Intelligent operation and maintenance management system and method for green power plant

Abstract

The invention relates to the technical field of intelligent operation and maintenance management, and discloses an intelligent operation and maintenance management system and method for a green power plant. The method comprises the following steps: acquiring a fault phenomenon description from an SCADA system, comparing the fault phenomenon description with a local fault phenomenon library, and selecting candidate historical cases; querying a propagation path template according to the candidate historical case, and positioning a source equipment component according to the propagation path template; and extracting a diagnosis rule matched with the source equipment component from a local expert maintenance rule base, and inputting the diagnosis rule into a locally deployed lightweight language model to generate a structured diagnosis report. The whole reasoning process is completed in the local server without accessing an external network, the problems of network delay and offline unavailability are eliminated, and safe and stable operation of equipment of the green power plant under the condition that the network is limited or is completely offline is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent operation and maintenance management, and particularly relates to an intelligent operation and maintenance management system and method for a green power plant. BACKGROUND

[0002] With the transformation of green power plants to a distributed unattended mode, the intelligent operation and maintenance of multi-energy complementary devices such as wind turbine generators, photovoltaic inverters and energy storage systems is increasingly urgent. The traditional operation and maintenance method relies on manual inspection and experience judgment, which is difficult to cope with the complexity and suddenness of device faults, resulting in long fault response time and low accuracy of maintenance decision.

[0003] Although the existing cloud expert system integrates a large amount of diagnostic knowledge, it still faces problems such as poor real-time performance and delayed fault diagnosis caused by unstable network in remote power stations, and cannot provide services at all when the network is interrupted. Meanwhile, the fault propagation mechanism among devices such as wind power fluctuation and energy storage system charge-discharge switching in green power plants is complex, and traditional methods are difficult to accurately trace the root cause of the fault, which is prone to misjudgment or omission of key fault devices. In addition, the rare faults of green power plant devices and the multi-device coupled faults lack sufficient historical samples, and traditional diagnostic models are difficult to cover various complex fault modes that occur in actual operation, and lack the ability to continuously learn and optimize from maintenance feedback, making it difficult to improve the diagnostic accuracy as the running time increases. SUMMARY

[0004] The main purpose of the present application is to provide an intelligent operation and maintenance management system and method for a green power plant. The entire reasoning process of the present application is completed on a local server without accessing external networks, eliminating network delays and offline unavailability problems, and ensuring the safe and stable operation of green power plant devices under network limited or completely offline conditions.

[0005] To achieve the above-mentioned purpose, the present application provides an intelligent operation and maintenance management method for a green power plant, comprising the following steps: Collecting fault phenomenon descriptions from a SCADA system, and comparing the fault phenomenon descriptions with a local fault phenomenon library to select candidate historical cases; Querying a propagation path template according to the candidate historical cases, and locating a source device component according to the propagation path template; Extracting diagnostic rules matched with the source device component from a local expert maintenance rule library, and inputting the diagnostic rules into a locally deployed lightweight language model to generate a structured diagnostic report.

[0006] Optionally, in the first implementation manner of the first aspect of the present application, the step of collecting fault phenomenon descriptions from a SCADA system, and comparing the fault phenomenon descriptions with a local fault phenomenon library to select candidate historical cases, comprises: The gearbox vibration acceleration, oil temperature, generator current fluctuation amplitude and environmental temperature parameters are acquired through a communication interface of the edge computing node and the SCADA system, and a normal operation interval baseline of each parameter is established according to equipment nameplate parameters and historical normal operation data statistical analysis; When the gearbox vibration acceleration exceeds the upper limit value of the normal operation interval baseline and the duration reaches a preset length of time, a hard threshold abnormality is marked, and when the oil temperature exceeds the upper limit and the rising rate exceeds a normal rising rate threshold, a trend abnormality is marked, and a parameter combination that simultaneously satisfies the hard threshold abnormality and the trend abnormality forms a fault phenomenon description; A candidate historical case with the highest similarity is selected from a local fault phenomenon library according to the fault phenomenon description.

[0007] Optionally, in a second implementation manner of the first aspect of the present application, selecting a candidate historical case with the highest similarity from a local fault phenomenon library according to the fault phenomenon description comprises: A current fault feature vector containing a main abnormal parameter value, a secondary abnormal parameter combination and a current environmental condition is extracted from the fault phenomenon description, a stored historical fault mode record in the local fault phenomenon library is searched, and a corresponding historical fault feature vector is extracted for each historical fault mode record; The similarity between the current fault feature vector and each historical fault feature vector is calculated, and the candidate historical case with the highest similarity is selected.

[0008] Optionally, in a third implementation manner of the first aspect of the present application, a propagation path template is queried according to the candidate historical case, and a source device component is located according to the propagation path template, comprising: A fault type identifier is extracted from the candidate historical case, and a propagation path template matching the fault type identifier is searched in the local fault propagation correlation library; A multi-level propagation node is extracted from the propagation path template, and each level of the multi-level propagation node contains a trigger condition definition, a state judgment basis and a propagation time delay parameter; The position of the current abnormal device in the green power plant topology is reversely matched with the propagation path template, and a source device component where a fault root cause is located is located.

[0009] Optionally, in a fourth implementation manner of the first aspect of the present application, after the multi-level propagation node is extracted from the propagation path template, further comprising: Resolving a parameter configuration of a root cause node and a first-level propagation node from the propagation path template, the root cause node defining a preset proportion of an output power fluctuation amplitude of the fan exceeding a rated power as a trigger condition, and the first-level propagation node defining a number of times of charge-discharge mode switching of the energy storage system exceeding a preset switching threshold as a state determination basis; Extracting a propagation time delay parameter of the root cause node pointing to the first-level propagation node to define a delay time range from power fluctuation detection to occurrence of a frequent switching state, and a second-level propagation node defining an increase proportion of a measured internal resistance value of a battery cell obtained through impedance testing exceeding a preset internal resistance threshold compared with an initial internal resistance value.

[0010] Optionally, in a fifth implementation manner of the first aspect of the present application, the position of the current abnormal equipment in the green power plant topology is reversely matched with the propagation path template, and a source equipment component where a root cause of the fault lies is located, including: Obtaining an upstream and downstream equipment connection relationship of the current abnormal equipment in the green power plant topology, comparing the gearbox vibration acceleration and the generator current fluctuation amplitude in the fault phenomenon description with the multi-level propagation nodes, and when it is found that the gearbox vibration acceleration and the current fluctuation amplitude respectively match characteristic parameters of the second-level propagation node and the third-level propagation node, reversely tracing to the root cause node along the multi-level propagation nodes; Querying iron element content and copper element content detection data in oil in a latest oil analysis report of the current abnormal equipment, comparing the iron element content with an iron element content threshold defined in the root cause node, comparing the copper element content with a copper element content threshold defined in the root cause node, and when the iron element content and the copper element content both exceed the corresponding thresholds, confirming that the source equipment component is a gearbox bearing.

[0011] Optionally, in a sixth implementation manner of the first aspect of the present application, a diagnostic rule matched with the source equipment component is extracted from a local expert maintenance rule library, and the diagnostic rule is input into a locally deployed lightweight language model to generate a structured diagnostic report, including: According to the source equipment component, a diagnostic rule completely matched with the source equipment component and the fault type is screened in the local expert maintenance rule library, the diagnostic rule containing condition judgment logic, maintenance time window suggestion, spare part list, and disposal step sequence; The diagnostic rule is input into a locally deployed lightweight language model for reasoning to generate a structured diagnostic report.

[0012] Optionally, in a seventh implementation manner of the first aspect of the present application, the diagnostic rule is input into a locally deployed lightweight language model for reasoning to generate a structured diagnostic report, including: The condition judgment logic in the diagnosis rule and the disposal step sequence, the fault phenomenon description and the maintenance measure record in the candidate historical case, and the propagation path analysis result in the multi-level propagation node are integrated into a context information data packet, and the context information data packet is input into a lightweight language model deployed in a local server to perform text understanding and comprehensive reasoning; The lightweight language model generates a structured diagnosis report after reading the context information data packet, and the structured diagnosis report includes a root cause analysis paragraph that explicitly indicates the source device component and failure mechanism, an influence range evaluation paragraph that lists the affected downstream devices and potential risk devices, and a step-by-step maintenance guidance paragraph that details the specific operation requirements for isolating the faulty device to the recovery operation in chronological order.

[0013] Optionally, in an eighth implementation manner of the first aspect of the present application, the intelligent operation and maintenance management method for the green power plant further includes: The maintenance execution feedback data input by the field operation and maintenance personnel through the mobile terminal is acquired, and when it is found that the source device component in the maintenance execution feedback data contains an additional fault component that is not predicted in the structured diagnosis report, it is determined that there is a deviation; If there is a deviation, a new fault mode item is generated, and the new fault mode item is supplemented to the local fault phenomenon library, and a propagation path record from the predicted component fault to the additional fault component is newly added in the local fault propagation correlation library, and a propagation time delay parameter and a propagation probability weight are set in the propagation path record.

[0014] The present application also provides an intelligent operation and maintenance management system for a green power plant, comprising: The acquisition module is configured to acquire fault phenomenon descriptions from the SCADA system and compare the fault phenomenon descriptions with a local fault phenomenon library to select candidate historical cases. The positioning module is configured to query a propagation path template according to the candidate historical cases and locate a source device component according to the propagation path template. The diagnosis module is configured to extract a diagnosis rule matched with the source device component from a local expert maintenance rule library and input the diagnosis rule into a lightweight language model deployed locally to generate a structured diagnosis report.

[0015] In summary, the present application realizes the self-precise fault diagnosis of green power plant equipment under complete offline conditions by constructing a local three-layer heterogeneous knowledge base and combining fault propagation correlation analysis. The first layer fault phenomenon library stores the threshold values and duration requirements of the characteristic parameters of typical faults such as fan gear box vibration, photovoltaic module hot spot, and abnormal energy storage battery voltage. The second layer fault propagation correlation library defines the causal chain and propagation time delay parameters between devices in a multi-energy complementary scene. The third layer expert maintenance rule library integrates the device manufacturer's technical manual and field maintenance experience to form structured diagnostic rules. The cooperative work of the three-layer knowledge base enables fault positioning to be promoted from a single device level to a multi-device propagation chain level. After extracting the candidate historical cases using the similarity matching algorithm, the root cause of the fault is located by reverse tracing the propagation path template, avoiding misdiagnosis caused by judging only based on surface abnormal phenomena, and ensuring accurate identification of the source device components. The diagnostic rules are input into the locally deployed lightweight language model to generate a structured diagnostic report containing root cause analysis, impact range evaluation, and step-by-step maintenance guidance. The entire reasoning process is completed on the local server without accessing external networks, eliminating network delays and offline availability problems. By collecting the actual root cause confirmation and replacement spare parts information entered by the maintenance personnel after performing maintenance, when there is a deviation between the actual root cause and the predicted root cause, new fault mode items are supplemented to the fault phenomenon library and the propagation time delay parameters and propagation probability weights in the propagation correlation library are corrected, so that the knowledge base coverage and diagnostic accuracy continuously improve with case accumulation, ensuring the safe and stable operation of green power plant equipment under network limited or complete offline conditions. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is an embodiment of the present application for the intelligent operation and maintenance management method steps of green power plants; Figure 2 is a structural block diagram of the intelligent operation and maintenance management system for green power plants in the embodiment of the present application.

[0017] The implementation of the present application, functional characteristics and advantages will be further described with reference to the accompanying drawings. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical scheme and advantages of the present application more clear, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0019] Referring to Figure 1 , the present embodiment provides an intelligent operation and maintenance management method for green power plants, comprising the following steps: S1, collecting fault phenomenon descriptions from the SCADA system, and comparing the fault phenomenon descriptions with the local fault phenomenon library to select candidate historical cases; S2, query the propagation path template based on the candidate historical cases, and locate the source equipment component based on the propagation path template; S3 extracts diagnostic rules that match the source equipment components from the local expert maintenance rule base and inputs the diagnostic rules into the locally deployed lightweight language model to generate a structured diagnostic report.

[0020] In one example, fault descriptions are collected from the SCADA system and compared with a local fault database to select candidate historical cases, including: The gearbox vibration acceleration, oil temperature, generator current fluctuation amplitude and ambient temperature parameters are obtained through the communication interface between the edge computing node and the SCADA system. Based on the equipment nameplate parameters and historical normal operation data, the normal operation range baseline of each parameter is established through statistical analysis. When the gearbox vibration acceleration exceeds the upper limit of the baseline of the normal operating range and the duration reaches the preset duration, it is marked as a hard threshold abnormality. When the oil temperature exceeds the limit and the rate of rise exceeds the normal rate of rise threshold, it is marked as a trend abnormality. The combination of parameters that simultaneously meet the hard threshold abnormality and trend abnormality forms a fault phenomenon description. Based on the description of the fault phenomenon, the candidate historical cases with the highest similarity are selected from the local fault phenomenon database.

[0021] In this example, the edge computing node establishes a data interaction connection with the SCADA system of the green power plant through the completed Modbus TCP communication interface, and real-time obtains the operating parameters of the target wind turbine generator set according to a sampling period of once per second, including the gear box vibration acceleration, the gear box oil temperature, the short-time fluctuation amplitude of the generator stator or rotor current, and the current environmental temperature, wherein the vibration acceleration is used to monitor the dynamic stability of the mechanical system, the oil temperature parameter reflects the lubrication and heat dissipation working condition, the current fluctuation amplitude reflects the response state of the power generation load side, and the environmental temperature is used as an external working condition variable. After the edge node receives the above parameters, it inputs them into the local running interval modeling module, which establishes the upper and lower limits of the normal running interval of each parameter based on the equipment nameplate parameters such as the maximum allowable oil temperature of the gear box, the allowable vibration intensity, the rated power generation current, and the applicable range of environmental conditions, and combines the recorded data of the equipment in stable running state in the past 30 days to establish the upper and lower limits of the normal running interval of each parameter through the mean plus three times standard deviation method. For example, if the past 30 days gear box vibration acceleration mean is 2.1 mm / s, and the standard deviation is 0.6 mm / s, then the running interval upper limit is 2.1+3×0.6=3.9 mm / s, when the real-time collection value exceeds the running interval upper limit and the duration reaches or exceeds the set 20 seconds, it can be determined as a hard threshold abnormality; similarly, the gear box oil temperature is set to the environmental temperature plus 55℃, i.e. 70℃, as the reference running upper limit under the condition of current environmental temperature of 15℃, if the current oil temperature rises to 88℃, the overrun amplitude is 18℃, and if this temperature rise occurs within 5 minutes, i.e. the per minute rising rate is 3.6℃, which is higher than the pre-set normal temperature rising rate upper limit of 0.8℃ / min, therefore the oil temperature abnormality not only triggers the static upper limit, but also meets the criterion of trend rising abnormality. When both the vibration acceleration and the oil temperature meet their respective abnormality determination conditions, a collaborative abnormality event is constituted, the model induces the feature combination composed of multiple indicators jointly abnormal into a structured fault phenomenon description, which contains fields such as abnormal parameter name, current value, overrun amplitude, rising rate, abnormal starting time, duration, and environmental temperature at the time of abnormality occurrence, and combines device identification such as WT-03 wind turbine and running state such as grid-connected power generation as supplementary information. The fault phenomenon description is input into the local first layer fault phenomenon knowledge base as a query condition, and the historical case matching process is carried out through the defined multi-feature similarity weighting formula, which considers the consistency of the main abnormal parameter type, the relative deviation of the abnormal value, the intersection proportion of the secondary abnormal parameter combination, and the matching degree of the environmental condition and gives different weights. For example, in one matching, if the vibration parameter is completely the same, it gets 1.0 points, the oil temperature deviation is (88-85) / 85=3.5%, which gets 0.965 points, the two secondary parameters coincide, which gets 0.667 points, and the environmental condition completely matches, which gets 1.0 point, the comprehensive similarity score is calculated.All fan-related historical fault modes in the fault phenomenon library are scored and sorted one by one, and the candidate case with the highest score is selected as the reference case of the current abnormal event.

[0022] Before extracting the current fault feature vector from the fault phenomenon description and retrieving the stored historical fault mode record in the local fault phenomenon library, it also includes: querying the device type, service life and historical fault frequency from the green power plant device account according to the abnormal device identification, dividing the local fault phenomenon library into fan fault partition, photovoltaic fault partition and energy storage fault partition according to the device type, establishing high-frequency fault index table and low-frequency fault index table in the corresponding partition according to the fault frequency from high to low, loading the fault mode record in the high-frequency fault index table into the local server memory to form a hot cache area; the change trend of the main abnormal parameter value extracted from the fault phenomenon description includes the continuous rising time length of the gearbox vibration acceleration and the rising rate of the oil temperature, according to the change trend, match the gradual fault feature or the sudden fault feature in the hot cache area, when the continuous rising time length exceeds the preset time length and the rising rate is lower than the preset rate threshold, it is determined as a gradual fault and the fault mode record marked as bearing wear in the high-frequency fault index table is preloaded, when the rising rate exceeds the preset rate threshold, it is determined as a sudden fault and the fault mode record marked as bearing fracture is preloaded; group the preloaded fault mode records according to the device component type, count the number of historical cases in each group, when the number of historical cases in a certain group is lower than the preset sample threshold, supplement the loading of fault mode records of corresponding device component type from the low-frequency fault index table, form a retrieval range set containing the hot cache area and the supplemented loading records, and subsequent similarity calculation is only performed in the retrieval range set without traversing the complete knowledge base.

[0023] Before selecting the candidate historical case with the highest similarity from the local fault phenomenon library according to the fault phenomenon description, it also includes: for rare fault types in the local fault phenomenon library, the number of samples of which is lower than a preset threshold, extracting device physical parameters associated with the rare fault types, including gear pair meshing stiffness of a gear box, bearing radial clearance, lubricating oil viscosity temperature coefficient, and electrochemical impedance model parameters of a battery monomer, establishing a device multi-physical field coupling simulation model according to the extracted device physical parameters; injecting a fault excitation in the device multi-physical field coupling simulation model, simulating bearing wear by reducing the bearing radial clearance, simulating battery aging by increasing the battery internal resistance, setting the environmental temperature and load power boundary conditions, running the simulation to obtain simulation time series data of the gear box vibration acceleration, oil temperature and battery monomer voltage, and extracting abnormal parameter combinations from the simulation time series data to form a virtual fault sample; labeling the virtual fault sample with device type, fault type, trigger condition and failure mechanism to generate a virtual fault mode entry, supplementing the virtual fault mode entry to the local fault phenomenon library, and setting a simulation source identifier and a confidence weight parameter in the virtual fault mode entry, so that the weight of the virtual fault sample is lower than that of the real fault sample in subsequent similarity matching.

[0024] In one example, selecting the candidate historical case with the highest similarity from the local fault phenomenon library according to the fault phenomenon description includes: extracting a current fault feature vector containing main abnormal parameter values, secondary abnormal parameter combinations and current environmental conditions from the fault phenomenon description, retrieving stored historical fault mode records in the local fault phenomenon library, and extracting corresponding historical fault feature vectors for each historical fault mode record; calculating the similarity between the current fault feature vector and each historical fault feature vector, and selecting the candidate historical case with the highest similarity.

[0025] In this example, the structured failure phenomenon description data records the type and real-time value of the primary abnormal parameter (e.g., the gear box oil temperature is 88°C), the combination of secondary abnormal parameters (e.g., vibration acceleration 5.2 mm / s and current fluctuation amplitude 9.3%), and the environmental conditions at the time of failure (e.g., ambient temperature 15°C, wind speed 8.5 m / s). These three types of information are encoded separately to form the feature vector of the current failure, where the type and value of the primary abnormal parameter are mapped to categorical encoding and normalized value respectively, the secondary abnormal parameter set is constructed by combining the hash method to form the vector dimension, and the ambient temperature and wind speed are normalized at a fixed ratio and appended to the end of the vector as continuous variables. From the local failure phenomenon library, read the stored historical failure mode entries in sequence. Each record in the knowledge base contains structured fields to support feature extraction operations, so for each historical failure mode, extract its primary abnormal parameter value, secondary abnormal parameter combination, and recording environment conditions in the same way to construct the feature vector of the historical failure. After completing the vector extraction, perform a one-by-one comparison operation, i.e., calculate the similarity index between the current failure feature vector and each historical failure feature vector. The similarity index is constructed in a weighted superposition form, with the primary abnormal parameter type matching degree weight set to 0.3, the primary parameter value proximity weight set to 0.4, the secondary parameter combination overlap weight set to 0.2, and the environmental condition consistency weight set to 0.1. The parameter type matching degree is assigned by judging whether the type is completely identical (score 1.0), the same subsystem (score 0.6), or completely different (score 0.2), the primary parameter value proximity is calculated as 1 minus the percentage deviation, the secondary parameter overlap is calculated as the ratio of the number of intersection elements to the number of union elements, and the environmental condition consistency is judged according to whether the current environment falls within the environmental range recorded in the historical case to determine whether the score is 1.0. Perform the similarity calculation process on all fault entries in the failure phenomenon library, sort the similarity values between each historical failure mode and the current failure feature, and select the top one or several as the optimal candidate historical case.

[0026] In one example, the candidate historical case is queried according to the propagation path template, and the source device component is located according to the propagation path template, including: Extracting the failure type identifier from the candidate historical case, retrieving the propagation path template matching the failure type identifier in the local failure propagation association library; Extracting the multi-level propagation node from the propagation path template, each level of the multi-level propagation node containing trigger condition definition, state determination basis and propagation delay parameter; Reverse matching the location of the current abnormal device in the green power plant topology with the propagation path template to locate the source device component of the root cause of the failure.

[0027] In this example, after completing the candidate historical case screening, the corresponding fault type identifier is extracted from the most similar case, for example, if the current case is a wind turbine gearbox bearing wear case, its fault type identifier is set to FT_WT_GB_BEARING_WEAR, which is input as a query keyword into the locally deployed fault propagation correlation library. In the fault propagation correlation library, the fault type identifier is used as a retrieval index field to associate fault propagation path templates. Each template describes a multi-level causal path of device state changes caused by the fault type in a graph structure, and constructs a complete topological relationship from the root cause node to the final system abnormality downstream in a node chain manner. After matching the corresponding propagation path template, each level of propagation node in the path is structured and extracted to build a propagation node sequence. Each propagation node in the propagation node sequence contains three types of core definitions: trigger condition definition, such as a node representing gearbox vibration enhancement, with a trigger condition set to vibration acceleration exceeding 4.5 mm / s and a main frequency between 200 Hz and 500 Hz; state judgment basis, such as a combination of quantitative parameters including vibration sensor output values from real-time data, frequency spectrum analysis results, and device operation log records; and propagation delay parameter, which defines the minimum and maximum time delay interval required for the state change from the upstream node to propagate to the current node, such as a delay of 12 hours to 48 hours from bearing wear to vibration enhancement. Once the above structured propagation path is established, the system performs reverse matching based on the topological location of the current abnormal device in the green power plant. During the topological matching process, the access structure information of the target wind turbine or photovoltaic device in the multi-energy complementary network of the green power plant is retrieved, such as the output of wind turbine WT-03 connected to the 35kV bus through the transformer, and coupled with energy storage PCS, photovoltaic inverter, and main transformer. This structure is used to constrain the physical reachability of the propagation path. According to the downstream nodes in the propagation path, the upstream nodes are traced back layer by layer to determine whether the propagation conditions at each level are met in the current device parameters, such as the observation of vibration enhancement, current fluctuation, and oil temperature rise in the current monitoring values, which are consistent with the second and third level propagation nodes in the propagation path template. Further verification is performed to determine whether the node trigger condition and time delay interval meet the definition. If the oil analysis data required by an upstream node such as "bearing wear" is also found to meet the corresponding metal element content standard in the historical maintenance record, the node is confirmed as the source device component of the fault based on the node matching degree, condition satisfaction degree, and topological consistency, thereby completing the root cause localization.

[0028] Before the current abnormal device position in the green power plant topology is reversely matched with the propagation path template after the multi-level propagation nodes are extracted from the propagation path template, further comprising: obtaining the real-time topology configuration of the current green power plant, including the device identification and connection relationship of the wind turbine, energy storage system and photovoltaic inverter, mapping the general device type node in the propagation path template to the specific device instance in the current topology, and generating an instantiated propagation path sequence according to the abnormal wind turbine output bus connection energy storage system identification and the energy storage system hanging battery cluster identification; querying the charge-discharge switching times of the energy storage system corresponding to the first level propagation node in the instantiated propagation path sequence from the SCADA system within a preset period before the fault occurs, comparing the queried charge-discharge switching times with the switching threshold defined in the propagation path template, and confirming that the first level propagation node is in an activated state when the charge-discharge switching times exceed the switching threshold; obtaining the latest internal resistance test record of the second level propagation node corresponding to the battery cell from the battery management system, calculating the increase ratio of the current internal resistance compared with the initial internal resistance and comparing it with the internal resistance threshold defined in the propagation path template, and confirming that the second level propagation node is in an activated state when the increase ratio exceeds the internal resistance threshold, and the proportion of the number of activated nodes in the instantiated propagation path sequence to the total number of nodes is calculated, and when the proportion exceeds a preset verification ratio, it is determined that the propagation path is effective.

[0029] In one example, after the multi-level propagation nodes are extracted from the propagation path template, further comprising: The parameters of the root cause node and the first level propagation node are parsed from the propagation path template, the root cause node defines the fluctuation amplitude of the wind turbine output power exceeding the preset proportion of the rated power as the trigger condition, and the first level propagation node defines the switching frequency of the energy storage system charge-discharge mode exceeding the preset switching threshold as the state determination basis; The propagation delay parameter of the root cause node pointing to the first level propagation node defines the delay time range from power fluctuation detection to the appearance of frequent switching state, and the second level propagation node defines that the increase ratio of the internal resistance measurement value of the battery cell obtained through impedance test compared with the initial internal resistance value exceeds the preset internal resistance threshold.

[0030] In this example, according to the candidate root cause case inferred from the current fault phenomenon, the propagation path template corresponding to the case fault type is loaded from the local fault propagation association library, the propagation path template is stored in a graph database structure, contains multiple nodes and edges, the node represents the device state change event, and the edge represents the propagation causal relationship between the nodes and the time delay parameter. Locate the root cause node and its directly connected downstream propagation nodes in the graph structure, and thus expand the item-by-item extraction of node parameter configuration. For the root cause node, its node attribute field defines an explicit trigger condition, where the root cause node represents the wind turbine output power fluctuation event, and the trigger condition is defined as the standard deviation value of the output power of the wind turbine generator set exceeding 15% of the rated power within a 10-minute sliding time window, and the fluctuation frequency cumulative value exceeding 8 times. The system will extract "power standard deviation threshold" as 15%, "time window" as 10 minutes, and "fluctuation frequency threshold" as 8 as parameter configuration items under the node, and archive them as the root cause trigger condition group. Locate the first level propagation node pointed by the root cause node, and the first level propagation node describes the generation mechanism of the frequent state of the energy storage system working mode, and the judgment basis is defined in the node configuration field as: the energy storage converter completes the charging to discharging or discharging to charging mode switching times within 1 hour after receiving the AGC (automatic generation control) scheduling instruction, which constitutes a frequent switching event, and the parameter field "switching statistical period" is extracted as 1 hour and "switching times threshold" is extracted as 20 times as the state judgment basis of the first level propagation node. Read the propagation time delay parameter from the connection edge between the root cause node and the first level propagation node, that is, the average delay time range between the occurrence of the high fluctuation event of the wind turbine output power and the entry of the energy storage system into the frequent switching state. The parameter is represented in the form of a time interval in the path template, for example, the lower limit is 15 minutes and the upper limit is 30 minutes. The system inserts the propagation time delay as a dynamic causal reasoning window into the subsequent time sequence verification module. Continue to descend along the propagation path, and the system reaches the second level propagation node, which represents the performance degradation of the battery monomer, that is, the increase of the battery internal resistance, and the judgment logic of the second level propagation node is defined in the propagation template as: after the energy storage system experiences frequent switching, the 1 kHz alternating current impedance test is performed on the battery monomer, and if the current measured internal resistance value increases by more than 30% compared with the reference internal resistance value under the initial working condition of the battery, the node state is entered. The 30% internal resistance growth ratio is extracted as "internal resistance growth threshold", and the impedance test method, test frequency and historical reference internal resistance value acquisition logic are recorded as the auxiliary judgment basis at the same time. Through the step-by-step analysis of node parameters and edge attributes, the propagation path template structure is structured and disassembled into a causal chain assembly body with a clear physical quantity correspondence.

[0031] Before the reverse matching of the position of the current abnormal device in the green power plant topology with the propagation path template, the method further comprises: querying time series data of the output power of the fan within a preset backtracking period before the abnormal start time from the SCADA system historical database, detecting a power fluctuation event in the time series data and recording the occurrence time of the power fluctuation event, querying a charge-discharge mode switching event within a preset backtracking period before the abnormal start time from the energy storage system charge-discharge record and recording the occurrence time of the switching event, and calculating the time interval between the occurrence time of the power fluctuation event and the occurrence time of the last switching event; comparing the calculated time interval with the propagation delay parameter from the root cause node to the first level propagation node defined in the propagation path template, determining that the first level propagation timing is consistent when the time interval falls within the delay time range defined by the propagation delay parameter, otherwise determining that the first level propagation timing is inconsistent, synchronously calculating the time interval between the occurrence time of the switching event and the battery cell internal resistance test time and comparing it with the propagation delay parameter of the second level propagation node to verify the second level propagation timing consistency; the proportion of the number of propagation levels with consistent timing in the total number of propagation levels in the multi-level propagation node sequence is calculated, and when the proportion exceeds the preset timing verification threshold, the propagation path template is determined to be an effective propagation path, and the propagation path template identifier and the timing verification score of the effective propagation path are recorded to the propagation path candidate list, and when the proportion is lower than the preset timing verification threshold, it is determined to be a false propagation path and is excluded from the propagation path candidate list. Only the effective propagation path in the propagation path candidate list is subjected to reverse matching and root cause positioning.

[0032] In one example, the reverse matching of the position of the current abnormal device in the green power plant topology with the propagation path template to locate the source device component of the root cause of the fault comprises: Obtaining the upstream and downstream device connection relationship of the current abnormal device in the green power plant topology, comparing the gearbox vibration acceleration and the generator current fluctuation amplitude in the fault phenomenon description with the multi-level propagation node, and when the gearbox vibration acceleration and the current fluctuation amplitude are found to match the characteristic parameters of the second level propagation node and the third level propagation node respectively, the multi-level propagation node is traced back to the root cause node; Querying the latest oil analysis report of the current abnormal device to obtain the iron element content and copper element content detection data in the oil, comparing the iron element content with the iron element content threshold defined in the root cause node, comparing the copper element content with the copper element content threshold defined in the root cause node, and when the iron element content and the copper element content both exceed the corresponding threshold, confirming that the source device component is the gearbox bearing.

[0033] In this example, after receiving the abnormal event diagnosis request, the topology connection structure of the current abnormal device is read from the locally deployed green power plant device topology database, taking the device identification as the retrieval index, for example, fan WT-03, whose output is connected to the 35kV bus through the box-type transformer, and the bus is connected to the photovoltaic inverter, energy storage converter and main transformer, to build the upstream and downstream device connection diagram centered on WT-03, and map it to a directed graph topology chain, which is used to constrain the node reachability in the propagation path template. The candidate propagation path template obtained by similarity screening is called, and the path node feature parameters are traversed level by level, compared with the vibration acceleration value and generator current fluctuation amplitude in the current fault phenomenon description, if the current vibration acceleration value exceeds 4.5mm / s, and the main frequency is between 200Hz and 500Hz, and the current fluctuation amplitude exceeds 8%, and the corresponding second level node "gearbox vibration intensification" and third level node "generator current fluctuation" in the propagation path meet the feature requirements, it is determined that the propagation path is highly matched with the current abnormal state. After the node parameters are aligned level by level, the root cause node is backtracked from the third level node to the root node along the propagation path, and the root cause node is defined as "gearbox bearing wear" in the template, and its necessary state confirmation conditions include: bearing wear exceeds the limit, and the concentration of metal particles in the oil increases. In order to verify whether the state of the root cause node is true, the oil detection record database associated with the current device WT-03 is called, and the latest oil analysis report is retrieved, which contains the test results of the concentration of various metal elements in the oil. Extract the iron and copper elements, and compare their values with the threshold values defined in the root cause node template, for example, the iron content of 235ppm is greater than the defined threshold value of 200ppm, and the copper content of 168ppm also exceeds the defined upper limit of 150ppm, so both metal pollution criteria are met. Based on the triple verification logic of the propagation path node matching result, topology path reachability verification result and oil analysis data matching, the root cause of the current abnormality is confirmed to be the mechanical wear fault of the input shaft bearing in the gearbox of the fan WT-03, and the conclusion is packaged as the field content in the "root cause positioning report", and the source device component, triggering basis, propagation chain and confidence score are marked.

[0034] Before the source device component is confirmed as a gearbox bearing and the diagnostic rule matched with the source device component is extracted from the local expert maintenance rule library, it also includes: extracting the historical propagation success rate of instantiating the nodes of each level of the propagation path sequence from the local fault propagation association library, calculating the conditional probability from the root cause node to the first level propagation node as the historical statistical probability that the root cause node exists fault when the first level node is activated, calculating the conditional probability from the first level to the second level node as the historical statistical probability that the first level node has triggered when the second level node is activated, multiplying the conditional probability of each level to obtain the joint propagation probability; for the root cause node matched by the reverse tracing, calculating the matching degree score of the feature parameters of the root cause node and the current abnormal device measurement parameters, multiplying the matching degree score and the joint propagation probability to obtain the confidence value of the current root cause node, when there are multiple propagation paths pointing to different root cause nodes, the confidence values of each root cause node are calculated respectively and sorted from high to low according to the confidence values; extracting the root cause node with the highest confidence value as the main root cause, when the difference between the second highest confidence value and the highest confidence value is less than a preset threshold, it is determined that there is a concurrent root cause, and the device identifier, component type and confidence value of the main root cause and the concurrent root cause are encapsulated as the root cause positioning result, and the maintenance guidance of the main root cause and the troubleshooting suggestion of the concurrent root cause are respectively output in the subsequent generated structured diagnostic report.

[0035] Before integrating the condition judgment logic in the diagnostic rule and the sequence of disposal steps, the fault phenomenon description and maintenance measure record in the candidate historical case, and the propagation path analysis result in the multi-level propagation node into the context information data package, it also includes: calculating the fault severity score according to the main abnormal parameter value in the fault phenomenon description, weighting and summing the ratio of the gear box vibration acceleration value to the preset vibration alarm threshold and the ratio of the oil temperature value to the preset temperature alarm threshold to obtain the fault severity score, determining the context information detail level according to the fault severity score, setting the high detail level when the fault severity score exceeds the preset severe fault threshold, and setting the standard detail level when the fault severity score is lower than the preset severe fault threshold; extracting the IF condition part of the condition judgment logic from the IF-THEN diagnostic rule record, including the judgment expression of the vibration threshold and the temperature threshold, extracting the key operation nodes of the disposal step sequence, including cutting off the power supply, disassembling the components, replacing the bearings and trial running verification, when the context information detail level is the high detail level, retaining the complete sub-steps and quality control points of the disposal step sequence, when the context information detail level is the standard detail level, only retaining the key operation nodes to form the simplified disposal step sequence; extracting the fault description keywords from the candidate historical case, including the equipment identification and the fault component name, extracting the maintenance measure key operation, including the replacement of spare parts model, discarding non-key information such as fault occurrence environment and operating personnel, assembling the context information data package according to the context information detail level, assembling the context information data package including the complete diagnostic rule, the first three candidate historical cases and the complete propagation path node for the high detail level, assembling the context information data package including the simplified diagnostic rule, the first candidate historical case and the key propagation node for the standard detail level, and controlling the reasoning time of the lightweight language model within the preset time limit by estimating the text length of the context information data package.

[0036] In one example, the diagnostic rule matching the source device component is extracted from the local expert maintenance rule library, and the diagnostic rule is input into the locally deployed lightweight language model to generate a structured diagnostic report, including: According to the source device component in the local expert maintenance rule library, the diagnostic rule that completely matches the source device component and the fault type is screened, and the diagnostic rule includes the condition judgment logic, the maintenance time window suggestion, the spare parts list and the disposal step sequence; The diagnostic rule is input into the locally deployed lightweight language model for reasoning to generate a structured diagnostic report.

[0037] In this example, the information fields in the root cause positioning report generated according to the root cause positioning stage include abnormal device identification, device type, source component name, and determined fault type identification (such as gear box input shaft deep groove ball bearing wear), and a hierarchical screening operation is performed in a local expert repair rule library using the device type and component type as joint indexes. The expert repair rule library uses a document type database structure to store all repair knowledge entries, each rule entry contains a device type label, a component name label, and a fault type label, and a nested structured field is used to describe the diagnosis logic and repair specification of this type of fault. From the rule library, a rule set completely matching the target device type (such as a wind turbine generator) is screened out, a subset of rules related to the target component (such as a gearbox) is screened out, and a final rule entry completely matching the bearing wear type is screened out according to the fault type, for example, the gearbox bearing wear diagnosis rule numbered R_WT_GB_001. The diagnosis rule entry content is composed of multiple key segments: the conditional judgment logic part contains joint criteria based on real-time operating parameters and maintenance detection data, for example, define when the gearbox temperature exceeds 85℃ and the vibration acceleration is greater than 4.5mm / s to trigger the oil analysis process, if the iron element concentration in the oil exceeds 200ppm and the copper element concentration exceeds 150ppm, the bearing wear is confirmed; the repair time window suggestion segment divides the shutdown arrangement strategy according to the fault severity level, if the vibration value is between 4.5mm / s and 6.0mm / s and the oil temperature is between 85℃ and 95℃, it is determined as a moderate fault, and it is recommended to arrange maintenance within 72 hours, if it exceeds this range, it is considered as a severe fault and requires immediate shutdown; the spare parts list segment clearly lists the component model and quantity to be replaced, for example, 2 sets of SKF-6328 bearings for the input shaft and 2 sets of SKF-6230 bearings for the output shaft for preventive replacement, and also includes lubricating oil, sealing gaskets, and standardized tools and equipment such as hydraulic pullers, bearing heaters, etc.; the disposal step sequence segment lists eight standard repair steps from shutdown, blade locking, gearbox opening, oil sample extraction, bearing disassembly, heating installation, oil injection, and cover closing, each step is accompanied by operation points and safety precautions. The structured rule entry is used as input, and the root cause positioning report, current abnormal parameter description, and similar historical case set are arranged into a unified input format and transmitted to the lightweight language model deployed on the GPU acceleration unit of the local edge computing node. The lightweight language model is a LLaMA-2 large language model architecture based on 7B parameters, which has been fine-tuned on green power plant operation and maintenance field knowledge corpus and has integrated context modeling capability.The language model automatically identifies the association between the diagnosis intent and the parameter field after receiving the input, performs context reasoning based on the rule text and the current state data, and automatically generates a structured diagnosis report containing four modules: a root cause analysis module that explains the process of vibration enhancement and oil temperature rise caused by bearing wear; an impact assessment module that quantifies the system-level consequences of power reduction, current fluctuation, and potential busbar impact; a maintenance priority module that proposes the fault emergency level and the recommended downtime window; and a step-by-step disposal module that outputs the complete maintenance operation sequence and marks the key quality control points. The diagnosis report is pushed to the operation and maintenance personnel APP, the remote monitoring center large screen, and the fan HMI interface through the interface.

[0038] In one example, the diagnosis rules are input into a locally deployed lightweight language model for reasoning to generate a structured diagnosis report, including: The condition judgment logic and disposal step sequence in the diagnosis rules, the fault phenomenon description and maintenance measure record in the candidate historical cases, and the propagation path analysis result in the multi-level propagation nodes are integrated into a context information data package. The context information data package is input into a lightweight language model deployed on a local server for text understanding and comprehensive reasoning. The lightweight language model generates a structured diagnosis report after reading the context information data package. The structured diagnosis report includes a root cause analysis paragraph that clearly identifies the source equipment components and failure mechanisms, an impact range assessment paragraph that lists the affected downstream equipment and potential risk equipment, and a step-by-step maintenance guidance paragraph that details the specific operation requirements for isolating the faulty equipment to the recovery operation in chronological order.

[0039] In this example, the root cause positioning module outputs the source device component name and the fault type as the screening basis, extracts the corresponding diagnostic rule entries from the expert repair rule library, where the condition judgment logic part is like temperature overrun + vibration overrun → perform oil analysis → metal element concentration threshold judgment wear is established, and the maintenance step sequence is like power off, lock paddle, open cover, sampling, replace bearing, heating installation, oil injection, trial operation, etc. is parsed item by item and classified by field; At the same time, the fault phenomenon description field (such as vibration 5.2mm / s, oil temperature 88℃, current fluctuation 9.3%, ambient temperature 15℃) and maintenance measures record field (such as replace SKF-6328 bearing, 200 liters of lubricating oil, replace time 6.5 hours, vibration after recovery to 2.3mm / s) with high similarity score are extracted from the historical fault cases, which have relevance to the current fault structure, as "analog support information" for model understanding. The matched node data in the fault propagation path template is extracted as a propagation chain sequence, including the root node (bearing wear), the second level node (vibration enhancement), the third level node (current fluctuation), the fourth level node (power reduction), etc. Each level node is equipped with trigger conditions, state confirmation standards and propagation delay parameters, which together form an upstream and downstream logical closed loop chain. The above three types of information (rules, cases, propagation path) are fused and packaged into a unified context information data packet, represented by structured JSON or multi-segment plain text nested format, each segment of content is attached with a content identification field for model to identify semantic boundaries, and input through memory mapping interface to the lightweight language model deployed on the edge server. The model is based on LLaMA-2 architecture, fine-tuned on green power plant fault knowledge corpus, and has the ability to understand industrial diagnostic semantics in an instruction manner. After receiving the data packet, the language model identifies the role relationship and causal direction of each segment of content at the semantic layer based on its context modeling ability and instruction alignment ability, for example, vibration enhancement and bearing wear are linked as a cause-effect pair, oil metal exceeding standard and maintenance measures are matched as a confirmation chain, and the propagation path is bound with the device topology relationship to form a complete semantic link.The final model automatically generates a structured diagnostic report text, which consists of three main paragraphs. The root cause analysis paragraph points out that the source equipment component is the gearbox input shaft deep groove ball bearing, and the failure mechanism is that the bearing gap increases due to fatigue wear, causing increased vibration, friction heating, and inducing current fluctuations. In addition, evidence of metal particle contamination in the oil confirms the validity of the failure. The impact range assessment paragraph lists the affected generator control loop and output power stability module based on the propagation chain analysis, and then proposes that the downstream 35kV bus-connected energy storage system and photovoltaic inverter have potential risks of frequent charge-discharge switching and bus voltage fluctuations. The step-by-step maintenance guidance paragraph details the entire process of how to perform the operation from power-off, locking the blades, opening the cover, sampling, disassembly, replacing parts, oil injection, and commissioning. It also marks the key quality control points and safety operation points, such as bearing heating temperature not less than 100℃, oil injection volume 200 liters, no-load operation for 30 minutes, and real-time monitoring of vibration and temperature parameters returning to the baseline range.

[0040] In one example, the intelligent operation and maintenance management method for green power plants further includes: Obtaining maintenance execution feedback data entered by field operation and maintenance personnel through a mobile terminal, and determining that there is a deviation when it is found that the source equipment component in the maintenance execution feedback data contains additional failure components that are not predicted in the structured diagnostic report; If there is a deviation, a new failure mode item is generated, and the new failure mode item is supplemented to the local fault phenomenon library. In addition, a propagation path record from the predicted component failure to the additional failure component is added in the local fault propagation correlation library, and the propagation time delay parameter and the propagation probability weight are set in the propagation path record.

[0041] In this example, during the maintenance task execution phase, each item of operation data entered by the field maintenance personnel through the mobile terminal APP is continuously monitored, including the maintenance task number, task start time, operation step completion status, field image upload record, component replacement, operation parameter recovery, and maintenance end time, etc. In the operation step feedback, the maintenance personnel can select the "fault component correction" function through the interface to actively mark the newly discovered additional fault components that are not included in the diagnosis report but are actually found during maintenance, and provide image proof, brief description and location identification. For example, when executing the bearing disassembly step, it is found that not only the input shaft bearing is severely worn, but also the intermediate shaft bearing retainer is broken. The system identifies the intermediate shaft bearing as an additional fault component that is not predicted in the structured diagnosis report. After analyzing the maintenance execution feedback data, the entered fault component set is compared with the predicted component set output by the root cause positioning module in the original structured diagnosis report at the field level. If the component number, installation location, type identification or failure characteristics are not in the predicted set, it is determined that the current maintenance result has a diagnosis deviation, and the deviation type is classified as "incomplete prediction". After confirming the existence of the deviation, a new fault mode entry is generated based on the feedback data, and the number is automatically generated according to the equipment type, timestamp and serial number, for example, CASE_WT_GB_0217. The new mode is registered in the fault phenomenon library in a structured form, including fault ID, associated equipment, main abnormal parameter combination, secondary abnormal parameter, environmental working condition, confirmed root cause, involved component list (such as input shaft bearing + intermediate shaft bearing), maintenance measure, recovery effect and data entry time, etc. fields, to ensure that the new mode has complete data closure. After the fault mode entry is supplemented to the local fault phenomenon library, the fault sample capacity of the wind turbine generator in the library increases from N to N+1, increasing the expression ability of double bearing type composite faults. At the same time, the association path supplement process is started in the fault propagation association library, the original predicted component (input shaft bearing) is taken as the upstream node in the propagation path graph, and the newly identified additional component (intermediate shaft bearing) is taken as the downstream node, to build a new propagation path from "input shaft bearing wear" to "intermediate shaft bearing abnormality". The new propagation path is supplemented by the following fields through the structure template: propagation path number, predecessor node ID, successor node ID, propagation time delay parameter (for example, set to 24 hours to 72 hours after the input shaft wear starts), propagation condition expression (such as the expansion of the axial vibration frequency to the intermediate shaft frequency band), propagation probability weight (such as assigning a value of 0.65 according to the frequency of the field feedback case, indicating that 65% of the input shaft faults will induce intermediate shaft abnormalities). After the path record is completed, it is appended to the Neo4j graph database structure.

[0042] Referring to Figure 2 The embodiment provides an intelligent operation and maintenance management system for a green power plant, comprising: The collection module 1 is configured to collect the fault phenomenon description from the SCADA system, compare the fault phenomenon description with the local fault phenomenon library, and select a candidate historical case; The positioning module 2 is configured to query a propagation path template according to the candidate historical case, and locate a source device component according to the propagation path template. The diagnosis module 3 is configured to extract a diagnosis rule matched with the source device component from a local expert maintenance rule library, and input the diagnosis rule into a locally deployed lightweight language model to generate a structured diagnosis report.

[0043] In the embodiment, the specific implementation of each unit in the system embodiment is described above in the method embodiment, and will not be described here.

[0044] It should be noted that in this paper, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, system, article or method including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or includes elements inherent to such process, system, article or method. Without more limitations, the element defined by the statement "including a" does not exclude the existence of other identical elements in the process, system, article or method including the element.

[0045] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation using the content of the present application specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A smart operation and maintenance management method for green power plants, characterized in that, include: Fault phenomenon descriptions are collected from the SCADA system and compared with the local fault phenomenon database to select candidate historical cases; Based on the candidate historical cases, query the propagation path template, and locate the source device component based on the propagation path template; Diagnostic rules matching the source equipment components are extracted from the local expert maintenance rule base, and the diagnostic rules are input into a locally deployed lightweight language model to generate a structured diagnostic report.

2. The intelligent operation and maintenance management method for green power plants according to claim 1, characterized in that, Fault descriptions are collected from the SCADA system and compared with the local fault database to select candidate historical cases, including: The gearbox vibration acceleration, oil temperature, generator current fluctuation amplitude and ambient temperature parameters are obtained through the communication interface between the edge computing node and the SCADA system. Based on the equipment nameplate parameters and historical normal operation data, the normal operation range baseline of each parameter is established through statistical analysis. When the gearbox vibration acceleration exceeds the upper limit of the baseline of the normal operating range and the duration reaches the preset duration, it is marked as a hard threshold anomaly. When the oil temperature exceeds the limit and the rate of rise exceeds the normal rate of rise threshold, it is marked as a trend anomaly. The combination of parameters that simultaneously satisfy the hard threshold anomaly and the trend anomaly forms a fault phenomenon description. Based on the description of the fault phenomenon, the candidate historical cases with the highest similarity are selected from the local fault phenomenon database.

3. The intelligent operation and maintenance management method for green power plants according to claim 2, characterized in that, Based on the description of the fault phenomenon, the candidate historical cases with the highest similarity are selected from the local fault phenomenon database, including: Extract the current fault feature vector containing the values ​​of the main abnormal parameters, the combination of secondary abnormal parameters and the current environmental conditions from the fault phenomenon description; retrieve the stored historical fault mode records in the local fault phenomenon database; and extract the corresponding historical fault feature vector for each historical fault mode record. Calculate the similarity between the current fault feature vector and each of the historical fault feature vectors, and select the candidate historical case with the highest similarity.

4. The intelligent operation and maintenance management method for green power plants according to claim 3, characterized in that, Based on the candidate historical cases, a propagation path template is queried, and the source device component is located based on the propagation path template, including: Extract fault type identifiers from the candidate historical cases, and retrieve propagation path templates that match the fault type identifiers from the local fault propagation association database; Multi-level propagation nodes are extracted from the propagation path template. Each level of the multi-level propagation node includes a trigger condition definition, a state determination basis, and a propagation delay parameter. The location of the current abnormal device in the green power plant topology is reverse-matched with the propagation path template to locate the source device component where the root cause of the fault lies.

5. The intelligent operation and maintenance management method for green power plants according to claim 4, characterized in that, After extracting multi-level propagation nodes from the propagation path template, the method further includes: The parameter configurations of the root cause node and the first-level propagation node are parsed from the propagation path template. The root cause node defines a preset proportion in which the output power fluctuation of the wind turbine exceeds the rated power as a trigger condition. The first-level propagation node defines a preset switching threshold in which the number of times the energy storage system switches between charging and discharging modes as a status determination basis. The propagation delay parameter of the root cause node pointing to the first-level propagation node is defined as the delay time range from the detection of power fluctuation to the occurrence of frequent switching state, and the second-level propagation node is defined as the increase ratio of the internal resistance measurement value obtained by the impedance test of the battery cell to the initial internal resistance value exceeds the preset internal resistance threshold.

6. The intelligent operation and maintenance management method for green power plants according to claim 5, characterized in that, The location of the current faulty device in the green power plant topology is reverse-matched with the propagation path template to pinpoint the source equipment component where the root cause of the fault lies, including: Obtain the upstream and downstream equipment connection relationship of the current abnormal equipment in the green power plant topology, compare the gearbox vibration acceleration and the generator current fluctuation amplitude in the fault phenomenon description with the multi-level propagation nodes, and when it is found that the gearbox vibration acceleration and the current fluctuation amplitude match the characteristic parameters of the second-level propagation node and the third-level propagation node respectively, trace back to the root cause node along the multi-level propagation nodes; The most recent oil analysis report of the currently abnormal equipment is queried to obtain the detection data of iron and copper content in the oil. The iron content is compared with the iron content threshold defined in the root cause node, and the copper content is compared with the copper content threshold defined in the root cause node. When both the iron content and the copper content exceed the corresponding threshold, the source equipment component is confirmed to be the gearbox bearing.

7. The intelligent operation and maintenance management method for green power plants according to claim 6, characterized in that, Diagnostic rules matching the source equipment components are extracted from the local expert maintenance rule base, and these diagnostic rules are input into a locally deployed lightweight language model to generate a structured diagnostic report, including: Based on the source equipment component, the diagnostic rules are selected from the local expert maintenance rule base and are completely matched with the source equipment component and the fault type. The diagnostic rules include condition judgment logic, maintenance time window suggestions, spare parts list and handling step sequence. The diagnostic rules are input into a locally deployed lightweight language model for inference to generate a structured diagnostic report.

8. The intelligent operation and maintenance management method for green power plants according to claim 7, characterized in that, The diagnostic rules are input into a locally deployed lightweight language model for inference to generate a structured diagnostic report, including: The condition judgment logic and handling step sequence in the diagnostic rules, the fault phenomenon description and maintenance measure record in the candidate historical cases, and the propagation path analysis results in the multi-level propagation nodes are integrated into a context information data packet. The context information data packet is then input into a lightweight language model deployed on a local server for text understanding and comprehensive reasoning. After reading the context information data packet, the lightweight language model generates a structured diagnostic report. The structured diagnostic report includes a root cause analysis section that clarifies the source equipment components and failure mechanisms, an impact scope assessment section that lists the affected downstream equipment and potentially risky equipment, and a step-by-step maintenance guidance section that details the specific operational requirements for isolating the faulty equipment and restoring operation in chronological order.

9. The intelligent operation and maintenance management method for green power plants according to claim 8, characterized in that, The intelligent operation and maintenance management method for green power plants also includes: Obtain maintenance execution feedback data entered by on-site maintenance personnel via mobile terminals. When it is found that the source equipment component in the maintenance execution feedback data contains an additional faulty component that was not predicted in the structured diagnostic report, it is determined that there is a deviation. If a deviation exists, a new fault mode entry is generated and added to the local fault phenomenon library. At the same time, a propagation path record from the predicted component fault to the additional fault component is added to the local fault propagation association library. The propagation delay parameter and propagation probability weight are set in the propagation path record.

10. An intelligent operation and maintenance management system for green power plants, characterized in that, The steps for implementing the intelligent operation and maintenance management method for green power plants according to any one of claims 1 to 9, wherein the intelligent operation and maintenance management system for green power plants comprises: The data acquisition module is used to acquire fault phenomenon descriptions from the SCADA system, compare the fault phenomenon descriptions with the local fault phenomenon database, and select candidate historical cases. The positioning module is used to query the propagation path template based on the candidate historical cases, and to locate the source device component based on the propagation path template. The diagnostic module is used to extract diagnostic rules that match the source equipment components from the local expert maintenance rule base, and input the diagnostic rules into a locally deployed lightweight language model to generate a structured diagnostic report.

Images