Intelligent abnormality alarm system for concrete pole deflection based on sparse sensing

Abstract

The present application belongs to the technical field of alarm and monitoring, and particularly relates to a concrete pole deflection abnormality intelligent alarm system based on sparse sensing. The system receives deflection related time series data of limited measuring points, forms a measuring point state sequence through time alignment, measuring point position association and pole body coordinate normalization; based on pole bottom boundary constraint, pole body deflection continuity constraint and measured data consistency constraint, the deflection sequence of virtual measuring points between limited measuring points is reconstructed by using deflection shape base function; the deflection extreme position, curvature mutation position, deflection gradient, shape weight deviation and reconstruction residual are extracted to form structure deflection abnormality confidence value and measuring point data abnormality confidence value, and the hierarchical alarm or state change information is generated through mutual exclusion verification, persistence verification and backfilling verification. The present application can represent the continuous deflection field of the pole body under the condition of sparse sensing, distinguish the deflection abnormality of non-measuring point area and the single measuring point data abnormality, and reduce the risk of missed alarm and false alarm.

Description

Technical Field

[0001] This invention belongs to the field of alarm and monitoring technology, specifically relating to an intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing. Background Technology

[0002] Concrete poles are commonly used load-bearing components in power distribution lines. Their deflection state is directly related to the pole's stress, foundation constraints, line tension, and long-term service condition. Existing deflection anomaly alarm schemes typically deploy tilt, displacement, or strain data collection points at a few locations on the pole. Data acquisition terminals collect data from these points at set intervals, and an alarm host compares the real-time collected values ​​with preset safety limits. When the data at a particular measurement point exceeds the limit, the alarm host generates an over-limit alarm; when the data returns to within the limit, the alarm is deactivated. This type of scheme relies primarily on single-point data analysis, and measurement points typically do not participate in joint calculations. Deflection changes in non-measurement areas of the pole are not directly represented. For concrete pole monitoring under sparse sensing conditions, while existing schemes can respond to abnormal changes in measurement point locations, the alarm basis is concentrated on the instantaneous values ​​of a limited number of measurement points, lacking reconstruction and verification of the overall pole deflection morphology.

[0003] In relatively similar conventional technical solutions, alarm systems filter, smooth, compare means, or determine trends from limited measurement point data to reduce the impact of acquisition noise on alarm results. Some solutions also set different alarm limits based on different installation heights or incorporate the rate of change of data at adjacent sampling times into the judgment process to distinguish between short-term disturbances and continuous anomalies. These processing methods still center on the data from the actual measurement points themselves, mainly changing the data preprocessing method or alarm criterion form, without establishing a constraint relationship between the measurement point data and the continuous deflection morphology of the concrete pole. When the abnormal deflection peak occurs between two measurement points, near the pole root, or in a region of weakened local stiffness, the measurement point data may only show a small amplitude change, and filtering and smoothing may actually weaken the abnormal characteristics. If a measurement point experiences installation misalignment, zero-point drift, or a momentary communication interruption, the alarm system may also directly interpret a single abnormal data point as a pole anomaly, leading to an inconsistency between the alarm result and the actual structural state.

[0004] Some existing technologies attempt to estimate pole deflection state through model calculations. These typically establish estimation relationships based on a pre-defined pole stress model or historical normal data, then input limited sampled data into the model to obtain state evaluation results. In engineering implementation, such solutions often use fixed pole type parameters, uniform boundary assumptions, or historical experience curves. Due to differences in foundation constraints, line tension, pole aging, and installation posture of concrete poles, the model output is easily affected by the initial assumptions. More importantly, conventional model estimations often use measured data as fitting input and the fitting results as the basis for alarms, lacking joint discrimination between fitting residuals, deflection deviations, and abnormal locations of virtual measuring points. If the model relies too heavily on limited measuring points, it can spread abnormalities at those points to an overall abnormality; if the model relies too heavily on pre-defined shapes, it may mask real local deflection abrupt changes, making it difficult to meet the alarm system's requirements for low false alarms and low false negatives.

[0005] Therefore, data collected from a limited number of measuring points cannot directly characterize the continuous deflection field of the pole. Single-point threshold judgment and conventional model estimation both struggle to reliably distinguish between deflection anomalies in non-measuring point areas and anomalies in measured data. This leads to insufficient identification of localized abnormal deflection in the pole by the alarm system, and it easily misjudges sensor drift or isolated noise as structural anomalies. The root cause of this problem is the lack of a closed-loop technical chain in the existing alarm process, consisting of sparse measured data, pole boundary constraints, continuous deflection constraints, virtual measuring point deflection reconstruction, and reconstruction residual discrimination. This prevents the use of the physical characteristic of the continuous variation of concrete pole deflection along the pole to constrain the alarm results. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing, which can effectively solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing includes: a sparse sensing data access unit, a deflection field reconstruction processor, an anomaly confidence discriminator, and an alarm output unit.

[0009] The sparse sensor data access device is used to receive deflection-related time-series data of a limited number of measuring points on a concrete pole and to map the data of different measuring points to a unified pole body coordinate system.

[0010] The deflection field reconstruction processor is used to generate virtual measurement point deflection data for the pole position between finite measurement points based on pole boundary constraints, deflection continuity constraints, and measured data consistency constraints.

[0011] The anomaly confidence discriminator is used to generate anomaly confidence results for deflection based on measured point data, virtual point deflection data, and reconstructed residuals.

[0012] The alarm output device is used to output alarm information for abnormal deflection of concrete poles based on the confidence result of the abnormal deflection.

[0013] Preferably, the sparse sensor data access device performs measurement point time alignment, measurement point position association, and pole coordinate normalization processing on the deflection-related time series data;

[0014] The pole coordinate normalization process includes using the base of the concrete pole as the coordinate reference, writing the acquisition time, height, deflection direction, and attitude identifier of each measuring point into the same measuring point state sequence, and providing the measuring point state sequence to the deflection field reconstruction processor so that the finite measuring point data and the virtual measuring point generation process have the same spatial reference basis.

[0015] Preferably, the deflection field reconstruction processor is configured with a set of deflection morphology basis functions, which includes at least the rod bottom rotation morphology, the rod body continuous bending morphology, the rod top load morphology, and the local stiffness attenuation morphology.

[0016] The deflection field reconstruction processor combines and solves the set of deflection morphology basis functions based on the measurement point state sequence. During the solution process, it simultaneously introduces the pole bottom boundary conditions, the pole deflection continuity conditions, and the measured point consistency conditions to obtain a virtual measurement point deflection sequence distributed along the pole height.

[0017] Preferably, when generating the virtual measurement point deflection sequence, the deflection field reconstruction processor sets a morphological weight associated with the fitting residual of the measured measurement point for each deflection morphological basis function, and performs temporal smoothing constraint on the virtual measurement point deflection sequence according to the change amplitude of the morphological weight at adjacent time points.

[0018] When the shape weight of any flexural shape basis function changes abruptly between adjacent time steps, the deflection field reconstruction processor marks the virtual measurement point deflection data at the corresponding time step as the reconstruction data to be verified and transmits it to the abnormal confidence discriminator.

[0019] Preferably, the anomaly confidence discriminator includes a deflection morphology feature extraction process. This process extracts the deflection extreme value position, the position of abrupt change in the rod curvature, the deflection gradient of adjacent virtual measurement points, the morphology weight deviation, and the reconstruction residual of the measured measurement points from the virtual measurement point deflection sequence. It also establishes a correspondence between the deflection extreme value position, the position of abrupt change in the rod curvature, and the reconstruction residual of the measured measurement points to form a discriminative feature set that distinguishes between deflection anomalies in non-measurement point areas and anomalies in measured measurement point data.

[0020] Preferably, the anomaly confidence discriminator constructs a dual-path confidence discrimination process based on the discrimination feature set, and the dual-path confidence discrimination process includes virtual measurement point anomaly paths and actual measurement point anomaly paths;

[0021] The virtual measuring point anomaly path forms a structural deflection anomaly confidence value based on the deflection extreme value in the non-measuring point area, the location of the abrupt change in the curvature of the pole, and the deflection gradient of adjacent virtual measuring points;

[0022] The abnormal path of the measured point is based on the residual of a single measured point reconstruction, the deviation of the morphological weight, and the persistence of the residual at adjacent time points to form the abnormal confidence value of the measured point data.

[0023] Preferably, the anomaly confidence discriminator performs mutual exclusion verification on the anomaly confidence values ​​of the structural deflection and the anomaly confidence values ​​of the measurement point data;

[0024] When the abnormal location corresponding to the abnormal confidence value of the structural deflection falls between a limited number of measuring points and the abnormal confidence value of the measuring point data does not meet the abnormal condition of the measuring point, a candidate result for structural deflection alarm is generated.

[0025] When the abnormal confidence values ​​of the measurement point data are concentrated at a single measured point and the corresponding virtual measurement point deflection sequence does not satisfy the continuous bending pattern of the rod, a candidate result of abnormal measurement point data is generated.

[0026] Preferably, the anomaly confidence discriminator performs continuous verification on the structural deflection alarm candidate results, and the continuous verification includes comparing the anomaly position drift range, the consistency of curvature change direction, and the stability of morphological weight ranking within a continuous sampling window;

[0027] When the abnormal position drift range, the consistency of the curvature change direction, and the stability of the morphological weight sorting satisfy the continuous abnormal conditions of the neighborhood of the same virtual measuring point, a deflection abnormality alarm confirmation result is generated, and the deflection abnormality alarm confirmation result is sent to the alarm output device.

[0028] Preferably, the alarm output device generates graded alarm content based on the deflection anomaly alarm confirmation result. The graded alarm content includes concrete pole identification, abnormal virtual measuring point location, corresponding deflection morphology type, structural deflection anomaly confidence value, measuring point data anomaly confidence value, and confidence change sequence within the continuous sampling window.

[0029] When the candidate result of the abnormal measurement point data is higher than the candidate result of the structural deflection alarm, the alarm output device outputs a measurement point verification prompt, and when the candidate result of the structural deflection alarm is higher than the candidate result of the abnormal measurement point data, it outputs a deflection abnormality alarm prompt.

[0030] Preferably, the abnormal confidence discriminator performs backfill verification between the output graded alarm content and the measured point data in the subsequent sampling window. The backfill verification includes comparing the consistency of the regenerated virtual point deflection sequence in the subsequent sampling window with the abnormal virtual point location, corresponding deflection morphology type and confidence change sequence corresponding to the output alarm.

[0031] When the consistency comparison result deviates, the structural deflection anomaly confidence value and the measurement point data anomaly confidence value are updated, and the alarm output device generates alarm status change information.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] 1. This invention transforms the alarm for abnormal deflection of concrete poles from a single-point limit judgment based on a limited number of measuring points to an anomaly identification based on a continuous deflection field through collaborative processing between sparse sensor data access, unified pole coordinate mapping, deflection field reconstruction, anomaly confidence judgment, and alarm output. The limited measuring point data, after time alignment, measuring point location association, and pole coordinate normalization, can serve as a common spatial basis for generating virtual measuring points. The deflection field reconstruction processor generates a virtual measuring point deflection sequence based on the pole base boundary conditions, pole deflection continuity conditions, and actual measuring point consistency conditions for the pole positions between the limited measuring points. The anomaly confidence judgment process then jointly judges the virtual measuring point deflection extremes, curvature abrupt change locations, deflection gradients of adjacent virtual measuring points, shape weight deviations, and actual measuring point reconstruction residuals. Therefore, when a local deflection peak or curvature abrupt change occurs in a non-measuring point area, it no longer depends on whether the adjacent measured points have reached a fixed limit. Instead, it can form structural deflection alarm candidate results through continuous morphological changes in the virtual measuring point sequence. When an abnormal reading occurs at a single measured point, abnormal measuring point data candidate results can be generated by reconstructing the residual concentration and the inconsistency of the continuous bending morphology of the pole body. This makes the alarm results more closely correlated with the actual deflection state of the pole body.

[0034] 2. This invention also enables continuous verification and status update mechanisms before and after alarm output. The anomaly confidence discriminator generates structural deflection anomaly confidence values ​​and measurement point data anomaly confidence values ​​through a dual-path confidence discrimination process, and then performs mutual exclusion verification on the two to avoid a single anomaly evidence directly triggering a structural alarm. For structural deflection alarm candidate results, the anomaly position drift range, curvature change direction consistency, and morphological weight ranking stability within the continuous sampling window are used for continuous verification, which can reduce alarm fluctuations caused by instantaneous noise, short-term disturbances, or reconstruction instability. The hierarchical alarm content generated by the alarm output device includes the location of the abnormal virtual measurement point, the corresponding deflection morphology type, the structural deflection anomaly confidence value, the measurement point data anomaly confidence value, and the confidence change sequence, giving the alarm information a clear data source and morphological basis. The virtual measurement point deflection sequence regenerated in subsequent sampling windows can also be compared for consistency with the anomaly location, deflection morphology type, and confidence change sequence corresponding to the output alarm, and the structural deflection anomaly confidence value and the measurement point data anomaly confidence value are updated accordingly, so that the alarm status can be corrected as the deflection field changes. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the overall processing flow of the intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing, as described in this invention.

[0036] Figure 2 This is a flowchart of the sparse sensing data access and measurement point state sequence generation of the present invention.

[0037] Figure 3 This is a flowchart of the continuous deflection field reconstruction based on the flexural morphology basis function of the present invention;

[0038] Figure 4 This is a flowchart of the deflection anomaly confidence judgment, graded alarm and backfill verification process of the present invention;

[0039] Figure 5 This is a continuous deflection field diagram of the present invention;

[0040] Figure 6 This is a gradient and curvature abrupt change diagram of the present invention;

[0041] Figure 7 This is a diagram showing the abnormal position drift range of the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Please refer to Figure 1 This embodiment provides an intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing. It includes a sparse sensing data receiver, a deflection field reconstruction processor, an anomaly confidence discriminator, and an alarm output device. The sparse sensing data receiver receives deflection-related time-series data from a limited number of measuring points on the concrete pole. This deflection-related time-series data can be pole lateral displacement data, tilt angle data, strain-equivalent deflection data, or a combination of these data, converted from existing deflection, attitude, or strain acquisition devices. The sparse sensing data receiver does not limit the sensor structure or installation structure; it only performs unified processing on the time, position, and direction of the data entering the alarm calculation. The deflection field reconstruction processor receives the unified processed data from the limited number of measuring points and, based on pole boundary constraints, deflection continuity constraints, and measured data consistency constraints, determines the deflection of the pole from the given points. Virtual measuring point deflection data is generated. The anomaly confidence discriminator inputs the measured measuring point data, virtual measuring point deflection data, and reconstructed residuals into the anomaly discrimination process to form a deflection anomaly confidence result. The alarm output device generates an alarm information for concrete pole deflection anomaly based on the deflection anomaly confidence result. In this embodiment, sparse sensor data is used to provide a real observation entry point, pole constraints are used to supplement the continuous deflection expression in non-measuring point areas, and reconstructed residuals are used to constrain the consistency relationship between measured data and reconstructed data. The three types of information are closed within the same alarm chain, so that the alarm object is transformed from a single measuring point exceeding the limit to an abnormal position and abnormal shape in the continuous deflection field of the pole. The advantage of this embodiment is that it can distinguish between deflection anomalies in non-measuring point areas and single measuring point data anomalies under limited measuring point conditions, avoiding the alarm logic from relying entirely on the instantaneous value of a single point.

[0044] Specifically, refer to Figure 2The sparse sensor data access device establishes a pole coordinate reference for each concrete pole, with the pole base position recorded as the pole coordinate origin. Normalized coordinates are established along the pole height direction. Each measurement point record entering the calculation includes the pole identifier, measurement point height, sampling time, measurement point direction identifier, original deflection value, and data quality identifier. The measurement point height is converted to normalized height during access, and the deflection direction is mapped to a signed direction consistent with the lateral force direction of the pole. The sampling time is aligned to a unified calculation time window. When multiple sampling values ​​exist within the same time window, the measurement point value for that time window is determined by the time distance and data quality identifier. When a measurement is missing within a time window, an alarm is not generated directly, but rather during subsequent reconstruction processing. The data is treated as missing observations. After receiving finite measurement point data under unified pole coordinates, the deflection field reconstruction processor establishes a discrete expression of the pole with actual measurement points as constraint points and virtual measurement points as estimation points. Virtual measurement points can be generated at equal intervals according to the normalized height of the pole or by densifying the pole's stress-sensitive areas. The anomaly confidence discriminator reads both actual and virtual points during calculation. The alarm output device outputs alarm information including at least the pole identification, anomaly source type, anomaly location, confidence result, and alarm status. The advantage of this embodiment is that it transforms scattered sparse observation data into an alarm calculation object with a unified spatiotemporal reference basis, reducing anomaly discrimination bias caused by inconsistencies in measurement point height, sampling time, or direction identification.

[0045] ;

[0046] in, Indicates the first Normalized pole height at each measured point Indicates the first The height position of each measured point relative to the bottom of the pole. This indicates the total height of the concrete pole used for deflection monitoring; for example, a measurement point located at a height of [missing information - likely a specific height on the pole]. The total height of the poles involved in the monitoring. ,but This indicates that the measured point is located at the midpoint of the pole height being monitored.

[0047] In one embodiment, the sparse sensor data access device performs measurement point time alignment, measurement point location association, and pole coordinate normalization processing on deflection-related time-series data. Measurement point time alignment is not a simple averaging of the original data, but rather maps data from different acquisition times to a unified calculation time window, and retains the sampling time deviation as a data quality factor to participate in the subsequent fitting weight determination. Measurement point location association establishes a fixed correspondence between each measurement point and the pole identifier, measurement point height, and measurement point direction. When there are changes in the measurement point order, communication disorder, or interleaving of multi-source data in the access data, the measurement point sequence can still be restored based on the measurement point identity and pole coordinates. Pole coordinate normalization processing writes the acquisition time, measurement point height, deflection direction, and measurement point attitude identifier of each measurement point into the same measurement point state sequence. The measurement point state sequence serves as the input of the deflection field reconstruction processor, avoiding the mixing and calculation of data from different spatial locations or different directions in the subsequent virtual measurement point generation process. In this embodiment, the measurement point state sequence can be organized using the fields shown in Table 1.

[0048] Table 1. Fields and Calculation Applications of Measurement Point Status Sequence

[0049]

[0050] Furthermore, the deflection field reconstruction processor generates an observation vector based on the measurement point state sequence. Each item in the observation vector corresponds to the signed deflection value of a valid measured point within the same calculation time window. If a measured point is missing, the measurement point position is retained in the observation vector, but the missing value is not forcibly filled to zero. During reconstruction, the missing item is excluded by the observation mask to avoid the missing measurement being mistakenly considered as no deflection of the pole. The measurement point time alignment can be achieved by using the selection of nearest time and the weighting of time deviation. That is, within the unified calculation time window, the sampled value closest to the center of the window and whose data quality identifier meets the access conditions is selected, and the sampling time deviation is used to reduce the influence of the sampled value on the reconstruction solution. The measurement point position association can be achieved by establishing a measurement point dictionary under the pole identifier. The measurement point dictionary records the correspondence between the measurement point number and the normalized height and direction identifier. Any data entering the calculation is backfilled with position through the measurement point dictionary. The advantage of this embodiment is that it organizes sparse sensing data into a standardized sequence that can directly participate in the continuous deflection field reconstruction and retains the data source information at the measurement point level for subsequent residual calculation.

[0051] In one embodiment, reference Figure 3The deflection field reconstruction processor configures a set of deflection morphology basis functions. This set includes pole-bottom rotation morphology, pole-body continuous bending morphology, pole-top load morphology, and local stiffness attenuation morphology. The pole-bottom rotation morphology expresses the low-order deflection distribution caused by the overall tilting of the pole body when the bottom constraint loosens. The pole-body continuous bending morphology expresses the smoothly varying bending deflection distribution along the pole body. The pole-top load morphology expresses the pole body deflection response caused by changes in the tension of the upper line. The local stiffness attenuation morphology expresses the morphological components of localized bending changes in a specific region of the pole body. The field reconstruction processor does not treat the above-mentioned form as a fixed conclusion, but rather uses it as a candidate basis function to participate in the combined solution. The system determines the combined weight of each basis function based on the measurement point state sequence within the same calculation window, and simultaneously adds the pole bottom boundary condition, pole deflection continuity condition, and measured point consistency condition during the combined solution, forming a virtual measurement point deflection sequence distributed along the pole height. The advantage of this embodiment is that it incorporates the deflection continuity of the concrete pole and the consistency of finite measurement point observations into the same reconstruction process, so that the virtual measurement point deflection data in the non-measurement point area has a traceable calculation source.

[0052] ;

[0053] in, Indicates time Lower normalized height The reconstructed deflection value at the location, This represents the total number of basis functions for the flexural shape. Indicates time Next Shape weights of a flexural shape basis function. Indicates the first The basis functions of the flexural morphology at the normalized height The value at; for example, taking At a certain normalized height , , The corresponding form weight is , , ,but . refer to Figure 5 ,Book Figure 5 The horizontal axis represents the normalized height from the bottom to the top of the pole, and the vertical axis represents the lateral deflection in millimeters. The curve represents the continuous deflection field generated by the rear end for the entire pole within the current calculation window, and the scatter points represent the measured results of available measuring points within the same window. During the calculation, the height of the measuring points is first uniformly calibrated, and then the deflection at non-measuring point locations is reconstructed using multiple morphological basis functions and window weights. If the peak value is located between two measuring points and the adjacent measuring points are still close to the curve, it indicates that there are pole morphological changes in the middle section that require key verification; if a single scatter point is far from the curve but the curve is still smooth, then the measuring point link verification is prioritized.

[0054] Preferably, when the deflection field reconstruction processor solves the basis function combination, it uses constrained minimum residual calculation to determine the shape weights. The consistency condition of the measured points is used to constrain the reconstructed deflection to be close to the measured deflection value at the height of the measured points. The pole bottom boundary condition is used to constrain the deflection or rotation angle at the pole bottom to meet the preset boundary expression. The pole body deflection continuity condition is used to constrain the deflection and curvature changes between adjacent virtual measuring points to remain continuous. The preset boundary expression does not need to change the concrete pole structure and is only used as the boundary condition input in the calculation model. When only the pole bottom position can be confirmed on site and not the actual position, the reconstructed deflection field reconstruction processor can be used to solve the problem. When considering the true boundary stiffness, weak boundary constraints can be used to allow the bar base conditions to participate in the solution with low weight, avoiding the fixed boundary assumption from having an excessively strong influence on the reconstruction results. The local stiffness attenuation pattern can be expressed by local basis functions located at different normalized height centers. Multiple local basis functions only have a large response in their respective neighborhoods, which facilitates the formation of local anomaly candidate locations in non-measurement point areas. The advantage of this embodiment is that it involves physical continuity constraints, measured observation constraints, and candidate flexural pattern constraints in parallel in the calculation, so that the reconstruction results are not completely driven by a single measurement point, nor are they completely controlled by the preset bar shape.

[0055] In one embodiment, the deflection field reconstruction processor sets morphological weights associated with the fitting residuals of the measured points for each deflection morphological basis function, and applies temporal smoothing constraints to the virtual measurement point deflection sequence based on the magnitude of morphological weight changes at adjacent time points. The morphological weights are not fixed parameters, but are obtained within each calculation window based on the observed vectors of the measured points and the basis function matrix. The fitting residuals are used to measure the degree of interpretation of the observed data by a certain set of morphological weights at the measured point location. When the residuals are large, it indicates that the current morphological combination is inconsistent with the observations of the limited number of measurement points. When the residuals are small, it indicates that the current morphological combination can be well verified by the observations of the limited number of measurement points. The temporal smoothing constraints are used to limit the morphological weights from jumping in a way that is inconsistent with the physical changes of the pole within adjacent calculation windows. If the morphological weight of a certain basis function changes abruptly at adjacent time points, the deflection field reconstruction processor marks the virtual measurement point deflection data at the corresponding time point as data to be reconstructed and transmits it to the anomaly confidence discriminator. The advantage of this embodiment is that it avoids a single anomalous sampling directly changing the complete pole deflection field, while retaining the abrupt reconstruction results for subsequent anomaly discrimination.

[0056] ;

[0057] in, Indicates time The target value for reconstruction is below. Indicates the number of measured points. Indicates the first Each measured point at time... Data quality weights Indicates the first Signed deflection values ​​at each measured point Indicates the first Normalized height of each measured point This represents the reconstructed deflection value at the measured height of the measuring point. Indicates the number of virtual measurement points. Indicates the first Normalized height of a virtual measuring point Indicates the weight of the deflection continuity constraint. This represents the morphological weights and temporal smoothing constraint weights. This indicates the time interval between adjacent calculation windows; for example, if there are two measured points at a certain moment, , , , The continuous constraint term is 0.04 and The morphological weight change term is 0.09 and ,but .

[0058] Specifically, the deflection field reconstruction processor establishes a basis function matrix formed by the measured values ​​of the measured points within each calculation window. Each row of the matrix corresponds to a measured point, and each column corresponds to a deflection morphology basis function. The morphology weights obtained are used to generate the reconstructed deflection values ​​of all virtual points. If the data quality flag of a measured point indicates sampling delay, short-term packet loss and re-sampling, or instantaneous change, the data quality weight is reduced, participates in the solution but does not dominate the solution result. If the number of effective points in a certain window is less than that required for the basis function solution, the morphology weight of the previous window can be temporarily used and the verification mark level can be increased, so that the anomaly confidence discriminator can make a more stringent judgment on the result of that window in subsequent processes. The mark of morphology weight change can be determined based on the weight difference between adjacent windows and the change of the reconstruction target value. Only when the weight change is accompanied by the concentration of reconstruction residuals or abnormal morphological changes of virtual points will it enter the candidate anomaly calculation. The advantage of this embodiment is that it can still maintain the continuous expression of the deflection field under the condition of incomplete data and sparse measurement points, and avoids directly converting low-quality observations into structural alarms.

[0059] In one embodiment, reference Figure 4The anomaly confidence discriminator performs a deflection morphology feature extraction process. This process extracts the deflection extreme value position, the position of abrupt change in the rod curvature, the deflection gradient of adjacent virtual measurement points, the morphology weight deviation, and the reconstruction residual of the measured measurement points from the virtual measurement point deflection sequence. The deflection extreme value position is determined by comparing the absolute values ​​of the reconstructed deflection of each virtual measurement point. The position of abrupt change in the rod curvature is determined by the gradient change of adjacent virtual measurement points. The deflection gradient of adjacent virtual measurement points is calculated by the deflection difference between adjacent virtual points and the normalized height difference. The morphology weight deviation is calculated by... The current morphological weight is obtained by comparing it with the morphological weight baseline in the historical stable window. The reconstructed residual of the measured measurement point is obtained by comparing the measured deflection value with the reconstructed deflection value at the same height. The anomaly confidence discriminator establishes a correspondence between the extreme deflection location, the location of the sudden change in pole curvature, and the reconstructed residual of the measured measurement point, forming a discriminant feature set that distinguishes between deflection anomalies in non-measurement point areas and anomalies in measured measurement point data. The advantage of this embodiment is that it does not use a single feature as the alarm basis, but instead puts virtual position anomalies, curvature shape anomalies, and measured residual anomalies into the same discriminant set. The source and discriminant meaning of the discriminant feature set are shown in Table 2.

[0060] Table 2. Sources and discriminative meanings of the discriminative feature sets.

[0061]

[0062] Furthermore, when constructing the discriminant feature set, the anomaly confidence discriminator performs neighborhood merging on the virtual measurement point deflection sequence. Neighborhood merging refers to grouping the positions with similar deflection extremes and consistent curvature directions among multiple adjacent virtual measurement points into the same candidate anomaly region, avoiding multiple similar alarm points due to the high density of discrete virtual points. The calculation of curvature change location can be completed using the second-order difference method. The morphological weight deviation is obtained by the distance between the current weight vector and the historical stable weight vector. The historical stable weight vector can be formed by calculating the median value of multiple weight vectors within the alarm-free window. The residual of the actual measurement point reconstruction records not only the residual magnitude but also the residual concentration location. If the residual is concentrated at a single actual measurement point while the virtual measurement point deflection sequence remains continuous, the feature is more biased towards measurement point data anomalies. If the residual is continuously distributed with the pole height at multiple measurement points and virtual measurement points exhibit extreme values ​​and curvature changes, the feature is more biased towards structural deflection anomalies. The advantage of this embodiment is that it forms an implementable anomaly discrimination basis through the common index of location, morphology, and residual.

[0063] ;

[0064] in, Indicates time Next Deflection gradient of a virtual measuring point interval, Indicates the first Reconstructed deflection values ​​of virtual measuring points Indicates the first Reconstructed deflection values ​​of virtual measuring points This represents the normalized height difference between two adjacent virtual measurement points; for example, , , , ,but This indicates that the unit normalized height deflection change over the virtual measuring point interval is 6. (Reference) Figure 6 , Figure 6 The horizontal axis represents the normalized height of the pole, and the vertical axis represents the curvature change between adjacent virtual points, used to locate local bending abrupt changes in a continuous deflection field. The curvature peak does not generate an event independently but is determined in conjunction with the location of deflection extrema, gradient smoothness, measurement point mass, residual concentration, and continuous window stability. If the peak remains within the same height neighborhood across multiple windows, spatial positioning is stable; if the peak drifts significantly with each window, the measurement point mass heatmap and target composition should be reviewed to determine if there are short-term disturbances or abrupt changes in mass weights.

[0065] In one embodiment, the anomaly confidence discriminator constructs a dual-path confidence discrimination process based on a discriminative feature set. The dual-path confidence discrimination process includes a virtual measurement point anomaly path and a measured measurement point anomaly path. The virtual measurement point anomaly path forms a structural deflection anomaly confidence value based on the deflection extreme value in the non-measurement point area, the location of the sudden change in the curvature of the pole, and the deflection gradient of adjacent virtual measurement points. The measured measurement point anomaly path forms a measurement point data anomaly confidence value based on the reconstruction residual of a single measured measurement point, the deviation of the morphological weight, and the persistence of the residual at adjacent time points. The calculation of the structural deflection anomaly confidence value focuses on whether the virtual measurement point forms a continuous abnormal shape in the non-measurement point area. The calculation of the measurement point data anomaly confidence value focuses on whether the abnormal residual is concentrated at a single measured point and is inconsistent with the continuous deflection shape of the pole. The dual-path confidence discrimination process generates two types of confidence values ​​in parallel, and then the alarm source is determined by a mutual exclusion verification process. The advantage of this embodiment is that it treats the real deflection anomaly in the non-measurement point area and the measurement point data anomaly as two competing interpretations, rather than directly interpreting any abnormal value as a pole failure.

[0066] ;

[0067] in, Indicates time The structural deflection anomaly confidence value is as follows: Represents the normalized mapping function. This represents the normalized value of the extreme deflection in the non-measuring point region. This represents the normalized value of the abrupt change in shaft curvature. This represents the normalized deflection gradient between adjacent virtual measuring points. This represents the centrally normalized quantity of residuals at a single measured point. , , and This indicates that the confidence weights of each input quantity are calculated and their meaning remains unchanged throughout the text; for example, taking... , , , , , , , The result inside the parentheses is , .

[0068] In a preferred embodiment, the anomaly confidence value of the measurement point data is jointly determined by residual concentration, morphological inconsistency, and residual persistence. Residual concentration indicates that the reconstructed residuals are mainly distributed in a single measured point. Morphological inconsistency indicates that the virtual measurement point sequence cannot form a continuous bending shape in the same direction as the anomaly of the measured point. Residual persistence indicates that the residuals of the same measured point appear repeatedly in adjacent time windows, but the position of the anomalous virtual measurement point is unstable. When calculating the anomaly confidence value of the measurement point data, the anomaly confidence discriminator does not directly identify large single-point residuals as measurement point anomalies. Instead, it requires that there is an inconsistent relationship between the residual and the continuous deflection field of the virtual measurement point. If the residual of a certain measured point is large, but the deflection extremum, curvature change position, and gradient direction of the virtual measurement point all support the existence of continuous bending in the same area, then the residual is regarded as observational evidence of structural deflection anomalies. If the residual of a single measured point is large and the adjacent virtual measurement points do not form continuous changes, then the residual is regarded as evidence of measurement point data anomalies. The advantage of this embodiment is that the same residual feature can enter different discrimination paths according to the context of the pole shape, reducing misjudgment of single features.

[0069] In one embodiment, the anomaly confidence discriminator performs mutual exclusion checks on the structural deflection anomaly confidence value and the measurement point data anomaly confidence value. When the anomaly location corresponding to the structural deflection anomaly confidence value falls within a limited number of measurement points and the measurement point data anomaly confidence value does not meet the measurement point anomaly condition, a structural deflection alarm candidate result is generated. When the measurement point data anomaly confidence value is concentrated at a single measured measurement point and the corresponding virtual measurement point deflection sequence does not meet the continuous bending pattern of the rod, a measurement point data anomaly candidate result is generated. The mutual exclusion check does not require that only one of the two types of confidence values ​​can exist. Instead, it determines the dominant candidate result based on the correspondence between the anomaly location, residual distribution, and continuous bending pattern. When the two types of confidence values ​​are close and the discrimination evidence is insufficient, the anomaly confidence discriminator can mark the current time window as an observation state and pass the data to subsequent continuous checks. The alarm state is changed after the subsequent sampling window provides more temporal evidence. The advantage of this embodiment is that it avoids the simultaneous triggering of structural alarms and measurement point anomaly prompts in the same insufficient evidence scenario, thus maintaining the clarity of the alarm output.

[0070] ;

[0071] in, Indicates time Abnormal confidence values ​​of the measurement point data below This represents the centrally normalized quantity of residuals at a single measured point. This indicates that the shape weights deviate from the normalized value. This represents the continuous normalization of the residuals at adjacent time points. This represents the normalized value of the extreme deflection in the non-measuring point region. , , and This indicates that the weights for outlier confidence values ​​of the measurement points are calculated and their meaning remains unchanged throughout the text; for example, taking... , , , , , , , The result inside the parentheses is , .

[0072] Furthermore, the anomaly confidence discriminator establishes a candidate result status table during mutual exclusion verification. The candidate result status includes structural deflection alarm candidates, measurement point data anomaly candidates, and candidates to be observed. The structural deflection alarm candidate records the location of the abnormal virtual measurement point, the neighborhood of curvature change, the corresponding deflection morphology type, and the structural deflection anomaly confidence value. The measurement point data anomaly candidate records the abnormal measured measurement point identifier, the degree of residual concentration, evidence of morphological inconsistency, and the measurement point data anomaly confidence value. The candidates to be observed record two types of confidence values, the current calculation time window, and the features required to enter the continuous verification. In this embodiment, the alarm output device does not directly output the structural alarm corresponding to the candidate to be observed, but waits for the continuous verification conclusion. If the subsequent window shows that the abnormal position is stable and the curvature direction is consistent, the candidate to be observed can be transferred to the structural deflection alarm candidate. If the subsequent window shows that the residual is continuously concentrated at the same measured measurement point and the virtual measurement point sequence is unstable, the candidate to be observed can be transferred to the measurement point data anomaly candidate. The advantage of this embodiment is that it retains the data state with ambiguous boundaries in the traceable candidate set, avoiding short-term data uncertainty from directly entering the alarm output.

[0073] In one embodiment, the anomaly confidence discriminator performs continuous verification on the structural deflection alarm candidate results. The continuous verification compares the anomaly location drift range, the consistency of curvature change direction, and the stability of morphological weight ranking within a continuous sampling window. The anomaly location drift range is used to determine whether the candidate anomaly stably exists in the same virtual measuring point neighborhood. The consistency of curvature change direction is used to determine whether the local bending direction of the rod remains in the same direction within multiple time windows. The stability of morphological weight ranking is used to determine whether the candidate deflection morphology maintains its dominant explanatory position within multiple time windows. If the anomaly location changes significantly within a continuous window, the candidate anomaly is more likely to come from noise or reconstruction instability. If the curvature change direction changes repeatedly, the candidate anomaly is more likely to come from short-term disturbances. If the morphological weight ranking changes frequently, the morphological interpretation of the candidate anomaly is unstable. Only when the anomaly location drift range, the consistency of curvature change direction, and the stability of morphological weight ranking all satisfy the continuous anomaly condition in the same virtual measuring point neighborhood, the anomaly confidence discriminator generates a deflection anomaly alarm confirmation result and sends it to the alarm output device. The advantage of this embodiment is that it transforms the single-time-window discrimination result into a continuous-time-window confirmation result. The input and confirmation conditions for continuous verification are shown in Table 3.

[0074] Table 3 Continuous Validation Input and Confirmation Conditions

[0075]

[0076] Specifically, the drift range of anomaly locations can be represented by the difference between the maximum and minimum normalized heights of the abnormal virtual measurement points within a continuous sampling window. The consistency of curvature mutation direction can be represented by the proportion of time windows with the same curvature mutation sign within a continuous window. The stability of morphological weight ranking can be represented by the ranking change of the dominant basis functions within a continuous window. The dominant basis function refers to the basis function with higher morphological weight in the current time window that participates in interpreting the abnormal virtual measurement points. When the dominant basis functions remain the same or change within the same category of basis functions within a continuous window, the morphological interpretation is considered stable. When the anomaly location is stable within a continuous window but the curvature direction is inconsistent, the anomaly confidence discriminator can maintain the candidate to be observed without generating a confirmation result. When the anomaly location, curvature direction, and morphological weight ranking all meet the continuous anomaly condition, the deflection anomaly alarm confirmation result includes the location of the abnormal virtual measurement point, the range of the continuous window, the dominant deflection morphology, and the structural deflection anomaly confidence value. The advantage of this embodiment is that it eliminates isolated reconstruction fluctuations through temporal consistency, so that the alarm confirmation result is based on continuous data evidence.

[0077] ;

[0078] in, Indicated by time The range of abnormal position drift within the continuous sampling window at the end time. This represents the set of sampling times within the continuous sampling window. Indicates time Normalized height of abnormal virtual measuring points corresponding to candidate results of substructure deflection alarm This indicates taking the maximum value. This indicates taking the minimum value; for example, if the normalized heights of abnormal virtual measurement points within a continuous sampling window are 0.62, 0.64, 0.63, and 0.65 respectively, then... This indicates that the abnormal position drift range within the window is 0.03 normalized height units. (Reference) Figure 7 ,Book Figure 7 The horizontal axis represents the window number or sampling time, and the vertical axis represents the drift range of the candidate extreme value position within the recent window, used to measure the stability of event localization. The backend tracks the extreme value height of each window and calculates the difference between the maximum and minimum positions within the rolling window; the smaller the difference, the more reliable the candidate source is in the time dimension. If the drift curve suddenly rises, it should be determined whether it is caused by delay, short-term disturbance, or sensor state conflict, in conjunction with the quality of the measurement points, residual concentration, and the composition of the reconstructed target; if the drift is stable and the structural confidence is high, the event level can be increased.

[0079] In one embodiment, the alarm output device generates graded alarm content based on the deflection anomaly alarm confirmation result. The graded alarm content includes concrete pole identification, abnormal virtual measuring point location, corresponding deflection morphology type, structural deflection anomaly confidence value, measuring point data anomaly confidence value, and confidence change sequence within the continuous sampling window. The abnormal virtual measuring point location is used to indicate the suspected deflection anomaly location in non-measuring point areas. The corresponding deflection morphology type is used to characterize which of the following morphologies the current anomaly is closer to: pole base rotation, continuous bending of the pole body, pole top load, or local stiffness attenuation. The structural deflection anomaly confidence value and the measuring point data anomaly confidence value are output simultaneously to express two types of interpretive evidence for the alarm result. The confidence change sequence within the continuous sampling window is used to reflect the temporal stability during the alarm confirmation process. When the measuring point data anomaly candidate result is higher than the structural deflection alarm candidate result, the alarm output device outputs a measuring point verification prompt. When the structural deflection alarm candidate result is higher than the measuring point data anomaly candidate result, the alarm output device outputs a deflection anomaly alarm prompt. The advantage of this embodiment is that the alarm information not only gives the conclusion of exceeding the limit, but also gives the anomaly location, morphology type, and confidence source.

[0080] Furthermore, the alarm output device can generate different alarm states based on the candidate result source. The deflection anomaly alarm prompts the confirmation result of the corresponding structural deflection alarm candidate after continuous verification. The measurement point verification prompts the corresponding actual measurement point data anomaly candidate. The alarm state change information corresponds to the situation where the existing alarm judgment is changed in the subsequent backfill verification. When the alarm output device generates alarm content, it does not change the structural deflection anomaly confidence value and the measurement point data anomaly confidence value generated by the anomaly confidence discriminator. Instead, it writes the two into the alarm record according to the alarm state so that the subsequent sampling window can perform consistency comparison. The deflection morphology type in the alarm content comes from the basis function category that ranks high in the morphology weight ranking and is related to the neighborhood of the abnormal virtual measurement point. If the abnormal virtual measurement point is located between finite measurement points and the local stiffness decay morphology weight is dominant, then the corresponding deflection morphology type in the alarm content is recorded as the local stiffness decay morphology. If the anomaly is mainly manifested as the overall tilt of the rod and the rod bottom rotation morphology weight is dominant, then it is recorded as the rod bottom rotation morphology. The advantage of this embodiment is that it makes the alarm record retain the key features in the reconstruction and discrimination process, which is convenient for the same source comparison when the state is updated in the future.

[0081] In one embodiment, the anomaly confidence discriminator performs backfill verification between the output graded alarm content and the measured point data in the subsequent sampling window. Backfill verification includes comparing the newly generated virtual point deflection sequence in the subsequent sampling window with the abnormal virtual point locations, corresponding deflection morphology types, and confidence change sequences corresponding to the output alarms. When the consistency comparison result deviates, the anomaly confidence discriminator updates the structural deflection anomaly confidence value and the measurement point data anomaly confidence value, and the alarm output device generates alarm status change information. Backfill verification is not a simple overwrite of historical alarms, but rather provides new data in the subsequent sampling window. After obtaining the measured data, coordinate normalization, deflection field reconstruction, feature extraction, and dual-path confidence discrimination are re-executed. The newly generated anomaly location, morphology type, and confidence sequence are compared with the existing alarm records. If the location of the anomaly virtual measurement point in the subsequent window is still within the original alarm neighborhood and the dominant deflection morphology remains consistent, the alarm status is maintained. If the anomaly location in the subsequent window moves to the vicinity of the measured measurement point and the residual concentration increases, the alarm status can be changed from a deflection anomaly alarm prompt to a measurement point verification prompt. The advantage of this embodiment is that the alarm status can be corrected with continuous observation data, reducing the status solidification caused by insufficient initial window data.

[0082] ;

[0083] in, Indicates time The following alarm consistency deviation This indicates the normalized height of the abnormal virtual measurement points redefined in subsequent sampling windows. This indicates that the normalized height of the abnormal virtual measuring point in the alarm record has been output. This indicates the confidence value of structural deflection anomalies recalculated in subsequent sampling windows. This indicates that the structural deflection anomaly confidence value has been output from the alarm log. This indicates the abnormal confidence value of the measurement point data recalculated in subsequent sampling windows. This indicates that the abnormal confidence value of the measurement point data in the alarm record has been output. This indicates the difference between the deflection morphology type in subsequent sampling windows and the deflection morphology type in the output alarm records. , , and This indicates that the consistency deviation from the calculated weights does not change the meaning throughout the text; for example, taking , , , , , , , , , , ,but .

[0084] In a preferred embodiment, backfill verification further includes extending the confidence change sequence of the output alarm record. After each reconstruction and discrimination by the subsequent sampling window, the new structural deflection anomaly confidence value and the measurement point data anomaly confidence value are appended to the corresponding alarm record. The anomaly confidence discriminator determines whether the alarm state is maintained, downgraded, converted to measurement point verification, or cancelled based on the appended confidence change sequence. If the structural deflection anomaly confidence value remains continuously in the structural anomaly dominant state and the abnormal virtual measurement point position is in the original neighborhood, the alarm state is maintained. If the structural deflection anomaly confidence value decreases and the measurement point data anomaly confidence value increases, and the residuals are concentrated in the same measured measurement point, the alarm state is converted to measurement point verification. If both types of confidence values ​​decrease and the virtual measurement point deflection sequence recovers to the historical stable state, the alarm state can generate cancellation information. When generating alarm state change information, the alarm output device simultaneously records the state before the change, the state after the change, the subsequent sampling window that triggered the change, and the main deviation characteristics. The advantage of this embodiment is that it transforms the alarm output from a one-time judgment to a continuous data-driven state management process.

[0085] In one embodiment, the set of deflection morphology basis functions can be initialized based on historical stable data. Historical stable data refers to the sequence of measurement point states formed in a sampling window where there are no deflection anomaly alarms, no abnormal measurement point data candidates, and the data quality identifier meets the access conditions. The deflection field reconstruction processor inputs the historical stable data into the same reconstruction objective function to obtain morphology weight vectors under multiple stable windows, and forms a morphology weight baseline using its median or truncated mean. The morphology weight deviation is determined by the distance between the current morphology weight vector and the morphology weight baseline. The initialization process does not require external labeling of abnormal samples or modification of the concrete pole structure; it only relies on the stable window data formed during system operation. If a pole has insufficient available stable windows, the stable weight baseline of poles of the same specification in the same line can be used as the initial baseline, and then replaced with the pole's own baseline after accumulating stable windows. The advantage of this embodiment is that the morphology weight deviation can reflect the long-term deflection state of a specific pole, rather than relying entirely on uniform empirical parameters.

[0086] ;

[0087] in, Indicates time The morphological weights deviate from the normalized value. Indicates time Next Shape weights of a flexural shape basis function. Indicates the first in the historical stable window Baseline weights of a flexural morphology basis function This represents the total number of basis functions for the flexural shape. This indicates a positive number to prevent the denominator from being zero; for example, taking... , , , , , Then the molecule is The denominator is , .

[0088] Furthermore, the selection of historical stable windows is constrained by both alarm status and data quality. When a window is in the period of deflection anomaly alarm prompt, measurement point verification prompt, candidate for observation, or alarm status change, it will not be included in the stable window set. When a window has multiple missing measurement points, excessive sampling time deviation, or measurement point attitude identifier conflict, it will not be included in the stable window set. The morphological weight baseline update can be performed using a sliding window method, but the window during the alarm period will not participate in the update to prevent abnormal data from being written into the normal baseline. If the baseline change is large after long-term operation, the deflection field reconstruction processor can retain the difference record between the old baseline and the new baseline and use it as part of the subsequent interpretation of the morphological weight deviation. When the abnormal confidence discriminator reads the morphological weight deviation, it also reads the baseline source identifier to distinguish between the adaptive baseline of this pole and the initial baseline of the same line. The advantage of this embodiment is that the alarm discrimination has the ability to adapt to the long-term state of the pole itself, while avoiding the pollution of the normal baseline by abnormal windows.

[0089] In one embodiment, the concentration of the reconstructed residuals of the measured points is determined by the residual energy distribution. The anomaly confidence discriminator calculates the sum of squared residuals of all valid measured points within the same calculation window and calculates the proportion of the squared residual of a single measured point in the total squared residual. If the proportion is high and the virtual measured points in the neighborhood of the measured point do not form continuous deflection extrema and curvature abrupt changes, then the residual concentration enters the abnormal path of the measured point data. If the squared residuals are continuously distributed with height among multiple measured points and the sequence of virtual measured points forms a stable abnormal neighborhood, then the residual concentration is not used as the dominant evidence of the measured point anomaly. The persistence of the residuals is determined by the repeated occurrence of the residual concentration of the same measured point within the continuous sampling window. If the residual concentration only occurs in a single time window, the anomaly confidence discriminator can mark it as to be observed instead of directly generating a measured point review prompt. The advantage of this embodiment is that it enables the judgment of measured point anomalies to consider the residual size, residual spatial distribution, and residual temporal continuity at the same time.

[0090] ;

[0091] in, Indicates time Normalized quantity of residuals at a single measured point Indicates the number of valid measured points. Indicates the first Each measured point at time... The reconstructed residuals below This indicates taking the maximum value from the squared residuals of all measured points. This indicates a positive number to prevent the denominator from being zero; for example, if there are 3 valid measured points with residuals of 0.2, 1.0, and 0.3 respectively, take... Therefore, the maximum residual squared is 1.0, and the total residual squared is 0.04 + 1.0 + 0.09 = 1.13. .

[0092] In a preferred embodiment, the anomaly confidence discriminator can perform directional consistency verification on the reconstructed residuals of the measured points. The directional consistency verification compares the sign of the residual of each measured point with the deflection gradient direction of adjacent virtual measured points. If the residual sign is consistent with the direction of change of the deflection field in the neighborhood, and the residuals of adjacent measured points change in the same direction, then the residual is more consistent with the observed behavior of structural deflection anomalies. If the residual sign is opposite to the direction of change of the deflection field in the neighborhood, and only a single measured point residual is prominent, then the residual is more consistent with the abnormal behavior of the measured point data. The directional consistency verification does not change the residual itself, but provides a residual interpretation label for the dual-path confidence discrimination process. When the anomaly confidence discriminator forms a candidate for abnormal measured point data, it simultaneously writes the residual interpretation label and the corresponding measured point identifier. The alarm output device reads the label when outputting the measured point verification prompt. The advantage of this embodiment is that it further distinguishes different sources of anomalies under the same residual size, reducing the false verification of measured points caused by real local changes in the pole.

[0093] In one embodiment, the virtual measuring points can be set in a fixed sequence based on the normalized height of the pole, or they can be reconstructed and encrypted based on historical anomaly candidate positions. The fixed sequence is used to ensure that the virtual measuring point positions are consistent across calculation windows. The reconstruction and encryption is used to increase the density of virtual measuring points in the candidate anomaly neighborhood to refine the anomaly position. The reconstruction and encryption only changes the calculation point position, without changing the actual measurement point layout or the concrete pole structure. When performing the reconstruction and encryption, the deflection field reconstruction processor retains the original virtual measuring point sequence and inserts additional virtual measuring points in the anomaly candidate neighborhood. The deflection value of the additional virtual measuring points is still calculated by the same deflection morphology basis function combination. When calculating the anomaly position drift range, the additional virtual measuring points can be mapped back to the original virtual measuring point neighborhood to ensure that the comparison basis for continuous verification is consistent. The advantage of this embodiment is that it refines the expression of anomaly positions in non-measuring point areas without increasing the number of actual measuring points, while maintaining cross-time window comparability.

[0094] Furthermore, after generating the virtual measurement point deflection sequence, the deflection field reconstruction processor can perform boundary consistency checks and morphological acceptability checks on the sequence. The boundary consistency check is used to confirm that the reconstructed deflection and reconstructed rotation angle at the bottom boundary of the rod do not deviate from the allowable range of the boundary constraints. The morphological acceptability check is used to confirm that the virtual measurement point sequence does not have high-frequency alternating fluctuations caused by solution instability. If the boundary consistency or morphological acceptability does not meet the calculation requirements, the virtual measurement point deflection data of the current time window is marked as reconstruction data to be verified. When reading the reconstruction data to be verified, the anomaly confidence discriminator increases the reliance on persistent evidence and does not directly form structural deflection alarm candidates based on the extreme values ​​of virtual measurement points in a single time window. The advantage of this embodiment is that it performs quality control on the reconstruction process itself, so that low-confidence reconstruction results will not be directly converted into alarm output.

[0095] In one embodiment, the alarm records of the alarm output device can be managed using a state machine approach. The alarm states include no alarm, pending observation, deflection anomaly alarm prompt, measurement point verification prompt, and alarm state change. After the anomaly confidence discriminator outputs candidate results, the alarm output device switches the alarm state based on the candidate source, continuous verification results, and backfill verification results. When the structural deflection alarm candidate is confirmed by continuous verification, the state switches from pending observation or no alarm to deflection anomaly alarm prompt. When the measurement point data anomaly candidate dominates, the state switches from pending observation or no alarm to measurement point verification prompt. When the backfill verification shows that the original structural deflection alarm prompt no longer meets the consistency requirements, the state switches to alarm state change. The alarm record retains the confidence value, anomaly location, and morphological type before and after each state switch. The advantage of this embodiment is that it transforms complex discrimination results into executable alarm states, ensuring that the alarm output corresponds to the continuous deflection field analysis results.

[0096] In a preferred embodiment, the alarm output device merges repeated alarms for the same concrete pole within adjacent time windows. The merging criteria are: identical pole identification, abnormal virtual measuring point locations in the same neighborhood, consistent corresponding deflection morphology types, and continuous structural deflection anomaly confidence change sequence. When the above conditions are met, the alarm output device updates the confidence change sequence and duration window range of the original alarm record without creating a new alarm record. If the subsequent window shows a significant migration of the abnormal location or a change in the corresponding deflection morphology type, a new alarm status change information is created and the reason for the change is recorded. The merging criteria for measuring point verification prompts are: the same measured measuring point identification and the continuous occurrence of residual concentration. If the measuring point verification prompt and the deflection anomaly alarm prompt compete within the same time window, the alarm output device outputs a single main state according to the dominant candidate result of mutual exclusion verification, and retains another type of confidence value as an auxiliary field in the alarm record. The advantage of this embodiment is that it reduces the redundancy of alarm records caused by repeated confirmation of the same anomaly in continuous sampling windows, while preserving the alarm evolution process.

[0097] In one embodiment, the system can be deployed at the station, the line-side data aggregation node, or the monitoring host. The sparse sensor data access device receives limited measurement point data through existing data interfaces. The deflection field reconstruction processor, the abnormal confidence discriminator, and the alarm output device can be different processing processes in the same processing program, or they can be data processing units in different service processes. Data is transmitted between processing processes through measurement point state sequences, virtual measurement point deflection sequences, discrimination feature sets, candidate results, and alarm records. The deployment method does not affect the aforementioned data processing logic. In this embodiment, the software processing process uses a calculation time window as the basic operating cycle. In each cycle, data access, pole coordinate normalization, deflection field reconstruction, feature extraction, dual-path confidence discrimination, mutual exclusion verification, continuous verification, and alarm output are completed. If there is insufficient data in a certain cycle, the output is to be observed or the original alarm state is maintained. The structural alarm is not generated based on missing measurement data. The advantage of this embodiment is that it decouples the specific software deployment form from the alarm calculation logic, making it easy to implement the same technical solution in different monitoring platforms.

[0098] Furthermore, each processing step in the aforementioned embodiments can be implemented by a processor executing a computer program stored in the storage medium. The computer program includes instructions for generating a sequence of measurement point states, instructions for solving the deflection shape weights, instructions for generating a virtual measurement point deflection sequence, instructions for extracting a discriminative feature set, instructions for calculating the structural deflection anomaly confidence value and the measurement point data anomaly confidence value, instructions for performing mutual exclusion checks and persistence checks, and instructions for generating alarm records and alarm status change information. When the program is executed, the input data is the deflection-related time-series data of a limited number of measurement points, and the output data is the concrete pole deflection anomaly alarm information, measurement point verification prompts, or alarm status change information. The program does not rely on adding new pole structure components, nor does it require changing the existing sensing acquisition objects. It only performs coordinateization, reconstruction, and confidence processing on sparse sensing data. The advantage of this embodiment is that it implements a complete alarm chain with a computer program, ensuring that the same logic can be repeatedly executed on different processing devices.

Claims

1. An intelligent alarm system for abnormal deflection of concrete utility poles based on sparse sensing, characterized in that, include: Sparse sensor data access unit, deflection field reconstruction processor, anomaly confidence discriminator and alarm output unit; The sparse sensor data access device is used to receive deflection-related time-series data of a limited number of measuring points on a concrete pole and to map the data of different measuring points to a unified pole body coordinate system. The deflection field reconstruction processor is used to generate virtual measurement point deflection data for the pole position between finite measurement points based on pole boundary constraints, deflection continuity constraints, and measured data consistency constraints. The anomaly confidence discriminator is used to generate anomaly confidence results for deflection based on measured point data, virtual point deflection data, and reconstructed residuals. The alarm output device is used to output alarm information for abnormal deflection of concrete poles based on the confidence result of the abnormal deflection.

2. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 1, characterized in that, The sparse sensor data access device performs measurement point time alignment, measurement point position association, and pole coordinate normalization on the deflection-related time series data. The pole coordinate normalization process includes using the base of the concrete pole as the coordinate reference, writing the acquisition time, height, deflection direction, and attitude identifier of each measuring point into the same measuring point state sequence, and providing the measuring point state sequence to the deflection field reconstruction processor so that the finite measuring point data and the virtual measuring point generation process have the same spatial reference basis.

3. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 2, characterized in that, The deflection field reconstruction processor is configured with a set of deflection morphology basis functions, which includes at least the pole bottom rotation morphology, the pole body continuous bending morphology, the pole top load morphology, and the local stiffness attenuation morphology. The deflection field reconstruction processor combines and solves the set of deflection morphology basis functions based on the measurement point state sequence. During the solution process, it simultaneously introduces the pole bottom boundary conditions, the pole deflection continuity conditions, and the measured point consistency conditions to obtain a virtual measurement point deflection sequence distributed along the pole height.

4. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 3, characterized in that, When generating the virtual measurement point deflection sequence, the deflection field reconstruction processor sets a morphological weight associated with the fitting residual of the measured measurement point for each deflection morphological basis function, and performs temporal smoothing constraint on the virtual measurement point deflection sequence according to the change amplitude of the morphological weight at adjacent time points. When the shape weight of any flexural shape basis function changes abruptly between adjacent time steps, the deflection field reconstruction processor marks the virtual measurement point deflection data at the corresponding time step as the reconstruction data to be verified and transmits it to the abnormal confidence discriminator.

5. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 3, characterized in that, The anomaly confidence discriminator includes a deflection morphology feature extraction process. This process extracts the deflection extreme value position, the position of abrupt change in the rod curvature, the deflection gradient of adjacent virtual measurement points, the morphology weight deviation, and the reconstruction residual of the measured measurement points from the virtual measurement point deflection sequence. It also establishes a correspondence between the deflection extreme value position, the position of abrupt change in the rod curvature, and the reconstruction residual of the measured measurement points to form a discriminative feature set that distinguishes between deflection anomalies in non-measurement point areas and anomalies in measured measurement point data.

6. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 5, characterized in that, The anomaly confidence discriminator constructs a dual-path confidence discrimination process based on the discrimination feature set. The dual-path confidence discrimination process includes virtual measurement point anomaly paths and actual measurement point anomaly paths. The virtual measuring point anomaly path forms a structural deflection anomaly confidence value based on the deflection extreme value in the non-measuring point area, the location of the abrupt change in the curvature of the pole, and the deflection gradient of adjacent virtual measuring points; The abnormal path of the measured point is based on the residual of a single measured point reconstruction, the deviation of the morphological weight, and the persistence of the residual at adjacent time points to form the abnormal confidence value of the measured point data.

7. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 6, characterized in that, The anomaly confidence discriminator performs mutual exclusion verification on the anomaly confidence values ​​of the structural deflection and the anomaly confidence values ​​of the measurement point data; When the abnormal location corresponding to the abnormal confidence value of the structural deflection falls between a limited number of measuring points and the abnormal confidence value of the measuring point data does not meet the abnormal condition of the measuring point, a candidate result for structural deflection alarm is generated. When the abnormal confidence values ​​of the measurement point data are concentrated at a single measured point and the corresponding virtual measurement point deflection sequence does not satisfy the continuous bending pattern of the rod, a candidate result of abnormal measurement point data is generated.

8. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 7, characterized in that, The anomaly confidence discriminator performs continuous verification on the candidate results of the structural deflection alarm. The continuous verification includes comparing the drift range of the anomaly location, the consistency of the curvature change direction, and the stability of the morphological weight ranking within a continuous sampling window. When the abnormal position drift range, the consistency of the curvature change direction, and the stability of the morphological weight sorting satisfy the continuous abnormal conditions of the neighborhood of the same virtual measuring point, a deflection abnormality alarm confirmation result is generated, and the deflection abnormality alarm confirmation result is sent to the alarm output device.

9. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 8, characterized in that, The alarm output device generates graded alarm content based on the deflection anomaly alarm confirmation result. The graded alarm content includes concrete pole identification, abnormal virtual measuring point location, corresponding deflection morphology type, structural deflection anomaly confidence value, measuring point data anomaly confidence value, and confidence change sequence within the continuous sampling window. When the candidate result of the abnormal measurement point data is higher than the candidate result of the structural deflection alarm, the alarm output device outputs a measurement point verification prompt, and when the candidate result of the structural deflection alarm is higher than the candidate result of the abnormal measurement point data, it outputs a deflection abnormality alarm prompt.

10. The intelligent alarm system for abnormal deflection of concrete poles based on sparse sensing according to claim 9, characterized in that, The abnormal confidence discriminator backfills and verifies the output graded alarm content with the measured data of the actual measurement points in the subsequent sampling window. The backfill verification includes comparing the consistency of the regenerated virtual measurement point deflection sequence in the subsequent sampling window with the abnormal virtual measurement point location, corresponding deflection morphology type and confidence change sequence corresponding to the output alarm. When the consistency comparison result deviates, the structural deflection anomaly confidence value and the measurement point data anomaly confidence value are updated, and the alarm output device generates alarm status change information.

Images