Overhaul method and system based on triple consistent evaluation

Abstract

The invention relates to the technical field of new energy equipment operation and maintenance, and discloses a maintenance method and system based on triple consistent evaluation, and the method comprises the steps: obtaining multi-source heterogeneous data of a new energy station, and compressing the multi-source heterogeneous data into an abnormal fingerprint vector; combining the abnormal fingerprint vector with the equipment mechanism knowledge to obtain a candidate root cause; calculating the interpretation degree and the time sequence conformity of the candidate root causes, and constructing a candidate root cause set; performing simulation verification to obtain a simulation consistency result; the interpretation degree, the time sequence conformity and the simulation consistency result are fused, a scoring function is constructed, candidate root causes are screened and sorted, and a high-confidence-coefficient root cause set is obtained; according to the method, the explanation degree, the time sequence conformity and the simulation consistency result are fused through the comprehensive scoring function, the high-confidence-coefficient root cause set is converted into the maintenance step sequence, and the scientificity and the performability of the maintenance scheme are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of operation and maintenance technology for new energy equipment, specifically to a maintenance method and system based on triple consistency assessment. Background Technology

[0002] As the installed capacity of new energy power plants continues to expand, equipment such as wind turbine generators, photovoltaic inverters, energy storage converters, and step-up transformers operate in complex environments for extended periods, and their operating status is directly related to the safety and stability of the power system.

[0003] To ensure the efficient operation of all equipment, it is crucial to identify potential safety hazards before equipment failures occur and to implement repair measures quickly after a failure occurs. This places higher demands on maintenance methods. Summary of the Invention

[0004] This invention provides a maintenance method and system based on triple consistency assessment to solve the problem of how to provide maintenance for new energy power plants.

[0005] In a first aspect, the present invention provides a maintenance method based on triple consistency assessment, the method comprising: Acquire multi-source heterogeneous data from new energy power plants and compress the multi-source heterogeneous data into anomaly fingerprint vectors; By combining abnormal fingerprint vectors with knowledge of device mechanisms, candidate root causes are obtained; Calculate the explanatory power and temporal conformity of candidate root causes, and construct a candidate root cause set for candidate root causes whose explanatory power and temporal conformity both meet the preset conditions; Simulation verification was performed on the candidate root causes in the candidate root cause set to obtain simulation consistency results; By integrating the results of explanatory power, temporal consistency, and simulation consistency, a scoring function is constructed to screen and rank candidate root causes, resulting in a set of high-confidence root causes. The high-confidence root cause set is converted into a maintenance step sequence, an execution schedule and resource allocation plan are generated, and then converted into a work order and sent to the on-site operation and maintenance system.

[0006] This invention collects multi-source heterogeneous data from new energy power plants to avoid the one-sidedness of single-data-driven approaches. It combines anomaly fingerprint vectors based on multi-source heterogeneous data compression with equipment mechanism knowledge to locate the root cause of the fault. Through dual screening of interpretability and temporal consistency, false root causes are filtered out. The candidate root causes are verified by simulation, upgrading the diagnostic results from probabilistic inference to physical verifiability. Based on a comprehensive scoring function, the interpretability, temporal consistency, and simulation consistency results are integrated. The resulting set of high-confidence root causes is converted into a sequence of maintenance steps, thereby ensuring the scientific nature and feasibility of the maintenance plan.

[0007] In one alternative implementation, compressing multi-source heterogeneous data into anomaly fingerprint vectors includes: Signal preprocessing for multi-source heterogeneous data; Single-mode feature extraction is performed on the multi-source heterogeneous data after signal preprocessing to obtain vibration features, electrical features, thermal imaging features and acoustic features; Vibration features, electrical features, thermal imaging features, and acoustic features are concatenated into an initial feature vector. A multimodal encoder is then used to fuse the initial feature vector into multiple modes to generate an anomaly fingerprint vector.

[0008] This invention preprocesses multi-source heterogeneous data to eliminate environmental interference, extracts single-modal features to extract core features, splices the extracted features, performs multimodal fusion, eliminates feature islands, and captures cross-modal coupling faults.

[0009] In one optional implementation, calculating the explanatory power and temporal consistency of candidate root causes includes: Constructing knowledge graphs; The abnormal fingerprint vector is decoded into a set of symptoms using a mapping function; The explanatory power of each candidate root cause to the symptom set is quantified by using the conditional probability in the knowledge graph, thus obtaining the explanatory power of the candidate root cause. Verify whether each candidate root cause and its corresponding symptom satisfy the time window constraint of the knowledge graph to obtain the temporal conformity of the candidate root cause.

[0010] This invention constructs a knowledge graph and uses a mapping function to decode anomaly fingerprint vectors, quantifying the interpretability and temporal consistency of candidate root causes to ensure the interpretability of root causes and to ensure that root causes conform to the physical laws of fault evolution.

[0011] In one optional implementation, simulation verification is performed on candidate root causes in the candidate root cause set to obtain simulation consistency results, including: A unified interface is used to assemble electromechanical coupling models, electromagnetic transient models, and thermodynamic models into a simulation platform; The simulation platform is injected with operating conditions to obtain the simulation fingerprint; Consistency index is obtained by performing consistency calculations between simulated fingerprints and measured abnormal fingerprints.

[0012] This invention integrates multidisciplinary mechanisms to fully simulate the multi-physics coupling characteristics of new energy equipment, solving the problem that a single model cannot reproduce cross-domain faults. It injects operating conditions into the simulation platform to achieve directional mapping between root causes and features, avoiding the disconnect between simulation and root causes, improving the realism of simulation, and using consistency calculation to quantify the matching degree between simulation and actual measurement, providing reliable evidence for root cause tracing.

[0013] In one alternative implementation, the consistency index is calculated according to the following formula:

[0014] in, As a consistency indicator, For simulated fingerprints, To test abnormal fingerprints, This is the set of candidate root causes.

[0015] In one alternative implementation, the scoring function is as follows:

[0016] in, For rating, , , These are the weights for interpretability, temporal compliance, and simulation consistency results, respectively. , For the sake of explanation, For timing compliance, This is a consistency indicator for digital twins.

[0017] In one alternative implementation, screening and ranking candidate root causes includes: Remove root causes where the explanatory power is less than the preset threshold, the timing compliance is less than the preset threshold, or the simulation consistency result is less than the preset threshold. Root causes are ranked based on the scoring results, and a bundle search is used to retain root causes that meet the conditions.

[0018] This invention selects root causes based on three indicators: interpretability, timing compliance, and simulation consistency results, avoiding misjudgment based on a single indicator. It also employs a bundle search sorting method to adapt to on-site operation and maintenance resources.

[0019] Secondly, the present invention provides a maintenance system based on triple consistency assessment, the system comprising: The acquisition module is used to acquire multi-source heterogeneous data from new energy power plants and compress the multi-source heterogeneous data into anomaly fingerprint vectors. The module combines abnormal fingerprint vectors with device mechanism knowledge to obtain candidate root causes; The construction module is used to calculate the explanatory power and temporal conformity of candidate root causes, and constructs a set of candidate root causes that meet the preset conditions in terms of both explanatory power and temporal conformity. The simulation module is used to simulate and verify the candidate root causes in the candidate root cause set, and obtain simulation consistency results; The fusion module is used to integrate the results of explanatory power, temporal compliance and simulation consistency, construct a scoring function, screen and sort candidate root causes, and obtain a set of high-confidence root causes. The distribution module is used to convert the high-confidence root cause set into a maintenance step sequence, generate an execution schedule and resource allocation plan, convert it into a work order, and distribute the work order to the on-site operation and maintenance system.

[0020] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the maintenance method based on triple consistency evaluation described in the first aspect or any corresponding embodiment thereof.

[0021] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the maintenance method based on triple consistency evaluation described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating the maintenance method based on triple consistency evaluation according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a maintenance process based on triple consistency assessment according to an embodiment of the present invention; Figure 3 This is a structural block diagram of a maintenance system based on triple consistency evaluation according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0026] In related technologies, anomaly detection is typically performed using a single data-driven approach or by employing rule-based reasoning that relies on expert experience to analyze equipment operating data and obtain diagnostic results. However, these methods have limitations: data-driven approaches are susceptible to noise and sample distribution, lacking interpretability; and rule-based reasoning based on expert experience has low accuracy when dealing with complex coupled faults. Furthermore, maintenance processes largely rely on manual experience to formulate solutions, lacking standardized modeling of work steps and automatic optimization of resource constraints, resulting in insufficient maintenance efficiency and accuracy.

[0027] According to an embodiment of the present invention, a maintenance method based on triple consistency evaluation is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] This embodiment provides a maintenance method based on triple consistency assessment. Figure 1 This is a flowchart of a maintenance method based on triple consistency evaluation according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain multi-source heterogeneous data from new energy power stations and compress the multi-source heterogeneous data into anomaly fingerprint vectors.

[0029] In this embodiment of the invention, multi-source heterogeneous data from new energy power plants is acquired. This data covers primary equipment, secondary systems, on-site sensing, and operation and maintenance support. Primary equipment includes wind turbines, photovoltaic inverters, energy storage converters, and step-up transformers; secondary systems include protection devices, monitoring systems, PLCs, and sensor networks; on-site sensing includes vibration, acoustic, infrared thermal imaging, visible light monitoring, and lidar point clouds; and operation and maintenance support includes maintenance work orders, spare parts storage data, operating environment, and meteorological data.

[0030] It should be noted that after acquiring multi-source heterogeneous data from new energy power plants, the multi-source heterogeneous data is preprocessed, including time unification, spatial mapping, and data standardization.

[0031] Time unification: GPS / BeiDou time source and IEEE 1588 PTP protocol are used to ensure millisecond-level alignment of different modal signals; for low-frequency (1 Hz SCADA) and high-frequency (25.6 kHz vibration) data, "time slices" are established through multi-level resampling and sliding window methods to ensure that cross-modal features can correspond under a unified timestamp; Spatial mapping: Point clouds and images are transformed into BIM / GIS coordinate systems through extrinsic parameter matrices and bound to equipment component IDs to achieve a "one-to-one physical correspondence" in space. For example, the hot spot temperature of a transformer corresponds precisely to the shell area scanned by the point cloud, facilitating subsequent anomaly location; Data standardization: The numerical values ​​are processed through normalization and physical unit conversion, and the interface adopts the unified JSON / OPC UA format, which not only facilitates subsequent calculations, but also allows for seamless integration with the existing plant monitoring system.

[0032] Preprocessing multi-source heterogeneous data solves the problems of fragmented multimodal data, inconsistent sampling frequencies, and inconsistent spatial coordinates in existing new energy power stations, providing high-quality and fusionable data input for intelligent diagnosis.

[0033] After preprocessing the multi-source heterogeneous data from new energy power plants, the multi-source heterogeneous data is compressed into a unified anomaly fingerprint vector, making different signals comparable and fusionable.

[0034] Step S102: Combine the abnormal fingerprint vector with the device mechanism knowledge to obtain candidate root causes.

[0035] In this embodiment of the invention, abnormal fingerprint vectors and device mechanism knowledge are combined to achieve "knowledge + data" dual fusion and obtain candidate root causes.

[0036] Step S103: Calculate the explanatory power and temporal conformity of the candidate root causes, and construct a candidate root cause set for the candidate root causes whose explanatory power and temporal conformity both meet the preset conditions.

[0037] In this embodiment of the invention, the explanatory power of candidate root causes for a set of abnormal symptoms is quantified to obtain the explanatory power of candidate root causes. The temporal conformity of candidate root causes is obtained by examining whether the candidate root causes and their corresponding symptoms satisfy the time window constraints defined in the knowledge graph. Whether the explanatory power and temporal conformity of candidate root causes meet preset conditions is determined, and candidate root causes that meet the preset conditions are constructed into a candidate root cause set.

[0038] Step S104: Perform simulation verification on the candidate root causes in the candidate root cause set to obtain simulation consistency results.

[0039] In this embodiment of the invention, a digital twin model is used to simulate and verify the candidate root causes in the candidate root cause set, and the possible results obtained from causal reasoning are converted into physically verifiable simulation evidence to obtain simulation consistency results.

[0040] Step S105: The explanatory power, temporal consistency and simulation consistency results are integrated to construct a scoring function, and the candidate root causes in the candidate root cause set are sorted and filtered to obtain a high-confidence root cause set.

[0041] In this embodiment of the invention, a unified measurement is applied to the explanatory power, temporal compliance, and simulation consistency of candidate root causes, and a comprehensive scoring function is defined. The scoring function is used to sort and filter candidate root causes in the candidate root cause set, resulting in a high-confidence root cause set. This achieves high-confidence, interpretable, and traceable root cause diagnosis, ensuring that the diagnostic results are applicable and persuasive in practical engineering.

[0042] Step S106: Convert the high-confidence root cause set into a maintenance step sequence, generate an execution schedule and resource allocation plan, convert them into work orders, and send the work orders to the on-site operation and maintenance system.

[0043] In this embodiment of the invention, the diagnosed high-confidence root cause set is converted into an executable and verifiable sequence of maintenance steps, outputting a maintenance recipe package. Combined with constraints on personnel, tools, spare parts, and power grid dispatch, an optimal execution schedule and resource allocation plan are generated to ensure that maintenance tasks are completed under feasible, safe, and efficient conditions. The execution schedule and resource allocation plan are then converted into standardized work orders and distributed to the field operation and maintenance system, thereby ensuring that the diagnostic results are truly translated into executable maintenance actions.

[0044] The maintenance method based on triple consistency assessment provided in this embodiment collects multi-source heterogeneous data from new energy power plants to avoid the one-sidedness of single data-driven approaches. It combines anomaly fingerprint vectors based on multi-source heterogeneous data compression and equipment mechanism knowledge to locate the root cause of the fault. Through dual screening of explanatory power and temporal consistency, false root causes are filtered out. The candidate root causes are verified by simulation, upgrading the diagnostic results from probabilistic inference to physical verifiability. Based on a comprehensive scoring function, the explanatory power, temporal consistency, and simulation consistency results are integrated, and the resulting set of high-confidence root causes is converted into a sequence of maintenance steps, thereby ensuring the scientific nature and feasibility of the maintenance plan.

[0045] This embodiment provides a maintenance method based on triple consistency evaluation, the process of which includes the following steps: Step S201: Obtain multi-source heterogeneous data from new energy power stations and compress the multi-source heterogeneous data into anomaly fingerprint vectors.

[0046] Specifically, step S201 includes: Step S2011: Perform signal preprocessing on the multi-source heterogeneous data.

[0047] Step S2012: Single-mode feature extraction is performed on the multi-source heterogeneous data after signal preprocessing to obtain vibration features, electrical features, thermal imaging features and acoustic features.

[0048] Step S2013: The vibration features, electrical features, thermal imaging features and acoustic features are concatenated into an initial feature vector. The initial feature vector is then fused using a multimodal encoder to generate an abnormal fingerprint vector.

[0049] In this embodiment of the invention, the multi-source heterogeneous data is first preprocessed. The vibration signal, electrical signal, thermal imaging signal and acoustic signal are preprocessed respectively. For the vibration signal, band-stop filtering is used to remove power frequency interference, and wavelet packet denoising is used to decompose the high-frequency and low-frequency components in the vibration signal. For the electrical signal, harmonic decomposition and zero drift correction are used for preprocessing. For the thermal imaging signal, emissivity correction and non-uniformity repair are used for preprocessing. For the acoustic signal, short-time Fourier transform is used for preprocessing to obtain the time spectrum.

[0050] Then, single-mode feature extraction is performed on the multi-source heterogeneous data after signal preprocessing, extracting vibration features, electrical features, thermal imaging features, and acoustic features. For vibration features, root mean square value, kurtosis, and envelope amplitude ratio are extracted; for electrical features, negative-sequence and zero-sequence components, and harmonic distortion rate are extracted; for thermal imaging features, maximum temperature rise difference, isothermal region area, and temperature gradient are extracted; and for acoustic features, spectral centroid and transient impulse rate are extracted. The extracted single-mode features not only cover amplitude information but also reflect frequency, spatial, and temporal characteristics, ensuring that anomalous features in different dimensions can be captured.

[0051] Next, the vibration features, electrical features, thermal imaging features, and acoustic features are concatenated into an initial feature vector. Anomaly fingerprint vectors are generated using a multimodal encoder (which integrates an attention mechanism and a feature compression structure):

[0052] in, These are vibration, electrical, thermal imaging, and acoustic characteristics, respectively.

[0053] The formula for generating abnormal fingerprint vectors not only compresses the data, but also preserves cross-modal dependencies through an attention mechanism, such as the coupling characteristics between current distortion and hot spot temperature rise.

[0054] By preprocessing signals from multi-source heterogeneous data to eliminate environmental interference, performing single-modal feature extraction to extract core features in a targeted manner, and then splicing the extracted features to perform multimodal fusion, feature islands are eliminated and cross-modal coupling faults are captured.

[0055] Step S202: Combine the abnormal fingerprint vector with the device mechanism knowledge to obtain candidate root causes.

[0056] Please see details Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0057] Step S203: Calculate the explanatory power and temporal conformity of the candidate root causes, and construct a candidate root cause set for the candidate root causes whose explanatory power and temporal conformity both meet the preset conditions.

[0058] Specifically, step S203 includes: Step S2031: Construct a knowledge graph.

[0059] Step S2032: Use a mapping function to decode the abnormal fingerprint vector into a set of symptoms.

[0060] Step S2033: Quantify the explanatory power of each candidate root cause on the symptom set using the conditional probability in the knowledge graph to obtain the explanatory power of the candidate root cause.

[0061] Step S2034: Verify whether each candidate root cause and its corresponding symptom satisfy the time window constraint of the knowledge graph, and obtain the temporal conformity of the candidate root cause.

[0062] In this embodiment of the invention, a knowledge graph is first constructed. The hierarchical structure of the knowledge graph nodes includes an equipment layer, a component layer, an operating condition layer, a symptom layer, and a fault mode layer. The equipment layer includes wind turbines, photovoltaic inverters, energy storage converters, and step-up transformers; the component layer includes gearboxes, bearings, cooling fans, and power modules; the operating condition layer includes load levels, ambient temperature, wind speed, and humidity; the symptom layer includes abnormal vibration characteristic frequency amplitude, increased negative sequence current, increased infrared hotspots, and abnormal acoustic pulses; and the fault mode layer includes bearing peeling, radiator blockage, poor busbar contact, and power module insulation degradation. Edge types include causal relationships, sequential relationships, mitigation relationships, and inhibition relationships. Each edge has attributes, including conditional probability. Time window parameters , Confidence weights. Storage is handled by a graph database (Neo4j or similar), supporting high-concurrency queries and probability calculations to ensure real-time inference after field data is accessed.

[0063] Then, the abnormal fingerprint vector is used to... Decoding into a set of symptoms That is, to identify which abnormal features correspond to known equipment malfunction symptoms, the symptom set is as follows:

[0064] For example, abnormal peak values ​​of high-frequency vibration envelope signals may correspond to "poor gear meshing", while current harmonic distortion may correspond to "power module failure".

[0065] For each candidate root cause Its explanatory power is determined by the conditional probabilities in the knowledge graph. Indicates a candidate root cause Abnormal fingerprints Corresponding set of symptoms The degree of explanation, The expression is as follows:

[0066] The above formula indicates that if the root cause If it occurs, then it corresponds to the set of signs in the abnormal fingerprint. The explanatory power is represented by the product of conditional probabilities. If all symptoms can be reasonably explained, The higher the value.

[0067] The edges in the knowledge graph also store the chronological order. We verify whether each candidate root cause and its corresponding symptom satisfy the time window constraints of the knowledge graph to obtain the temporal logical compliance. If the root cause... At any moment If it occurs, then there are signs. Must appear Within the time window. During the calculation. At the same time, the time series verification results are superimposed. If it does not meet the requirements, the root cause will be automatically removed.

[0068] The candidate root cause set obtained at this time is: ,in, , This is the threshold parameter. Each candidate root cause is accompanied by one or more evidence paths (root cause-symptom chain) to facilitate tracing and interpretation.

[0069] By constructing a knowledge graph and using a mapping function to decode anomaly fingerprint vectors, the interpretability and temporal consistency of candidate root causes are quantified to ensure the interpretability of root causes and to ensure that root causes conform to the physical laws of fault evolution.

[0070] Step S204: Perform simulation verification on the candidate root causes in the candidate root cause set to obtain simulation consistency results.

[0071] Specifically, step S204 includes: Step S2041: Use a unified interface to assemble the electromechanical coupling model, electromagnetic transient model, and thermodynamic model into a simulation platform.

[0072] Step S2042: Inject operating conditions into the simulation platform to obtain the simulation fingerprint.

[0073] Step S2043: Perform consistency calculation between the simulated fingerprint and the measured abnormal fingerprint to obtain the consistency index.

[0074] In this embodiment of the invention, the digital twin model consists of three sub-models: an electromechanical coupling model, an electromagnetic transient model, and a thermodynamic model. The electromechanical coupling model establishes the rotation equations of the wind turbine drive chain and the generator, considering the dynamic characteristics of bearings, gears, and rotors. The electromagnetic transient model describes the current and voltage transients of the converter and transformer when there are voltage drops, short circuits, or sudden changes in contact resistance. The thermodynamic model simulates the heat transfer and convection processes of the heat sink, windings, and oil cooling system, and outputs the temperature distribution.

[0075] Each sub-model is assembled into a multi-physics coupled simulation platform through a unified interface. It can run independently or be coupled collaboratively to ensure the reproducibility of different failure modes.

[0076] For candidate root causes Operating condition injection is achieved through parameter perturbation: like =“Bearing spalling”, then introduce periodic impact stiffness variation into the electromechanical coupling model; like =“Radiator blockage” means that the effective heat transfer area is reduced and the thermal resistance is increased in the thermodynamic model; like If the cable has poor contact, then the contact resistance is increased and the thermal coupling effect is introduced into the electromagnetic transient model.

[0077] After the digital twin model runs, it outputs the corresponding simulation feature signal, which is processed by the same encoder as the abnormal fingerprint generation unit to obtain the simulation fingerprint. .

[0078] Simulated fingerprints Abnormal fingerprints as measured Comparisons are performed within the same feature space, and a consistency index is defined:

[0079] in, As a consistency indicator, For simulated fingerprints, To test abnormal fingerprints, This is the set of candidate root causes.

[0080] Its value is located in the range [0, 1], and the higher the value, the closer the simulation result is to the measured data. This index not only provides a similarity quantification, but also reveals the relative differences in the fingerprint space.

[0081] For each candidate root cause, a consistency index is calculated. If the consistency index is greater than or equal to a preset threshold, then... If the root cause is positive, it is retained; otherwise, it is downgraded or eliminated.

[0082] Meanwhile, the verification results will be stored as "simulation evidence," including injected parameters, model version number, simulation results, and consistency values, ensuring the traceability of the diagnostic results.

[0083] By integrating multidisciplinary mechanisms, the system fully simulates the multi-physics coupling characteristics of new energy equipment, solving the problem that a single model cannot reproduce cross-domain faults. By injecting operating conditions into the simulation platform, the system achieves a directional mapping between root causes and characteristics, avoiding the disconnect between simulation and root causes, and improving the realism of the simulation. By using consistency calculations to quantify the matching degree between simulation and actual measurement, the system provides reliable evidence for root cause tracing.

[0084] Step S205: The explanatory power, temporal consistency and simulation consistency results are integrated to construct a scoring function, and the candidate root causes in the candidate root cause set are sorted and filtered to obtain a high-confidence root cause set.

[0085] In this embodiment of the invention, the scoring function is as follows:

[0086] in, For rating, , , These are the weights for interpretability, temporal consistency, and simulation consistency results, respectively. These weights can be automatically optimized through historical work order statistics, expert experience, or reinforcement learning methods. , For the sake of explanation, For timing compliance, This is a consistency indicator for digital twins.

[0087] The higher the value, the stronger the consistency between the candidate root cause and the measured data, causal mechanism, and simulation evidence.

[0088] To sort and filter candidate root causes, a threshold filter is first applied to the candidate root cause set. , or In this case, the candidate root cause will be directly eliminated.

[0089] In the root cause set that passes the threshold, according to Sort the data and use the Beam Search method to retain the top results. k One result:

[0090] in, k The range is typically 5-10, used to balance coverage and accuracy.

[0091] After obtaining the high-confidence root cause, an "evidence tree" is generated, with the root cause node located at the root and each symptom node located at the leaf, with conditional probabilities labeled on the sides. Time window parameters, simulation consistency values The "evidence tree" path demonstrates a complete logical chain of reasoning, verification, and screening. This allows for a clear tracing of the source of diagnostic results, enhancing the transparency and interpretability of conclusions.

[0092] In traditional technologies, relying solely on probabilistic reasoning may lead to misjudgments of "high probability but not reproducible"; relying solely on simulation verification may result in the loss of on-site complexity due to model simplification; and relying solely on sequential logic may overlook complex situations involving cross-component coupling.

[0093] Through formula Through multi-source fusion, this platform achieves high-confidence, interpretable, and traceable root cause diagnosis, ensuring that the diagnostic results are applicable and persuasive in engineering practice.

[0094] Step S206: Convert the high-confidence root cause set into a maintenance step sequence, generate an execution schedule and resource allocation plan, convert them into work orders, and send the work orders to the on-site operation and maintenance system.

[0095] Specifically, the diagnosed set of high-confidence root causes is converted into an executable and verifiable sequence of maintenance steps. The Standard Operation Block (SOB) library is called to dynamically assemble maintenance tasks and automatically insert safety verification nodes to ensure that the solution is operable while meeting the safety regulations of the power industry.

[0096] 1. Standard Operation Block Library Design The library adopts a modular design, with each work block storing the following fields: Work step number and name: such as "shutdown and power off", "oil level check", "replace radiator fan"; Required spare parts: model, quantity, and inventory location; Required tools and equipment: such as hoisting tools, insulation tester, torque wrench; Personnel qualification requirements: such as high-voltage electrician certificate, hot work permit; Estimated working hours: given by historical work orders and statistical data; Safety permit type: such as power off permit, hot work permit, confined space permit; Pre- and post-dependencies: used to describe that this step can only be executed after certain other work is completed.

[0097] The job block library supports incremental expansion, and the platform dynamically corrects the working hours and resource fields through the work order feedback mechanism to achieve self-evolution.

[0098] 2. Root cause to job block mapping When a root cause is identified as the primary source of failure, the system will automatically retrieve the set of job blocks corresponding to it.

[0099] For example: Radiator blockage: The sequence of work blocks called is "shutdown and power off → oil level check → radiator disassembly → radiator cleaning or replacement → reinstallation and refueling → grid connection restoration"; Bearing spalling: The sequence of work blocks called is "Stop and isolate → Disassembly and repair → Bearing replacement → Dynamic balance test → Complete machine reset"; Poor busbar contact: The sequence of work blocks to be called is "Power outage maintenance → Busbar connection point inspection → Contact resistance test → Replacement or reinforcement → Insulation restoration → Grid connection commissioning".

[0100] This mapping relationship is stored in the "Root Cause-Work Block Index Table" to ensure the automatic generation of maintenance plans.

[0101] 3. Assemble the work diagram The set of job blocks is assembled in the form of a directed acyclic graph (DAG), and topological sorting ensures that predecessor and predecessor dependencies are satisfied:

[0102] in, This represents the job block node, and SafetyChecks represents the mandatory safety check node.

[0103] For example, before the "disassembly" step, "power-off isolation" and "residual energy release" verifications must be performed; before the "hot work welding" step, "hot work permit verification" must be completed.

[0104] 4. Security verification mechanism The platform will automatically insert safety nodes: Power outage verification: requires on-site electronic signature to confirm that power isolation has been completed; Hot work permit: requires approval record of hot work permit from the upstream supplier before proceeding; Confined space access: inserts detection nodes before inspecting enclosed spaces such as boxes and pipes to confirm oxygen concentration and harmful gas indicators.

[0105] These safety checks are tied to the work steps; if a check fails, the task cannot proceed.

[0106] 5. Output Results The final maintenance recipe package is output in a standardized data structure, including: an ordered sequence of work steps; a list of required resources (spare parts, tools, personnel qualifications); a safety ticket link; and a task dependency graph.

[0107] This recipe package serves as input for subsequent resource and scheduling optimization, ensuring that optimization goals are based on real, actionable steps.

[0108] By transforming high-confidence root cause sets into a sequence of maintenance steps, and converting abstract diagnostic results into actionable engineering steps, the solution is guaranteed to be standardized, safe, and traceable. Through DAG models and safety verification mechanisms, safety accidents caused by incorrect operation sequences or missing documentation are avoided.

[0109] Based on the maintenance formula package, and taking into account on-site personnel, tools, spare parts, and power grid dispatch constraints, an optimal execution schedule and resource allocation plan are generated to ensure that maintenance tasks are completed under feasible, safe, and efficient conditions.

[0110] 1. Input content This unit receives standardized output from the maintenance recipe generation unit, including: the sequence of work steps and their dependencies; required spare parts and their inventory status; required tools and quantities; personnel qualifications and team information; safety permit requirements; and estimated working hours.

[0111] 2. Constraints During the scheduling process, the following constraints must be met simultaneously: Resource capacity constraints: The number of similar tools and workstations used at the same time must not exceed the capacity limit; Personnel qualification constraints: For example, "high-voltage electrical work" can only be performed by certified electricians; Inventory constraints: The amount of spare parts used must not exceed the current inventory or the expected arrival quantity; Downtime window constraints: Maintenance must be completed within the downtime window allowed by the power grid dispatch; Logical constraints: If the steps rely Then it must satisfy .

[0112] 3. Optimize objectives The system models the scheduling problem as a multi-objective optimization task, with the objective function being:

[0113] in, To comprehensively consider labor costs, spare parts costs, and downtime losses, This represents the total downtime of the equipment. The risk reduction before and after maintenance is evaluated by comparing the failure rate function. The safety score is calculated based on indicators such as ticket compliance rate and hazardous operation risk factors. For the weight parameters, satisfying .

[0114] 4. Optimization Algorithm For small-scale scenarios (number of tasks < 50): a mixed-integer linear programming (MILP) model is used to directly solve for the globally optimal schedule in a commercial optimizer.

[0115] Large-scale scenarios: Conservative Q-Learning (CQL) is adopted to train scheduling strategies based on historical work order data, ensuring that near-optimal results are obtained under safety constraints and that real-time adaptability is available.

[0116] 5. Output Results The final schedule includes: each job step start and end times ; Assigned work teams and personnel; Corresponding resource set (tools, spare parts); Insertion location of safety ticket verification nodes.

[0117] The results are output in a table or JSON format and directly integrated with the work order execution system to ensure on-site implementation.

[0118] Based on the maintenance prescription package, an execution schedule and resource allocation plan are generated to rationally allocate limited resources, avoid conflicts and waiting, shorten downtime under the premise of safety and compliance, reduce overall operation and maintenance costs, improve risk reduction efficiency, generate a standardized schedule, and directly enter the execution stage to ensure that the diagnostic results are truly implemented.

[0119] The execution schedule and resource allocation plan are transformed into work orders and sent to the on-site operation and maintenance system. During the execution process, execution feedback is collected in real time to complete the feedback and update of knowledge and experience, thereby forming a self-learning and evolvable closed loop.

[0120] 1. Work order issuance The optimized schedule and recipe package are pushed to the company's existing CMMS (Computerized Operation and Maintenance Management System) or EAM (Enterprise Asset Management System) in the form of structured work orders.

[0121] The work order includes: the sequence and dependencies of maintenance steps; a list of required spare parts and tools; the qualification requirements for the work team and personnel; the safety permit chain (power outage permit, hot work permit, confined space permit, etc.); and the planned start and end times and resource usage of the task.

[0122] The work order format supports standardized interfaces such as JSON / XML, ensuring compatibility with operation and maintenance management platforms of different plants and stations.

[0123] 2. Execution process monitoring During on-site execution, the system monitors the execution status through mobile terminals, industrial tablets, and smart work cards: the start and completion time of each step; the confirmation of the operator's identity and qualifications; the electronic signing status of safety tickets; and the dynamic recovery status of equipment operating parameters (such as a decrease in temperature rise, a reduction in vibration amplitude, and the return of current distortion to normal).

[0124] Critical steps (such as power outage isolation, residual energy release, and hot work operations) must be confirmed through methods such as scanning codes and electronic signatures to ensure safety and traceability.

[0125] 3. Perform feedback collection After a work order is completed, the platform automatically collects and records the following feedback: the deviation between actual and planned working hours; key performance indicators such as work order completion rate and rework rate; and the degree of equipment performance improvement after maintenance (e.g., risk reduction). Compare with predicted values); unexpected situations or temporary adjustments on site (such as additional materials or replacement of component models).

[0126] 4. Knowledge reinjection mechanism Feedback is fed back into the system through three dimensions: Causal knowledge graph updates: adjusting root cause-symptom edge weights based on verification results to improve the accuracy of the diagnostic process; Standard operating procedure (SOP) block revisions: correcting time estimates, resource allocation, and safety ticket requirements to make subsequent maintenance plans more closely aligned with actual site conditions; and risk model corrections: using the post-maintenance equipment performance to calibrate the risk prediction function, enhancing the scientific rigor of future scheduling. Simultaneously, the system automatically records the twin model version, diagnostic evidence tree, and execution feedback to ensure traceability throughout the entire process.

[0127] In this way, a complete closed loop of diagnosis, recommendation, optimization, execution, and feedback is achieved, ensuring that the diagnostic results are truly implemented into actionable maintenance actions. On-site feedback enhances the system's adaptability to different sites and operating conditions, and a self-evolutionary mechanism of "data, knowledge, simulation, decision-making, and execution" is established, enabling the platform to continuously improve diagnostic accuracy and recommendation rationality during long-term operation.

[0128] like Figure 2 As shown, the maintenance method based on triple consistency assessment includes multi-source data access and alignment, anomaly fingerprint generation, causal reasoning and candidate root cause generation, digital twin verification, triple consistency assessment, maintenance recipe package splicing, resource and scheduling optimization, work order execution and knowledge feedback, forming a closed-loop structure that connects the front and back ends, enabling full-process diagnosis and management of equipment failures in the complex operating environment of new energy power plants.

[0129] Compared with traditional prevention methods, this invention improves the interpretability and accuracy of anomaly diagnosis by introducing causal knowledge and mechanism models; it reduces the risk of misjudgment and omission by performing secondary verification of candidate root causes through simulation verification; it achieves high-confidence fault tracing results through the fusion evaluation of multi-source evidence, ensuring the reliability of diagnostic conclusions; and it ensures the scientific nature and feasibility of maintenance plans through standardized maintenance formulas and schedule optimization.

[0130] Furthermore, this invention enables the reinjection of operation and maintenance knowledge and the dynamic updating of the model through a work order execution and feedback mechanism, giving the system the ability to learn and continuously optimize. This overcomes the shortcomings of existing maintenance methods, such as reliance on manual experience, disconnect between diagnosis and maintenance, and difficulty in utilizing feedback information. It realizes intelligent management of new energy power stations throughout the entire process of fault diagnosis, maintenance recommendation, and execution feedback, and has high engineering application value.

[0131] This embodiment also provides a maintenance system based on triple consistency evaluation, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that performs a predetermined function. Although the systems described in the following embodiments are preferably implemented in software, hardware implementations, or a combination of software and hardware, are also possible and contemplated.

[0132] This embodiment provides a maintenance system based on triple consistency assessment, such as... Figure 3 As shown, it includes: The acquisition module 301 is used to acquire multi-source heterogeneous data from new energy power plants and compress the multi-source heterogeneous data into anomaly fingerprint vectors.

[0133] Module 302 is used to combine the abnormal fingerprint vector with device mechanism knowledge to obtain candidate root causes.

[0134] The construction module 303 is used to calculate the explanatory power and temporal conformity of candidate root causes, and constructs a set of candidate root causes that meet the preset conditions for both explanatory power and temporal conformity.

[0135] The simulation module 304 is used to perform simulation verification on the candidate root causes in the candidate root cause set and obtain simulation consistency results.

[0136] The fusion module 305 is used to fuse the results of explanatory power, timing compliance and simulation consistency, construct a scoring function, screen and sort candidate root causes, and obtain a set of high-confidence root causes.

[0137] The distribution module 306 is used to convert the high-confidence root cause set into a maintenance step sequence, generate an execution schedule and resource allocation plan, convert them into work orders, and distribute the work orders to the on-site operation and maintenance system.

[0138] In some optional implementations, the acquisition module 301 includes: The signal preprocessing unit is used to preprocess multi-source heterogeneous data.

[0139] The feature extraction unit is used to extract single-mode features from multi-source heterogeneous data after signal preprocessing, and obtain vibration features, electrical features, thermal imaging features and acoustic features.

[0140] The multimodal fusion unit is used to stitch together vibration features, electrical features, thermal imaging features and acoustic features into an initial feature vector. The initial feature vector is then fused using a multimodal encoder to generate an anomaly fingerprint vector.

[0141] In some alternative implementations, building module 303 includes: Building units are used to construct knowledge graphs.

[0142] The decoding unit is used to decode the abnormal fingerprint vector into a set of symptoms using a mapping function.

[0143] The quantization unit is used to quantify the explanatory power of each candidate root cause on the symptom set using the conditional probability in the knowledge graph, and to obtain the explanatory power of the candidate root cause.

[0144] The verification unit is used to verify whether each candidate root cause and its corresponding symptom satisfy the time window constraint of the knowledge graph, and to obtain the temporal conformity of the candidate root cause.

[0145] In some alternative implementations, simulation module 304 includes: Assembly units are used to assemble electromechanical coupling models, electromagnetic transient models, and thermodynamic models into a simulation platform using a unified interface.

[0146] The operating condition injection unit is used to inject operating conditions into the simulation platform to obtain simulation fingerprints.

[0147] The consistency calculation unit is used to perform consistency calculations between simulated fingerprints and measured abnormal fingerprints to obtain consistency indices.

[0148] In some alternative implementations, the fusion module 305 includes: The filtering unit is used to remove root causes that have an interpretability less than a preset threshold, a timing compliance less than a preset threshold, or a simulation consistency result less than a preset threshold.

[0149] The sorting unit is used to sort root causes based on the scoring results, and uses bundle search to retain root causes that meet the conditions.

[0150] The maintenance system based on triple consistency evaluation provided in this invention can execute the maintenance method based on triple consistency evaluation provided in any embodiment of this invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0151] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0152] The following is a detailed reference. Figure 4 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0153] Typically, the following systems can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0154] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a memory 408, or installed from a ROM 402. When the computer program is executed by the processor 401, it performs the functions defined in the maintenance method based on triple consistency evaluation according to embodiments of the present invention.

[0155] Figure 4The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0156] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the troubleshooting method based on triple consistency evaluation shown in the above embodiments is implemented.

[0157] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0158] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and all such modifications and variations fall within the scope defined by the appended invention.

Claims

1. A maintenance method based on triple consistency assessment, characterized in that, The method includes: Acquire multi-source heterogeneous data from new energy power stations and compress the multi-source heterogeneous data into anomaly fingerprint vectors; By combining the abnormal fingerprint vector with knowledge of device mechanisms, candidate root causes are obtained; Calculate the explanatory power and temporal conformity of candidate root causes, and construct a candidate root cause set for candidate root causes whose explanatory power and temporal conformity both meet the preset conditions; The candidate root causes in the candidate root cause set are verified by simulation to obtain simulation consistency results; The explanatory power, temporal consistency, and simulation consistency results are integrated to construct a scoring function, which is used to screen and sort candidate root causes to obtain a set of high-confidence root causes. The high-confidence root cause set is converted into a maintenance step sequence, an execution schedule and resource allocation plan are generated, and then converted into a work order, which is then sent to the on-site operation and maintenance system.

2. The method according to claim 1, characterized in that, The step of compressing the multi-source heterogeneous data into anomaly fingerprint vectors includes: The multi-source heterogeneous data is subjected to signal preprocessing; Single-mode feature extraction is performed on the multi-source heterogeneous data after signal preprocessing to obtain vibration features, electrical features, thermal imaging features and acoustic features; The vibration features, electrical features, thermal imaging features, and acoustic features are concatenated into an initial feature vector. The initial feature vector is then fused using a multimodal encoder to generate an abnormal fingerprint vector.

3. The method according to claim 1, characterized in that, The calculation of the explanatory power and temporal consistency of candidate root causes includes: Constructing knowledge graphs; The abnormal fingerprint vector is decoded into a set of symptoms using a mapping function; The explanatory power of each candidate root cause to the symptom set is quantified by using the conditional probability in the knowledge graph, thus obtaining the explanatory power of the candidate root cause. Verify whether each candidate root cause and its corresponding symptom satisfy the time window constraint of the knowledge graph to obtain the temporal conformity of the candidate root cause.

4. The method according to claim 1, characterized in that, The step of performing simulation verification on candidate root causes in the candidate root cause set to obtain simulation consistency results includes: A unified interface is used to assemble electromechanical coupling models, electromagnetic transient models, and thermodynamic models into a simulation platform; The simulation platform is subjected to working condition injection to obtain a simulation fingerprint; The consistency index is obtained by performing consistency calculations between the simulated fingerprint and the measured abnormal fingerprint.

5. The method according to claim 4, characterized in that, The consistency index is calculated using the following formula: in, As a consistency indicator, For simulated fingerprints, To test abnormal fingerprints, This is the set of candidate root causes.

6. The method according to claim 4, characterized in that, The scoring function is as follows: in, For rating, , , These are the weights for interpretability, temporal compliance, and simulation consistency results, respectively. , For the sake of explanation, For timing compliance, This is a consistency indicator for digital twins.

7. The method according to claim 1, characterized in that, The process of screening and ranking candidate root causes includes: Remove root causes where the explanatory power is less than the preset threshold, the timing compliance is less than the preset threshold, or the simulation consistency result is less than the preset threshold. Root causes are ranked based on the scoring results, and a bundle search is used to retain root causes that meet the conditions.

8. A maintenance system based on triple consistency evaluation, characterized in that, The system includes: The acquisition module is used to acquire multi-source heterogeneous data from new energy power stations and compress the multi-source heterogeneous data into anomaly fingerprint vectors. The module combines the abnormal fingerprint vector with device mechanism knowledge to obtain candidate root causes; The construction module is used to calculate the explanatory power and temporal conformity of candidate root causes, and constructs a set of candidate root causes that meet the preset conditions in terms of both explanatory power and temporal conformity. The simulation module is used to perform simulation verification on the candidate root causes in the candidate root cause set and obtain simulation consistency results; The fusion module is used to fuse the explanatory power, temporal compliance and simulation consistency results, construct a scoring function, screen and sort the candidate root causes, and obtain a set of high-confidence root causes. The distribution module is used to convert the high-confidence root cause set into a maintenance step sequence, generate an execution schedule and resource allocation plan, convert it into a work order, and distribute the work order to the on-site operation and maintenance system.

9. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory stores computer instructions, and the processor executes the computer instructions to perform the maintenance method based on triple consistency evaluation as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the maintenance method based on triple consistency assessment as described in any one of claims 1 to 7.

Images