Offshore wind power junction box data acquisition unit operation state monitoring method and device

Abstract

This application discloses a method and device for monitoring the operational status of an offshore wind power combiner box data acquisition unit, relating to the field of power data acquisition equipment status monitoring technology. The method includes: constructing a corresponding prior causal knowledge model based on the internal structure and operating environment of the offshore wind power combiner box data acquisition unit; instantiating the prior causal knowledge model into a dynamic causal graph; sampling key nodes in the offshore wind power combiner box data acquisition unit at a first frequency; obtaining high-frequency sampling data when anomalies exist at key nodes; and performing real-time causal inference on the high-frequency sampling data based on the dynamic causal graph to obtain a diagnostic result of the operational status of the offshore wind power combiner box data acquisition unit. This addresses the problems of insufficient monitoring of the data acquisition unit itself, strong fault concealment, and difficulty in causal localization.

Description

Technical Field

[0001] This application relates to the field of power data acquisition equipment status monitoring technology, and in particular to a method and equipment for monitoring the operating status of a data acquisition unit for offshore wind power combiner boxes. Background Technology

[0002] Offshore wind power is an important direction for the development of renewable energy. Offshore wind farms typically consist of dozens to hundreds of wind turbine generators. Each turbine is equipped with a combiner box (also known as a tower base cabinet) to collect the electrical output of the turbine and perform data acquisition. The combiner box data acquisition unit (also known as an RTU, DTU, or intelligent acquisition terminal) is responsible for collecting parameters such as current, voltage, and temperature from the combiner box and uploading them to a remote monitoring platform. It is a key link connecting field equipment and the operation and maintenance system.

[0003] However, monitoring the operational status of the data acquisition unit in the combiner box of an offshore wind farm faces the following technical challenges: First, there is the issue of blind spots in monitoring. Existing monitoring solutions only focus on the electrical parameters of the combiner box itself, neglecting the health status of the acquisition unit itself. Maintenance personnel cannot detect the degradation trend of the acquisition unit in advance.

[0004] Second, the faults are highly concealed. Problems such as oxidation of the communication port connectors of the acquisition unit, drying out of the electrolytic capacitors in the power module, and drift of the reference source do not cause complete data interruption in the early stages, but only manifest as occasional packet loss or slight measurement deviation. Traditional threshold alarm methods are difficult to detect these progressive degradation characteristics, and they are often only discovered when the data is completely interrupted, at which point the root cause of the fault is difficult to trace.

[0005] Third, cause-and-effect localization is difficult. When abnormal data is detected in the combiner box, it is impossible to distinguish whether the problem lies with a sensor malfunction in the combiner box itself or with a fault in the data acquisition unit. The handling methods for different causes of failure vary greatly, and the lack of accurate root cause localization capabilities leads to low operation and maintenance efficiency. Summary of the Invention

[0006] This application provides a method and device for monitoring the operational status of a data acquisition unit in an offshore wind power combiner box, in order to solve the following technical problems: how to solve the problems of lack of self-monitoring of the combiner box data acquisition unit, strong fault concealment, and difficulty in causal location in the existing technology in offshore wind power scenarios.

[0007] In a first aspect, embodiments of this application provide a method for monitoring the operational status of an offshore wind power combiner box data acquisition unit. The method includes: constructing a priori causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit based on its internal structure and operating environment; instantiating the priori causal knowledge model into a dynamic causal graph corresponding to the offshore wind power combiner box data acquisition unit; sampling key nodes in the offshore wind power combiner box data acquisition unit at a first frequency to obtain a baseline operational status of the offshore wind power combiner box data acquisition unit, wherein the key nodes are multi-dimensional physical state parameters defined in the dynamic causal graph to characterize the operational status of the offshore wind power combiner box data acquisition unit itself; sampling target nodes at a second frequency when the key nodes are abnormal to obtain high-frequency sampling data, wherein the second frequency is higher than the first frequency, and the target node is a node determined based on the dynamic causal graph and associated with the abnormal key node; and performing real-time causal reasoning on the high-frequency sampling data based on the dynamic causal graph to obtain a diagnostic result of the operational status of the offshore wind power combiner box data acquisition unit.

[0008] Secondly, embodiments of this application also provide an operational status monitoring device for an offshore wind power combiner box data acquisition unit. The device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the operational status monitoring method for an offshore wind power combiner box data acquisition unit as described in the first aspect above.

[0009] Thirdly, embodiments of this application also provide a computer storage medium storing computer-executable instructions, which, when executed, implement a method for monitoring the operational status of an offshore wind power combiner box data acquisition unit as described in the first aspect above.

[0010] The method for monitoring the operational status of a data acquisition unit for an offshore wind power combiner box provided in this application has the following beneficial effects: In this embodiment, a priori causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit can be constructed based on its internal structure and operating environment. This priori causal knowledge model is then instantiated into a dynamic causal graph and a digital twin causal model. Key nodes within the offshore wind power combiner box data acquisition unit are sampled at a first frequency, enabling monitoring of its operational status. Subsequently, in cases of anomalies at key nodes, associated target nodes are sampled at a second frequency, yielding high-frequency sampling data. This allows for the sensitive detection of early, progressive, and cross-domain coupled hidden faults. Following this, real-time causal inference is performed on the high-frequency sampling data based on the dynamic causal graph to obtain a diagnostic result of the offshore wind power combiner box data acquisition unit's operational status. This enables precise root cause localization and differentiation of fault origins. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a method for monitoring the operational status of a data acquisition unit for offshore wind power combiner boxes, as provided in this application embodiment; Figure 2 A flowchart illustrating a method for real-time causal inference of high-frequency sampled data, provided in an embodiment of this application; Figure 3 A flowchart illustrating a method for obtaining deep diagnostic results using a digital twin causal model, provided in an embodiment of this application; Figure 4 This is a schematic diagram of the internal structure of a data acquisition unit for monitoring the operation status of an offshore wind power combiner box, provided in an embodiment of this application. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0014] Figure 1 This document presents a flowchart illustrating a method for monitoring the operational status of a data acquisition unit for offshore wind power combiner boxes, as provided in an embodiment of this application. Figure 1 As shown in the figure, the method for monitoring the operating status of a data acquisition unit for an offshore wind power combiner box provided in this application embodiment specifically includes the following steps: Step 101: Based on the internal structure and operating environment of the offshore wind power combiner box data acquisition unit, construct the prior causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit.

[0015] In practical applications, the prior causal knowledge model is a model that includes all possible causal relationships of the offshore wind turbine combiner box data acquisition unit, but whose parameters are still theoretical or nominal values. This prior causal knowledge model is built upon the internal structure and operating environment of the offshore wind turbine combiner box data acquisition unit; that is, it originates from structural design, physical laws, and expert knowledge, and does not rely on a large amount of historical fault data, thus solving the problem of limited fault samples in offshore wind turbine combiner box data acquisition units. Furthermore, this prior causal knowledge model can clearly distinguish the cause and effect of faults within the data acquisition unit and describe the intensity and time delay of the impact, providing possibilities for subsequent interpretative diagnosis and fault tracing of the offshore wind turbine combiner box data acquisition unit.

[0016] Step 102: Instantiate the prior causal knowledge model into a dynamic causal graph corresponding to the offshore wind power combiner box data acquisition unit.

[0017] In practical applications, the prior causal knowledge model is a general knowledge model. However, there are differences between offshore wind turbine combiner box data acquisition units. For example, manufacturing tolerances, component batches, installation stress, and microenvironment can all cause individual differences. Through instantiation, the above prior causal knowledge model can be calibrated using data from a specific data acquisition unit to meet the specific requirements of that offshore wind turbine combiner box data acquisition unit. Among them, the dynamic causal graph is a simplified version of the prior causal knowledge model that can perform fast state reasoning in real time and with low power consumption. It can generally be deployed in the local MCU of the offshore wind turbine combiner box data acquisition unit.

[0018] Step 103: Sample the key nodes in the offshore wind power combiner box data acquisition unit at a first frequency to obtain the baseline operating status of the offshore wind power combiner box data acquisition unit.

[0019] The key nodes are multi-dimensional physical state parameters defined in the dynamic causal graph, used to characterize the operating state of the offshore wind power combiner box data acquisition unit itself.

[0020] Under normal circumstances, to conserve resources, it is not necessary to sample all nodes of the offshore wind turbine combiner box data acquisition unit; sampling only key nodes is sufficient. These key nodes are multi-dimensional physical state parameters defined in the dynamic cause-effect graph, which characterize the operational status of the offshore wind turbine combiner box data acquisition unit itself. They can cover the physical domain (electrical, thermal, mechanical, and chemical) and indicate whether the physical domain is functioning normally. The number of these key monitoring nodes is relatively small (generally only 5%-20% of the total nodes), ensuring ultra-low power consumption in normal operation. This allows for resource conservation while acquiring the baseline operational status of the offshore wind turbine combiner box data acquisition unit. In practical applications, the initial sampling frequency can be once every 5-30 minutes, without specific limitations. With all key nodes operating normally, the operational status of the offshore wind turbine combiner box data acquisition unit can be temporarily determined to be normal.

[0021] Step 104: In the event of an anomaly at the critical node, sample the target node at a second frequency to obtain high-frequency sampling data.

[0022] Wherein, the second frequency is higher than the first frequency, and the target node is a node that is associated with the key node that has an anomaly, as determined based on the dynamic causal graph.

[0023] In practical applications, if the values ​​corresponding to a critical node meet preset conditions, it can be determined that the critical node is abnormal. At this point, a dynamic cause-effect graph can be used to identify nodes causally related to the critical node, nodes that may cause the critical node's abnormality, or nodes that may be affected by the critical node. For example, suppose an offshore wind turbine combiner box data acquisition unit has 50 monitorable points, among which 5 critical nodes are preset for routine monitoring: K1 (power input current), K2 (main processor temperature), K3 (chassis vibration), K4 (internal humidity), and K5 (communication error rate). During routine monitoring, a critical node K2 (main processor temperature) is found to have an abnormally high temperature, meeting the preset conditions. In this case, by querying the dynamic cause-effect graph, the nodes directly related to K2 can be found: A1 (CPU utilization, not monitored normally, non-critical node), A2 (cooling fan PWM duty cycle, not monitored normally, non-critical node), K1 (power input current, a critical node, already under routine monitoring), as well as nodes R1 (processor throttling flag) and R2 (adjacent memory chip temperature). Thus, the target nodes can be identified as K2, A1, A2, K1, R1, and R2. Under normal circumstances, to save power, only a few key points are monitored, resulting in blind spots. Once an anomaly is detected in a key node, the monitoring range can be immediately and automatically extended to the relevant blind spot nodes, gaining a global view. Furthermore, it allows for extreme resource optimization. While monitoring only key nodes under normal conditions and monitoring key nodes during anomalies, although the instantaneous power consumption is high, the duration is short, and the overall power consumption is lower than that of long-term monitoring of all nodes. Moreover, by sampling the target nodes at a high frequency (a second frequency higher than the first frequency, for example, once per minute, without specific limitations), high-frequency sampling data can be obtained, providing more complete status information (high-frequency sampling data) of the offshore wind turbine combiner box data acquisition unit, facilitating subsequent analysis of the operational status of the offshore wind turbine combiner box data acquisition unit.

[0024] Step 105: Perform real-time causal reasoning on the high-frequency sampled data based on the dynamic causal graph to obtain the diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit.

[0025] In practical applications, real-time causal reasoning analysis can be performed locally on the offshore wind turbine combiner box data acquisition unit based on a dynamic causal graph. This allows for anomaly confirmation and preliminary root cause tracing, yielding a diagnostic result of the offshore wind turbine combiner box data acquisition unit's operational status. For example, the local dynamic causal graph can be invoked to quickly analyze the high-frequency sampled data, identifying multiple candidate cause nodes that may have caused the anomaly. Then, within the dynamic causal graph, the posterior probability of each candidate cause node can be calculated. For instance, calculating "power supply problem probability: 65%, sensor failure probability: 30%, other probability: 5%" yields the final diagnostic result of the offshore wind turbine combiner box data acquisition unit's operational status. This result may include the 1-3 most likely root cause hypotheses and their confidence levels. In practical applications, the diagnostic result can also include the urgency of the anomaly.

[0026] In this embodiment, a priori causal knowledge model corresponding to the offshore wind turbine combiner box data acquisition unit can be constructed based on its internal structure and operating environment. This priori causal knowledge model is then instantiated into a dynamic causal graph and a digital twin causal model. Key nodes in the offshore wind turbine combiner box data acquisition unit are sampled at a first frequency, enabling monitoring of its operational status under normal conditions. Subsequently, in cases of anomalies at key nodes, associated target nodes are sampled at a second frequency to obtain high-frequency sampling data. This allows for the sensitive detection of early, progressive, and cross-domain coupled hidden faults. Following this, real-time causal reasoning is performed on the high-frequency sampling data based on the dynamic causal graph to obtain a diagnostic result of the operational status. This enables precise root cause localization and differentiation of fault origins.

[0027] In one possible implementation, the prior causal knowledge model corresponding to the offshore wind turbine combiner box data acquisition unit is constructed based on the internal structure and operating environment of the unit, including: The circuit diagram and / or structural diagram of the offshore wind power combiner box data acquisition unit are extracted to obtain the components corresponding to the offshore wind power combiner box data acquisition unit and the circuit or physical connection relationship between the components. Using the aforementioned components as vertices of the structural layer and the aforementioned circuit or physical connection relationships as directed edges of the structural layer, a structural layer diagram corresponding to the offshore wind power combiner box data acquisition unit is constructed. Based on the knowledge of the operating environment and equipment failure modes of the offshore wind power combiner box data acquisition unit, an environmental failure mapping table corresponding to the offshore wind power combiner box data acquisition unit is established. The operating environment includes specific environmental spectrum data of offshore wind power. The environmental failure mapping table includes the environmental factors, failure modes and observable symptoms caused by the failure modes corresponding to the offshore wind power combiner box data acquisition unit. Based on the environmental fault mapping table and the physical and / or chemical changes involved in the operation of the offshore wind power combiner box data acquisition unit, a causal relationship and influence function library corresponding to the offshore wind power combiner box data acquisition unit are constructed. The causal relationship includes the inducing relationship between the environmental factors and the fault mode and the causal relationship between the fault mode and the observable symptoms. The influence function library includes influence functions, which are calculation formulas that match the causal relationship. Using the physical quantities, component states, environmental factors, and fault modes corresponding to the offshore wind power combiner box data acquisition unit as nodes, and the causal relationships as directed edges, a priori causal knowledge model framework corresponding to the offshore wind power combiner box data acquisition unit is constructed. The physical quantities are used to characterize the measurable parameters within the offshore wind power combiner box data acquisition unit, and the component states are used to characterize the module health status evaluated based on the physical quantity nodes. The physical quantities include: power supply voltage, current, ripple, temperature, vibration, humidity, and communication port impedance. The nodes are configured to belong to the components, and the circuit or physical connection relationships in the structural layer diagram are configured as topological constraints for the directed edges in the priori causal knowledge model framework. Traverse each directed edge in the prior causal knowledge model framework, match and associate the influence functions corresponding to the directed edges from the influence function library, assign initial values ​​to the directed edges, and generate the prior causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit.

[0028] In practical applications, when constructing a priori causal knowledge models, it is necessary to first gather original knowledge from different dimensions and transform it into structured information. First, the internal structure of the offshore wind turbine combiner box data acquisition unit can be analyzed to extract its physical components (such as power modules, main control chips, RS-485 buses, IP65 sealing rings) and the physical connection relationships between components. For example, the circuit diagram, PCB layout diagram, mechanical structure 3D model, and bill of materials of the data acquisition unit can be processed to analyze the circuit, identify key signal chains and power supply chains, and clarify the electrical connection relationships between components (such as "the output terminal of the U1 power chip is connected to the C101 filter capacitor"). Then, the mechanical structure can be analyzed to identify the thermal path (heating element → heat conduction interface → heat sink / shell) and the force transmission path (the path of vibration transmitted from the installation point to the internal PCB). Finally, all sealing interfaces (such as the box joint surface) and their material properties can be identified. In this way, a structured list of component connection relationships can be obtained, recording the circuit or physical connection relationships between various components of the offshore wind turbine combiner box data acquisition unit.

[0029] Using the aforementioned components as vertices, directed edges can be constructed between nodes where there is direct physical energy / signal transmission, based on the aforementioned circuit or physical connection relationships. This describes the flow and conversion relationships of energy, signals, or matter under healthy and normal operating conditions, thereby constructing a structural layer diagram.

[0030] It should be noted that the structural layer graph exists only as a constraint during the model building phase. In the final generated prior causal knowledge model, the structural layer information has been internalized as physical reachability constraints of causal edges and boundary conditions of influence functions, and will not increase the computational burden of online inference.

[0031] Secondly, domain-specific prior knowledge can be analyzed to obtain knowledge of the operating environment and equipment failure modes of the offshore wind turbine combiner box data acquisition unit. The operating environment includes specific environmental spectrum data for offshore wind power, such as the typical salt spray concentration distribution, temperature-humidity cycle curves, and vibration spectrum data caused by wind turbine operation and wave loads in the sea area where the offshore wind turbine combiner box data acquisition unit is located. The equipment failure mode knowledge can include failure mode and impact analysis reports for this model of offshore wind turbine combiner box data acquisition unit (e.g., the official manual for this model of data acquisition unit, extracting all possible failure modes, root causes, and final effects), high-frequency failure cases from historical maintenance records, etc. This knowledge is integrated into an environmental failure mapping table. An example of this table is as follows: Environmental factor: continuous high temperature (>T℃), Failure mode: accelerated drying of electrolyte in electrolytic capacitor C1, Observable symptoms: increased effective value of power supply output ripple voltage, slower voltage regulation response. The observable symptoms are the symptoms of the offshore wind turbine combiner box data acquisition unit obtained based on data collected by sensors deployed inside or around the unit.

[0032] In practical applications, based on the aforementioned environmental fault mapping table, causal relationships can be formally defined at the model level, providing semantic rules for the "edges" in subsequent model construction. These causal relationships can be either an inducing relationship of "environmental factors → fault modes," specifying which environmental conditions (e.g., salt spray, high temperature) are potential causes of specific failures (e.g., corrosion, aging). For example, defining "the combined effect of salt spray concentration and high humidity is a necessary condition for inducing electrochemical corrosion of connector J1"; or a causative relationship of "fault modes → observable symptoms," specifying which measurable physical parameters (e.g., increased impedance, increased ripple) are directly caused by each specific failure. For example, defining "an increase in the ESR of capacitor C105 directly leads to an increase in the 5V power supply ripple voltage."

[0033] Subsequently, based on the physicochemical laws involved in the operation of the offshore wind turbine combiner box data acquisition unit, parameterized multiphysics coupling and degradation influence functions corresponding to the induced and causative relationships can be defined, thereby forming an influence function template library. Specifically, environmental induced relationships are generally material and component degradation functions (such as corrosion, fatigue, and aging kinetic equations). The material and component degradation influence function templates are configured to: receive environmental state input, calculate the time-varying development parameters of the failure mode based on the physicochemical degradation kinetic equations; thus, the development of failures can be predicted. For example, the electrochemical corrosion function based on the Arrhenius equation and electrochemical kinetics can be used to calculate the growth rate of connector contact resistance based on salt spray concentration and temperature; the thermo-mechanical fatigue function based on the Coffin-Manson equation can be used to calculate the cumulative fatigue damage of electronic component solder joints caused by temperature cycling.

[0034] Fault-cause relationships are typically established using fault-symptom transfer functions (such as circuit parameter degradation functions or performance degradation models), which map internal fault states to external measurable signals. Fault-symptom transfer function templates are configured to receive fault state input and, based on circuit, thermodynamic, or signal transmission principles, calculate predicted changes in observable physical quantity nodes. For example, a ripple transfer function based on switching power supply topology principles is used to calculate the increment of power supply ripple voltage based on the increase in capacitor ESR; an impedance-bit error rate transfer function based on signal integrity is used to calculate the increase in communication bit error rate based on the increase in communication port impedance.

[0035] Ultimately, the structured knowledge described above can be integrated into a unified graph model framework. Based on the above output, four types of nodes can be defined: Physical quantity node: A model representation of measurable parameters inside the data acquisition unit of the offshore wind turbine combiner box, such as power supply voltage, current, ripple, temperature, vibration, humidity and communication port impedance.

[0036] The component status node represents the module's health status based on a comprehensive evaluation of physical quantity nodes. For example, it may include the health status of the power module and the health status of the communication link. This node is generally calculated based on physical quantity nodes.

[0037] Fault mode nodes: characterize specific failure mechanisms. For example, Fault_Corrosion_J1 (corrosion of connector J1) and Fault_ESRIncrease_C1 (increased ESR of capacitor C1) are directly derived from the above environmental fault mapping table; they can be accurate down to specific components.

[0038] Environmental factor nodes: These are external inputs to the model, representing uncontrollable external stresses. Examples include salt spray concentration and ambient temperature. They are derived from the environmental fault mapping table mentioned above.

[0039] Each node is associated with one or more specific components. For example, the power supply voltage in the physical quantity node can be associated with the power module; the power module health status in the component status node can be associated with the power module and the filter capacitor; the salt spray concentration can be associated with the chassis surface and the heat dissipation holes; and the fault mode node can be associated with the power module. This attribution relationship is used to constrain the physical realizability of causal relationships, but it does not exist as a directed edge in the prior causal knowledge model framework. Essentially, it maps the symbols (nodes) in the prior causal knowledge model one-to-one with the physical objects (the actual components of the offshore wind power combiner box data acquisition unit), proving that the model is an engineering diagnostic tool "based on physical entities."

[0040] In the above embodiments, directed edges can be established based on causal relationships. Edges are drawn between nodes with causal relationships according to these relationships. The direction of the edge represents the "cause" pointing to the "effect." This can describe how harsh external environments lead to the accumulation of internal damage, ultimately causing specific faults, and how internal faults manifest as measurable external anomalies. All causal edge establishment is based on physicochemical laws (such as the principle of electrochemical corrosion) and environmental fault mapping tables (such as salt spray-induced increases in contact resistance), rather than simple statistical correlations. Constraints on physical connection edges ensure the engineering interpretability of causal relationships. For any directed causal edge from node A to node B in the above prior causal knowledge model framework, there must exist a physical connection path in the structural layer graph from the component belonging to node A to the component belonging to node B.

[0041] In this way, we can obtain an initial prior causal knowledge model framework that includes a set of nodes V and a set of directed edges E.

[0042] It should be noted that the key nodes for routine system monitoring are: physical quantity nodes (e.g., power input ripple voltage, internal humidity of the chassis) and component status nodes (e.g., power module health).

[0043] After obtaining the aforementioned prior causal knowledge model framework, it can be initialized. First, for each directed edge, its corresponding influence function is associated from the aforementioned influence function library. Each directed edge (regardless of whether its semantics are "environmentally induced" or "fault-induced") must be associated with a quantitative influence function; otherwise, the entire model will be unable to perform any numerical calculations or inferences. For example, each directed edge e(A→B) in the aforementioned prior causal knowledge model framework can be traversed, and the most suitable influence function can be matched from the aforementioned influence function library based on the types of its source node A and target node B and the semantics of the edge. For example, for the edge [current (load)] → [power consumption (chip)], the electro-thermal conversion function is matched, in the form P = I. 2 * R. For the edge [Ambient salt spray concentration] → [Fault (connector corrosion)], match the electrochemical corrosion kinetic function, which may be in the form of corrosion depth = f(salt spray concentration, time, temperature); for the edge [Fault (increased capacitor ESR)] → [Symptom (power supply ripple)], match the transfer function or algebraic relationship based on circuit theory, such as ΔV ripple = k * ΔESR.

[0044] Subsequently, initial values ​​can be assigned to the aforementioned directed edges, such as initial values ​​for the function parameters, edge weights, and time delays of all influencing functions. The function parameters are derived from design nominal values ​​(e.g., resistance, thermal resistance) and simulation calculation results. Initial values ​​are assigned to the function parameters of the "environment → fault" edge, taking into account the environmental magnitude, material properties, and historical data in the environmental fault mapping table. For example, the reaction rate constant of the salt spray corrosion function can be initialized.

[0045] Directed edges can also be assigned edge weights, which can be initially set to 1; at the same time, degenerate edges ("environmental factors → failure mode" edges) can be assigned corresponding causal time delays, which can be set based on the time constant of the physical process. For example, the time delay of temperature conduction may be a few minutes, while the time delay of electrochemical corrosion may be tens to hundreds of hours.

[0046] Simultaneously, a prior probability P(Fault) can be assigned to each fault mode node. This value can be derived from the historical fault statistics frequency associated with the aforementioned environmental fault mapping table. For example, if historical data shows that the corrosion failure rate of this type of connector is approximately 2%, then P(Fault) can be set as follows: Corrosion = 0.02.

[0047] In practical applications, the aforementioned prior causal knowledge model is parameterized and learnable. Its structure remains relatively stable after initialization, but the weights, time delays, and associated function parameters of its directed edges can be dynamically updated and optimized based on the actual operating data of the offshore wind power combiner box data acquisition unit through online learning or batch learning.

[0048] In practical applications, a complete and ideal prior causal knowledge model has been constructed, defining all corresponding nodes of the offshore wind turbine combiner box data acquisition unit, although it is practically impossible to monitor all of them. Subsequently, key nodes for monitoring can be selected based on engineering constraints (cost, space). Examples include internal humidity (direct evidence of seal failure), power supply voltage (power supply health), and salt spray concentration (if cost allows, deploy corrosion sensors; otherwise, abandon the plan). Then, the selected nodes can be mapped to specific sensor hardware, and their installation locations determined. For example, for the internal humidity node, a series of temperature and humidity sensors can be deployed, installed in the airflow area in the middle of the chassis, avoiding direct contact with heat-generating components. For the ripple coefficient node, a high-frequency ADC sampling circuit can be deployed on the PCB trace at the output end of the power module. For the communication impedance node, an impedance measurement circuit can be deployed at the RS-485 communication port (indirectly assessing contact corrosion by measuring the differential line resistance).

[0049] In one possible implementation, instantiating the prior causal knowledge model into a dynamic causal graph corresponding to the offshore wind power combiner box data acquisition unit includes: Traverse all nodes of the prior causal knowledge model, remove unobservable nodes of the offshore wind power combiner box data acquisition unit that do not have corresponding physical sensors deployed, and remove all causal edges connected to the unobservable nodes to generate an observable causal subgraph. The influence function associated with each edge in the observable causal subgraph is simplified; The factory calibration data and health baseline of the offshore wind power combiner box data acquisition unit are injected into the simplified observable causal subgraph to generate a dynamic causal graph for the offshore wind power combiner box data acquisition unit.

[0050] In practical applications, all nodes in the prior causal knowledge model can be traversed. For each node, it is checked whether the offshore wind turbine combiner box data acquisition unit is equipped with a corresponding physical sensor to directly or indirectly measure it. For example, nodes with corresponding sensors, such as power supply voltage and chassis temperature, are retained. Nodes without corresponding sensors, such as chip junction temperature, core loss, and microcrack length in the sealing ring, which are purely theoretical or state-related, need to be removed. At the same time, all causal edges of source or target nodes that have been pruned need to be removed because the states involved in these edges are unobservable and cannot be verified and reasoned about in real time at the edges. In this way, an observable causal subgraph containing only observable nodes can be obtained, and this observable causal subgraph matches the actual capabilities of the current offshore wind turbine combiner box data acquisition unit.

[0051] Then, the influence functions associated with each edge in the observable causal subgraph can be simplified. This simplification can be achieved by dynamically adjusting the available computing resources and memory of the microcontroller at the edge of the offshore wind turbine combiner box data acquisition unit to strike a balance between inference real-time performance and model accuracy. Specifically, the influence functions bound to each edge in the observable subgraph can be simplified to ensure that a single calculation is completed within microseconds. For example, for complex nonlinear functions, the input-output values ​​can be pre-calculated and stored as a lookup table. Linearization / polynomial fitting can also be used, or high-order dynamic systems can be simplified to first-order inertial elements. Furthermore, while ensuring the propagation paths of critical faults, clustering of some nodes can further reduce graph complexity.

[0052] Finally, personalized parameter tuning can be performed by retrieving equipment calibration data (such as factory test reports and initial health operation data) from the digital archive of the offshore wind turbine combiner box data acquisition unit, and replacing the theoretical nominal values ​​of corresponding parameters in the model with these actual measured values. For example, the general thermal resistance value can be replaced with the actual thermal resistance fitted to the equipment using initial data. Simultaneously, health baselines (such as normal temperature range and vibration spectrum baselines) for key nodes of the offshore wind turbine combiner box data acquisition unit can be pre-calculated or downloaded from the cloud. These baselines serve as initial normal references for nodes in the model and can be integrated into the initialization parameters of the dynamic causal graph.

[0053] In practical applications, the aforementioned dynamic cause-effect graph can be deployed on the edge side of the offshore wind power combiner box data acquisition unit, i.e., deployed locally, which can ensure the real-time nature of monitoring and diagnosis.

[0054] In one possible implementation, after constructing the prior causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit, the following steps are included: instantiating the prior causal knowledge model into a digital twin causal model corresponding to the offshore wind power combiner box data acquisition unit. After obtaining the diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit, the process includes: using the digital twin causal model, utilizing the high-frequency sampling data and the diagnostic results, and through multi-hypothesis simulation verification and Bayesian inference, to obtain the in-depth diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit.

[0055] In practical applications, a digital twin causal model is a virtual mirror that is highly consistent with the data acquisition unit of the offshore wind turbine combiner box and can be used for deep scientific computing. It is generally deployed in the cloud.

[0056] When an anomaly occurs at a critical node, high-frequency sampling data and diagnostic results can be used in the cloud to extrapolate within the aforementioned digital twin causal model, pinpointing the root cause to a specific component and obtaining more accurate and specific in-depth diagnostic results.

[0057] In one possible implementation, instantiating the prior causal knowledge model into a digital twin causal model corresponding to the offshore wind power combiner box data acquisition unit includes: Replace the influence function associated with multiple directed edges in the prior causal knowledge model with the physical solver corresponding to the influence function; The factory calibration data of the offshore wind power combiner box data acquisition unit is injected into the prior causal knowledge model. The initial healthy operation data of the offshore wind power combiner box data acquisition unit is used to optimize the parameters in the prior causal knowledge model that cannot be directly measured through parameter identification algorithm, so that the simulation output matches the actual measurement data in a healthy state. A data assimilation algorithm for real-time synchronization with the actual state of the offshore wind power combiner box data acquisition unit is integrated into the prior causal knowledge model, and a multi-hypothesis diagnostic reasoning engine is configured in the prior causal knowledge model to obtain a digital twin causal model corresponding to the offshore wind power combiner box data acquisition unit.

[0058] In practical applications, based on the aforementioned prior causal knowledge model, independent computing resources can be allocated to the offshore wind power combiner box data acquisition unit in the cloud to create a configurable digital twin instance container. A globally unique ID is not generated for this instance and is not strongly bound to the physical device's serial number, establishing a one-to-one, tamper-proof mapping relationship to avoid unnecessary errors during use.

[0059] In practical applications, the influence functions of multiple directed edges in a priori causal knowledge models can be replaced with a physical solver. This physical solver can receive state data from the source nodes in the offshore wind turbine combiner box data acquisition unit, and output high-precision state values ​​of the target nodes by solving the governing equations (e.g., partial differential equations or high-level constitutive equations) representing the physical processes of the edge. For example, the algebraic formula for "current → junction temperature" can be replaced with a three-dimensional transient thermo-stress coupled finite element solution physical solver. This physical solver can use the current chip power consumption distribution and ambient temperature as loads and boundary conditions, and use the precise three-dimensional parametric geometric model corresponding to the data acquisition unit to solve the heat conduction equation and elasticity equation, outputting a high-precision distribution field of chip junction temperature and PCB thermal stress. For corrosion prediction edges, an electrochemical-transmission coupled solver can be used, employing diffusion-reaction equations that include salt spray concentration, humidity, and temperature fields to output predictions of the corrosion pit depth evolution over time.

[0060] In practical applications, for edges in the model that do not require high precision (such as partially linear transitive relationships), their original parameterized influence functions can be retained to improve overall computational efficiency.

[0061] Then, actual measured values ​​of key components (such as the actual resistance of precision resistors, the actual output voltage of the reference voltage source, and the gain / offset error of the ADC) can be extracted from the factory calibration data (e.g., digital factory test report) of the offshore wind turbine combiner box data acquisition unit, replacing the theoretical nominal values ​​of the corresponding components in the aforementioned prior causal knowledge model. Finally, the parameter identification algorithm can be run using the operating data of the first month after the equipment is put into operation (confirmed to be healthy). This parameter identification algorithm automatically adjusts those gray box parameters that cannot be directly measured but have a significant impact, such as the actual contact thermal resistance from the chip to the heat sink, the comprehensive heat transfer coefficient of the chassis to the external environment, and the damping coefficient of the vibration transmission path, by comparing the simulation output of the digital twin causal model with the measured data. After calibration, the behavior of the model in a healthy state is highly consistent with the physical entity. The above data can also be statistically processed (e.g., mean, variance, spectral characteristics) and stored in an auxiliary database as a unique health fingerprint of the equipment, serving as a benchmark for judging "normal" in subsequent "virtual-real comparison".

[0062] Ultimately, data assimilation algorithms can be integrated, such as embedding state estimation algorithms (e.g., ensemble Kalman filtering, particle filtering) into the model. This algorithm uses continuously received edge sensor data as "observations," compares them with the twin model's "predictions," and adjusts a large number of internal state variables in the model that cannot be directly measured (e.g., chip junction temperature, current value of equivalent series resistance of capacitors, aging degree of sealing materials). This allows the state of the virtual digital twin causal model to be dynamically synchronized with the real internal state of the physical device.

[0063] A multi-hypothesis diagnostic inference engine can also be configured for the model, enabling it to automatically generate multiple fault hypotheses in a virtual environment based on the input abnormal data, and calculate the posterior probability of each hypothesis through parallel simulation to achieve accurate diagnosis. For example, when an abnormal signal is received, the aforementioned multi-hypothesis diagnostic inference engine can automatically trigger a diagnostic inference closed loop of "hypothesis generation → simulation verification → probability inference" based on its graph structure to achieve diagnostic inference.

[0064] In practical applications, a counterfactual simulation control engine can also be configured for the model. This counterfactual simulation control engine can be used to modify the model's inputs, parameters, or structure according to diagnostic or decision-making needs, and drive the model to re-execute the simulation to extrapolate the future state under different intervention measures. In this way, it is convenient to modify the model parameters (simulate maintenance) or external conditions (simulate environmental changes) and rerun the simulation to evaluate the consequences of "if...then...".

[0065] In practical applications, the aforementioned digital twin causal model can be deployed in the cloud, such as on cloud servers or edge servers at the site, and allocated sufficient cloud computing resources to enable it to run in the cloud.

[0066] In one possible implementation, after sampling key nodes in the offshore wind turbine combiner box data acquisition unit at a first frequency to obtain a baseline operating state of the offshore wind turbine combiner box data acquisition unit, the method further includes: If the observed value at the critical node exceeds the preset hardware safety limit threshold, it is determined that there is a safety anomaly in the offshore wind power combiner box data acquisition unit, triggering a preset hardware protection action. If the abnormal pattern present at the critical node matches the preset instantaneous interference pattern library, it is determined that there is instantaneous interference in the offshore wind power combiner box data acquisition unit, and logs are recorded. If the abnormal pattern existing at the critical node matches a single fault pattern in the preset fault knowledge base, the handling suggestion corresponding to the single fault pattern is triggered. The key node is determined to be abnormal if the abnormal pattern of the key node satisfies at least two of the following conditions: persistence, correlation, and predictive value.

[0067] In one possible implementation, the persistence condition is that the duration of the abnormal state of the critical node exceeds a first time threshold, or the number of times it occurs repeatedly within a time window exceeds a frequency threshold. The correlation condition is that, in the dynamic causal graph, there is a direct causal edge connection between the key node and at least one other key node. The predictive value condition is that the abnormal behavior of the critical node is a unidirectional trend drift, or that the critical node will exceed the hardware security limit threshold within a second time threshold in the future.

[0068] In practical applications, the hardware safety limit threshold (also called the alarm threshold) is an absolute value that cannot be exceeded, set based on the maximum rated parameters of the components within the offshore wind turbine combiner box data acquisition unit. Preset hardware protection actions include immediate power-off, hardware reset, switching to a backup redundant unit, or entering a minimum power consumption safe state. When observed values ​​at certain critical points exceed the absolute hardware safety limit—for example, when the internal temperature of the chassis momentarily exceeds 125°C (the junction temperature limit of the chip), or the input voltage exceeds 40V (the withstand voltage limit of the offshore wind turbine combiner box data acquisition unit)—preset hardware protection actions (such as power cut-off, emergency shutdown, and switching to a backup circuit) can be executed immediately and unconditionally. This response is rule-based, reflective, and speed-prioritized, allowing for timely system maintenance and ensuring system stability. After the protection action is executed, this critical event can be recorded, and after the equipment recovers, the data before and after the event can be used as an extreme case for subsequent optimization of the prior causal knowledge model. In the above embodiment, immediate causal reasoning is not required in such emergency situations.

[0069] In practical applications, if anomalies at critical nodes are transient, isolated, and clearly attributable to known external interference, and are already stored in a transient interference pattern library—for example, a voltage spike lasting milliseconds and within a safe range caused by the start-up or shutdown of nearby high-power equipment; or a single communication error induced by a lightning strike—then simply logging or triggering a low-level log event is sufficient. There is no need to collect high-frequency sampling data for immediate causal inference, as the root cause (external interference) is clear and transient, and has no ongoing diagnostic value.

[0070] In practical applications, if the fault mode corresponding to a critical node is singular and clear, that is, it matches a pre-defined single fault mode in the fault knowledge base. For example, a cooling fan's "zero speed" alarm (fan hardware failure) or a device restart caused by timeout. In this case, the fault knowledge base can be directly queried for the corresponding handling suggestions for the single fault mode, automatically generating a simple work order for replacing the fan or checking the code. This process is based on searching the fault knowledge base and does not require locating the root cause, because the "effect" (fan stopping) and the "cause" (fan failure) are almost synonymous. The fault knowledge base can be stored locally on the offshore wind turbine combiner box data acquisition unit for easy retrieval.

[0071] In practical applications, an anomaly can trigger a complete causal analysis process when it meets the following characteristics: Persistence: The anomaly is not instantaneous but persistent or recurring. Correlation: The anomaly affects multiple parameters, or is expected to affect multiple parameters (according to a dynamic causal graph). For example, an anomaly of "increased internal humidity" may not be severe in itself, but according to a dynamic causal graph, it is associated with "communication impedance" and "corrosion risk," thus having high diagnostic value. Predictive value: The anomaly can exhibit a slow, gradual trend, indicating potential future functional loss. For example, a value moves slowly but continuously in one direction. During the calculation, linear regression can be used to calculate the slope of the data within the sliding time window; or a CUSUM control chart can be used, which can sensitively capture small, persistent shifts in the process mean. If the slope is statistically significantly non-zero (p-value < 0.01) and the direction is unfavorable, it is determined to be a trend drift. Alternatively, extrapolation prediction can be performed by combining current monitoring values ​​and the physical / statistical degradation models corresponding to key nodes. The second time threshold is an acceptable maximum lead time or minimum remaining time defined by the operation and maintenance strategy. For example, "an early warning must be issued at least 7 days before a fault occurs," then the "second time threshold" is 7 days. In practical applications, the conditions met by critical nodes can also be of unknown origin, meaning that the anomalies at the critical node cannot be directly correlated to a clear and simple fault point. For example, the "effect" of "increased power ripple" may be caused by multiple "causes" such as "aging input capacitors," "sudden load changes," and "unstable control loops," requiring further reasoning.

[0072] For example, the alarm threshold (or hardware safety limit threshold) for the internal humidity of a data acquisition unit in an offshore wind turbine combiner box is 85%RH, and its health baseline (the normal characteristic pattern corresponding to key nodes when the equipment is in a confidently healthy and normal working state) is 61%RH (±3). During monitoring, the internal humidity slowly and continuously rises from 61% to 69%. At this point, it does not exceed the safety limit (<85%RH), is not a transient interference, and is not a simple and obvious fault (it is unknown whether it is a sealing ring problem or an environmental problem). Furthermore, it triggers the characteristics of causal analysis: it meets the continuity condition; it meets the correlation condition (correlated with "communication impedance" and "corrosion"); and it has predictive value (the internal humidity slowly and monotonously rises from 61%RH to 68%RH within a week. Although each step is within the baseline fluctuation range, the overall trend is abnormal, indicating future corrosion). At this point, it can be determined that there is an anomaly at the key node. Thus, through the aforementioned pre-emptive, lightweight intelligent filtering, it is ensured that only those complex, correlated, predictive, and unexplained problems can be further diagnosed in depth. This design ensures extremely high monitoring intelligence while also guaranteeing the overall system's efficiency and engineering practicality, preventing the waste of computing resources and alarm fatigue among maintenance personnel. This addresses the issue of highly concealed faults and enables predictive maintenance.

[0073] In one possible implementation, the step of performing real-time causal inference on the high-frequency sampled data based on the dynamic causal graph to obtain the diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit includes: The time-domain and frequency-domain features of the high-frequency sampled data are calculated to generate a feature vector corresponding to the offshore wind power combiner box data acquisition unit, wherein the feature vector includes the feature value of the target node at the current time; Each feature value in the feature vector is compared with the health baseline corresponding to the feature value to obtain the observed abnormal nodes whose difference from the health baseline is greater than a preset threshold, so as to generate a set of evidence of observed abnormalities. Starting from the current state indicated by the feature vector, a reverse search is performed along the causal edges in the dynamic causal graph to obtain multiple candidate cause nodes that can explain this abnormal event. Based on the prior probability of the candidate cause nodes, the posterior probability of the candidate cause nodes causing the current abnormal event is calculated, and the candidate cause nodes are sorted according to the posterior probability to obtain a sorted root cause list. Starting from the observed abnormal nodes in the observed abnormal evidence set, state propagation is carried out along the causal direction in the dynamic causal graph to predict the state change trend of the nodes in the dynamic causal graph after a future time period, and to determine whether the nodes are at risk of exceeding the safety threshold, so as to generate forward prediction results. Based on the sorted root cause list and the forward prediction results, a diagnostic result is generated for the offshore wind power combiner box data acquisition unit.

[0074] In practical applications, high-frequency sampled data (such as voltage, temperature, and vibration sampled at 1kHz) can be aligned and buffered. All channel data is strictly aligned on the time axis (to compensate for sensor sampling delay) and stored in a short-term sliding window buffer (e.g., data from the most recent 5 seconds). Then, for each target node's data in the buffer, its time-domain characteristics are calculated, including mean, root mean square, peak value, rise / fall slope, and number of threshold crossings. Frequency-domain characteristics can also be calculated, for example, by using FFT to calculate the amplitude of the main spectral components. This yields a feature vector V(t), where each element corresponds to the current feature value of a target node. For example: V(t) = {"Ripple_RMS": 120mV, "Temperature_Shell Temperature": 68.5℃, "Vibration_1xAmplitude": 0.15g,...}.

[0075] Then, each feature value in the current feature vector V(t) can be compared with its corresponding healthy baseline. Target nodes that deviate significantly from the baseline (with a gap greater than a preset threshold) are marked as observed abnormal nodes, thereby constructing abnormal evidence. Each piece of abnormal evidence includes observed abnormal node, abnormal features, degree of deviation, and timestamp. In this way, an observed abnormal evidence set O can be formed.

[0076] Thus, nodes that have clearly exceeded their healthy baseline can be selected from the current feature vector V(t), forming a set O of observed anomalous nodes. Then, starting from each observed anomalous node in set O, a backward search (backtracking) can be performed along the "result ← cause" edge in the dynamic causal graph. All possible upstream cause nodes are traversed, all nodes found through the backtracking path are merged, and duplicates are removed, forming a set C of candidate cause nodes. The topological structure of the causal graph ensures that the search does not become endless.

[0077] Subsequently, for each candidate root cause c in set C... i Calculate a posterior probability P(c) that can explain all observed anomalies O. i | O). Calculation based on: Prior probability P(c i ): The inherent probability of this fault occurring (derived from the assignment during model initialization, which can be obtained based on historical fault data).

[0078] Likelihood P(O | c) i ): If c i If true, the probability of observing O is 0. Due to limited edge computing power, estimation can be performed by querying the dynamic causal graph from c. iDetermine the path to each anomalous node in O, the influence function of the edges, and the initial weights, and estimate c. i To what extent and how quickly can these anomalies occur? The shorter the path and the higher the weight, the higher the likelihood. In practical applications, this can be simplified to rule-based matching or small-scale conditional probability table queries.

[0079] Applying the basic idea of ​​Bayes' theorem, an approximate calculation is performed, with the formula: P(c i | O) ∝ P(O |c i ) * P(c i In practical applications, the results can be normalized to obtain the relative posterior probability of each candidate root cause.

[0080] In practical applications, candidate cause nodes can be sorted from high to low according to their calculated posterior probabilities to generate a sorted root cause list, for example: [(node: power supply filter capacitor C105, probability: 0.70), (node: load current sensor, probability: 0.25),...].

[0081] Then, starting from the current state described by the current feature vector V(t), one or more steps of state propagation calculation can be performed on the dynamic causal graph along the "cause → effect" edge. Using the simplified influence function of the directed edge association in the dynamic causal graph, the future state is calculated recursively in a forward manner. Through propagation, the predicted state vector V'(t+Δt) of each node in the graph is calculated after a short time interval Δt (e.g., 10 seconds or 1 minute). Special attention is paid to nodes that are currently normal but are predicted to be abnormal.

[0082] Then, the predicted state V'(t+Δt) can be compared with the safe operating threshold corresponding to each node. If it is predicted that a node will exceed the threshold, a predictive warning can be generated, such as "the power module temperature is predicted to exceed the limit in 60 seconds", which can trigger the system to enter a higher alert level in advance.

[0083] In practical applications, the diagnostic results mentioned above may include the sorted root cause list and the forward prediction results.

[0084] In practical applications, the urgency level of the current abnormal event (e.g., "high / medium / low") can be calculated based on a set of predefined rules. Evaluation factors may include: the confidence level of the most likely root cause (e.g., whether the probability is >0.8); the number of anomalous nodes and their impact on functionality; and the results of forward prediction (whether it will lead to more serious consequences).

[0085] For example, the diagnostic results may also include the top-ranked candidate node and its probability; a list of other candidate nodes; an urgency assessment (high, medium, low level); and (if any) a prediction of the problems that will occur.

[0086] In practical applications, if the urgency level is high, preset protection actions can be executed immediately without waiting for cloud commands, such as switching to backup circuits, restarting faulty modules, and sending emergency shutdown signals, to ensure the normal operation of the offshore wind turbine combiner box data acquisition unit or reduce losses. It should be noted that regardless of the urgency level, high-frequency sampling data and diagnostic results must be uploaded to the digital twin causal model for in-depth analysis.

[0087] In one possible implementation, the digital twin causal model is used, along with the high-frequency sampling data and the diagnostic results, to obtain a deep diagnostic result of the operating status of the offshore wind power combiner box data acquisition unit through multi-hypothesis simulation verification and Bayesian inference, including: The high-frequency sampling data is used as an observation value and input into the digital twin causal model. The state of the digital twin causal model is synchronized with the state of the offshore wind power combiner box data acquisition unit during the fault period through a data assimilation algorithm, and the state variables that cannot be directly measured inside the digital twin causal model are estimated. Based on the diagnostic results, a set of possible fault hypotheses are generated. For each fault hypothesis, the parameters corresponding to the fault hypothesis are modified in the digital twin causal model to simulate the fault. The multiphysics transient simulation is then rerun to obtain the virtual observation data corresponding to the fault hypothesis. For each of the fault hypotheses, the matching likelihood between the virtual observation data and the high-frequency sampling data is calculated. Combined with the prior probability of occurrence of the fault hypothesis, the posterior probability that the fault hypothesis is the true root cause is calculated using Bayes' theorem. The fault hypothesis with the highest posterior probability is selected as the precise root cause result, and the digital twin causal model is used to analyze the precise root cause diagnosis result to obtain the in-depth diagnosis result of the operating status of the offshore wind power combiner box data acquisition unit.

[0088] In practical applications, ensemble Kalman filtering or particle filtering algorithms can be used for data assimilation. This algorithm uses the digital twin causal model as a "predictor" and high-frequency sampled data as observations for iterative calculations. Through assimilation, the algorithm can back-estimate and update hundreds of internal unobservable state variables in the twin model, ensuring that the virtual model's internal state at the point of failure is as consistent as possible with the physical device. For example, it can accurately estimate the chip's true junction temperature at the moment of failure, the current equivalent series resistance of the capacitor, and the real-time aging degree of the sealing material. At this point, the digital twin has achieved state synchronization with the physical world, laying the foundation for subsequent accurate simulations.

[0089] Generate a set of fault hypotheses: Using the root cause list from the diagnostic results above as a seed, and combining it with the fault knowledge base, enumerate a set of specific, simulable fault hypotheses H = {h1, h2, ..., hn}. Each hi represents a specific component fault or parameter drift, for example: h1: The ESR of the power supply filter capacitor C105 increases from 50mΩ to 200mΩ.

[0090] h2: Increased bearing friction in cooling fan FAN1 caused a 20% decrease in speed.

[0091] h3: The reference voltage source Vref of communication chip U201 drifts by -5%.

[0092] On the digital twin copy that has been synchronized with the state, create a separate simulation branch for each hypothesis hi. In each branch, modify the parameters corresponding to the model to accurately simulate the fault (e.g., change the ESR parameter of C105 to 200mΩ in branch 1).

[0093] On each branch following fault injection, starting from the fault injection moment, the transient multiphysics simulation is rerun to simulate the entire process from fault occurrence to data upload, thus obtaining a set of virtual sensor data D. i That is, "if hi is true, what should each sensor observe?"

[0094] Likelihood calculation: For each hypothesis hi, calculate its likelihood P(observed data O | hi). That is, compare the degree of matching between the virtual observed data Di under this hypothesis and the actual observed data O. The higher the matching degree, the greater the likelihood. The matching degree calculation can take into account sensor noise, model error, etc.

[0095] Based on the prior probability P(hi) of this failure hypothesis (from historical statistics or FMEA), the posterior probability is calculated using Bayes' theorem: P(hi | O) = [P(O | hi) * P(hi)] / Σ[P(O | hj) * P(hj)] The posterior probabilities of all hypotheses are normalized so that their sum equals 1. The hypothesis h* with the highest posterior probability P(h* | O) is selected as the most likely root cause, and its probability value is the confidence level of the diagnosis. Simultaneously, the top 3-5 hypotheses and their probabilities are obtained, providing a complete perspective for judgment. This allows for precise root cause localization, in the following format: Faulty component: C105; Fault mode: Increased ESR; Parameter estimation: 215mΩ ±10mΩ; Confidence level: 94%. This achieves a leap from the module level to the component level, and from qualitative to quantitative analysis.

[0096] In practical applications, after obtaining accurate root cause results, the complete causal path from the root cause h* to all observed anomaly nodes can be traced back based on the aforementioned digital twin causal model, enabling fault propagation chain reconstruction and quantifying the contribution and time delay of each edge. Simultaneously, natural language descriptions can be generated, such as "Due to the aging of C105, ..., which in turn causes ...". Furthermore, the current impact of this fault on system performance, safety, and efficiency can be quantified (e.g., "leading to a 3% increase in data acquisition error"). Starting from the current diagnosed state, a component degradation model can be invoked to perform accelerated life simulation within the digital twin causal model, predicting the time distribution of component failure. Output: Remaining useful life: 65 days. In practical applications, counterfactual reasoning can be used to generate maintenance strategy recommendations (such as "immediate replacement," "planned maintenance," "reduced mileage operation," and their effectiveness evaluation). Thus, deep diagnostic results can include accurate root cause results, quantified confidence levels, fault propagation chains, RUL predictions, and strategy recommendations. In practical applications, the content of deep diagnostic results is not specifically limited.

[0097] It's important to note that inference using dynamic causal graphs identifies anomalies and narrows down the general range of suspects within seconds, prioritizing speed over accuracy. It can also assess the urgency of the anomaly and trigger local protection mechanisms. In contrast, inference using digital twin causal models requires precise, component-level diagnosis and quantitative prediction, prioritizing accuracy over speed. Bayesian inference is applied in both, but at different depths and through different processes. In dynamic causal graph inference, it primarily involves lightweight probability assessment to rank candidate hypotheses and identify the most likely nodes. In digital twin causal model inference, it serves as the top-level logic driving and managing a series of simulation experiments. Simulations provide likelihood estimates for Bayesian inference, which then determines which hypothesis to test next. Edge diagnostic results (based on simplified Bayesian inference) provide a high-quality search starting point and problem focus for deep analysis in the cloud (based on simulation Bayesian inference), avoiding blind searches in a vast hypothesis space. The accurate results from the cloud can, in turn, correct and update the prior probabilities and parameters of the dynamic causal graph, making it more accurate in subsequent iterations.

[0098] In one possible implementation, after obtaining the in-depth diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit, the method further includes: Obtain maintenance resource information associated with the offshore wind power combiner box data acquisition unit; Using a rule engine, based on the deep diagnostic results and the maintenance resource information, the executable instructions corresponding to the offshore wind power combiner box data acquisition unit are determined, wherein the executable instructions are used to resolve this abnormal event; The complete data of multiple abnormal events are linked and stored to form tagged fault diagnosis cases; Perform trend analysis on the long-term operating data of the key nodes to identify the slow drift of the key nodes due to aging, and adjust the health baseline corresponding to the key nodes. Based on the operational data of the offshore wind power combiner box data acquisition unit within the sliding time window, the statistical correlation between nodes in the prior causal knowledge model is recalculated to update the weights and time delay parameters of the directed edges in the dynamic causal graph and the digital twin causal model. Monitor and analyze abnormal patterns that are not explained by the prior causal knowledge model. When the abnormal pattern is detected, add, delete and optimize the local structure of the prior causal knowledge model and update it synchronously to the dynamic causal graph and the digital twin causal model.

[0099] In practical applications, maintenance resource information associated with the offshore wind turbine combiner box data acquisition unit can be obtained first, such as the inventory quantity, location, minimum order quantity, and procurement lead time of the spare parts corresponding to the faulty components; the skills and qualifications, location, and current workload of the available maintenance team; and future weather windows, sea conditions, and safe operation requirements.

[0100] Then, the rules engine is used for analysis. For example, in practical applications, when the remaining useful life is below the first threshold or the failure has caused functional loss, the decision can be to perform maintenance immediately and schedule resources with the highest priority; when the remaining useful life is above the first threshold but below the second threshold, and there is a future available planned downtime window, the decision can be to perform maintenance within the planned window; when the remaining useful life is above the second threshold, the decision can be not to perform maintenance immediately, but to put the equipment on the key monitoring list and shorten the inspection cycle.

[0101] For example, in practical applications, a rule engine can be a system with a large number of pre-defined "IF-THEN" business logic statements. Example rules are as follows: IF Emergency Level == High OR RUL < 24 hours THEN Final Alert Level = Red (Emergency).

[0102] If urgency level == medium AND RUL > 7 days AND planned shutdown window THEN final warning level = orange (warning).

[0103] If urgency level == low AND RUL > 30 days THEN, final alert level = yellow (caution).

[0104] The rule engine can also match maintenance measures based on the precise root cause results in the deep diagnostic results. For example, if the root cause is "C105 ESR increased", then the measure is "Execute SOP-ELC-001: Replace electrolytic capacitor", associated tools [soldering iron, desoldering pump], and safety procedures [power off, discharge, voltage test].

[0105] Example rules in a rule engine can also be: IF Warning Level == Red THEN Execution Mode = "Emergency Repair", Scheduling = "Immediate", Resource Scheduling = "Highest Priority".

[0106] IF Warning Level == Orange AND Spare Parts in Stock AND Planned Window Within the Next 14 Days THEN Execution Mode = "Planned Maintenance", Planned Time = "Next Available Window", Automatically Reserve Spare Parts.

[0107] If spare parts are out of stock, the "Purchase Request" sub-process is triggered, and the RUL and schedule are recalculated.

[0108] Finally, based on the output obtained by the rule engine through in-depth diagnostic results and maintenance resource information, executable instructions can be generated. For example, executable instructions may include: execution mode: emergency repair / planned maintenance; suggested time window: A; required resources: name, code, quantity, storage location; personnel skill requirements: electrical maintenance qualification; completion verification standard: such as "after replacement, the power ripple value monitored by the system should be less than 50mV".

[0109] After repairs are completed, maintenance personnel can upload a completion report and verification data (such as the measured ripple value after replacement). The system automatically compares the verification data with the expected recovery value (from digital twin causal model simulation prediction). If they match, the work order is automatically closed, and the complete "diagnosis-repair-verification" case is archived for subsequent model optimization. If they do not match, an alarm is triggered, requiring a re-inspection.

[0110] In practical applications, complete data related to each anomaly event can be archived. This complete data can include: high-frequency sampling data, preliminary analysis results and in-depth diagnostic results, executable instructions, and post-maintenance verification data (monitoring data after equipment restart / maintenance). Once a maintenance work order is closed and the maintenance personnel have reported the actual cause of the fault (e.g., "on-site confirmation of C105 replacement"), this ground truth value can be automatically associated with the previous diagnostic report, forming a tagged "diagnosis-maintenance-verification" closed-loop case. The digital twin archive of the offshore wind turbine combiner box data acquisition unit can record the components involved in the maintenance, the time, and the post-maintenance performance baseline. This forms a unique "health history" for the equipment.

[0111] In practical applications, long-term trend analysis can be continuously performed on the health indicators (such as power efficiency, ADC baseline stability, and background noise) corresponding to key nodes of the equipment, using time series models (such as exponential smoothing and ARIMA) to distinguish between short-term fluctuations and long-term drift. For key nodes using threshold judgment, if their monitored values ​​show slow, unidirectional drift (such as a temperature slowly rising with the seasons), the upper / lower limits of their normal range are adaptively relaxed according to preset rules (such as the statistical distribution of a sliding window). For nodes using a predictive model as a baseline, the predictive model is retrained with new data or updated online (such as linear regression coefficients) so that the model's predictions can track the aging trajectory of the equipment. In practical applications, baseline updates are slow, conservative, and subject to upper and lower bound constraints. The aim is to allow the equipment to "age healthily," avoiding continuous false alarms for reasonable performance degradation, while ensuring that accelerated deterioration or sudden failures can still be detected.

[0112] In practical applications, sliding time windows (such as data from the most recent 90 days) can be used to recalculate the statistical correlations or physical parameter sensitivities between observable nodes in a dynamic causal graph. The weights and time delay parameters of corresponding directed edges are then smoothly updated using the new statistical results. For example, if vibration is found to have a more significant impact on a contact resistance than previously thought, the weight of that edge can be increased. Long-term operational data (especially data on slow performance degradation) can also be used to refit or Bayesian update the parameters of the physical degradation models built into the digital twin causal model (such as the ESR growth model for capacitors or the aging model for sealing materials). This makes the model's prediction of the remaining lifespan of the equipment increasingly accurate over time. In practical applications, the prior probabilities of each failure mode in the failure knowledge base can be updated based on statistics from historical failure cases. If a certain type of failure occurs repeatedly, its prior probability needs to be increased, making the system more vigilant about it in the future.

[0113] In practical applications, anomaly detection algorithms can be continuously run to identify recurring anomalous data patterns that cannot be explained by existing prior causal knowledge models. If a new pattern is statistically significant and has engineering implications (e.g., a specific vibration spectrum accompanied by a specific current harmonic), it is marked as a candidate for a "potential new failure mode." For this new pattern, efforts can be made to discover other variables with strong statistical correlations in a broader dataset. Combining domain knowledge, hypotheses about its possible causes and consequences are generated, and attempts are made to find "gap" positions in the existing dynamic causal graph. Then, the discovered "potential new failure modes" and their hypothetical causal relationships are compiled into a knowledge update suggestion report and submitted to engineers for review. After confirmation, the aforementioned prior causal knowledge model is updated, for example, by adding nodes representing this new failure mode or new physical quantity; adding newly discovered causal edges, or deleting causal edges verified as false; and, in rare cases, modifying local connectivity relationships. The above updates need to be simultaneously synchronized to the digital twin causal model and the dynamic causal graph.

[0114] To provide a more detailed explanation of the offshore wind power combiner box data acquisition unit operation status monitoring method in the embodiments of this application, the following supplementary descriptions are also provided in the embodiments of this application: Figure 2 A flowchart of a method for real-time causal inference of high-frequency sampled data provided in this application embodiment is included, the method comprising the following steps: Step 201: Perform data alignment and caching on the high-frequency sampling data of the target node.

[0115] Step 202: Calculate the time-domain and frequency-domain characteristics of the high-frequency sampling data.

[0116] Step 203: Assemble the time-domain features and frequency-domain features to form the feature vector corresponding to the target node.

[0117] Step 204: Using the feature vector as input, perform forward deduction along the dynamic causal graph to obtain candidate cause nodes.

[0118] Step 205: Calculate the posterior probability of each candidate cause node based on a simplified Bayesian network.

[0119] Step 206: Sort the candidate cause nodes according to the posterior probability to obtain the sorted root cause list.

[0120] Step 207: Identify the abnormal observation nodes.

[0121] Step 208: Starting from the observed abnormal node, propagate the state along the causal direction in the dynamic causal graph to predict the state of each node after the future event segment.

[0122] Step 209: Determine if any nodes are about to cross the boundary in order to generate forward prediction results.

[0123] Step 210: Based on the sorted root cause list and the forward prediction results, assess the urgency of this abnormal event.

[0124] Step 211: Generate diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit.

[0125] Figure 3 A flowchart of a method for obtaining deep diagnostic results using a digital twin causal model, provided in an embodiment of this application, is included. This method may include the following steps: Step 301: Input the high-frequency sampling data as observations into the digital twin causal model, and drive the state of the digital twin causal model to synchronize with the state of the data acquisition device during the fault period through the data assimilation algorithm.

[0126] Step 302: Estimate the state variables that cannot be directly measured within the digital twin causal model.

[0127] Step 303: Enumerate the set of fault hypotheses.

[0128] Step 304: For each fault hypothesis, modify the parameters in the digital twin causal model to simulate the fault.

[0129] Step 305: Run the simulation and record the virtual observation data corresponding to the fault assumption.

[0130] Step 306: Compare the matching degree between the virtual observation data and the high-frequency sampling data.

[0131] Step 307: Calculate the posterior probability corresponding to the fault hypothesis.

[0132] Step 308: Select the fault hypothesis with the highest posterior probability as the exact root cause result.

[0133] Step 309: Based on the precise root cause results, reconstruct the fault propagation chain.

[0134] Step 310: Based on the precise root cause results, assess the impact of this event and the remaining lifespan of the components.

[0135] Step 311: Generate maintenance strategy suggestions through counterfactual reasoning.

[0136] Step 312: Generate in-depth diagnostic results.

[0137] The following specific embodiment will further explain the above-mentioned method for monitoring the operational status of the offshore wind power combiner box data acquisition unit.

[0138] Step 1: Scene and Initial State.

[0139] Monitoring object: Data acquisition unit (ID: DAU-05) inside the combiner box of wind turbine #05 in an offshore wind farm.

[0140] Environmental background: The wind farm is located in a high-salinity, foggy sea area. It is currently not typhoon season, but sea fog is frequent.

[0141] System Status: The system has completed initialization, built a personalized digital twin causal model for DAU-05, and loaded a dynamic causal graph. The device is operating stably and is in normal monitoring mode.

[0142] Normal monitoring mode: Key node data (internal humidity of the chassis, main power supply voltage, chassis temperature, etc.) are collected every 5 minutes and compared with the health baseline. Everything is normal.

[0143] Step 2: Detection of anomalies and causal drivers.

[0144] Day T: The internal humidity of the enclosure was monitored to slowly rise from a stable 65%RH to 78%RH over 12 hours. This value is still below the general alarm threshold (usually 85%RH), but has been deviating from the health baseline of the device over the past month (mean 62%RH, standard deviation ±5%RH).

[0145] Anomaly detection: An anomaly has been detected in the "Humidity" node.

[0146] Focused Perception: Based on a dynamic cause-effect graph, the system automatically triggers synchronous high-frequency sampling (at 1-minute intervals) of nodes strongly correlated with "abnormal humidity". Electrical connections: communication port (RS485) impedance, power input ripple.

[0147] Chemical correlation: Corrosion risk index (calculated in real time by temperature and humidity model).

[0148] Environmental correlation: Synchronously read the external ambient humidity provided by the airborne weather station.

[0149] Step 3: Real-time edge reasoning to obtain diagnostic results.

[0150] Focused sensing lasted for 2 hours. Data showed that: internal humidity remained at 78-80%RH; communication port impedance slowly and monotonically increased from a stable 1.2Ω to 1.9Ω within 2 hours; corrosion risk index increased from 20 to 40; external environmental humidity returned to normal (65%RH); power ripple was normal.

[0151] Reverse diagnosis: Using "increased communication impedance" as the primary abnormal evidence, reverse reasoning was performed in the dynamic causal graph. The model calculations showed that pin corrosion of connector J1 (RS485) and loose connector were the two most likely candidate root causes, with probabilities of 0.65 and 0.25, respectively.

[0152] Forward prediction: The model predicts that if humidity conditions are maintained, the communication impedance will reach a critical value of 3.0Ω in 5-7 days, which may cause occasional communication interruptions.

[0153] Urgency assessment: Due to the involvement of corrosion (an irreversible process) and the predicted impact time within a week, the system is assessed as having a "medium" urgency level.

[0154] Initial action: The edge device does not perform an emergency shutdown (its functionality has not yet been affected), but it packages the preliminary diagnostic report (suspected connector corrosion, 65% probability) and all high-frequency sampling data, marks it as "early warning" level, and immediately reports it to the cloud.

[0155] Step 4: Precise cloud-based analysis.

[0156] Data Injection Twin: After receiving the data in the cloud, it is injected into the digital twin causal model corresponding to DAU-05.

[0157] State synchronization: An ensemble Kalman filter is run to synchronize the state of the digital twin causal model with the physical device. The model estimates unmeasurable parameters such as the current oxide film thickness on the connector surface and confirms the internal high humidity state.

[0158] Multi-hypothesis simulation verification: (1) Based on the preliminary edge diagnosis, three specific hypotheses are generated: H1: Electrochemical corrosion of the gold plating layer at RS485 port J1 (simulated corrosion depth 120nm).

[0159] H2: The internal spring stress of port J1 is relaxed (simulated contact pressure decreases by 30%).

[0160] H3: Moisture on the PCB board leads to a decrease in insulation, creating a leakage path (simulating a decrease in insulation resistance).

[0161] (2) The above fault parameters are injected into three parallel digital twin copies respectively, and the simulation is rerun from the "fault start time" to simulate the virtual data of the past 2 hours under their respective fault assumptions.

[0162] (3) Bayesian inference: The simulated virtual impedance curve and internal humidity were rigorously compared with the actual reported high-frequency sampling data. The calculation results show that the simulation data of H1 (corrosion) matches the measured data by more than 90%, while the matching degree of H2 and H3 is less than 50%.

[0163] In-depth diagnostic results: Final diagnosis generated in the cloud: Precise root cause: Electrochemical corrosion of pin J1 on RS485 port.

[0164] Quantitative status: The current corrosion depth is approximately 120 nm, resulting in an increase in contact resistance of approximately 0.7 Ω. Confidence level: 92%.

[0165] Fault chain: Salt spray intrusion from typhoon → Micro-leakage of sealing ring leading to high internal humidity → Corrosion of J1 pin → Increased contact resistance.

[0166] Remaining service life prediction: Based on the corrosion kinetics model, at the current rate, the port impedance is predicted to reach the functional failure threshold (5Ω) in 15 days. Remaining safe operating time: 10 days (90% confidence interval: 7-15 days).

[0167] Step 5: The rule engine makes a decision and generates executable instructions.

[0168] Rule engine trigger: Input the above deep diagnostic results (root cause, RUL, confidence level) into the rule engine.

[0169] Contextual query: The engine automatically queries: There is a planned maintenance shutdown window for wind turbine #12 within the next two weeks (8 days later); there are spare parts for the RS485 interface module of this model in the warehouse.

[0170] Rule matching and decision-making: Safety decision: RUL (10 days) > planning window (8 days), no immediate safety risk, decision is "planned maintenance".

[0171] Strategy matching: Matching rule "IF (RUL>Schedule window) THEN Schedule maintenance within the schedule window".

[0172] Resource verification: Spare parts are sufficient and tools are available.

[0173] Specific measures are generated: Match the standard operating procedure for "replacing the faulty communication interface module".

[0174] Generate executable instructions: The engine automatically generates a structured maintenance work order and pushes it to the user's system via an interface. The content of the above structured maintenance work order can be: Work order number: XXXX Title: [Planned Maintenance] Replacement of RS485 Communication Interface Module J1 in the 12-fan combiner box DAU Planned downtime: 8 days from now (T+8, 09:00-12:00) Required materials: RS485 interface module (P / N: IF-485-01), sealing ring (P / N: SEAL-2020) Work instructions: Includes detailed disassembly, replacement, and testing steps, and includes a digital twin link that can highlight the J1 position in 3D.

[0175] Verification criteria: After repair, the communication port impedance must be restored to <1.5Ω, and the internal humidity of the chassis must be <70%RH.

[0176] Step 6, Model Evolution (Post-hoc).

[0177] (1) Maintenance execution and verification: Eight days later, the maintenance team completed the replacement according to the work order and uploaded the verification data: the impedance was restored to 1.1Ω and the humidity was reduced to 65%RH.

[0178] (2) Case archiving: The system will store the complete data chain of this “early warning-diagnosis-repair-verification”, the intermediate state of the model, and the maintenance records together to form a high-quality fault diagnosis case.

[0179] (3) Model evolution: Health baseline update: The system will slowly adjust the personalized baseline upper limit of the internal humidity of the chassis from 62%RH to 65%RH to reflect the sealing performance of the new seals.

[0180] Causal model optimization: The time delay parameter of the edge from "salt spray concentration" to "sealing aging" in the prior causal knowledge model was fine-tuned to make it more consistent with the corrosion development rate observed in this study.

[0181] Knowledge base update: The specific timeline of this corrosion development (from abnormal humidity to measurable increase in impedance) has been recorded and used to optimize similar prediction models for DAU of other wind turbines in the same sea area.

[0182] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a monitoring device for the operational status of a data acquisition unit for offshore wind power combiner boxes, the structure of which is as follows: Figure 4 As shown.

[0183] Figure 4 This is a schematic diagram of the internal structure of a data acquisition unit for monitoring the operational status of an offshore wind power combiner box, provided as an embodiment of this application. Figure 4 As shown, the device includes: At least one processor 401; And a memory 402 that is communicatively connected to at least one processor; The memory 402 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 401 so that at least one processor 401 can: execute the above-described method for monitoring the operating status of the offshore wind power combiner box data acquisition unit.

[0184] In one possible implementation, the processor can construct a priori causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit based on the internal structure and operating environment of the unit; instantiate the priori causal knowledge model into a dynamic causal graph corresponding to the unit; sample key nodes in the unit at a first frequency to obtain the baseline operating state of the unit, wherein the key nodes are multi-dimensional physical state parameters defined in the dynamic causal graph to characterize the operating state of the unit; when an anomaly exists in the key nodes, sample target nodes at a second frequency to obtain high-frequency sampling data, wherein the second frequency is higher than the first frequency, and the target nodes are nodes determined based on the dynamic causal graph that are associated with the anomaly-existing key nodes; perform real-time causal reasoning on the high-frequency sampling data based on the dynamic causal graph to obtain a diagnostic result of the operating state of the unit.

[0185] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium stores computer-executable instructions, which are configured to execute the above-mentioned method for monitoring the operating status of the offshore wind power combiner box data acquisition unit.

[0186] In one possible implementation, the aforementioned processor computer-executable instructions are configured to: construct a priori causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit based on its internal structure and operating environment; instantiate the priori causal knowledge model into a dynamic causal graph corresponding to the offshore wind power combiner box data acquisition unit; sample key nodes in the offshore wind power combiner box data acquisition unit at a first frequency to obtain the baseline operating state of the offshore wind power combiner box data acquisition unit, wherein the key nodes are multi-dimensional physical state parameters defined in the dynamic causal graph to characterize the operating state of the offshore wind power combiner box data acquisition unit itself; in the case of an anomaly in the key nodes, sample target nodes at a second frequency to obtain high-frequency sampling data, wherein the second frequency is higher than the first frequency, and the target node is a node determined based on the dynamic causal graph that is associated with the key node that has an anomaly; and perform real-time causal reasoning on the high-frequency sampling data based on the dynamic causal graph to obtain a diagnostic result of the operating state of the offshore wind power combiner box data acquisition unit.

[0187] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0188] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0189] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0190] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0191] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0192] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0193] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0194] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0195] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0196] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0197] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for monitoring the operational status of a data acquisition unit for an offshore wind power combiner box, characterized in that, The method includes: Based on the internal structure and operating environment of the offshore wind power combiner box data acquisition unit, a priori causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit is constructed. The prior causal knowledge model is instantiated as a dynamic causal graph corresponding to the offshore wind power combiner box data acquisition unit; The key nodes in the offshore wind power combiner box data acquisition unit are sampled at a first frequency to obtain the baseline operating state of the offshore wind power combiner box data acquisition unit. The key nodes are multi-dimensional physical state parameters defined in the dynamic causal graph to characterize the operating state of the offshore wind power combiner box data acquisition unit itself. In the event of an anomaly at the critical node, the target node is sampled at a second frequency to obtain high-frequency sampling data, wherein the second frequency is higher than the first frequency, and the target node is a node that is associated with the critical node that is anomaly, as determined based on the dynamic cause-effect graph. Based on the dynamic causal graph, real-time causal reasoning is performed on the high-frequency sampled data to obtain the diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit.

2. The method according to claim 1, characterized in that, Based on the internal structure and operating environment of the offshore wind power combiner box data acquisition unit, a priori causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit is constructed, including: The circuit diagram and / or structural diagram of the offshore wind power combiner box data acquisition unit are extracted to obtain the components corresponding to the offshore wind power combiner box data acquisition unit and the circuit or physical connection relationship between the components. Using the aforementioned components as vertices of the structural layer and the aforementioned circuit or physical connection relationships as directed edges of the structural layer, a structural layer diagram corresponding to the offshore wind power combiner box data acquisition unit is constructed. Based on the knowledge of the operating environment and equipment failure modes of the offshore wind power combiner box data acquisition unit, an environmental failure mapping table corresponding to the offshore wind power combiner box data acquisition unit is established. The operating environment includes specific environmental spectrum data of offshore wind power. The environmental failure mapping table includes the environmental factors, failure modes and observable symptoms caused by the failure modes corresponding to the offshore wind power combiner box data acquisition unit. Based on the environmental fault mapping table and the physical and / or chemical changes involved in the operation of the offshore wind power combiner box data acquisition unit, a causal relationship and influence function library corresponding to the offshore wind power combiner box data acquisition unit are constructed. The causal relationship includes the inducing relationship between the environmental factors and the fault mode and the causal relationship between the fault mode and the observable symptoms. The influence function library includes influence functions, which are calculation formulas that match the causal relationship. Using the physical quantities, component states, environmental factors, and fault modes corresponding to the offshore wind power combiner box data acquisition unit as nodes, and the causal relationships as directed edges, a priori causal knowledge model framework corresponding to the offshore wind power combiner box data acquisition unit is constructed. The physical quantities are used to characterize the measurable parameters within the offshore wind power combiner box data acquisition unit, and the component states are used to characterize the module health status evaluated based on the physical quantity nodes. The physical quantities include: power supply voltage, current, ripple, temperature, vibration, humidity, and communication port impedance. The nodes are configured to belong to the components, and the circuit or physical connection relationships in the structural layer diagram are configured as topological constraints for the directed edges in the priori causal knowledge model framework. Traverse each directed edge in the prior causal knowledge model framework, match and associate the influence functions corresponding to the directed edges from the influence function library, assign initial values ​​to the directed edges, and generate the prior causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit.

3. The method according to claim 2, characterized in that, The instantiation of the prior causal knowledge model into a dynamic causal graph corresponding to the offshore wind power combiner box data acquisition unit includes: Traverse all nodes of the prior causal knowledge model, remove unobservable nodes of the offshore wind power combiner box data acquisition unit that do not have corresponding physical sensors deployed, and remove all directed edges connected to the unobservable nodes to generate an observable causal subgraph. The influence function associated with each directed edge in the observable causal subgraph is simplified; The factory calibration data and health baseline of the offshore wind power combiner box data acquisition unit are injected into the simplified observable causal subgraph to generate a dynamic causal graph for the offshore wind power combiner box data acquisition unit.

4. The method according to claim 2, characterized in that, After constructing the prior causal knowledge model corresponding to the offshore wind power combiner box data acquisition unit, the following steps are included: instantiating the prior causal knowledge model into a digital twin causal model corresponding to the offshore wind power combiner box data acquisition unit; After obtaining the diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit, the process includes: using the digital twin causal model, utilizing the high-frequency sampling data and the diagnostic results, and through multi-hypothesis simulation verification and Bayesian inference, to obtain the in-depth diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit.

5. The method according to claim 4, characterized in that, The instantiation of the prior causal knowledge model into a digital twin causal model corresponding to the offshore wind power combiner box data acquisition unit includes: Replace the influence function associated with multiple directed edges in the prior causal knowledge model with the physical solver corresponding to the influence function; The factory calibration data of the offshore wind power combiner box data acquisition unit is used as prior knowledge and injected into the prior causal knowledge model. The initial healthy operation data of the offshore wind power combiner box data acquisition unit is used to optimize the parameters in the prior causal knowledge model that cannot be directly measured through parameter identification algorithm, so that the simulation output matches the actual measurement data in a healthy state. A data assimilation algorithm for real-time synchronization with the actual state of the offshore wind power combiner box data acquisition unit is integrated into the prior causal knowledge model, and a multi-hypothesis diagnostic reasoning engine is configured in the prior causal knowledge model to obtain a digital twin causal model corresponding to the offshore wind power combiner box data acquisition unit.

6. The method according to claim 1, characterized in that, After sampling key nodes in the offshore wind power combiner box data acquisition unit at a first frequency to obtain the baseline operating status of the offshore wind power combiner box data acquisition unit, the method further includes: If the measured value at the critical node exceeds the preset hardware safety limit threshold, it is determined that there is a safety anomaly in the offshore wind power combiner box data acquisition unit, triggering the preset hardware protection action. If the abnormal pattern present at the critical node matches the preset instantaneous interference pattern library, it is determined that there is instantaneous interference in the offshore wind power combiner box data acquisition unit, and logs are recorded. If the abnormal pattern existing at the critical node matches a single fault pattern in the preset fault knowledge base, the handling suggestion corresponding to the single fault pattern is triggered. The key node is determined to be abnormal if the abnormal pattern of the key node satisfies at least two of the following conditions: persistence, correlation, and predictive value.

7. The method according to claim 1, characterized in that, The diagnostic results of the operational status of the offshore wind power combiner box data acquisition unit obtained by performing real-time causal inference on the high-frequency sampled data based on the dynamic causal graph include: The time-domain and frequency-domain features of the high-frequency sampled data are calculated to generate a feature vector corresponding to the offshore wind power combiner box data acquisition unit, wherein the feature vector includes the feature value of the target node at the current time; Each feature value in the feature vector is compared with the health baseline corresponding to the feature value to obtain the observed abnormal nodes whose difference from the health baseline is greater than a preset threshold, so as to generate a set of evidence of observed abnormalities. Starting from the current state indicated by the feature vector, a reverse search is performed along the causal edges in the dynamic causal graph to obtain multiple candidate cause nodes that can explain this abnormal event. Based on the prior probability of the candidate cause nodes, the posterior probability of the candidate cause nodes causing the current abnormal event is calculated, and the candidate cause nodes are sorted according to the posterior probability to obtain a sorted root cause list. Starting from the observed abnormal nodes in the observed abnormal evidence set, state propagation is carried out along the causal direction in the dynamic causal graph to predict the state change trend of the nodes in the dynamic causal graph after a future time period, and to determine whether the nodes are at risk of exceeding the safety threshold, so as to generate forward prediction results. Based on the sorted root cause list and the forward prediction results, a diagnostic result is generated for the offshore wind power combiner box data acquisition unit.

8. The method according to claim 4, characterized in that, The method of using the digital twin causal model, utilizing the high-frequency sampling data and the diagnostic results, and through multi-hypothesis simulation verification and Bayesian inference, obtains in-depth diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit, including: The high-frequency sampling data is used as an observation value and input into the digital twin causal model. The state of the digital twin causal model is synchronized with the state of the offshore wind power combiner box data acquisition unit during the fault period through a data assimilation algorithm, and the state variables that cannot be directly measured inside the digital twin causal model are estimated. Based on the diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit, a set of possible fault hypotheses are generated. For each fault hypothesis, the parameters corresponding to the fault hypothesis are modified in the digital twin causal model to simulate the fault. The multiphysics transient simulation is then rerun to obtain the virtual observation data corresponding to the fault hypothesis. For each of the fault hypotheses, the matching likelihood between the virtual observation data and the high-frequency sampling data is calculated. Combined with the prior probability of occurrence of the fault hypothesis, the posterior probability that the fault hypothesis is the true root cause is calculated using Bayes' theorem. The fault hypothesis with the highest posterior probability is selected as the precise root cause result, and the digital twin causal model is used to analyze the precise root cause diagnosis result to obtain the in-depth diagnosis result of the operating status of the offshore wind power combiner box data acquisition unit.

9. The method according to claim 8, characterized in that, After obtaining the in-depth diagnostic results of the operating status of the offshore wind power combiner box data acquisition unit, the method further includes: Obtain maintenance resource information associated with the offshore wind power combiner box data acquisition unit; Using a rule engine, based on the deep diagnostic results and the maintenance resource information, the executable instructions corresponding to the offshore wind power combiner box data acquisition unit are determined, wherein the executable instructions are used to resolve this abnormal event; The complete data of multiple abnormal events are linked and stored to form tagged fault diagnosis cases; Perform trend analysis on the long-term operating data of the key nodes to identify the slow drift of the key nodes due to aging, and adjust the health baseline corresponding to the key nodes. Based on the operational data of the offshore wind power combiner box data acquisition unit within the sliding time window, the statistical correlation between nodes in the prior causal knowledge model is recalculated to update the weights and time delay parameters of the directed edges in the dynamic causal graph and the digital twin causal model. Monitor and analyze abnormal patterns that are not explained by the prior causal knowledge model. When the abnormal pattern is detected, add, delete and optimize the local structure of the prior causal knowledge model and update it synchronously to the dynamic causal graph and the digital twin causal model.

10. A monitoring device for the operational status of a data acquisition unit for an offshore wind power combiner box, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a method for monitoring the operating status of a data acquisition unit for offshore wind power combiner boxes as described in any one of claims 1-9.

Images