Hydropower station intelligent comprehensive alarm and operation and maintenance response method based on execution engine

Abstract

The present application belongs to the field of automatic monitoring and operation and maintenance management of hydropower stations, and particularly relates to a method for intelligent comprehensive alarm and operation and maintenance response of hydropower stations based on an execution engine. The method comprises: constructing a physical-logical fusion topology model; monitoring signals of the whole plant in real time, and when a first alarm signal is detected, opening an analysis window with the triggering time of the first alarm signal as the center, temporally and spatially aligning concurrent alarm signals in the analysis window to form a set of signals to be analyzed; using a directed graph algorithm to analyze the topology paths between alarm nodes in the set of signals to be analyzed, thereby locking the root cause alarm, and marking physical and logical secondary alarms based on the root cause alarm; based on the locked root cause alarm, the execution engine automatically suppresses the independent pop-up of alarm prompts corresponding to the secondary alarms, and according to the characteristic information of the root cause alarm, matches the corresponding operation and maintenance response plan, and executes the operation and maintenance response. The present application effectively suppresses secondary alarms, and improves the emergency response efficiency and operation safety of hydropower stations.

Description

Technical Field

[0001] This invention belongs to the field of automated monitoring and operation and maintenance management of hydropower stations, and specifically relates to an intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine. Background Technology

[0002] With the continuous improvement of the intelligence level of hydropower stations, establishing an efficient and reliable monitoring and alarm system is of great significance for ensuring the safe and stable operation of the station. As a complex system integrating hydraulics, mechanics, electrical systems, and control, the real-time perception and anomaly identification of a hydropower station's operating status directly affect the timeliness and accuracy of fault handling. In modern hydropower stations, integrating various sensor data and constructing an alarm response mechanism covering the entire plant has become an important technological direction for improving emergency management capabilities.

[0003] Currently, most alarm monitoring systems used in hydropower stations are based on threshold criteria and time-series processing logic. When a sudden equipment failure occurs, due to the tight physical connections and control coupling within the system, a large number of related alarm signals are often triggered in a short period of time, forming a so-called alarm avalanche phenomenon. Traditional systems struggle to effectively distinguish the true root cause of the fault from the secondary alarms caused by the massive concurrent signals, resulting in redundant information occupying the operation and maintenance interface and significantly increasing the difficulty for maintenance personnel to identify critical faults. Furthermore, after an alarm is triggered, the corresponding operation and maintenance response plan of existing alarm systems usually relies on manual retrieval and experience-based judgment, lacking automated response capabilities deeply coupled with the alarm analysis mechanism, which affects the efficiency and accuracy of fault handling.

[0004] Therefore, how to quickly pinpoint the root cause alarm from burst alarm signals, automatically suppress secondary interference, and achieve coordinated alarm analysis and operation and maintenance response are technical problems that urgently need to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a smart integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine, so as to effectively solve the problems in the background art mentioned above.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for intelligent integrated alarm and operation and maintenance response of hydropower stations based on an execution engine, comprising: S1. Construct a physical-logical integrated topology model; S2. Based on the physical logic fusion topology model, the execution engine monitors all plant signals in real time. When the first alarm signal is detected, an analysis window is opened with the trigger time of the first alarm signal as the center. The concurrent alarm signals in the analysis window are spatiotemporally aligned to form a set of signals to be analyzed. S3. Based on the physical-logical fusion topology model, the topological path between each alarm node in the signal set to be analyzed is analyzed using a directed graph algorithm, thereby locking the root cause alarm and marking the corresponding physical secondary alarm and logical secondary alarm based on the root cause alarm. S4. Based on the root cause alarm, the execution engine automatically suppresses the independent pop-up of alarm prompts corresponding to secondary alarms, and matches the corresponding operation and maintenance response plan according to the characteristic information of the root cause alarm, and executes the operation and maintenance response.

[0007] Preferably, step S1 specifically includes: S101. Construct a four-level device tree structure that reflects the digital abstraction of the controlled and perceived objects across the entire hydropower station area; S102. Define and establish physical and logical connections between the nodes of the device tree structure to form a fused topology diagram; S103. Store the fused topology graph in a graph database, with nodes representing device entities and edges representing physical or logical relationships between devices, and assign direction and weight attributes to each edge to obtain a physical-logical fused topology model.

[0008] Preferably, the device tree structure in step S101 is divided from top to bottom into plant level, unit or common system level, specific equipment level, and measurement point signal and status level. Among them, the plant level stores the global metadata of the plant; the unit or common system level divides the entire station into relatively independent logical function blocks; the specific equipment level lists in detail the physical components that make up each unit, including primary equipment, secondary equipment, and process auxiliary systems; the measurement point signal and status level includes all analog data reported by sensors, switch status bits, hard-wired signals of protection actions, and internal logic variables. In step S102, the physical connections include electrical connections for power transmission, rigid connections for mechanical energy transmission, and pipeline connections for fluid medium transmission; the logical connections include command interaction relationships for control loops, constraint relationships for interlocking loops, and triggering causal relationships for protection loops. In step S103, a weight attribute value is assigned to each edge according to the preset density calculation rules. Specifically, for any edge, its weight is... The weights are determined by the base weight, distance attenuation factor, and dielectric conductivity, and the calculation formula is as follows: , In the formula, The weight attribute value of the edge. It is based on the preset weight of the connection type, which includes electrical connection, mechanical connection, fluid pipeline connection and logical connection; The distance attenuation factor is a parameter used to quantify the attenuation effect of physical or electrical distance between nodes on the association strength. It is calculated based on the actual physical length of the edge. The medium conduction coefficient is used to characterize the difference in the transmission efficiency of different transmission media to fault disturbances or control signals. It is determined according to the type of transmission medium, which includes copper conductors, fluid pipelines with different roughness, and logic signals.

[0009] Preferably, step S2 specifically includes: S201, The execution engine acquires the signal data stream of the entire plant in real time through the distributed acquisition gateway; S202. Set the width of the sliding time window, and dynamically determine the width of the sliding time window based on the weighted average of the electromagnetic transient response time and thermodynamic response time of the hydropower station. The calculation formula is as follows: , In the formula, The width of the sliding time window. This refers to the typical transient process time of circuit breaker operation or voltage drop. This represents the typical process time for changes in bearing temperature or oil temperature. and These are the weighting coefficients, and ; S203. When the execution engine detects the first alarm signal, it locks the trigger time of the first alarm signal and uses that time as the analysis start point. Then, it opens an analysis window based on the width of the sliding time window, with the analysis start point as the center. S204. In the time dimension, a high-precision time synchronization protocol is used to project all signals falling within the analysis window onto a unified reference time axis, eliminating the deviation of each signal in time recording. S205. In the spatial dimension, taking the device node where the first alarm signal is located as the center, the physical logic fusion topology model is used to search for all relevant nodes within the preset topology distance, and to extract all accompanying alarm signals generated by all relevant nodes during the analysis window. S206. The execution engine synchronously associates the time-aligned information from step S204 with the accompanying alarm signal obtained from spatial aggregation in step S205, thereby forming a set of signals to be analyzed with a unified time reference and a clear spatial topology.

[0010] Preferably, step S3 specifically includes: S301. Map all alarm signals in the signal set to be analyzed one by one to the corresponding nodes of the physical logic fusion topology model, extract the local subgraph containing only the current alarm node, and express the relationship between nodes as a directed connection according to the preset direction attribute of each edge to form an alarm propagation map. S302. Search for all nodes with an in-degree of zero in the alarm propagation graph and extract all nodes with an in-degree of zero as root cause candidate nodes. S303. If there is only one root cause candidate node, directly use that root cause candidate node as the root cause alarm; otherwise, calculate the comprehensive root cause weight of each root cause candidate node, and select the root cause candidate node with the highest comprehensive root cause weight value as the root cause alarm. S304. After locking the root cause alarm, traverse downstream along the directed edge of the alarm propagation map, and determine whether other alarm signals on the path belong to physical secondary alarms according to the preset physical propagation law. The preset physical propagation law includes thermodynamic conduction delay law, fluid pressure fluctuation propagation law and electromagnetic transient response law. Specifically, for an alarm signal that conforms to any one of the physical propagation laws of thermodynamic conduction delay law, fluid pressure fluctuation propagation law and electromagnetic transient response law, if the alarm signal is located downstream of another alarm signal in the alarm propagation map and the difference between the triggering times of the two signals satisfies the preset time delay interval of the corresponding physical law, then the alarm signal is marked as a physical secondary alarm caused by an upstream fault. S305. After locking the root cause alarm, the logic secondary alarm is marked based on the control logic rules. Specifically, the action feedback sequence that the relevant controlled object should generate is predicted according to the protection logic entry that matches the current root cause alarm type in the logic rule library. Then, the execution engine compares the actual monitored signal with the predicted action feedback sequence. If the timestamp deviation between the two is less than the preset threshold and the state value is completely consistent, then these actual monitored alarm signals that match the predicted sequence are uniformly marked as logic secondary alarms.

[0011] Preferably, in step S303, the comprehensive root cause weight is jointly determined by the topological compactness weight and the time priority weight, wherein the topological compactness weight is calculated as follows: , In the formula, For topologically compact weights, The total shortest path length from the candidate root cause node to each downstream alarm node in the alarm propagation graph. The formula for calculating the overall root cause weight is: , In the formula, To weigh the root causes; This is a preset coordination coefficient, ranging from 0 to 1, used to balance the relative importance of topological and time factors in decision-making; This is a time priority weight.

[0012] Preferably, the total shortest path length from any root cause candidate node in the alarm propagation graph to all downstream alarm nodes is calculated using one of the following two methods, depending on whether edge weights are configured: (1) Unweighted path length: Used when no edge weights are configured, it is the sum of the number of edges contained in the path. The specific calculation method is as follows: The execution engine searches all directed paths from the root cause candidate node to each reachable downstream alarm node, determines the directed path with the fewest number of edges as the corresponding shortest path, and adds up the number of edges contained in the shortest path from the root cause candidate node to each reachable downstream alarm node to obtain the total length of the shortest path of the root cause candidate node. (2) Weighted path length: Used when edge weights are configured for each edge, it is the sum of the equivalent propagation costs of all edges contained in the path. The specific calculation method is as follows: The execution engine calculates the sum of the equivalent propagation costs of all edges contained in each directed path, and determines the directed path with the smallest sum of equivalent propagation costs as the corresponding weighted shortest path. The weighted shortest path lengths from the root cause candidate node to each reachable downstream alarm node are added together to obtain the total weighted shortest path length of the root cause candidate node. The formula for calculating the equivalent propagation cost of any edge is as follows: , In the formula, For the equivalent propagation cost of any edge, This is a preset minimum positive number used to avoid a denominator of zero; Step S303 also includes incorporating maintenance history data to correct the comprehensive root cause weights, specifically as follows: (1) The execution engine obtains maintenance records and defect ledgers from the power plant operation and management system through cross-disciplinary data interfaces; (2) Based on the defects and maintenance records of the equipment corresponding to each root cause candidate node within a preset number of historical days, assign a corresponding historical correction weight to the root cause candidate node, specifically: If the device corresponding to a certain root cause candidate node has only defect records within a preset number of historical days, then the historical correction weight of that root cause candidate node is... ; If the device corresponding to a certain root cause candidate node has only maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node is... ; If the device corresponding to a root cause candidate node has both defect records and maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node is adjusted. ; If the device corresponding to a certain root cause candidate node has no defect records and no maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node is adjusted. ; (3) Substitute the historical adjusted weights into the formula for calculating the comprehensive root cause weights for correction. The formula for calculating the corrected comprehensive root cause weights is as follows: , In the formula, This is a historical correction factor. Adjust the weights based on historical data.

[0013] Preferably, step S3 further includes a loop-breaking search strategy to handle closed-loop structures existing in the alarm propagation spectrum, specifically: (1) The execution engine traverses the alarm propagation graph. When a directed closed loop consisting of several nodes is detected, the state change timestamps of all nodes in the directed closed loop are extracted, and the node that first triggers the alarm on the time axis is defined as the logical starting point of the directed closed loop. (2) Temporarily remove the incoming edges pointing to the logical starting point in the alarm propagation graph, and transform the original cyclic graph into a temporary directed acyclic graph; (3) Perform the root cause search and secondary alarm marking process on the temporary directed acyclic graph. After completing the causal analysis of the alarm signals in the directed closed loop, restore all the reverse edges that have been temporarily released.

[0014] Preferably, step S4 specifically includes: S401. The execution engine suppresses secondary alarms on the operation interface of the operation and maintenance personnel and pops them up as independent alarm entries. It then collapses and displays them in the details list of the corresponding main alarm entry. Subsequently, it generates the corresponding main alarm entry based on the identified root cause alarm and displays it. The information items of the main alarm entry include the full name of the identified root cause alarm signal, the time of occurrence, the physical location involved, the root cause confidence value, the type identifier, the professional category to which the equipment belongs, and the fault level. S402. Using the root cause alarm type identifier, the equipment's professional category, and the fault level as the joint search key, search the preset intelligent operation and maintenance knowledge base and match the corresponding operation and maintenance response plan. S403. The matched operation and maintenance response plan is pushed to the operation and maintenance personnel through a multi-terminal collaboration mechanism. Each push instruction in step S403 contains a unique task identification code, which is bound to the root cause alarm event and used to track the response status of the root cause alarm event throughout the process. On-site inspection personnel use handheld terminals to scan the device identification code on the faulty device to locate and confirm the location, and feed back the processing progress to the execution engine in real time.

[0015] Preferably, step S1 further includes: introducing operation mode perception variables into the physical-logical fusion topology model, and establishing a mapping library between operation modes and model configurations, so that the model can dynamically adapt to the changing operating conditions of the hydropower station. The operation modes include power generation mode, pumping mode, phase adjustment mode and maintenance mode.

[0016] In summary, this application includes the following beneficial technical effects: 1. By integrating the physical logic topology model with the sliding time window, this invention can accurately identify causal relationships from the massive amount of "alarm avalanche" information and automatically suppress secondary alarms. This allows maintenance personnel to focus only on key root cause alarms under complex fault conditions, greatly reducing the decision-making pressure caused by information overload.

[0017] 2. By utilizing directed graph algorithms combined with the specific physical propagation laws and control logic of hydropower stations, this invention elevates alarm analysis from simple sequence arrangement to in-depth logical deduction, enabling the identification of the fault's starting point within seconds, effectively eliminating interference from redundant information, and achieving a leap from phenomenon perception to essence recognition.

[0018] 3. Through deep coupling of the execution engine and the operation and maintenance response plan, this invention can automatically match and push targeted handling suggestions based on the identified root causes of faults. That is, alarm analysis results are directly converted into executable operation and maintenance instructions, eliminating the intermediate links of manual alarm interpretation and plan retrieval in the traditional method. This reduces the fault response time from minutes to seconds and avoids secondary risks caused by human judgment errors, realizing a closed loop from "information presentation" to "action guidance". At the same time, the multi-terminal push and feedback mechanism ensures the traceability of the response process. Furthermore, by introducing historical maintenance data to correct the root cause weight and combining it with the intelligent operation and maintenance knowledge base to achieve accurate matching of plans, the system can make full use of the power station's historical operating information and established handling experience, significantly enhancing the hydropower station's safety defense capabilities in the face of sudden accidents.

[0019] 4. The physical-logical fusion topology model supports switching between multiple operating modes, enabling the invention to automatically adjust the analysis strategy according to different output states of the hydropower station. Simultaneously, the introduction of historical maintenance data and sensor status monitoring further improves the identification accuracy under non-ideal operating conditions, ensuring the system's reliability and robustness. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall technical solution for a hydropower station intelligent integrated alarm and operation and maintenance response method based on an execution engine. Figure 2 It is a schematic diagram illustrating the principle of physical-logical fusion topology model construction and hierarchical abstraction; Figure 3This is a flowchart illustrating the logic of aligning the sliding time window signal monitoring with the alarm signal in time and space. Figure 4 This is a schematic diagram illustrating the core principle of alarm cause-and-effect tracing and secondary alarm labeling based on directed graph algorithms; Figure 5 This is a diagram illustrating the interaction and data flow between the matching and push of operation and maintenance response plans and the collaborative response of multiple terminals. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 5 The present invention will be further described in detail below with reference to specific embodiments.

[0022] Example 1

[0023] The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on the execution engine is implemented according to the following steps: First, step S1 is executed: constructing a physical-logical fusion topology model. In the intelligent integrated alarm and operation and maintenance response method for hydropower stations based on the execution engine, the core task of step S1 is to construct a digital model reflecting the physical connections and control logic relationships of all equipment in the hydropower station, i.e., a physical-logical fusion topology model. This model is the foundation for the intelligent alarm system to perform analysis, tracing, and decision-making. Its construction process is specifically broken down into the following sub-steps.

[0024] Step S101: Construct a hierarchical device tree structure to complete the digital abstract representation of all controlled and perceived objects in the hydropower station. This device tree is divided into four levels from top to bottom.

[0025] The first level is the plant / site level, which is used to store the plant / site's global metadata, mainly including the plant / site's geographical location information, total installed capacity, access voltage level, and the whole site's single-line diagram index.

[0026] The second level is the unit or common system level, which divides the entire station into relatively independent logical functional blocks, such as Unit 1, Unit 2, 500 kV switch station system, plant power system, etc.

[0027] The third level is the specific equipment level, which lists in detail the physical components that make up each unit. Among them, primary equipment includes turbines, generators, main transformers, circuit breakers, disconnect switches, and busbars; secondary equipment includes relay protection devices, computer monitoring system units, automatic dispatching devices, and excitation regulation devices; and process auxiliary systems include technical water supply systems, turbine oil systems, compressed air systems, and leakage drainage systems.

[0028] The fourth level is the measurement point signal and status level, which includes all analog data reported by sensors, switch status bits, hard-wired signals of protection actions, and internal logic variables.

[0029] In this embodiment of the application, primary equipment refers to the main equipment that directly participates in the generation, transmission and distribution of electrical energy, while secondary equipment refers to auxiliary equipment that monitors, controls, protects and regulates the primary equipment.

[0030] Step S102: Define and establish two types of connection relationships between the nodes of the established device tree, namely physical connection and logical connection, so as to form a fusion topology diagram that can simultaneously reflect the interaction between the real physical field and the control system.

[0031] Physical connections are specifically manifested in electrical connections for power transmission, rigid connections for mechanical energy transmission, and pipeline connections for fluid medium transmission. These connections are represented in the database through strong edges.

[0032] Logical connections are specifically manifested in the command interaction relationships of control loops, the constraint relationships of interlocking loops, and the triggering causal relationships of protection loops. These connections are represented by weakly associated edges or conditionally associated edges.

[0033] Step S103: The constructed fused topology graph, carrying physical and logical connections, is completely stored in a preset graph database. This graph database adopts a non-relational structure, with nodes representing device entities and edges representing physical or logical connections between devices. Each edge is assigned a weight attribute value and a direction attribute.

[0034] The direction attribute is used to specify the direction of transmission of energy flow, information flow, or signal flow.

[0035] The weight attribute value is used to characterize the tightness of the connection. The weight is determined according to a preset tightness calculation rule, specifically: for any edge, its weight... The weights are determined by the base weight, distance attenuation factor, and dielectric conductivity, and the calculation formula is as follows:

[0036] in, Based on the preset base weights of the connection type, the weights differ depending on the type of electrical connection, mechanical connection, fluid piping connection, and logical connection. They are all different; The distance attenuation factor is a parameter used to quantify the attenuation effect of physical or electrical distance between nodes on the association strength. It is calculated based on the actual physical length of the edge, such as the cable length or pipe length. The medium conduction coefficient is used to characterize the difference in the transmission efficiency of different transmission media to fault disturbances or control signals.

[0037] In this embodiment, the basic weight values ​​of the instruction interaction relationship of the control loop, the constraint relationship of the interlocking loop, and the triggering causal relationship of the protection loop are equal.

[0038] In this embodiment of the application, the distance attenuation factor The value is not directly taken from the physical length, but is calculated according to different benchmarks based on the physical characteristics of the connection type, so as to accurately reflect the real law of fault propagation with distance in different fields of hydropower stations.

[0039] Specifically, for electrical connections (such as power cables or enclosed busbars), a reference length of 100 meters is used as the baseline. ,in The actual laying length of the electrical connection is given in meters; for mechanical connections (such as the drive shaft between a turbine and a generator), a reference length of 1 meter is used. ,in This refers to the actual physical length of the shaft system, in meters; for fluid piping connections (such as technical water supply pipes or turbine oil pipes), a reference length of 10 meters is used. ,in This refers to the actual length of the pipeline, in meters. For logic control connections, including control loops, interlocking loops, and protection loops, since their transmission medium is electrical or optical signals, the impact of physical distance on the time delay of fault causal transmission is negligible. This embodiment uses 1000 meters as the standard reference length. ,in The actual length of the signal cable is in meters. Through the above differential calculations, the attenuation factor of physical connections is usually significantly greater than that of logical connections. Therefore, in subsequent directed graph analysis, the system is more inclined to classify upstream physical faults as logic signals whose root cause is less affected by distance.

[0040] In this embodiment, the dielectric conductivity is... Instead of real-time calculation using continuous functions, values ​​are directly assigned based on preset empirical values ​​according to the actual physical transmission medium of the connection path and the fault propagation characteristics of various media in hydropower station engineering practice. Specifically: For electrical connections, if the transmission medium is a copper busbar or copper core cable, it is chosen because it has the highest conductivity and the strongest fault current conduction capability. If the conductor is an aluminum core (such as the low-voltage side busbar of some older dry-type transformers), then take... .

[0041] For mechanical connections, if the coupling is made of rigid steel or the main shaft of the transmission, mechanical vibration and torque disturbances can be transmitted efficiently. If the connection contains a flexible coupling or damping device, the value should be appropriately reduced according to the degree of damping, for example, take... .

[0042] For fluid pipeline connections, the conduction efficiency depends on the compressibility and viscosity of the fluid medium. In clean water pipelines such as technical water supply systems, pressure waves propagate relatively quickly, so... For oil pipelines such as turbine oil, due to their high viscosity and rapid pressure wave attenuation, [the following is used]: Compressed air pipelines, due to the high compressibility of gas and significant lag in disturbance transmission, require... .

[0043] For protection circuits, control circuits, and interlocking circuits that are logically connected, the transmission medium is always an electrical signal or an optical signal, resulting in extremely fast transmission speed. However, the transmission efficiency varies slightly due to the influence of signal processing stages (such as relay action time, PLC scan cycle, and communication protocol stack delay). Specifically, the trip signals of protection circuits are mostly pure hard-wired direct connections and do not undergo complex logical operations. The command signals of the control loop need to be processed by the host computer or the local control unit to obtain... The constraint signal of the interlocking circuit often needs to pass through multiple auxiliary contacts in series and involves electrical conversion of mechanical position. .

[0044] For example, when an inter-turn short circuit occurs in the generator stator winding, the fault current is conducted along the copper core cable towards the main transformer. Its dielectric conductivity is taken as 1.0, which is much higher than the conductivity of cooling water temperature changes (0.5). This makes the weight of directly measured electrical signals significantly higher than that of indirect thermodynamic measurement signals in subsequent weight calculations, thus facilitating more accurate location of the electrical fault source. Through this calculation rule, the weight of each edge is quantified, providing a reference for subsequent alarm source tracing analysis.

[0045] Preferably, an operational mode-aware variable is introduced into the physical-logical fusion topology model, and a mapping library between operational modes and model configurations is established, enabling the model to dynamically adapt to the changing operating conditions of the hydropower station. Specific operational modes include at least power generation mode, pumping mode, phase regulation mode, and maintenance mode.

[0046] The technical mechanism for achieving automatic switching is as follows: In the graph database, for each group of relational edges or alarm suppression rules that will change under different operating modes, an associated mode identifier is created. The execution engine internally maintains an operating mode configuration mapping library, which is indexed by the operating mode and stores the set of edges that should be activated or deactivated under that mode, as well as the alarm weight adjustment coefficients for each device node.

[0047] When the overall control system or manual input changes the current operating mode, the execution engine reads the configuration parameters of the corresponding mode from the mapping library, automatically loads the edge connection relationships and alarm weight coefficients of that mode, and thus dynamically reorganizes the topology.

[0048] For example, when a unit switches to maintenance mode, all device IDs of that unit in the mapping library are retrieved, and the execution engine sets the logical priority of all alarm signals originating from these IDs to the lowest level, so that they do not trigger the subsequent root cause analysis process, thereby avoiding interference with normally operating units.

[0049] Thus, through step S1, a well-structured, clearly interconnected, and adaptive physical-logical fusion topology model has been constructed. This model depicts the spatial hierarchy of equipment, structurally expresses the direction and density of energy and signal transmission, and can dynamically adjust its topology and rules according to changes in operating conditions. This provides an analytical foundation for the execution engine.

[0050] In subsequent step S2, when the execution engine captures the first alarm signal in the entire plant using a sliding time window, it will open an analysis window based on this model and perform spatiotemporal alignment on the concurrent alarm signals within the window, thereby preparing a set of signals to be analyzed for the upcoming causal tracing and deduction.

[0051] Next, step S2 is executed: the execution engine monitors all plant signals in real time through a sliding time window, and opens an analysis window when the first alarm is detected to perform spatiotemporal alignment of concurrent alarms within the window.

[0052] It should be noted that in this application, "first alarm signal" refers to the first alarm signal newly detected and used to trigger the current alarm analysis cycle when the execution engine is not currently processing an alarm analysis cycle, rather than the absolute first alarm signal generated during the operation of the hydropower station. After the execution engine detects the first alarm signal of the current alarm analysis cycle, it determines the analysis window centered on the trigger time of the first alarm signal, and analyzes other alarm signals that fall within the analysis window and meet the preset spatial range as accompanying alarm signals of the current alarm analysis cycle. Before the completion of the current alarm analysis cycle, a new alarm analysis cycle will not be started with the accompanying alarm signals. For alarm signals newly detected outside the current analysis window, the execution engine stores them in the analysis queue in sequence according to their trigger times. After the completion of the current alarm analysis cycle, the alarm signal with the earliest trigger time in the analysis queue that has not yet been assigned to other alarm analysis cycles is determined as the first alarm signal of the next alarm analysis cycle, and the next alarm analysis is started accordingly. Alarm signals within the analysis window that are located before the trigger time of the first alarm signal are used as historical correlation signals for this causal analysis and do not affect the determination that the first alarm signal is the trigger signal for the current alarm analysis cycle.

[0053] Step S2 is used to accurately define a spatiotemporal range to be analyzed when a massive influx of signals arrives, and to unify asynchronous multi-source signals under the same benchmark, providing clean and accurate input for subsequent directed graph algorithms. The specific implementation process can be broken down into the following sub-steps.

[0054] Step S201: Deploy a distributed data acquisition gateway to access the plant's real-time data streams. The execution engine uses a distributed data acquisition gateway to acquire real-time data streams from the monitoring system, protection system, and various smart sensors. These data streams are connected to the execution engine's buffer via a unified communication protocol bus, ensuring that signals from different sources and in different formats can be aggregated within the same channel.

[0055] In this embodiment, the execution engine is a software platform that integrates data acquisition, intelligent analysis, decision control, and collaborative command. It interacts with the local system through a distributed acquisition gateway to achieve a complete closed loop from alarm perception and root cause analysis to contingency plan execution and tracking.

[0056] Step S202: Set the basic parameters of the sliding time window and establish a dynamic adjustment mechanism. The width of the sliding time window is set to the time window width. (i.e., preset time threshold). The value is determined based on the weighted average of the electromagnetic transient response time and the thermodynamic response time of the hydropower station. The specific weighting calculation method is as follows: .in, This refers to the typical transient process time of circuit breaker operation or voltage drop. This represents the typical process time for changes in bearing temperature or oil temperature. and These are weighting coefficients adjusted based on the characteristics of the power plant equipment, and they satisfy... .

[0057] The step size of the sliding time window, i.e., the time interval between the start times of two adjacent analysis windows, is dynamically adjusted by the system based on real-time calculated load. The preset load threshold is greater than the preset stability threshold; both thresholds are manually set by maintenance personnel based on historical power plant operating data. When the surge intensity (i.e., the number of newly triggered alarms per unit time) exceeds the preset load threshold, the execution engine automatically shortens the width of the analysis window to improve real-time response efficiency; when the surge intensity is lower than the preset stability threshold, the execution engine maintains the standard window width to ensure the depth and comprehensiveness of the analysis.

[0058] Step S203: Lock the analysis start point and trigger the opening of the analysis window. When the execution engine detects that the status bit of any signal has flipped or the value has exceeded the preset threshold, it immediately locks that moment as the analysis start point. And taking that moment as the center, based on the time window width determined in step S202 Extend forward Extend backward Define a total width as The theoretical analysis window range, i.e. Because the window covers future periods after the first alarm, the execution engine uses a sliding cache mechanism: Read immediately from the cache Historical signal data within the interval is simultaneously monitored and cached in real time. New alarm signals entering the interval; waiting for actual time to advance to Once all window data is ready, the execution engine merges the two data segments into complete analysis window data and outputs it uniformly to the subsequent spatiotemporal alignment steps.

[0059] For example, if the first alarm is triggered at the 20th second (i.e. (seconds), and If the window spans seconds, then the theoretical analysis window covers data from second 15 to second 25: the execution engine locks the window starting point at second 15 at second 20 and reads the existing cache, while continuously listening for new data from second 20 to second 25. Once second 25 arrives, the complete window data is formed. This method employs a centrally symmetric window strategy to fully capture the causal landscape before and after the fault occurs; the cost is the introduction of... The inherent analysis delay, which extends across the time window width. The dynamic setting has been taken into account in terms of real-time tolerance.

[0060] Step S204: Perform consistency correction on the signals within the window in the time dimension. With the opening of the analysis window, the execution engine initiates a spatiotemporal alignment operation. In the time dimension, the execution engine extracts precise synchronization timestamps from signals from different sources. Since different subsystems use different sampling frequencies and experience transmission delays, the execution engine utilizes high-precision time synchronization protocols, such as Network Time Protocol (NTP) or Precision Time Protocol (PTP), to project all signals onto a unified reference time axis, eliminating minor deviations in the time recording of each signal.

[0061] Step S205: The signals within the window are correlated and aggregated spatially. Spatially, the execution engine, centered on the device node containing the first alarm signal, searches for all relevant nodes within a preset topological distance using the physical-logical fusion topology model constructed in step S1, and extracts all accompanying alarm signals generated by these relevant nodes during the analysis window.

[0062] In this embodiment, the preset topological distance refers to the number of edges traversed when traversing outwards along the associated edges from the central node in the graph database. This distance is dynamically determined manually based on the type of the first alarm signal. The execution engine extracts all accompanying alarm signals generated by the aforementioned nodes during the analysis window, and aggregates these signals with the first alarm signal to form a set of signals to be analyzed.

[0063] Specifically, for electrical quantity signals, including overcurrent, overvoltage, and frequency anomalies, whose fault impact is relatively wide, the topology distance is set to 3 to 5 hops; for logic quantity signals, including protection actions, switch changes, and control commands, which may propagate across multiple levels, the topology distance is set to 3 to 5 hops; for mechanical quantity signals, including vibration, sway, and speed anomalies, whose vibration transmission range is limited, the topology distance is set to 1 to 2 hops; for thermal quantity signals, including temperature and oil temperature rises, whose transmission speed is slow and affects local areas, the topology distance is set to 1 hop; for flow and pressure signals, including water pressure, oil pressure, and air pressure anomalies, the topology distance is set to 2 to 3 hops. When the first alarm signal simultaneously possesses multiple types of characteristics, the maximum corresponding topology distance value is taken.

[0064] In step S206, the execution engine synchronously associates the time-aligned information from step S204 with the accompanying alarm signal obtained from spatial aggregation in step S205, thereby forming a set of signals to be analyzed with a unified time reference and a clear spatial topology.

[0065] It should be noted that steps S204 and S205 employ a parallel processing mechanism to perform dual filtering of concurrent alarm signals in the analysis window in both the time and spatial domains. Subsequently, the execution engine performs an intersection operation on alarm signals that simultaneously meet the criteria of "time axis aligned" and "node located within a predetermined topological distance," and aggregates them into a set of signals to be analyzed.

[0066] In summary, through step S2, the execution engine, based on a sliding time window and spatiotemporal alignment mechanism, completes the temporal and spatial dual-dimensional aggregation of alarm signals. This process uses the static topology model constructed in step S1 as a spatial index, combined with dynamic time window technology, to effectively solve the problem of disordered signal influx during alarm avalanches. Once the set of signals to be analyzed is locked, the system naturally transitions to step S3, which uses a directed graph algorithm to perform topology path search and causal inference within this signal set to distinguish between root cause alarms and secondary alarms.

[0067] Subsequently, step S3 is executed: based on the topology model and directed graph algorithm, the topological path between each alarm node is analyzed, and the signal located downstream of the path and conforming to the control logic deduction result is marked as a secondary alarm.

[0068] Step S3 follows the signal set to be analyzed identified in step S2 and proceeds to the fault causal reasoning stage. The execution engine utilizes the physical-logical fusion topology model constructed in step S1 as a knowledge framework, combining directed graph algorithms and preset physical and control rules to distinguish causal chains from alarm avalanches and pinpoint the source of the fault. Its specific implementation process can be broken down into the following sub-steps.

[0069] Step S301: Map the signals to be analyzed to the topology model to construct a directional alarm propagation map. The execution engine maps all alarm signals in the signal set to be analyzed generated in step S2 to the corresponding nodes of the physical-logical fusion topology model constructed in step S1, thereby extracting a local subgraph containing only the current alarm node.

[0070] In this local subgraph, the execution engine expresses the relationships between nodes as directional connections based on the preset direction attributes of each edge. Thus, the subgraph is transformed into an alarm propagation graph, where the in-degree of each node represents the number of upstream alarms that directly or indirectly affect that node, and the out-degree represents the number of downstream nodes that this alarm may affect.

[0071] Step S302: Initially identify root cause candidate nodes through in-degree search. A directed graph algorithm traverses and analyzes the alarm propagation graph, focusing on searching for all nodes with an in-degree of 0. An in-degree of 0 means that the alarm signal represented by this node has no upstream reference from other alarm signals in the graph, indicating that it was not caused by any other monitored signals in this event. The execution engine extracts all these nodes with an in-degree of 0 as a potential root cause candidate set.

[0072] If, after the search, there is only one node with an in-degree of 0, the execution engine will directly mark the node as a root cause alarm and proceed to step S304 to determine the secondary alarm.

[0073] If there are multiple nodes with an in-degree of 0, proceed to step S303 for further sorting and identification.

[0074] If there are no nodes with an in-degree of 0, it indicates that the alarm propagation graph currently presents a fully closed-loop structure. The execution engine immediately calls the loop-breaking search strategy preset in step S3 to which this step belongs, performs loop-breaking processing on the closed loop, transforms the loop-bearing graph into a temporary directed acyclic graph, and re-executes the in-degree search of this step in the transformed graph. If, after loop-breaking processing and re-searching, no nodes with an in-degree of 0 are still found, the execution engine uses the node with the earliest trigger time as the filtering condition as the default root cause alarm and proceeds to step S304 to determine the secondary alarm.

[0075] Preferably, if the execution engine uses the earliest trigger time as the filtering condition and selects multiple candidate nodes, then the node with the highest topological centrality is further selected as the default root cause alarm.

[0076] In this embodiment, the topological centrality is calculated by any centrality metric, such as degree centrality, proximity centrality, betweenness centrality, or eigenvector centrality. In practical applications, the appropriate metric can be selected based on the needs of those skilled in the art.

[0077] Step S303: When multiple root cause candidate nodes exist, they are ranked by weight based on correlation density and timing rules. When multiple nodes with an in-degree of 0 exist in the alarm propagation graph, the execution engine needs to identify the most likely fault initiation point. The execution engine calculates the comprehensive root cause weight for each root cause candidate node. Due to topologically dense weights and time priority weight It consists of two parts, and the calculation formula is: ,in This is a preset coordination coefficient, ranging from 0 to 1, used to balance the relative importance of topological and time factors in decision-making.

[0078] The topology compactness weight depends on the common path length from the candidate node to other alarm nodes. It is calculated as follows: in the alarm propagation graph, starting from the candidate node with an in-degree of 0, calculate the total shortest path length from it to each downstream alarm node. Then topologically compact weights .in The smaller the value, the more tightly the node is connected to all alarm events in the topology, and the more likely it is to be the source of the alarm propagation chain.

[0079] In this embodiment of the application, the total shortest path length from any root cause candidate node in the alarm propagation graph to each downstream alarm node is calculated using one of the following two methods, depending on whether edge weights are configured: (1) Unweighted path length: Used when no edge weights are configured, it is the sum of the number of edges contained in the path. The specific calculation method is as follows: The execution engine searches all directed paths from the root cause candidate node to each reachable downstream alarm node, determines the directed path with the fewest number of edges as the corresponding shortest path, and adds up the number of edges contained in the shortest path from the root cause candidate node to each reachable downstream alarm node to obtain the total length of the shortest path of the root cause candidate node. (2) Weighted path length: Used when edge weights are configured for each edge. It is the sum of the equivalent propagation costs of each edge contained in the path. The specific calculation method is as follows: The execution engine calculates the sum of the equivalent propagation costs of each edge contained in each directed path, determines the directed path with the smallest sum of equivalent propagation costs as the corresponding weighted shortest path, and adds the weighted shortest path lengths from the root cause candidate node to each reachable downstream alarm node to obtain the total weighted shortest path length of the root cause candidate node.

[0080] The formula for calculating the equivalent propagation cost of any edge is shown below:

[0081] In the formula, For the equivalent propagation cost of any edge, This is a preset minimum positive number used to avoid a denominator of zero.

[0082] It should be noted that a larger edge weight indicates a stronger connection between the corresponding nodes and a lower equivalent propagation cost; conversely, a smaller edge weight indicates a weaker connection between the corresponding nodes and a higher equivalent propagation cost. For multiple directed paths from a root cause candidate node to any downstream alarm node, the path with the minimum sum of equivalent propagation costs is selected as the shortest path between them. Therefore, the base weight, distance attenuation factor, and medium conduction coefficient participate in the calculation of the total shortest path length and topological compactness weight through the edge weights. A smaller total shortest path length indicates a stronger overall connection between the root cause candidate node and all downstream alarm nodes, a higher topological compactness weight, and a greater likelihood of it becoming the source of the alarm propagation chain.

[0083] The time priority weight depends on the triggering time of the alarm signals corresponding to the candidate nodes. The execution engine sorts all candidate nodes with an in-degree of 0 according to their triggering time from earliest to latest, assigning higher priority to nodes that are ranked earlier. Values ​​can be assigned using a linearly decreasing ranking method, for example, assigning values ​​of 1.0, 0.8, 0.6, 0.4, and 0.2 respectively based on the trigger time from earliest to latest, with a minimum of 0.2; alternatively, a relative time difference normalization method can be used, where the earliest trigger time is recorded as... The latest trigger time is recorded as For any candidate node Triggering time According to the formula Calculation, where This is a very small positive number (e.g., 0.001 seconds) to avoid the denominator being zero. Both methods ensure that the node with the earliest trigger time receives the highest time priority weight, which is closely related to the topology weight. Jointly participate in the comprehensive root cause weighting The calculation. The specific assignment method can be preset by the operations and maintenance personnel during system configuration.

[0084] Preferably, step S303 also introduces maintenance history data as historical correction weight to improve the accuracy of root cause judgment. In order to further improve the identification accuracy of the root cause locking link under non-ideal working conditions, the execution engine obtains maintenance records and defect ledgers in the power plant operation management system through cross-professional data interfaces.

[0085] Specifically, in step S303, when sorting the root cause candidate nodes by weight, the execution engine obtains the maintenance records and defect ledgers from the power plant operation management system through a cross-disciplinary data interface, queries the maintenance and defect status of the equipment corresponding to each candidate node within a preset number of historical days, and performs the sorting according to the criteria defined in step S303. The assignment rules calculate the historical modified weights of each candidate node and substitute them into the comprehensive root cause weight formula for recalculation. The node with the highest adjusted weight is selected as the root cause alarm. Adjusted overall root cause weight. The calculation formula is ,in; This is the historical correction factor, preset to 0.2; The historical correction weight is calculated by querying the maintenance records and defect ledgers of the equipment within a preset number of historical days through a cross-disciplinary data interface.

[0086] In this embodiment of the application, if the device corresponding to a certain root cause candidate node only has defect records within a preset number of historical days, then the historical correction weight of the root cause candidate node is... If the device corresponding to a certain root cause candidate node only has maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node is adjusted. If the device corresponding to a certain root cause candidate node has both defect records and maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node will be adjusted. If the device corresponding to a certain root cause candidate node has no defect records and no maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node is adjusted. .

[0087] The execution engine selects the node with the highest overall root cause weight as the final root cause alarm, and uses this node as the sole causal starting point for the subsequent steps S304 (physical secondary alarm determination) and S305 (logical secondary alarm determination). That is, all "downstream path traversal" and "control logic prediction sequence" in subsequent steps S304 and S305 are calculated solely based on this final root cause alarm. Nodes with an in-degree of 0 do not participate in the secondary alarm propagation path traversal analysis in steps S304 and S305, and are marked as independent concurrent faults or temporarily suspended depending on the specific scenario.

[0088] In this embodiment of the application, for the remaining candidate nodes with an in-degree of 0, the execution engine performs the following traffic splitting determination based on their topological association with the final root cause alarm node, specifically: If there is no connection path between the node and the final root cause alarm node in the alarm propagation graph, the alarm is determined to be an independent concurrent fault unrelated to this fault, and the execution engine marks it as an independent concurrent fault. If the node and the final root cause alarm node have a common downstream alarm node or a shared propagation path in the alarm propagation graph, the root cause relationship of the alarm is determined to be temporarily pending, and the execution engine marks it as pending.

[0089] In this embodiment, for a node marked as an independent concurrent fault, the execution engine generates an independent secondary alarm entry for the alarm signal corresponding to that node. This entry is mounted as a collapsed list below the main alarm entry in the operation interface. It does not trigger an independent operation and maintenance response plan matching process, but retains complete alarm details and spatiotemporal tags for operation and maintenance personnel to review and process after the main root cause event has been handled.

[0090] In this embodiment, for nodes marked as pending, the execution engine serializes and stores a snapshot of the node's state (including trigger time, associated devices, weight scores, and associated upstream and downstream node information) into a pending queue. This snapshot does not participate in the suppression and contingency plan matching process of this root cause alarm. After the closed-loop handling of this root cause alarm event is completed, the operation and maintenance personnel manually trigger the generation of a secondary analysis task through the operation interface. At that time, the execution engine takes the pending node as the starting point and re-executes the directed graph search and secondary alarm marking process in step S3. This secondary analysis is not constrained by the first alarm signal and analysis window described in step S2 and independently starts a new local analysis window.

[0091] Through this mechanism, when multiple candidate nodes with an in-degree of 0 have similar original comprehensive root cause weights, the node with recent maintenance or defect records will be given priority as the root cause.

[0092] Step S304: Mark secondary physical alarms based on physical propagation laws. After identifying the root cause alarm, the execution engine begins to traverse downstream along the directed edges of the alarm propagation graph, determining whether other alarms on the path are secondary physical alarms according to preset physical propagation laws. These physical propagation laws specifically include thermodynamic conduction delay laws, fluid pressure fluctuation propagation laws, and electromagnetic transient response laws.

[0093] Regarding the thermodynamic conduction delay law, the execution engine invokes the thermodynamic delay calculation logic during the judgment process. For example, in a scenario where the bearing temperature rises after the cooling water flow disappears, this logic is based on the specific heat capacity of the bearing material. and thermal resistance Calculate the expected delay time If the actual trigger time of the temperature alarm falls after the trigger time of the flow alarm... If the temperature alarm is within the preset tolerance range, it is determined that the temperature alarm conforms to the thermodynamic conduction law and is marked as a physical secondary alarm caused by the disappearance of flow.

[0094] Following the same logic, the propagation laws of fluid pressure fluctuations and electromagnetic transient response are used to determine the secondary propagation relationships of pressure alarms and electrical quantity alarms, respectively. If an alarm signal is located downstream of another alarm signal in the alarm propagation spectrum, and the difference between their trigger times satisfies the preset time delay interval of the corresponding physical law, then the signal is determined to be a secondary alarm caused by an upstream fault and is marked accordingly.

[0095] Step S305: Mark secondary alarms based on control logic rules. The automatic control system of the hydropower station will perform preset protection actions in case of fault conditions, and these actions will also generate corresponding alarm signals. For the identification of such signals, the execution engine calls the logic judgment mechanism, which has a pre-set control logic rule library covering automatic reclosing logic, unit shutdown protection logic, and automatic backup power supply logic.

[0096] When a root cause alarm is triggered, the execution engine predicts the action feedback sequence that the relevant controlled object should produce based on the protection logic entries that match the current root cause type in the logic rule base. For example, the protection logic entry corresponding to a certain type of fault is "start the shutdown solenoid valve → close the guide vane → disconnect the demagnetizing switch → trip the main transformer high-voltage side circuit breaker".

[0097] The execution engine compares the actually monitored signals with the predicted action feedback sequence. If the timestamp deviation is less than a preset threshold and the state values ​​are completely consistent, then these actually monitored alarm signals that match the predicted sequence are uniformly marked as logical secondary alarms. This indicates that these signals are the result of the protection system operating according to predetermined logic, rather than new independent faults.

[0098] It should be noted that the secondary alarms marked in step S304 are determined based on physical propagation laws (thermodynamic conduction delay, fluid pressure fluctuation propagation, electromagnetic transient response), and are classified as physical secondary alarms. The signal types they target are accompanying signals caused by the natural propagation of the physical effects of the fault, such as temperature increases, pressure fluctuations, and changes in electrical quantities. The controlled secondary signals marked in step S305 are determined based on the control logic rule base (automatic reclosing logic, unit shutdown protection logic, automatic backup power supply switching logic), and are classified as logic secondary alarms. The signal types they target are control commands and action feedback signals actively issued by the automatic control system after a fault occurs, such as protection trips, valve closures, and equipment switching. In this embodiment, these two types of signals are filtered through independent parallel filtering paths and are both suppressed and popped up in step S4.

[0099] Preferably, when a closed-loop structure exists in the alarm propagation graph, such as a feedback adjustment loop, the bidirectional connection of directed edges can cause causal judgment to fall into a logical loop. A loop-breaking search strategy is used to handle the loop topology. Specifically, the execution engine traverses the alarm propagation graph. When a directed closed loop consisting of several nodes is detected, the timestamps of the state changes of all nodes within the loop are extracted. The execution engine defines the node that first triggered the alarm on the timeline as the logical starting point of the loop, and then temporarily removes all reverse edges pointing to this logical starting point in the alarm propagation graph, thereby transforming the original loop-bearing graph into a temporary directed acyclic graph. On this temporary directed acyclic graph, the root cause search and secondary labeling process from steps S302 to S305 is re-executed to complete the causal analysis of the alarm signals within the loop. After the analysis is completed, the temporarily removed reverse edges are restored.

[0100] By employing a loop-breaking search strategy to handle loop topologies, the directed graph algorithm gains the ability to process complex closed-loop relationships, avoiding analysis dead loops in special structures such as regulation systems and redundant power supplies.

[0101] In summary, through step S3, the execution engine uses a directed graph algorithm as its core to search for nodes with an in-degree of 0 in the alarm propagation graph to locate root cause candidate nodes. It then sorts multiple candidates based on topological density and time order to determine the root cause alarm. Next, it uses physical propagation laws and control logic rules to perform physical and logical secondary marking on the downstream signals of the root cause alarm. Finally, it uses a loop-breaking search strategy to solve the analysis problem of closed-loop topology.

[0102] At this point, the system has extracted the root cause alarm and the complete causal chain from the alarm avalanche. Once the root cause is identified, the system proceeds to step S4, where the execution engine automatically matches the operation and maintenance response plan based on the root cause alarm and pushes it to multiple terminals of the operation and maintenance personnel.

[0103] Finally, step S4 is executed: the execution engine automatically suppresses the independent pop-up of alarm prompts corresponding to secondary alarms, and matches the corresponding operation and maintenance response plan according to the determined root cause alarm signal.

[0104] This step follows the root cause alarm and complete causal chain labeling results output in step S3, entering the alarm presentation and maintenance response phase. The execution engine performs hierarchical filtering of alarm information on the interface, quickly retrieves handling solutions based on root cause alarm signals, and closes the task loop to on-site personnel through a multi-terminal collaboration mechanism. Its specific implementation process can be broken down into the following sub-steps.

[0105] Step S401: Generate a primary alarm entry and collapse secondary alarms. The execution engine suppresses all pop-ups marked as physical secondary alarms in step S304 and logical secondary alarms in step S305 on the operation interface of the maintenance personnel. Based on the root cause alarm locked in step S3, the execution engine generates a corresponding primary alarm entry. The information items of the primary alarm entry include the full name of the identified root cause alarm signal, the time of occurrence, the physical location involved, the root cause confidence value, the type identifier, the professional category of the equipment, and the fault level. All signals marked as secondary alarms are collapsed and stored in the subordinate details list of the primary alarm entry. This details list is in a collapsed and hidden state by default, and is only expanded and displayed when the maintenance personnel actively retrieve it through click operation or voice command. This design effectively prevents alarm avalanche from interfering with personnel judgment.

[0106] It should be noted that the root cause confidence score is calculated by those skilled in the art based on actual needs. In the embodiments of this application, the root cause confidence score is calculated by normalizing the comprehensive root cause weight of each root cause candidate node.

[0107] In this embodiment, the type identifier is the functional protection category or fault nature classification label to which the root cause alarm signal belongs. This label is pre-stored in the attribute field of the corresponding node in the physical logic fusion topology model. After the execution engine determines the root cause alarm, it directly reads it from the node attribute, including but not limited to differential protection action, overcurrent protection action, zero sequence protection action, temperature over-limit, pressure abnormality, cooling system complete shutdown, control loop disconnection, etc. The professional category of the equipment is automatically determined by the execution engine according to the hierarchical branch of the node where the root cause alarm is located in the equipment tree structure, including primary electrical, secondary electrical, hydraulic machinery, auxiliary equipment and automation control professional categories.

[0108] In this embodiment, the fault level is determined as follows: the execution engine comprehensively calculates and maps the fault level to the corresponding level in the preset level set based on the criticality level of the node device corresponding to the root cause alarm in the device tree, the nature and category of the alarm signal, and the degree to which the signal value deviates from the preset normal operating range. The preset level set includes three levels: severe fault, general fault, and minor fault. The specific determination rules are as follows: (i) If the device corresponding to the root cause alarm belongs to the critical main equipment in the equipment tree (including water turbine, generator, main transformer, circuit breaker) and meets at least one of the following conditions, it shall be judged as a serious fault: the alarm nature is a protection trip signal or emergency shutdown signal, or the analog quantity exceeds the limit by more than 20% of the rated value; (ii) If the device corresponding to the root cause alarm belongs to the critical main equipment or important auxiliary equipment in the equipment tree (including technical water supply pump, speed governor hydraulic device, excitation system), and meets at least one of the following conditions, it shall be judged as a general fault: the alarm nature is a general over-limit signal or an abnormal status change signal, or the analog quantity over-limit amplitude exceeds 5% but does not exceed 20% of the rated value; (iii) If the device corresponding to the root cause alarm is a general auxiliary device (including drainage pumps, ventilation equipment, and lighting systems), and meets at least one of the following conditions, it shall be judged as a minor fault: the alarm is a warning signal or a reminder of the over-limit signal, or the over-limit amplitude of the analog quantity does not exceed 5% of the rated value; (iv) If the root cause alarm signal is a sensor fault or communication interruption signal, it shall be judged as a minor fault regardless of the equipment level.

[0109] Step S402: Based on the characteristics of the root cause alarm signal, retrieve the matching operation and maintenance response plan from the intelligent operation and maintenance knowledge base.

[0110] When the execution engine performs a pre-plan retrieval, it uses the type identifier of the root cause alarm signal, the professional field to which the equipment belongs, and the fault level determined in step S401 as the joint retrieval key to perform a full-text matching retrieval in the preset intelligent operation and maintenance knowledge base.

[0111] Scenario setting: The root cause alarm of a hydropower station is "differential protection of main transformer No. 1 activated", and the fault level is judged as a serious fault.

[0112] Search key extraction: Type Identifier: Differential Protection Equipment Category: Primary Electrical Equipment / Transformer Fault severity: Severe Knowledge base matching results: The knowledge base contains multiple solutions, which are sorted by matching degree after retrieval as follows: Contingency Plan A: "Transformer Differential Protection Action Handling Procedures" (Type Identifier Fully Matched, Professional Category Fully Matched, Level Matched), with a matching degree of 95%, was automatically selected as the recommended contingency plan. Contingency Plan B: "Transformer Overcurrent Protection Handling Procedures" (Type identification partially matches, professional category fully matches, and grade partially matches), matching degree 60%. Contingency Plan C: "Generator Differential Protection Action Handling Procedure" (Type identification fully matches, professional mismatch, level matches), matching degree 50%. Recommended contingency plan content (structured fields of contingency plan A): Preliminary diagnosis: An inter-turn short circuit or phase-to-phase short circuit occurred inside the transformer windings, causing the differential current to exceed the limit. Recommended isolation measures: Confirm that all circuit breakers on the main transformer are tripped, disconnect the isolating switches, and install grounding wires on both the high-voltage and low-voltage sides. Spare parts list: Main transformer high-voltage side bushing (spare part number TB-2024-03), gas relay (spare part number BJ-2024-07), transformer insulating oil (25#, 200L) Inspection Operating Procedures: Check the main transformer for abnormal noises and oil spraying; check the gas relay for gas accumulation; take an oil sample for chromatographic analysis; check the protection device action reports and fault recording waveforms. Push: The execution engine will simultaneously push the above contingency plans to the central control room's large screen, the duty computer terminal, and the handheld terminals of on-site inspection personnel.

[0113] Each maintenance response plan includes the following structured fields: preliminary fault diagnosis conclusion, recommended isolation measures, a list of required spare parts, and corresponding inspection procedures. The execution engine sorts the plans in descending order of search matching degree, and selects the plan with the highest matching degree or the top few plans as candidates, with the plan with the highest matching degree being automatically selected as the recommended plan.

[0114] In step S403, the operation and maintenance response plan is pushed to the operation and maintenance personnel through a multi-terminal collaboration mechanism. The execution engine synchronously pushes the operation and maintenance response plan matched in step S402 to multiple terminals, including the central control room large screen, the duty room computer terminal, and the portable handheld mobile terminal of the on-site inspection personnel.

[0115] Each push notification contains a unique task identifier. This identifier is linked to the root cause alarm event within the system and used to track the task's response status throughout the process. The push notification explicitly requires on-site inspectors to scan the device identification code on the faulty equipment using a handheld terminal upon arrival at the fault location to confirm their location and ensure they are indeed at the correct location. On-site inspectors can use their handheld terminals to provide real-time feedback to the execution engine, including processing progress, photographic records, and voice descriptions. Upon receiving this feedback, the execution engine updates the task status.

[0116] Preferably, step S4 further includes: for controlled objects with remote operation permissions, generating remote control instruction suggestions and executing a security confirmation process; the operation and maintenance response plan also includes automated remote control instruction suggestions. For controlled objects with remote operation permissions, the execution engine generates pre-instructions based on the plan logic. For example, when a water pump overcurrent alarm is identified as the root cause, the plan will automatically generate a remote control pre-instruction to switch to the standby pump.

[0117] The pre-instruction is first pushed to the operator's terminal as a suggestion, and is not directly executed. The system waits for confirmation from the operator on the terminal. Only after receiving confirmation from the operator does the execution engine work with the computer monitoring system to automatically perform device isolation or switching operations.

[0118] In summary, through step S4, the execution engine completes the entire process from alarm presentation to closed-loop operation and maintenance response. The primary alarm entry generation and secondary alarm folding mechanism protects operation and maintenance personnel from information overload; the contingency plan matching module and maintenance history weighting mechanism provide support for decision-making from the perspectives of knowledge base and historical records, respectively; multi-terminal collaborative push, task identification code binding, device identification code location confirmation, and feedback closed loop enable operation and maintenance instructions to be tracked and executed; and the remote control security confirmation process achieves a balance between automation and security.

[0119] After steps S1 to S4, the intelligent integrated alarm and operation and maintenance response method for hydropower stations proposed in this invention completes a full cycle of alarm triggering, source tracing, suppression, and response. This method achieves a closed-loop process across the entire chain, from alarm avalanche to precise root cause localization and automated operation and maintenance response, through the collaborative work of physical-logical fusion topology models, spatiotemporal alignment of sliding time windows, directed graph causal inference, and execution engine-driven pre-plan matching and push.

[0120] Furthermore, the method also involves monitoring the state of sensor measurement accuracy. The execution engine compares data from multiple redundant sensors within the same physical area and uses deviation analysis logic to determine the authenticity of alarm signals. For example, for generator outlet voltage, it compares data from three independent voltage transformers. If the deviation of one set of measurements from the arithmetic mean of the other two sets exceeds 10%, and this state persists for more than a preset threshold, it is determined to be a false alarm caused by a sensor malfunction. The execution engine directly marks it as an illegal signal at the source, preventing it from entering the subsequent topology analysis and suppression process.

[0121] Example 2

[0122] Based on Example 1, this example further refines the calculation process of node attributes and their dynamic weights in the physical-logical fusion topology model, and provides an application illustration in the context of a hydropower station's "transformer cooling system complete shutdown" fault scenario.

[0123] When constructing the physical-logical fusion topology model in step S1, the system pre-defines a multi-dimensional state vector space for each type of device node. The state vector includes a real-time operating condition vector, a static physical constraint vector, and an environmental impact vector. The real-time operating condition vector contains continuously changing analog data such as current, voltage, speed, flow rate, pressure, and temperature. The static physical constraint vector includes the device's rated power, temperature rise tolerance, material strength limits, and geometric topology. The environmental impact vector includes external inputs such as ambient temperature and humidity, power grid frequency fluctuations, and vibration interference.

[0124] In graph databases, edge attributes include not only direction but also dynamic weight coefficients. Dynamic weight coefficients The calculation method is as follows: in step S103, the static weights Based on this, multiply by an equipment health coefficient related to the equipment's lifecycle. ,Right now .in, Adjustments are made based on the current stage of the equipment's lifecycle; the calculation formula is as follows: Here This refers to the cumulative operating hours of the equipment. The rated design life hours of the equipment. This is an aging acceleration factor, with an empirical value ranging from 0.1 to 0.5, determined based on equipment type and operating environment. Exceed hour, The continuous increase in weighting means that, during alarm analysis, older devices that have been running for a longer period of time have a higher weighting coefficient for their physical connection edges, making them more likely to be identified as sources of failure.

[0125] In step S2, the execution engine incorporates signal fingerprinting technology while monitoring signals across the entire plant. For each acquired analog data stream, the execution engine not only monitors whether it exceeds the alarm threshold but also uses a sliding window to calculate the signal's rate of change (first derivative) and acceleration (second derivative). When the signal's rate of change exceeds a preset abrupt change threshold, the execution engine will preheat the analysis window even if the value has not yet reached the alarm limit. For switch signals, the execution engine records their jitter frequency. If more than three flips occur within 500 milliseconds, it is automatically marked as signal interference and smoothed and filtered before topology analysis to prevent the generation of a large number of false alarm nodes.

[0126] It should be noted that "preheating" refers to the operation of preparing resources and conducting pre-analysis in advance when the system detects the beginning of a signal change but has not yet triggered an alarm. The purpose is to shorten the response delay between alarm triggering and formal analysis.

[0127] In this embodiment, the switching signal has only two states: 0 or 1, open or closed, active or reset. Examples include: circuit breaker position signal (closed / open), pressure switch signal (normal / abnormal), and valve position signal (fully open / fully closed). "Flip" refers to a switching signal changing from one state to another, for example, from 0 to 1 (active) and from 1 to 0 (reset). Each state change is considered a flip. Jitter refers to multiple unexpected and rapid flips of the switching signal within a short period.

[0128] In this embodiment of the application, the cause of jitter is as follows: Mechanical contact bounce: When the mechanical contacts of circuit breakers, relays, contactors and other equipment are closed or opened, due to the bouncing effect, they will make and separate multiple times in a very short time (tens to hundreds of milliseconds), causing the signal to jump rapidly between 0 and 1.

[0129] Electromagnetic interference: Strong electromagnetic fields (such as circuit breaker opening and closing operations, lightning strikes) induce pulse interference on signal lines, causing brief signal jumps.

[0130] Loose wiring or poor contact: Oxidation or loosening at the signal line connection causes intermittent interruption.

[0131] For example, when a circuit breaker closes normally, the auxiliary contact should smoothly transition from 0 to 1. However, due to contact bounce, the actual signal may jump multiple times within 50 milliseconds from 0 to 1 to 0 to 1 until the contact stabilizes completely. If this is not addressed, the system will mistakenly interpret it as the circuit breaker experiencing multiple abnormal operations.

[0132] Normal equipment state changes are typically a one-time toggle (e.g., circuit breaker closing: 0→1 toggle once), or, in rare cases, a brief jitter within 100 milliseconds. If more than 3 toggles (i.e., at least 4 state changes) occur within 500 milliseconds, the signal is highly likely a false jump caused by mechanical contact bounce or electromagnetic interference, rather than a genuine equipment state change. Therefore, the execution engine marks it as signal interference and filters it before subsequent analysis to prevent these false toggles from being included in the signal set to be analyzed as real alarm signals, thereby avoiding the generation of a large number of false alarm nodes.

[0133] In this embodiment of the application, "smoothing filtering processing" refers to the process by which the execution engine performs de-jitter processing (anti-bouncing processing) on ​​signals marked as interference before topology analysis. The method is as follows: Method 1, Time Delay Determination (Software Debouncing): Set a debouncing time window (e.g., 50 milliseconds). Within this time, the signal must stably maintain a certain state before the state is considered valid. Specifically: record the current stable state value. When a flip is detected, do not immediately update the state. Instead, start a 50-millisecond timer. If the signal still maintains the new state after the flip when the timer ends, the flip is considered valid and the state is updated; if it flips back to the original state before the timer ends, the flip is ignored.

[0134] For example, a circuit breaker contact bounces, causing a rapid 0→1→0→1 transition. When the system detects the first transition (0→1), it does not immediately assume the circuit breaker is closed, but instead starts a 50-millisecond timer. Since the signal flips back to 0 after 30 milliseconds, the system ignores this 0→1 transition and continues monitoring. Finally, after the signal stabilizes at 1 for more than 50 milliseconds, the system confirms that the circuit breaker is truly closed and outputs a valid transition.

[0135] Method 2: Multiple sampling and voting: Sample the signal multiple times within a short time window (e.g., 10 milliseconds), take the state that appears most frequently as the true state at that moment, and filter out single-transition pulses.

[0136] Before performing topology path analysis in step S3, the execution engine first performs the aforementioned smoothing and filtering process on all signals in the plant. This ensures that all signals entering the subsequent causal inference stage are genuine and valid state changes, preventing false flips from generating a large number of alarm nodes. Without filtering, the five flips generated by a circuit breaker's closing operation might be misjudged as five independent alarm signals, interfering with root cause analysis. After filtering, the five flips are merged into one valid closing event, accurately reflecting the equipment status and significantly reducing false alarms.

[0137] In step S3, regarding the causal tracing process of the "transformer cooling system complete shutdown," the execution engine uses a directed graph algorithm combined with electromagnetic-thermal coupling logic for deduction. When the main transformer cooler power supply fault alarm is triggered, the system automatically identifies the affected cooling fan group, submersible oil pump group, and transformer winding temperature rise node in the topology graph. The directed graph algorithm calculates an expected temperature rise trajectory curve based on the physical inertia of transformer oil temperature changes. The calculation method is: based on the initial oil temperature before the cooler stops. Ambient temperature and the thermal time constant of the transformer Calculate the expected oil temperature curve according to the exponential law. ,in The time elapsed since the cooling system was shut down, its unit is the transformer thermal time constant. The units are the same (usually seconds or minutes) to ensure that the exponent is dimensionless. The execution engine continuously compares the measured temperature rise curve with the expected trajectory in real time.

[0138] If the measured rate of increase in temperature is within the expected curve If the temperature rise alarm is within the preset allowable deviation range, it is marked as a "confirmed secondary signal." If the measured temperature rise rate is abnormally high, exceeding the upper limit of the allowable deviation, the execution engine will search upstream through the topology path to determine whether there is an internal winding short circuit or other accompanying faults. When marking secondary alarms, the execution engine also considers the redundancy design of the hydropower station. For example, if the simultaneous loss of two plant power supplies causes the cooler to stop, the execution engine will, based on the logic deduction of the automatic switching system (ATS), attribute all instantaneous undervoltage alarms during the power switching process to the primary cause of "plant power supply switching."

[0139] In step S3 above, for signal identification that conforms to the control logic deduction results, the execution engine uses a built-in lightweight simulator to mirror the logic instructions of the PLC. This lightweight simulator is implemented by pre-converting the ladder diagram programs of each PLC in the hydropower station into equivalent Boolean logic expressions and timer / counter combinations using an offline parsing tool, and storing them in the simulator as a directed acyclic graph.

[0140] When a root cause alarm is generated, the simulator uses the alarm signal as the simulation trigger condition. It performs a directed traversal of the graph, calculating and outputting step-by-step based on the truth tables of each logic gate and the timer settings, ultimately generating the expected protection action sequence, such as "start the shutdown solenoid valve → close the guide vane → disconnect the demagnetizing switch → trip the main transformer high-voltage side circuit breaker". If the actual generated alarm sequence deviates by less than 200 milliseconds in time and the logical order is completely consistent, the execution engine will merge and fold this series of action signals, displaying only "Unit protection action shutdown" on the alarm interface, while hiding the specific trip signal in the secondary list.

[0141] In step S4, the matching logic for the operation and maintenance response plan incorporates a semantic recognition model based on deep learning. This deep learning-based semantic matching model can replace or combine with the full-text matching retrieval in Example 1 to achieve a semantic-level mapping from unstructured fault descriptions to response plans. The training data for the model is constructed as follows: historical fault reports from the past 10 years are collected, each report is manually annotated, and the "fault phenomenon description" is extracted as the input text, and the "final operation and maintenance response plan number" is used as the supervision label.

[0142] The input text is processed by a BERT pre-trained language model, which maps each fault description into a 768-dimensional feature vector. Structurally, a fully connected classification layer is chained above this BERT encoding layer. Its output dimension equals the total number of solutions N in the knowledge base. The output vector is then normalized using Softmax to obtain the matching probability of each solution. The cross-entropy loss function is used, and the training objective is to minimize the cross-entropy between the model's predicted probability distribution and the one-hot encoding of the true solution labels.

[0143] After model training is complete, the execution engine vectorizes the current root cause alarm features during online inference, searches for the fault case with the highest matching probability in the semantic space, and retrieves the corresponding operation and maintenance response plan. The plan pushed to the handheld terminal not only includes the text operation procedure, but also historical fault photos of the fault point, 3D structural model cross-sectional diagrams, and the process requirements of the most recent maintenance. When on-site inspection personnel execute the plan, they point the handheld terminal's AR augmented reality function at the faulty equipment, and the system automatically overlays and displays the equipment's internal structure, real-time current and voltage values, and bolt torque values ​​that should be checked, achieving seamless integration of information and physical reality.

[0144] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0145] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for intelligent integrated alarm and operation and maintenance response of hydropower stations based on an execution engine, characterized in that, include: S1. Construct a physical-logical integrated topology model; S2. Based on the physical logic fusion topology model, the execution engine monitors all plant signals in real time. When the first alarm signal is detected, an analysis window is opened with the trigger time of the first alarm signal as the center. The concurrent alarm signals in the analysis window are spatiotemporally aligned to form a set of signals to be analyzed. S3. Based on the physical-logical fusion topology model, the topological path between each alarm node in the signal set to be analyzed is analyzed using a directed graph algorithm, thereby locking the root cause alarm and marking the corresponding physical secondary alarm and logical secondary alarm based on the root cause alarm. S4. Based on the root cause alarm, the execution engine automatically suppresses the independent pop-up of alarm prompts corresponding to secondary alarms, and matches the corresponding operation and maintenance response plan according to the characteristic information of the root cause alarm, and executes the operation and maintenance response.

2. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine as described in claim 1, characterized in that, Step S1 is as follows: S101. Construct a four-level device tree structure that reflects the digital abstraction of the controlled and perceived objects across the entire hydropower station area; S102. Define and establish physical and logical connections between the nodes of the device tree structure to form a fused topology diagram; S103. Store the fused topology graph in a graph database, with nodes representing device entities and edges representing physical or logical relationships between devices, and assign direction and weight attributes to each edge to obtain a physical-logical fused topology model.

3. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine as described in claim 2, characterized in that, The device tree structure in step S101 is divided from top to bottom into plant level, unit or common system level, specific equipment level, and measurement point signal and status level. Among them, the plant level stores the global metadata of the plant; the unit or common system level divides the entire station into relatively independent logical function blocks; the specific equipment level lists in detail the physical components that make up each unit, including primary equipment, secondary equipment, and process auxiliary systems; the measurement point signal and status level includes all analog data reported by sensors, switch status bits, hard-wired signals of protection actions, and internal logic variables. In step S102, the physical connections include electrical connections for power transmission, rigid connections for mechanical energy transmission, and pipeline connections for fluid medium transmission; the logical connections include command interaction relationships for control loops, constraint relationships for interlocking loops, and triggering causal relationships for protection loops. In step S103, a weight attribute value is assigned to each edge according to the preset density calculation rules. Specifically, for any edge, its weight is... The weights are determined by the base weight, distance attenuation factor, and dielectric conductivity, and the calculation formula is as follows: ， In the formula, The weight attribute value of the edge; It is based on the preset weight of the connection type, which includes electrical connection, mechanical connection, fluid pipeline connection and logical connection; The distance attenuation factor is a parameter used to quantify the attenuation effect of physical or electrical distance between nodes on the association strength. It is calculated based on the actual physical length of the edge. The medium conduction coefficient is used to characterize the difference in the transmission efficiency of different transmission media to fault disturbances or control signals. It is determined according to the type of transmission medium, which includes copper conductors, fluid pipelines with different roughness, and logic signals.

4. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine according to claim 1, characterized in that, Step S2 specifically includes: S201, The execution engine acquires the signal data stream of the entire plant in real time through the distributed acquisition gateway; S202. Set the width of the sliding time window, and dynamically determine the width of the sliding time window based on the weighted average of the electromagnetic transient response time and thermodynamic response time of the hydropower station. The calculation formula is as follows: ， In the formula, The width of the sliding time window. This refers to the typical transient process time of circuit breaker operation or voltage drop. This represents the typical process time for changes in bearing temperature or oil temperature. and These are the weighting coefficients, and ; S203. When the execution engine detects the first alarm signal, it locks the trigger time of the first alarm signal and uses that time as the analysis start point. Then, it opens an analysis window based on the width of the sliding time window, with the analysis start point as the center. S204. In the time dimension, a high-precision time synchronization protocol is used to project all signals falling within the analysis window onto a unified reference time axis, eliminating the deviation of each signal in time recording. S205. In the spatial dimension, taking the device node where the first alarm signal is located as the center, the physical logic fusion topology model is used to search for all relevant nodes within the preset topology distance, and to extract all accompanying alarm signals generated by all relevant nodes during the analysis window. S206. The execution engine synchronously associates the time-aligned information from step S204 with the accompanying alarm signal obtained from spatial aggregation in step S205, thereby forming a set of signals to be analyzed with a unified time reference and a clear spatial topology.

5. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine according to claim 1, characterized in that, Step S3 specifically includes: S301. Map all alarm signals in the signal set to be analyzed one by one to the corresponding nodes of the physical logic fusion topology model, extract the local subgraph containing only the current alarm node, and express the relationship between nodes as a directed connection according to the preset direction attribute of each edge to form an alarm propagation map. S302. Search for all nodes with an in-degree of zero in the alarm propagation graph and extract all nodes with an in-degree of zero as root cause candidate nodes. S303. If there is only one root cause candidate node, directly use that root cause candidate node as the root cause alarm; otherwise, calculate the comprehensive root cause weight of each root cause candidate node, and select the root cause candidate node with the highest comprehensive root cause weight value as the root cause alarm. S304. After locking the root cause alarm, traverse downstream along the directed edge of the alarm propagation map, and determine whether other alarm signals on the path belong to physical secondary alarms according to the preset physical propagation law. The preset physical propagation law includes thermodynamic conduction delay law, fluid pressure fluctuation propagation law and electromagnetic transient response law. Specifically, for an alarm signal that conforms to any one of the physical propagation laws of thermodynamic conduction delay law, fluid pressure fluctuation propagation law and electromagnetic transient response law, if the alarm signal is located downstream of another alarm signal in the alarm propagation map and the difference between the triggering times of the two signals satisfies the preset time delay interval of the corresponding physical law, then the alarm signal is marked as a physical secondary alarm caused by an upstream fault. S305. After locking the root cause alarm, the logic secondary alarm is marked based on the control logic rules. Specifically, the action feedback sequence that the relevant controlled object should generate is predicted according to the protection logic entry that matches the current root cause alarm type in the logic rule library. Then, the execution engine compares the actual monitored signal with the predicted action feedback sequence. If the timestamp deviation between the two is less than the preset threshold and the state value is completely consistent, then these actual monitored alarm signals that match the predicted sequence are uniformly marked as logic secondary alarms.

6. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine according to claim 5, characterized in that, In step S303, the comprehensive root cause weight is jointly determined by the topological compactness weight and the time priority weight. The topological compactness weight is calculated as follows: ， In the formula, For topologically compact weights, The total shortest path length from the candidate root cause node to each downstream alarm node in the alarm propagation graph. The formula for calculating the overall root cause weight is: ， In the formula, To weigh the root causes; This is a preset coordination coefficient, ranging from 0 to 1, used to balance the relative importance of topological and time factors in decision-making; This is a time priority weight.

7. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine as described in claim 6, characterized in that, The total shortest path length from any root cause candidate node in the alarm propagation graph to all downstream alarm nodes is calculated using one of the following two methods, depending on whether edge weights are configured: (1) Unweighted path length: Used when no edge weights are configured, it is the sum of the number of edges contained in the path. The specific calculation method is as follows: The execution engine searches all directed paths from the root cause candidate node to each reachable downstream alarm node, determines the directed path with the fewest number of edges as the corresponding shortest path, and adds up the number of edges contained in the shortest path from the root cause candidate node to each reachable downstream alarm node to obtain the total length of the shortest path of the root cause candidate node. (2) Weighted path length: Used when edge weights are configured for each edge, it is the sum of the equivalent propagation costs of all edges contained in the path. The specific calculation method is as follows: The execution engine calculates the sum of the equivalent propagation costs of all edges contained in each directed path, and determines the directed path with the smallest sum of equivalent propagation costs as the corresponding weighted shortest path. The weighted shortest path lengths from the root cause candidate node to each reachable downstream alarm node are added together to obtain the total weighted shortest path length of the root cause candidate node. The formula for calculating the equivalent propagation cost of any edge is as follows: ， In the formula, For the equivalent propagation cost of any edge, This is a preset minimum positive number used to avoid a denominator of zero; Step S303 also includes incorporating maintenance history data to correct the comprehensive root cause weights, specifically as follows: (1) The execution engine obtains maintenance records and defect ledgers from the power plant operation and management system through cross-disciplinary data interfaces; (2) Based on the defects and maintenance records of the equipment corresponding to each root cause candidate node within a preset number of historical days, assign a corresponding historical correction weight to the root cause candidate node, specifically: If the device corresponding to a certain root cause candidate node has only defect records within a preset number of historical days, then the historical correction weight of that root cause candidate node is... ; If the device corresponding to a certain root cause candidate node has only maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node is... ; If the device corresponding to a root cause candidate node has both defect records and maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node is adjusted. ; If the device corresponding to a certain root cause candidate node has no defect records and no maintenance records within a preset number of historical days, then the historical correction weight of that root cause candidate node is adjusted. ; (3) Substitute the historical adjusted weights into the formula for calculating the comprehensive root cause weights for correction. The formula for calculating the corrected comprehensive root cause weights is as follows: ， In the formula, This is a historical correction factor. Adjust the weights based on historical data.

8. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine according to claim 1, characterized in that, Step S3 also includes a loop-breaking search strategy to handle closed-loop structures in the alarm propagation map, specifically: (1) The execution engine traverses the alarm propagation graph. When a directed closed loop consisting of several nodes is detected, the state change timestamps of all nodes in the directed closed loop are extracted, and the node that first triggers the alarm on the time axis is defined as the logical starting point of the directed closed loop. (2) Temporarily remove the incoming edges pointing to the logical starting point in the alarm propagation graph, and transform the original cyclic graph into a temporary directed acyclic graph; (3) Perform the root cause search and secondary alarm marking process on the temporary directed acyclic graph. After completing the causal analysis of the alarm signals in the directed closed loop, restore all the reverse edges that have been temporarily released.

9. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine according to claim 1, characterized in that, Step S4 specifically includes: S401. The execution engine suppresses secondary alarms on the operation interface of the maintenance personnel and pops them up as independent alarm entries. It then collapses and displays them in the details list of the corresponding main alarm entry. Subsequently, it generates the corresponding main alarm entry based on the locked root cause alarm and displays it. The information items of the main alarm entry include the full name of the identified root cause alarm signal, the time of occurrence, the physical location involved, the root cause confidence value, the type identifier, the professional category of the equipment, and the fault level. S402. Using the root cause alarm type identifier, the equipment's professional category, and the fault level as the joint search key, search the preset intelligent operation and maintenance knowledge base and match the corresponding operation and maintenance response plan. S403. The matched operation and maintenance response plan is pushed to the operation and maintenance personnel through a multi-terminal collaboration mechanism. Each push instruction in step S403 contains a unique task identification code, which is bound to the root cause alarm event and used to track the response status of the root cause alarm event throughout the process. On-site inspection personnel use handheld terminals to scan the device identification code on the faulty device to locate and confirm the location, and feed back the processing progress to the execution engine in real time.

10. The intelligent integrated alarm and operation and maintenance response method for hydropower stations based on an execution engine according to claim 1, characterized in that, Step S1 also includes: introducing operation mode perception variables into the physical-logical fusion topology model, and establishing a mapping library between operation modes and model configurations, so that the model can dynamically adapt to the changing operating conditions of the hydropower station. The operation modes include power generation mode, pumping mode, phase adjustment mode and maintenance mode.

Images