A knowledge graph-based intelligent control method and system for wastewater treatment plants

By constructing a wastewater treatment process knowledge graph based on the STOMP algorithm and an improved AMNet model, the shortcomings of existing technologies in identifying abnormal process objects and control relationships in wastewater treatment plants are addressed. This enables the clear identification of abnormal propagation paths and the traceability of control commands, thereby improving the operational safety and accuracy of control strategies in wastewater treatment.

CN123117532APending Publication Date: 2026-11-17ANHUI GREEN TITAN ENVIRONMENTAL TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611288558.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The control systems of existing wastewater treatment plants struggle to promptly identify the location and path of anomalies when there are sudden changes in influent water quality, load fluctuations, or a decline in equipment efficiency. This leads to delayed adjustments, inaccurate selection of controlled objects, and waste of reagents and energy. Furthermore, they lack a continuous expression of the relationship between operating segments, process sequences, and control responses.

Method used

Using the STOMP algorithm and an improved AMNet model, a wastewater treatment process knowledge graph is constructed, generating a set of evidence anchor points for wastewater treatment segments, a stable operating condition subgraph, a disturbance operating condition subgraph, and a disturbance propagation difference graph. Abnormal process objects and control relationships are identified, and executable control instructions are generated through the control attribution chain.

Benefits of technology

It achieves clear abnormal propagation paths, accurate control attribution, and traceable control commands, thereby improving the operational safety of wastewater treatment and the accuracy of control strategies.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117532A_ABST
    Figure CN123117532A_ABST
Patent Text Reader

Abstract

This invention discloses an intelligent control method and system for wastewater treatment plants based on knowledge graphs, belonging to the field of knowledge graph technology. The method includes: collecting industrial operation data from the wastewater treatment plant to obtain a standardized industrial operation time series; executing the STOMP algorithm to obtain a set of evidence anchor points for wastewater treatment segments; constructing a wastewater treatment process knowledge graph, extracting stable operating condition subgraphs and disturbed operating condition subgraphs; constructing an improved AMNet model to generate a disturbance propagation difference graph; performing process causal consistency verification to obtain an effective disturbance propagation subgraph; constructing a control attribution chain based on the effective disturbance propagation subgraph; and determining the wastewater treatment control strategy to obtain executable control commands. This invention, by introducing the STOMP algorithm and an improved AMNet model, achieves the identification of abnormal disturbance propagation paths, control attribution analysis, and generation of executable control commands in wastewater treatment plants.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to an intelligent control method and system for wastewater treatment plants based on knowledge graphs. Background Technology

[0002] Wastewater treatment plants primarily remove pollutants through screens, equalization tanks, biological reaction tanks, sedimentation tanks, disinfection tanks, and supporting aeration, reflux, chemical dosing, and sludge removal equipment. Current operation and management typically rely on online instruments to collect data on influent water quality, flow rate, dissolved oxygen, sludge concentration, and effluent parameters. Equipment operation is adjusted through manual inspections, fixed threshold alarms, empirical rules, or conventional automatic control methods. While this approach can meet the stable operation requirements under normal conditions, it often only detects effluent exceeding standards when there are sudden changes in influent water quality, load fluctuations, decreased equipment efficiency, or deviations in process status. It struggles to promptly determine the location and path of the anomaly, easily leading to delayed adjustments, inaccurate selection of controlled objects, and waste of chemicals and energy.

[0003] With the development of industrial data analysis, intelligent control, and graph structure modeling technologies, some wastewater treatment systems have begun to utilize historical operating data to establish relationships between process objects and use data analysis models to assist in judging water quality changes, equipment anomalies, and control effectiveness. Current technologies typically analyze treatment sections, equipment, and pollutant indicators as independent monitoring objects, or trigger control strategies based on single alarm results. This lack of expression of the continuous relationships between operating segments, process sequences, control responses, and effluent changes leads to unclear correspondence between anomaly causes and control actions, and the traceability and safety verification capabilities of control strategies still have room for improvement.

[0004] Therefore, how to provide a knowledge graph-based intelligent control method and system for wastewater treatment plants is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent control method and system for wastewater treatment plants based on knowledge graphs. This invention utilizes the STOMP algorithm, a wastewater treatment process knowledge graph, and an improved AMNet model to describe in detail the process of obtaining standardized industrial operation time series from industrial operation data of wastewater treatment plants. It further generates a set of evidence anchor points for wastewater treatment segments, a stable operating condition subgraph, a disturbance operating condition subgraph, a disturbance propagation difference graph, an effective disturbance propagation subgraph, and a control attribution chain. This enables the identification of associations between abnormal process objects, abnormal control relationships, and abnormal effluent results in wastewater treatment. Based on the control attribution chain, safety constraint verification is performed to obtain executable control commands. This method possesses advantages such as clear abnormal propagation paths, accurate control attribution, traceable control commands, and high operational safety in wastewater treatment.

[0006] According to an embodiment of the present invention, a knowledge graph-based intelligent control method for wastewater treatment plants includes: Collect industrial operation data from wastewater treatment plants, perform preprocessing on the wastewater treatment plant industrial operation data, and obtain standardized industrial operation time series; Process phase constraints, operation segment binding, and control response correlation are performed on standardized industrial operation time series. The STOMP algorithm is used to calculate the matrix contour distance between operation segments and perform segment similarity retrieval to obtain the set of evidence anchor points for wastewater treatment segments. Based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, a knowledge graph of wastewater treatment processes is constructed, and subgraphs of steady-state and disturbed states are extracted. Based on the stable operating condition subgraph and the disturbance operating condition subgraph, an improved AMNet model is constructed, and multi-frequency relationship extraction, subgraph difference processing and process causal conflict resolution are performed to generate a disturbance propagation difference map. Based on the process flow direction, pollutant transformation relationship, equipment adjustment object and segment occurrence time, the disturbance propagation difference diagram is subjected to process causal consistency verification to obtain an effective disturbance propagation sub-diagram; Identify anomalous process objects, anomalous control relationships, and anomalous effluent results in the effective disturbance propagation subgraph, and construct a control attribution chain. Based on the control attribution chain, the wastewater treatment control strategy is determined, the safety constraint verification is performed, and the executable control instructions are obtained.

[0007] Optionally, the industrial operation data of the wastewater treatment plant specifically includes influent water quality data, influent flow rate data, process section status data, equipment operation data, control action data, effluent index data, and collection timestamp data.

[0008] Optionally, the preprocessing of industrial operation data of wastewater treatment plants specifically includes data format standardization, timestamp correction, abnormal data removal, missing data completion, sampling interval standardization, unit standardization, and time series sorting.

[0009] Optionally, the obtained set of wastewater treatment fragment evidence anchor points includes: Based on standardized industrial operation time series, process phase constraints are implemented to divide the current operation segment, historical stable operation segment, historical abnormal operation segment, historical effective control segment, and historical inefficient control segment. Each operation segment is then bound to the process object, segment time, and process phase to obtain a set of phase-constrained segments. Read the current running segment, historical stable running segment, and historical abnormal running segment from the phase constraint segment set. Perform stable action replacement and reverse action replacement on the control action change segment to obtain the control ablation segment and the control reverse enhancement segment. Associate the control action change segment with the effluent response segment and use the STOMP algorithm to calculate the matrix profile distance to obtain the three-state evidence data of the control action. Identify abnormal effluent results and candidate abnormal cause objects in the current operation segment, retrieve historical operation segments with the same influent load change direction, the same abnormal effluent results, and no abnormal changes in candidate abnormal cause objects, and obtain abnormal propagation counter-evidence data; Based on the current operating segment, historical stable operating segments, historical abnormal operating segments, three-state evidence data of control actions, and abnormal propagation counter-evidence data, a candidate disturbance propagation chain is constructed. The process objects and equipment objects in the candidate disturbance propagation chain are replaced with stable segments of the same phase, and the changes in similar distance and the connectivity state of the propagation chain before and after the replacement are calculated to obtain propagation cut evidence data. Historical effective control segments and historical inefficient control segments are extracted from the phase constraint segment set. The STOMP algorithm is used to perform invertibility matrix contour retrieval to obtain control invertibility evidence data. Combined with control action three-state evidence data, anomaly propagation counter-evidence data, propagation cut evidence data and segment identification data, a wastewater treatment segment evidence anchor point set is obtained.

[0010] Optionally, the extraction of the steady-state subgraph and the disturbance subgraph includes: Based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, process object nodes and segment evidence anchor point nodes are extracted, and process association edges are established according to process flow direction, pollutant transformation, equipment adjustment and control response to obtain a wastewater treatment process knowledge graph. The evidence data and fragment identification data in the fragment evidence anchor node are mapped to the corresponding process object node and process association edge to obtain the evidence mapping graph. Extract the corresponding nodes and process-related edges of historical stable operation segments and historical effective control segments from the evidence mapping graph to form a stable operating condition subgraph. Extract the corresponding nodes and process-related edges of the current operating segment, historical abnormal operating segment, and historical inefficient control segment to form a disturbance operating condition subgraph.

[0011] Optionally, generating the perturbation propagation difference map includes: An improved AMNet model is constructed, which includes a dual-condition evidence input layer, an evidence modulation multi-frequency extraction layer, a difference constraint processing layer, and a difference map output layer. The dual-condition evidence input layer is based on the stable condition subgraph, the disturbance condition subgraph, and the wastewater treatment segment evidence anchor point set. It performs evidence polarity recoding on the evidence data to obtain support disturbance identifiers, counter-evidence exclusion identifiers, propagation blocking identifiers, and reversible control identifiers, and writes them into the corresponding nodes and process-related edges to obtain the dual-condition evidence input graph. The evidence modulation multi-frequency extraction layer is based on the dual-condition evidence input graph. It performs evidence polarity frequency band routing according to the support disturbance flag, the rebuttal evidence exclusion flag, the propagation blocking flag, and the reversible control flag. It extracts low-frequency stable relations, high-frequency disturbance relations, and propagation cut sensitivity relations to obtain the dual-condition multi-frequency relation representation. The differential constraint processing layer is based on the dual-condition multi-frequency relationship representation. It performs mirror differential processing on nodes with consistent object identification and process-related edges with consistent connected objects. It also performs process causal conflict resolution according to process flow direction, pollutant conversion relationship, equipment adjustment object, fragment occurrence time, counter-evidence exclusion mark and propagation blocking mark, and filters the dominant disturbance path to obtain node difference data, edge difference data and effective propagation difference data. The difference graph output layer is based on node difference data, edge difference data, and effective propagation difference data. It retains the effective abnormal propagation paths that connect abnormal water discharge results, propagation cut objects, and reversible control objects, and deletes abnormal propagation paths with counter-evidence exclusion markers to obtain the disturbance propagation difference graph. The improved AMNet model was trained using perturbation propagation difference map reconstruction error, anomaly propagation path identification error, and control attribution chain matching error as joint optimization objectives. The parameters of the dual-condition evidence input layer, evidence modulation multi-frequency extraction layer, differential constraint processing layer, and difference map output layer were continuously optimized. When the decrease of the joint optimization objective in 5 consecutive training rounds was less than 0.001, or when the number of training rounds reached 200, training was stopped and the improved AMNet model was obtained.

[0012] Optionally, obtaining the effective perturbation propagation subgraph includes: Based on the node difference data, edge difference data, effective propagation difference data and segment occurrence time in the disturbance propagation difference graph, candidate disturbance paths are extracted to connect abnormal water discharge results, abnormal process objects, abnormal control relationships and equipment adjustment objects. Based on the process flow direction, pollutant transformation relationship, equipment regulation object and segment occurrence time, perform path consistency verification on candidate disturbance paths, delete candidate disturbance paths that do not conform to the process sequence, transformation relationship, regulation object relationship and time sequence, and obtain causally consistent disturbance paths; Extract nodes, process-related edges, and effective propagation difference data from the causal consistent perturbation path to form an effective perturbation propagation subgraph.

[0013] Optionally, constructing the control attribution chain includes: Based on the nodes, process-related edges, effective propagation difference data, and segment identification data in the effective disturbance propagation subgraph, abnormal process objects, abnormal control relationships, and abnormal effluent results are identified according to the node object type. Based on process flow direction, pollutant transformation relationship, equipment adjustment object and segment occurrence time, abnormal process object, abnormal control relationship and abnormal effluent result are sorted into paths to obtain candidate control attribution paths; Based on the effective propagation difference data and process-related edge connection order in the candidate control attribution paths, candidate control attribution paths that continuously connect abnormal process objects, abnormal control relationships and abnormal effluent results are selected, and control attribution chains are constructed.

[0014] Optionally, obtaining executable control instructions includes: Based on the abnormal process objects, abnormal control relationships, and abnormal effluent results in the control attribution chain, candidate control strategies for wastewater treatment are determined according to the equipment adjustment objects corresponding to the abnormal control relationships. Based on the equipment operating boundary, process operating boundary, and effluent discharge index limit, the safety constraint verification is performed on the candidate control strategy for wastewater treatment. Candidate control strategies that exceed the equipment operating boundary, process operating boundary, or effluent discharge index limit are deleted to obtain the safety control strategy. Based on safety control strategies and equipment operation data, the safety control strategies are converted into corresponding equipment adjustment parameters to obtain executable control commands.

[0015] An intelligent control system for a wastewater treatment plant based on a knowledge graph, according to an embodiment of the present invention, includes the following modules: The industrial time series management module is used to collect industrial operation data from wastewater treatment plants and perform preprocessing to obtain standardized industrial operation time series. The phase anchor point parsing module is used to process standardized industrial operation time series using the STOMP algorithm to obtain a set of evidence anchor points for wastewater treatment segments. The process graph twin module is used to construct a wastewater treatment process knowledge graph based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, and to extract the steady-state subgraph and the disturbed state subgraph. The perturbation difference inference module is used to process the stable operating condition subgraph and the perturbation operating condition subgraph by improving the AMNet model, and generate a perturbation propagation difference graph. The causality consistency verification module is used to perform process causality consistency verification on the disturbance propagation difference graph to obtain a valid disturbance propagation subgraph; The control attribution and tracing module is used to identify abnormal process objects, abnormal control relationships and abnormal effluent results in the effective disturbance propagation subgraph, and to construct the control attribution chain; The safety instruction decision module is used to determine the wastewater treatment control strategy based on the control attribution chain, and to perform safety constraint verification to obtain executable control instructions.

[0016] The beneficial effects of this invention are: This invention discloses a knowledge graph-based intelligent control method and system for wastewater treatment plants. By collecting industrial operation data from the wastewater treatment plant and obtaining standardized industrial operation time series, it enables influent water quality, influent flow rate, process stage status, equipment operation, control actions, and effluent indicators to be analyzed on the same time scale, reducing the impact of time inconsistencies from different data sources on anomaly detection. The invention also utilizes the STOMP algorithm to generate a set of evidence anchor points for wastewater treatment segments, associating current operation segments, historical stable operation segments, historical abnormal operation segments, historical effective control segments, and historical inefficient control segments with similarity, allowing for a quantifiable expression of the relationship between changes in control actions and effluent response.

[0017] This invention uses a wastewater treatment process knowledge graph to represent the process relationships between treatment sections, equipment, pollutant indicators, and control actions. By improving the AMNet model, it identifies the differences in disturbance propagation between stable and disturbed operating condition subgraphs. This allows for the screening of abnormal process objects, abnormal control relationships, and abnormal effluent results from the graph structure, reducing false alarms caused by relying solely on single threshold alarms. Through process causal consistency verification and control attribution chain construction, this invention can form traceable anomaly propagation paths and perform safety constraint checks before generating executable control commands, thereby improving the accuracy, safety, and interpretability of wastewater treatment control strategies. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a knowledge graph-based intelligent control method for wastewater treatment plants proposed in this invention. Figure 2 This is a schematic diagram of the structure of an improved AMNet model for a knowledge graph-based intelligent control method for wastewater treatment plants proposed in this invention. Figure 3 This is a schematic diagram of the structure of an intelligent control system for a wastewater treatment plant based on a knowledge graph, as proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 and Figure 2 A knowledge graph-based intelligent control system for wastewater treatment plants includes: Collect industrial operation data from wastewater treatment plants, perform preprocessing on the wastewater treatment plant industrial operation data, and obtain standardized industrial operation time series; Process phase constraints, operation segment binding, and control response correlation are performed on standardized industrial operation time series. The STOMP algorithm is used to calculate the matrix contour distance between operation segments and perform segment similarity retrieval to obtain the set of evidence anchor points for wastewater treatment segments. Based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, a knowledge graph of wastewater treatment processes is constructed, and subgraphs of steady-state and disturbed states are extracted. Based on the stable operating condition subgraph and the disturbance operating condition subgraph, an improved AMNet model is constructed, and multi-frequency relationship extraction, subgraph difference processing and process causal conflict resolution are performed to generate a disturbance propagation difference map. Based on the process flow direction, pollutant transformation relationship, equipment adjustment object and segment occurrence time, the disturbance propagation difference diagram is subjected to process causal consistency verification to obtain an effective disturbance propagation sub-diagram; Identify anomalous process objects, anomalous control relationships, and anomalous effluent results in the effective disturbance propagation subgraph, and construct a control attribution chain. Based on the control attribution chain, the wastewater treatment control strategy is determined, the safety constraint verification is performed, and the executable control instructions are obtained.

[0021] In this embodiment, the industrial operation data of the wastewater treatment plant specifically includes influent water quality data, influent flow rate data, process section status data, equipment operation data, control action data, effluent index data, and data collection timestamp data, wherein: Process status data refers to the data reflecting the operating status of each treatment section in the wastewater treatment process. Specifically, it includes influent level data, dissolved oxygen data, sludge concentration data, pH data, sludge interface height data, turbidity data, and effluent water quality stability data. Control action data refers to the data generated by the control equipment of a wastewater treatment plant during operation when it performs adjustment actions. Specifically, it includes aeration volume adjustment data of aeration equipment, return ratio adjustment data of return equipment, dosing volume adjustment data of dosing equipment, sludge discharge time adjustment data of sludge discharge equipment, and corresponding control action execution status data.

[0022] In this embodiment, the preprocessing of industrial operation data of wastewater treatment plants specifically includes data format unification, timestamp correction, abnormal data removal, missing data completion, sampling interval unification, unit unification, and time series sorting.

[0023] In this embodiment, obtaining the set of evidence anchor points for wastewater treatment segments includes: Based on standardized industrial operation time series, process phase constraints are applied, dividing the operation into current operation segments, historical stable operation segments, historical abnormal operation segments, historical effective control segments, and historical inefficient control segments. Each operation segment is then bound to a process object, segment time, and process phase, resulting in a set of phase-constrained segments, where: Process phase constraint refers to writing corresponding process phase identifiers for operating segments according to the treatment sequence of wastewater from the influent section, biochemical reaction section, sedimentation and separation section to the effluent section, and restricting operating segments from participating in segment division and segment retrieval within the same process phase or adjacent process phases; The process phase constraint is executed as follows: Read standardized industrial operation time series data, sort them from earliest to latest according to the collected timestamp data, and read the effective sampling time intervals of online instruments in the wastewater treatment plant within 30 consecutive operating days. Remove time intervals caused by communication interruptions, duplicate uploads, and null value uploads. Use the time interval with the most frequent occurrences and a data missing rate of less than 2% as the unified sampling interval. Read the response time of the effluent indicators starting to decline continuously after the execution of control actions in historical effective control segments, and use the median of the response time as the operation segment coverage duration. Ensure that the operation segment coverage duration is not less than 30 minutes and not more than 1 minute. For every 20 minutes, the duration of each running segment is divided by the uniform sampling interval and the nearest integer is taken to obtain the number of sampling points in each running segment. The number of sampling points in each running segment is not less than 6. The process segment status data in each running segment is read, and the occurrence frequency of the influent segment, biological reaction segment, sedimentation separation segment and effluent segment in the running segment is counted. The process segment status with an occurrence frequency of not less than 80% of the number of sampling points in the running segment is taken as the main process segment of the segment. When the occurrence frequency of all four types of process segment status is less than 80%, the running segment is removed. When the main process section of a segment is the inlet section, an inlet phase identifier is written for the running segment; when the main process section of a segment is the biochemical reaction section, a biochemical reaction phase identifier is written for the running segment; when the main process section of a segment is the sedimentation and separation section, a sedimentation and separation phase identifier is written for the running segment; when the main process section of a segment is the effluent section, an effluent phase identifier is written for the running segment. The running segment whose timestamp data is closest to the current control time is designated as the current running segment. For operational segments where effluent indicators did not exceed the effluent discharge limits for 30 consecutive operating days, the intra-segment fluctuation coefficients for Chemical Oxygen Demand (COD), ammonia nitrogen (ANH), and total phosphorus (TP) were calculated. The intra-segment fluctuation coefficient was calculated as the standard deviation of the same effluent indicator across all sampling points within the operational segment, divided by the average. The 95th percentile of the intra-segment fluctuation coefficient was used as the stable fluctuation threshold. When the stable fluctuation threshold was below 0.03, it was set to 0.03; when it was above 0.08, it was set to 0.08. Operational segments where COD was no greater than 50 mg / L, ANH no greater than 5 mg / L, and TP no greater than 0.5 mg / L at all sampling points, and where the intra-segment fluctuation coefficients for COD, ANH, and TP were all below the stable fluctuation threshold, were classified as historically stable operational segments. Operating segments with chemical oxygen demand greater than 50 mg / L, ammonia nitrogen greater than 5 mg / L, or total phosphorus greater than 0.5 mg / L are classified as historical abnormal operating segments. Historical control action records and effluent index recovery records are read. The median of the decrease in effluent index within three consecutive operating segments after the execution of the control action in the confirmed effective control records is taken as the effective improvement threshold. When the effective improvement threshold is less than 10%, the effective improvement threshold is set to 10%. Operating segments with a decrease in effluent index of not less than the effective improvement threshold within three consecutive operating segments after the change in control action data are classified as historical effective control segments. The effective improvement threshold is multiplied by 0.3 to obtain the inefficient improvement threshold. Operating segments with a decrease in effluent index of less than the inefficient improvement threshold within three consecutive operating segments after the change in control action data are classified as historical inefficient control segments. The process object refers to the corresponding treatment section or regulating equipment in the wastewater treatment plant, including the influent section, biochemical reaction section, sedimentation and separation section, effluent section, aeration equipment, reflux equipment, dosing equipment and sludge discharge equipment. Process phase refers to the positional identifier of an operating segment in the wastewater treatment process, including the influent phase, biochemical reaction phase, sedimentation and separation phase, and effluent phase; The set of phase-constrained segments is obtained as follows: The main process segment of the operating segment is taken as the process object of the treatment segment. The aeration equipment, recirculation equipment, dosing equipment or sludge discharge equipment whose control action data changes within the operating segment are taken as the equipment process objects. The collection timestamp of the first sampling point of the operating segment is taken as the segment start time, and the collection timestamp of the 12th sampling point is taken as the segment end time. The segment time is composed of the segment start time and the segment end time. The current operating segment, historical stable operating segment, historical abnormal operating segment, historical effective control segment and historical inefficient control segment are taken as segment objects. The process object, segment time, process phase identifier, influent water quality data, influent flow rate data, process segment status data, equipment operation data, control action data and effluent index data corresponding to each segment object are composed of segment records. All segment records are composed of a phase-constrained segment set. The current running segment, historical stable running segment, and historical abnormal running segment are read from the phase-constrained segment set. Stable action replacement and reverse action replacement are performed on the control action change segments to obtain control ablation segments and control reverse enhancement segments. These control action change segments are associated with the effluent response segments, and the STOMP algorithm is used to calculate the matrix profile distance to obtain the three-state evidence data of the control action, where: A control action change segment refers to an operational segment in the phase constraint segment set where the aeration volume, reflux ratio, dosage, or sludge discharge time changes significantly. A significant change is defined as a change in the range of any one control action data between two adjacent sampling points of not less than 5%. The range of change is calculated by subtracting the value before the change from the value after the change and then dividing by the value before the change and taking the absolute value. Perform stable action replacement and reverse action replacement on the control action change segment, specifically: Read the process phase identifier, equipment process object, and changed control action data in the control action change segment. Select segments with the same process phase identifier and the same equipment process object from the historical stable operation segments. Sort the same control action data in the selected segments from smallest to largest value, and take the value in the middle of the sort as the stable action value. Copy the control action change segment. From the first sampling point where the control action changes to the end of the segment, change the changed control action data to the stable action value, and keep the unchanged data with the original value to obtain the control ablation segment. Read the values ​​before and after the control action change. When the value after the change is greater than the value before the change, subtract the increase in change from the stable action value to obtain the reverse candidate value. When the value after the change is less than the value before the change, add the decrease in change to the stable action value to obtain the reverse candidate value. Before writing the reverse candidate value, compare the reverse candidate value with the minimum and maximum values ​​of the same control action data in the historical stable operation segment. If the reverse candidate value is lower than the minimum value, take the minimum value; if it is higher than the maximum value, take the maximum value; if it is between the minimum and maximum values, take the reverse candidate value to obtain the reverse action value. Copy the control action change segment. From the first sampling point where the control action change occurs to the end of the segment, change the control action data that has changed to the reverse action value, and keep the original value of the data that has not changed to obtain the control reverse enhancement segment. The effluent response segment refers to the operational segment used to record the changes in chemical oxygen demand (COD), ammonia nitrogen (NH3), and total phosphorus (TP) after the control action change segment ends. When the process phase identifier of the control action change segment is the influent phase, the effluent response segment corresponds to the 36th to 47th sampling points after the end of the segment. When the process phase identifier is the biochemical reaction phase, the effluent response segment corresponds to the 24th to 35th sampling points after the end of the segment. When the process phase identifier is the sedimentation and separation phase, the effluent response segment corresponds to the 12th to 23rd sampling points after the end of the segment. When the process phase identifier is the effluent phase, the effluent response segment corresponds to the 6th to 17th sampling points after the end of the segment. The STOMP algorithm is used to calculate the matrix contour distance, specifically: The control action change segment, control ablation segment, and control reverse enhancement segment are respectively used as the search segments. For each search segment, the process phase identifier and equipment process object are read first. Then, segments with the same process phase identifier and the same equipment process object are selected from the historical stable operation segments as stable search objects. Segments with the same process phase identifier and the same equipment process object are selected from the historical abnormal operation segments as abnormal search objects. To retrieve the fragment, control action data, equipment operation data, chemical oxygen demand, ammonia nitrogen, and total phosphorus from 12 sampling points are read and arranged in chronological order to form a fragment sequence containing 60 standardized values. For each stable retrieval object, the same data is read and arranged into a stable historical fragment sequence. The 60 sets of values ​​at the same position in the fragment sequence and the stable historical fragment sequence are subtracted respectively, and the 60 differences are squared. The sum of the 60 squared values ​​is divided by 60, and the square root of the division result is taken to obtain the distance value between the fragment to be retrieved and one stable retrieval object. The minimum value among all stable retrieval object distances is taken as the matrix contour distance from the fragment to be retrieved to the historical stable operation fragment. The distance value for each abnormal retrieval object is calculated in the same way, and the minimum value among all abnormal retrieval object distances is taken as the matrix contour distance from the fragment to be retrieved to the historical abnormal operation fragment. The three-state evidence data for the control action were obtained, specifically: The control action change segment is taken as the original action state, the control ablation segment is taken as the stable replacement state, and the control reverse enhancement segment is taken as the reverse enhancement state. The matrix contour distance from the three states to the historical stable operation segment, the matrix contour distance to the historical abnormal operation segment, and the change direction of the effluent index in the effluent response segment are recorded respectively. The change direction of the effluent index is determined according to chemical oxygen demand, ammonia nitrogen, and total phosphorus respectively. The average value of the same effluent index in the control action change segment is read as the average value before the change. The average value of the same effluent index in the effluent response segment is read as the average value after the response. When the increase of the average value after the response relative to the average value before the change is not less than 5%, it is recorded as an increase. When the decrease is not less than 5%, it is recorded as a decrease. When both the increase and decrease are less than 5%, it is recorded as stable. When the matrix profile distance from the original action state to the historical abnormal operation segment is less than the matrix profile distance from the stable replacement state to the historical abnormal operation segment, and the direction of change of any one of the effluent indicators is upward, it is recorded as evidence of disturbance control. When the matrix profile distance from the original action state to the historical stable operation segment is less than the matrix profile distance from the reverse enhancement state to the historical stable operation segment, and the direction of change of any one of the effluent indicators is downward, it is recorded as evidence of disturbance suppression control. When the difference in matrix profile distance between the three states is less than 0.02, it is recorded as evidence of invalid control. The evidence of disturbance control, the evidence of disturbance suppression control, and the evidence of invalid control constitute the three-state evidence data of control action. Identify abnormal effluent results and candidate abnormal cause objects in the current operating segment, retrieve historical operating segments with the same influent load change direction, the same abnormal effluent results, and no abnormal changes in candidate abnormal cause objects, and obtain anomaly propagation counter-evidence data, including: Identify abnormal water discharge results and candidate abnormal cause objects in the current running segment, specifically as follows: Read the effluent index data in the current operating segment. When the chemical oxygen demand is greater than 50 mg / L, ammonia nitrogen is greater than 5 mg / L, or total phosphorus is greater than 0.5 mg / L, the effluent index exceeding the corresponding value is recorded as an abnormal effluent result. Read the influent water quality data, influent flow rate data, process section status data, equipment operation data, and control action data in the current operating segment, and read the median of the corresponding data in the historical stable operating segments of the same process phase. When the change of the influent water quality data or influent flow rate data in the current operating segment relative to the median of the corresponding data is not less than 10%, the influent segment is recorded as a candidate abnormal cause object. When the change of the process section status data, equipment operation data, or control action data relative to the median of the corresponding data is not less than 10%, the corresponding treatment segment or equipment is recorded as a candidate abnormal cause object. Historical operating segments refer to operating segments in the phase constraint segment set whose acquisition timestamp is earlier than the current control time, including historical stable operating segments, historical abnormal operating segments, historical effective control segments, and historical inefficient control segments; The data obtained as evidence of anomalous propagation are as follows: Read the influent water quality data and influent flow rate data in the current operating segment. Multiply the chemical oxygen demand, ammonia nitrogen, and total phosphorus in the influent water quality data by the influent flow rate data to obtain the corresponding pollutant influent load. Read the median of the corresponding pollutant influent load in the historical stable operating segments of the same process phase. When the influent load of any pollutant in the current operating segment increases by no less than 5% relative to the corresponding median, the direction of the influent load change is recorded as an increase. When the decrease is no less than 5%, it is recorded as a decrease. When both the increase and decrease are less than 5%, it is recorded as stable. Select segments from the historical operating segments with the same influent load change direction, the same abnormal effluent result, and the corresponding data of the candidate abnormal cause object changes by less than 5% relative to the median of the historical stable operating segments of the same process phase. Combine the selected historical operating segment identifier, the same influent load change direction, the same abnormal effluent result, the record of no abnormal change of the candidate abnormal cause object, and the corresponding matrix contour distance to form the abnormal propagation counter-evidence data. Based on current operating segments, historical stable operating segments, historical abnormal operating segments, three-state evidence data of control actions, and anomaly propagation counter-evidence data, candidate disturbance propagation chains are constructed. In-phase stable segments are replaced for the process objects and equipment objects in the candidate disturbance propagation chains, and the changes in similarity distance and the connectivity state of the propagation chains before and after the replacement are calculated to obtain propagation cut evidence data, including: Constructing the candidate perturbation propagation chain is as follows: Read the abnormal effluent results, candidate abnormal cause objects, three-state evidence data of control actions, and abnormal propagation counter-evidence data from the current operating segment. Treat the abnormal effluent results as the tail of the chain, and the candidate abnormal cause objects and the disturbance-inducing and suppression control action objects from the three-state evidence data of control actions as the front of the chain. Read historical abnormal operating segments with the same abnormal effluent results, and read the median of the corresponding data for the same process object and the same equipment object in historical stable operating segments of the same process phase. When the corresponding data of the process object or equipment object in the historical abnormal operating segment is relative to the median of the corresponding data... When the change is not less than 10%, the process object or equipment object is taken as the disturbance propagation object. The change is calculated by subtracting the median of the corresponding data from the object data in the historical abnormal operation segment, dividing by the median of the corresponding data, and taking the absolute value. In the order of influent phase, biochemical reaction phase, sedimentation separation phase, and effluent phase, and in the order of the segment occurrence time from earliest to latest, the pre-chain object, disturbance propagation object, and abnormal effluent result are connected to form a candidate disturbance propagation chain. When there is a record of no abnormal change in the candidate abnormal cause object in the abnormal propagation counter-evidence data, the chain segment where the candidate abnormal cause object is located is marked as the counter-evidence chain segment. Equipment refers to the equipment in a wastewater treatment plant that participates in the execution of control actions and changes the process operating state, including aeration equipment, reflux equipment, dosing equipment, and sludge removal equipment; Perform in-phase stable segment replacement on the process objects and equipment objects in the candidate disturbance propagation chain, specifically as follows: Read the process objects and equipment objects in the candidate disturbance propagation chain, and read the process phase identifier of the operating segment where each object is located. Select segments with the same process phase identifier and the same object from the historical stable operating segments. Read the process segment status data, equipment operation data and control action data corresponding to the objects in the selected segments. Take the median of the data at the same sampling time position in the 12 sampling points to obtain the same-phase stable replacement data. Copy the candidate disturbance propagation chain according to the number of objects to be replaced. One object to be replaced corresponds to one copied candidate disturbance propagation chain. In the copied candidate disturbance propagation chain, change the data of the object to be replaced to the same-phase stable replacement data. Other process objects, equipment objects and abnormal effluent results keep the original values ​​to obtain the replaced candidate disturbance propagation chain. Calculate the change in similarity distance and the connectivity state of the propagation chain before and after the replacement, specifically as follows: Read the process segment status data, equipment operation data, control action data, chemical oxygen demand, ammonia nitrogen, and total phosphorus from the candidate disturbance propagation chain before replacement. Arrange these data into a candidate disturbance propagation chain sequence before replacement in chronological order. Calculate the matrix contour distance from the candidate disturbance propagation chain sequence before replacement to the historical abnormal operation segment using the same calculation method as the matrix contour distance in the three-state evidence data of control actions. Read the same data items from each candidate disturbance propagation chain after replacement, arrange them into a candidate disturbance propagation chain sequence after replacement in the same order, and calculate the candidate disturbance propagation chain sequence after replacement using the same calculation method as the matrix contour distance in the three-state evidence data of control actions. The matrix contour distance to the historical abnormal operation segment is calculated by subtracting the matrix contour distance from the candidate disturbance propagation chain sequence before replacement to the historical abnormal operation segment from the matrix contour distance after replacement. This yields the similarity distance change. When the similarity distance change is not less than 0.05, and the correspondence between the preceding object in the candidate disturbance propagation chain after replacement and the abnormal effluent result does not satisfy the relationships of continuous process flow direction, continuous segment time, or equipment adjustment object, the object to be replaced is recorded as the propagation cut object. The propagation cut object, similarity distance change, broken chain segment, counter-evidence chain segment, and corresponding segment identifier constitute the propagation cut evidence data, where: Propagation cut evidence data refers to evidence data formed based on the changes in matrix contour distance and the connectivity status of the propagation chain after performing in-phase stable segment replacement on process objects or equipment objects in the candidate perturbation propagation chain; Historically effective and inefficient control segments are extracted from the phase-constrained segment set. The STOMP algorithm is used to perform invertibility matrix contour retrieval to obtain control invertibility evidence data. This data, combined with control action three-state evidence data, anomaly propagation counter-evidence data, propagation cut evidence data, and segment identification data, yields a set of wastewater treatment segment evidence anchor points, where: The STOMP algorithm is used to perform invertible matrix contour retrieval, specifically as follows: Read the abnormal water discharge results, control action three-state evidence data, and propagation cut evidence data from the currently running segment. Select the segment sequence containing the propagation cut object, abnormal water discharge results, and control action data as the reversible segment to be retrieved. From historical effective control segments, select segments with the same abnormal water discharge results and the same equipment object as effective control retrieval objects. From historical inefficient control segments, select segments with the same abnormal water discharge results and the same equipment object as inefficient control retrieval objects. Calculate the matrix contour distance from the reversible segment to the effective control retrieval object and the matrix contour distance to the inefficient control retrieval object using the same calculation method as in the control action three-state evidence data. When the reversible segment to be retrieved reaches the effective control retrieval object... If the matrix contour distance of the target object is less than the matrix contour distance to the inefficient control retrieval object, the reversibility distance difference is obtained by subtracting the matrix contour distance from the reversible fragment to the efficient control retrieval object from the matrix contour distance from the reversible fragment to the efficient control retrieval object. If the reversibility distance difference is not less than the reversibility distance difference threshold, the propagation cut object is recorded as a reversible control object. If the reversibility distance difference is less than the reversibility distance difference threshold, or if the matrix contour distance from the reversible fragment to the inefficient control retrieval object is even smaller, the propagation cut object is recorded as a low-reversibility control object. The reversible control object, the low-reversible control object, the corresponding control action data, and the corresponding matrix contour distance constitute the control reversibility evidence data, where: The reversibility distance difference threshold is a threshold used to determine whether a reversible fragment to be retrieved is significantly closer to a historical effective control fragment. The reversibility distance difference threshold is set by reading historical reversibility distance differences, deleting historical reversibility distance differences less than or equal to 0, arranging the remaining historical reversibility distance differences in ascending order, and taking the values ​​at the 25th percentile of the sorted values ​​as the reversibility distance difference threshold. The set of evidence anchor points for wastewater treatment is obtained as follows: Read the three-state evidence data of control actions, the evidence data of abnormal propagation rebuttal, the evidence data of propagation segmentation, the evidence data of control reversibility, and the fragment identification data. The fragment identification data includes fragment type, fragment time, process phase identifier, process object, and equipment object. Combine the disturbing control evidence, the disturbing control evidence, the invalid control evidence, the evidence of rebuttal, the propagation segmentation object, the reversible control object, and the low reversibility control object with the corresponding fragment identification data to obtain fragment evidence anchor points. All fragment evidence anchor points constitute the wastewater treatment fragment evidence anchor point set, in which: Fragment evidence anchor points refer to evidence records formed by binding evidence data in an operational segment with segment identification data. Evidence data includes control action three-state evidence data, abnormal propagation counter-evidence data, propagation cut evidence data, and control reversibility evidence data. Fragment identification data includes segment type, segment time, process phase identifier, process object, and equipment object.

[0024] In this embodiment, the extraction of the steady-state subgraph and the disturbance subgraph includes: Based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, process object nodes and segment evidence anchor point nodes are extracted. Process association edges are then established according to process flow direction, pollutant transformation, equipment adjustment, and control response to obtain a wastewater treatment process knowledge graph, in which: Wastewater treatment process object data refers to the corresponding relationship data between treatment sections, regulating equipment, pollutant indicators and control actions within a wastewater treatment plant, including influent section, biochemical reaction section, sedimentation and separation section, effluent section, aeration equipment, reflux equipment, dosing equipment, sludge discharge equipment, chemical oxygen demand, ammonia nitrogen, total phosphorus, aeration volume, reflux ratio, dosing volume and sludge discharge time. Extract process object nodes and fragment evidence anchor point nodes, specifically: Read the wastewater treatment process object data, extract the influent section, biochemical reaction section, sedimentation and separation section and effluent section as treatment section nodes, extract the aeration equipment, reflux equipment, dosing equipment and sludge discharge equipment as equipment nodes, extract chemical oxygen demand, ammonia nitrogen and total phosphorus as pollutant index nodes, extract aeration volume, reflux ratio, dosing volume and sludge discharge time as control action nodes, read the wastewater treatment segment evidence anchor point set, and extract the disturbing control evidence, the suppressing control evidence, the ineffective control evidence, the counter-evidence chain segment, the propagation cut object, the reversible control object and the low reversible control object as segment evidence anchor point nodes; Process-related edges are established based on process flow direction, pollutant transformation, equipment adjustment, and control response, specifically as follows: Following the process sequence of influent section, biochemical reaction section, sedimentation and separation section, and effluent section, process flow edges are established between adjacent treatment section nodes. The average values ​​of chemical oxygen demand (COD), ammonia nitrogen (NH3), and total phosphorus (TP) in adjacent treatment sections are read. When the variation range between the average value of the index in the previous treatment section and the average value of the same index in the next treatment section is not less than 3%, a pollutant conversion edge is established. Equipment adjustment edges are established between aeration equipment and aeration volume, between reflux equipment and reflux ratio, between dosing equipment and dosing volume, and between sludge discharge equipment and sludge discharge time. Three-state evidence data of control actions are read. When the COD, NH3, or TP in the effluent response segment is recorded as increasing or decreasing after the control action changes, a control response edge is established between the control action, equipment, treatment section, pollutant index, and segment evidence anchor point. The wastewater treatment process knowledge graph is obtained as follows: The wastewater treatment process knowledge graph is obtained by using the processing segment node, equipment node, pollutant index node, control action node and fragment evidence anchor point node as graph nodes, and the process flow edge, pollutant conversion edge, equipment adjustment edge and control response edge as graph edges. The corresponding graph nodes are connected according to the graph edges. The evidence data and fragment identifier data in the fragment evidence anchor node are mapped to the corresponding process object node and process association edge to obtain the evidence mapping graph, where: Evidence data includes control action three-state evidence data, abnormal propagation counter-evidence data, propagation cut evidence data, and control reversibility evidence data; fragment identification data includes fragment type, fragment time, process phase identifier, process object, and equipment object. The evidence mapping spectrum is obtained as follows: Read the evidence data and fragment identification data from the fragment evidence anchor node. According to the process object and equipment object in the fragment identification data, associate the evidence data with the process object node with the same object name. According to the process phase identifier and fragment time in the fragment identification data, associate the evidence data with the process association edge with the same process phase identifier and connected fragment time to obtain the evidence mapping graph. From the evidence mapping graph, extract the corresponding nodes and process-related edges of historical stable operation segments and historical effective control segments to form a stable operating condition subgraph. Extract the corresponding nodes and process-related edges of the current operating segment, historical abnormal operating segments, and historical inefficient control segments to form a disturbance operating condition subgraph, where: The steady-state subplot is composed of the following: Read the segment evidence anchor node of segment type historical stable operation segment and historical effective control segment from the evidence mapping graph, extract the process object node and process association edge associated with historical stable operation segment and historical effective control segment, and form a stable operating condition subgraph; The disturbance condition sub-diagram is composed of the following: Extract the segment evidence anchor nodes of the segment types of current running segment, historical abnormal running segment, and historical inefficient control segment from the evidence mapping graph, extract the process object nodes and process association edges associated with the current running segment, historical abnormal running segment, and historical inefficient control segment, and form a disturbance condition subgraph.

[0025] In this embodiment, generating the disturbance propagation difference map includes: An improved AMNet model is constructed, comprising a dual-condition evidence input layer, an evidence modulation multi-frequency extraction layer, a difference constraint processing layer, and a difference map output layer, wherein: The improved AMNet model is constructed as follows: Based on the graph data input layer of the traditional AMNet model, the single graph input method is changed to a combined input of the stable operating condition subgraph, the disturbed operating condition subgraph, and the evidence anchor point set of the wastewater treatment segment, resulting in a dual-operating-condition evidence input layer. Based on the multi-frequency relationship extraction layer of the traditional AMNet model, the method of extracting multi-frequency features solely based on graph connectivity is changed to extracting multi-frequency relationships modulated according to evidence identifiers, resulting in an evidence-modulated multi-frequency extraction layer. Based on the adaptive fusion layer of the traditional AMNet model, the multi-frequency feature fusion method is changed to subgraph differential processing between the stable operating condition subgraph and the disturbed operating condition subgraph, and process causal conflict resolution is added, resulting in a differential constraint processing layer. Based on the anomaly output layer of the traditional AMNet model, the anomaly score output method is changed to a disturbance propagation difference graph output method, resulting in a difference graph output layer, thus obtaining the improved AMNet model, where: AMNet model refers to the adaptive multi-frequency graph neural network model. The basic structure of the traditional AMNet model includes a graph data input layer, a multi-frequency relationship extraction layer, an adaptive fusion layer, and an anomaly output layer. The dual-condition evidence input layer, based on the stable condition subgraph, the disturbed condition subgraph, and the wastewater treatment segment evidence anchor point set, performs evidence polarity recoding on the evidence data to obtain support disturbance markers, disproven evidence exclusion markers, propagation blocking markers, and reversible control markers. These are then written into the corresponding nodes and process-related edges, resulting in the dual-condition evidence input graph, where: The dual-condition evidence input layer includes: Subgraph reader: Reads the steady-state subgraph and the disturbance subgraph; Evidence Re-encoder: Reads the set of evidence anchor points for wastewater treatment segments and converts disturbance control evidence, counter-evidence chains, propagation cut objects, and reversible control objects into disturbance support identifiers, counter-evidence exclusion identifiers, propagation blocking identifiers, and reversible control identifiers, respectively. Identifier Writer: Writes perturbation identifiers, counter-evidence exclusion identifiers, propagation blocking identifiers, and reversible control identifiers to the corresponding nodes and process-related edges to obtain a dual-condition evidence input graph. The evidence data is recoded to reflect its polarity, specifically as follows: Read the steady-state subgraph, the disturbance subgraph, and the wastewater treatment segment evidence anchor point set. Then, read the control action three-state evidence data, anomaly propagation counter-evidence data, propagation cut evidence data, and control reversibility evidence data from the wastewater treatment segment evidence anchor point set. When the evidence data is disturbing control evidence, the supporting disturbance flag is recorded as 1. When the evidence data is a counter-evidence chain segment, the counter-evidence exclusion flag is recorded as 1. When the evidence data is a propagation cut object, the propagation blocking flag is recorded as 1. When the evidence data is a reversible control object, the reversible control flag is recorded as 1. The flag for evidence data that does not correspond is recorded as 0. The supporting disturbance flag, counter-evidence exclusion flag, propagation blocking flag, and reversible control flag are obtained. The dual-condition evidence input diagram is obtained as follows: Read data supporting disturbance identifiers, counter-evidence exclusion identifiers, propagation blocking identifiers, reversible control identifiers, and fragment identifiers. Fragment identifier data includes fragment type, fragment time, process phase identifier, process object, and equipment object. According to the process object and equipment object in the fragment identifier data, write the four types of evidence identifiers into nodes with the same object name. According to the process phase identifier and fragment time in the fragment identifier data, write the four types of evidence identifiers into process association edges with the same process phase identifier and the same fragment time. The stable operating condition subgraph and the disturbance operating condition subgraph after writing the evidence identifiers form a dual-operating condition evidence input graph. The evidence modulation multi-frequency extraction layer, based on a dual-condition evidence input graph, performs evidence polarity frequency band routing according to support disturbance flags, rebuttal evidence exclusion flags, propagation blocking flags, and reversible control flags. It extracts low-frequency stable relationships, high-frequency disturbance relationships, and propagation cut-sensitive relationships to obtain a dual-condition multi-frequency relationship representation, where: The evidence modulation multi-frequency extraction layer includes: Stable Relationship Filter: Extracts nodes and process-related edges that support disturbance flags of 0 and propagation blocking flags of 0, forming low-frequency stable relationships; Disturbance Relationship Filter: Extracts nodes and process-related edges that support a disturbance flag of 1 and have a counter-evidence exclusion flag of 0, forming high-frequency disturbance relationships; Sensitive Relationship Filter: Extracts nodes with a propagation blocking flag of 1 or a reversible control flag of 1, as well as process-related edges, to form propagation cut sensitive relationships; Relation aggregator: Aggregates low-frequency stable relations, high-frequency perturbation relations, and propagation cut-sensitive relations to obtain a dual-condition multi-frequency relation representation. Evidence polarity frequency band routing refers to the process of assigning nodes and process-related edges in the dual-condition evidence input graph to low-frequency stable relationships, high-frequency disturbance relationships, and propagation cut-sensitive relationships based on support disturbance flags, disinformation exclusion flags, propagation blocking flags, and reversible control flags. The execution of evidence polarity frequency band routing is as follows: Read the nodes, process-related edges, support disturbance markers, disproving evidence exclusion markers, propagation blocking markers, and reversible control markers from the dual-condition evidence input graph. In the stable condition subgraph, the nodes with support disturbance markers of 0 and propagation blocking markers of 0, along with the process-related edges, form a low-frequency stable relationship. In the disturbance condition subgraph, the nodes with support disturbance markers of 1 and disproving evidence exclusion markers of 0, along with the process-related edges, form a high-frequency disturbance relationship. The nodes with propagation blocking markers of 1 or reversible control markers of 1, along with the process-related edges, form a propagation cut-sensitive relationship. The nodes with disproving evidence exclusion markers of 1, along with the process-related edges, form a disproving evidence exclusion record. Delete the nodes and process-related edges that are the same as the disproving evidence exclusion record from the high-frequency disturbance relationship. The dual-condition multi-frequency relationship is represented as follows: The evidence modulation multi-frequency extraction layer reads low-frequency stable relationships, high-frequency perturbation relationships, and propagation cut-sensitive relationships, and extracts the node identifier, process-related edge identifier, process phase identifier, fragment time, and corresponding evidence identifier from each type of relationship to form low-frequency stable relationship records, high-frequency perturbation relationship records, and propagation cut-sensitive relationship records. The three types of relationship records form a dual-condition multi-frequency relationship representation. The differential constraint processing layer, based on a dual-condition multi-frequency relationship representation, performs mirror differential processing on nodes with consistent object identifiers and process-related edges with consistent connected objects. It then performs process causal conflict resolution according to process flow direction, pollutant conversion relationship, equipment adjustment object, fragment occurrence time, counter-evidence exclusion markers, and propagation blocking markers, filtering the dominant disturbance path to obtain node difference data, edge difference data, and effective propagation difference data, among which: The differential constraint processing layer includes: Node Differentiator: Compares the differences in the identifiers of nodes with the same object identifier to obtain node difference data; Edge Differencer: Compares the segment time and identifier differences of edges with consistent process associations to obtain edge difference data; Conflict resolver: Deletes candidate disturbance paths with counter-evidence exclusion markers and those that do not conform to process flow direction, pollutant conversion relationship, equipment adjustment object, or segment time sequence; Dominant path extractor: retains candidate perturbation paths for connection propagation cut objects, reversible control objects, and abnormal water discharge results to obtain effective propagation difference data; For nodes with consistent object identifiers and process-related edges with consistent connected objects, perform mirroring and differential processing, specifically as follows: Read the dual-condition multi-frequency relationship representation, match nodes with consistent object identifiers in the stable condition subgraph and the disturbance condition subgraph, compare the supporting disturbance identifier, the counter-evidence exclusion identifier, the propagation blocking identifier, and the reversible control identifier of the corresponding nodes. When there are inconsistent identifier values, record the node object identifier, process phase identifier, fragment time, and inconsistency identifier to obtain node difference data. Match process association edges with consistent start and end objects in the stable condition subgraph and the disturbance condition subgraph, compare the fragment time and evidence identifier of the corresponding process association edges, and when there are inconsistencies, record the edge start object, edge end object, fragment time, and inconsistency identifier to obtain edge difference data. The pollutant transformation relationship refers to the corresponding relationship of chemical oxygen demand, ammonia nitrogen, and total phosphorus as they change with the treatment sequence in the influent section, biochemical reaction section, sedimentation and separation section, and effluent section. The equipment adjustment objects refer to the equipment and treatment sections directly affected by the control actions, including aeration equipment and biochemical reaction sections corresponding to aeration volume, reflux equipment and biochemical reaction sections corresponding to reflux ratio, dosing equipment and sedimentation separation sections corresponding to dosage, and sludge discharge equipment and sedimentation separation sections corresponding to sludge discharge time. Process causal conflict resolution refers to screening candidate disturbance paths according to process flow direction, pollutant transformation relationship, equipment regulation object, fragment occurrence time, counter-evidence exclusion marker, and transmission blocking marker, and deleting candidate disturbance paths that do not conform to the process sequence, transformation relationship, regulation object relationship, time sequence, or have counter-evidence exclusion marker; The implementation of process causal conflict resolution is specifically as follows: Read node difference data, edge difference data, process flow direction, pollutant transformation relationship, equipment adjustment object, fragment occurrence time, disproven exclusion marker, and propagation blockage marker. Extract abnormal process objects, propagation cut objects, reversible control objects, and abnormal effluent results from the node difference data. Extract process association edges connecting these objects from the edge difference data and connect them into candidate disturbance paths according to the fragment occurrence time from earliest to latest. When the same abnormal effluent result corresponds to more than two candidate disturbance paths, delete the candidate disturbance paths with disproven exclusion markers. Delete candidate disturbance paths that do not conform to the process flow direction, pollutant transformation relationship, equipment adjustment object, or fragment time order. Among the remaining candidate disturbance paths, retain the path connecting the propagation cut object, reversible control object, and abnormal effluent result as the dominant disturbance path. The node difference data, edge difference data, and path connection relationship in the dominant disturbance path constitute the effective propagation difference data. The difference graph output layer, based on node difference data, edge difference data, and effective propagation difference data, retains effective abnormal propagation paths connecting abnormal water discharge results, propagation cut objects, and reversible control objects, while deleting abnormal propagation paths marked with proof of contradiction, thus obtaining the disturbance propagation difference graph, where: The difference map output layer includes: Path filter: Reads the dominant perturbation path in the valid propagation difference data; Dissent Deletor: Deletes abnormal propagation paths marked with a dissent exclusion flag; Difference Outputter: Based on the retained nodes, process-related edges, node difference data, edge difference data, and effective propagation difference data, outputs a disturbance propagation difference map; The perturbation propagation difference diagram is obtained as follows: Read node difference data, edge difference data, and effective propagation difference data. Extract the dominant disturbance path from the effective propagation difference data. Keep the nodes and process-related edges corresponding to the dominant disturbance path. Delete abnormal propagation paths with counter-evidence exclusion markers and their corresponding process-related edges. The retained nodes, process-related edges, node difference data, edge difference data, and effective propagation difference data form a disturbance propagation difference graph. The improved AMNet model was trained using perturbation propagation difference map reconstruction error, anomaly propagation path identification error, and control attribution chain matching error as joint optimization objectives. The parameters of the dual-condition evidence input layer, evidence modulation multi-frequency extraction layer, differential constraint processing layer, and difference map output layer were continuously optimized. Training was stopped when the decrease in the joint optimization objective was less than 0.001 for five consecutive training rounds, or when the training rounds reached 200. The resulting improved AMNet model was then obtained. The improved AMNet model is trained as follows: The stable condition subgraph, the disturbed condition subgraph, and the evidence anchor point set of the wastewater treatment segment are input into the improved AMNet model, outputting a training disturbance propagation difference map. Node difference data, edge difference data, and effective propagation difference data in the training disturbance propagation difference map are read and compared line by line with the node difference data, edge difference data, and effective propagation difference data in the labeled disturbance propagation difference map. The number of inconsistent data points is counted and then divided by the sum of the number of node difference data points, edge difference data points, and effective propagation difference data points in the labeled disturbance propagation difference map to obtain the disturbance propagation difference map reconstruction error. The dominant disturbance path in the training disturbance propagation difference map is read and compared line by line with the labeled dominant disturbance path. The number of inconsistent paths is counted and then divided by the number of labeled dominant disturbance paths to obtain the abnormal propagation path identification error. The control attribution chain corresponding to the training disturbance propagation difference map is read and compared line by line with the labeled control attribution chain according to the object connection order. The number of inconsistent object identifiers is counted and then divided by the number of object positions included in the labeled control attribution chain to obtain the control attribution chain matching error. Where: The disturbance propagation difference diagram, the dominant disturbance path, and the control attribution chain are generated from historical operation segments, historical alarm records, historical control action records, equipment maintenance records, and effluent index recovery records, and are reviewed and confirmed by process personnel. The perturbation propagation difference map reconstruction error is multiplied by 0.4, the anomaly propagation path identification error is multiplied by 0.3, and the control attribution chain matching error is multiplied by 0.3, and then summed to obtain the joint optimization objective. After each training round, the parameters of the dual-condition evidence input layer, evidence modulation multi-frequency extraction layer, differential constraint processing layer, and difference map output layer are adjusted in the direction of decreasing joint optimization objective. The step size of parameter adjustment is 0.0005. When the decrease of joint optimization objective in 5 consecutive training rounds is less than 0.001, or when the number of training rounds reaches 200, training is stopped and the improved AMNet model is obtained.

[0026] In this embodiment, obtaining the effective perturbation propagation subgraph includes: Based on the node difference data, edge difference data, effective propagation difference data, and segment occurrence time in the disturbance propagation difference graph, candidate disturbance paths are extracted to connect abnormal effluent results, abnormal process objects, abnormal control relationships, and equipment adjustment objects. Extract candidate disturbance paths from abnormal effluent results, abnormal process objects, abnormal control relationships, and equipment adjustment objects, specifically as follows: Read the node difference data, edge difference data, effective propagation difference data, and segment occurrence time from the disturbance propagation difference graph. Take the pollutant index node with abnormal effluent result identifier as the path endpoint, the treatment section node and equipment node with node difference data as abnormal process objects, the process association edge with edge difference data that connects the control action node, equipment node, treatment section node and pollutant index node as abnormal control relationship, and the equipment node in the abnormal control relationship as equipment adjustment object. Connect the abnormal process object, equipment adjustment object, abnormal control relationship and abnormal effluent result in order of segment occurrence time from early to late to obtain the candidate disturbance path. Candidate disturbance paths are validated for consistency based on process flow direction, pollutant transformation relationships, equipment regulation targets, and segment occurrence times. Candidate disturbance paths that do not conform to the process sequence, transformation relationships, regulation target relationships, and time sequence are deleted, resulting in causally consistent disturbance paths. The path consistency check is performed as follows: Read the nodes, process association edges, process flow direction, pollutant conversion relationship, equipment adjustment object and segment occurrence time in the candidate disturbance path, delete the candidate disturbance path that does not conform to the process sequence of influent section, biochemical reaction section, sedimentation separation section and effluent section, delete the candidate disturbance path that the pollutant index does not meet the pollutant conversion relationship with the treatment section, delete the candidate disturbance path that the control action is inconsistent with the equipment adjustment object, and delete the candidate disturbance path whose segment occurrence time of abnormal process object or abnormal control relationship is later than the segment occurrence time of abnormal effluent result. The causally consistent perturbation path is obtained as follows: The candidate disturbance paths retained after path consistency verification are arranged in the connection order of abnormal process objects, equipment adjustment objects, abnormal control relationships and abnormal effluent results, and the corresponding nodes, process association edges, fragment occurrence time and effective propagation difference data are retained to obtain causal consistent disturbance paths. Nodes, process-related edges, and effective propagation difference data are extracted from the causal consistent perturbation path to form an effective perturbation propagation subgraph, where: The effective perturbation propagation subgraph is composed as follows: Read the nodes, process-related edges, and effective propagation difference data in the causal consistent perturbation path, retain the interconnected nodes and process-related edges within the same path, and retain the node difference data, edge difference data, fragment occurrence time, and effective propagation difference data to form an effective perturbation propagation subgraph.

[0027] In this embodiment, constructing the control attribution chain includes: Based on the nodes, process-related edges, effective propagation difference data, and segment identification data in the effective disturbance propagation subgraph, abnormal process objects, abnormal control relationships, and abnormal effluent results are identified according to node object type, where: Identify abnormal process objects, abnormal control relationships, and abnormal effluent results according to node object type, specifically: Read the nodes, process association edges, effective propagation difference data and segment identification data in the effective disturbance propagation subgraph. Identify the treatment segment nodes and equipment nodes that carry node difference data or propagation cut object identification as abnormal process objects. Identify the process association edges that carry edge difference data or effective propagation difference data and connect control action nodes, equipment nodes, treatment segment nodes and pollutant index nodes as abnormal control relationships. Identify the pollutant index nodes with chemical oxygen demand greater than 50 mg / L, ammonia nitrogen greater than 5 mg / L or total phosphorus greater than 0.5 mg / L as abnormal effluent results. Based on process flow direction, pollutant transformation relationships, equipment regulation targets, and segment occurrence time, abnormal process objects, abnormal control relationships, and abnormal effluent results are ranked to obtain candidate control attribution paths, among which: The candidate control attribution paths are obtained as follows: Read the abnormal process objects, abnormal control relationships, abnormal effluent results, process flow direction, pollutant transformation relationships, equipment adjustment objects, and segment occurrence time; first arrange the abnormal process objects and abnormal control relationships from earliest to latest according to the segment occurrence time, and then correct the arrangement results according to the process sequence of influent section, biochemical reaction section, sedimentation separation section, and effluent section; connect the abnormal process objects, equipment adjustment objects, abnormal control relationships, and abnormal effluent results in sequence. When the connection sequence simultaneously satisfies the correspondence between pollutant transformation relationships and equipment adjustment objects, a candidate control attribution path is obtained. Based on the effective propagation difference data and process-related edge connection order in the candidate control attribution paths, candidate control attribution paths that continuously connect abnormal process objects, abnormal control relationships, and abnormal effluent results are selected, and a control attribution chain is constructed, wherein: Constructing a control attribution chain, specifically: Read the effective propagation difference data and process association edge connection order in the candidate control attribution path, retain the candidate control attribution path that continuously connects abnormal process objects, abnormal control relationships and abnormal effluent results, delete the candidate control attribution path that disconnects process association edges, lacks effective propagation difference data or has reversed segment occurrence time, and form a control attribution chain by combining the retained candidate control attribution paths in the order of abnormal process objects, abnormal control relationships and abnormal effluent results.

[0028] In this embodiment, obtaining executable control instructions includes: Based on the abnormal process objects, abnormal control relationships, and abnormal effluent results in the control attribution chain, candidate control strategies for wastewater treatment are determined according to the equipment regulation objects corresponding to the abnormal control relationships, wherein: The determination of candidate control strategies for wastewater treatment is as follows: Read the abnormal process objects, abnormal control relationships, and abnormal effluent results in the control attribution chain, determine the equipment adjustment objects corresponding to the abnormal control relationships, generate an aeration rate adjustment strategy when the equipment adjustment object is an aeration device, generate a return ratio adjustment strategy when the equipment adjustment object is a return flow device, generate a dosing rate adjustment strategy when the equipment adjustment object is a dosing device, and generate a sludge discharge duration adjustment strategy when the equipment adjustment object is a sludge discharge device. Read the control action data with the same abnormal effluent results and the same equipment adjustment object in the historical effective control segments, take the median of the change range of the control action as the candidate adjustment range, take 3% when the candidate adjustment range is less than 3%, and take 15% when it is greater than 15%, to obtain the candidate control strategies for wastewater treatment. Based on equipment operating boundaries, process operating boundaries, and effluent discharge limits, safety constraints are checked on candidate control strategies for wastewater treatment. Candidate control strategies that exceed equipment operating boundaries, process operating boundaries, or effluent discharge limits are deleted, resulting in a safe control strategy, where: The equipment operating boundary is the allowable range of equipment operating data, including aeration equipment load rate not exceeding 95%, reflux equipment load rate not exceeding 90%, chemical dosing equipment load rate not exceeding 90%, and sludge discharge time of sludge discharge equipment not exceeding 30 minutes per cycle. The process operating boundary is the allowable range of process section status data, including dissolved oxygen in the biochemical reaction section not less than 2 mg / L and not more than 4 mg / L, reflux ratio not less than 50% and not more than 150%, and chemical dosing after adjustment not exceeding 120% of the current dosing. The effluent discharge index limits are chemical oxygen demand not exceeding 50 mg / L, ammonia nitrogen not exceeding 5 mg / L, and total phosphorus not exceeding 0.5 mg / L. The safety constraints of the candidate control strategies for wastewater treatment are checked, specifically as follows: The system reads the equipment adjustment objects, control actions, and candidate adjustment ranges from the candidate control strategies for wastewater treatment, and reads the equipment operation data, process segment status data, and effluent index data. It then overlays the candidate adjustment ranges onto the current control action data to obtain the candidate execution values. When the candidate execution value causes the equipment operation data to exceed the equipment operation boundary, the corresponding wastewater treatment candidate control strategy is deleted. When the candidate execution value causes the process segment status data to exceed the process operation boundary, the corresponding wastewater treatment candidate control strategy is deleted. When the chemical oxygen demand, ammonia nitrogen, or total phosphorus in the historical effective control segment corresponding to the candidate execution value exceeds the effluent discharge index limit, the corresponding wastewater treatment candidate control strategy is deleted. The security control strategy is as follows: The wastewater treatment candidate control strategies retained after the safety constraint verification are used as safety control strategies. When there are more than two safety control strategies corresponding to the same equipment adjustment object, the safety control strategy with the smallest candidate adjustment range and the largest decrease in the effluent index of the historical effective control segment is retained. Based on safety control strategies and equipment operation data, the safety control strategies are converted into corresponding equipment adjustment parameters to obtain executable control commands, wherein: The executable control instructions are obtained, specifically: The system reads the equipment adjustment object, control action, and candidate adjustment range from the safety control strategy, and reads the current equipment parameter value from the equipment operation data. When the control action is to increase, the current equipment parameter value is multiplied by the candidate adjustment range to obtain the parameter increase amount. Then, the current equipment parameter value is added to the parameter increase amount to obtain the equipment adjustment parameter. When the control action is to decrease, the current equipment parameter value is multiplied by the candidate adjustment range to obtain the parameter decrease amount. Then, the current equipment parameter value is subtracted from the parameter decrease amount to obtain the equipment adjustment parameter. When the equipment adjustment object is an aeration device, the equipment adjustment parameter is used as the aeration rate setpoint. When the equipment adjustment object is a recirculation device, the equipment adjustment parameter is used as the recirculation ratio setpoint. When the equipment adjustment object is a chemical dosing device, the equipment adjustment parameter is used as the chemical dosing rate setpoint. When the equipment adjustment object is a sludge discharge device, the equipment adjustment parameter is used as the sludge discharge duration setpoint. The equipment identifier, equipment adjustment object, control action, equipment adjustment parameter, execution time, and safety control strategy identifier are combined to form an executable control command.

[0029] refer to Figure 3 A knowledge graph-based intelligent control system for wastewater treatment plants includes the following modules: The industrial time series management module is used to collect industrial operation data from wastewater treatment plants and perform preprocessing to obtain standardized industrial operation time series. The phase anchor point parsing module is used to process standardized industrial operation time series using the STOMP algorithm to obtain a set of evidence anchor points for wastewater treatment segments. The process graph twin module is used to construct a wastewater treatment process knowledge graph based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, and to extract the steady-state subgraph and the disturbed state subgraph. The perturbation difference inference module is used to process the stable operating condition subgraph and the perturbation operating condition subgraph by improving the AMNet model, and generate a perturbation propagation difference graph. The causality consistency verification module is used to perform process causality consistency verification on the disturbance propagation difference graph to obtain a valid disturbance propagation subgraph; The control attribution and tracing module is used to identify abnormal process objects, abnormal control relationships and abnormal effluent results in the effective disturbance propagation subgraph, and to construct the control attribution chain; The safety instruction decision module is used to determine the wastewater treatment control strategy based on the control attribution chain, and to perform safety constraint verification to obtain executable control instructions.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a continuous operation simulation scenario of a wastewater treatment plant. The wastewater treatment plant adopts a treatment process consisting of an influent section, a biological reaction section, a sedimentation and separation section, and an effluent section. The control equipment includes aeration equipment, reflux equipment, dosing equipment, and sludge removal equipment. During one continuous operation cycle, the system collected data showing that the influent chemical oxygen demand increased from 286 mg / L to 421 mg / L, the influent ammonia nitrogen increased from 24.8 mg / L to 39.2 mg / L, the influent total phosphorus increased from 3.6 mg / L to 5.9 mg / L, the influent flow rate increased from 815 m³ / h to 962 m³ / h, while the reflux ratio decreased from 94% to 69%, and the dissolved oxygen in the biological reaction section decreased from 2.7 mg / L to 1.6 mg / L. After a 60-minute response, the effluent ammonia nitrogen rose to 6.3 mg / L and the effluent total phosphorus rose to 0.73 mg / L, exceeding the effluent limits of 5 mg / L and 0.5 mg / L respectively, indicating that the wastewater treatment plant experienced a combination of influent load shock and abnormal equipment regulation.

[0031] The system standardized the sampling interval to 5 minutes, with each 12 sampling points forming one operational segment. The 960 sampling points created 80 operational segments, including 1 current operational segment, 51 historical stable operational segments, 16 historical abnormal operational segments, 8 historical effective control segments, and 4 historical inefficient control segments, totaling 80 segments. In the training samples, the stable samples had an influent COD of 292 mg / L, ammonia nitrogen of 25.1 mg / L, total phosphorus of 3.8 mg / L, a reflux ratio of 96%, and effluent ammonia nitrogen of 3.4 mg / L and effluent total phosphorus of 0.32 mg / L. The abnormal samples had an influent COD of 408 mg / L, ammonia nitrogen of 37.6 mg / L, total phosphorus of 5.5 mg / L, a reflux ratio of 71%, and effluent ammonia nitrogen of 6.4 mg / L and effluent total phosphorus of [missing data]. With a dosage of 0.69 mg / L, in the effective control sample, after increasing the reflux ratio by 10% and the aeration rate by 6%, the effluent ammonia nitrogen decreased from 6.2 mg / L to 4.3 mg / L and the effluent total phosphorus decreased from 0.68 mg / L to 0.41 mg / L in the three operating segments. In the inefficient control sample, after increasing the dosage of the single chemical dosing by 14%, the effluent total phosphorus decreased from 0.70 mg / L to only 0.58 mg / L and the effluent ammonia nitrogen decreased from 6.1 mg / L to only 5.9 mg / L.

[0032] The system uses the STOMP algorithm to retrieve the similarity relationship between the current running segment and historical running segments. The matrix profile distance from the current running segment to the historical stable running segment is 0.228, and the matrix profile distance to the historical abnormal running segment is 0.067. Therefore, the current running segment is determined to be in an abnormal running state. The system performs stabilizing action replacement and reversing action replacement on segments with decreasing reflux ratios. The distance from the original reflux ratio action state to the historical abnormal running segment is 0.064. After the reflux ratio is replaced with a stable value of 94%, the distance increases to 0.146. After the reflux ratio is reversed and enhanced to 102%, the distance to the historical stable running segment decreases to 0.081. Therefore, the system records a decrease in reflux ratio as evidence of disturbance control and an increase in reflux ratio as evidence of disturbance suppression control. The system performs the same processing on segments with changes in dosage. The distance difference between the three states is less than 0.02. Therefore, a change in dosage alone is recorded as invalid control evidence.

[0033] When identifying abnormal effluent results in the current operating segment, the system recorded effluent ammonia nitrogen (6.3 mg / L) and total phosphorus (0.73 mg / L) as abnormal results. The system read the median of historical stable operating segments for the same process phase and found that the reflux ratio decreased by 26.6% relative to 94%, dissolved oxygen in the biochemical reaction section decreased by 40.7% relative to 2.7 mg / L, and influent ammonia nitrogen load increased by 43.1% relative to the historical median. Therefore, the reflux equipment, biochemical reaction section, and influent section were recorded as candidate causes of abnormality. The system retrieved 11 historical operating segments with the same direction of influent load change and identical abnormal effluent results. Among these, 7 segments did not show abnormal changes in the dosing equipment but still exhibited excessive total phosphorus, providing evidence that the dosing equipment was not the primary cause of the abnormality.

[0034] After constructing candidate perturbation propagation chains, the system performs in-phase stable segment replacements on the reflux unit, biochemical reaction section, and precipitation separation section. After replacing the reflux unit with a stable reflux ratio of 94%, the distance from the candidate perturbation propagation chain to the historical abnormal operating segment increased from 0.067 to 0.153, and the similarity distance changed to 0.086. After replacing the dissolved oxygen in the biochemical reaction section with 2.7 mg / L, the similarity distance changed to 0.061, and after replacing the precipitation separation section, the similarity distance changed to 0.018. The system records the reflux unit and biochemical reaction section as propagation cut targets. The system inputs the stable operating condition subgraph, the disturbed operating condition subgraph, and the set of evidence anchor points for the wastewater treatment segment into the improved AMNet model, obtaining 6 node difference data, 5 edge difference data, and 3 effective propagation difference data. The output disturbance propagation difference map retains the dominant disturbance path of increased influent load - decreased reflux ratio - decreased dissolved oxygen in the biochemical reaction section - excessive ammonia nitrogen in effluent, and deletes the path of abnormal dosing equipment - excessive total phosphorus with the counter-evidence exclusion mark.

[0035] After performing a process causal consistency check on the disturbance propagation difference graph, the system obtains an effective disturbance propagation subgraph and constructs a control attribution chain: abnormal reflux equipment - decreased reflux ratio - decreased biochemical reaction section processing capacity - ammonia nitrogen and total phosphorus exceedances. Based on the control attribution chain, the system generates candidate control strategies to increase the reflux ratio by 10% and the aeration rate by 6%. Safety constraint checks show that the reflux equipment load rate increases from 71% to 81%, falling below 90%; the aeration equipment load rate increases from 78% to 83%, falling below 95%; and the dissolved oxygen in the biochemical reaction section is expected to increase from 1.6 mg / L to 2.4 mg / L, falling within the range of 2 mg / L to 4 mg / L. Therefore, the system outputs executable control commands: the reflux ratio setpoint is adjusted from 69% to 75.9%, and the aeration rate setpoint is adjusted from 5200 cubic meters per hour to 5512 cubic meters per hour, with an execution cycle of 3 operating segments.

[0036] Under the same simulation data, the traditional fixed threshold control method only triggers a 15% increase in dosage after the total phosphorus in the effluent exceeds 0.5 mg / L. It takes 85 minutes to identify the source of the anomaly, with an accuracy rate of 62.0%. It takes 175 minutes to restore effluent ammonia nitrogen to below 5 mg / L and 146 minutes to restore effluent total phosphorus to below 0.5 mg / L. The conventional automatic control method increases the aeration rate by 10% but fails to identify the abnormal return ratio. It has an accuracy rate of 74.6% for identifying the abnormal source, and it takes 132 minutes to restore effluent ammonia nitrogen and 118 minutes to restore effluent total phosphorus. The method of this invention takes 36 minutes to identify the abnormal source, has an accuracy rate of 91.8%, and it takes 82 minutes to restore effluent ammonia nitrogen and 76 minutes to restore effluent total phosphorus. In 30 consecutive simulation cycles, the traditional fixed threshold control method showed 216 effluent exceeding the limit sampling points, the conventional automatic control method showed 131 effluent exceeding the limit sampling points, and the method of this invention showed 47 effluent exceeding the limit sampling points, proving that the present invention can more accurately identify abnormal propagation paths and generate safe and executable control commands.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based intelligent control method for wastewater treatment plants, characterized in that, include: Collect industrial operation data from wastewater treatment plants, perform preprocessing on the wastewater treatment plant industrial operation data, and obtain standardized industrial operation time series; Process phase constraints, operation segment binding, and control response correlation are performed on standardized industrial operation time series. The STOMP algorithm is used to calculate the matrix contour distance between operation segments and perform segment similarity retrieval to obtain the set of evidence anchor points for wastewater treatment segments. Based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, a knowledge graph of wastewater treatment processes is constructed, and subgraphs of steady-state and disturbed states are extracted. Based on the stable operating condition subgraph and the disturbance operating condition subgraph, an improved AMNet model is constructed, and multi-frequency relationship extraction, subgraph difference processing and process causal conflict resolution are performed to generate a disturbance propagation difference map. Based on the process flow direction, pollutant transformation relationship, equipment adjustment object and segment occurrence time, the disturbance propagation difference diagram is subjected to process causal consistency verification to obtain an effective disturbance propagation sub-diagram; Identify anomalous process objects, anomalous control relationships, and anomalous effluent results in the effective disturbance propagation subgraph, and construct a control attribution chain. Based on the control attribution chain, the wastewater treatment control strategy is determined, the safety constraint verification is performed, and the executable control instructions are obtained.

2. The intelligent control method for wastewater treatment plants based on knowledge graphs according to claim 1, characterized in that, The industrial operation data of the wastewater treatment plant specifically includes influent water quality data, influent flow rate data, process section status data, equipment operation data, control action data, effluent index data, and data collection timestamp data.

3. The intelligent control method for wastewater treatment plants based on knowledge graphs according to claim 1, characterized in that, The preprocessing of industrial operation data from wastewater treatment plants specifically includes data format standardization, timestamp correction, abnormal data removal, missing data completion, sampling interval standardization, unit standardization, and time series sorting.

4. The intelligent control method for wastewater treatment plants based on knowledge graphs according to claim 1, characterized in that, The obtained set of evidence anchor points for wastewater treatment fragments includes: Based on standardized industrial operation time series, process phase constraints are implemented to divide the current operation segment, historical stable operation segment, historical abnormal operation segment, historical effective control segment, and historical inefficient control segment. Each operation segment is then bound to the process object, segment time, and process phase to obtain a set of phase-constrained segments. Read the current running segment, historical stable running segment, and historical abnormal running segment from the phase constraint segment set. Perform stable action replacement and reverse action replacement on the control action change segment to obtain the control ablation segment and the control reverse enhancement segment. Associate the control action change segment with the effluent response segment and use the STOMP algorithm to calculate the matrix profile distance to obtain the three-state evidence data of the control action. Identify abnormal effluent results and candidate abnormal cause objects in the current operation segment, retrieve historical operation segments with the same influent load change direction, the same abnormal effluent results, and no abnormal changes in candidate abnormal cause objects, and obtain abnormal propagation counter-evidence data; Based on the current operating segment, historical stable operating segments, historical abnormal operating segments, three-state evidence data of control actions, and abnormal propagation counter-evidence data, a candidate disturbance propagation chain is constructed. The process objects and equipment objects in the candidate disturbance propagation chain are replaced with stable segments of the same phase, and the changes in similar distance and the connectivity state of the propagation chain before and after the replacement are calculated to obtain propagation cut evidence data. Historical effective control segments and historical inefficient control segments are extracted from the phase constraint segment set. The STOMP algorithm is used to perform invertibility matrix contour retrieval to obtain control invertibility evidence data. Combined with control action three-state evidence data, anomaly propagation counter-evidence data, propagation cut evidence data and segment identification data, a wastewater treatment segment evidence anchor point set is obtained.

5. The intelligent control method for wastewater treatment plants based on knowledge graphs according to claim 1, characterized in that, The extraction of the steady-state subgraph and the disturbance subgraph includes: Based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, process object nodes and segment evidence anchor point nodes are extracted, and process association edges are established according to process flow direction, pollutant transformation, equipment adjustment and control response to obtain a wastewater treatment process knowledge graph. The evidence data and fragment identification data in the fragment evidence anchor node are mapped to the corresponding process object node and process association edge to obtain the evidence mapping graph. Extract the corresponding nodes and process-related edges of historical stable operation segments and historical effective control segments from the evidence mapping graph to form a stable operating condition subgraph. Extract the corresponding nodes and process-related edges of the current operating segment, historical abnormal operating segment, and historical inefficient control segment to form a disturbance operating condition subgraph.

6. The intelligent control method for wastewater treatment plants based on knowledge graphs according to claim 1, characterized in that, The generation of the perturbation propagation difference map includes: An improved AMNet model is constructed, which includes a dual-condition evidence input layer, an evidence modulation multi-frequency extraction layer, a difference constraint processing layer, and a difference map output layer. The dual-condition evidence input layer is based on the stable condition subgraph, the disturbance condition subgraph, and the wastewater treatment segment evidence anchor point set. It performs evidence polarity recoding on the evidence data to obtain support disturbance identifiers, counter-evidence exclusion identifiers, propagation blocking identifiers, and reversible control identifiers, and writes them into the corresponding nodes and process-related edges to obtain the dual-condition evidence input graph. The evidence modulation multi-frequency extraction layer is based on the dual-condition evidence input graph. It performs evidence polarity frequency band routing according to the support disturbance flag, the rebuttal evidence exclusion flag, the propagation blocking flag, and the reversible control flag. It extracts low-frequency stable relations, high-frequency disturbance relations, and propagation cut sensitivity relations to obtain the dual-condition multi-frequency relation representation. The differential constraint processing layer is based on the dual-condition multi-frequency relationship representation. It performs mirror differential processing on nodes with consistent object identification and process-related edges with consistent connected objects. It also performs process causal conflict resolution according to process flow direction, pollutant conversion relationship, equipment adjustment object, fragment occurrence time, counter-evidence exclusion mark and propagation blocking mark, and filters the dominant disturbance path to obtain node difference data, edge difference data and effective propagation difference data. The difference graph output layer is based on node difference data, edge difference data, and effective propagation difference data. It retains the effective abnormal propagation paths that connect abnormal water discharge results, propagation cut objects, and reversible control objects, and deletes abnormal propagation paths with counter-evidence exclusion markers to obtain the disturbance propagation difference graph. The improved AMNet model was trained using perturbation propagation difference map reconstruction error, anomaly propagation path identification error, and control attribution chain matching error as joint optimization objectives. The parameters of the dual-condition evidence input layer, evidence modulation multi-frequency extraction layer, differential constraint processing layer, and difference map output layer were continuously optimized. When the decrease of the joint optimization objective in 5 consecutive training rounds was less than 0.001, or when the number of training rounds reached 200, training was stopped and the improved AMNet model was obtained.

7. The intelligent control method for wastewater treatment plants based on knowledge graphs according to claim 1, characterized in that, The obtained effective perturbation propagation subgraph includes: Based on the node difference data, edge difference data, effective propagation difference data and segment occurrence time in the disturbance propagation difference graph, candidate disturbance paths are extracted to connect abnormal water discharge results, abnormal process objects, abnormal control relationships and equipment adjustment objects. Based on the process flow direction, pollutant transformation relationship, equipment regulation object and segment occurrence time, perform path consistency verification on candidate disturbance paths, delete candidate disturbance paths that do not conform to the process sequence, transformation relationship, regulation object relationship and time sequence, and obtain causally consistent disturbance paths; Extract nodes, process-related edges, and effective propagation difference data from the causal consistent perturbation path to form an effective perturbation propagation subgraph.

8. The intelligent control method for wastewater treatment plants based on knowledge graphs according to claim 1, characterized in that, The construction of the control attribution chain includes: Based on the nodes, process-related edges, effective propagation difference data, and segment identification data in the effective disturbance propagation subgraph, abnormal process objects, abnormal control relationships, and abnormal effluent results are identified according to the node object type. Based on process flow direction, pollutant transformation relationship, equipment adjustment object and segment occurrence time, abnormal process object, abnormal control relationship and abnormal effluent result are sorted into paths to obtain candidate control attribution paths; Based on the effective propagation difference data and process-related edge connection order in the candidate control attribution paths, candidate control attribution paths that continuously connect abnormal process objects, abnormal control relationships and abnormal effluent results are selected, and control attribution chains are constructed.

9. The intelligent control method for wastewater treatment plants based on knowledge graphs according to claim 1, characterized in that, The obtained executable control instructions include: Based on the abnormal process objects, abnormal control relationships, and abnormal effluent results in the control attribution chain, candidate control strategies for wastewater treatment are determined according to the equipment adjustment objects corresponding to the abnormal control relationships. Based on the equipment operating boundary, process operating boundary, and effluent discharge index limit, the safety constraint verification is performed on the candidate control strategy for wastewater treatment. Candidate control strategies that exceed the equipment operating boundary, process operating boundary, or effluent discharge index limit are deleted to obtain the safety control strategy. Based on safety control strategies and equipment operation data, the safety control strategies are converted into corresponding equipment adjustment parameters to obtain executable control commands.

10. A knowledge graph-based intelligent control system for a wastewater treatment plant, executing the knowledge graph-based intelligent control method for a wastewater treatment plant as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The industrial time series management module is used to collect industrial operation data from wastewater treatment plants and perform preprocessing to obtain standardized industrial operation time series. The phase anchor point parsing module is used to process standardized industrial operation time series using the STOMP algorithm to obtain a set of evidence anchor points for wastewater treatment segments. The process graph twin module is used to construct a wastewater treatment process knowledge graph based on the set of evidence anchor points for wastewater treatment segments and the data of wastewater treatment process objects, and to extract the steady-state subgraph and the disturbed state subgraph. The perturbation difference inference module is used to process the stable operating condition subgraph and the perturbation operating condition subgraph by improving the AMNet model, and generate a perturbation propagation difference graph. The causality consistency verification module is used to perform process causality consistency verification on the disturbance propagation difference graph to obtain a valid disturbance propagation subgraph; The control attribution and tracing module is used to identify abnormal process objects, abnormal control relationships and abnormal effluent results in the effective disturbance propagation subgraph, and to construct the control attribution chain; The safety instruction decision module is used to determine the wastewater treatment control strategy based on the control attribution chain, and to perform safety constraint verification to obtain executable control instructions.