Multi-stage filter layer denitrification equipment fault early warning method and system based on edge computing

Abstract

The application provides a multi-stage filter layer denitrification equipment fault early warning method and system based on edge computing. The method comprises: performing time sequence alignment, data reliability evaluation and windowing processing on multi-source operation data at an edge computing node to generate filter layer operation state data; constructing a filter layer state atlas at the edge computing node and mapping the filter layer operation state data to filter layer nodes and associated edges in the filter layer state atlas; jointly representing the hydraulic state, denitrification conversion state, operation control state and monitoring confidence state of each filter layer node to generate filter layer state representation; calculating mechanism residuals representing the deviation degree of filter layer operation and generating fault evolution representation; inputting the fault evolution representation into a fault early warning model deployed at the edge computing node to determine the fault location, fault type and fault risk level and generate multi-stage filter layer denitrification equipment fault early warning results. The application can improve fault positioning accuracy, reduce false positive and false negative rates and shorten on-site early warning delay.

Description

Technical Field

[0001] This application relates to the field of intelligent operation and maintenance technology for wastewater treatment equipment, and in particular to a fault early warning method and system for multi-stage filter denitrification equipment based on edge computing. Background Technology

[0002] Multi-stage filter bed denitrification equipment is widely used in advanced wastewater treatment and reclaimed water reuse. It typically achieves further removal of pollutants such as nitrate nitrogen and total nitrogen through processes including multi-stage filter media retention, biofilm denitrification, carbon source regulation, and backwashing recovery. With increasingly stringent effluent quality standards, denitrification equipment needs to maintain stable operation under complex conditions such as fluctuating influent loads, low temperatures, changes in carbon sources, and filter bed aging. Therefore, timely monitoring of equipment operation and early warning of faults are of great importance.

[0003] Current multi-stage filter bed denitrification equipment typically relies on online data such as effluent quality, flow rate, differential pressure, dissolved oxygen, oxidation-reduction potential, and valve / pump status for operation monitoring. Equipment malfunctions are determined using fixed thresholds, manual experience rules, or centralized monitoring platforms. Some technologies further employ data analysis models to identify anomalies in historical operating data, helping to detect operational problems such as filter bed clogging, insufficient carbon source, abnormal backwashing, or instrument malfunction.

[0004] However, existing technologies mostly rely on overall equipment or effluent indicators as alarm criteria, making it difficult to accurately locate abnormal filter layers and fault propagation paths; fixed thresholds are difficult to adapt to changes in influent load and process status, easily leading to false alarms or missed alarms; centralized analysis is highly dependent on communication links, making it difficult to meet the needs of rapid on-site early warning; at the same time, existing models lack joint constraints on filter layer structure relationships, media flow relationships, and denitrification reaction mechanisms, making it difficult to distinguish between real process faults, sensor drift, and short-term disturbances, resulting in insufficient timeliness, accuracy, and interpretability of fault warnings. Summary of the Invention

[0005] In view of this, embodiments of this application provide a fault early warning method and system for multi-stage filter denitrification equipment based on edge computing, in order to solve the problems of inaccurate fault location, high false alarm and false alarm rates, and delayed on-site early warning in the prior art.

[0006] A first aspect of this application provides a fault early warning method for a multi-stage filter bed denitrification device based on edge computing, comprising: acquiring multi-source operational data of the multi-stage filter bed denitrification device during operation, and performing time-series alignment, data reliability assessment, and windowing processing on the multi-source operational data at an edge computing node to generate filter bed operational status data; constructing a filter bed state map at the edge computing node based on the filter bed structure relationship, media flow relationship, and denitrification reaction relationship of the multi-stage filter bed denitrification device, and mapping the filter bed operational status data to the filter bed nodes and associated edges in the filter bed state map; Based on the filter layer state map, the hydraulic state, denitrification conversion state, operation and control state, and monitoring reliability state of each filter layer node are jointly characterized to generate a filter layer state characterization. According to the filter layer state characterization and the baseline state of the corresponding filter layer nodes, the mechanism residual representing the degree of deviation of filter layer operation is calculated, and based on the spatial distribution and temporal changes of the mechanism residual in the filter layer state map, a fault evolution characterization is generated. The fault evolution characterization is input into the fault early warning model deployed on the edge computing node to determine the fault location, fault type, and fault risk level, and generate fault early warning results for multi-level filter layer denitrification equipment.

[0007] A second aspect of this application provides a fault early warning system for a multi-stage filter bed denitrification device based on edge computing, comprising: an acquisition module, configured to acquire multi-source operational data of the multi-stage filter bed denitrification device during operation, and perform time-series alignment, data reliability assessment, and windowing processing on the multi-source operational data at an edge computing node to generate filter bed operational status data; a construction module, configured to construct a filter bed state map at an edge computing node based on the filter bed structure relationship, media flow relationship, and denitrification reaction relationship of the multi-stage filter bed denitrification device, and map the filter bed operational status data to the filter bed nodes and associated edges in the filter bed state map; and a characterization module. The system is used to jointly characterize the hydraulic state, denitrification conversion state, operation and control state, and monitoring reliability state of each filter layer node based on the filter layer state map, and generate a filter layer state characterization; the calculation module is used to calculate the mechanism residual representing the degree of deviation of filter layer operation based on the filter layer state characterization and the baseline state of the corresponding filter layer node, and generate a fault evolution characterization based on the spatial distribution and temporal changes of the mechanism residual in the filter layer state map; the generation module is used to input the fault evolution characterization into the fault early warning model deployed on the edge computing node, determine the fault location, fault type and fault risk level, and generate fault early warning results for multi-level filter layer denitrification equipment.

[0008] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: By acquiring multi-source operational data of the multi-stage filter bed denitrification equipment during operation, and performing time-series alignment, data reliability assessment, and windowing processing on the edge computing nodes, filter bed operational status data is generated. Based on the filter bed structure relationship, media flow relationship, and denitrification reaction relationship of the multi-stage filter bed denitrification equipment, a filter bed state map is constructed on the edge computing nodes, and the filter bed operational status data is mapped to the filter bed nodes and associated edges in the filter bed state map. Based on the filter bed state map, the hydraulic state, denitrification conversion state, operation control state, and monitoring reliability state of each filter bed node are jointly characterized to generate a filter bed state characterization. Based on the filter bed state characterization and the baseline state of the corresponding filter bed nodes, the mechanistic residual representing the degree of deviation of filter bed operation is calculated, and based on the spatial distribution and temporal changes of the mechanistic residual in the filter bed state map, a fault evolution characterization is generated. The fault evolution characterization is input into a fault early warning model deployed on the edge computing nodes to determine the fault location, fault type, and fault risk level, generating a fault early warning result for the multi-stage filter bed denitrification equipment. This application can improve the accuracy of fault location, reduce the false alarm and missed alarm rates, and shorten the on-site early warning delay. Attached Figure Description

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

[0010] Figure 1 This is a flowchart illustrating the fault early warning method for multi-stage filter denitrification equipment based on edge computing provided in this application embodiment; Figure 2 This is a schematic diagram of the fault early warning system for a multi-stage filter bed denitrification equipment based on edge computing provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] Currently, multi-stage filter bed denitrification equipment typically achieves deep denitrification through processes such as multi-stage filter media retention, biofilm denitrification, carbon source addition and regulation, and backwashing recovery. Existing operational monitoring methods largely rely on online data such as effluent water quality, flow rate, pressure difference, dissolved oxygen, oxidation-reduction potential, and valve and pump status, and determine whether the equipment is abnormal through fixed thresholds, manual experience rules, or centralized monitoring platforms. Although some solutions introduce data analysis models for anomaly identification, they usually still rely on the overall equipment status or effluent indicators as the main basis for judgment.

[0013] However, operational anomalies in multi-stage filter bed denitrification equipment often exhibit characteristics of inter-layer conduction, process state coupling, and gradual evolution. Existing technologies struggle to accurately correlate monitoring data with specific filter bed nodes and fault propagation paths, leading to inaccurate fault location. Fixed thresholds are ill-suited to dynamic changes in influent load, filter bed aging, carbon source variations, and backwash recovery, easily resulting in false alarms or missed alarms. Centralized analysis methods are heavily reliant on communication links and remote computing resources, making it difficult to meet timely on-site fault early warning needs. Therefore, existing technologies suffer from inaccurate fault location, high false alarm / missed alarm rates, and delayed on-site early warning.

[0014] To address the aforementioned issues, this application provides a fault early warning method for multi-stage filter bed denitrification equipment based on edge computing. This method acquires multi-source operational data during the operation of the multi-stage filter bed denitrification equipment at edge computing nodes, and performs time-series alignment, data reliability assessment, and windowing processing on the multi-source operational data to generate filter bed operational status data. A filter bed state map is constructed based on the filter bed structure relationships, media flow relationships, and denitrification reaction relationships, and the filter bed operational status data is mapped to filter bed nodes and associated edges in the filter bed state map. Furthermore, each filter bed node is jointly characterized based on the filter bed state map to generate a filter bed state representation, and the mechanistic residual is calculated based on the filter bed state representation and the baseline state of the corresponding filter bed nodes. Then, a fault evolution representation is generated based on the spatial distribution and temporal changes of the mechanistic residual in the filter bed state map. The fault evolution representation is input into a fault early warning model deployed on edge computing nodes to generate a fault early warning result containing fault location, fault type, and fault risk level.

[0015] Through the above technical solution, this application can perform correlation analysis between multi-source operating data and multi-stage filter layer structure and denitrification process mechanism, so that fault judgment is transformed from single-point threshold alarm to edge-side early warning based on filter layer state spectrum and mechanism residual, thereby improving fault location accuracy, reducing false alarm and missed alarm rate, shortening on-site early warning delay, and enhancing the interpretability and on-site applicability of fault early warning results of multi-stage filter layer denitrification equipment.

[0016] The technical solution of this application will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0017] Figure 1This is a schematic flowchart of a fault early warning method for a multi-stage filter bed denitrification equipment based on edge computing, provided in an embodiment of this application. Figure 1 As shown, the method may specifically include: S101: Acquire multi-source operation data of the multi-stage filter bed denitrification equipment during operation, and perform time-series alignment, data credibility assessment and windowing processing on the multi-source operation data at the edge computing node to generate filter bed operation status data; S102, Based on the filter layer structure relationship, media flow relationship and denitrification reaction relationship of the multi-stage filter layer denitrification equipment, construct the filter layer state map at the edge computing node, and map the filter layer operation status data to the filter layer node and associated edge in the filter layer state map; S103, based on the filter layer state map, jointly characterizes the hydraulic state, denitrification conversion state, operation and control state and monitoring reliability state of each filter layer node to generate a filter layer state characterization. S104. Based on the filter layer state characterization and the baseline state of the corresponding filter layer nodes, calculate the mechanism residuals that characterize the degree of deviation of the filter layer operation, and generate a fault evolution characterization based on the spatial distribution and temporal changes of the mechanism residuals in the filter layer state map. S105 inputs the fault evolution characterization into the fault early warning model deployed on the edge computing node, determines the fault location, fault type and fault risk level, and generates fault early warning results for multi-stage filter layer denitrification equipment.

[0018] In some embodiments, multi-source operational data of the multi-stage filter bed denitrification equipment during operation is acquired, and the multi-source operational data is time-series aligned, data reliability assessed, and windowed at an edge computing node to generate filter bed operational status data, including: Based on the filter layer distribution and control correlation of the multi-stage filter layer denitrification equipment, operational status data, process control data and equipment response data are acquired at the monitoring locations corresponding to each filter layer, and data source identifiers corresponding to filter layer nodes are configured for multi-source operational data. Based on a unified time reference for edge computing nodes, sampling period matching and time synchronization are performed on multi-source running data to generate synchronized running data corresponding to each filter layer node; Based on the data integrity, continuity of change, and consistency of process correlation of the synchronous operation data, the corresponding data credibility is determined, and anomaly removal and missing data correction are performed on the synchronous operation data based on the data credibility. According to the preset operation analysis window, the state features of the corrected synchronous operation data are extracted to generate filter layer operation status data.

[0019] Specifically, when the multi-stage filter bed denitrification equipment is in continuous operation, the edge computing node connects to various sensors, actuator controllers and equipment status acquisition modules through the field communication bus, and establishes data acquisition tasks according to the filter bed distribution and control correlation.

[0020] For example, taking a three-stage denitrification filter as an example, the first, second, and third filter layers are respectively equipped with influent-side monitoring points, effluent-side monitoring points, and differential pressure monitoring points. The carbon source addition pipeline is equipped with flow rate and pump frequency monitoring points, and the backwash pipeline is equipped with air wash intensity, water wash intensity, and backwash duration acquisition points. While collecting operating status data such as nitrate nitrogen, total nitrogen, dissolved oxygen, oxidation-reduction potential, pH, temperature, flow rate, liquid level, and interlayer pressure difference, the edge computing node also collects process control data such as carbon source pump frequency, valve opening, reflux adjustment amount, and backwash start status, as well as equipment response data such as pump current, valve feedback position, and sensor self-test status.

[0021] For each piece of collected data, the edge computing node generates a data source identifier based on the collection location, device number, and data type, and establishes a mapping relationship between the data source identifier and the corresponding filter layer node in the filter layer status map. For example, the nitrate nitrogen data on the effluent side of the second filter layer is marked as the effluent water quality source of the second filter layer node, and the inlet and outlet pressure difference data of the second filter layer is marked as the hydraulic resistance source of the second filter layer node.

[0022] Edge computing nodes establish a unified time base based on a local clock synchronization module and perform sampling period matching on multi-source operating data according to a preset synchronization cycle. For flow rate, differential pressure, and valve opening data collected at the second level, the edge computing node calculates the average value, maximum change amplitude, and change slope within a 10-second synchronization cycle. For nitrate nitrogen, redox potential, and dissolved oxygen data collected at the minute level, the edge computing node maps adjacent sampled values ​​to the corresponding synchronization cycle based on the timestamp and completes the current cycle data through linear interpolation or keep-updated methods.

[0023] In some examples, for discrete control data such as backwash start-up and carbon source pump start-up / shutdown, edge computing nodes write the data into the corresponding synchronization cycle according to the event occurrence time, forming synchronized operation data corresponding to each filter layer node. If the pressure difference of the first filter layer is uploaded at a 5-second interval and nitrate nitrogen is uploaded at a 60-second interval, the edge computing node can use 60 seconds as the basic cycle for operation analysis, and aggregate pressure difference trends, flow fluctuations, and water quality responses within this cycle, thereby keeping data from different sampling frequencies comparable within the same time window.

[0024] After generating synchronized operation data, the edge computing nodes further assess data reliability. Data integrity is used to determine whether necessary data is complete within the current synchronization cycle. If a filter node lacks key water quality data but has valid data from adjacent cycles, it is marked as correctable. Change continuity is used to determine whether data changes conform to equipment operating inertia. If a pressure difference suddenly increases to three times its historical stable value within a synchronization cycle, while flow rate, valve opening, and pump current do not show corresponding changes, the pressure difference is marked as low-reliability data. Process correlation consistency is used to determine whether different data conform to the denitrification process constraints. If the carbon source dosage remains stable, influent nitrate nitrogen does not increase significantly, but effluent nitrate nitrogen increases suddenly at a single point, the reliability of that water quality data is preferentially reduced. For low-reliability data, the edge computing nodes use adjacent filter layer correlation data, historical stable baselines, and current control status to remove anomalies or correct missing data, and retain correction labels to prevent subsequent fault warning models from misjudging instrument noise as actual filter layer faults.

[0025] After correction, the edge computing nodes extract state features according to a preset operation analysis window. The operation analysis window can be set as a combination of short and long windows; short windows are used to extract abrupt changes, while long windows are used to extract gradual changes. For example, within a 5-minute short window, the instantaneous rise rate of differential pressure, flow fluctuation amplitude, and redox potential shift are extracted; within a 60-minute long window, the trends in nitrate nitrogen removal, carbon source utilization, and differential pressure recovery after backwashing are extracted. The edge computing nodes encapsulate the features obtained in each window along with the data source identifier, filter layer node identifier, and data reliability to generate filter layer operation status data, which is then output to the subsequent filter layer state map mapping process.

[0026] Through the above embodiments, operational data with multiple sampling frequencies, multiple sources, and multiple process attributes can be converted into filter layer operational status data with filter layer positioning relationships and confidence indicators at the edge side, reducing the impact of communication delays and data noise on fault identification, and providing a stable and accurate data foundation for subsequent mechanism residual calculation and filter layer-level fault early warning.

[0027] In some embodiments, based on the filter layer structure relationship, media flow relationship, and denitrification reaction relationship of the multi-stage filter layer denitrification device, a filter layer state map is constructed at the edge computing node, including: Obtain structural configuration data and process operation constraints of multi-stage filter bed denitrification equipment, and determine the hierarchical position and interconnection relationship of each filter bed in the treatment process; Based on the direction of medium transmission and the operational control effect between filter layers, associated edges are generated to characterize the state transmission between filter layers; Based on the influence of the denitrification reaction on the state change of the filter layer, state attributes and edge weight parameters are configured for filter layer nodes and associated edges; The filter layer nodes, associated edges, state attributes, and edge weight parameters are assembled into a graph structure to generate the filter layer state graph.

[0028] Specifically, before the multi-stage denitrification equipment is put into operation, the edge computing node first reads the equipment configuration file and on-site process parameters. The equipment configuration file can be entered by the commissioning personnel based on the filter design drawings, pipeline connection diagrams, and control logic tables, while the on-site process parameters can be automatically issued by the equipment control system. Taking a three-stage denitrification filter as an example, the structural configuration data should at least record the number of each filter layer, filter media type, filter layer thickness, support layer position, inlet and outlet water interfaces, differential pressure monitoring points, carbon source addition interfaces, and backwash interfaces; the process operation constraints should at least record the design treatment flow rate, allowable differential pressure range, residence time range, carbon-nitrogen ratio control range, backwash trigger conditions, and effluent total nitrogen control target. Based on the above data, the edge computing node determines that the first filter layer is located at the front-end load reduction position, the second filter layer is located at the middle stable denitrification position, and the third filter layer is located at the end deep denitrification position, and determines that adjacent filter layers are connected sequentially through the water distribution structure, pipeline valves, and flow control units.

[0029] After determining the hierarchical location, the edge computing nodes generate associated edges based on the direction of media transmission within the equipment. For a water flow path from the outlet side of the first filter layer to the inlet side of the second filter layer, a media conduction edge is generated from the first filter layer node to the second filter layer node; for a water flow path from the outlet side of the second filter layer to the inlet side of the third filter layer, a media conduction edge is generated from the second filter layer node to the third filter layer node; for cases where the carbon source dosing unit acts on multiple filter layers, an operational control edge is generated based on the dosing point location, mixing distance, and flow direction delay; for the impact of the backwashing unit on filter layer differential pressure recovery and flux recovery, a backwash recovery edge is generated. Associated edges not only represent pipeline connectivity but also record the direction of media transmission, action delay, controlled object, and range of influence. For example, when the carbon source is added from the front end of the first filter layer and sequentially enters subsequent filter layers with the water flow, the edge computing nodes set the carbon source control effect to a decreasing conduction along the directions of the first, second, and third filter layers to reflect the weakening impact of the carbon source on subsequent filter layers after its consumption in the preceding filter layers.

[0030] Edge computing nodes are further configured with state attributes and edge weight parameters based on the influence of denitrification reactions on filter bed state changes. The state attributes of filter bed nodes can include hydraulic load state, nitrogen removal state, anoxic reaction state, carbon source adaptation state, backwash recovery state, and monitoring reliability state. Edge weight parameters are used to characterize the state transfer intensity between filter beds, and the initial values ​​of the edge computing nodes can be determined based on historical operating data and process constraints.

[0031] For example, in an equipment with a design flow rate of 80 cubic meters per hour, where the first filter layer contributes 45% to nitrate nitrogen removal on average, the second filter layer contributes 35%, and the third filter layer contributes 20%, the nitrogen load transmission weight from the first to the second filter layer is higher than that from the second to the third filter layer. When the pressure difference of the first filter layer approaches the backwash threshold and the flow rate fluctuation increases, the hydraulic disturbance transmission weight from the first filter layer to the downstream filter layer also increases. The weight parameters can be determined by using the equipment design parameters as the basic weights, and then dynamically correcting them by combining the flow rate stability, pressure difference rate of change, and nitrate nitrogen removal gradient within the recent operating window.

[0032] After configuring nodes, associated edges, state attributes, and edge weight parameters, the edge computing node assembles them into a filter layer state map. In the map, each filter layer node corresponds to an updatable state container, and each associated edge corresponds to a computable state transmission channel. Both nodes and associated edges retain timestamps and version numbers for updates during operation based on backwashing, maintenance, or process parameter tuning results. For example, when the filter media in the third filter layer is replaced, the edge computing node updates the filter media attributes and baseline state of the third filter layer node, and simultaneously recalculates the state transmission edge weights from the second to the third filter layer, ensuring the map reflects the latest equipment structure and process status. After the filter layer state map is generated, subsequently collected filter layer operating status data can be directly written to the corresponding node or associated edge according to the data source identifier for filter layer state characterization, mechanism residual calculation, and fault evolution analysis.

[0033] Through the above embodiments, the physical structure, media flow path, and denitrification reaction relationship of the multi-stage filter bed denitrification equipment can be transformed into a calculable graph structure at the edge side, so that the operating data has a clear filter bed positioning basis and conduction analysis basis, improving the accuracy of subsequent fault location, risk propagation judgment, and early warning interpretation.

[0034] In some embodiments, mapping filter layer operating state data to filter layer nodes and associated edges in the filter layer state graph includes: The corresponding target filter layer node is determined based on the data source identifier and collection location identifier of the filter layer operation status data; According to the hierarchical position of the target filter layer node in the filter layer state map, the data representing the filter layer body state is written into the target filter layer node to generate the node state vector. Based on the media transfer direction and operation control function corresponding to the filter layer operation status data, determine the target association edge related to the target filter layer node; The data characterizing the inter-layer state propagation and operational influence relationship is written into the target associated edge to generate edge state parameters; Based on the node state vectors and edge state parameters, the consistency of the filter layer state graph is verified, and the filter layer state graph after data mapping is generated.

[0035] Specifically, after the edge computing node generates the filter layer operation status data, it first reads the data source identifier, acquisition location identifier, and time window identifier carried by each data point and matches them with the node index table in the filter layer status map. Taking a three-stage denitrification filter as an example, if a data source identifier corresponds to the online nitrate nitrogen meter on the effluent side of the second filter layer, and the acquisition location identifier corresponds to the effluent pipe section of the second filter layer, then the edge computing node determines the second filter layer node as the target filter layer node; if a data source identifier corresponds to the differential pressure measurement point between the first and second filter layers, and the acquisition location identifier is located in the connecting pipe section between the two filter layers, then the edge computing node determines the associated edge between the first and second filter layer nodes as the target associated edge. For data where the acquisition location has an interaction relationship with multiple filter layers, such as the carbon source dosing pump frequency, backwash air volume, and total influent flow rate, the edge computing node determines one main-acting filter layer node and at least one affected associated edge based on the dosing point location, backwash branch number, and media transfer direction.

[0036] After the edge computing node determines the target filter layer node, it writes the data characterizing the filter layer's physical state into the node state container according to the target filter layer node's hierarchical position in the filter layer state map. The node state container can be configured with fields according to hydraulic state, denitrification state, regulation state, and monitoring reliability state. Taking the second filter layer node as an example, the inlet and outlet pressure difference, flow rate, and liquid level change are written into the hydraulic state field; the inlet and outlet nitrate nitrogen difference, total nitrogen change, and redox potential are written into the denitrification state field; the carbon source effect, valve opening, and backwash recovery flag are written into the regulation state field; and the data integrity score, sensor self-test results, and correction flag are written into the monitoring reliability state field.

[0037] Furthermore, the edge computing nodes normalize and encode the aforementioned fields to generate a node state vector for the second filter layer node within the current operating analysis window. For example, if the differential pressure of the second filter layer increases from 12 kPa to 17 kPa in a certain window, the nitrate nitrogen removal rate decreases from 42% to 31%, and the redox potential increases from -80 mV to -30 mV, the system encodes these changes as node state vectors representing increased hydraulic resistance, decreased denitrification capacity, and weakened anoxic conditions.

[0038] Regarding the interlayer state transmission and operational influence relationships, edge computing nodes determine the target association edges based on the media transmission direction and operational control effect corresponding to the filter layer operational state data. If the effluent flow fluctuation of the first filter layer is transmitted to the influent side of the second filter layer, the fluctuation data is written into the media transmission edge from the first filter layer node to the second filter layer node; if the carbon source addition point is located at the front end of the first filter layer, and the carbon source acts on each filter layer sequentially with the water flow, the carbon source action amount and transmission delay are written into the carbon source control association edge; if the backwash operation only acts on the third filter layer, the backwash intensity, duration, and differential pressure recovery amplitude are written into the corresponding backwash recovery association edge.

[0039] Edge computing nodes calculate edge state parameters based on data change magnitude, effect delay, and data reliability. These edge state parameters characterize the conduction intensity, influence direction, duration, and attenuation degree. For example, when the pressure difference of the first filter layer increases rapidly and the effluent flow rate fluctuates more, the hydraulic disturbance conduction parameter from the first filter layer to the second filter layer increases. If there is a time delay matching between the increase in downstream nitrate nitrogen and the fluctuation in upstream flow rate, the corresponding nitrogen load conduction parameter increases synchronously.

[0040] After writing the node state vector and edge state parameters, the edge computing node performs consistency verification on the filter layer state map. The verification process includes consistency between the node and the acquisition location, consistency of the state change direction of adjacent nodes, consistency of edge state parameters and media flow direction, and consistency of control actions and state responses. For example, when the nitrate nitrogen in the effluent of the second filter layer increases but the inlet load, carbon source activity, and redox potential of the second filter layer do not change accordingly, the edge computing node reduces the contribution of this water quality data to the node state vector and marks it as data to be verified. When the flow rate of the first filter layer decreases and the flow rate of the second filter layer decreases synchronously, and the direction of the associated edge is consistent with the equipment flow direction, the edge computing node confirms that the state transmission is effective. The verified data is solidified into the filter layer state map after data mapping within the current operating window and submitted to the subsequent joint state characterization and mechanism residual calculation process.

[0041] Through the above embodiments, the filter layer operating status data can be accurately written into the corresponding filter layer nodes and associated edges, transforming multi-source data from isolated monitoring values ​​into spectral status data with hierarchical location, transmission direction and process meaning, thereby improving the accuracy of filter layer status identification and providing a reliable data foundation for subsequent fault source location and fault propagation analysis.

[0042] In some embodiments, based on the filter layer state map, the hydraulic state, denitrification conversion state, operational control state, and monitoring reliability state of each filter layer node are jointly characterized to generate a filter layer state characterization, including: Based on the node state vector of each filter node and the edge state parameters of adjacent associated edges in the filter state diagram, the operational influence range of the corresponding filter node is determined. Based on the scope of operational influence, the operational status data of the filter layer corresponding to the filter layer node is decomposed into hydraulic state components, denitrification and transformation state components, operational control state components, and reliable monitoring state components. Based on the process coupling relationship and data reliability between each state component, the state components are weighted and fused to generate the node joint state of the corresponding filter layer node. Based on the hierarchical transmission relationship in the filter layer state diagram, the joint states of each node are aggregated using a graph structure to generate a filter layer state representation.

[0043] Specifically, after the edge computing nodes complete the writing of the filter layer operation status data, they determine the operational influence range of each filter layer node within the current operation analysis window according to the node state vector of each filter layer node and the edge state parameters of adjacent associated edges in the filter layer state map. Taking a three-stage denitrification filter as an example, the operational influence range of the second filter layer node includes not only the inlet and outlet water quality, interlayer pressure difference, local flow rate, and valve response of the second filter layer itself, but also the media conduction edge from the first filter layer to the second filter layer, the carbon source regulation edge, and the nitrogen load conduction edge from the second filter layer to the third filter layer.

[0044] In some examples, if the effluent flow rate of the first filter layer fluctuates by more than 15% within 10 minutes and the nitrate nitrogen at the inlet of the second filter layer increases accordingly, the edge computing node will include the effluent status of the first filter layer in the upstream influence range of the second filter layer node; if the total nitrogen at the outlet of the third filter layer increases with lag, the edge computing node will synchronously expand the downstream influence range of the second filter layer node on the third filter layer.

[0045] After determining the scope of operational impact, the edge computing nodes decompose the operational status data of the filter layers corresponding to the target filter layer nodes. The hydraulic state component, generated from flow rate, liquid level, interlayer pressure difference, pressure difference change rate, and flux recovery amplitude after backwashing, characterizes the filter layer's water resistance and flux stability. The denitrification conversion state component, generated from nitrate nitrogen removal gradient, total nitrogen change trend, redox potential shift, and temperature correction coefficient, characterizes the filter layer's denitrification reaction capacity. The operational control state component, generated from carbon source dosage, valve opening, reflux regulation, backwashing action, and pump response, characterizes the effect of external control on filter layer operation.

[0046] The reliable state components for monitoring are generated based on data integrity, continuity of change, sensor self-test results, and consistency of process correlation. They are used to characterize whether the current data is suitable for participating in state judgment. For example, if the pressure difference of the second filter layer increases from 13 kPa to 18 kPa within a 30-minute window, the nitrate nitrogen removal rate decreases from 40% to 29%, the redox potential increases from -90 mV to -35 mV, and the carbon source pump frequency does not increase with the load, the edge computing nodes will generate state components for increased hydraulic resistance, weakened denitrification conversion, insufficient regulation compensation, and reliable monitoring data, respectively.

[0047] Subsequently, the edge computing nodes perform weighted fusion based on the process coupling relationship and data reliability between each state component. If the hydraulic state component is abnormal but the denitrification conversion state component remains stable, the system reduces the weight of the filter layer biochemical fault judgment and increases the weight of the hydraulic disturbance judgment; if the denitrification conversion state component is abnormal and the operation control state component shows insufficient carbon source supply, the system increases the weight of carbon source adaptation anomaly; if the monitoring reliability state component is low, the system reduces the contribution of the corresponding water quality or pressure difference data in the fusion process.

[0048] In some examples, during actual calculations, edge computing nodes can determine their basic weights based on historical stable operating conditions, and then adjust them according to the current data reliability and edge state parameters to generate the joint state of each filter layer node. For example, when the reliability of the redox potential data in the second filter layer is 0.9 and the reliability of the differential pressure data is 0.6, the contribution of the denitrification conversion state to the joint state of the nodes is higher than the contribution of a single-point differential pressure change, thereby avoiding misjudgment of filter layer blockage caused by instantaneous noise of the differential pressure gauge.

[0049] Furthermore, after generating the joint states of each node, the edge computing nodes perform graph structure aggregation according to the hierarchical transmission relationships in the filter layer state graph. The aggregation process sequentially processes the first, second, and third filter layers along the media flow direction, while introducing the joint states of adjacent filter layer nodes and the edge state parameters of associated edges. For the second filter layer node, the system integrates the joint state of the second filter layer itself with the output influence of the first filter layer and the feedback risk of the third filter layer to form a graph aggregation state with upstream and downstream correlation significance; for the third filter layer node, the system focuses on integrating the upstream nitrogen load transmission state and the end-effect water risk state. The edge computing nodes uniformly encode the graph aggregation states of all filter layer nodes to form a filter layer state representation that includes the filter layer's own state, neighborhood transmission state, and data reliability state, and send the filter layer state representation to the mechanism residual calculation module.

[0050] Through the above embodiments, the local monitoring data of a single filter layer, the conduction relationship between adjacent filter layers, and the impact of operation and regulation can be uniformly expressed as a filter layer status characterization, so that subsequent fault judgment no longer depends on isolated monitoring values, improves the accuracy of filter layer status identification under complex operating conditions, and enhances the hierarchical positioning capability and process interpretation capability of fault early warning results.

[0051] In some embodiments, graph structure aggregation is performed on the joint states of each node according to the hierarchical transmission relationship in the filter layer state graph to generate a filter layer state representation, including: Based on the hierarchical position of each filter node and the edge weight parameters of the associated edges in the filter state graph, the propagation order and aggregation weight of the joint state of the nodes are determined. According to the transmission order, the joint state of the upstream filter layer node and the adjacent filter layer node is transmitted to the target filter layer node to generate the neighborhood state of the target filter layer node. Based on the aggregation weight, the joint state and neighborhood state of the target filter layer node are fused to generate the graph aggregation state of the target filter layer node. The graph aggregation state of each filter layer node is uniformly encoded to generate a filter layer state representation.

[0052] Specifically, after obtaining the joint state of each filter layer node, the edge computing node first determines the graph structure aggregation order based on the hierarchical position, media flow direction, and edge weight parameters of the associated edges recorded in the filter layer state map. Taking a three-stage denitrification filter as an example, the first filter layer node is located at the front-end load reduction position, the second filter layer node is located at the middle stable denitrification position, and the third filter layer node is located at the end deep denitrification position. The edge computing node takes the first filter layer to the second filter layer and the second filter layer to the third filter layer as the main transmission direction, and at the same time takes the carbon source addition unit to each filter layer node and the backwashing unit to the target filter layer node as the control transmission direction.

[0053] For each associated edge, the edge computing node determines the aggregation weight based on the edge weight parameter, which can be jointly modified by the design flow rate, recent differential pressure change rate, nitrogen load transfer intensity, and data reliability. For example, when the nitrate nitrogen in the effluent of the first filter layer increases and the inlet load of the second filter layer increases simultaneously, the nitrogen load transfer weight from the first filter layer to the second filter layer increases. When the redox potential response of the second filter layer is significant after the carbon source pump frequency changes, while the response of the third filter layer is weak, the aggregation weight of the carbon source control edge on the second filter layer node is higher than that on the third filter layer node.

[0054] Furthermore, after determining the transmission sequence and aggregation weights, the edge computing nodes transmit the joint state of each node according to the operation analysis window. For the second filter layer node, the edge computing node reads the joint state of the nodes of the second filter layer body, and at the same time receives the upstream hydraulic disturbance state, nitrogen load change state and carbon source consumption state transmitted from the first filter layer node along the main transmission direction, and receives the end-effect risk state transmitted from the third filter layer node along the adjacent feedback direction.

[0055] In some examples, for the first filter layer node, the edge computing node mainly receives the total influent load status and the initial carbon source status; for the third filter layer node, the edge computing node mainly receives the output status of the second filter layer and the end-effect constraint status. If the differential pressure of the first filter layer continues to rise by 18% within 30 minutes, the inlet flow fluctuation of the second filter layer increases, and the total nitrogen in the effluent of the third filter layer approaches the control limit, the edge computing node integrates the above-mentioned correlation information into the neighborhood status of the second and third filter layers to reflect the combined impact of upstream disturbances and downstream risk constraints on the target filter layer node.

[0056] Subsequently, the edge computing nodes fuse the joint state and neighborhood state of the target filter layer node based on aggregation weights. This fusion process can be implemented using a lightweight edge-side graph aggregation model. This model retains the basic weights for the target filter layer node's intrinsic state and applies modified weights to the neighborhood state based on the transmission strength of associated edges. Taking the second filter layer node as an example, if the denitrification conversion state of the second filter layer decreases, and the nitrogen load state transmitted from the first filter layer increases while the carbon source regulation edge shows insufficient compensation, the graph aggregation model marks the graph aggregation state of the second filter layer as having enhanced load transmission and insufficient regulation compensation. If the pressure difference of the second filter layer is abnormal but the upstream flow, downstream water quality, and backwash response show no significant changes, the graph aggregation model reduces the support level of the neighborhood state for this anomaly while retaining the instrument verification tendency. In this way, the graph aggregation state of the target filter layer node includes not only its intrinsic operating state but also upstream transmission, neighboring responses, and regulatory influences.

[0057] After generating the aggregated state of each filter layer node, the edge computing node performs a unified encoding on the aggregated state. This unified encoding can be implemented using a fixed-dimensional state vector, which includes the filter layer hierarchy encoding, current hydraulic state encoding, nitrogen removal transformation encoding, regulation response encoding, neighborhood conduction encoding, and monitoring reliability encoding, along with the runtime analysis window number and data version number. For example, the encoding result of the third filter layer node can reflect the terminal deep nitrogen removal capacity, the nitrogen load conduction intensity from the second filter layer, the backwash recovery state, and the proximity of effluent risk. The edge computing node combines the unified encoding of all filter layer nodes according to the hierarchical order of the filter layer state map to generate a filter layer state representation, which is then sent to the mechanism residual calculation module.

[0058] Through the above embodiments, the local state and interlayer conduction state of each filter layer node can be aggregated into a graph structure to form a filter layer state representation that includes filter layer body features, neighborhood influence and process conduction relationship. This improves the completeness of multi-level filter layer operation status judgment, reduces misjudgment caused by single-point data fluctuation, and provides more accurate state input for subsequent fault evolution identification.

[0059] In some embodiments, based on the filter layer state characterization and the baseline state of the corresponding filter layer nodes, a mechanistic residual characterizing the degree of deviation in filter layer operation is calculated, and a fault evolution characterization is generated based on the spatial distribution and temporal variation of the mechanistic residual in the filter layer state map, including: Based on the filter state characterization of each filter node, the baseline state of the corresponding filter node is called to generate the state deviation. Based on the matching relationship between the state deviation and the nitrogen removal process constraints, the mechanism residuals of the corresponding filter layer nodes are calculated. Based on the hierarchical transmission relationship in the filter layer state diagram, spatial correlation analysis is performed on the mechanistic residuals of each filter layer node to generate a residual distribution characterization. Based on the trend of the residual distribution characterization within the continuous operation analysis window, the fault accumulation state and fault propagation state are determined, and a fault evolution characterization is generated.

[0060] Specifically, after obtaining the state characterization of the filter layer, the edge computing node calls the baseline state of the corresponding filter layer node according to the node identifier of each filter layer node in the filter layer state map. The baseline state can be generated by combining stable operating data during the equipment trial operation phase, normal operating data that has been manually confirmed in the recent period, and stable data after backwashing recovery, and is saved separately according to the filter layer level.

[0061] For example, in a three-stage denitrification filter, the baseline state of the first filter layer records the differential pressure range, nitrate nitrogen reduction gradient, and backwash recovery rate under normal flow conditions; the baseline state of the second filter layer records the redox potential range, carbon source response intensity, and nitrogen load carrying capacity during the stable denitrification stage; and the baseline state of the third filter layer records the terminal deep denitrification capacity and effluent risk buffering capacity. Edge computing nodes compare the current filter layer state characterization with the corresponding baseline state to generate a state deviation, which includes the magnitude deviation, trend deviation, and persistence deviation of the current state relative to the baseline state.

[0062] After generating the state deviation, the edge computing nodes calculate the theoretical residuals based on the denitrification process constraints. Denitrification process constraints can include hydraulic resistance constraints, nitrogen conversion constraints, carbon source adaptation constraints, and backwash recovery constraints. Taking the second filter layer as an example, if the baseline pressure difference of this filter layer under stable operating conditions is 12 kPa to 15 kPa, the baseline nitrate nitrogen removal rate is 38% to 45%, and the baseline oxidation-reduction potential is -120 mV to -70 mV, and in the current operating window the pressure difference rises to 19 kPa, the nitrate nitrogen removal rate drops to 28%, the oxidation-reduction potential rises to -35 mV, and the carbon source pump frequency does not increase with the increase of influent nitrogen load, then the edge computing nodes generate hydraulic deviations, denitrification conversion deviations, and carbon source adaptation deviations respectively, and calculate the theoretical residuals based on the process constraint matching results. If the pressure difference deviates from the flow rate change but the water quality change lags behind, the mechanism residual is more likely to characterize hydraulic blockage; if the pressure difference is stable but the nitrate nitrogen removal rate decreases and the redox potential increases, the mechanism residual is more likely to characterize insufficient carbon source or weakened anoxic reaction.

[0063] The edge computing nodes further perform spatial correlation analysis on the mechanistic residuals of each filter layer node according to the hierarchical transmission relationship in the filter layer state map. During the analysis, the edge computing nodes compare the residual types, residual occurrence times, and residual intensity changes of adjacent filter layer nodes along the medium flow direction. For example, if the hydraulic residual of the first filter layer continuously increases, followed by increased fluctuations in the inlet flow of the second filter layer, and the total nitrogen in the effluent of the third filter layer approaches the control limit, then the edge computing nodes determine that the residuals are transmitted from the first filter layer to the subsequent filter layers, and generate a residual distribution characterization with the first filter layer as the residual source point and the second and third filter layers as the influence range. If the pressure difference residuals of the three filter layers are not significant, but the oxidation-reduction potentials at each level increase synchronously and are accompanied by an increase in effluent nitrate nitrogen, then the edge computing nodes determine the residual distribution characterization as insufficient global anoxic response or insufficient carbon source regulation, rather than single filter layer blockage.

[0064] During continuous operation analysis, edge computing nodes track the trend of residual distribution. If the mechanistic residual of the same filter node continuously increases across multiple windows without being eliminated by backwashing or control actions, it is identified as a fault accumulation state. If the residual appears sequentially from upstream filter nodes to downstream filter nodes, and the time interval matches the media residence time, it is identified as a fault propagation state. Edge computing nodes encode the fault accumulation state, fault propagation state, residual source point, impact range, and duration to generate a fault evolution characterization, which is then output to the fault early warning model for subsequent fault location, fault type, and fault risk level determination.

[0065] Through the above embodiments, the current filter layer state can be jointly compared with the individualized baseline state and the denitrification process constraints to identify the underlying mechanism of filter layer operation deviation, and the fault evolution process can be characterized based on the residual spatial distribution and temporal changes, thereby improving the accuracy of early fault identification, the reliability of fault propagation judgment, and the interpretability of early warning results.

[0066] In some embodiments, spatial correlation analysis is performed on the mechanistic residuals of each filter layer node according to the hierarchical transmission relationship in the filter layer state map to generate a residual distribution characterization, including: Based on the mechanistic residuals and corresponding residual types of each filter layer node, determine the residual nodes to be analyzed; Based on the hierarchical transmission relationship and the edge weight parameters of the associated edges in the filter layer state diagram, the residual association path between the residual nodes to be analyzed is determined; Directional propagation matching of residuals along residual correlation paths is performed to determine residual source points, residual diffusion ranges, and residual decay states; A residual distribution characterization is generated based on the residual source point, residual diffusion range, and residual decay state.

[0067] Specifically, after calculating the mechanistic residuals for each filter layer node, the edge computing node first filters the residual nodes to be analyzed based on the residual amplitude, residual type, and data reliability. Mechanism residuals can include hydraulic resistance residuals, denitrification conversion residuals, carbon source adaptation residuals, backwash recovery residuals, and monitoring consistency residuals. Taking a three-stage denitrification filter as an example, if the hydraulic resistance residual of the first filter layer is higher than the dynamic threshold for three consecutive operating analysis windows, and the pressure difference data reliability reaches 0.85, the edge computing node marks the first filter layer node as the residual node to be analyzed.

[0068] If the residual nitrate nitrogen removal in the second filter layer increases but the corresponding water quality data confidence level is below 0.5, the edge computing node first marks this node as a residual node to be reviewed, reducing the weight of this residual in the spatial correlation analysis. For cases where multiple residual nodes appear within the same operating window, the edge computing node determines the analysis priority based on the residual duration and the degree of impact of the residual on the effluent risk nodes.

[0069] After identifying the residual nodes to be analyzed, the edge computing node determines the residual correlation paths based on the hierarchical transmission relationships and edge weight parameters of the associated edges in the filter layer state map. Residual correlation paths can be established along the direction of media flow, the direction of regulation, and the direction of backwash recovery. If the hydraulic transmission edge weight between the first and second filter layers is 0.78, the nitrogen load transmission edge weight between the second and third filter layers is 0.62, and the carbon source regulation edge has an influence weight of 0.71 on the second filter layer, then the edge computing node prioritizes analyzing the residual correlation paths between the first and second filter layers, between the second and third filter layers, and between the carbon source unit and the second filter layer. Paths with edge weights lower than the preset correlation threshold can be retained as candidate paths but not participated in the main path matching to avoid misjudging fault sources due to weakly correlated nodes.

[0070] Furthermore, after the residual correlation path is determined, the edge computing node performs directional transmission matching of the residuals of each mechanism along the residual correlation path. The matching content includes the order of occurrence of residuals, the consistency of residual types, the relationship of residual intensity changes, and the matching relationship of transmission time. For example, if the hydraulic resistance residual of the first filter layer increases in the first window, the inlet flow fluctuation residual of the second filter layer increases in the second window, and the nitrogen conversion residual of the effluent of the third filter layer increases in the fourth window, and the time interval of each window is basically consistent with the hydraulic residence time of the equipment, then the edge computing node judges that the residual is transmitted along the direction of the first filter layer, the second filter layer, and the third filter layer, and determines the first filter layer node as the residual source point. If the redox potential residual increases simultaneously in each filter layer, but the pressure difference residual does not change synchronously, the edge computing node judges that the residual source point is more likely to be located in the carbon source control unit or the influent load end, rather than a single filter layer blockage.

[0071] The edge computing nodes further determine the residual diffusion range and residual decay state based on the intensity changes of the residuals along the associated paths. If the hydraulic resistance residual of the first filter layer is 0.92, the corresponding residual of the second filter layer is 0.54, and the corresponding residual of the third filter layer is 0.21, the system determines that the residual diffusion range covers the first to the third filter layers and exhibits a progressive decay state. If the downstream residual intensity does not decay but instead increases, accompanied by an increase in the residual of the effluent risk node, the system expands the diffusion range to the end effluent risk node and increases the fault propagation risk indicator. For paths where the residual decreases rapidly after backwashing, the edge computing nodes mark the residual decay state as recoverable; for paths where the residual continues to increase after backwashing, the edge computing nodes mark the residual decay state as continuously spreading.

[0072] Finally, the edge computing nodes encode the residual source point, residual diffusion range, residual decay state, residual type, and path reliability to generate a residual distribution representation. Through the above embodiments, the mechanistic residuals dispersed across different filter layer nodes can be organized into a spatial distribution with variations in direction, range, and intensity. This improves the accuracy of fault source point location, reduces misjudgments caused by local noise, and provides a reliable basis for subsequent identification of fault accumulation and propagation states.

[0073] In some embodiments, the fault evolution characterization is input into a fault early warning model deployed on an edge computing node to determine the fault location, fault type, and fault risk level, generating fault early warning results for multi-stage filter bed denitrification equipment, including: Edge-side feature compression is performed on the fault evolution representation to generate an early warning input representation corresponding to the filter layer state map; The warning input is used to characterize the fault warning model, and the candidate fault location is determined based on the anomaly contribution of each filter layer node and the propagation contribution of the associated edge. The candidate fault type is determined based on the mechanism residual type, fault accumulation state, and fault propagation state corresponding to the candidate fault location. Based on the impact of candidate fault types on water effluent risk nodes and the confidence level of early warning, the fault risk level is determined. The fault location, fault type, fault risk level, and corresponding early warning basis are associated and encapsulated to generate fault early warning results for multi-stage filter denitrification equipment.

[0074] Specifically, after obtaining the fault evolution representation, the edge computing node first performs edge-side feature compression to adapt to the computing power and storage conditions of the field gateway. The fault evolution representation typically includes the residual source point, residual diffusion range, residual decay state, fault accumulation state, fault propagation state, and path reliability of each filter layer node. The edge computing node converts the above information into a fixed-length early warning input representation according to the node order of the filter layer state map.

[0075] For example, taking a three-stage denitrification filter as an example, the first, second, and third filter layers each correspond to a set of node codes. Each set of node codes includes the current residual intensity, the number of residual duration windows, the neighborhood conduction intensity, and the correlation degree of effluent risk. The correlation edge codes include the upstream-to-downstream conduction weight, the conduction delay matching degree, and the edge state credibility. For data with low contribution or low credibility, the edge computing nodes are compressed and dimensionality reduced, retaining only the residual peak value, trend slope, and path identifier, thereby reducing the input size of the fault warning model.

[0076] After completing the early warning input representation, the edge computing nodes input the early warning input representation into the fault early warning model. The fault early warning model can be deployed in the field edge gateway using a combination of a lightweight graph attention network and a rule verification module. The graph attention network calculates the node anomaly contribution based on the anomaly intensity, residual duration, and data reliability of each filter layer node, and calculates the propagation contribution based on the edge weight parameters of the associated edges, the residual directionality matching results, and the propagation delay.

[0077] For example, in a certain operational analysis, the hydraulic resistance residual of the first filter layer increased for four consecutive windows, the hydraulic disturbance transmission contribution from the first to the second filter layer reached 0.76, the inlet flow fluctuation of the second filter layer appeared in the next window, and the total nitrogen risk in the effluent of the third filter layer increased simultaneously. The model rated the abnormal contribution of the first filter layer node as the highest and identified the first filter layer as the candidate fault location. If there is no obvious abnormality in the pressure difference of each filter layer, but the contribution of the carbon source regulation edge is high, and the denitrification conversion residual of multiple filter layers increases simultaneously, then the model points the candidate fault location to the carbon source addition-related location or the location affected by the influent load.

[0078] Furthermore, after determining the candidate fault locations, the edge computing nodes determine the candidate fault types based on the mechanistic residual type, fault accumulation state, and fault propagation state corresponding to the candidate fault locations. If the candidate fault location is in the first filter layer, the mechanistic residual is mainly hydraulic resistance residual, and the residual decrease after backwashing is insufficient, the system determines the candidate fault type as filter layer blockage or abnormal backwash recovery. If the candidate fault location is in the second filter layer, the mechanistic residual is mainly denitrification conversion residual, the redox potential is too high, and the carbon source regulation state is not compensated in time, the system determines the candidate fault type as insufficient carbon source adaptation or decreased denitrification activity. If the candidate fault location corresponds to a single sensor node, and the states of adjacent filter layers do not show consistent changes, the system determines the candidate fault type as inaccurate monitoring data or instrument drift. In the case of multiple candidate fault types coexisting, the edge computing nodes sort them according to the mechanistic residual matching degree, the similarity of historical fault samples, and the reliability of field data, and output the main fault type and alternative fault types.

[0079] Subsequently, the edge computing nodes determine the fault risk level based on the impact of candidate fault types on the effluent risk nodes and the early warning confidence level. If the fault only manifests as a slight increase in local residuals and does not spread downstream, the risk level is set to the alert level; if the residuals accumulate continuously and have affected adjacent filter layers, the risk level is set to the attention level; if the predicted value of the effluent risk node is close to the control limit and the fault propagation state matches the media residence time, the risk level is set to the handling level; if the effluent risk node has shown a high-risk trend and is accompanied by abnormalities in key execution units, the risk level is set to the shutdown protection level. The edge computing nodes associate and encapsulate the fault location, fault type, fault risk level, residual change trend, contribution of related edge propagation, and data confidence to form a fault early warning result for multi-level filter layer denitrification equipment, and output it through the on-site human-machine interface, operation and maintenance platform, or control system interface.

[0080] Through the above embodiments, the fault evolution representation can be compressed into an early warning input representation suitable for real-time reasoning on the edge side, and the fault location, type identification and risk classification can be completed by combining the node anomaly contribution and conduction contribution, thereby improving the real-time performance, accuracy and interpretability of fault early warning for multi-stage filter denitrification equipment.

[0081] In some embodiments, the warning input is represented by the input fault warning model, and the candidate fault location is determined based on the anomaly contribution of each filter layer node and the propagation contribution of the associated edge, including: The warning input is used to characterize the fault warning model, and attention weighting is applied to the state deviation intensity and residual persistence of each filter layer node to generate the node anomaly contribution. Based on the edge weight parameters of the associated edges and the residual propagation matching results, the propagation contribution of the associated edges is generated. Based on the causal matching relationship between node anomaly contribution and conduction contribution, the anomaly source points in the filter layer state map are sorted. The abnormal source points that meet the preset location conditions in the sorting results are identified as candidate fault locations.

[0082] Specifically, after obtaining the early warning input representation, the edge computing node inputs the early warning input representation into the fault early warning model deployed in the field gateway. The fault early warning model may include a node attention calculation layer, an associated edge propagation calculation layer, and an anomaly source ranking layer. The node attention calculation layer first reads the state deviation intensity, residual persistence, data credibility, and effluent risk correlation of each filter layer node, and assigns dynamic attention weights to different dimensions.

[0083] For example, taking a three-stage denitrification filter as an example, if the hydraulic resistance residual of the first filter layer increases from 0.45 to 0.88 within four consecutive operating analysis windows, and the pressure difference data reliability is 0.91, the node attention calculation layer increases the weight of the hydraulic resistance residual and the number of continuous windows; if the denitrification conversion residual of the second filter layer only increases within a single window, and the nitrate nitrogen data has a correction indicator, then the abnormal weight of that node is reduced to avoid single-point water quality fluctuations being misjudged as fault sources.

[0084] In the calculation of node anomaly contribution, edge computing nodes can jointly encode the intensity of state deviation and the persistence of residuals. The intensity of state deviation reflects the magnitude of the current filter layer node's deviation from the baseline state, while the persistence of residuals reflects whether the anomaly has a cumulative trend. For the first filter layer node, if the differential pressure residual continues to increase and the backwash recovery residual increases synchronously, the model superimposes the hydraulic state deviation and backwash recovery deviation into a higher node anomaly contribution. For the third filter layer node, if the total nitrogen in the effluent is close to the control limit but the upstream nitrogen load transfer is strong, the model sets the anomaly contribution of the third filter layer node to medium and retains it as an affected node rather than a source node. In this way, the fault early warning model can distinguish between actively anomalous nodes and passively responding nodes.

[0085] Meanwhile, the associated edge transmission calculation layer generates transmission contribution based on the edge weight parameters of the associated edges in the filter layer state map and the residual transmission matching results. The edge weight parameters reflect the intensity of medium flow, nitrogen load transmission intensity, and control effect between filter layers, while the residual transmission matching results reflect whether the order of residual occurrence is consistent with the transmission path. For example, if the hydraulic resistance residual of the first filter layer increases in the first window, the inlet flow fluctuation of the second filter layer increases in the second window, and the denitrification conversion residual of the third filter layer increases in the fourth window, with the time interval consistent with the hydraulic residence time of the equipment, the transmission contribution of the associated edges from the first filter layer to the second filter layer and from the second filter layer to the third filter layer is increased. If the residual of the second filter layer appears first and the residual of the first filter layer appears later, the transmission contribution decreases, preventing downstream local anomalies from being traced back to upstream faults.

[0086] The anomaly source ranking layer ranks the anomaly sources in the filter layer state map based on the causal matching relationship between the node's anomaly contribution and its conduction contribution. The causal matching relationship can be determined based on the order of anomaly occurrence, consistency of residual type, consistency of conduction direction, and consistency of influence range. If a node has a high node anomaly contribution, and the associated edge originating from that node has a high conduction contribution, and the residuals of downstream nodes can be explained by the residuals of that node, then that node is ranked higher. Taking a warning as an example, the anomaly contribution of the first filter layer node is 0.87, the conduction contribution from the first filter layer to the second filter layer is 0.74, and the conduction contribution from the second filter layer to the third filter layer is 0.58, and the downstream residuals decrease progressively. The model ranks the first filter layer as the first anomaly source. If the contribution of the carbon source regulation edge is higher than the contribution of any filter layer node, then the carbon source addition associated location is included in the ranking as an anomaly source.

[0087] The edge computing nodes further compare the ranking results with preset location conditions. These preset location conditions may include contribution thresholds, number of consecutive windows, path reliability, and correlation with effluent risk. If a top-ranked anomaly source meets conditions such as a node anomaly contribution greater than 0.7, a conduction contribution greater than 0.6, at least three consecutive anomaly windows, and a path reliability greater than 0.8, the edge computing node identifies that anomaly source as a candidate fault location. If multiple anomaly sources simultaneously meet these conditions, the primary candidate fault location and secondary candidate fault locations are output based on the correlation with effluent risk and fault handling priority.

[0088] Through the above embodiments, attention-weighted methods can be used to highlight filter layer anomalies that are persistent and have clear process correlations. Combined with the contribution of correlation edge transmission, the causal source of the fault in the filter layer state map can be identified, thereby improving the accuracy of candidate fault location determination and reducing the risk of mislocation caused by passive response nodes or low-reliability monitoring data.

[0089] The following are system embodiments of this application, which can be used to execute the method embodiments of this application. For details not disclosed in the system embodiments of this application, please refer to the method embodiments of this application.

[0090] Figure 2 This is a schematic diagram of the fault early warning system for a multi-stage filter bed denitrification equipment based on edge computing provided in an embodiment of this application. Figure 2 As shown, the system includes: The acquisition module 201 is used to acquire multi-source operating data of the multi-stage filter bed denitrification equipment during operation, and perform time-series alignment, data credibility assessment and windowing processing on the multi-source operating data at the edge computing node to generate filter bed operating status data. Module 202 is used to construct a filter layer state map at the edge computing node based on the filter layer structure relationship, media flow relationship and denitrification reaction relationship of the multi-stage filter layer denitrification equipment, and to map the filter layer operation status data to the filter layer nodes and associated edges in the filter layer state map; The characterization module 203 is used to jointly characterize the hydraulic state, denitrification conversion state, operation and control state and monitoring reliability state of each filter layer node based on the filter layer state map, and generate the filter layer state characterization. The calculation module 204 is used to calculate the mechanism residuals that characterize the degree of deviation of the filter layer operation based on the filter layer state characterization and the baseline state of the corresponding filter layer nodes, and to generate a fault evolution characterization based on the spatial distribution and temporal changes of the mechanism residuals in the filter layer state map. The generation module 205 is used to input the fault evolution characterization into the fault early warning model deployed on the edge computing node, determine the fault location, fault type and fault risk level, and generate fault early warning results for the multi-stage filter layer denitrification equipment.

[0091] In some embodiments, Figure 2 The acquisition module 201 acquires operating status data, process control data, and equipment response data at the monitoring locations corresponding to each filter layer according to the filter layer distribution and control correlation of the multi-stage filter layer denitrification equipment, and configures data source identifiers corresponding to filter layer nodes for the multi-source operating data; based on the unified time reference of the edge computing nodes, it performs sampling period matching and time sequence synchronization of the multi-source operating data to generate synchronous operating data corresponding to each filter layer node; based on the data integrity, change continuity, and process correlation consistency of the synchronous operating data, it determines the corresponding data credibility, and performs anomaly removal and missing data correction on the synchronous operating data based on the data credibility; according to the preset operation analysis window, it extracts state features from the corrected synchronous operating data to generate filter layer operating status data.

[0092] In some embodiments, Figure 2 The construction module 202 acquires the structural configuration data and process operation constraints of the multi-stage filter bed denitrification equipment, determines the hierarchical position and interconnection relationship of each filter bed in the treatment process; generates associated edges to characterize the state conduction between filter beds based on the transmission direction and operation control effect of the medium between each filter bed; configures state attributes and edge weight parameters for filter bed nodes and associated edges based on the influence relationship of denitrification reaction on filter bed state changes; and assembles the filter bed nodes, associated edges, state attributes and edge weight parameters into a graph structure to generate a filter bed state map.

[0093] In some embodiments, Figure 2The construction module 202 determines the corresponding target filter layer node based on the data source identifier and acquisition location identifier of the filter layer operation status data; according to the hierarchical position of the target filter layer node in the filter layer state map, it writes the data representing the filter layer body state into the target filter layer node to generate a node state vector; based on the medium transmission direction and operation control effect corresponding to the filter layer operation status data, it determines the target association edge related to the target filter layer node; it writes the data representing the inter-layer state transmission and operation influence relationship into the target association edge to generate edge state parameters; based on the node state vector and edge state parameters, it performs consistency verification on the filter layer state map to generate the filter layer state map after data mapping is completed.

[0094] In some embodiments, Figure 2 The characterization module 203 determines the operational influence range of the corresponding filter layer node based on the node state vector of each filter layer node and the edge state parameters of adjacent associated edges in the filter layer state map. Based on the operational influence range, the module decomposes the filter layer operational state data corresponding to the filter layer node to generate hydraulic state components, denitrification conversion state components, operational control state components, and monitoring reliability state components. Based on the process coupling relationship and data reliability between each state component, the module performs weighted fusion of each state component to generate the node joint state of the corresponding filter layer node. According to the hierarchical transmission relationship in the filter layer state map, the module performs graph structure aggregation of each node joint state to generate the filter layer state characterization.

[0095] In some embodiments, Figure 2 The representation module 203 determines the propagation order and aggregation weight of the joint state of nodes based on the hierarchical position of each filter layer node and the edge weight parameters of the associated edges in the filter layer state graph. According to the propagation order, the joint state of the upstream filter layer node and the adjacent filter layer node is transmitted to the target filter layer node to generate the neighborhood state of the target filter layer node. Based on the aggregation weight, the joint state of the target filter layer node and the neighborhood state are fused to generate the graph aggregation state of the target filter layer node. The graph aggregation state of each filter layer node is uniformly encoded to generate the filter layer state representation.

[0096] In some embodiments, Figure 2 The calculation module 204, based on the filter layer state characterization of each filter layer node, calls the baseline state of the corresponding filter layer node to generate the state deviation; based on the matching relationship between the state deviation and the denitrification process constraints, it calculates the mechanistic residual of the corresponding filter layer node; according to the hierarchical transmission relationship in the filter layer state map, it performs spatial correlation analysis on the mechanistic residual of each filter layer node to generate the residual distribution characterization; based on the changing trend of the residual distribution characterization in the continuous operation analysis window, it determines the fault accumulation state and fault transmission state, and generates the fault evolution characterization.

[0097] In some embodiments, Figure 2The calculation module 204 determines the residual nodes to be analyzed based on the mechanistic residuals and corresponding residual types of each filter layer node; it determines the residual association path between the residual nodes to be analyzed based on the hierarchical transmission relationship and the edge weight parameters of the associated edges in the filter layer state map; it performs directional transmission matching on each mechanistic residual along the residual association path to determine the residual source point, residual diffusion range and residual decay state; and it generates a residual distribution characterization based on the residual source point, residual diffusion range and residual decay state.

[0098] In some embodiments, Figure 2 The generation module 205 performs edge-side feature compression on the fault evolution representation to generate an early warning input representation corresponding to the filter layer state map. The early warning input representation is input into the fault early warning model, and candidate fault locations are determined based on the abnormal contribution of each filter layer node and the transmission contribution of the associated edges. The candidate fault type is determined according to the mechanism residual type, fault accumulation state and fault transmission state corresponding to the candidate fault location. The fault risk level is determined based on the degree of influence of the candidate fault type on the effluent risk node and the early warning confidence level. The fault location, fault type, fault risk level and corresponding early warning basis are associated and encapsulated to generate fault early warning results for multi-stage filter layer denitrification equipment.

[0099] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0100] Figure 3 This is a schematic diagram of the electronic device 3 provided in an embodiment of this application. Figure 3 As shown, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various system embodiments described above.

[0101] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that... Figure 3 This is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or different components.

[0102] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0103] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or memory of the electronic device 3. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the electronic device 3. The memory 302 can also include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device.

[0104] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0105] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which may be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0106] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A fault early warning method for multi-stage filter bed denitrification equipment based on edge computing, characterized in that, include: Acquire multi-source operational data of the multi-stage filter bed denitrification equipment during operation, and perform time-series alignment, data credibility assessment and windowing processing on the multi-source operational data at the edge computing node to generate filter bed operational status data; Based on the filter layer structure relationship, media flow relationship and denitrification reaction relationship of the multi-stage filter layer denitrification equipment, a filter layer state map is constructed at the edge computing node, and the filter layer operation status data is mapped to the filter layer node and associated edge in the filter layer state map; Based on the filter layer state map, the hydraulic state, denitrification conversion state, operation and control state and monitoring reliability state of each filter layer node are jointly characterized to generate the filter layer state characterization. Based on the state characterization of the filter layer and the baseline state of the corresponding filter layer nodes, the mechanism residuals characterizing the degree of deviation of the filter layer operation are calculated, and the fault evolution characterization is generated based on the spatial distribution and temporal changes of the mechanism residuals in the filter layer state map. The fault evolution characterization is input into the fault early warning model deployed on edge computing nodes to determine the fault location, fault type and fault risk level, and generate fault early warning results for multi-stage filter denitrification equipment.

2. The method according to claim 1, characterized in that, The process involves acquiring multi-source operational data of the multi-stage filter bed denitrification equipment during operation, and performing time-series alignment, data reliability assessment, and windowing processing on the multi-source operational data at an edge computing node to generate filter bed operational status data, including: Based on the filter layer distribution and control correlation of the multi-stage filter layer denitrification equipment, operational status data, process control data and equipment response data are acquired at the monitoring locations corresponding to each filter layer, and data source identifiers corresponding to filter layer nodes are configured for multi-source operational data. Based on a unified time reference for edge computing nodes, sampling period matching and time synchronization are performed on multi-source running data to generate synchronized running data corresponding to each filter layer node; Based on the data integrity, continuity of change, and consistency of process correlation of the synchronous operation data, the corresponding data credibility is determined, and anomaly removal and missing data correction are performed on the synchronous operation data based on the data credibility. According to the preset operation analysis window, the state features of the corrected synchronous operation data are extracted to generate filter layer operation status data.

3. The method according to claim 1, characterized in that, The process of constructing a filter layer state map at the edge computing nodes based on the filter layer structure relationship, media flow relationship, and denitrification reaction relationship of the multi-stage filter layer denitrification equipment includes: Obtain structural configuration data and process operation constraints of multi-stage filter bed denitrification equipment, and determine the hierarchical position and interconnection relationship of each filter bed in the treatment process; Based on the direction of medium transmission and the operational control effect between filter layers, associated edges are generated to characterize the state transmission between filter layers; Based on the influence of the denitrification reaction on the state change of the filter layer, state attributes and edge weight parameters are configured for filter layer nodes and associated edges; The filter layer nodes, associated edges, state attributes, and edge weight parameters are assembled into a graph structure to generate the filter layer state graph.

4. The method according to claim 3, characterized in that, The process of mapping filter layer operating status data to filter layer nodes and associated edges in the filter layer status graph includes: The corresponding target filter layer node is determined based on the data source identifier and collection location identifier of the filter layer operation status data; According to the hierarchical position of the target filter layer node in the filter layer state map, the data representing the filter layer body state is written into the target filter layer node to generate the node state vector. Based on the media transfer direction and operation control function corresponding to the filter layer operation status data, determine the target association edge related to the target filter layer node; The data characterizing the inter-layer state propagation and operational influence relationship is written into the target associated edge to generate edge state parameters; Based on the node state vectors and edge state parameters, the consistency of the filter layer state graph is verified, and the filter layer state graph after data mapping is generated.

5. The method according to claim 1, characterized in that, The method, based on the filter layer state map, jointly characterizes the hydraulic state, denitrification conversion state, operational control state, and reliable monitoring state of each filter layer node to generate a filter layer state characterization, including: Based on the node state vector of each filter node and the edge state parameters of adjacent associated edges in the filter state diagram, the operational influence range of the corresponding filter node is determined. Based on the scope of operational influence, the operational status data of the filter layer corresponding to the filter layer node is decomposed into hydraulic state components, denitrification and transformation state components, operational control state components, and reliable monitoring state components. Based on the process coupling relationship and data reliability between each state component, the state components are weighted and fused to generate the node joint state of the corresponding filter layer node. Based on the hierarchical transmission relationship in the filter layer state diagram, the joint states of each node are aggregated using a graph structure to generate a filter layer state representation.

6. The method according to claim 5, characterized in that, The step of performing graph structure aggregation on the joint states of each node according to the hierarchical transmission relationship in the filter layer state graph to generate a filter layer state representation includes: Based on the hierarchical position of each filter node and the edge weight parameters of the associated edges in the filter state graph, the propagation order and aggregation weight of the joint state of the nodes are determined. According to the transmission order, the joint state of the upstream filter layer node and the adjacent filter layer node is transmitted to the target filter layer node to generate the neighborhood state of the target filter layer node. Based on the aggregation weight, the joint state and neighborhood state of the target filter layer node are fused to generate the graph aggregation state of the target filter layer node. The graph aggregation state of each filter layer node is uniformly encoded to generate a filter layer state representation.

7. The method according to claim 1, characterized in that, The step involves calculating the mechanistic residuals representing the degree of deviation in filter layer operation based on the filter layer state characterization and the baseline state of the corresponding filter layer nodes. Then, based on the spatial distribution and temporal changes of the mechanistic residuals in the filter layer state map, a fault evolution characterization is generated, including: Based on the filter state characterization of each filter node, the baseline state of the corresponding filter node is called to generate the state deviation. Based on the matching relationship between the state deviation and the nitrogen removal process constraints, the mechanism residuals of the corresponding filter layer nodes are calculated. Based on the hierarchical transmission relationship in the filter layer state diagram, spatial correlation analysis is performed on the mechanistic residuals of each filter layer node to generate a residual distribution characterization. Based on the trend of the residual distribution characterization within the continuous operation analysis window, the fault accumulation state and fault propagation state are determined, and a fault evolution characterization is generated.

8. The method according to claim 7, characterized in that, The step involves performing spatial correlation analysis on the mechanistic residuals of each filter layer node according to the hierarchical transmission relationship in the filter layer state diagram, generating a residual distribution characterization, including: Based on the mechanistic residuals and corresponding residual types of each filter layer node, determine the residual nodes to be analyzed; Based on the hierarchical transmission relationship and the edge weight parameters of the associated edges in the filter layer state diagram, the residual association path between the residual nodes to be analyzed is determined; Directional propagation matching of residuals along residual correlation paths is performed to determine residual source points, residual diffusion ranges, and residual decay states; A residual distribution characterization is generated based on the residual source point, residual diffusion range, and residual decay state.

9. The method according to claim 1, characterized in that, The fault evolution characterization is input into the fault early warning model deployed on edge computing nodes to determine the fault location, fault type, and fault risk level, and generate fault early warning results for multi-stage filter layer denitrification equipment, including: Edge-side feature compression is performed on the fault evolution representation to generate an early warning input representation corresponding to the filter layer state map; The warning input is used to characterize the fault warning model, and the candidate fault location is determined based on the anomaly contribution of each filter layer node and the propagation contribution of the associated edge. The candidate fault type is determined based on the mechanism residual type, fault accumulation state, and fault propagation state corresponding to the candidate fault location. Based on the impact of candidate fault types on water effluent risk nodes and the confidence level of early warning, the fault risk level is determined. The fault location, fault type, fault risk level, and corresponding early warning basis are associated and encapsulated to generate fault early warning results for multi-stage filter denitrification equipment.

10. A fault early warning system for a multi-stage filter bed denitrification equipment based on edge computing, characterized in that, include: The acquisition module is used to acquire multi-source operating data of the multi-stage filter bed denitrification equipment during operation, and perform time-series alignment, data credibility assessment and windowing processing on the multi-source operating data at the edge computing node to generate filter bed operating status data. The module is used to construct a filter layer state map at the edge computing node based on the filter layer structure relationship, media flow relationship and denitrification reaction relationship of the multi-stage filter layer denitrification equipment, and to map the filter layer operation status data to the filter layer nodes and associated edges in the filter layer state map; The characterization module is used to jointly characterize the hydraulic state, denitrification conversion state, operation and control state and monitoring reliability state of each filter layer node based on the filter layer state map, and generate the filter layer state characterization. The calculation module is used to calculate the mechanism residuals that characterize the degree of deviation of the filter layer operation based on the filter layer state characterization and the baseline state of the corresponding filter layer nodes, and to generate a fault evolution characterization based on the spatial distribution and temporal changes of the mechanism residuals in the filter layer state map. The generation module is used to input the fault evolution characterization into the fault early warning model deployed on the edge computing node, determine the fault location, fault type and fault risk level, and generate fault early warning results for multi-stage filter denitrification equipment.

Images