On-line monitoring method and device for state of current collection system of mountain power station

Abstract

The invention provides a mountain power station current collection system state online monitoring method and device, and relates to the field of data processing. According to the method, distributed optical fiber data, environmental meteorological data and electrical measurement data are acquired to construct a multi-mode time sequence matrix, an operation state diagram is formed through graph structure representation and graph signal processing, a working condition mode cluster is obtained by combining dynamic clustering, a constraint diagram optimization framework is input, thermal, electrical and mechanical constraints are introduced, and an operation state track is generated. And constructing and fusing a non-stationary degradation path library and environment exposure history to form a dynamic degradation evolution model, and finally outputting in-service reliability and predicting residual life by a risk inference engine to realize online monitoring of the state of the current collection system of the mountain power station. By implementing the technical scheme provided by the invention, the accuracy of online monitoring of the state of the current collection system of the mountain power station is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a mountain power station power collection system state online monitoring method and device. BACKGROUND

[0002] Under the background of global energy transformation and low-carbon development, mountain power stations, as an important form of clean energy utilization, are widely used in wind energy, water energy, photovoltaic and other multi-energy complementary scenarios.

[0003] In the actual application of online monitoring of the state of the mountain power station power collection system, the traditional state estimation model based on the steady-state assumption usually relies on fixed parameters and linear assumptions for deduction. However, in the mountain power station scenario, the operating conditions are jointly affected by microclimate mutations, extreme weather disturbances, and geological environmental changes, making it difficult for the state estimation model based on the steady-state assumption to accurately capture the real dynamic characteristics, resulting in significant mismatch between the estimation results and the actual state, and low monitoring accuracy.

[0004] Therefore, there is an urgent need for a mountain power station power collection system state online monitoring method and device. SUMMARY

[0005] The present application provides a mountain power station power collection system state online monitoring method and device, which facilitates improving the accuracy of online monitoring of the state of the mountain power station power collection system.

[0006] In a first aspect of the present application, a mountain power station power collection system state online monitoring method is provided, the method comprising: acquiring distributed optical fiber data, environmental meteorological data and electrical measurement data for the mountain power station power collection system, and constructing a multi-modal time series matrix according to the distributed optical fiber data, environmental meteorological data and electrical measurement data; constructing a graph structure representation based on the multi-modal time series matrix, and processing the graph structure representation through a graph signal processing algorithm to obtain a structured operating state graph; performing dynamic clustering on the operating state graph to obtain a working condition mode cluster; inputting the operating state graph and the working condition mode cluster into a constrained graph optimization framework, introducing thermal constraints, electrical constraints and mechanical constraints in the optimization process to obtain an operating state trajectory; constructing a non-stationary degradation path library that changes with time and working conditions according to the operating state trajectory, and fusing the non-stationary degradation path library with environmental exposure history to form a dynamic degradation evolution model; inputting the dynamic degradation evolution model into a risk reasoning engine to generate in-service reliability and predicted remaining life, and completing online monitoring of the state of the mountain power station power collection system based on the in-service reliability and the predicted remaining life.

[0007] In a second aspect of the present application, a mountain power station power collection system state online monitoring device is provided, the device comprising an acquisition module and a processing module, wherein the acquisition module is configured to acquire distributed optical fiber data, environmental meteorological data and electrical measurement data for a mountain power station power collection system, and construct a multi-modal time series matrix based on the distributed optical fiber data, the environmental meteorological data and the electrical measurement data; the processing module is configured to construct a graph structure representation based on the multi-modal time series matrix, and process the graph structure representation through a graph signal processing algorithm to obtain a structured operating state graph; the processing module is further configured to perform dynamic clustering on the operating state graph to obtain a working condition mode cluster; the processing module is further configured to input the operating state graph and the working condition mode cluster into a constrained graph optimization framework, and introduce thermal constraints, electrical constraints and mechanical constraints in the optimization process to obtain an operating state trajectory; the processing module is further configured to construct a non-stationary degradation path library that varies with time and working condition based on the operating state trajectory, and fuse the non-stationary degradation path library with an environmental exposure history to form a dynamic degradation evolution model; the processing module is further configured to input the dynamic degradation evolution model into a risk reasoning engine to generate an in-service reliability and a predicted remaining life, and complete online monitoring of the mountain power station power collection system state based on the in-service reliability and the predicted remaining life.

[0008] In a third aspect of the present application, an electronic device is provided, the electronic device comprising a processor, a memory, a user interface and a network interface, the memory being configured to store instructions, the user interface and the network interface both being configured to communicate with other devices, and the processor being configured to execute the instructions stored in the memory to enable the electronic device to perform the method described above.

[0009] In a fourth aspect of the present application, a computer-readable storage medium is provided, the computer-readable storage medium storing instructions which, when executed, perform the method described above.

[0010] In summary, one or more technical solutions provided in the present application have at least the following technical effects or advantages: By synchronous acquisition and fusion of distributed optical fiber data, environmental meteorological data and electrical measurement data, a multi-modal time series matrix covering thermal, electrical and mechanical dimensions can be constructed to realize all-around perception of the operation state of the power collection system of the mountain power station and avoid information loss and deviation caused by single sensing data. By using graph structure representation and graph signal processing algorithms to process the multi-modal time series matrix, the operation state data has spatial topological constraints and time continuity characteristics, so that the coupling relationship between devices and sections can be more accurately expressed, and the precision of abnormal detection and state recognition is improved. By dynamically clustering to extract working condition mode clusters, different operating modes in complex mountain environments can be adaptively identified, and the stability of the monitoring results can be maintained during mode switching, avoiding model failure caused by environmental disturbance and load mutation. By introducing a constraint graph optimization framework with thermal, electrical and mechanical constraints, the physical laws are combined with operation data, so that the estimation results of the state trajectory not only depend on the observation data, but also meet the requirements of heat conduction, electrical continuity and mechanical stability, improving the credibility and physical interpretability of the operation state estimation. By constructing a non-stationary degradation path library and fusing environmental exposure history, the non-stationary evolution characteristics of life-related variables in time and working conditions can be accurately captured, making the degradation modeling results more consistent with the actual complex operating environment, overcoming the limitations of traditional steady-state assumption models. By inputting the dynamic degradation evolution model into the risk reasoning engine, the in-service reliability and predicted remaining life can be output, realizing quantitative evaluation of the health degree and potential failure risk of the equipment, providing a reliable basis for preventive maintenance and operation strategy, and finally realizing intelligent, dynamic and traceable online monitoring of the power collection system of the mountain power station. Therefore, the accuracy of online monitoring of the power collection system of the mountain power station is improved. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 A flowchart of a mountain power station power collection system state online monitoring method provided by an embodiment of the present application is shown in the figure. Figure 2 A module diagram of a mountain power station power collection system state online monitoring device provided by an embodiment of the present application is shown in the figure. Figure 3 A structural diagram of an electronic device provided by an embodiment of the present application is shown in the figure.

[0012] Explanation of reference numerals: 21, acquisition module; 22, processing module; 31, processor; 32, communication bus; 33, user interface; 34, network interface; 35, memory. DETAILED DESCRIPTION

[0013] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be clearly and completely described below in conjunction with the drawings in the specification. Obviously, the described embodiments are only some of the embodiments of the present application, not all.

[0014] In the description of the embodiments of the present application, the words such as "for example" or "for instance" are used to represent an example, illustration or description. Any embodiment or design scheme described as "for example" or "for instance" in the embodiments of the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the words such as "for example" or "for instance" are intended to present the relevant concept in a specific manner.

[0015] In the description of the embodiments of the present application, the term "a plurality of" means two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. In addition, the terms "first" and "second" are used for description purposes only and should not be interpreted as indicating or implying relative importance or implicitly indicating the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more features. The terms "include", "contain", "have" and their variants mean "include but are not limited to", unless otherwise specifically emphasized.

[0016] To solve the above technical problems, the present application provides a mountain power station power collection system state online monitoring method, referring to Figure 1 , Figure 1 A flowchart of a mountain power station power collection system state online monitoring method provided by the embodiments of the present application. The method is applied to a server and includes steps S110 to S160, which are as follows: S110, acquiring distributed optical fiber data, environmental meteorological data and electrical measurement data for the mountain power station power collection system, and constructing a multi-modal time series matrix according to the distributed optical fiber data, the environmental meteorological data and the electrical measurement data.

[0017] Specifically, the server refers to a cluster of computing devices that undertake data reception, storage, calculation and scheduling functions. In the scene of the mountain power station power collection system, it is used to uniformly access distributed optical fiber data, environmental meteorological data and electrical measurement data, perform time synchronization, spatial registration and quality control, and provide data services and analysis task arrangement externally; for example, the data center server deployed at the station control layer is responsible for long-period data archiving and model calculation, and the edge type server deployed at the power collection line junction point is responsible for high-frequency data buffering and event triggered backhaul.

[0018] The power collection system of a mountain power station refers to the collection of power lines and equipment that collect the power generated by wind turbines, photovoltaic arrays, or small hydropower units and send it to the booster station. It includes cables, junction boxes, ring network cabinets, box-type transformers, towers, and connectors. It has the characteristics of long lines, large drop, multiple angles, and strong environmental load. For example, in a mountainous photovoltaic and wind power hybrid station, multiple branches are connected to the same box-type transformer through medium-voltage cables, and then sent to the booster station through the power collection line. Distributed optical fiber data refers to continuous spatial resolution monitoring data generated by distributed optical fiber sensing along the cable sheath or trench. It usually includes temperature channels, strain channels, mileage coordinates, quality identifiers, and event annotations, which are used to represent the spatial and temporal changes in the thermal and mechanical state of the cable along the line. For example, a short-term temperature jump and a small increase in sheath strain are detected near the K23+400 mileage post, which can be used to locate potential joint heating or icing-induced mechanical stress concentration.

[0019] Environmental meteorological data refers to external environmental observation data related to the microclimate of the power collection line. It usually includes wind speed, wind direction, air temperature, relative humidity, rainfall, irradiance, ice thickness, ground temperature, and lightning activity frequency, and is accompanied by site coordinates and data quality identifiers. It is used to explain the external driving of the equipment state. For example, the increase in ice thickness during periods of strong wind and low temperature can be correlated with temperature anomalies and strain anomalies in distributed optical fiber data. Electrical measurement data refers to electrical quantity measurement data collected by transformers and online monitoring devices. It usually includes three-phase current, three-phase voltage, active power, reactive power, power factor, phase angle, frequency, harmonic index, joint temperature monitoring, and partial discharge count, and is accompanied by loop identifier and topology location. It is used to describe the electrical operating load and insulation health status. For example, a certain loop exhibits rapid power fluctuations around sunset accompanied by harmonic rise, which, together with the slow rise in temperature from distributed optical fiber data, indicates potential inadequate heat dissipation.

[0020] Multi-modal time series matrix refers to a structured data matrix indexed by time and with column sets as distributed optical fiber data channel clusters, environmental meteorological data channel clusters, and electrical measurement data channel clusters. Each row is accompanied by spatial coordinates, topology location, data source, and quality identifier metadata, meeting the input requirements of rolling window calculation, event playback, and graph structure construction. For example, at the same time step, the matrix contains temperature and strain values at K23+400, wind speed and ice thickness at the corresponding weather station, three-phase current and harmonic index of the corresponding loop, and carries a "high confidence" quality identifier, which is used to construct the operating state graph and execute the graph signal processing algorithm.

[0021] Furthermore, in the process of collecting distributed optical fiber data along the power collection line using distributed optical fiber, the distributed optical fiber is continuously laid between the sheath and shielding layer of the power collection line to complete the connection between the distributed optical fiber and the demodulation device, the photoelectric conversion device and the data acquisition device. A mileage index along the route is established as a spatial coordinate, and continuous sampling of the distributed optical fiber is triggered with a unified time reference to form a distributed optical fiber data channel. The distributed optical fiber data channel includes a temperature channel and a strain channel. At the same time, the signal quality returned by the demodulation device is measured and the optical fiber data quality identifier is calculated to indicate availability. The optical fiber data quality identifier is generated according to the following formula and written into the data stream over time. The specific formula is as follows:

[0022] in, Mileage location With time Fiber optic data quality identifier, This is the normalized index for signal-to-noise ratio. The residual normalization index, For missing measurement indication, These are non-negative weighting coefficients. This calculation, based on the principles of signal-to-noise ratio enhancement, residual suppression, and missing measurement penalty, maps sampling stability to a quality identifier; the distributed fiber optic data channel and the fiber optic data quality identifier serve as inputs for the next step of time synchronization and spatial registration.

[0023] In the process of collecting environmental meteorological data through meteorological acquisition equipment, local automatic weather stations and available telemetry meteorological sources are configured. Wind speed, wind direction, temperature, relative humidity, rainfall, irradiance, and icing thickness are selected as standard fields for environmental meteorological data. All environmental meteorological data are sampled using a unified time reference and accompanied by the spatial coordinates of the meteorological stations. At the same time, representative weighting is established to match the spatial coordinates of the power collection lines to generate representative environmental meteorological data along the line. The representative weighting is calculated by the following formula and used for weighted aggregation:

[0024] in, Mileage location For the first The representative weight of meteorological stations Mileage location With weather stations Spatial distance, For distance attenuation scale, This represents the angle between the route alignment and the station's orientation. A weighted average is used to generate the location. Representative environmental meteorological data are used to improve spatial matching; the representative environmental meteorological data and their weights are used as inputs for the next step of time synchronization and spatial registration.

[0025] In the process of collecting electrical measurement data through current transformers, voltage transformers, temperature probes, partial discharge monitoring devices, and box-type transformer monitoring devices, a topological index is established for each device to identify its circuit, joint, and tower location. Continuous sampling is performed using a unified time reference. The electrical measurement data includes fields such as three-phase current, three-phase voltage, active power, reactive power, power factor, phase angle, frequency, harmonic index, joint temperature, and partial discharge count. To map the device measurements to locations along the route, a topological mapping matrix from device to location is constructed. Multiple device measurements within the same mileage neighborhood are weighted and aggregated. The aggregated load vector is defined by the following formula to generate mileage-time electrical measurement data:

[0026] in, Mileage location With time Aggregated electrical measurement data vector, These are elements of the topology mapping matrix, representing device nodes. Mileage location Contribution weight, For device nodes At any moment The electrical measurement data vector is aggregated based on topological and physical adjacency to form electrical measurement data consistent with the distributed fiber optic data on the mileage coordinates. The aggregated electrical measurement data and the device topology index serve as inputs for the next step of time synchronization and spatial registration.

[0027] In the implementation of time synchronization and spatial registration based on a unified time reference, and attaching spatial coordinates, topology indexes, and quality identifiers at the same time step to form a multimodal time series matrix, a unified time axis is first established and a target sampling interval is set. Resampling and alignment are performed on distributed fiber optic data, representative environmental meteorological data, and aggregated electrical measurement data. Resampling employs a consistent interpolation and fragmentation strategy that preserves event boundaries. Simultaneously, alignment is performed spatially based on mileage and topology indexes to ensure a one-to-one correspondence between the three types of data in terms of mileage location and equipment relationship. For each time step, a row vector containing distributed fiber optic data channels, environmental meteorological data channels, and electrical measurement data channels is constructed, along with spatial coordinates, topology indexes, and cross-source quality identifiers. The cross-source quality identifiers are summarized using the following formula to reflect the overall availability of that time step:

[0028] in, Mileage location With time Cross-source quality identification, For fiber optic data quality identification, As a quality indicator for environmental meteorological data, For electrical measurement data quality identification, These are non-negative weighting coefficients used to reflect the importance of each channel under the target operating condition. All time steps and all mileage positions are stacked using time as the primary index to obtain a multimodal time series matrix that meets the input requirements for subsequent graph structure representation construction and graph signal processing algorithms.

[0029] S120. Construct a graph structure representation based on a multimodal time-series matrix, and process the graph structure representation using a graph signal processing algorithm to obtain a structured running state graph.

[0030] Specifically, a graph structure is a data structure composed of a set of nodes, a set of edges, adjacency relationships, a node attribute matrix, and an edge attribute matrix. It is used to explicitly encode the spatial topology and coupling relationships between devices, segments, and loops, and to carry observational attributes that change over time. For example, the node set contains entities such as "joint J12, tower P45, and box-type transformer B03", the edge set contains "the physical topology edge from J12 to P45, the spatial proximity edge from P45 to P46, and the cross-domain coupling edge between temperature and current", the node attribute matrix records the distributed fiber temperature, distributed fiber strain, and joint temperature of each node within a time window, and the edge attribute matrix records the synchronization consistency and cooperative change intensity of adjacent nodes.

[0031] Graph signal processing algorithms refer to a family of algorithms that treat node and edge attributes as signals defined on a graph. They utilize the graph's adjacency relationships to preprocess, denoise, smooth, enhance, correct for consistency, and identify anomalies. These algorithms can reduce measurement jitter while preserving event boundaries and alarm markers, and can also determine source priority in cross-source conflicts. For example, constrained smoothing of distributed fiber temperature along physical topology edges can suppress isolated spikes while preserving abrupt changes near joints; consistency enhancement can be applied to the synergistic relationship between "temperature increase and current increase" based on cross-domain coupling edges; and deweighting propagation can be performed on node attributes with lower quality ratings to reduce their impact on overall graph judgment.

[0032] A structured operational state graph refers to a graph data object that has been processed and solidified within the current time sliding window. It includes a set of nodes, a set of edges, a node attribute matrix, an edge attribute matrix, an anomaly label set, multi-scale mapping relationships, and window metadata. It can be directly used as standardized input for dynamic clustering, constraint graph optimization, and degradation modeling. For example, a structured operational state graph for a certain time window shows that "the distributed fiber temperature in the section where connector J12 is located is rising synchronously with the connector temperature and is consistent with the increase in upstream load." The edge set indicates the connectivity between this section and adjacent sections, the anomaly label set marks the location reference of "suspected heat dissipation limitation," and retains a traceable mapping with the rows and columns of the multimodal time series matrix.

[0033] Furthermore, in the process of generating nodes from the multimodal time series matrix according to spatial coordinates, topological location, and device identifier, records with the same spatial location and the same device identifier are first aggregated based on the spatial coordinates and topological index attached to the row to obtain a node set. A node attribute vector and node metadata are established for each node. The node attribute vector is formed by concatenating the alignment values ​​of the distributed optical fiber data channel, environmental meteorological data channel, and electrical measurement data channel at the same time step in a predetermined order, and is then weighted and standardized using quality identifiers to suppress the influence of low-confidence channels. The node metadata synchronously records spatial coordinates, topological location, device identifier, data source identifier, and version identifier. The output node index table and node attribute matrix serve as inputs for subsequent edge generation. The specific formula is as follows:

[0034] in, For nodes At any moment The node attribute vector, This is a distributed fiber optic data channel vector. This is the channel vector for environmental meteorological data. For electrical measurement data channel vectors, For each channel, a quality identifier vector is provided. This indicates a decentralized and standardized approach that uses quality identifiers as the basis for authority.

[0035] In the process of generating edges based on the physical topology, spatial proximity, and energy coupling relationships of power collection lines, a set of physical topology edges is constructed based on loop connection relationships, a set of spatial proximity edges is constructed based on spatial coordinate proximity relationships, and a set of energy coupling edges is constructed based on cross-domain statistical correlations such as temperature and current, and temperature and wind speed. Edge weights are calculated for each edge to quantify connectivity strength and written into the edge attribute matrix. Simultaneously, the edge type and generation basis are registered in the edge metadata. The output edge index table and edge attribute matrix serve as inputs for sliding window assembly. The specific formulas are as follows:

[0036] in, For nodes With nodes The right to the border, For physical topological adjacency indication, The spatial distance between nodes. For distance attenuation scale, The correlation strength of cross-domain coupling, These are non-negative weighting coefficients.

[0037] In the implementation of dividing the multimodal time series matrix into sliding windows and constructing a graph structure representation including a node index table, an edge index table, a node attribute matrix, and an edge attribute matrix, the window length and sliding step size are first set according to a unified time base. Then, the node attribute matrix and the edge attribute matrix are sliced ​​within each window while maintaining index consistency, resulting in a time-ordered graph structure representation sequence. The start and end times, data coverage, and synchronization quality indicators are recorded for each window. The graph structure representation sequence serves as the input to the graph signal processing algorithm, as shown in the following formula:

[0038] in, For the first A set of time indices for a sliding window. The sliding step size, This is the window length.

[0039] In the implementation process of performing node attribute verification, anomaly labeling propagation, cross-source conflict adjudication and local repair according to graph signal processing algorithms, and performing graph domain smoothing and consistency correction on node attributes based on edge attributes to form a denoised graph structure representation, firstly, quality labels and anomaly labels are used to reduce the weight of suspicious dimensions and propagate them to the first-order neighborhood. Then, cross-source conflicts are arbitrated based on source credibility. Subsequently, graph domain smoothing is performed on node attributes in the current window to suppress isolated noise. At the same time, consistency correction is performed on inconsistent observations based on edge weights. The output graph structure representation with significantly reduced noise and improved edge-node compatibility is shown in the following formula:

[0040] in, This is a matrix of node attributes within the window. This is the node attribute matrix after graph domain smoothing. It is the identity matrix. For balance coefficient, The Laplace matrix of an undirected graph. Let be the edge weight matrix. The degree matrix is ​​used; this processing follows the principles of preserving boundaries, suppressing noise, and enhancing consistency.

[0041] In the process of constructing multi-scale views at the spatial and topological levels based on the denoised graph structure representation to form a family of multi-scale graph structure representations that can be locally magnified and viewed globally, the following steps are taken: First, upward aggregation relationships and downward mapping relationships are constructed according to the hierarchical principle of towers, segments, and loops. Then, node attributes and edge attributes are hierarchically summarized according to the aggregation weight to obtain multi-scale views at the measurement point level, tower level, segment level, and loop level. Each view retains a traceable mapping table for backtracking and positioning. The specific formula is as follows:

[0042] in, For aggregation nodes Hierarchical attribute vectors, The aggregated set of underlying nodes. For nodes The denoised attribute vector, For aggregate weights, For node quality identification, For node degree, The exponential parameter is used to suppress high connection bias.

[0043] In the process of performing anomaly identification and temporal consistency processing on a multi-scale graph structure representation family, generating anomaly label sets and solidifying them into a structured operational state graph containing node sets, edge sets, attribute matrices, anomaly labels, and multi-scale mapping relationships, candidate anomalies are first generated based on neighborhood prediction residuals and hierarchical consistency deviations. Then, consistency is performed according to temporal continuity and mode switching boundaries to remove duplicates and false triggers. Subsequently, confirmed anomalies are written back to the diagnostic fields of the node attribute matrix and edge attribute matrix, and solidified together with the multi-scale mapping relationships into a structured operational state graph, serving as standardized input for dynamic clustering and constraint graph optimization. The specific formula is as follows:

[0044] in, For nodes At any moment Abnormal scores, For nodes The neighborhood group, For border rights, For nodes Robust scaling estimation of residual distribution; this metric identifies local anomalies based on neighborhood consistency and ensures the stability and traceability of labels on the time axis through temporal consistency.

[0045] S130. Perform dynamic clustering on the operating state diagram to obtain the operating condition pattern cluster.

[0046] Specifically, dynamic clustering refers to an adaptive clustering process that performs online incremental clustering on the graph embedding representation generated by the operating state map based on a time sliding window, and recursively updates the cluster centers, cluster scale, and cluster members as new data arrives. It includes sample assignment updates, cluster merging, cluster splitting, and hysteresis control, used to maintain the stability and traceability of operating condition identification under non-stationary conditions. Unlike one-time static clustering, dynamic clustering does not immediately generate jitter when mode switching occurs, but filters short-term disturbances through hysteresis control and time continuity constraints. For example, before and after a strong wind, dynamic clustering incorporates the graph embedding representation of the operating state map within three consecutive windows into the same candidate cluster and slowly moves the cluster center; when the icing thickness and distributed fiber strain increase synchronously over a longer residence time, the algorithm triggers cluster splitting, forming new icing-related clusters and recording the switching boundary.

[0047] Operating condition clusters refer to stable cluster sets output by dynamic clustering. Each cluster represents an operating condition mode with consistent statistical characteristics and physical meaning in time and space, and includes a cluster center, cluster boundary, dwell time, label profile, and a list of switching events. These clusters provide a priori domains for subsequent constraint graph optimization and degradation modeling. Operating condition clusters can correspond to "high humidity environment mode," "icing environment mode," "high wind environment mode," "high radiation environment mode," or "load switching mode," etc., and aggregate statistics are provided on the abnormal label distribution of nodes within the cluster and the consistency strength of edges. For example, a certain section is classified into the "icing environment mode" operating condition cluster during two weeks of overlapping low nighttime temperatures and high humidity. The cluster center exhibits high icing thickness, slow increase in distributed fiber strain, and a distributed fiber temperature plateau due to limited heat dissipation. The cluster boundary and mode switching events are used to guide the weighting of thermal constraint factors and robust handling of transition domains in subsequent constraint graph optimization.

[0048] Furthermore, in the process of constructing a time-consistent graph feature set and generating a graph embedding representation by combining the node attribute matrix and edge attribute matrix in the operational state graph, the structured operational state graph output from the previous stage is used as input. For each time sliding window, distributed fiber temperature, distributed fiber strain, three-phase current, three-phase voltage, and quality identifiers are aggregated at the node layer, and spatial proximity strength, physical topology connectivity strength, and energy coupling strength are aggregated at the edge layer. A time-consistent graph feature set is formed through time alignment. For each node, node features and neighborhood weighted features are spliced ​​in a fixed channel order and quality-weighted normalization is performed. Then, a graph embedding representation is obtained using linear projection or a shallow encoder. The graph embedding representation maintains a one-to-one correspondence with the node index and edge index and retains a traceable mapping relationship. The specific formula is as follows:

[0049] in, For nodes At any moment Graph embedding representation, To embed the projection matrix, For node attribute vectors, For nodes The neighborhood group, For border rights, This is a statistical feature of local consistency and energy coupling constructed based on the edge attribute matrix. This processing obtains a robust graph embedding representation by uniformly encoding local observations and neighborhood cooperative information in the same time step.

[0050] In the process of offline clustering of graph embedding representations within the cold start window to obtain the initial candidate set of operating condition modes, all graph embedding representations within the cold start window are used as input. Clustering based on squared error objective is used to solve for cluster centers, and the number of clusters is determined according to the model selection criterion. The output is an initial candidate set of operating condition modes containing cluster centers, cluster scales, and cluster member sets. The specific formula is as follows: ; ;

[0051] in, For the first Graph embedding representation of each sample, For the first Cluster center, For the sample Cluster identifier, For the sample size, For selecting the number of clusters Bayesian information criterion, For Gaussian likelihood, For shared estimation of intra-cluster covariance, The above steps extract the stable statistical structure within the cold start window into a candidate set of initial operating conditions, representing the model's degrees of freedom.

[0052] Based on the initial candidate set of operating conditions, the graph embedding representation of the current time sliding window and the clustering result of the previous time sliding window are jointly input and online incremental clustering is performed to obtain the temporary set of operating conditions. During this process, the cluster centers and cluster scales of the previous time sliding window are used as priors, and the graph embedding representation of the current time sliding window is used for soft affiliation updates, and the cluster centers and cluster scales are updated recursively. At the same time, the continuity of cluster labels is maintained to reduce short-term jitter. The specific formula is as follows: ; ;

[0053] in, For the sample At any moment Belongs to cluster The soft attribution probability, The graph embedding representation of the current time sliding window. and This provides an estimate of the cluster centers and intra-cluster variances for the previous time-sliding window. To be according to Weighted in-cluster sample mean, To be according to Weighted within-cluster sample variance The coefficients are updated recursively; this process outputs a set of temporary operating conditions and preserves the cluster evolution trajectory.

[0054] In the process of performing merging and splitting operations on the temporary operating condition pattern set to generate a standardized operating condition pattern set and pattern dictionary, the inter-cluster center distance and intra-cluster scale are first calculated. If the center distance between adjacent clusters is less than the merging threshold set according to the cluster scale, merging is performed. If the intra-cluster scale exceeds the splitting threshold and exhibits bimodality or strong heterogeneity, splitting is performed. After completion, a unique identifier is assigned to each operating condition pattern, and the center description and residence statistics are recorded to form a searchable pattern dictionary. The specific formula is as follows: ; ;

[0055] in, For clusters with cluster The center distance, For clusters Intra-cluster scale, The number of samples within the cluster. For clusters The sample set, The Davies-Bouldin index is used to evaluate cluster separability; the merging and splitting are stably completed through the joint criterion of distance and separability, and a normalized set of case patterns and a pattern dictionary are generated.

[0056] In the process of identifying mode switching events based on a pattern dictionary and setting hysteresis control to suppress jittery switching, the cluster affiliation sequence of the continuous time sliding window is smoothed by voting. Mode switching is only confirmed when the proportion of the new cluster within the hysteresis window reaches a threshold. The confirmed mode switching event is written into the boundary time and effective time list as a priori boundaries for subsequent optimization and modeling. The specific formula is as follows:

[0057] in, For a moment Instantaneous cluster prediction, For clusters At any moment The probability of belonging, The final cluster sequence after hysteresis control. The length of the lag window. To confirm the threshold, This is an indicator function; this mechanism effectively suppresses short-term jitter switching under non-stationary conditions.

[0058] During the implementation process of standardizing the set of working condition models and solidifying the results to output the working condition model cluster, consistency checks are performed on the cluster labels of adjacent time sliding windows, including the overlap of sample sets across windows, the drift amplitude of cluster centers, and the continuity of the model dictionary. When the consistency index reaches the threshold, the working condition model cluster of the current window is solidified and the mode switching event is registered; otherwise, local recalculation and threshold callback are triggered. Finally, the working condition model cluster used for subsequent constraint graph optimization and non-stationary degradation path library construction is output. The specific formula is as follows:

[0059] in, For clusters The overlap of samples in adjacent time sliding windows For clusters The center of the cross-window drift amplitude, and These are the cluster sample sets for the current and previous time sliding windows, respectively. By combining the constraints of overlap and drift amplitude, the stability of the operating condition pattern cluster in time and the interpretability in space are ensured, and a searchable, traceable, and usable final result for downstream modeling is formed.

[0060] S140. Input the running state diagram and the operating condition mode cluster into the constraint diagram optimization framework, and introduce thermal constraints, electrical constraints and mechanical constraints during the optimization process to obtain the running state trajectory.

[0061] Specifically, the constraint graph optimization framework takes the operating state graph and the cluster of operating modes as input, and simultaneously applies thermal, electrical, and mechanical constraints within a unified graph structure. Through robust optimization, it obtains physically consistent state estimates for each node and segment, and generates continuous outputs that slide over time. This framework receives the node set, edge set, node attribute matrix, edge attribute matrix, and mode priors. It assigns higher weights to time slices in stable dwell modes and robustly suppresses time slices crossing mode boundaries. For example, within the transition window from strong winds to calm winds, the framework reduces the influence of uncertain segments, prioritizing the convergence of stable segments.

[0062] Thermal constraints refer to physical consistency limitations established based on heat conduction paths, current-carrying heat generation, and heat dissipation boundary conditions. These constraints limit the spatial distribution and temporal evolution of temperature rise along the cable path, ensuring that temperature estimates conform to cable structure, joint locations, and environmental heat transfer conditions. Constraints focus on the hot ends of box-type transformers, weak points in joint heat dissipation, and insufficiently cooled sections shielded by terrain. For example, when wind speed drops sharply and irradiance increases, thermal constraints limit the temperature gradient within the same section and the temperature continuity between adjacent sections, preventing non-physical temperature peaks caused by instantaneous noise.

[0063] Electrical constraints refer to electrical rationality limitations established based on loop continuity, power balance, and phasor consistency. They are used to constrain the correlation between current, voltage, active power, reactive power, and phase angle, ensuring that load changes in a section are consistent with upstream bus measurements, and that joint contact performance estimates are consistent with signs of localized heating. For example, if the upstream three-phase current increases after a load switch but the section power does not change proportionally, electrical constraints will cause the estimation to backtrack to topology or measurement anomalies, rather than misjudging them as temperature anomalies.

[0064] Mechanical constraints refer to the reasonable mechanical limitations established based on strain compatibility, displacement limits, and structural stability. These constraints are used to regulate the distribution of sheath strain at adjacent measuring points, tower spans, and corner sections, limiting large curvature corners and span displacements to acceptable ranges, and ensuring consistency across the domain with icing thickness and gust loads. For example, under the combined effects of icing and crosswinds, mechanical constraints require that the increase in strain at corner sections remain continuous with that of adjacent spatial sections, avoiding abnormal peak values ​​at isolated measuring points that violate structural compatibility.

[0065] Furthermore, when using the mode identifier, dwell time, and switching boundary provided by the operating mode cluster as prior information for domain partitioning, time indicators for the stable constraint domain and the transitional constraint domain are first generated on the time axis. Then, a stability weighting factor and a boundary proximity weighting factor are calculated for each time step, ensuring that subsequent constraint factors are strengthened in the stable constraint domain and suppressed in the transitional constraint domain. The stability weighting factor and the boundary proximity weighting factor can be defined as follows:

[0066] in, For a moment The stable weighting factor For stability sensitivity coefficient, For a moment Duration of stay under the operating conditions This is the threshold for dwell time. For a moment The neighbor weighting factor of the relative mode switching boundary, For a moment Time distance to the nearest switching boundary The boundary effect attenuation scale. and The combination of these factors determines the weight mask for the stable constraint domain and the transition constraint domain, and the weight mask is written into the window metadata of the running state graph, which is used as the time-weighted input for subsequent constraint factors.

[0067] When constructing thermal, electrical, and mechanical constraint factors within the stable and transitional constraint domains, respectively, residual terms consistent with the node attribute matrix and edge attribute matrix are defined around the distributed fiber optic temperature channel, box-type transformer temperature channel, load event annotation channel, current channel, voltage channel, joint contact performance channel, loop topology, sheath strain channel, spatial coordinate mapping, and tower segment structural information. Exemplary thermal, electrical, and mechanical residuals can be written as:

[0068] in, For nodes At any moment Thermal residuals This refers to the node temperature of the distributed fiber optic temperature channel. For ambient temperature in environmental meteorological data, Let be the node current of the current path. The equivalent coefficient for current-carrying heating. The neighborhood temperature is obtained by weighting the neighbor temperatures based on the edge attribute matrix. is the equivalent coefficient of heat transfer in the neighborhood.

[0069]

[0070] in, For the edge At any moment Electrical residuals The measured active power is obtained from electrical measurement data. The voltage at the upstream node. For edge current, Let be the phase angle between the current and voltage at the edge.

[0071]

[0072] in, For nodes At any moment Mechanical residuals For the nodal strain of the sheath strain channel. The equivalent coefficient for thermally induced strain is... This represents the increase in node temperature relative to ambient temperature. The neighborhood strain is obtained by weighting the neighbor strains based on spatial coordinate mapping and tower segment structural information. The strain-compatible equivalent coefficients are used. The three types of residuals mentioned above are respectively bound to the factor nodes of the thermal constraint factor, electrical constraint factor, and mechanical constraint factor, and inherit the time weights of the stable constraint domain and the transition constraint domain.

[0073] When assembling a sparse constraint graph based on thermal, electrical, and mechanical constraint factors, as well as node and edge attribute matrices, and solving for intermediate running state variables, the node-level and edge-level states to be estimated are uniformly written into the state vector. A weighted objective function is constructed using a time sliding window, and sparse solution is used to output the intermediate running state variables. The objective function can be written as:

[0074] in, For the weighted objective function, The global weights of the three types of constraint factors. The weights of the data fitting terms, To obtain the observation consistency residuals between the node attribute matrix and the edge attribute matrix, For observation indexing. Minimize using sparse iterative methods such as Gauss-Newton or conjugate gradients. During the iteration process, a larger trust domain is used for the stable constraint domain and a stronger robust weight function is used for the transition constraint domain, thereby obtaining intermediate operating state variables that include cable temperature estimation, sheath strain estimation, joint contact performance estimation and loop flow consistency estimation.

[0075] When performing boundary condition checks and physical feasibility checks on intermediate operating state variables to obtain the operating state trajectory, a hard threshold test is first performed on the safety boundary, followed by a soft consistency test on the physical relationships. The final output is a time series that passes the checks, accompanied by a confidence level label. The safety boundary check and physical consistency score can be expressed as: ; ;

[0076] in, For safety boundary inspection indication quantity, For indicator functions, This is the upper limit threshold temperature. For nodes The lower and upper limits of the strain threshold, For nodes Estimation of joint contact performance, These are the lower and upper limits of the joint contact performance threshold; Scoring for physical consistency , , These are the weighted average absolute values ​​of the corresponding residuals within the time sliding window. This represents the consistency score weighting coefficient. It satisfies... and Time steps that do not exceed a preset threshold are identified as valid points. After being spliced ​​together in chronological order and inheriting the weight metadata of the stable constraint domain and the transition constraint domain, a running state trajectory is formed. Low confidence segments are marked on the trajectory for use by the subsequent non-stationary degradation path library and dynamic degradation evolution model.

[0077] S150. Based on the operational status trajectory, construct a non-stationary degradation path library that varies with time and operating conditions, and integrate the non-stationary degradation path library with the environmental exposure history to form a dynamic degradation evolution model.

[0078] Specifically, non-stationarity refers to the fact that statistical characteristics change over time and under different operating conditions, with the mean, variance, and correlation structure not being constant. This leads to different fluctuations and trends in the same variable at different times or under different operating conditions. For example, after the flood season arrives, the variance of temperature noise increases, and the temperature and strain baseline rise simultaneously when staying in an icing environment mode. Operating conditions refer to identifiable patterns of operation and environment, including high humidity environment mode, icing environment mode, high wind environment mode, high radiation environment mode, and load switching mode, used to characterize the stable and transitional segments of external drives and internal responses. For example, staying in an icing environment mode for two consecutive weeks leads to a continuous and gradual increase in strain.

[0079] The degradation path library refers to a collection of lifespan-related variable evolution curves compiled based on operational status trajectories, categorized by component, segment, and loop dimensions. Path types include progressive degradation paths, intermittently accelerated degradation paths, seasonally amplified degradation paths, and stress-triggered degradation paths. Each degradation path records the segment source, segment weight, and segment confidence level, supporting versioning and reuse. For example, a prototype of a "progressive degradation path under long-term high humidity exposure" can be generated for the same type of joint, with weights updated upon the arrival of new data. Changes with time and operating conditions refer to the systematic changes in the morphology and parameters of degradation paths under different time periods and operating conditions, with clear mode switching boundaries and residence time constraints. For example, after switching from a normal operating scenario to an icing environment mode, the path slope increases significantly and remains consistent throughout the residence period.

[0080] Environmental exposure history refers to the time series of external stresses that are consistent with the spatial reference of the operational trajectory, including wind speed, wind direction, temperature, relative humidity, rainfall, irradiance, icing thickness, lightning activity, and hydrological processes, along with station coordinates, representative weights, and quality indicators, used to explain the external factors contributing to degradation paths; for example, the icing thickness and peak gust time index of K35+600 in January match the strain lift segment. Fusion refers to aligning and jointly interpreting the degradation path library and environmental exposure history under a unified time axis, spatial coordinates, and operating condition model. This is done in two layers: segment-level fusion and path-level fusion. The former is used to adjust segment weights and segment confidence, while the latter is used to correct long-term trend labels and mode switching sensitivity descriptions; for example, the interpretation priority of local degradation segments is increased during thunderstorm weeks, while the weights are reduced under normal operating scenarios to avoid overfitting to extreme events.

[0081] The dynamic degradation evolution model refers to the updated, traceable, and multi-scale degradation description and prediction model formed after fusion. It outputs component-level curves, segment-level curves, and loop-level curves, and includes segment weights, segment confidence, long-term trend labels, mode switching sensitivity, and change point locations for subsequent risk inference and lifetime extrapolation. For example, it provides the degradation range and potential risk inflection point for the next 30 days for two scenarios: "continuous residence in icing environment mode" and "frequent load switching".

[0082] Furthermore, in the implementation process of segmenting and indexing the operational status trajectories according to component dimensions, spatial coordinates, topological location, and operating condition mode clusters, and selecting operational status trajectory segments with long-term evolution characteristics to form a degradation candidate list, firstly, a multi-level index is established for the operational status trajectories based on component dimensions, spatial coordinates, and topological location; then, the time axis is divided according to the dwell interval of the operating condition mode cluster to obtain initial segments; subsequently, the long-term score of each segment is calculated, and operational status trajectory segments with gradual, slow-changing, or dwelling trends are selected accordingly. The long-term score is defined as:

[0083] in, For long-term scoring, The slope obtained by performing linear regression on the main variables (such as cable temperature estimation or sheath strain estimation) of this segment's operating state trajectory. The standard deviation of the regression residuals. To avoid dividing by zero, the smallest positive number, This refers to the dwell time of the corresponding operating mode cluster. This is the length of the window segment. The explanatory power of the seasonal component for variance. The weighting coefficient is non-negative; when L exceeds the threshold, the segment is included in the degradation candidate list, and the segment start and end time, operating mode cluster identifier and quality identifier are recorded.

[0084] In the process of extracting degradation prototypes based on the degradation candidate list and archiving progressive degradation prototypes, intermittently accelerated degradation prototypes, seasonally amplified degradation prototypes, and stress-triggered degradation prototypes into a degradation prototype library, the degradation candidate trajectories are scale-aligned and time-aligned. Shape consistency distance is calculated and clustered according to distance to obtain prototype centers. The shape consistency distance is defined as:

[0085] in, Shift the two candidate segments within the allowed time. The mean absolute difference below The normalized trajectory within the segment after removing the mean and unifying the scale. For the common overlap length, To minimize the time shift difference, the shape parameters (slope, convexity, inflection point density, burst ratio) of the cluster center are automatically labeled as progressive, intermittent acceleration, seasonal amplification, or stress-triggered, and the center curve and statistical attributes are archived to form a degenerate prototype library.

[0086] In the process of splicing the degraded prototype library with the running state trajectory segments to form a non-stationary degradation path that varies with time and operating conditions, a segment sequence is established for each component on the time axis according to the switching boundary of the operating mode cluster. The degraded prototypes most similar to the segments are continuously spliced ​​at the boundary and a smooth transition is made at the boundary. The continuity cost of splicing is defined as:

[0087] in, As a cost of boundary continuity, This is the normalized curve of the measured segment. This is a degenerate prototype curve. and These are the closest points on the left and right sides of the boundary, respectively. The first time derivative, Match weights to the slope; to minimize The prototype is selected and the boundary transition length is determined in a way that yields a segmented continuous non-stationary degradation path, and each segment is assigned a segment weight and segment confidence.

[0088] In the process of constructing a time-continuous exposure sequence from meteorological records, icing records, lightning records, hydrological records, and extreme event records, and aligning it with a non-stationary degradation path library to establish a paired index of degradation and exposure segments, the exposure data from each source are first resampled temporally and weighted spatially representatively to form an exposure sequence consistent with the spatial coordinates of the components. Then, the optimal time lag is estimated using lag correlation, and time alignment is completed. The time lag estimate is defined as:

[0089] in, For optimal time delay, The maximum allowable time delay, The Pearson correlation coefficient is used. The normalized curve for a non-stationary degradation path. The normalized intensity sequence of environmental exposure history; After alignment, a fragment-level pairing index is established based on temporal overlap and spatial consistency, and coverage and quality indicators are recorded.

[0090] In the implementation process of generating fusion rules based on paired indexes, adjusting fragment weights and fragment confidence at the fragment level, and correcting long-term trends and mode switching sensitivity at the path level to obtain fused non-stationary degradation paths, an exposure response function is defined to adaptively correct fragment weights, while a path-level gradual variation coefficient is used to adjust the trend. The fragment weight correction rules and path trend correction rules are as follows:

[0091] in, For the first The corrected segment weights of each segment. The original segment weights, This represents the average intensity of the environmental exposure history within this segment. The standard deviation of the environmental exposure history within this segment. and For reference strength and reference fluctuation, This refers to the segment-level weight correction coefficient; These are the trend parameters after path-level long-term trend correction. The original trend parameters, The average exposure intensity during the path stay period. This is the path-level trend correction coefficient; based on this, a fused non-stationary degradation path that simultaneously encodes external influences and internal degradation patterns is obtained.

[0092] In the process of outputting the fused non-stationary degradation path as a dynamic degradation evolution model, multi-scale curves are generated for each component at the measurement point, tower, section, and loop levels. Each curve is assigned segment weights, segment confidence scores, long-term trend parameters, mode switching sensitivity, and change point locations. Simultaneously, model confidence scores are calculated as a unified quality quantification index. The model confidence score is defined as:

[0093] in, The model confidence score for the dynamic degradation and evolution model. For the first Fragment confidence of each segment For non-stationary degradation paths that are fused, For the aligned environmental exposure history, For covariance, and These are the standard deviations of the two, To ensure effective coverage duration, Total duration The non-negative weighting coefficients are used; ultimately, the curve set, parameter set, and model confidence score constitute a versioned artifact of the dynamic degradation evolution model, which serves as the direct input to the risk inference engine and lifetime extrapolation.

[0094] S160. Input the dynamic degradation evolution model into the risk inference engine to generate in-service reliability and predicted remaining life, so as to complete the online monitoring of the status of the power collection system of the mountain power station based on the in-service reliability and predicted remaining life.

[0095] Specifically, a risk inference engine refers to a computing system that performs structured reasoning on dynamic degradation evolution models and external evidence and outputs risk quantification results. It typically includes sub-modules such as feature mapping, mechanism alignment, evidence fusion, uncertainty clearing, and result calibration. It can compare degradation trajectories, environmental exposures, and historical failure samples in the same domain and form in-service risk trajectories. For example, under continuous low wind and high radiation scenarios, the risk inference engine strengthens the weight of thermal evidence, thereby increasing the contribution of the "heat dissipation limitation" related risk segment in the final result.

[0096] In-service reliability refers to the quantification of the probability that a target object will remain functionally intact within a given time frame under current operating conditions and environmental exposure. It is typically given as a time sequence with confidence and coverage labels, used to characterize the level of confidence that "no functional failure will occur from now until a certain period in the future." For example, the in-service reliability of segment A slowly decreases from high to median within a three-day window of large diurnal temperature variation, and is simultaneously labeled as "medium data coverage" and "high model confidence." Predicted remaining life refers to the quantification of the time from the current moment to reaching the failure or maintenance threshold under a given operating scenario. It is typically given a point estimate and upper and lower limit ranges, distinguishing between normal operating scenarios, extreme exposure scenarios, and degraded operating scenarios, used to arrange preventative maintenance and spare parts. For example, the predicted remaining life range for connector J12 under icing environment mode is 18 to 26 days, extrapolated to 24 to 34 days under degraded operating scenario.

[0097] Furthermore, in the process of inputting long-term trend labels, dwell interval descriptions, mode switching sensitivity descriptions, segment weights, and segment confidence scores, along with environmental exposure profiles, from the dynamic degradation and evolution model into the risk inference engine to obtain a set of risk feature vectors, the curves of the same component dimension, segment dimension, and loop dimension are first aligned on a unified time base, and then concatenated in a fixed order to obtain the risk feature vectors, with weight decay applied to low-confidence channels; the risk feature vector can be represented as:

[0098] in, For a moment The risk feature vector, This is the numerical encoding vector for long-term trend labels. A numerical vector describing the dwell time interval (including dwell time and remaining dwell time). This is a vector describing the sensitivity to mode switching. For the segment weight vector, For the segment confidence vector, Multiply corresponding elements. This is an environmental exposure profile vector (aligned values ​​for wind speed, air temperature, irradiance, ice thickness, etc.); this vector is used as the primitive input for subsequent mechanism alignment and evidence fusion.

[0099] In the process of aligning the risk feature vector set with the failure mechanism ontology and load response mapping table, and injecting environmental exposure weight correction rules, operating condition mode residence enhancement rules, and transition boundary suppression rules to form a mechanism-enhanced risk feature vector set, the feature dimensions are first projected onto mechanism coordinates such as "sheath aging, joint contact degradation, insulation moisture, and heat dissipation limitation" based on the failure mechanism ontology. Then, the weights of different mechanisms are modified according to the scenario using rules. The mechanism enhancement mapping can be written as:

[0100] in, For mechanism-enhanced risk feature vectors, This is the linear or piecewise linear mapping matrix corresponding to the load response mapping table. This is a rule correction matrix (its diagonal blocks implement the amplification or suppression of environmental exposure weight correction rules, operating condition mode dwelling enhancement rules, and transition boundary suppression rules). The definition is the same as above; this mapping ensures that the strength of evidence for different mechanisms is comparable under the same scale.

[0101] In the process of dividing the set of risk feature vectors enhanced by mechanism into a sequence of risk evidence according to time windows, the window length and sliding step size are set. Mechanism enhancement vectors are aggregated for each window while preserving variance and coverage, resulting in time-ordered risk evidence. Window aggregation can be expressed as:

[0102] in, For the first Aggregated risk evidence vectors for each time window For the first A set of time indices for a time window For time The weighting coefficient (usually the product of the segment confidence and the data coverage) is used. The definition is the same as above; the aggregation results form a sequence of risk evidence for use in the propagation of uncertainty.

[0103] In the process of performing uncertainty propagation and confidence clearing on a sequence of risk evidence, and arbitrating conflicting cross-source evidence based on source priority and historical credibility to obtain a draft risk trajectory, evidence from different sources is first synthesized at the covariance level, and then the conflict dimension is reweighted based on priority and historical credibility; the synthesis and arbitration can be expressed as: ;

[0104] in, For the first Risk evidence vector after the uncertainty clears within a time window For source In the The aggregated risk evidence vector of each window, For source The evidence covariance matrix in this window, For source The overall weight (obtained by multiplying source priority by historical credibility); By connecting along the timeline and smoothing it slightly over time, a draft risk trajectory is obtained.

[0105] In the process of unifying the risk trajectory draft with the residual statistics, weight evolution trajectory, and operating mode cluster switching boundaries from the constraint diagram optimization stage to form the in-service risk trajectory, penalties are imposed on segments that contradict the physical constraint conclusions, peak execution times that cross mode switching boundaries are aligned, and inconsistencies with weight evolution trends are downweighted; the unification correction can be expressed as:

[0106] in, For the first The in-service risk vector for each window. For the draft vector of the risk trajectory, This is the residual statistics vector (a windowed statistical representation of the thermal, electrical, and mechanical residuals from the constraint graph optimization phase). This is a mode switching boundary proximity indicator vector (representing the distance and direction from the window to the nearest switching boundary). This is the weight evolution offset vector (representing the deviation of the evidence weights in this window from the historical trend). Three sets of consistent coefficient matrices are used; after correction, in-service risk trajectories with consistent structure, aligned events, and consistent context are obtained.

[0107] In the implementation process of generating in-service reliability based on in-service risk trajectory and combining long-term trend labels and mode switching sensitivity descriptions of dynamic degradation evolution model to generate multi-scenario prediction lifetime extrapolation trajectory, the in-service risk vector is first mapped to instantaneous risk rate, and then accumulated to obtain in-service reliability sequence. At the same time, several extrapolation scenarios are constructed and the risk vector of future windows is adjusted accordingly. The mapping and extrapolation can be expressed as:

[0108] in, For the first The instantaneous risk rate of each window. For the Sigmoid function, For risk mapping parameter vectors, For the in-service risk vector, For window duration, To accumulate to the number The in-service reliability of each window; For the context The future risk vector below, For the context The environmental and operational assumption vectors (such as "normal operation scenario", "extreme exposure scenario", and "reduced operating scenario") The scenario sensitivity matrix is ​​derived from long-term trend labels and mode switching sensitivity descriptions; based on this, lifespan extrapolation trajectories for multiple scenarios can be generated.

[0109] In the process of performing order-preserving calibration and quantile envelope fitting on the lifetime extrapolation trajectory to output the predicted remaining lifetime, the failure time distribution under each scenario is uniformly calibrated monotonically, and upper and lower envelopes are generated using quantile regression to obtain point estimates and interval estimates; the calibration and fitting can be expressed as: ; ;

[0110] in, To estimate the predicted failure time point after sequence-preserving calibration, For indicator functions, The empirical failure distribution function value is obtained from the extrapolated trajectory. For time sampling points; quantiles The envelope function, The quantile loss function, extrapolating the trajectory in time Degradation intensity at the site; comprehensive and It can output the predicted remaining lifetime with a confidence region, and together with the in-service reliability, it can be used to complete the online monitoring and alarm triggering of the status of the power collection system of the mountain power station.

[0111] This application also provides an online monitoring device for the status of a mountain power station's power collection system, referring to... Figure 2 , Figure 2 This application provides a schematic diagram of a module for an online monitoring device for the status of a mountain power station collection system. The device is a server, comprising an acquisition module 21 and a processing module 22. The acquisition module 21 acquires distributed optical fiber data, environmental meteorological data, and electrical measurement data for the mountain power station collection system, and constructs a multimodal time-series matrix based on these data. The processing module 22 constructs a graph structure representation based on the multimodal time-series matrix and processes this representation using a graph signal processing algorithm to obtain a structured operational status graph. The processing module 22 also performs dynamic processing on the operational status graph. Clustering is performed to obtain operating condition pattern clusters. Processing module 22 is also used to input the operating state diagram and operating condition pattern clusters into the constraint graph optimization framework, and introduce thermal constraints, electrical constraints and mechanical constraints during the optimization process to obtain the operating state trajectory. Processing module 22 is also used to construct a non-stationary degradation path library that changes with time and operating conditions based on the operating state trajectory, and to fuse the non-stationary degradation path library with the environmental exposure history to form a dynamic degradation evolution model. Processing module 22 is also used to input the dynamic degradation evolution model into the risk inference engine to generate in-service reliability and predicted remaining life, so as to complete the online monitoring of the status of the mountain power station collection system based on the in-service reliability and predicted remaining life.

[0112] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0113] This application also provides an electronic device, with reference to... Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: at least one processor 31, at least one network interface 34, a user interface 33, a memory 35, and at least one communication bus 32.

[0114] The communication bus 32 is used to enable communication between these components.

[0115] The user interface 33 may include a display screen and a camera. Optionally, the user interface 33 may also include a standard wired interface and a wireless interface.

[0116] The network interface 34 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0117] The processor 31 may include one or more processing cores. The processor 31 connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in the memory 35, and calling data stored in the memory 35 to perform various server functions and process data. Optionally, the processor 31 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 31 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content to be displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 31 and may be implemented as a separate chip.

[0118] The memory 35 may include random access memory (RAM) or read-only memory. Optionally, the memory 35 may include a non-transitory computer-readable storage medium. The memory 35 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 35 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 35 may also be at least one storage device located remotely from the aforementioned processor 31. Figure 3 As shown, the memory 35, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for an online monitoring method of the status of a mountain power station power collection system.

[0119] exist Figure 3 In the electronic device shown, the user interface 33 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 31 can be used to call the application program stored in the memory 35 for an online monitoring method of the status of a mountain power station collection system. When executed by one or more processors, the electronic device executes one or more methods as described in the above embodiments.

[0120] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0121] This application also provides a computer-readable storage medium storing instructions. When executed by one or more processors, these instructions cause an electronic device to perform one or more of the methods described in the above embodiments.

[0122] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0123] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0124] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0125] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0126] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0127] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for online monitoring of the status of a mountain power station's power collection system, characterized in that, The method includes: Acquire distributed optical fiber data, environmental meteorological data, and electrical measurement data for the power collection system of a mountain power station, and construct a multimodal time series matrix based on the distributed optical fiber data, environmental meteorological data, and electrical measurement data; A graph structure representation is constructed based on the multimodal time series matrix, and the graph structure representation is processed by a graph signal processing algorithm to obtain a structured running state graph. Dynamic clustering is performed on the operating state diagram to obtain operating condition pattern clusters; The operating state diagram and the operating condition mode cluster are input into the constraint diagram optimization framework. Thermal constraints, electrical constraints and mechanical constraints are introduced during the optimization process to obtain the operating state trajectory. Based on the operational trajectory, a non-stationary degradation path library that varies with time and operating conditions is constructed, and the non-stationary degradation path library is integrated with the environmental exposure history to form a dynamic degradation evolution model. The dynamic degradation evolution model is input into the risk inference engine to generate in-service reliability and predicted remaining lifetime, so as to complete the online monitoring of the status of the mountain power station collection system based on the in-service reliability and the predicted remaining lifetime.

2. The method for online monitoring of the status of a mountain power station's power collection system according to claim 1, characterized in that, The acquisition of distributed optical fiber data, environmental meteorological data, and electrical measurement data for the power collection system of the mountain power station, and the construction of a multimodal time series matrix based on the distributed optical fiber data, environmental meteorological data, and electrical measurement data, specifically includes: Data from the distributed optical fiber is collected along the power collection line using distributed optical fiber. The environmental meteorological data are obtained by collecting wind speed, wind direction, temperature, relative humidity, rainfall, irradiance, and icing using meteorological data acquisition equipment. The electrical measurement data are collected using current transformers, voltage transformers, temperature probes, partial discharge monitoring devices, and box-type transformer monitoring devices. Based on a unified time reference, the distributed optical fiber data, the environmental meteorological data, and the electrical measurement data are synchronized in time and registered in space, and spatial coordinates, topology indexes, and quality identifiers are attached at the same time step to form the multimodal time series matrix.

3. The method for online monitoring of the status of a mountain power station's power collection system according to claim 1, characterized in that, The process of constructing a graph structure representation based on the multimodal time-series matrix and processing the graph structure representation using a graph signal processing algorithm to obtain a structured operational state graph specifically includes: The multimodal time series matrix is ​​used to generate nodes according to spatial coordinates, topological location and device identifier. The nodes are used to record node attributes and node metadata. Edges are generated based on the physical topology, spatial proximity, and energy coupling of the power collection lines. These edges are used to record edge attributes and maintain connectivity between nodes. The multimodal time series matrix is ​​divided into sliding windows to construct a graph structure representation that includes a node index table, an edge index table, a node attribute matrix, and an edge attribute matrix; According to the graph signal processing algorithm, the graph structure representation is subjected to node attribute verification, anomaly label propagation, cross-source conflict adjudication and local repair, and the node attributes are smoothed and corrected based on the edge attributes to form a denoised graph structure representation. Based on the denoised graph structure representation, multi-scale views are constructed at the spatial and topological levels to form a family of multi-scale graph structure representations that can be locally magnified and globally viewed. Anomaly identification and temporal consistency processing are performed on the multi-scale graph structure representation family to generate anomaly label set. The anomaly label set is then combined with the node attribute matrix and edge attribute matrix to solidify it into a structured running state graph containing a set of nodes, a set of edges, an attribute matrix, anomaly labels, and multi-scale mapping relationships.

4. The method for online monitoring of the status of a mountain power station's power collection system according to claim 1, characterized in that, The step of performing dynamic clustering on the operating state diagram to obtain operating condition pattern clusters specifically includes: Construct a graph feature set with consistent time by using the node attribute matrix and edge attribute matrix in the running state graph, and generate a graph embedding representation based on the graph feature set. The graph embedding representation is used to maintain a one-to-one correspondence with the node index and edge index. Offline clustering is performed on the graph embedding representation within the cold start window to obtain an initial set of candidate operating conditions; According to the initial set of candidate working conditions, the graph embedding representation of the current time sliding window and the clustering result of the previous time sliding window are jointly input, and online incremental clustering is performed to update the working condition mode attribution, so as to obtain a temporary set of working conditions. The temporary operating condition pattern set is merged and split to generate a standardized operating condition pattern set. A unique identifier and central description are established for each operating condition pattern in the standardized operating condition pattern set to form a searchable pattern dictionary. Based on the pattern dictionary, mode switching events are identified, and hysteresis control is set for the mode switching events to suppress jitter switching. The confirmed mode switching events are used as the boundary and effective time of the operating mode cluster. The standardized set of operating condition modes is verified for consistency and the results are solidified to output the operating condition mode cluster.

5. The method for online monitoring of the status of a mountain power station's power collection system according to claim 1, characterized in that, The optimization framework for inputting the operating state diagram and the operating condition mode cluster into the constraint diagram introduces thermal constraints, electrical constraints, and mechanical constraints during the optimization process to obtain the operating state trajectory, specifically including: Using the mode identifier, dwell time and switching boundary provided by the operating mode cluster as prior information, the operating state diagram is divided into domains to form stable constraint domains and transition constraint domains. Thermal constraint factors, electrical constraint factors, and mechanical constraint factors are constructed in the stable constraint domain and the transition constraint domain, respectively. The thermal constraint factors are driven by the distributed optical fiber temperature channel, the box-type transformer temperature channel, and the load event labeling channel. The electrical constraint factors are driven by the current channel, the voltage channel, the joint contact performance channel, and the loop topology relationship. The mechanical constraint factors are driven by the sheath strain channel, spatial coordinate mapping, and tower section structural information. A sparse constraint graph is assembled based on the thermal constraint factor, electrical constraint factor, mechanical constraint factor, node attribute matrix, and edge attribute matrix, and intermediate running state variables are output according to the sparse constraint graph. Boundary condition checks and physical feasibility checks are performed on the intermediate running state variables to obtain the running state trajectory.

6. The method for online monitoring of the status of a mountain power station's power collection system according to claim 1, characterized in that, The step involves constructing a non-stationary degradation path library that varies with time and operating conditions based on the operational state trajectory, and then integrating this non-stationary degradation path library with the environmental exposure history to form a dynamic degradation evolution model. Specifically, this includes: The operational trajectory is segmented and indexed according to component dimension, spatial coordinates, topological location and operating mode cluster, and operational trajectory segments with long-term evolution characteristics are selected to form a degradation candidate list. Degradation prototypes are extracted based on the degradation candidate list, and progressive degradation prototypes, intermittent accelerated degradation prototypes, seasonally amplified degradation prototypes, and stress-triggered degradation prototypes are archived as a degradation prototype library. The degradation prototype library is spliced ​​with the running state trajectory segment to form a non-stationary degradation path that changes with time and operating conditions; The environmental exposure history is constructed from meteorological records, icing records, lightning records, hydrological records and extreme event records into a time-continuous exposure sequence, and aligned with the non-stationary degradation path library to establish a pairing index between degradation segments and exposure segments; Based on the pairing index, a fusion rule is generated. At the fragment level, the fragment weight and fragment confidence are adjusted, and at the path level, the long-term trend and mode switching sensitivity are corrected to obtain the fused non-stationary degradation path. The fused non-stationary degradation paths are output as the dynamic degradation evolution model.

7. The method for online monitoring of the status of a mountain power station's power collection system according to claim 1, characterized in that, The step of inputting the dynamic degradation evolution model into the risk inference engine to generate in-service reliability and predicted remaining lifetime specifically includes: The long-term trend labels, dwell interval descriptions, mode switching sensitivity descriptions, segment weights and segment confidences, along with the environmental exposure profiles from the dynamic degradation and evolution model, are input into the risk inference engine to obtain a set of risk feature vectors. Align the risk feature vector set with the failure mechanism ontology and load response mapping table, and inject environmental exposure weight correction rules, operating condition mode residence enhancement rules and transition boundary suppression rules to form a mechanism-enhanced risk feature vector set. The set of risk feature vectors enhanced by mechanism is divided into risk evidence sequences according to time windows; Uncertainty propagation and confidence clearing are performed on the risk evidence sequence, and cross-source conflict evidence is arbitrated based on source priority and historical credibility to obtain a draft risk trajectory; The draft risk trajectory is consistent with the residual statistics, weight evolution trajectory and operating mode cluster switching boundary of the constraint graph optimization stage to form the in-service risk trajectory. The in-service reliability is generated based on the in-service risk trajectory, and combined with the long-term trend label and mode switching sensitivity description of the dynamic degradation evolution model, a multi-scenario predicted lifetime extrapolation trajectory is generated. The lifetime extrapolation trajectory is subjected to order-preserving calibration and quantile envelope fitting to output the predicted remaining lifetime.

8. An online monitoring device for the status of a mountain power station's power collection system, characterized in that, The device is used to perform the online monitoring method for the status of a mountain power station collection system as described in any one of claims 1 to 7, the device comprising an acquisition module and a processing module, wherein... The acquisition module is used to acquire distributed optical fiber data, environmental meteorological data and electrical measurement data for the power collection system of the mountain power station, and to construct a multimodal time series matrix based on the distributed optical fiber data, environmental meteorological data and electrical measurement data; The processing module is used to construct a graph structure representation based on the multimodal time series matrix, and process the graph structure representation through a graph signal processing algorithm to obtain a structured running state graph. The processing module is also used to perform dynamic clustering on the running state diagram to obtain a cluster of operating condition patterns; The processing module is also used to input the running state diagram and the working condition mode cluster into the constraint diagram optimization framework, and introduce thermal constraints, electrical constraints and mechanical constraints during the optimization process to obtain the running state trajectory; The processing module is also used to construct a non-stationary degradation path library that varies with time and operating conditions based on the running state trajectory, and to integrate the non-stationary degradation path library with the environmental exposure history to form a dynamic degradation evolution model. The processing module is also used to input the dynamic degradation evolution model into the risk inference engine to generate in-service reliability and predicted remaining lifetime, so as to complete the online monitoring of the status of the mountain power station collection system based on the in-service reliability and the predicted remaining lifetime.

9. An electronic device, characterized in that, The electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions. The user interface and the network interface are both used to communicate with other devices. The processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images