Power plant health state anomaly ai monitoring method based on large model data analysis

Abstract

The application relates to the technical field of power station intelligent monitoring, and discloses a power station health state anomaly AI monitoring method based on large model data analysis. The method comprises the following steps: synchronously acquiring power station multi-modal operation data and environment background information, constructing a power station global operation situation topology graph reflecting the dynamic correlation relationship between devices, and extracting multi-granularity abnormal signs therefrom; using a large model to analyze historical fault cases, constructing and dynamically updating an abnormal knowledge base containing fault evolution modes and strategies; matching and reasoning the real-time abnormal signs and the knowledge base modes, generating a multi-layer abnormal decision flow; optimizing the decision flow through a collaborative analysis engine, forming a monitoring conclusion, outputting a power station health state portrait and a corresponding operation and maintenance decision cluster, and building a complete numerical sequence physical model for spatial intelligence of a smart power station, so that precise perception and intelligent diagnosis of global and correlation anomalies of the power station are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring of power stations, in particular to a power station health state anomaly AI monitoring method based on large model data analysis. BACKGROUND

[0002] Current power station anomaly monitoring generally relies on threshold alarm and isolated analysis of discrete device operating parameters. The conventional technical solution sets static alarm thresholds at each monitoring point, and makes over-limit judgments on individual data such as current, voltage, and power of inverters and strings. This method regards the power station as a simple collection of independent monitoring points, and can only capture explicit faults of the device itself, and cannot depict the mutual influence and implicit correlation between devices due to electrical connection and environmental coupling. When the anomaly is transmitted through the internal network of the power station or caused by multiple device slight deterioration, isolated analysis is difficult to locate the source, and is prone to produce a large number of false alarms, and cannot reflect the overall operation health situation of the power station.

[0003] Existing fault diagnosis relies on a pre-set rule base or a classification model based on historical data, and its knowledge base is usually composed of simple "symptom-fault" correspondences summarized by humans, which is difficult to update and maintain. The decision logic of this method is single and linear, and lacks the ability to analyze complex fault evolution processes and multi-factor coupling. In the face of new faults or complex scenarios where multiple abnormal signs are intertwined, the static rule base is difficult to match, and the diagnosis conclusion is often one-sided, and cannot generate an optimal operation and maintenance decision sequence that comprehensively considers fault probability, evolution risk, and disposal cost. Power station operation and maintenance still needs to rely on human experience for secondary research and judgment, and there are bottlenecks in efficiency and accuracy. SUMMARY

[0004] The purpose of the present application is to provide a power station health state anomaly AI monitoring method based on large model data analysis to solve the problems raised in the background.

[0005] To achieve the above purpose, the present application provides a power station health state anomaly AI monitoring method based on large model data analysis, which comprises:

[0006] Synchronously acquiring multi-modal operating data generated by devices in the power station during operation and associated environmental background information;

[0007] Using the multi-modal operating data to construct a power station global operating situation topology graph that can reflect the global dynamic correlation of the power station;

[0008] Based on the power station global operating situation topology graph and the environmental background information, extracting power station operating multi-granularity abnormal signs with different semantic levels;

[0009] Deep analysis of historical fault cases using large models, construction and dynamic updating of an abnormal knowledge base of power stations containing fault evolution patterns and coping strategies;

[0010] Real-time matching and reasoning of the multi-granularity abnormal signs of power station operation with the fault evolution patterns in the abnormal knowledge base of power station;

[0011] According to the matching and reasoning results, a multi-layer abnormal decision flow for the current abnormal situation is generated;

[0012] Start a collaborative analysis engine to fuse and optimize the multi-layer abnormal decision flow, and generate the final monitoring conclusion;

[0013] Based on the monitoring conclusion, output the power station health status portrait representing the overall health status of the power station and the operation and maintenance decision cluster corresponding thereto.

[0014] Preferably, the multi-modal operation data refers to a heterogeneous data set covering electrical parameters, device status, operation logs and communication messages collected by different types of sensors and monitoring devices during the operation of the power station.

[0015] Preferably, the power station global operation situation topology graph capable of reflecting the global dynamic correlation relationship of the power station is constructed using the multi-modal operation data, which specifically includes:

[0016] Clean and align the multi-modal operation data to eliminate the misalignment of data in time and space dimensions;

[0017] Identify the correlation relationship between each physical device, electrical connection and logical control unit in the power station from the cleaned and aligned data;

[0018] Based on the correlation relationship, calculate the dynamic influence weight between each correlation node;

[0019] According to the dynamic influence weight and the correlation relationship, generate a power station global operation situation topology graph composed of nodes and edges with dynamically variable weights.

[0020] Preferably, based on the power station global operation situation topology graph and the environmental background information, the power station operation multi-granularity abnormal signs with different semantic levels are extracted, which specifically includes:

[0021] Traverse the nodes and edges in the power station global operation situation topology graph to identify abnormal nodes and abnormal connection edges whose operation parameters deviate from the predetermined reference range;

[0022] Analyze the inducing or aggravating effect of the environmental background information on the abnormal nodes and the abnormal connection edges to form environmental coupling abnormal features;

[0023] In combination with the abnormal nodes, the abnormal connection edges, and the environmental coupling abnormal features, multi-granularity abnormal signs of power station operation are mined from three dimensions of equipment level, branch level, and system level, respectively, to represent local abnormality, correlation abnormality, and system-level risk.

[0024] Preferably, a large model is used to conduct in-depth analysis on historical fault cases, and an abnormal knowledge base of power stations is constructed and dynamically updated, which includes fault evolution patterns and coping strategies. Specifically, the abnormal knowledge base of power stations includes:

[0025] Historical fault case reports, maintenance records, and corresponding complete operation data sequences of power stations are collected.

[0026] A large model is used to jointly analyze the historical fault case reports and the complete operation data sequences, and the initial cause, development path, key turning point, and final disposal strategy of the fault are extracted.

[0027] The extracted information is structured and stored to form an abnormal knowledge base of power stations indexed by fault patterns, which includes fault development chains and coping operation sequences.

[0028] When a new fault case is processed and confirmed, an update process is automatically started to integrate the analysis results of the new case into the abnormal knowledge base of power stations.

[0029] Preferably, the multi-granularity abnormal signs of power station operation are matched and reasoned with the fault evolution patterns in the abnormal knowledge base of power stations in real time. Specifically, the matching and reasoning include:

[0030] The multi-granularity abnormal signs of power station operation are input as queries for multidimensional retrieval in the abnormal knowledge base of power stations.

[0031] The similarity between the multi-granularity abnormal signs of power station operation and each fault evolution pattern in the abnormal knowledge base of power stations in terms of sign performance, development timing, and environmental context is calculated.

[0032] According to the similarity, a high-confidence candidate fault evolution pattern set is selected.

[0033] A large model is used to deduce the candidate fault evolution pattern set to predict the most likely development path and potential risk points of the current abnormality.

[0034] Preferably, according to the matching and reasoning results, a multi-layer abnormal decision flow for the current abnormal situation is generated. Specifically, the multi-layer abnormal decision flow includes:

[0035] According to the matched fault evolution patterns and their predicted development paths, a preliminary disposal suggestion sequence of different emergency levels is generated.

[0036] analyzing the power plant global operation situation topology, evaluating the cascading effects that the execution of the preliminary treatment suggestion sequence may cause to the non-faulty part of the power plant;

[0037] fusing the preliminary treatment suggestion sequence and the evaluation results of the cascading effects to generate a multi-layer abnormal decision flow containing execution steps, expected effects, risk prompts and backup solutions.

[0038] Preferably, a collaborative analysis engine is started to fuse and optimize the multi-layer abnormal decision flow to generate the final monitoring conclusion, which specifically includes:

[0039] The collaborative analysis engine simulates the operation rules and safety constraints of the power plant, and performs logical consistency checking and conflict detection on each operation in the multi-layer abnormal decision flow;

[0040] On the premise of meeting all operation rules and safety constraints, the operation steps with conflicts or redundancies are merged, sorted and optimized;

[0041] Combined with the real-time updated multi-modal operation data, the optimized decision flow is pre-evaluated for effect;

[0042] Output the final monitoring conclusion after checking, optimizing and pre-evaluating.

[0043] Preferably, based on the monitoring conclusion, the power plant health status portrait representing the overall health status of the power plant is output, which specifically includes:

[0044] According to the monitoring conclusion, determine the comprehensive influence level of the current abnormality on the power plant generation efficiency, equipment safety and system stability;

[0045] From the dimensions of equipment availability, system reliability and power loss, the current health degree index of the power plant is quantitatively generated;

[0046] Fuse the comprehensive influence level and the health degree index to form a visual power plant health status portrait that directly displays the overall health status, weak links and risk distribution of the power plant.

[0047] Preferably, based on the monitoring conclusion, the operation and maintenance decision cluster corresponding to it is output, which specifically includes:

[0048] Classify the operation instructions in the monitoring conclusion according to the execution subject, execution priority and execution time window;

[0049] Configure detailed operation parameters, execution conditions and acceptance standards for each type of operation instruction;

[0050] Pack all classified operation instructions with detailed configurations to form a structured operation and maintenance decision cluster that can be directly parsed and scheduled by the operation and maintenance system.

[0051] Compared with the prior art, the present application has the beneficial effects that:

[0052] By fusing multi-modal operation data to construct a dynamically weighted power station global operation situation topology graph, the monitoring object is upgraded from a discrete device node to a visualized dynamic correlation network. The weight of the edge in the graph is driven by real-time electrical, thermal, and other data, which can quantitatively represent the propagation path and influence strength of the abnormality in the physical topology. This technology realizes the capture of global and implicit abnormalities caused by correlation relationship degradation, such as accurately locating the root node that causes the system efficiency to decline in a chain due to a branch fault, and changes the limitation of traditional point-type alarms that cannot reveal the abnormal conduction rule.

[0053] By using a large model to deeply analyze unstructured historical cases, a multi-stage fault evolution pattern containing time series and causal relationship is extracted, and a dynamically expandable knowledge graph is constructed. The multi-granularity abnormal signs extracted in real time are matched with the graph to generate multiple candidate diagnosis decision flows with confidence and evidence chains, and then the collaborative analysis engine is used for fusion and optimization. This scheme simulates the reasoning process of expert consultation, can deconstruct complex and coupled abnormalities, and output an optimal decision sequence that comprehensively considers diagnosis accuracy, treatment urgency, and operation economy, so that the system has the generalization and evolution ability to handle unknown combined faults. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 A workflow diagram of the power station health state abnormality AI monitoring method based on large model data analysis described in the present application;

[0055] Figure 2 A flowchart for constructing a power station global operation situation topology graph;

[0056] Figure 3 A flowchart for constructing and dynamically updating a power station abnormality knowledge base;

[0057] Figure 4 A power grid risk and power generation loss impact analysis graph;

[0058] Figure 5 A power station health degree quantitative analysis comparison graph. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0060] Please refer toFigure 1 The application provides a power station health state anomaly AI monitoring method based on large model data analysis, which comprises the following steps: synchronously acquiring multi-modal operation data generated by equipment in a power station during operation and associated environmental background information, wherein the multi-modal operation data covers electrical parameters, equipment states, operation logs and communication messages, and the environmental background information includes illumination intensity, environmental temperature and meteorological data. A power station global operation situation topology graph capable of reflecting global dynamic correlation of the power station is constructed by using the multi-modal operation data. The topology graph represents the correlation between each physical device, electrical connection and logical control unit in the power station in the form of nodes and edges, and the weight of the edge is a dynamic influence weight. Based on the power station global operation situation topology graph and the environmental background information, power station operation multi-granularity anomaly signs with different semantic levels are extracted from three dimensions of equipment level, branch level and system level.

[0061] A large model is used to deeply analyze historical fault case reports and corresponding complete operation data sequences, extract initial causes, development paths, key turning points and final disposal strategies of faults, and then construct and dynamically update a power station anomaly knowledge base indexed by fault modes and containing fault evolution modes and coping strategies. The power station operation multi-granularity anomaly signs extracted in real time are matched and reasoned with the fault evolution modes in the power station anomaly knowledge base in real time, the similarity in sign performance, development time sequence and environmental context is calculated, candidate modes are screened and development paths are predicted. According to the matching and reasoning results, a multi-layer anomaly decision flow containing execution steps, expected effects, risk prompts and backup schemes is generated. A collaborative analysis engine is started to perform logical consistency checking, conflict detection, operation optimization and effect pre-evaluation on the multi-layer anomaly decision flow, and finally a monitoring conclusion is generated. Based on the monitoring conclusion, the comprehensive influence level is determined from the dimensions of power generation efficiency, equipment safety and system stability, and the device availability, system reliability and power loss are quantitatively generated, and a visual power station health state portrait is formed by fusion. Meanwhile, the operation instructions in the monitoring conclusion are classified according to the execution subject, priority and time window, and a structured operation and maintenance decision cluster is generated after the detailed parameters are configured.

[0062] In some embodiments, the multi-modal operation data is collected from a monitoring system of the power station. The multi-modal operation data includes electrical parameter data, device state data, operation log data, and communication message data. The electrical parameter data includes DC-side voltage and DC-side current of the inverter, AC-side voltage and AC-side current of the inverter, active power and reactive power of the inverter, voltage, current, power factor, and frequency of the box transformer and the collection line. The device state data includes internal insulated gate bipolar transistor module temperature of the inverter, DC circuit breaker opening and closing position of the busbar cabinet, winding temperature monitoring value of the box transformer, cooling fan rotating speed signal, and alarm signal and tripping signal of the protection device. The operation log data records inverter start event, inverter stop event, maximum power point tracking algorithm mode switching record, and grid-connected contactor closing and opening sequence. The communication message data captures command and response frames between the monitoring host and the inverter based on Modbus TCP protocol, polling and reporting messages between the busbar cabinet and the data collector based on RS485 bus, and manufacturing message specification messages between the box transformer protection device and the station control layer based on IEC61850 standard.

[0063] In some embodiments, the multi-modal operation data is collected from a monitoring system of the power station. The multi-modal operation data includes electrical parameter data, device state data, operation log data, and communication message data. The electrical parameter data includes DC-side voltage and DC-side current of the inverter, AC-side voltage and AC-side current of the inverter, active power and reactive power of the inverter, voltage, current, power factor, and frequency of the box transformer and the collection line. The device state data includes internal insulated gate bipolar transistor module temperature of the inverter, DC circuit breaker opening and closing position of the busbar cabinet, winding temperature monitoring value of the box transformer, cooling fan rotating speed signal, and alarm signal and tripping signal of the protection device. The operation log data records inverter start event, inverter stop event, maximum power point tracking algorithm mode switching record, and grid-connected contactor closing and opening sequence. The communication message data captures command and response frames between the monitoring host and the inverter based on Modbus TCP protocol, polling and reporting messages between the busbar cabinet and the data collector based on RS485 bus, and manufacturing message specification messages between the box transformer protection device and the station control layer based on IEC61850 standard. In some embodiments, the electrical parameter data includes DC input voltage value of 800 volts, DC input current value of 200 amperes, AC output voltage value of 400 volts, AC output current value of 150 amperes, and output active power value of 95 kilowatts of a group of inverters at 12:00 noon, and DC input voltage value of 600 volts, DC input current value of 50 amperes, AC output voltage value of 400 volts, AC output current value of 20 amperes, and output active power value of 8 kilowatts of the same group of inverters at 18:00.

[0064] In some embodiments, specific examples of the device status data include an internal temperature sensor reading of 65 degrees Celsius, a cooling fan speed sensor reading of 3000 revolutions per minute, and a DC side breaker position signal of closed state for inverter number A001; in contrast, an internal temperature sensor reading of 80 degrees Celsius, a cooling fan speed sensor reading of 4000 revolutions per minute, and a DC side breaker position signal of closed state for inverter number A002. Specific examples of the operation log data record that inverter number B001 performed an automatic restart operation sequence at timestamp 10:00:00 and completed grid synchronization at timestamp 10:00:05; in contrast, another entry in the operation log records that inverter number B001 triggered a derating operation instruction due to a temperature overload alarm at timestamp 14:30:20. Specific examples of the communication message data show that the monitoring host internet protocol address 192.168.1.10 sent a read holding register command with function code 0x03 to the inverter internet protocol address 192.168.1.20, with a starting address of 0x0001 and a register quantity of 0x0002, and the inverter returned a message data field containing a DC voltage register value of 800 and a DC current register value of 200; in contrast, the monitoring host internet protocol address 192.168.1.10 sent a read input register command with function code 0x04 to the combiner box internet protocol address 192.168.1.30, with a starting address of 0x1000 and a register quantity of 0x0006, and the combiner box returned a message data field containing six string current values of 5.1 amperes, 5.2 amperes, 5.0 amperes, 5.3 amperes, 0.1 amperes, and 5.1 amperes.

[0065] Optionally, the sampling frequency of the electrical parameter data in the multi-modal operation data is set to once per second, the sampling frequency of the device status data is set to once every 5 seconds, the operation log data is written into the log file in real time at the time point of event occurrence, and the communication message data is captured by a network packet capture tool and attached with a timestamp in milliseconds. Optionally, the storage format of the multi-modal operation data stores the electrical parameter data and the device status data in a time series database, stores the operation log data in a relational database, and stores the communication message data in a binary file. In specific implementation, the multi-modal operation data is preprocessed by a data integration formula, which is expressed as: wherein: represents the aggregated multi-modal operation data set, K represents the data category index and belongs to the set , E represents the electrical parameter data category, S represents the device status data category, L represents the operation log data category, C represents the communication message data category, and T represents a unified time axis sequence, represents the original data set under category K, represents a function of time alignment and format conversion of data of category K. Function Map different categories of data to a unified timestamp and structured record. It can be understood that the heterogeneity of multi-modal operation data is reflected in the difference in data value types. Electrical parameter data is mainly floating point values, device state data is mainly Boolean or integer enumeration values, operation log data is mainly string type text, and communication message data is mainly hexadecimal byte stream.

[0066] Embodiment 2: refer to Figure 2 Clean and align the multi-modal operation data, eliminate the misalignment of data in time and space dimensions, including unified calibration of time scale, filling of short-term data missing due to communication interruption, and elimination of outliers obviously exceeding the physical range. Identify the correlation between each physical device, electrical connection and logical control unit in the power station from the cleaned and aligned data, such as identifying the physical connection relationship between all strings under the same combiner box, all combiner boxes under the same inverter, and all inverters under the same transformer, as well as the logical control and communication relationship between the monitoring host and each lower device. Based on the correlation, calculate the dynamic influence weight between each correlation node, which is quantified according to the closeness of electrical coupling between nodes, the frequency of data flow interaction and the dependence of control logic. According to the dynamic influence weight and the correlation, generate a power station global operation situation topology graph with dynamic variable weight composed of nodes and edges. Traverse the nodes and edges in the power station global operation situation topology graph, identify abnormal nodes and abnormal connection edges whose operation parameters deviate from the predetermined reference range, which is set according to the rated parameters of the device, historical normal operating data and environmental conditions. Analyze the inducing or aggravating effect of the environmental background information on the abnormal nodes and the abnormal connection edges to form environmental coupling abnormal features, such as analyzing the contribution of high temperature environment to the temperature abnormal node of the device, or analyzing the rationality of the power abnormal edge under low radiation conditions. Combine the abnormal nodes, the abnormal connection edges and the environmental coupling abnormal features to mine power station operation multi-granularity abnormal signs representing local abnormality, correlation abnormality and system-level risk from three dimensions of device level, branch level and system level, such as device-level signs like single inverter module over-temperature, branch-level signs like imbalance of current consistency of multiple strings under a combiner box, and system-level signs like systematic deviation of the whole station power curve from the theoretical value.

[0067] In specific implementations, the multi-modal operation data, including electrical parameter data, equipment status data, operation logs, and communication messages, are cleaned and aligned to eliminate misalignments in time and space dimensions. In specific implementations, the time alignment operation addresses clock biases existing in different monitoring equipment, for example, the time stamp of the environmental background information collected by the weather station is 10:00:00, while the time stamp of the power data collected by inverter No. A01 is 10:00:03. The alignment process corrects the time stamps of all data to the uniform time of 10:00:00 based on the master clock of the power station monitoring system. For the case where the irradiance data of the weather station at 10:00:00 is missing, the data at 10:00:05 and 09:59:55 are used to calculate the estimated irradiance value at 10:00:00 by linear interpolation. The space alignment operation addresses the problem of inconsistent data source identifiers, for example, the operation log records “inverter A01 failure”, while the device address identifier in the communication message is “INV-01”. The alignment process establishes a device alias mapping table to associate “inverter A01” and “INV-01” to the uniform device identifier “INV_A_01”. The outlier rejection operation identifies and processes data points that are outside the physical range, for example, a group of string current sensors reports a current value of 125 amperes at 10:00:00, while the rated maximum current of this model group is 15 amperes. The cleaning process marks this 125 amperes data point as invalid and removes it from the data sequence. In some embodiments, the comparison of the cleaned and aligned data examples is that the DC voltage of the inverter at 10:00:00 in the original data is 800 volts, while the DC voltage of the same inverter at 10:00:01 is recorded as 0.8 volts. The alignment and cleaning logic determines that 0.8 volts is a dimension recording error and corrects it to 800 volts based on historical data patterns. In some embodiments, the association between each physical device, electrical connection, and logical control unit in the power station is identified from the cleaned and aligned data. The identification process establishes physical connection relationships based on power station design drawings and configuration tables, for example, it identifies that string numbers PV_String_001, PV_String_002 to PV_String_016 belong to combiner box number CJB_01, the output of combiner box number CJB_01 is connected to the DC input terminal of inverter number INV_A_01, and the AC output of inverter number INV_A_01 is connected to the low voltage side of box-type transformer number T_01. The logical control relationship is parsed from the communication messages, for example, the start-stop control command message sent by the monitoring host to inverter number INV_A_01 is parsed, thereby establishing the logical control relationship of the monitoring host to inverter number INV_A_01, and the data collector periodically polls the messages of each current in combiner box number CJB_01, thereby establishing the data collection association relationship between the data collector and combiner box number CJB_01.

[0068] Based on the association relationship, the dynamic influence weight between each associated node is calculated, the associated node refers to a physical device or a logical unit, and the dynamic influence weight is quantified according to the closeness of electrical coupling between nodes, the frequency of data flow interaction and the dependency of control logic. The closeness of electrical coupling is calculated by electrical parameter correlation, for example, the Pearson correlation coefficient between the current of each string of the road group under the busbar number CJB_01 and the total output current of the busbar number CJB_01. The frequency of data flow interaction is calculated by the number and type of communication messages per unit time, for example, the number of heartbeat messages and the number of power control instructions exchanged between the monitoring host and the inverter number INV_A_01 per minute. The dependency of control logic is determined by analyzing the sequence of event triggering and conditional judgment logic in the operation log. The calculation of the dynamic influence weight is realized by an integrated formula, which is expressed as: Wherein: represents the dynamic influence weight between node m and node n, represents the absolute value of the correlation coefficient of the key electrical parameter sequence of node m and node n in the selected time window, represents the communication interaction frequency of node m and node n per unit time, represents the maximum communication interaction frequency between all nodes in the station for normalization, represents the control dependency of node m to node n, its value is assigned to 0 to 1 according to the priority of the control instruction and whether it is a critical control path, and α, β, γ are the weighting coefficients of electrical coupling, communication interaction and control dependency respectively, and satisfy ​determination. In the example, the topology graph contains a node "PV_String_001", a node "CJB_01", and an edge connecting the two nodes, the weight of the edge is calculated based on the correlation coefficient of the string current and the total current of the junction box, the data reporting frequency between the string and the junction box. In a specific implementation, the nodes and edges in the global operation situation topology graph of the power station are traversed, and abnormal nodes and abnormal connection edges whose operation parameters deviate from the predetermined reference range are identified. The predetermined reference range is set according to the rated parameters of the equipment, historical normal working condition data and environmental conditions, for example, the internal temperature of the inverter number INV_A_01 is set to 0-70 degrees Celsius, and the internal temperature of the inverter is 65 degrees Celsius under the historical normal working condition of the ambient temperature of 25 degrees Celsius and the output power of 80 kilowatts. If the internal temperature of the inverter number INV_A_01 is detected to be 85 degrees Celsius, the node "INV_A_01" is determined to be an abnormal node. For the connection edge, the predetermined reference range can be set to the normal difference range between the associated parameters, for example, the current difference of each string under the junction box number CJB_01 under uniform illumination should be less than 10% of the average value. If the current of the string number PV_String_005 is 1 ampere, and the current of the other 15 strings is 5 ampere, it is determined that the edge connecting "PV_String_005" and "CJB_01" is an abnormal connection edge. Analyze the inducing or aggravating effect of environmental background information on abnormal nodes and abnormal connection edges to form environmental coupling abnormal characteristics, for example, when the internal temperature abnormal node of the inverter number INV_A_01 is identified, the environmental temperature data in the environmental background information is read as 40 degrees Celsius, and the environmental coupling abnormal characteristic can be quantified as an "environmental high temperature contribution factor". The factor is based on the device thermal model to calculate the theoretical contribution proportion of the current environmental temperature to the device temperature rise. The device thermal model first determines the baseline thermal equilibrium state of the device under standard environmental conditions according to the design specifications and historical normal operation data of the device, including the heat generation of the device itself, the heat dissipation capacity and the heat exchange characteristics of the surrounding environment. When the device temperature anomaly is monitored, the model introduces the current environmental temperature data in real time, compares the actual temperature value with the expected temperature value under the baseline thermal equilibrium state, and analyzes the driving effect of environmental factors on device temperature rise. Optionally, combined with abnormal nodes, abnormal connection edges and environmental coupling abnormal characteristics, multi-granularity abnormal signs of power station operation are mined from three dimensions of device level, branch level and system level. The device level abnormal sign represents local abnormalities, for example, "the internal temperature of the inverter number INV_A_01 exceeds the safety threshold, and the environmental high temperature contribution factor is 0.3". The branch level abnormal sign represents associated abnormalities, for example, "the current of the string number PV_String_005 under the junction box number CJB_01 is significantly low, only 20% of the average current of the branch, and there is no shadow blocking record at the location of the string".System-level abnormality symptom characterizes system-level risks, such as "the total output power of the power plant shows a continuous downward trend during the continuous high irradiance in the afternoon, deviating from the theoretical power curve by more than 15%, and the temperature of multiple inverter nodes shows an upward trend". It can be understood that the extraction of power plant operation multi-granularity abnormality symptoms is based on cleaned and aligned data, dynamic topology association and environmental coupling analysis.

[0069] Embodiment 3: refer to Figure 3 , collect historical fault case reports, maintenance records and corresponding complete operation data sequences of the power plant, which cover the whole process before, during and after the fault occurs. Use a large model to jointly analyze the historical fault case reports and the complete operation data sequences. The large model analyzes the text description in the fault report and correlates and maps it with the operation data change trend in the corresponding time period, thereby extracting the initial cause, development path, key turning point and final disposal strategy of the fault. The extracted information is structured and stored to form a power plant abnormality knowledge base indexed by fault mode, containing fault development chain and response operation sequence. Each knowledge record contains fault phenomenon, evolution stage characteristics, associated environmental factors, disposal measures and effects. When a new fault case is processed and confirmed, an update process is automatically started to integrate the analysis results of the new case into the power plant abnormality knowledge base, realizing the accumulation and evolution of knowledge. The real-time extracted power plant operation multi-granularity abnormality symptoms are used as query inputs for multi-dimensional retrieval in the power plant abnormality knowledge base, with retrieval dimensions including abnormal parameter type, abnormality occurring device level and associated environmental background. The similarity between the power plant operation multi-granularity abnormality symptoms and each fault evolution mode in the power plant abnormality knowledge base in terms of symptom manifestation, development timing and environmental context is calculated, which integrates feature matching degree, time series shape distance and context relevance. According to the similarity, a high-confidence candidate fault evolution mode set is selected. Use a large model to deduce the candidate fault evolution mode set, combined with the structure and state of the current power plant global operation situation topology graph, to predict the most likely development path and potential risk points of the current abnormality.

[0070] In a specific implementation, historical fault case reports, maintenance records, and corresponding complete operation data sequences of a power station are collected, which cover the whole process before, during, and after the fault occurs. The historical fault case reports are stored in the form of text files, which record the phenomenon description, occurrence time, handling personnel, cause analysis, and final disposal measures of the fault. The maintenance records are stored in the form of database tables, which include fault equipment identification, maintenance operation steps, replacement component information, and test results after maintenance. The corresponding complete operation data sequences are extracted from the power station monitoring historical database, which include multi-modal operation data of several hours to several days before and after the fault event time point. The historical fault case report of example one is described as “the AC side overcurrent protection tripping of inverter number INV_C_03 occurred at 14:20 on July 15, 2023, and on-site inspection found that the cooling fan was stuck, the internal insulated gate bipolar transistor module temperature was too high, and after processing, the fan was replaced and restarted to recover”, and the corresponding complete operation data sequence includes the internal temperature data, output current data, fan speed data, and event records in the operation log of inverter number INV_C_03 from 12:00 to 16:00 on July 15. The historical fault case report of example two is described as “the output current of multiple groups of strings under the combiner box number CJB_12 was continuously low, and inspection found that the DC fuse was blown and the connection point was corroded”, and the corresponding complete operation data sequence includes the daily current curve, environmental temperature and humidity data of each group of strings under the combiner box number CJB_12 one week before the fault occurred, and the comparison current data of other normal branches in the same combiner box. In some embodiments, a large model is used to jointly analyze the historical fault case report and the complete operation data sequence. The large model analyzes the text description in the fault report, such as extracting the key entities “cooling fan” and “insulated gate bipolar transistor module” and the states “stuck” and “temperature too high” from the report text “cooling fan stuck, internal insulated gate bipolar transistor module temperature too high”. At the same time, the large model correlates the complete operation data sequence in the corresponding time period, identifies that the internal temperature data of inverter number INV_C_03 continuously rises from 65 degrees Celsius to 90 degrees Celsius within two hours before tripping, the fan speed data decreases from 3000 revolutions per minute to 0 revolutions per minute within half an hour before the fault, and the output current appears a sharp peak at the moment before tripping. By jointly analyzing the text semantics and data time series changes, the large model extracts the initial cause of the fault as “cooling fan stop”, the development path as “fan stop leads to heat dissipation failure, insulated gate bipolar transistor module temperature continuously rises, temperature too high triggers overcurrent protection”, the key turning point as “fan speed drops to zero”, and the final disposal strategy as “replace the cooling fan and restart the inverter”.In some embodiments, the extracted information is structured and stored to form a power station abnormality knowledge base indexed by failure modes, containing failure development chains and response operation sequences, each knowledge record containing a failure mode name, a failure phenomenon description, an evolution stage characteristic, an associated environmental factor, a sequence of disposal measures and an effect feedback. For example, for the above example, a knowledge record is constructed, the failure mode index is "inverter heat dissipation failure mode A", the failure phenomenon description field contains "AC side overcurrent protection tripping", "inverter internal temperature is too high", the evolution stage characteristic field contains "fan speed gradually descending stage", "temperature linearly rising stage", "protection triggering tripping stage", the associated environmental factor field contains "environmental temperature higher than 35 degrees Celsius", the disposal measure sequence field contains "1. remotely confirm fan failure alarm; 2. send on-site inspection work order; 3. replace the failed fan on site; 4. clear the failure record and remotely restart the inverter", and the effect feedback field contains "the inverter temperature returns to normal after replacement, and the grid connection is successful after restart".

[0071] In a specific implementation, when a new fault case is processed and confirmed by the on-site operation personnel, an update process is automatically started. The trigger condition of the update process is that the maintenance record state field is updated to "closed loop" and the time from the fault occurrence is more than 24 hours. The automatic start of the update process calls the large model analysis service, inputs the new fault case report and complete operation data sequence, and the large model outputs the new knowledge item. The knowledge management system compares the new item with the existing items in the power station abnormal knowledge base and fuses them. If it is a new mode, an index is added. If it is a supplementary case of an existing mode, the evolution characteristics or disposal measures of the original record are updated. Optionally, the real-time extracted power station operation multi-granularity abnormal signs are taken as query inputs and are searched in the power station abnormal knowledge base in multiple dimensions, including abnormal parameter type, abnormal device level, and associated environmental background. For example, the real-time extracted power station operation multi-granularity abnormal signs include "the internal temperature of inverter number INV_D_05 exceeds 75 degrees Celsius, the fan speed is 2800 revolutions per minute, and the environmental temperature is 38 degrees Celsius". The search process matches all fault modes containing the keywords "temperature too high" and "fan" in the power station abnormal knowledge base, and further filters the records with the device level of "inverter" and the environmental factors containing "high temperature". In a specific implementation, the similarity of the power station operation multi-granularity abnormal signs and each fault evolution mode in the power station abnormal knowledge base is calculated in terms of sign manifestation, development time sequence, and environmental context. The sign manifestation similarity compares the closeness of the specific parameters of the current abnormal signs and the historical fault phenomenon parameters recorded in the knowledge base. The development time sequence similarity compares the matching degree of the change trend of the current abnormal sign parameters and the historical fault development stage time sequence pattern recorded in the knowledge base. The environmental context similarity compares the consistency of the environmental background information when the current abnormality occurs and the historical fault associated environmental factors recorded in the knowledge base. The similarity calculation is realized through a multi-dimensional feature distance integration formula, which is expressed as: wherein S represents the comprehensive similarity score, represents the Euclidean distance between the current abnormal sign feature vector X and the historical sign feature vector Y of a fault mode in the knowledge base. The feature vector includes normalized numerical values of specific parameters such as temperature, speed, etc., represents the dynamic time warping distance between the current abnormal parameter change curve and the fault development time sequence curve recorded in the knowledge base represents the cosine similarity distance between the current environmental context vector and the historical fault case environmental context vector are the weight coefficients of the sign manifestation, development time sequence, and environmental context dimensions, respectively, and satisfy ​​This can be understood as filtering out a set of candidate fault evolution patterns with high confidence based on similarity, and setting a similarity score threshold, for example, setting a threshold. =0.8, all fault modes with an S value greater than 0.8 are selected as the candidate fault evolution mode set. Optionally, a large model is used to extrapolate the candidate fault evolution mode set to predict the most likely development path and potential risk points of the current anomaly. The input of the large model includes the historical development chain of each mode in the candidate fault evolution mode set, the structure and status of the current power plant's overall operating topology map, and the latest multimodal operating data. For example, if the candidate set includes "inverter heat dissipation fault mode A" and "DC-side insulation fault mode B", the large model analyzes the position of inverter number INV_D_05 in the circuit and the status of its adjacent equipment based on the topology map. Combined with the current continuous high temperature conditions, it extrapolates that if it develops according to "inverter heat dissipation fault mode A", inverter number INV_D_05 may trigger over-temperature derating within 2 hours, affecting the power generation of its branch; if there are early characteristics of "DC-side insulation fault mode B", it may develop into a short circuit to ground risk. The large model output prediction result is "the most likely development path is inverter heat dissipation failure mode A, and the potential risk point is that over-temperature protection will be triggered within 2 hours, resulting in power loss".

[0072] Example 4: Based on the matched fault evolution pattern and its predicted development path, preliminary handling suggestion sequences of different urgency levels are generated. These sequences are derived from the corresponding fault mode response operation sequences in the power plant anomaly knowledge base. The power plant's overall operational topology is analyzed to assess the potential cascading effects of executing the preliminary handling suggestion sequences on non-faulty parts of the power plant. For example, the impact of disconnecting a faulty branch on the power quality and system stability at the grid connection point is assessed, or the impact of derated operation of a certain device on overall power generation is assessed. The preliminary handling suggestion sequences and the cascading impact assessment results are integrated to generate a multi-layered anomaly decision flow containing execution steps, expected effects, risk warnings, and backup plans. The decision flow clearly defines the operation objects, operation instructions, execution conditions, and expected results for each step. A collaborative analysis engine is activated. This engine simulates the power plant's operating rules and safety constraints, performing logical consistency checks and conflict detection on each operation in the multi-layered anomaly decision flow. For example, it detects whether mutually exclusive instructions are issued to the same device simultaneously, or whether the operation sequence violates electrical operation safety regulations. Under the premise of satisfying all operational rules and safety constraints, conflicting or redundant operational steps are merged, sorted, and optimized to form a logically consistent and efficient operational sequence. Combined with the real-time updated multimodal operational data, the optimized decision flow is pre-evaluated to simulate possible changes in the power plant's operational status after execution. The collaborative analysis engine outputs the final monitoring conclusions after verification, optimization, and pre-evaluation.

[0073] In a specific implementation, a preliminary treatment suggestion sequence of different emergency levels is generated according to the matched fault evolution pattern and its predicted development path, the fault evolution pattern is derived from the records in the power plant anomaly knowledge base, for example, the matched fault pattern is “inverter heat dissipation fault pattern A”, and its predicted development path is “over-temperature reduction within 2 hours”, then the generated preliminary treatment suggestion sequence includes high emergency level “immediately remotely reduce the output power of inverter number INV D 05 to 80% of the rated value”, medium emergency level “send a patrol work order to check the heat dissipation fan of inverter number INV D 05”, and low emergency level “arrange night shutdown for heat dissipation system cleaning”. The preliminary treatment suggestion sequence is stored in a structured list form, each suggestion contains an operation object, an operation instruction, and an emergency level identifier. In some embodiments, the global operation situation topology of the power plant is analyzed to evaluate the cascading effects that the execution of the preliminary treatment suggestion sequence may cause to the non-fault part of the power plant, the evaluation process is based on the dynamic influence weight and electrical connection relationship between nodes in the topology, for example, evaluating the operation of remotely reducing the output power of inverter number INV D 05, inverter number INV D 05 is the largest capacity inverter under box transformer number T 02, reducing its power may cause the load rate of box transformer number T 02 to drop, thereby affecting the power output plan required by the power grid dispatching, the cascading effect evaluation result is quantified as “the total output power of box transformer number T 02 is expected to decrease by 50 kW, which may trigger the power grid assessment”. In some embodiments, the preliminary treatment suggestion sequence and the cascading effect evaluation result are fused to generate a multi-layer anomaly decision flow containing execution steps, expected effects, risk prompts and backup solutions, the multi-layer anomaly decision flow adopts a hierarchical structure, the top layer is the strategic target layer, the middle layer is the tactical operation layer, and the bottom layer is the specific execution layer. The strategic target layer defines the overall treatment target such as “prevent equipment damage while minimizing power generation loss”, the tactical operation layer arranges the priority and dependency relationship of the operation steps, and the specific execution layer details the parameters and execution conditions of each step. The fusion process associates each suggestion in the preliminary treatment suggestion sequence with the corresponding cascading effect evaluation result, and adds backup solutions for high-risk operations, for example, adding a risk prompt “may cause the output of box transformer number T 02 to be substandard” for the “remotely reduce the output power of inverter number INV D 05” step, and setting a backup solution “if the power grid allows, preferentially adjust the power of other branches for compensation”. Referring to Table 1, the multi-layer anomaly decision flow table.

[0074] Table 1: Multi-layer anomaly decision flow table

[0075] Decision flow hierarchy Operational step description Expected effect Risk alert Contingency plan Strategic goal layer Balance equipment safety and power generation efficiency Avoid overheating failure, loss controllable Power fluctuation may affect the power grid None Tactical operation layer Step 1: Run the inverter INV_D_05 at reduced capacity Reduce the temperature rise rate Box transformer T_02 total output drops Step 1a: Adjust the inverter INV_D_04 power Tactical operation layer Step 2: Assign a fan inspection work order Confirm and repair the root cause Risk continues during inspection Step 2a: Prepare a backup fan Specific execution layer Set the INV_D_05 power limit to 80% Temperature stabilizes below 70°C Real-time power data monitoring Execute step 1a

[0076] In a specific implementation, a collaborative analysis engine is initiated, which simulates power plant operation rules and safety constraints, including electrical operation safety procedures, equipment operation limits, grid dispatch protocols, and safety constraints including insulation safety distance, maximum allowable load, protection setting range. The collaborative analysis engine performs logical consistency check and conflict detection for each operation in the multi-layered abnormal decision flow. The logical consistency check checks whether the operation sequence conforms to the operation logic, for example, checking whether there is a reverse sequence operation of "disconnecting the bus breaker" before "confirming that the bus current is zero". The conflict detection identifies the mutual exclusivity between operation instructions, for example, detecting that there are two operations of "closing the inverter AC contactor" and "performing the inverter insulation resistance test" in the multi-layered abnormal decision flow, while the safety rule requires that the AC side connection must be disconnected during the insulation test. The conflict detection quantifies the conflict degree between operation steps by a conflict score formula, which is expressed as:

[0077]

[0078] wherein: represents the operation step and the operation step , the conflict score between represents the total number of power plant operation rules and safety constraints, represents the Kth rule or constraint, is an indicator function, the function value is 1 when the operation and simultaneously violates or potentially violates the rule , otherwise it is 0. For example, the rule is "the AC side must be disconnected during the insulation test", the operation is "close the inverter AC contactor", and the operation is "perform the inverter insulation resistance test", then =1. Optionally, under the premise of meeting all operating rules and safety constraints, the operation steps with conflicts or redundancies are merged, sorted and optimized, the merging operation integrates multiple operations for the same device into a composite operation, for example, merging “reduce power to 80%” and “enable standby cooling mode” into “set power limit 80% and enable standby cooling”, the sorting operation rearranges the step order according to the operation dependency and emergency level, and the optimization operation deletes redundant steps, for example, deleting repeated “confirm device status” steps. The optimization process aims to minimize the conflict score and maximize the execution efficiency. Optionally, in combination with real-time multi-modal operating data, the optimized decision flow is pre-evaluated for effectiveness, the pre-evaluation simulates the execution of each step in the optimized decision flow, and predicts the power plant state changes based on real-time data, for example, after simulating the execution of “set inverter number INV_D_05 power limit 80%”, read the real-time irradiance data and inverter efficiency curve, predict the change amplitude of the total power generation of the power plant within the next hour, and predict the temperature change trend of inverter number INV_D_05. The effectiveness pre-evaluation output quantitative indicators such as expected power loss, risk reduction probability. It can be understood that the final monitoring conclusion output by the collaborative analysis engine after verification, optimization and pre-evaluation is a comprehensive statement, including the confirmed abnormal type, the recommended operation sequence, the expected result, and the remaining risk explanation.

[0079] Referring to Figure 4 In the cascading impact assessment of reducing the output power of inverter INV_D_05, the impact of grid assessment risk and power generation loss is quantified by multi-dimensional indicators. In the specific analysis, “power quality”, “power output”, “voltage stability”, “frequency fluctuation” and “grid compliance” are taken as influence categories, respectively corresponding to the quantification of grid assessment risk score (0-100 points) and power generation loss ratio (%): the grid assessment risk (purple column) represents the risk degree with 0-100 points, reaching the highest (about 74 points) in the “power output” category, indicating that the operation has the most significant impact on the power output dimension of the grid assessment; while the risk is the lowest (about 5 points) in the “frequency fluctuation” category, reflecting that the operation has less disturbance on frequency stability. The power generation loss (orange line) quantifies the loss amplitude in percentage, which is the highest (about 8%) in the “power output” category, consistent with the peak dimension of the grid assessment risk; the loss is the lowest (about 0.3%) in the “frequency fluctuation” category, corresponding to the risk distribution. In the process of indicator quantification, the grid assessment risk is calculated based on the dynamic influence weight of the topology graph nodes and the grid dispatching protocol threshold, and the power generation loss is derived combined with the device capacity ratio and real-time irradiance data, and the associated distribution of the two directly presents the influence difference of the operation on different dimensions of the power plant system.

[0080] Example 5: According to the monitoring conclusion, determine the comprehensive influence level of the current anomaly on the power plant's power generation efficiency, equipment safety, and system stability, which is divided according to the power loss scale, equipment damage risk, and system outage probability that the anomaly may cause. From the equipment availability, system reliability, and power loss, etc. multiple dimensions, the health index of the power plant is quantitatively generated. The equipment availability is calculated based on the proportion of unavailable equipment capacity, the system reliability is evaluated based on the frequency and severity of recent abnormal events, and the power loss is calculated based on the difference between the current actual power and the ideal power. The comprehensive influence level and the health index are fused to form a visual power plant health status portrait that intuitively displays the overall health status, weak links, and risk distribution of the power plant. The portrait is displayed through a graphical interface, with different colors and legends identifying the health status and risk level of each area. The operation instructions in the monitoring conclusion are classified according to the execution subject, execution priority, and execution time window. The execution subject includes the automatic control system in the station, the robot inspection system, the remote monitoring center, or the on-site operation and maintenance personnel. Detailed operation parameters, execution conditions, and acceptance standards are configured for each type of operation instruction. Operation parameters such as set power limit, target temperature threshold, execution conditions such as meeting specific environmental conditions or equipment status, and acceptance standards such as the range that key parameters should recover after operation are configured. All classified operation instructions with detailed configurations are packaged to form a structured operation and maintenance decision cluster that can be directly parsed and scheduled by the operation and maintenance system. The decision cluster supports being issued to the corresponding execution system or personnel.

[0081] In a specific implementation, a comprehensive impact level of the current anomaly on the power plant's power generation efficiency, equipment safety, and system stability is determined according to the monitoring conclusion, which is derived from the final monitoring conclusion output by the collaborative analysis engine. For example, a monitoring conclusion confirms that "the heat dissipation of inverter number INV_D_05 is poor, and it is recommended to run at reduced capacity and arrange for inspection, which is expected to cause a 20% decrease in power generation power of the branch where it is located, and there is a risk of overheating damage, which has no direct impact on the stability of the whole station system." Based on this conclusion, the comprehensive impact level is determined according to the preset quantification rules. The impact level on power generation efficiency is divided according to the power loss ratio, for example, a loss of less than 5% is "low impact", a loss of between 5% and 20% is "medium impact", and a loss of more than 20% is "high impact". In the example, a 20% power loss corresponds to a "medium impact" level. The impact level on equipment safety is divided according to the risk and consequences of equipment failure, for example, "there is a risk of overheating damage" corresponds to a "medium-high risk" level. The impact level on system stability is divided according to whether it causes power grid fluctuations or protection actions, and in the example, "it has no direct impact on the stability of the whole station system" corresponds to a "no impact" level. Based on the above three dimensions, a decision matrix is used to map "medium impact on power generation efficiency", "medium-high risk on equipment safety", and "no impact on system stability" to a comprehensive impact level of "level two alarm". In some embodiments, a health index of the power plant is quantitatively generated from the device availability, system reliability, and power loss. The device availability index calculates the percentage of unavailable device capacity to the total installed capacity of the power plant, for example, the total installed capacity of the power plant is 10 megawatts, and the effective capacity of inverter number INV_D_05 is reduced from 1 megawatt to 0.8 megawatt due to reduced capacity operation, which is considered as 0.2 megawatt of unavailable capacity. Assuming that there is no other device shutdown in the station, the device availability rate is (10-0.2) / 10*100%=98%. The system reliability index is calculated based on the frequency and severity of recent abnormal events, for example, the number of alarm events of different levels occurring in the past 24 hours is counted and weighted, and the formula is expressed as:

[0082]

[0083] wherein R represents the system reliability index score range 0-100, N represents the total number of alarm event types occurring in the statistical period, represents the number of times of the i-th type of alarm event occurring in the statistical period, The weight coefficient of the severity of the i-th type of alarm event is pre-defined according to the severity of the alarm impact, for example, the weight of "first-level alarm" is 5, the weight of "second-level alarm" is 3, and the weight of "third-level alarm" is 1. The power loss index calculates the percentage difference between the current actual total power generation and the theoretical maximum power generation under the current environmental conditions, for example, the current actual total power generation is 8.5 MW, and the theoretical maximum power generation calculated according to the real-time irradiance and temperature is 9.0 MW, then the power loss = (9.0-8.5) / 9.0*100%≈5.56%. Optionally, the comprehensive impact level and the health index are fused to form a visual power station health status image that intuitively displays the overall health status, weak links and risk distribution of the power station. The power station health status image is presented by a graphical interface, and the center of the interface is a simplified topology of the power station electrical single-line diagram. Different device icons in the topology diagram are represented by their colors and shapes, for example, an inverter in normal state is displayed as a green square, an inverter INV_D_05 in reduced capacity operation state is displayed as an orange square and flashes, and a device in fault offline state is displayed as a red square. The interface side displays the real-time values of the device availability 98%, the system reliability index score 85, the power loss 5.56% and the health index in the form of a dashboard. The other side of the interface lists the current active alarms and their comprehensive impact levels in the form of a list, for example, "inverter INV_D_05 abnormal heat dissipation, comprehensive impact level: second-level alarm". The color depth of the device icon in the topology diagram can also map the historical alarm frequency of the device, and the deeper the color, the more frequent the recent alarms, thereby identifying weak links. The risk distribution is superimposed on the related position in the topology diagram in the form of a red semi-transparent area with different transparency, and the size and color transparency of the area represent the breadth and severity of the risk impact.

[0084] In specific implementations, based on the monitoring conclusion, an operation and maintenance decision cluster corresponding thereto is output, and the monitoring conclusion contains a recommended operation sequence. The operation instructions in the monitoring conclusion are classified according to execution subject, execution priority, and execution time window. The execution subject refers to the recipient and executor of the operation instruction, and the classification includes the in-station automatic control system, the robot inspection system, the remote monitoring center, and the on-site operation and maintenance personnel. The execution priority is classified according to the operation urgency, for example, “execute immediately”, “execute within 1 hour”, “execute within today”, and “plan to execute”. The execution time window specifies the specific or relative time range in which the operation is allowed to be executed. For example, for the operation “remotely reduce the output power of inverter number INV_D_05 to 80%” in the above monitoring conclusion, the execution subject is classified as “in-station automatic control system”, the execution priority is classified as “execute immediately”, and the execution time window is classified as “immediately after the temperature continues to exceed the threshold value”. The operation “assign an inspection work order to check the cooling fan of inverter number INV_D_05” has an execution subject classified as “on-site operation and maintenance personnel”, an execution priority classified as “execute within today”, and an execution time window classified as “during the daytime working hours today”. In some embodiments, detailed operation parameters, execution conditions, and acceptance standards are configured for each type of operation instruction. For the instruction “remotely reduce the output power of inverter number INV_D_05 to 80%”, the configured operation parameter is “power limit ratio: 80%”, the execution condition is “the internal temperature of inverter number INV_D_05 continues to be higher than 75 degrees Celsius for 5 minutes and the fan speed is lower than the rated value”, and the acceptance standard is “within 2 minutes after the instruction is issued, the monitoring data shows that the output power of inverter number INV_D_05 is reduced to 80% ± 2% of the rated value, and the internal temperature stops rising or starts to decrease”. For the instruction “assign an inspection work order to check the cooling fan of inverter number INV_D_05”, the configured operation parameter is “work order type: equipment inspection; target equipment: INV_D_05; inspection items: cooling fan operation status, fan inlet and outlet air ducts”, the execution condition is “the work order is received and distributed to available operation and maintenance personnel by the dispatching system”, and the acceptance standard is “the operation and maintenance personnel feed back the inspection results, including photos and descriptions, through a mobile terminal, and close the work order”. It can be understood that all the classified operation instructions with detailed configurations are packaged to form a structured operation and maintenance decision cluster that can be directly parsed and scheduled by an operation and maintenance system. The operation and maintenance decision cluster adopts a standardized data exchange format, such as JSON or XML format, defines the root node as “operation and maintenance decision cluster”, and contains multiple “operation instruction” child nodes under it. Each “operation instruction” child node contains “execution subject”, “execution priority”, “execution time window”, “operation parameter”, “execution condition”, and “acceptance standard”.Optionally, the operation and maintenance system parses the operation and maintenance decision cluster, and routes the instruction to the corresponding automatic execution module or generates an artificial work order to be distributed to the corresponding personnel according to the "execution subject" field, arranges the execution queue according to the "execution priority" and "execution time window", issues control commands according to the "operation parameters" in the execution process, judges whether to trigger the execution according to the "execution condition", and verifies the execution effect according to the "acceptance standard" after the execution.

[0085] Referring to Figure 5 In the quantitative analysis of the health degree of the power station, the normal value, the current value and the alarm threshold of the three core health degree indexes of equipment availability, system reliability score and power loss are presented in the form of multi-dimensional column chart. Specifically, in the equipment availability dimension, the normal value and the current value are close to 100% and higher than the alarm threshold; in the system reliability score dimension, the normal value is 100, the current value is about 85, and the alarm threshold is 80; in the power loss dimension, the current value is about 5%, and the alarm threshold is about 10%. The quantitative comparison of these indexes intuitively reflects the current running state of the power station: the equipment availability is at a good level, the system reliability score is close to the alarm threshold and needs attention, and the power loss is within a controllable range. Through the chart, the weak link of the health state of the power station can be quickly located, and data support is provided for operation and maintenance decision.

[0086] It should be noted that in this text, relational terms such as first and second are used merely to distinguish one entity or action from another, without necessarily requiring or implying that there is any such actual relationship or order between or among the entities or actions. Also, the terms "comprises", "comprising", or any other variations thereof are intended to cover non-exclusive inclusions, so that a process, method, article, or apparatus that comprises a list of elements does not only include those elements, but also includes other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0087] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring abnormal health status of power plants based on large-scale model data analysis, characterized in that, Perform the following procedure: Simultaneously acquire multimodal operating data and associated environmental background information generated by equipment within the power plant during operation; The multimodal operation data is used to construct a topology map of the power plant's overall operation status that reflects the dynamic relationships of the power plant. Based on the power plant's overall operational status topology map and the environmental background information, multi-granularity anomaly symptoms with different semantic levels are extracted, specifically including: Traverse the nodes and edges in the topology diagram of the power plant's overall operation status, and identify abnormal nodes and abnormal connection edges whose operating parameters deviate from the predetermined benchmark range; The environmental background information is analyzed to determine its inducing or aggravating effect on the abnormal nodes and abnormal connecting edges, thus forming an abnormal environmental coupling feature. Combining the abnormal nodes, abnormal connection edges, and environmental coupling anomaly features, we can mine multi-granularity anomaly signs of power plant operation from three dimensions: equipment level, branch level, and system level, respectively, to represent local anomalies, related anomalies, and system-level risks. By using large models to conduct in-depth analysis of historical failure cases, a power plant anomaly knowledge base containing failure evolution patterns and response strategies is constructed and dynamically updated. The process involves real-time matching and reasoning between the multi-granularity anomaly symptoms of the power plant operation and the fault evolution patterns in the power plant anomaly knowledge base, specifically including: The multi-granularity abnormal symptoms of the power plant operation are used as query inputs, and a multi-dimensional search is performed in the power plant anomaly knowledge base. Calculate the similarity between the multi-granularity abnormal symptoms of the power plant operation and each fault evolution mode in the power plant anomaly knowledge base in terms of symptom manifestation, development sequence and environmental context; A set of candidate fault evolution patterns with high confidence is selected based on the similarity. The evolution of the candidate fault patterns is extrapolated using a large model to predict the most likely development path and potential risk points of the current anomaly. Based on the matching and inference results, a multi-layered anomaly decision flow is generated for the current abnormal situation, specifically including: Based on the matched fault evolution patterns and their predicted development paths, a sequence of preliminary handling recommendations for different levels of urgency is generated; Analyze the overall operational status topology of the power plant and assess the potential cascading effects of implementing the preliminary handling recommendations on the non-faulty parts of the power plant; By integrating the preliminary treatment recommendation sequence with the assessment results of the chain reaction, a multi-layered abnormal decision flow is generated, which includes execution steps, expected effects, risk warnings, and backup plans. Launch a collaborative analysis engine to fuse and optimize the multi-layered anomaly decision flow and generate the final monitoring conclusions; Based on the monitoring results, a power plant health status profile representing the overall health status of the power plant and the corresponding operation and maintenance decision cluster are output.

2. The AI ​​monitoring method for abnormal power plant health status based on large model data analysis as described in claim 1, characterized in that, The multimodal operation data refers to the heterogeneous data set collected by different types of sensors and monitoring equipment during the operation of the power plant, covering electrical parameters, equipment status, operation logs and communication messages.

3. The AI ​​monitoring method for abnormal power plant health status based on large model data analysis as described in claim 1, characterized in that, The construction of a power plant global operational status topology map that reflects the dynamic relationships of the entire power plant using the aforementioned multimodal operational data specifically includes: The multimodal operation data is cleaned and aligned to eliminate misalignment in time and space. The relationships between various physical devices, electrical connections, and logic control units within the power plant were identified from the cleaned and aligned data. Based on the aforementioned relationships, the dynamic influence weights between each associated node are calculated; Based on the dynamic influence weights and the associated relationships, a topology map of the entire power plant's operational status, consisting of nodes and edges with dynamically variable weights, is generated.

4. The AI ​​monitoring method for abnormal power plant health status based on large model data analysis as described in claim 1, characterized in that, By leveraging large-scale models to conduct in-depth analysis of historical failure cases, a power plant anomaly knowledge base containing failure evolution patterns and response strategies is constructed and dynamically updated. Specifically, this includes: Collect historical fault case reports, maintenance records, and corresponding complete operational data sequences of the power plant; The historical failure case reports and the complete operational data sequence are jointly analyzed using a large model to extract the initial causes, development paths, key turning points and final handling strategies of the failures. The extracted information is stored in a structured manner to form a power plant anomaly knowledge base indexed by fault modes, containing fault development chains and response operation sequences. Once a new fault case has been processed and confirmed, the update process is automatically initiated, integrating the analysis results of the new case into the power plant anomaly knowledge base.

5. The AI ​​monitoring method for abnormal power plant health status based on large model data analysis as described in claim 1, characterized in that, A collaborative analysis engine is launched to fuse and optimize the multi-layered anomaly decision flow, generating the final monitoring conclusions, specifically including: The collaborative analysis engine simulates the power plant's operating rules and safety constraints, and performs logical consistency verification and conflict detection on each operation in the multi-layered abnormal decision flow. Under the premise of satisfying all operating rules and safety constraints, conflicting or redundant operation steps are merged, sorted and optimized. By combining the real-time updated multimodal operational data, the effect of the optimized decision flow is pre-evaluated; Output the final monitoring conclusions after verification, optimization, and pre-evaluation.

6. The AI ​​monitoring method for abnormal power plant health status based on large model data analysis as described in claim 1, characterized in that, Based on the monitoring findings, the power plant health status profile, which characterizes the overall health status of the power plant, specifically includes: Based on the monitoring findings, determine the overall impact level of the current anomaly on the power plant's power generation efficiency, equipment safety, and system stability. The current health indicators of the power plant are quantified from the dimensions of equipment availability, system reliability, and power loss. By integrating the comprehensive impact level with the health index, a visual health status profile of the power plant is formed, which intuitively displays the overall health status, weak links, and risk distribution of the power plant.

7. The AI ​​monitoring method for abnormal power plant health status based on large model data analysis as described in claim 1, characterized in that, Based on the monitoring findings, the corresponding operation and maintenance decision clusters output specifically include: The operation instructions in the monitoring conclusions are classified according to the executing entity, execution priority, and execution time window; Configure detailed operating parameters, execution conditions, and acceptance criteria for each type of operation instruction; All categorized operation instructions with detailed configurations are packaged into a structured operation and maintenance decision cluster that can be directly parsed and scheduled by the operation and maintenance system.

Images