causal topology map. In the process causal topology graph, starting from the marked device entity node, the link reverse propagation process is performed along the reverse direction of the directed transmission edge, and the edge constraint relationship verification process is performed on each upstream device entity node to obtain the causal fork identification node. Based on the causal bifurcation identification node, the disturbance propagation path is traced back, the operating status information of each device entity node on the disturbance propagation path is extracted and semantic synthesis processing is performed to generate disturbance propagation interpretation semantics. The disturbance propagation interpretation semantics includes a description of the propagation direction of the abnormal disturbance in the process causal topology map and a description of the key bifurcation position. The root cause matching logic tree is invoked to perform root cause logic matching processing on the disturbance propagation interpretation semantics to obtain the corresponding operation primitive sequence, and the operation primitive sequence is converted into control instructions and written into the target control system position to regulate the production equipment.

[0006] This invention provides a production control system, comprising: A processor; a storage device on which a computer program is stored; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement the above-described production control method for online environmental monitoring.

[0007] The present invention provides a readable storage medium on which a program or instruction is stored, and when the program or instruction is executed by a processor, it implements the above-described production control method for online environmental monitoring.

[0008] Compared with existing technologies, the beneficial effects of this invention are: by constructing a process causal topology graph, the material transfer, energy coupling, and implicit correlations between equipment entity nodes in the production process are explicitly transformed into a directed topology structure, providing a global causal reasoning basis for tracing the source of abnormal emissions; based on this, abnormal emission indicators collected by environmental monitoring sensors are anchored to the corresponding equipment entity nodes in the form of causal injection events, and link propagation and edge constraint verification are performed in reverse along the directed transmission edges, thereby accurately locating causal branching nodes and avoiding misjudgments caused by relying solely on single equipment data or simple temporal correlations in traditional methods; furthermore, by tracing back the disturbance propagation path and Semantic synthesis is performed on the operational status information of each device entity node along the path to generate disturbance propagation interpretation semantics containing descriptions of propagation direction and bifurcation position. This elevates the abnormal propagation process from the numerical level to an interpretable structured semantic level, providing a complete chain of propagation evidence for root cause determination. Finally, the root cause matching logic tree is invoked to match the disturbance propagation interpretation semantics, and the matched operation primitive sequence is converted into control commands that directly act on the target control system site. This achieves closed-loop control throughout the entire process, from anomaly identification, propagation path tracing, semantic interpretation generation to control command output, effectively improving the accuracy and interpretability of emission anomaly response during production control. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1This is a flowchart of a production control method for online environmental monitoring provided in an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram illustrating the generation of the causal branching identifier node provided in an embodiment of the present invention.

[0012] Figure 3 This is a schematic diagram of a production control method for online environmental monitoring provided in an embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] The production control method for online environmental monitoring provided in this invention can be applied to various industrial environmental online monitoring scenarios with continuous production processes, such as chemical production parks, thermal power plants, or pharmaceutical production workshops, where real-time monitoring and automatic control of emission indicators are required. In this invention embodiment, a continuous production process in a chemical plant is used as the application scenario throughout the text. This chemical plant includes a raw material pretreatment unit, a reaction synthesis unit, a separation and purification unit, and a tail gas treatment unit. These units are connected by pipelines and conveying equipment to form a complete production chain. The physical equipment nodes in each unit include reaction vessels, heat exchangers, compressors, distillation columns, and absorption towers.

[0015] Please see Figure 1 , Figure 1 This is a flowchart of a production control method for online environmental monitoring provided by an embodiment of the present invention. The method can be executed by a production control system and includes steps 110-150.

[0016] Step 110: Generate a set of directed transit edges from upstream equipment entity nodes to downstream equipment entity nodes based on the process association information between equipment entity nodes in the production process, and generate a process causal topology graph based on the equipment entity nodes and the set of directed transit edges.

[0017] A process causal topology map is a topological structure used to characterize the material, energy, or control information transfer relationships between various equipment entities in a production process. Equipment entities correspond to actual physical equipment in the production process, such as reactors, heat exchangers, compressors, or distillation columns. Process correlation information describes the upstream and downstream dependencies between these equipment entities formed due to material flow, energy supply, or state coupling.

[0018] Step 111: Obtain the process design document for the production process, and extract the equipment identification list and material flow path description information from the process design document.

[0019] Process design documents are technical documents formed during the production process design phase. They record the models, quantities, installation locations, and connection relationships of all equipment involved in the production process, as well as the complete flow route of materials from the initial raw materials through each process to the final product.

[0020] In chemical production scenarios, process design documents are typically presented in the form of process flow diagrams or piping and instrumentation diagrams. When parsing the equipment identification list from the process design document, each equipment legend in the process flow diagram is traversed, and the corresponding equipment number, name, and functional description are read. All this information is then compiled into an equipment identification list. Simultaneously, material flow path description information is extracted from the process design document. This description information specifically records the direction, medium, and conditions of material transfer between adjacent equipment. The material flow path description information can be represented as material pipeline connection records in the process flow diagram. Each record includes the starting equipment number, ending equipment number, pipeline specifications, and transport medium type.

[0021] Step 112: Generate one-to-one corresponding device entity nodes based on the device identifier list, and assign a unique device identifier to each device entity node.

[0022] For each equipment record in the equipment identification list, create a corresponding equipment entity node. Each equipment entity node is represented in the data structure as an object containing a set of node attributes. This object stores all attribute information of the equipment, including equipment number, equipment name, equipment type, rated power, design temperature, design pressure, installation location coordinates, and the process stage number of the equipment in the production process.

[0023] Each device entity node is assigned a unique device identifier that can be identified in the entire topology graph. The device identifier is in the form of device type abbreviation plus numerical sequence number. For example, R101 represents the first reactor in the reaction process, E201 represents the first heat exchanger in the heat exchange process, and C301 represents the first compressor in the compression process. The hundreds digit of the numerical part represents the process stage number to which the device belongs.

[0024] Step 113: parse the material flow path description information to determine the flow direction of the material from the upstream equipment physical node to the downstream equipment physical node.

[0025] The material flow path description information is parsed line by line. For each material pipeline connection record, the starting and ending equipment numbers of the pipeline connection are determined. The starting equipment is the equipment from which the material flows out, and it is designated as the upstream equipment entity node; the ending equipment is the equipment from which the material flows in, and it is designated as the downstream equipment entity node.

[0026] If an intermediate buffer tank is involved in one of the material pipeline connection records, then the intermediate buffer tank also participates in the determination of the flow direction as a physical equipment node. After the material flows out of the upstream reactor, it enters the intermediate buffer tank for temporary storage through the pipeline, and then flows from the intermediate buffer tank into the downstream distillation column through another pipeline, thus forming two directional flow directions: the direction from the reactor to the buffer tank and the direction from the buffer tank to the distillation column.

[0027] Step 114: For each group of equipment entity nodes with direct material flow relationship, create a directed transfer edge from the upstream equipment entity node to the downstream equipment entity node, and gather all directed transfer edges into a directed transfer edge set.

[0028] For each determined material flow direction, a directed transfer edge is created between the corresponding upstream and downstream equipment entity nodes. This directed transfer edge is represented in the data structure as an object containing attributes such as start node identifier, end node identifier, transfer medium type, transfer capacity, and transfer delay time. All created directed transfer edges are collected in a directed transfer edge set data structure, where each element is a complete directed transfer edge object. The directed transfer edge set can be stored as an array or a linked list, with each element in the array corresponding to a directed transfer edge.

[0029] Step 115: Using the device entity node as the topology node and the directed transit edges in the set of directed transit edges as the topology directed edges, generate an initial process causal topology graph.

[0030] All equipment entity nodes are arranged as topological nodes in the topological space. The relative layout positions of the nodes are determined according to the actual sequence of each equipment in the production process. Equipment nodes belonging to the same process stage are arranged at the same level, with nodes in earlier process stages arranged upstream and nodes in later process stages arranged downstream. All directed transit edges are used as directed topological edges to connect the corresponding topological nodes.

[0031] This forms an initial process causal topology graph where nodes represent devices and directed edges represent the direction of material or energy transfer. This initial process causal topology graph is stored using an adjacency list data structure. The adjacency list contains N entries, where N equals the total number of device entity nodes. Each entry corresponds to a device entity node and stores an adjacency list. The adjacency list contains information on directed transfer edges from that node to all its downstream device entity nodes. Each entry includes the target node identifier and edge attributes.

[0032] Step 116: Identify and complete the implicit process-related edges of the initial process causal topology graph to obtain the process causal topology graph.

[0033] In chemical production processes, in addition to the direct material flow relationships explicitly recorded in the process design documents, there are also implicit process relationships caused by factors such as energy recovery systems, thermally coupled heat exchange networks, recirculated material backflow, or shared utility systems. These implicit relationships are not recorded in the process design documents in the form of explicit material pipeline records, but they have a significant impact on the transmission of abnormal emission indicators.

[0034] For example, temperature fluctuations at the hot end outlet of one heat exchanger can affect the hot end inlet temperature of another heat exchanger through a thermal coupling network, even though there is no direct material pipeline connection between the two heat exchangers. Therefore, it is necessary to identify and complete the implicit process correlation edges that are not directly recorded in the initial process causal topology graph but actually exist.

[0035] Step 1161: Extract the energy consumption time-series characteristics of each equipment entity node from the production process execution system. The energy consumption time-series characteristics include the sequence of changes in the energy medium consumption per unit time of the equipment entity node in multiple production batches.

[0036] A production process execution system (PPI) is a computer system that collects and stores real-time equipment operation data during the production process. This system records the operating parameters of each piece of equipment at a fixed sampling period. The system reads the historical energy consumption records of each equipment node across multiple production batches. For each equipment node, its energy consumption per unit time in each production batch is arranged chronologically to form an energy consumption time series characteristic. In chemical production scenarios, the energy medium can be represented as steam consumption, electricity consumption, or cooling water consumption, and the unit time can be measured in minutes. Each energy consumption time series characteristic is a sequence containing multiple data points, each data point including a timestamp and the corresponding energy consumption measurement value.

[0037] Step 1162: Extract the material throughput timing features of each equipment entity node from the production process execution system. The material throughput timing features include the change sequence of the material processing volume per unit time of the equipment entity node in the multiple production batches.

[0038] The system reads the historical material handling volume records of each equipment entity node within the same multiple production batches from the production process execution system. For each equipment entity node, its material handling volume per unit time in each production batch is arranged into a sequence according to time, forming a material throughput time series feature. Each material throughput time series feature is a sequence containing multiple data points, each data point in the sequence containing a timestamp and the corresponding material handling volume measurement value.

[0039] Step 1163: Standardize and eliminate fluctuations for the energy consumption time series characteristics and the material throughput time series characteristics to obtain detrended energy fluctuation sequences and detrended material fluctuation sequences.

[0040] Both energy consumption and material throughput time-series characteristics contain low-frequency trend components caused by factors other than equipment condition changes, such as production plan adjustments, ambient temperature variations, or raw material batch differences. These low-frequency trend components can mask the true fluctuation correlations between equipment caused by condition coupling. Standardized fluctuation elimination is achieved by performing local detrending processing on the original time-series data.

[0041] In specific processing, for each data point in the energy consumption time series characteristics, a local window is formed by taking a predetermined number of neighboring data points before and after that data point as the center. The weighted average of all data points within this window is calculated, and this weighted average is used as the trend estimate for that point. Then, this trend estimate is subtracted from the original measurement value for that point to obtain the detrended fluctuation value. After performing the above operation on all data points, all detrended fluctuation values ​​are arranged in chronological order to form a detrended energy fluctuation sequence. The same processing is performed on the material throughput time series characteristics to obtain the detrended material fluctuation sequence.

[0042] Step 1164: For any first device entity node and second device entity node in the initial process causal topology graph that are not directly connected by a directed transitive edge, calculate the energy fluctuation coherence degree between the detrended energy fluctuation sequence of the first device entity node and the detrended energy fluctuation sequence of the second device entity node.

[0043] Energy fluctuation coherence measures the degree of synchronization between two device entity nodes in terms of energy consumption fluctuations. In calculation, the detrended energy fluctuation sequences of the first and second device entity nodes are aligned on the time axis. Then, a sliding window approach is used to divide the two sequences into multiple overlapping time windows, each containing corresponding data point pairs from both sequences within that time period. For each time window, the covariance between corresponding data points of the two sequences within that window is calculated and divided by the product of their respective standard deviations to obtain the correlation coefficient for that window. The average of the correlation coefficients from multiple time windows yields the energy fluctuation coherence value.

[0044] Step 1165: Calculate the material fluctuation coherence degree between the detrended material fluctuation sequence of the first equipment entity node and the detrended material fluctuation sequence of the second equipment entity node.

[0045] The calculation method for material fluctuation synergy is the same as that for energy fluctuation synergy. The detrended material fluctuation sequence of the first equipment entity node and the detrended material fluctuation sequence of the second equipment entity node are aligned on the time axis. Using the same sliding window method, the correlation coefficient between corresponding data points of the two sequences within each time window is calculated. The average of the correlation coefficients of each time window is taken to obtain the value of material fluctuation synergy.

[0046] Step 1166: Perform weighted fusion processing on the energy fluctuation synergy degree and the material fluctuation synergy degree to generate the energy-material synergy causal strength.

[0047] During weighted fusion processing, corresponding fusion weights are assigned based on the confidence levels of the energy fluctuation synergy and material fluctuation synergy in representing process correlations. In chemical production scenarios, energy fluctuation synergy is highly sensitive to reflecting thermal coupling correlations between equipment, while material fluctuation synergy is highly sensitive to reflecting material circulation correlations between equipment. The confidence weights for both are predetermined based on the specific process type. Multiplying the energy fluctuation synergy by its corresponding confidence weight yields the energy-weighted component, and multiplying the material fluctuation synergy by its corresponding confidence weight yields the material-weighted component. Adding the energy-weighted component and the material-weighted component yields the energy-material synergistic causal strength.

[0048] Step 1167: When the energy-material synergistic causal strength exceeds a preset synergistic significance threshold, perform time lag mutual information analysis on the detrended energy fluctuation sequence of the first device entity node and the detrended material fluctuation sequence of the second device entity node to determine the energy conduction lag direction.

[0049] The preset synergistic significance threshold is determined based on the statistical distribution of energy and material synergistic causal intensity between unrelated equipment in historical normal operation data. The synergistic causal intensity values ​​between all unrelated equipment pairs during historical normal operation are extracted from the production process execution system. The distribution of these values ​​is calculated, and the 95th percentile is used as the threshold. When the energy and material synergistic causal intensity exceeds this threshold, it indicates a significant implicit correlation between the two equipment pairs, requiring further determination of the transmission direction. Time-lag mutual information analysis is used to determine the temporal direction of the predictive power between two fluctuation sequences. Specifically, multiple different time-lag offsets are set. For each offset, the detrended material fluctuation sequence is shifted forward by that offset relative to the detrended energy fluctuation sequence, and the mutual information value between the two sequences is calculated. This mutual information value measures the amount of information that the detrended energy fluctuation sequence can predict the detrended material fluctuation sequence. Simultaneously, the detrended energy fluctuation sequence is shifted forward by the same offset relative to the detrended material fluctuation sequence, and the mutual information value between the two sequences is calculated again. Compare the magnitudes of the two sets of mutual information values. If the maximum value of the former at a certain offset is greater than the maximum value of the latter at the same offset, it indicates that the energy fluctuation leads the material fluctuation.

[0050] Step 1168: Perform time lag mutual information analysis on the detrended material fluctuation sequence of the first equipment entity node and the detrended material fluctuation sequence of the second equipment entity node to determine the direction of material transmission lag.

[0051] The method for determining the material conduction lag direction is the same as that for energy conduction lag direction. Multiple different time lag offsets are set. For each offset, the detrended material fluctuation sequence of the second equipment entity node is shifted forward by that offset relative to the detrended material fluctuation sequence of the first equipment entity node, and the mutual information value is calculated. Simultaneously, the detrended material fluctuation sequence of the first equipment entity node is shifted forward by the same offset relative to the detrended material fluctuation sequence of the second equipment entity node, and the mutual information value is calculated again. The magnitudes of the two sets of mutual information values ​​are compared, and the time lag offset and its direction that maximize the mutual information value are taken as the material conduction lag direction.

[0052] Step 1169: If the energy conduction lag direction is consistent with the material conduction lag direction, then an implicit process association edge is generated from the equipment entity node corresponding to the leading fluctuation to the equipment entity node corresponding to the lagging fluctuation, and the implicit process association edge is added to the initial process causal topology graph to obtain the process causal topology graph.

[0053] When the energy conduction lag direction coincides with the material conduction lag direction, it indicates a conduction pattern from the same device to another in both energy and material dimensions. This confirms the existence of an implicit process correlation between the two devices caused by energy coupling or material circulation. A directed edge is generated from the device entity node corresponding to the leading fluctuation to the device entity node corresponding to the lagging fluctuation, serving as the implicit process correlation edge. This implicit process correlation edge is added to the directed transitive edge set of the initial process causal topology graph, and the adjacency list of the corresponding node in the adjacency list is updated. After performing the same implicit process correlation edge identification and addition operation on all device pairs that meet the conditions, a complete process causal topology graph is obtained.

[0054] Step 117: Persistently store the process causal topology map to the causal topology storage unit, and establish a corresponding causal index table for each device entity node.

[0055] The causal topology storage unit is a storage medium used to persistently store the causal topology graph of a process. It can be a relational database or a key-value store. The adjacency list structure of the process causal topology graph is serialized and written into the causal topology storage unit.

[0056] Simultaneously, a corresponding causal index table is established for each device entity node in the process causal topology graph. This causal index table uses the device identifier of the device as the primary key in the storage unit. The table records the device identifier of the device entity node, all the adjacency information of the node in the process causal topology graph, and the attribute parameters of all directed transit edges associated with the node.

[0057] Step 120: Perform abnormal fluctuation pattern recognition processing on the emission index data stream collected by the environmental monitoring sensor, generate an abnormal trigger signal, and mark the abnormal trigger signal as a causal injection event on the corresponding equipment entity node in the process causal topology map.

[0058] Emission indicator data streams are sequences of monitoring data collected and uploaded in real time by environmental monitoring sensors deployed at various emission outlets or monitoring points throughout the production process. In chemical production scenarios, environmental monitoring sensors may manifest as volatile organic compound (VOC) concentration detectors installed on the exhaust pipes of tail gas treatment units, chemical oxygen demand (COD) analyzers installed at wastewater discharge outlets, or particulate matter monitors installed at plant boundaries. Abnormal fluctuation pattern recognition processing is the process of detecting and identifying changes in the emission indicator data stream that deviate from the normal operating range.

[0059] Step 121: Receive emission index data stream uploaded by environmental monitoring sensors in time series format, wherein the emission index data stream contains multiple emission monitoring value points arranged in time sequence.

[0060] Each emission monitoring data point in the emission index data stream contains timestamp information and the corresponding emission index measurement value. Multiple emission monitoring data points are arranged consecutively according to the chronological order of their timestamps, forming a continuous time-series data stream. This data stream is transmitted from the environmental monitoring sensor to the receiving end in units of data frames, with each data frame carrying one or more emission monitoring data points.

[0061] Step 122: Perform time-series segmentation processing on the emission index data stream, and extract multiple continuous and partially overlapping emission index data segments from the emission index data stream according to a preset observation window duration.

[0062] The preset observation window duration is determined based on the normal fluctuation cycle of emission indicators in the production process. This normal fluctuation cycle is obtained through power spectral density analysis of historical emission data, and the cycle duration corresponding to the peak power spectral density is taken as the normal fluctuation cycle. The observation window duration is twice this normal fluctuation cycle. Starting from the beginning of the emission indicator data stream, the first emission indicator data segment is extracted with the observation window duration as its length. Then, the window start point is slid backward by a predetermined overlap duration, which is usually half of the observation window duration, to extract the second emission indicator data segment. This process continues, with the window start point slid backward by the overlap duration each time to extract the next segment, until the window end reaches the end of the data stream, thus forming multiple consecutive emission indicator data segments with partial temporal overlap between adjacent segments.

[0063] Step 123: Perform fluctuation baseline extraction processing on each emission index data segment to obtain the fluctuation baseline shape that represents the normal change trend of the emission index within the segment. Based on the deviation relationship between the fluctuation baseline shape and the actual emission monitoring value points within the emission index data segment, extract the abnormal fluctuation profile where the deviation exceeds the allowable deviation boundary.

[0064] The fluctuating baseline extraction process employs a locally weighted regression smoothing method. For each data point in the emission index data segment, a local neighborhood window is formed by taking a predetermined number of neighboring data points before and after that data point as the center. Weights are assigned to the data points within this window, with the weights decreasing according to the distance from each data point to the center point using a Gaussian kernel function; closer data points are assigned greater weights. After assigning weights, a weighted linear regression is performed on the data points within the window to obtain a smoothed estimate of the center point. After performing the above operation on all data points in the emission index data segment, the smoothed estimates of all data points are connected to form a smooth curve. This curve represents the fluctuating baseline shape, characterizing the regular changing trend of the emission index within that segment.

[0065] After obtaining the fluctuation baseline shape, the difference between each actual emission monitoring value point within the emission index data segment and the smoothed estimate of the corresponding time position on the fluctuation baseline shape is calculated to obtain the deviation magnitude of each data point. The allowable deviation boundary is determined based on the standard deviation of the deviation magnitude of the emission index under normal operating conditions, and three times the standard deviation is taken as the allowable deviation boundary. The continuous data point region where the deviation magnitude exceeds the allowable deviation boundary is extracted, and the contour formed by the data points in these regions is the abnormal fluctuation contour.

[0066] Step 124: Perform morphological matching processing on the abnormal fluctuation profile and multiple emission anomaly patterns in the preset emission anomaly pattern library, and calculate the morphological difference degree between the abnormal fluctuation profile and each emission anomaly pattern.

[0067] The emission anomaly pattern library stores various emission anomaly pattern templates pre-extracted from historical anomaly data, such as spike-jump patterns, continuous climb patterns, periodic oscillation patterns, and step change patterns. Each emission anomaly pattern is stored as a sequence of multi-dimensional feature vectors. Each feature vector in this sequence includes normalized values ​​in multiple dimensions, such as amplitude change rate, time span, rise steepness, and fall steepness. The anomaly fluctuation profile is also extracted as a sequence of feature vectors of the same dimensions. That is, for each data point in the anomaly fluctuation profile, the amplitude change rate of that point relative to the fluctuation baseline is calculated, and the relative time position of that point in the profile is recorded, forming the feature vector for that point.

[0068] Then, the feature vector sequence of the abnormal fluctuation profile is dynamically time-warped and aligned with the feature vector sequence of each emission anomaly pattern. That is, the optimal alignment path between the two sequences is found to minimize the cumulative distance between the aligned sequences. The cumulative distance value is used as the morphological difference between the abnormal fluctuation profile and the emission anomaly pattern.

[0069] Step 125: Select the emission anomaly pattern with the smallest morphological difference as the matching emission anomaly pattern, and extract the anomaly category label and anomaly triggering condition corresponding to the matching emission anomaly pattern.

[0070] Among all emission anomaly patterns, the one with the smallest morphological difference value is identified as the pattern that best matches the current anomaly fluctuation profile. Each emission anomaly pattern is pre-associated with an anomaly category label and an anomaly trigger condition. The anomaly category label describes the type of emission anomaly corresponding to the pattern, such as a sudden spike in VOC concentration, a continuous rise in COD, or periodic oscillations in particulate matter. The anomaly trigger condition determines whether the anomaly has reached the severity level required to trigger a response; this condition includes trigger conditions for amplitude thresholds and duration.

[0071] Step 126: When the abnormal fluctuation profile meets the abnormal triggering condition, an abnormal triggering signal is generated that includes the abnormal category label, the abnormal start timestamp, and the abnormal duration parameter.

[0072] The maximum deviation of the abnormal fluctuation profile is compared with the amplitude threshold in the abnormal triggering conditions, and the duration of the abnormal fluctuation profile is compared with the duration threshold in the abnormal triggering conditions. When the maximum deviation exceeds the amplitude threshold and the duration exceeds the duration threshold, it indicates that the current emission anomaly has reached a severity level requiring control measures. At this time, an abnormal trigger signal is generated, which encapsulates an anomaly category label, the timestamp corresponding to the first data point in the abnormal fluctuation profile as the anomaly start timestamp, and the time span between the last data point and the first data point in the abnormal fluctuation profile as the anomaly duration parameter.

[0073] Step 127: Based on the anomaly category label in the anomaly trigger signal, locate the equipment entity node in the process causal topology map that has a predefined monitoring association with the anomaly category label.

[0074] Predefined monitoring associations refer to the pre-association of one or more emission monitoring point identifiers with each equipment entity node in the process causal topology map. Each emission monitoring point identifier corresponds to an anomaly category label. This association is established during system initialization based on the emission monitoring point layout scheme of the production process. The system searches the predefined monitoring association table based on the anomaly category label to find the corresponding emission monitoring point identifier. Then, based on the mapping relationship between the emission monitoring point identifier and the equipment entity node, it locates the equipment entity node that generated the abnormal emission.

[0075] Step 128: Encapsulate the abnormal trigger signal into a causal injection event, mark the causal injection event on the located device entity node, and activate the link reverse propagation status identifier for the marked device entity node.

[0076] A causal injection event is a data structure used to represent external abnormal input signals. This data structure includes an event type field, an event payload field, and an event timestamp field. The entire content of the abnormal trigger signal is placed in the event payload field of the causal injection event, and the generation time of the abnormal trigger signal is placed in the event timestamp field. The causal injection event is then marked in the attribute field of the located device entity node. A causal injection event list is added to the node's attribute field, and the causal injection event is appended to this list. Simultaneously, a link backpropagation status flag is set in the status field of the device entity node, and the value of this status flag is set to active. This status flag indicates that this node will serve as the starting node for inference in subsequent link backpropagation processing.

[0077] Step 130: In the process causal topology graph, starting from the marked device entity node, perform link reverse propagation processing along the reverse direction of the directed transmission edge, and perform edge constraint relationship verification processing on each upstream device entity node to obtain the causal fork identification node.

[0078] Link reverse propagation processing refers to the process of tracing upstream from the marked device entity node in the process causal topology graph, following the direction opposite to the direction of the directed propagation edges. Edge constraint relationship verification processing verifies the validity of each directed propagation edge traversed during the reverse propagation process.

[0079] Step 131: Obtain all device entity nodes in the link reverse propagation state in the process causal topology map as the set of reverse inference starting nodes.

[0080] Traverse the status fields of all device entity nodes in the process causal topology graph, filter out those device entity nodes whose link reverse propagation status indicators are active, and gather these nodes into a set of reverse inference starting nodes.

[0081] Step 132: Starting from each of the reverse reasoning starting nodes in the set of reverse reasoning starting nodes, perform an upstream traversal along the opposite direction of the directed transit edge pointing to that reverse reasoning starting node to obtain multiple reverse reasoning paths.

[0082] For each backward inference starting node in the set of backward inference starting nodes, first query the incoming edge list of that node in the adjacency list, which is the set of all source nodes for directed transitive edges pointing to that node. Starting from each upstream node in the source node set, continue querying the incoming edge list of that upstream node, recursively tracing upstream until reaching the top node in the process causal topology graph that has no incoming edges. The sequence of nodes from each backward inference starting node to all its possible top nodes constitutes a backward inference path, and the nodes in each backward inference path are arranged in order from downstream to upstream.

[0083] Step 133: For each reverse reasoning path, extract the directed transit edges between two directly connected device entity nodes on the reverse reasoning path and read the pre-configured edge constraint relationship definition.

[0084] Edge constraint definitions are pre-configured sets of constraints for each directed transitive edge in the process causal topology graph. These constraints specify the allowable deviation range between the operating state parameters of upstream and downstream equipment entity nodes, as well as the allowed direction of fluctuation propagation. The edge constraint definitions for directed transitive edges between adjacent nodes on each reverse reasoning path are read from the adjacency list storage; these definitions are stored in the form of constraint condition expressions.

[0085] Step 134: Use the edge constraint relationship definition to determine whether the extracted directed transit edge satisfies the unidirectional transit constraint condition of the disturbance. If it does not satisfy the constraint, mark the directed transit edge as a constraint violation edge.

[0086] The unidirectional perturbation propagation constraint stipulates that on a directed propagation edge, a perturbation can only propagate unidirectionally from the upstream device entity node to the downstream device entity node. If there is also observable perturbation propagation in the direction from the downstream device entity node to the upstream device entity node, then the directed propagation edge does not satisfy the unidirectional perturbation propagation constraint. During the determination, the allowed perturbation propagation direction in the edge constraint definition is compared with the actual observed perturbation propagation direction between the two device entity nodes. If the actual observed direction is opposite to the allowed direction or there is bidirectional propagation, then the directed propagation edge is determined to be a constraint violation edge.

[0087] Step 135: During the traversal, record the upstream device entity node connected to the first constraint violation edge on each reverse inference path as a candidate fork node. Summarize all candidate fork nodes generated by the reverse inference paths, merge and deduplicate identical candidate fork nodes to obtain a set of candidate fork nodes.

[0088] During the upstream traversal from the starting node of the reverse inference, when a constraint violation edge is encountered for the first time on one of the reverse inference paths, the upstream device entity node connected to that constraint violation edge is recorded, i.e., the starting node of that edge, as a candidate fork node on that reverse inference path. After performing the same recording operation on all reverse inference paths, all recorded candidate fork nodes are merged, and duplicate node identifiers are removed to obtain a set of candidate fork nodes.

[0089] Step 136: For each candidate fork node in the candidate fork node set, retrieve the corresponding causal index table from the causal topology storage unit based on the device identifier of the candidate fork node, and extract the number of downstream directed transit edges associated with the candidate fork node.

[0090] For each candidate fork node in the candidate fork node set, the device identifier of the node is used as the query key to look up the corresponding causal index table record in the causal topology storage unit. The adjacency list of the candidate fork node is read from the causal index table, and the total number of directed transit edges pointing to downstream device entity nodes in the adjacency list is counted to obtain the number of downstream directed transit edges associated with the candidate fork node.

[0091] Step 137: Identify candidate fork nodes with a number of downstream directed transmission edges greater than the single path determination threshold as causal fork identification nodes, and add a fork identification mark to the causal fork identification node in the process causal topology graph.

[0092] The single path determination threshold is determined based on the distribution of downstream edges of normal nodes in the process causal topology graph. The median of the downstream edge distribution of normal nodes plus one standard deviation is taken as the single path determination threshold. When the number of downstream directed edges associated with a candidate fork node exceeds this threshold, it indicates that the candidate fork node has multiple downstream branch paths and exhibits forking characteristics, thus being identified as a causal fork identifier node. A fork identifier is added to this causal fork identifier node in the process causal topology graph, and this identifier is written into the node's attribute field as an attribute label.

[0093] Please refer to the following: Figure 2 This illustrates a schematic diagram of the generation of causal branching identifier nodes. Figure 2 The flowchart illustrates the complete processing chain from the triggering of abnormal emission indicators to the final generation of the causal branching identification node. Figure 2 The overall process is divided into three sub-stages (3.1, 3.2, 3.3) from top to bottom according to the timeline. The core elements and flow relationships of each stage are as follows: In sub-phase 3.1, environmental monitoring sensors continuously collect emission index data streams from the production process. This data stream is processed by abnormal fluctuation pattern recognition to generate anomaly trigger signals. Simultaneously, the system constructs a process causal topology graph connected by directed transitive edges based on the process association information between equipment entity nodes in the production process. The anomaly trigger signals are encapsulated as causal injection events and correspondingly marked on the relevant equipment entity nodes in the process causal topology graph, thereby anchoring the abnormal state of emission indicators to specific topology nodes and providing a starting point for reverse tracing. In sub-phase 3.2, starting from the marked device entity node, the link reverse propagation process is performed along the reverse direction of the directed propagation edges. During the reverse traversal, edge constraint relationship verification is performed on each directed propagation edge traversed. Figure 2The code details the decision branch for this verification: if the edge constraint relationship verification passes, continue tracing upstream along the current reverse path; if the verification fails, mark the directed transit edge as a constraint violation edge. This branch's decision logic is executed recursively until all reverse reasoning paths have been traversed. In sub-stage 3.3, the upstream device entity nodes that first appear constraint violation edges on each reverse inference path are summarized, and the determination is made in combination with the number of downstream directed transmission edges associated with each node: when the number of downstream directed transmission edges exceeds a preset threshold, the node is determined as a causal fork marker node; otherwise, if the upstream correlation does not exceed the threshold, the disturbance propagation path backtracking process continues. Figure 2 The bottom is marked with a prominent symbol indicating the final causal bifurcation identification node. This node marks the key location where the abnormal disturbance occurs path bifurcation in the process causal topology map, providing a clear topological anchor point for tracing the disturbance propagation path and locating the root cause.

[0094] Step 140: Based on the causal bifurcation identification node, trace back the disturbance propagation path, extract the operating status information of each device entity node on the disturbance propagation path and perform semantic synthesis processing to generate disturbance propagation interpretation semantics. The disturbance propagation interpretation semantics includes a description of the propagation direction of the abnormal disturbance in the process causal topology map and a description of the key bifurcation position.

[0095] In this embodiment of the invention, the perturbation propagation interpretation semantics is a data structure used to describe the complete process of an abnormal perturbation propagating from the starting node to the causal fork identifier node. Step 140 further includes: Step 141: Using the causal fork identifier node as the backtracking endpoint and the marked device entity node as the backtracking starting point, extract the complete node link from the backtracking starting point to the backtracking endpoint in the process causal topology graph along the positive direction of the directed transmission edge, as the disturbance propagation path.

[0096] In the process causal topology graph, starting from the marked device entity node, the process traverses along the forward direction of the directed propagation edges. During the traversal, a breadth-first search strategy is used to explore downstream layer by layer until the causal branch marker node is reached. All nodes traversed from the marked device entity node to the causal branch marker node are connected in traversal order to form a complete node link, which is the disturbance propagation path.

[0097] Step 142: Extract the device identifier corresponding to each device entity node on the disturbance propagation path, and obtain the running status record of each device entity node during the period covered by the abnormal duration parameter from the production process execution system based on the device identifier.

[0098] For each device entity node along the disturbance propagation path, extract its device identifier. Use this device identifier to initiate a query request to the production process execution system, with the query time range being the period covered by the exception start timestamp plus the exception duration parameter. Obtain the device's operating status record for that period from the query results returned by the production process execution system. This operating status record includes time-series data of multiple operating parameters.

[0099] Step 143: Perform state parameter extraction processing on the acquired operating status records to obtain the operating parameter deviation vector of each device entity node. The operating parameter deviation vector is used to characterize the offset direction and degree of the device entity node relative to its standard operating state.

[0100] During the status parameter extraction process, for each operating parameter in the operating status record of each device entity node, the standard parameter value of that device under normal operating conditions is extracted from the production process execution system. The difference between the actual measured value and the standard parameter value at each time point in the operating status record is calculated. After normalizing the difference sequence, the deviation value sequence of that parameter is obtained. The deviation value sequences of all operating parameters of the same device entity node are concatenated along the parameter dimension to form a multi-dimensional operating parameter deviation vector.

[0101] Step 144: The operating parameter deviation vectors of all device entity nodes on the disturbance propagation path are spliced ​​together in node order to obtain a disturbance propagation deviation vector sequence.

[0102] Following the node order from the start to the end of the disturbance propagation path, the operating parameter deviation vectors of each device entity node are arranged sequentially. For two adjacent nodes, the operating parameter deviation vectors of the preceding and following nodes are concatenated along the vector dimension to form the propagation deviation vector between the adjacent node pairs. Arranging the propagation deviation vectors between all adjacent node pairs according to node order yields the disturbance propagation deviation vector sequence.

[0103] Step 145: Perform time-series causal correlation analysis on the disturbance propagation deviation vector sequence to identify whether there is consistency in the causal propagation direction between the operating parameter deviation vectors of adjacent device entity nodes.

[0104] During the temporal causal association analysis, for each pair of adjacent nodes in the perturbation propagation deviation vector sequence, the cosine of the angle between the changing direction of the operating parameter deviation vector of the preceding node and the changing direction of the operating parameter deviation vector of the following node is calculated. When the cosine of the angle is greater than a predetermined directional consistency threshold, it is determined that there is causal propagation directional consistency between the operating parameter deviation vectors of the two nodes, that is, the parameter deviation of the preceding node is propagated to the following node along the direction of the directed propagation edge.

[0105] Step 146: Based on the consistency identification result of the causal propagation direction, generate a perturbation propagation direction description from the starting node to the branching node. The perturbation propagation direction description includes semantic annotations of propagation continuity segments and propagation breakpoint segments.

[0106] In the disturbance propagation path, all adjacent node pairs that are determined to have consistent causal propagation directions are marked as propagation coherent segments, and adjacent node pairs that are determined not to have consistent causal propagation directions are marked as propagation breakpoint segments. After marking all node pairs, the propagation coherent segments are connected in the path order to form a coherent description, and the propagation breakpoint segments are connected in the path order to form a breakpoint description. These are then merged into a disturbance propagation direction description from the starting node to the bifurcation node.

[0107] Step 147: Extract the local adjacent substructure of the causal fork identifier node in the process causal topology graph, determine the composition information of each downstream branch in the local adjacent substructure, and obtain the description of the key fork position.

[0108] A local adjacency substructure is a subgraph structure centered on a causal fork identifier node, containing all directly downstream device entity nodes of that node, as well as the directly downstream device entity nodes below each of those downstream device entity nodes. From this local adjacency substructure, the starting node identifier, branch length, and ending node identifier of each downstream branch are extracted, and this information is summarized into a key fork location description.

[0109] Step 148: Semantically merge the description of the disturbance propagation direction and the description of the key bifurcation location, and generate disturbance propagation interpretation semantics by combining the abnormal category label in the abnormal trigger signal. The disturbance propagation interpretation semantics includes the disturbance source direction, the assessment of the propagation path coherence, and the overview of the downstream impact branches of the bifurcation.

[0110] The description of the perturbation propagation direction is used as part of the assessment of the propagation path coherence; the description of key bifurcation locations is used as part of the overview of the downstream branches affected by the bifurcation; and the anomaly category label is used as the classification basis for the perturbation source. These three parts are assembled according to a predefined semantic template to form a structured semantic data object for perturbation propagation interpretation.

[0111] Step 150: Call the root cause matching logic tree to perform root cause logic matching processing on the disturbance propagation interpretation semantics to obtain the corresponding operation primitive sequence, and convert the operation primitive sequence into control instructions and write them into the target control system position to regulate the production equipment.

[0112] In step 150, the root cause matching logic tree is a decision tree structure used to map perturbation propagation interpretation semantics to root cause determination conclusions and operational primitives. Specifically, step 150 includes: Step 151: Load a preset root cause matching logic tree. The root node of the root cause matching logic tree corresponds to the emission anomaly category, the branch nodes correspond to the transmission path feature items in the disturbance transmission interpretation semantics, and the leaf nodes correspond to the root cause determination conclusion.

[0113] The pre-built root cause matching logic tree data file is read from the storage medium. The root cause matching logic tree is represented as a multi-branch tree in terms of data structure. Its root node uses the emission anomaly category as the discrimination condition, each branch node uses the transmission path feature item as the discrimination condition, and the deepest leaf node stores the root cause determination conclusion. Each branch node contains a transmission path feature item template and a set of logical edges pointing to its child nodes.

[0114] Step 152: Align the source of disturbance in the disturbance propagation interpretation semantics with the root node of the root cause matching logic tree by category, and locate the corresponding starting matching branch.

[0115] During category alignment, the abnormal category label in the perturbation propagation interpretation semantics is matched with the emission abnormal category identifier associated with the root node of the root cause matching logic tree. The child node corresponding to the root node that matches the abnormal category label is used as the starting matching branch node.

[0116] Step 153: Under the initial matching branch, extract the propagation path coherence assessment and the downstream influence branch overview in the perturbation propagation interpretation semantics to obtain the semantic feature set.

[0117] Extract the cumulative proportion of transmission coherence segments and the variance of transmission breakpoint intervals from the semantic interpretation of disturbance transmission. Extract the total number of downstream branches and the type distribution vector of the device entity nodes pointed to by each downstream branch from the overview of downstream influence branches. Summarize all extracted features into a semantic feature set.

[0118] Step 154: Use the semantic feature set and the propagation path feature items on the root cause matching logic tree branch nodes to perform layer-by-layer matching processing. For each matched node, migrate downwards along the corresponding logical edge until the leaf node is reached.

[0119] It can be understood that the layer-by-layer matching process is an iterative matching process that starts from the initial matching branch node and proceeds downwards level by level.

[0120] Step 1541: Obtain the starting matching branch node located by the category alignment processing and set it as the current level node; construct the migrated node identifier sequence; add the node identifier of the current level node to the migrated node identifier sequence; and read the transmission path feature item stored in the current level node.

[0121] Set the starting matching branch node as the current level node. Create a migrated node identifier sequence data structure, initially empty, and append the node identifier of the current level node to the end of the sequence. Read the stored propagation path feature from the attribute field of the current level node, which contains propagation coherence constraints and branching structure constraints.

[0122] Step 1542: Extract coherence features from the conduction path coherence assessment contained in the semantic feature set, obtain the cumulative proportion of conduction coherence segments and the variance of conduction breakpoint intervals in the conduction path coherence assessment, and combine the cumulative proportion of conduction coherence segments and the variance of conduction breakpoint intervals into a coherence feature vector.

[0123] In the path continuity assessment, the total length of continuous conduction segments and the total length of discontinuous conduction segments were recorded. The total length of continuous conduction segments was calculated by dividing it by the sum of the total lengths of continuous conduction segments and discontinuous conduction segments to obtain the cumulative proportion of continuous conduction segments. Simultaneously, the positional distribution of discontinuous conduction segments along the path was extracted, and the variance of the interval length between each discontinuous conduction segment was calculated to obtain the interval variance of discontinuous conduction segments. The cumulative proportion of continuous conduction segments and the interval variance of discontinuous conduction segments were combined as two dimensions to form a continuity feature vector.

[0124] Step 1543: Extract bifurcation features from the bifurcation downstream influence branch overview contained in the semantic feature set, obtain the total number of bifurcation downstream branches and the type distribution vector of the device entity node pointed to by each downstream branch in the bifurcation downstream influence branch overview, and concatenate the total number of bifurcation downstream branches with the type distribution vector to generate a bifurcation structure feature vector.

[0125] The total number of downstream branches is read from the overview of downstream influence branches. For each downstream branch, the device type identifier of the terminal device entity node is obtained, and the frequency of each device type in all downstream branches is counted to form a type distribution vector. The total number of downstream branches is used as the first dimension, and all dimensions of the type distribution vector are arranged sequentially after it, and concatenated to form the fork structure feature vector.

[0126] Step 1544: Calculate the interval boundary offset between the coherence feature vector and the coherence reference ratio interval in the conduction coherence constraint, and the over-limit ratio of the conduction breakpoint segment interval variance in the coherence feature vector relative to the allowable upper limit of the breakpoint interval variance. Generate a coherence matching metric based on the interval boundary offset and the over-limit ratio.

[0127] The conduction coherence constraint includes a coherence reference proportion interval and an upper limit for the variance tolerance of breakpoint intervals. The cumulative proportion of conduction coherent segments in the coherence eigenvector is compared with the upper and lower limits of the coherence reference proportion interval. The absolute offset of the cumulative proportion of conduction coherent segments exceeding the interval boundary is calculated, and the ratio of this absolute offset to the interval width is taken as the interval boundary offset. The variance of the conduction breakpoint intervals in the coherence eigenvector is compared with the upper limit for the variance tolerance of breakpoint intervals. The ratio of the amount by which the variance of the conduction breakpoint intervals exceeds the upper limit to the upper limit is taken as the over-limit ratio. The interval boundary offset and the over-limit ratio are weighted and summed to obtain the coherence matching metric.

[0128] Step 1545: Calculate the offset between the total number of downstream branches in the bifurcation structure feature vector and the number interval of the branch number reference interval, and calculate the Mahalanobis distance between the type distribution vector in the bifurcation structure feature vector and the device type distribution tolerance deviation matrix. Generate a bifurcation structure matching metric based on the number interval offset and the Mahalanobis distance.

[0129] The bifurcation structure constraints include a reference interval for the number of branches and a tolerance matrix for equipment type distribution. The total number of downstream branches in the bifurcation structure eigenvector is compared with the upper and lower limits of the reference interval for the number of branches. The absolute offset exceeding the interval boundary is calculated, and the ratio of this offset to the interval width is taken as the quantity interval offset. The Mahalanobis distance between the type distribution vector in the bifurcation structure eigenvector and the tolerance matrix for equipment type distribution is calculated. This Mahalanobis distance is obtained by calculating the covariance-weighted distance between the type distribution vector and the center vector of the tolerance matrix. The quantity interval offset and the Mahalanobis distance are then weighted and summed to obtain the bifurcation structure matching metric.

[0130] Step 1546: Extract the coherence contribution weight and bifurcation structure contribution weight associated with the current level node from the root cause matching logic tree. Based on the coherence contribution weight and bifurcation structure contribution weight, perform a weighted synthesis of the coherence matching metric and the bifurcation structure matching metric to obtain the node comprehensive matching degree.

[0131] Each branch node is pre-associated with a set of weight coefficients, including a coherence contribution weight and a bifurcation structure contribution weight, the sum of which equals 1. The coherence matching metric is multiplied by the coherence contribution weight to obtain the coherence weighted component, and the bifurcation structure matching metric is multiplied by the bifurcation structure contribution weight to obtain the bifurcation structure weighted component. The two weighted components are added together to obtain the overall node matching degree.

[0132] Step 1547: Calculate the matching migration threshold based on the level depth of the current level node in the root cause matching logic tree, and compare the overall matching degree of the node with the matching migration threshold.

[0133] The matching migration threshold is dynamically determined based on the depth of the current level node in the logic tree. The deeper the level, the lower the matching migration threshold. Specifically, it is obtained by multiplying the baseline threshold by the inverse of the level depth. The calculated overall node matching degree is then compared numerically with this matching migration threshold.

[0134] Step 1548: When the overall matching degree of the node is greater than or equal to the matching migration threshold, the current level node is identified as a matching passed node, and the direct child node is obtained along the logical edge led out by the matching passed node. The direct child node is determined as the current node of the next level, and the node identifier of the direct child node is recorded in the migrated node identifier sequence. When it is determined that the node type of the direct child node does not belong to the leaf node type, the matching process from reading the transmission path feature item to calculating the overall matching degree of the node is repeated with the direct child node as the current level node.

[0135] When the overall matching degree of a node is greater than or equal to the matching migration threshold, it indicates that the node at the current level has passed the matching verification. All direct child nodes of the current level node are retrieved from its logical edge set, and the first direct child node is designated as the current node for the next level. The node identifier of this direct child node is appended to the end of the migrated node identifier sequence. The node type attribute of this direct child node is checked. If the attribute indicates that the node is not a leaf node type, this direct child node is used as the current level node, and the process jumps back to step 1541 to restart the layer-by-layer matching process.

[0136] Step 1549: When the overall matching degree of the node is less than the matching migration threshold, stop the matching migration along the current logical edge and output a matching termination indication.

[0137] When the overall matching degree of a node is less than the matching migration threshold, it indicates that the node at the current level has failed the matching verification. At this time, the migration along the current logical edge is stopped, and a matching termination indication signal is generated. This signal contains the node identifier and the overall matching degree value of the node at the current level.

[0138] Step 15410: Before repeatedly executing the matching process with the direct child node as the current level node, check whether the node identifier of the direct child node already exists in the migrated node identifier sequence. If it already exists, trigger the logical edge loop termination operation and stop the layer-by-layer matching process.

[0139] Before determining a direct child node as the current node of the next level, traverse all node identifiers in the migrated node identifier sequence and check if there is a record with the same node identifier as the direct child node. If a record with the same identifier exists, it indicates that a loop has occurred in the matching process, triggering the logical edge loop termination operation and immediately stopping the entire layer-by-layer matching process.

[0140] Step 15411: When it is determined that the node type of the direct child node belongs to the leaf node type, the layer-by-layer matching process is terminated, and the direct child node is recorded as the leaf node reached by the layer-by-layer matching.

[0141] When the node type attribute of a direct child node indicates that the node belongs to the leaf node type, it means that the end of the logic tree has been reached. At this time, the layer-by-layer matching process is terminated, and the node identifier of the leaf node is recorded as the matched leaf node identifier.

[0142] Step 155: When a leaf node is reached, extract the root cause determination conclusion stored in the leaf node and record the logical edge sequence that leads to the leaf node as a matching path.

[0143] The root cause determination conclusion is read from the attribute field of the matched leaf node. At the same time, all node identifiers from the starting matching branch node to the leaf node are extracted from the migrated node identifier sequence, and these node identifiers are arranged in order to form a logical edge sequence, which is recorded as the matching path.

[0144] Step 156: Index the corresponding set of operation primitives from the preset root cause response strategy mapping table according to the root cause determination conclusion.

[0145] The root cause response strategy mapping table is a pre-built key-value pair mapping table, where the key is the identifier of the root cause determination, and the value is a set of operation primitives consisting of multiple operation primitives for that root cause. Using the root cause determination identifier stored in the leaf nodes of the logical edge sequence recorded in the matching path as the query key, a lookup is performed in the root cause response strategy mapping table to obtain the corresponding set of operation primitives.

[0146] Step 157: Select the sequence of operation primitives with the highest execution priority from the set of operation primitives based on the propagation bifurcation severity factor reflected in the matching path.

[0147] The propagation fork severity factor is extracted from the fork structure matching metric of each branch node in the matching path. The maximum value of the fork structure matching metric of all branch nodes in the matching path is taken as the severity factor. Each operation primitive in the set of operation primitives is pre-associated with an applicable severity range. The severity factor is compared with the applicable severity range of each operation primitive. One or more operation primitives with the highest execution priority among the operation primitives whose applicable severity range contains the severity factor are selected and arranged in descending order of priority to form an operation primitive sequence.

[0148] Step 158: Convert each operation primitive in the selected operation primitive sequence into a control instruction adapted to the target control system site communication protocol, and embed the abnormal start timestamp and the device identifier of the marked device entity node in the abnormal trigger signal into the control instruction.

[0149] A target control system endpoint refers to the controller interface in the production process used to receive and execute control commands. For each operation primitive in the sequence of operation primitives, the target device identifier and control parameter value are extracted from that primitive. Based on the communication protocol format supported by the target control system endpoint, the control parameter value is encapsulated into a data frame structure specified by the communication protocol. The exception start timestamp value from the exception trigger signal and the device identifier value of the tagged device entity node are written into the designated fields of the data frame to form a complete control command data frame.

[0150] Step 159: The generated control commands are written sequentially into the corresponding target control system positions according to the execution order of the operation primitive sequence, so as to adjust the operating parameters of the relevant equipment entity nodes in the production process.

[0151] Following the order of the operation primitives in the operation primitive sequence, the corresponding control command data frames are sent to the target control system endpoints one by one. After receiving the control command, each target control system endpoint adjusts the operating parameters of its corresponding equipment entity node according to the control parameter values ​​in the command, such as adjusting the stirring speed of the reactor, adjusting the opening of the cooling water valve of the heat exchanger, or adjusting the outlet pressure setpoint of the compressor.

[0152] Please refer to the following: Figure 1 and Figure 3 , Figure 3 The process of constructing the process causal topology graph is illustrated. As detailed in step 110, the system first extracts the equipment identification list and material flow path description information from the process design document based on the process association information between equipment entity nodes in the production process, generates one-to-one corresponding equipment entity nodes, and creates directed transit edges from upstream to downstream for each group of equipment with direct material flow relationships, collecting them into a set of directed transit edges. Based on this, Figure 3The process of identifying and completing implicit process-related edges in the initial process causal topology map is further illustrated. The implicit related edges identified after the fusion analysis of energy consumption time series characteristics and material throughput time series characteristics are added to the map, and finally a complete process causal topology map is obtained.

[0153] Figure 3 The anomaly injection and reverse tracing process is also illustrated, corresponding to the core logic of steps 120 to 130. As described in step 120, the emission index data stream collected by the environmental monitoring sensor is processed by anomaly fluctuation pattern recognition to generate an anomaly trigger signal. This signal is encapsulated as a causal injection event and marked on the corresponding equipment entity node in the process causal topology map. Figure 3 The arrows clearly indicate the propagation path from emission index data through the anomaly trigger signal to the causal injection event. Then, as detailed in step 130, reverse propagation is performed along the directed propagation edges, starting from the marked node. Figure 3 The diagram illustrates the logic for verifying the edge constraint relationship of each directed transit edge using a branching graph: if the verification passes, the process continues to trace upstream along the reverse path; if the verification fails, a fork identification is triggered, and finally, a causal fork identifier node is output, thereby accurately locating the key position where the abnormal disturbance occurs at the path fork in the process causal topology graph.

[0154] Figure 3 The perturbation propagation semantic generation and root cause matching process is further illustrated, corresponding to the processing logic of steps 140 to 150. As described in step 140, the system traces the perturbation propagation path with the causal bifurcation marker node as the tracing endpoint, extracts the operating status information of each device entity node on the path, and generates perturbation propagation explanatory semantics containing propagation direction description and key bifurcation position description after state parameter extraction, deviation vector splicing and time-series causal association parsing. Figure 3 The right-hand "information card" schematically illustrates the two core components of this semantics: a description of the propagation direction and a description of the key bifurcation location, elevating the anomaly propagation process from a numerical level to an interpretable, structured semantic level. Subsequently, as described in step 150, the system calls the root cause matching logic tree to perform layer-by-layer matching processing on this semantics. Upon successful matching, the corresponding operation primitive sequence is output, which is then converted into control instructions adapted to the communication protocol of the target control system and finally written into the target control system location to regulate the operating parameters of the production equipment.

[0155] at last, Figure 3 The repeated appearance of equipment entity nodes and directed transit edges in the graph is marked by dashed boxes at the bottom, indicating that these two types of elements run through the entire process of constructing the process causal topology graph, tracing anomalies, and executing control measures. They play a pivotal role as the foundation of the data structure and the channel for causal reasoning at each stage.

[0156] Based on steps 110-150, the method further includes: Step 210: Obtain the timing data of the control response generated by the equipment entity node in the production process after the target control system site executes the control command, and extract the control transition data segment corresponding to the control transition period from the timing data of the control response.

[0157] Within a predetermined time window after the control command is executed, the system continuously receives control response timing data from the production process execution system. This control response timing data includes records of changes in operating parameters of the controlled equipment entity node and its downstream equipment entity nodes during the control process. A time period from the moment the control command is written to the moment the emission index data stream returns to a steady state is extracted from this timing data as the control transition period. All data points within this period are then extracted as the control transition data segment.

[0158] Step 220: Perform stationarity transition analysis on the control transition data segment, identify the transition time anchor point in the control transition data segment where the emission index data stream migrates from an abnormal fluctuation mode to a steady-state convergence mode, and divide the control transition data segment into a control response early segment and a control response late segment based on the transition time anchor point.

[0159] During the stationarity transition analysis, each data point in the control transition data segment is traversed in chronological order, and the absolute value of the rate of change for each data point is calculated sequentially. When the absolute value of the rate of change continuously decreases and eventually falls below a predetermined rate of change threshold, the time position corresponding to that data point is marked as the transition time anchor point. The region of data points in the control transition data segment before the transition time anchor point is divided into the early segment of the control response, and the region of data points after the transition time anchor point is divided into the late segment of the control response.

[0160] Step 230: Extract the emission index fluctuation attenuation slope features of the early segment of the control response and the emission index steady-state residual distribution features of the later segment of the control response, and perform multi-dimensional feature alignment processing on the emission index fluctuation attenuation slope features and the emission index steady-state residual distribution features to generate a multi-dimensional feature descriptor of the control effect.

[0161] For the early stage of the control response, the rate of change of each data point relative to time is calculated, and the median value of the rate of change is taken as the emission index fluctuation attenuation slope feature. For the later stage of the control response, the difference between each data point and the emission index standard value is calculated, and the root mean square value of the difference is taken as the emission index steady-state residual distribution feature. The emission index fluctuation attenuation slope feature and the emission index steady-state residual distribution feature are arranged in a predetermined feature dimension order to form a multi-dimensional feature descriptor of the control effect.

[0162] Step 240: Associate and encapsulate the multidimensional feature descriptor of the regulation effect with the anomaly category label in the causal injection event to generate a regulation effect feedback record containing the anomaly category, the multidimensional feature descriptor of the regulation effect, and the device identifier of the fork identifier node. Persist the regulation effect feedback record to the causal index table corresponding to the causal fork identifier node in the causal topology storage unit.

[0163] Create a data structure for the feedback record of the control effect. This data structure includes an anomaly category field, a multi-dimensional feature descriptor field for the control effect, and a device identifier field for the bifurcation identification node. Write the anomaly category label from the causal injection event to the anomaly category field, write the generated multi-dimensional feature descriptor of the control effect to the multi-dimensional feature descriptor field, and write the device identifier of the causal bifurcation identification node to the device identifier field. Store this control effect feedback record in the causal index table corresponding to the device identifier of the causal bifurcation identification node in the causal topology storage unit.

[0164] Based on steps 110-150, the method further includes: Step 310: Monitor the emission index data stream of the marked equipment entity nodes in the process causal topology map after the target control system site executes the control command. When an abnormal fluctuation pattern of the emission index data stream is detected and fails to converge to a steady state within a preset control tolerance time, generate an insufficient control indication signal.

[0165] The preset tolerance period is determined based on the historical control response time distribution of the emission index, taking twice the median of the historical control response time distribution as the tolerance period. Timing begins from the moment the control command is written, continuously monitoring the emission index data stream of the marked equipment entity node. If the fluctuation amplitude of the emission index data stream continuously exceeds the allowable deviation boundary within the tolerance period, it is determined that it has not converged to a steady state, and an insufficient control indication signal is generated.

[0166] Step 320: Using the insufficient regulation indication signal as a trigger condition, extract the set of adjacent device entity nodes that have a directed transitive edge connection relationship with the marked device entity node from the process causal topology map, and obtain the real-time operating parameter set of each adjacent device entity node in the set of adjacent device entity nodes from the production process execution system.

[0167] When the insufficient control indication signal is triggered, the adjacency table of the marked equipment entity node is queried, and all downstream and upstream adjacent nodes are aggregated into a set of adjacent equipment entity nodes. For each adjacent equipment entity node in this set, a real-time data request is sent to the production process execution system to obtain the current operating parameter values ​​of the node, forming the real-time operating parameter set of the node.

[0168] Step 330: Perform cross-node cross-response analysis on the emission index data stream of the marked device entity node and the real-time operating parameter set of each adjacent device entity node to identify delay-related parameter items in the real-time operating parameter set that have a response delay associated with the residual abnormal fluctuations of the emission index data stream.

[0169] During cross-node response analysis, the current fluctuation pattern of the emission index data stream of the marked equipment entity node is subjected to time-lag cross-correlation analysis with the parameter values ​​in the real-time operating parameter set of each adjacent equipment entity node. The cross-correlation coefficient of each parameter value sequence relative to the emission index data stream at different lag time offsets is calculated. The lag time offset at which the cross-correlation coefficient reaches its maximum value is selected as the response delay of that parameter item. If the response delay is within a predetermined effective delay interval, the parameter item is marked as a delay-related parameter item.

[0170] Step 340: Determine the target device entity node in the set of adjacent device entity nodes that generates the control compensation requirement based on the delay association parameter item, generate a compensation operation primitive for the target device entity node, and convert the compensation operation primitive into a compensation control instruction and write it into the control system position corresponding to the target device entity node.

[0171] Based on the device entity node to which the delay-related parameter item belongs, the node that generated the delay-related parameter item is located in the set of adjacent device entity nodes as the compensation target device entity node. For this compensation target device entity node, a compensation operation primitive is generated based on the response delay of the delay-related parameter item and the current deviation of the parameter item. This compensation operation primitive includes the compensation control parameter value and the compensation execution direction. The compensation operation primitive is encapsulated into a compensation control instruction according to the communication protocol format of the target control system point, and the compensation control instruction is written into the control system point corresponding to the compensation target device entity node.

[0172] The models involved in this invention include emission anomaly pattern matching models from the emission anomaly pattern library, root cause matching logic trees, and correlation response analysis modules in cross-node cross-response analysis. To enable those skilled in the art to clearly and completely implement this invention, the specific architecture of these models, the technical implementation of each module, the training data and process, and the application methods of the models are described below: I. The emission anomaly pattern matching model in the emission anomaly pattern library adopts a nearest neighbor matching architecture based on dynamic time warping distance. This matching model consists of a feature extraction submodule and a distance calculation submodule. The feature extraction submodule extracts feature vector sequences from the abnormal fluctuation contours to be matched, converting the original contour data into a standardized feature vector sequence containing dimensions of amplitude change rate, time span, rising edge steepness, and falling edge steepness. The distance calculation submodule uses a dynamic time warping algorithm to calculate the optimal alignment cumulative distance between the feature vector sequence to be matched and the feature vector sequences of each stored pattern in the pattern library. During the model construction phase, M groups of labeled samples containing various emission anomaly forms are extracted from the historical production process execution system. Each group of samples contains the original data of the abnormal fluctuation contours and its corresponding anomaly category label. The feature vector sequences of each group of samples are associated with their corresponding anomaly category labels and stored in the emission anomaly pattern library as reference patterns. During the application phase, the abnormal fluctuation contours extracted in real time are input into the feature extraction submodule to obtain a feature vector sequence. The distance calculation submodule calculates the dynamic time-normalized distance between this sequence and the feature vector sequences of each reference pattern in the pattern library in turn, and outputs the abnormal category label corresponding to the reference pattern with the smallest distance value as the matching result.

[0173] II. The root cause matching logic tree adopts a multi-branch decision tree architecture, consisting of a root node layer, branch node layers, and leaf node layers. The root node uses the emission anomaly category as the discriminant attribute, the branch nodes use the transmission path feature items as the discriminant attribute, and the leaf nodes store the root cause determination conclusions. Each branch node contains a matching determination submodule and a logical edge selection submodule. The matching determination submodule calculates the comprehensive matching degree of the branch node based on the input semantic feature set, and the logical edge selection submodule selects the sub-branch to enter based on the comparison result of the comprehensive matching degree and the matching migration threshold. During the construction phase, process experts manually annotate and extract rules based on historical anomaly handling cases, organizing the root cause determination conclusions corresponding to different transmission path features under each anomaly category into a tree structure. During the application phase, after extracting the perturbation transmission interpretation semantics into a semantic feature set, it is input layer by layer from the root node to the matching determination submodule of the branch node. The logical edge selection submodule migrates downwards along the logical edges according to the matching results until it reaches the leaf node and outputs the root cause determination conclusion stored in the leaf node.

[0174] III. The correlation response analysis module in the cross-node cross-response analysis adopts a computational architecture based on time-lag cross-correlation analysis. This module consists of a lag cross-correlation calculation submodule and a delay correlation determination submodule. The lag cross-correlation calculation submodule is used to calculate the cross-correlation coefficient between the current fluctuation sequence of the emission index data stream of the marked equipment entity node and the real-time operating parameter sequence of each adjacent equipment entity node under different lag offsets. The range of lag offset values ​​is set to all integer values ​​from the upper limit of negative offset to the upper limit of positive offset. The delay correlation determination submodule is used to select the offset that reaches the maximum value according to the cross-correlation coefficient corresponding to each lag offset as the response delay of the parameter item, and compares the response delay with the predetermined effective delay interval to determine whether the parameter item belongs to the delay correlation parameter item. During the parameter configuration phase, this module determines the upper and lower boundary values ​​of the effective delay interval and the significance threshold of the cross-correlation coefficient based on the statistical distribution of response delay between devices in historical normal operation data. During the application phase, when the insufficient control indication signal is triggered, the correlation response analysis module reads the sequence data parameter by parameter from the real-time operating parameter set of each adjacent device entity node. The lag cross-correlation calculation submodule calculates the lag cross-correlation coefficient between each parameter sequence and the emission index data stream fluctuation sequence in turn. The delay correlation determination submodule outputs a list of identifiers for the delay correlation parameter items.

[0175] This invention constructs a process causal topology map, explicitly representing the material transfer, energy coupling, and implicit correlations between equipment entities in the production process as a directed topological structure. This provides a global causal reasoning basis for tracing the source of abnormal emissions. Based on this, abnormal emission indicators collected by environmental monitoring sensors are anchored to the corresponding equipment entities as causal injection events. Link propagation and edge constraint verification are then performed in reverse along the directed transmission edges to accurately locate causal branching nodes, avoiding misjudgments caused by relying solely on single-device data or simple temporal correlations in traditional methods. Furthermore, by tracing the disturbance propagation path and examining each device along the path… The operational status information of entity nodes is semantically synthesized to generate disturbance propagation interpretation semantics that include descriptions of propagation direction and bifurcation location. This elevates the abnormal propagation process from the numerical level to an interpretable, structured semantic level, providing a complete chain of propagation evidence for root cause determination. Finally, the root cause matching logic tree is invoked to match the disturbance propagation interpretation semantics. The matched sequence of operational primitives is converted into control commands that directly act on the target control system sites. This achieves closed-loop control throughout the entire process, from anomaly identification, propagation path tracing, semantic interpretation generation to control command output, effectively improving the accuracy and interpretability of emission anomaly response during production control.

[0176] This invention also provides a production control system: a processor; a storage device storing a computer program thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor implements any of the aforementioned production control methods for online environmental monitoring.

[0177] Based on the above, a readable storage medium is provided, on which a program or instructions are stored, and when the program or instructions are executed by a processor, the steps of the above method are implemented.

[0178] Furthermore, it should be noted that this embodiment of the invention also provides a computer program product, which may include a computer program that can be stored in a computer-readable storage medium. The processor of the production control system reads the computer program from the computer-readable storage medium, and the processor can execute the computer program, causing the production control system to perform the aforementioned... Figure 1 The methods described in the corresponding embodiments are already known, and therefore will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer program product embodiments related to this invention, please refer to the description of the method embodiments of this invention.

[0179] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

Claims

1. A production control method applied to online environmental monitoring, characterized in that, The method includes: Based on the process association information between equipment entity nodes in the production process, a set of directed transit edges from upstream equipment entity nodes to downstream equipment entity nodes is generated, and a process causal topology graph is generated based on the equipment entity nodes and the set of directed transit edges. The emission index data stream collected by environmental monitoring sensors is processed for abnormal fluctuation pattern recognition, an abnormal trigger signal is generated, and the abnormal trigger signal is marked as a causal injection event on the corresponding equipment entity node in the process causal topology map. In the process causal topology graph, starting from the marked device entity node, the link reverse propagation process is performed along the reverse direction of the directed transmission edge, and the edge constraint relationship verification process is performed on each upstream device entity node to obtain the causal fork identification node. Based on the causal bifurcation identification node, the disturbance propagation path is traced back, the operating status information of each device entity node on the disturbance propagation path is extracted and semantic synthesis processing is performed to generate disturbance propagation interpretation semantics. The disturbance propagation interpretation semantics includes a description of the propagation direction of the abnormal disturbance in the process causal topology map and a description of the key bifurcation position. The root cause matching logic tree is invoked to perform root cause logic matching processing on the disturbance propagation interpretation semantics to obtain the corresponding operation primitive sequence, and the operation primitive sequence is converted into control instructions and written into the target control system position to regulate the production equipment.

2. The method according to claim 1, characterized in that, The step of generating a set of directed transitive edges from upstream equipment entity nodes to downstream equipment entity nodes based on the process association information between equipment entity nodes in the production process, and generating a process causal topology graph based on the equipment entity nodes and the set of directed transitive edges, includes: Obtain the process design document for the production process, and extract the equipment identification list and material flow path description information from the process design document; Based on the device identifier list, generate one-to-one corresponding device entity nodes and assign each device entity node a unique device identifier; The material flow path description information is parsed to determine the flow direction of materials from upstream equipment physical nodes to downstream equipment physical nodes; For each group of equipment entity nodes with direct material flow relationship, create a directed transfer edge from the upstream equipment entity node to the downstream equipment entity node, and gather all directed transfer edges into a directed transfer edge set. Using the device entity nodes as topology nodes and the directed transit edges in the set of directed transit edges as topology directed edges, an initial process causal topology graph is generated. The implicit process-related edges of the initial process causal topology graph are identified and completed to obtain the process causal topology graph. The process causal topology map is persistently stored in the causal topology storage unit, and a corresponding causal index table is established for each device entity node.

3. The method according to claim 2, characterized in that, The process of identifying and completing the implicit process-related edges in the initial process causal topology graph to obtain the process causal topology graph includes: The energy consumption time-series characteristics of each equipment entity node are extracted from the production process execution system. The energy consumption time-series characteristics include the sequence of changes in the energy medium consumption per unit time of the equipment entity node in multiple production batches. Material throughput timing features of each equipment entity node are extracted from the production process execution system. The material throughput timing features include the change sequence of material processing volume per unit time of the equipment entity node in the multiple production batches. The energy consumption time series characteristics and the material throughput time series characteristics are respectively subjected to standardized fluctuation elimination processing to obtain detrended energy fluctuation series and detrended material fluctuation series; For any first device entity node and second device entity node in the initial process causal topology graph that are not directly connected by a directed transit edge, calculate the energy fluctuation coherence degree between the detrended energy fluctuation sequence of the first device entity node and the detrended energy fluctuation sequence of the second device entity node. Calculate the material fluctuation coherence degree between the detrended material fluctuation sequence of the first equipment entity node and the detrended material fluctuation sequence of the second equipment entity node; The energy fluctuation synergy degree and the material fluctuation synergy degree are weighted and fused to generate the energy-material synergy causal strength; When the energy material synergistic causal strength exceeds a preset synergistic significance threshold, time lag mutual information analysis is performed on the detrended energy fluctuation sequence of the first equipment entity node and the detrended energy fluctuation sequence of the second equipment entity node to determine the energy conduction lag direction. Time lag mutual information analysis is performed on the detrended material fluctuation sequence of the first equipment entity node and the detrended material fluctuation sequence of the second equipment entity node to determine the direction of material transmission lag. If the energy conduction lag direction is consistent with the material conduction lag direction, an implicit process association edge is generated from the equipment entity node corresponding to the leading fluctuation to the equipment entity node corresponding to the lagging fluctuation, and the implicit process association edge is added to the initial process causal topology graph to obtain the process causal topology graph.

4. The method according to claim 1, characterized in that, The process of identifying and processing abnormal fluctuation patterns in the emission index data stream collected by environmental monitoring sensors to generate anomaly trigger signals, and marking these anomaly trigger signals as causal injection events on the corresponding equipment entity nodes in the process causal topology map, includes: Receive emission index data streams uploaded by environmental monitoring sensors in time series format, wherein the emission index data streams contain multiple emission monitoring value points arranged in time sequence; The emission index data stream is subjected to time-series segmentation processing, and multiple continuous and partially overlapping emission index data segments are extracted from the emission index data stream according to a preset observation window duration. For each emission index data segment, a fluctuation baseline extraction process is performed to obtain the fluctuation baseline shape that represents the normal change trend of the emission index within the segment. Based on the deviation relationship between the fluctuation baseline shape and the actual emission monitoring value points within the emission index data segment, abnormal fluctuation profiles with deviations exceeding the allowable deviation boundary are extracted. The abnormal fluctuation profile is matched with multiple emission anomaly patterns in a pre-set emission anomaly pattern library to calculate the morphological difference between the abnormal fluctuation profile and each emission anomaly pattern. The emission anomaly pattern with the smallest morphological difference is selected as the matching emission anomaly pattern, and the anomaly category label and anomaly triggering condition corresponding to the matching emission anomaly pattern are extracted. When the abnormal fluctuation profile meets the abnormal triggering condition, an abnormal triggering signal is generated that includes the abnormal category label, the abnormal start timestamp, and the abnormal duration parameter. Based on the anomaly category label in the anomaly trigger signal, locate the equipment entity node in the process causal topology map that has a predefined monitoring association with the anomaly category label; The abnormal trigger signal is encapsulated as a causal injection event, the causal injection event is marked on the located device entity node, and the link reverse propagation status identifier is activated on the marked device entity node.

5. The method according to claim 1, characterized in that, In the process causal topology graph, starting from the marked device entity node, the link reverse propagation process is performed along the reverse direction of the directed propagation edge. Edge constraint relationship verification is performed on each upstream device entity node to obtain the causal branch identification node, including: Obtain all device entity nodes in the link reverse propagation state in the process causal topology map as the set of reverse inference starting nodes; Starting from each of the reverse reasoning starting nodes in the set of reverse reasoning starting nodes, perform an upstream traversal in the opposite direction of the directed propagation edge pointing to that reverse reasoning starting node to obtain multiple reverse reasoning paths. For each reverse reasoning path, extract the directed transit edges between two directly connected device entity nodes on the reverse reasoning path and read the pre-configured edge constraint relationship definition. The edge constraint relationship is used to determine whether the extracted directed transit edge satisfies the one-way transit constraint condition of the disturbance. If it does not satisfy the constraint condition, the directed transit edge is marked as a constraint violation edge. During the traversal, the upstream device entity node connected to the first constraint violation edge on each reverse reasoning path is recorded as a candidate fork node. All candidate fork nodes generated by the reverse reasoning paths are summarized, and duplicate candidate fork nodes are merged and deduplicated to obtain a set of candidate fork nodes. For each candidate fork node in the candidate fork node set, the corresponding causal index table is retrieved from the causal topology storage unit based on the device identifier of the candidate fork node, and the number of downstream directed transmission edges associated with the candidate fork node is extracted. Candidate fork nodes with a number of downstream directed transmission edges greater than the single path determination threshold are identified as causal fork identification nodes, and fork identification marks are added to the causal fork identification nodes in the process causal topology map.

6. The method according to claim 1, characterized in that, The process involves tracing the disturbance propagation path based on the causal fork identifier node, extracting the operational status information of each device entity node along the disturbance propagation path, and performing semantic synthesis processing to generate disturbance propagation interpretation semantics, including: Using the causal fork identifier node as the backtracking endpoint and the marked device entity node as the backtracking starting point, the complete node link from the backtracking starting point to the backtracking endpoint is extracted along the positive direction of the directed transmission edge in the process causal topology graph, as the disturbance propagation path; Extract the device identifier corresponding to each device entity node on the disturbance propagation path, and obtain the running status record of each device entity node during the period covered by the abnormal duration parameter from the production process execution system based on the device identifier. The acquired operating status records are processed by extracting status parameters to obtain the operating parameter deviation vector of each device entity node. The operating parameter deviation vector is used to characterize the offset direction and degree of the device entity node relative to its standard operating state. The operating parameter deviation vectors of all device entity nodes on the disturbance propagation path are spliced ​​together in node order to obtain a disturbance propagation deviation vector sequence. The disturbance propagation deviation vector sequence is subjected to time-series causal correlation analysis to identify whether there is consistency in the causal propagation direction between the operating parameter deviation vectors of adjacent device entity nodes; Based on the consistency identification results of the causal propagation direction, a perturbation propagation direction description is generated from the starting node to the branching node. The perturbation propagation direction description includes semantic annotations of propagation continuity segments and propagation breakpoint segments. Extract the local adjacent substructure of the causal fork identifier node in the process causal topology graph, determine the composition information of each downstream branch in the local adjacent substructure, and obtain the description of the key fork position; The perturbation propagation direction description and the key bifurcation location description are semantically merged, and the perturbation propagation interpretation semantics are generated by combining the perturbation propagation direction description and the key bifurcation location description. The perturbation propagation interpretation semantics include the perturbation source direction, the propagation path coherence assessment, and the downstream impact branch overview of the bifurcation.

7. The method according to claim 1, characterized in that, The step of calling the root cause matching logic tree to perform root cause logic matching processing on the disturbance propagation interpretation semantics to obtain the corresponding operation primitive sequence, and converting the operation primitive sequence into control instructions and writing them into the target control system position to regulate the production equipment, includes: Load a preset root cause matching logic tree, wherein the root node of the root cause matching logic tree corresponds to the emission anomaly category, the branch nodes correspond to the transmission path feature items in the disturbance transmission interpretation semantics, and the leaf nodes correspond to the root cause determination conclusion. The source of disturbance in the disturbance propagation interpretation semantics is aligned with the root node of the root cause matching logic tree to locate the corresponding starting matching branch. Under the initial matching branch, extract the assessment of the continuity of the transmission path and the overview of the downstream influence of the bifurcation in the semantic interpretation of the perturbation transmission to obtain the semantic feature set; The semantic feature set is used to perform a layer-by-layer matching process with the propagation path feature items on the branch nodes of the root cause matching logic tree. For each matched node, the process moves down along the corresponding logical edge until a leaf node is reached. When a leaf node is reached, the root cause determination conclusion stored in the leaf node is extracted, and the logical edge sequence reaching the leaf node is recorded as a matching path; the corresponding set of operation primitives is indexed from the preset root cause response strategy mapping table according to the root cause determination conclusion; and the operation primitive sequence with the highest execution priority is selected from the set of operation primitives according to the propagation fork severity factor reflected in the matching path. Each operation primitive in the selected sequence of operation primitives is converted into a control instruction that is compatible with the target control system position communication protocol. The abnormal start timestamp and the device identifier of the marked device entity node are embedded in the control instruction. The generated control commands are sequentially written into the corresponding target control system positions according to the execution order of the operation primitive sequence, so as to adjust the operating parameters of the relevant equipment entity nodes in the production process.

8. The method according to claim 7, characterized in that, The step of performing a layer-by-layer matching process using the semantic feature set and the propagation path feature items on the branch nodes of the root cause matching logic tree, where each matched node is migrated downwards along the corresponding logical edge until a leaf node is reached, includes: Obtain the starting matching branch node located by the category alignment processing and set it as the current level node; construct the migrated node identifier sequence; add the node identifier of the current level node to the migrated node identifier sequence; and read the transmission path feature item stored in the current level node. The coherence feature is extracted from the conduction path coherence assessment contained in the semantic feature set to obtain the cumulative proportion of conduction coherence segments and the variance of conduction breakpoint segments in the conduction path coherence assessment, and the cumulative proportion of conduction coherence segments and the variance of conduction breakpoint segments are combined into a coherence feature vector. For the overview of the downstream influence branches contained in the semantic feature set, the fork feature is extracted to obtain the total number of downstream branches and the type distribution vector of the device entity node pointed to by each downstream branch. The total number of downstream branches and the type distribution vector are then concatenated to generate the fork structure feature vector. Calculate the interval boundary offset between the coherence feature vector and the coherence reference ratio interval in the conduction coherence constraint, and the over-limit ratio of the conduction breakpoint segment interval variance in the coherence feature vector relative to the allowable upper limit of the breakpoint interval variance. Generate a coherence matching metric based on the interval boundary offset and the over-limit ratio. Calculate the offset between the total number of downstream branches in the bifurcation structure feature vector and the number interval of the branch number reference interval, and calculate the Mahalanobis distance between the type distribution vector in the bifurcation structure feature vector and the device type distribution tolerance deviation matrix. Generate a bifurcation structure matching metric based on the number interval offset and the Mahalanobis distance. Extract the coherence contribution weight and bifurcation structure contribution weight associated with the current level node from the root cause matching logic tree. Based on the coherence contribution weight and bifurcation structure contribution weight, weight the coherence matching metric and the bifurcation structure matching metric to obtain the node comprehensive matching degree. The matching migration threshold is calculated based on the depth of the current level node in the root cause matching logic tree. The overall matching degree of the node and the matching migration threshold are compared. When the overall matching degree of the node is greater than or equal to the matching migration threshold, the current level node is identified as a matching passed node, and the direct child node is obtained along the logical edge derived from the matching passed node. The direct child node is determined as the current node of the next level, and the node identifier of the direct child node is recorded in the migrated node identifier sequence. When it is determined that the node type of the direct child node does not belong to the leaf node type, the matching process from reading the transmission path feature item to calculating the overall matching degree of the node is repeated with the direct child node as the current level node. When the overall matching degree of the node is less than the matching migration threshold, the matching migration along the current logical edge is stopped, and a matching termination indication is output. Before repeatedly executing the matching process with the direct child node as the current level node, check whether the node identifier of the direct child node already exists in the migrated node identifier sequence. If it already exists, trigger the logical edge loop termination operation and stop the layer-by-layer matching process. When it is determined that the node type of the direct child node belongs to the leaf node type, the layer-by-layer matching process is terminated, and the direct child node is recorded as the leaf node reached by the layer-by-layer matching.

9. A production control system, characterized in that, include: A processor; a storage device having a computer program stored thereon; a network interface for providing network communication functions; when the computer program is executed by the processor, the processor enables the processor to implement the production control method for online environmental monitoring as described in any one of claims 1-8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions, which, when executed by a processor, implement the production control method for online environmental monitoring as described in any one of claims 1-8.

Images