Passive plate-shaped base station antenna fault diagnosis method and system based on sparse sensing

Abstract

The application relates to the technical field of antenna fault diagnosis, in particular to a passive plate-shaped base station antenna fault diagnosis method and system based on sparse sensing. The application proposes the following scheme: acquiring a transmitting response sequence of a passive plate-shaped base station antenna in multiple directions, frequency points and times; calculating an array element response behavior matrix through an array element shaping contribution model; performing sparse sensing processing on the array element response behavior matrix, extracting an array element region with an abnormal track opening degree feature, and generating an abnormal sparse matrix; and judging and marking a fault array element based on an abnormal cumulative value. The application focuses on a modeling region through a track opening degree index driven tension diagram, constructs a sparse sensing model with a regularized weighting factor, realizes lightweight diagnosis of array element behavior abnormalities, and improves fault positioning efficiency and edge adaptability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of antenna fault diagnosis, in particular to a passive plate-shaped base station antenna fault diagnosis method and system based on sparse sensing. BACKGROUND

[0002] As a core device in modern communication base stations, passive plate-shaped base station antennas are usually composed of large-scale element arrays, and the stability of their electrical performance and mechanical structure is directly related to the coverage capability and communication quality of the system. Existing antenna fault detection methods mostly rely on centralized testing in laboratories, such as high-low temperature cycling, vibration, salt spray, rain test, etc. standard tests, which need to transport samples to a dedicated test center for completion, with long cycle and high cost, and it is difficult to reflect the antenna state in real running environment in time. At the same time, some methods only test under single working condition or static condition, lacking joint simulation of complex communication scenarios and multiple environmental factors, resulting in insufficient representativeness of the detection results.

[0003] On the other hand, with the development of edge computing technology, it is possible to sink the data processing capability to the base station side. If we can combine the advantages of edge computing in near-end real-time processing and analysis, and construct an in-situ fault detection method for passive plate-shaped base station antennas, we can not only shorten the test feedback cycle, but also dynamically evaluate the fault detection indicators of the antenna in the real running environment, thereby providing stronger support for product design improvement, operation and maintenance, and quality traceability.

[0004] For example, the Chinese patent application with publication number CN119109530A proposes a phased array antenna fault detection method based on sparse signal reconstruction algorithm, the implementation steps are: initializing parameters; constructing amplitude square data vector; constructing sparse signal reconstruction model based on sparse signal reconstruction algorithm; solving the sparse signal reconstruction model; obtaining phased array antenna fault detection result. The invention is based on sparse signal reconstruction algorithm, by introducing the sparsity prior knowledge of sparse signal matrix in the signal reconstruction model rewritten by fault factors to construct the sparse signal reconstruction model, and using convex optimization algorithm to solve it, using the solving result to judge the fault state of the antenna unit, since the sparsity prior knowledge is introduced in the objective function, additional guidance information is added in the solving process, without more measurement data, compared with the prior art, the detection efficiency is improved under the premise of ensuring the detection accuracy.

[0005] The above existing technologies all have the problems proposed in the background: it is difficult to effectively extract atypical fault features, and misjudgment or missed detection is easy to occur. To solve the above problems, the present application designs a passive plate-shaped base station antenna fault diagnosis method and system based on sparse sensing. SUMMARY

[0006] The technical problem this application aims to solve is to address the shortcomings of existing technologies by providing a method and system for fault diagnosis of passive plate base station antennas based on sparse sensing. This method involves acquiring the transmit response sequences of a passive plate base station antenna under multiple directions, frequencies, and times; calculating the element response behavior matrix using an element shaping contribution model; performing sparse sensing processing on the element response behavior matrix to extract element regions with abnormal orbital spread characteristics, generating an abnormal sparse matrix; and identifying and marking faulty elements based on abnormal cumulative values. This application uses a tension map driven by the orbital spread index to focus on the modeling region and constructs a sparse sensing model with regularized weighting factors to achieve lightweight diagnosis of abnormal element behavior, improving fault location efficiency and edge adaptability.

[0007] To achieve the above objectives, this application provides the following technical solution:

[0008] A method for fault diagnosis of passive plate-shaped base station antennas based on sparse sensing, the method comprising:

[0009] A pole-mounted fixture is used to mount a passive plate-shaped base station antenna on a vibration table.

[0010] The passive plate base station antenna is tested by the power supply of the electronically regulated module. The test includes applying a specified number of impacts to the passive plate base station antenna in both positive and negative directions on three mutually perpendicular axes according to a preset test severity level.

[0011] The transmission response sequence of the passive plate base station antenna under multiple directions, frequencies, and times is obtained. Combined with the preset reliability evaluation criteria, the fault detection result of the passive plate base station antenna under the impact test is determined.

[0012] The passive plate-shaped base station antenna is arranged at the same angle as the preset maximum tilt angle.

[0013] The impact interval between two adjacent tests is adjusted according to the degree of impact response.

[0014] Based on the transmission response sequence and a preset array element shaping contribution model, the array element response behavior matrix is ​​calculated; the array element response behavior matrix is ​​subjected to sparse sensing processing to obtain an abnormal sparse matrix, wherein the sparse sensing processing identifies array element response regions with abnormal sparse structures in the array element response behavior matrix based on the behavioral characteristics of each array element; and the fault detection result of the passive plate base station antenna is determined based on the abnormal sparse matrix.

[0015] The element response behavior matrix is ​​subjected to sparse sensing processing to obtain an anomalously sparse matrix, including:

[0016] The response value of each element in the element response behavior matrix is ​​subjected to delayed embedding processing to generate the corresponding high-dimensional behavior trajectory;

[0017] calculating a trajectory expansion index according to the expansion degree of the high-dimensional behavior trajectory in the embedding space, wherein the trajectory expansion index comprises at least one of trajectory envelope volume, principal component energy distribution or manifold dimension estimation;

[0018] calculating a sparse-aware candidate region according to the trajectory expansion index, performing sparse-aware on the sparse-aware candidate region to generate an anomaly sparse matrix.

[0019] performing delay embedding processing on the response value of each array element in the array element response behavior matrix to generate a corresponding high-dimensional behavior trajectory, comprising:

[0020] a response sequence of the response value of each array element in the time dimension, with a preset embedding dimension d and a time delay step constructing an embedding vector sequence, wherein the i th embedding vector of the embedding vector sequence is:

[0021] ,

[0022] wherein, represents the i th embedding vector, represents the response value at the i th time point;

[0023] arranging the embedding vector sequence in space to generate a point set trajectory, which constitutes the high-dimensional response behavior trajectory of the corresponding array element, wherein represents a d-dimensional real number.

[0024] calculating a sparse-aware candidate region according to the trajectory expansion index, performing sparse-aware on the sparse-aware candidate region to generate an anomaly sparse matrix, comprising:

[0025] generating an array behavior structure tension map according to the trajectory expansion index, wherein the tension value of each array element is calculated according to the gradient difference of the trajectory expansion index between the array element and the adjacent array element;

[0026] processing the array behavior structure tension map to take the tension concentration region as a sparse-aware candidate sub-domain;

[0027] configuring corresponding sparse-aware processing parameters according to the trajectory expansion index corresponding to the center sub-domain of the sparse-aware candidate sub-domain, wherein the sparse-aware processing parameters comprise scale factor of sparse structure extraction, sparse constraint strength and region-aware window;

[0028] The sparse sensing candidate subdomain is sparsely sensed according to the sparse sensing processing parameters to obtain a sparse structure region. The array elements corresponding to the sparse structure region are summarized to obtain an abnormally sparse matrix. The sparse sensing is based on a preset sparse sensing model, which includes a rank constraint term and a sparse structure term. The sparse sensing embeds the regularization weighting factor of the sparse structure term into the sparse sensing processing parameters and solves the sparse sensing model.

[0029] Based on the orbital unfolding index corresponding to the central subdomain of the sparse sensing candidate subdomain, configure the corresponding sparse sensing processing parameters, including:

[0030] A sparse sensing weight matrix is ​​generated based on the orbital expansion index of the central array element in the sparse sensing candidate subdomain.

[0031] Based on the sparse sensing weight matrix and the objective function of the sparse sensing model, the size of the region sensing window is calculated using the local mean and variance of the orbital spread index. The scale factor is calculated using the gradient vector of the orbital spread index, and the sparse constraint strength is calculated using the value range of the orbital spread index.

[0032] The array element shaping contribution model is constructed using historical measurement data and includes a mapping matrix used to map the observation direction response to the response of each array element.

[0033] The computational matrix of element response behavior includes:

[0034] Flatten the transmission response sequence according to the direction and frequency to obtain the observation vector corresponding to each time point;

[0035] The response value of the corresponding time-point matrix element is obtained by inverting the observation vector through the mapping relationship matrix.

[0036] The response values ​​are arranged according to the element numbers to obtain the element response behavior matrix.

[0037] Based on the anomalous sparse matrix, the fault detection result of the passive plate base station antenna is determined, including:

[0038] Statistical analysis is performed on the sparse feature strengths corresponding to the remaining array elements in the abnormal sparse matrix, and the abnormal cumulative value of each array element is calculated over the entire time window.

[0039] The accumulated abnormal value is compared with a preset intensity threshold to determine whether the accumulated abnormal value meets the reliability assessment criteria.

[0040] The array element meeting the reliability evaluation criterion is marked as a faulty array element, and the failure detection result of the passive plate-shaped base station antenna during the entire impact test is calculated according to the cumulative eigenvalue;

[0041] The reliability evaluation criterion at least includes one of the following:

[0042] The abnormal cumulative value of the array element in the preset time window is greater than or equal to a first intensity threshold value;

[0043] The abnormal response value of the array element is greater than or equal to a second intensity threshold value at a plurality of continuous time points;

[0044] The difference between the abnormal response value of the array element and the average response value of its adjacent array element is greater than or equal to a third spatial difference threshold value;

[0045] The sparse feature distribution of the abnormal sparse matrix corresponding to the array element accounts for more than or equal to a fourth regional abnormal proportion threshold value in the array area to which it belongs.

[0046] A passive plate-shaped base station antenna fault diagnosis system based on sparse sensing, the system comprising:

[0047] A response acquisition module for acquiring the transmission response sequence of the passive plate-shaped base station antenna in multiple directions, frequency points and time, and pre-processing the observation data;

[0048] A behavior modeling module for deconstructing the transmission response sequence in combination with a preset array element shaping contribution model, generating an array element response behavior matrix, and performing delay embedding processing on the response value of each array element, constructing a high-dimensional behavior track and calculating a track expansion degree index;

[0049] A sparse sensing module for constructing an array behavior structure tension diagram based on the track expansion degree index, determining a sparse sensing candidate sub-domain, configuring sparse sensing processing parameters, and extracting an abnormal sparse matrix;

[0050] A fault detection module for statistically analyzing the array elements according to the abnormal sparse matrix, marking the faulty array elements according to the preset reliability evaluation criterion, and generating the fault detection result of the passive plate-shaped base station antenna.

[0051] Compared with the prior art, the beneficial effects of the present application are:

[0052] On the one hand, by constructing an array element response behavior matrix and performing delay embedding processing, the time behavior information of the array element can be fully retained, the track expansion degree index is introduced to represent the high-dimensional evolution characteristics of the response track, and the recognition ability of the nonlinear and non-steady state fault mode is improved.

[0053] On the other hand, the proposed array behavior structure tension diagram is used for identifying the unfolding degree mutation area, realizing dynamic focusing on the local abnormal area, effectively reducing the spatial redundancy of sparse modeling, and significantly reducing the computing load of the edge device. BRIEF DESCRIPTION OF DRAWINGS

[0054] Other features, objects, and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments, when read in conjunction with the accompanying drawings:

[0055] Figure 1 An exemplary application scenario of an embodiment of the application is shown in the following figure:

[0056] Figure 2 An antenna fault detection principle diagram based on a sparse signal reconstruction algorithm of an embodiment of the application is shown in the following figure:

[0057] Figure 3 A flowchart of a passive plate-shaped base station antenna fault diagnosis method based on sparse sensing of an embodiment of the application is shown in the following figure:

[0058] Figure 4 A flowchart of sparse sensing processing of an embodiment of the application is shown in the following figure:

[0059] Figure 5 A high-dimensional response behavior track diagram of an embodiment of the application is shown in the following figure:

[0060] Figure 6 A generation principle diagram of an array behavior structure tension diagram of an embodiment of the application is shown in the following figure. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments.

[0062] In this document, the term“embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor does it necessarily refer to a particular alternative embodiment. It will be explicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0063] In the example technology, the method of the present application is not only suitable for conventional array data analysis, but also can be directly embedded into the reliability test flow of the passive plate-shaped base station antenna.

[0064] To better illustrate the application scenario of the method of the present application, a complete embodiment is given below in combination with reliability test standards. The embodiment takes a passive plate-shaped base station antenna as the test object, and performs loading and evaluation according to a preset reliability test procedure under laboratory conditions. It should be understood that the test procedure and method of the embodiment can be appropriately adjusted according to different product models and use environments.

[0065] In one specific embodiment, the antenna to be tested is first installed on a dedicated test tool, such as a pole clamp of a vibration table, an impact test table platform, a high-low temperature environmental test box, a rain test device, etc. The antenna installation method should be consistent with the actual engineering installation to ensure that the stress and constraint conditions are reasonable. Before the test, initial detection of the appearance, electrical performance and mechanical performance of the antenna is required to compare with the detection results after the test.

[0066] The first is a low-temperature working test:

[0067] The test condition is a temperature of −40℃, a duration of 16 hours, and a temperature change rate of not more than 1℃ / min. For special market needs, the low-temperature condition can be increased to −55℃. After the antenna is exposed to a stable temperature in the low-temperature box, electrical performance testing is performed, and the electrical adjustment module is kept working normally. During the test, it should be ensured that the electrical downtilt angle adjustment function is available, and there is no alarm in communication. After the test, it is checked whether there are cracks and deformation in the appearance, and whether the electrical performance (vswr, isolation, passive intermodulation) meets the technical index.

[0068] The second is a high-temperature working test:

[0069] The test condition is a temperature of +60℃, a duration of 16 hours, and a temperature change rate of not more than 1℃ / min. The test method is the same as the low-temperature working test, and the final pass criteria also include no defects in appearance, electrical performance meeting the index, and stable mechanical structure.

[0070] The third is a temperature cycle test:

[0071] The test condition is alternating cycles of −40℃ and +60℃, a temperature change rate of ≥1℃ / min, a temperature maintained for 3 hours at each temperature, and 6 cycles. During the entire cycle process, electrical performance and electrical downtilt angle adjustment tests should be performed regularly to ensure that the electrical adjustment function can work repeatedly under extreme conditions. After the test is completed, the appearance, electrical performance and mechanical performance are checked, and the judgment criteria are consistent with those of the low-temperature working test.

[0072] The fourth is an alternating damp-heat test:

[0073] Test conditions are temperature +55℃, relative humidity ≥95%, and the cycle is repeated twice. The specific process includes humidification at 25℃ to ≥95%, followed by an increase to 55℃ within 3 hours and a 12-hour maintenance. Throughout the cycle, the relative humidity must be maintained at ≥90%. The antenna is kept working during the test, and after the test, the electrical adjustment function is checked to see if it is normal, there is no crack or falling off, and the electrical performance meets the requirements.

[0074] It can be understood that an initial detection is required before all tests, including:

[0075] Appearance inspection: no cracking, deformation, and surface coating peeling off for radome body, end cover, and sealing glue;

[0076] Electrical performance: test parameters such as standing wave ratio, isolation, and passive intermodulation (PIM);

[0077] Mechanical performance: including structural integrity and normality of electrical downtilt angle adjustment.

[0078] In addition to temperature and humidity environmental stress, the mechanical bearing capacity of the antenna also needs to be verified in the test. The sinusoidal vibration test examines the reliability of the antenna when it encounters periodic mechanical vibration during transportation or use by setting the frequency range of reciprocating vibration. The random vibration test simulates the complex and variable vibration environment during transportation by wide-spectrum random acceleration input to verify the structural strength and reliability of the connecting parts.

[0079] In terms of mechanical impact, impact testing is one of the focuses of reliability testing. The antenna bears a specified number of half-sine wave impacts in three mutually perpendicular directions, with a peak acceleration of up to 10g and a pulse duration of 6ms. After the test, the antenna's appearance, electrical performance, and mechanical connections are comprehensively checked to ensure that it can still work normally after repeated impacts. Unlike the traditional method of relying only on appearance and electrical detection, the scheme of the present application emphasizes monitoring the response behavior during impact loading, thereby providing more dynamic and detailed data support for reliability evaluation.

[0080] In addition, wind load tests and rain tests can also be carried out. The wind load test verifies the structural reliability of the antenna under high wind conditions by applying aerodynamic pressure at the design wind speed or even extreme wind speed in a wind tunnel environment. The rain test examines the waterproof performance and sealing reliability of the antenna shell by simulating continuous rainfall conditions to ensure that there is no performance degradation due to water seepage in long-term outdoor environments.

[0081] In a specific embodiment of the impact test, the antenna to be tested is installed in a test tool, such as a vibration table, an impact platform, or a high-low temperature environmental chamber, and is subjected to multi-condition loading according to a preset test severity level (including temperature cycling, sinusoidal / random vibration, mechanical impact, and salt spray and rain). During the loading process, the antenna's emission response sequence in multiple directions, frequency points, and time is collected in real time, and the collected data is recorded and preprocessed synchronously.

[0082] In a specific example, the test conditions need to meet:

[0083] Impact waveform: half-sine wave;

[0084] Peak acceleration: 10g;

[0085] Pulse duration: 6ms;

[0086] Test times: 150 times in the ±X and ±Y axis directions, and 200 times in the ±Z axis direction, for a total of 1000 times.

[0087] The present application is mainly applied to reliability tests of impact, and in a specific embodiment, a passive plate-shaped base station antenna is installed on a holding pole tool of a vibration table, and mechanical impact of a specified number of times is applied in the positive and negative directions of the three mutually perpendicular axes according to a preset impact test severity level. After each impact, the antenna's emission response sequence in different directions, frequency points, and time is obtained through a response acquisition module. Compared with existing impact tests based only on appearance inspection or electrical indicators, the present application can extract and analyze the dynamic behavior characteristics of the array elements in real time during the impact loading process, thereby providing more detailed, dynamic, and quantifiable reliability evaluation results.

[0088] The present application is applicable to base station antenna systems with dense distribution characteristics of array structures and multi-dimensional response paths of emission behavior, and is particularly suitable for real-time monitoring scenarios deployed in edge computing environments.

[0089] It should be noted that the test method of the present application does not depend on a specific array element arrangement structure, a static electrical performance model, or historical template samples, but rather constructs a high-dimensional behavior track and calculates its spread index to guide subsequent abnormal structure focusing and sparse modeling.

[0090] It can be understood that the sparse perception logic of the present application is not based on hard-coded judgments of array element positions or single numerical thresholds, but rather guides the sparse extraction path of the corresponding array element through the structural evolution characteristics of the response behavior.

[0091] The selection of the application scenario of the present application is determined based on the common characteristics of antennas, and the selected application scenario needs to have at least one of the following characteristics:

[0092] The response behavior of the array system presents irregular disturbance, which is difficult to be distinguished by linear modeling;

[0093] The fault influence range has spatial local aggregation but time-domain polymorphism of the array structure;

[0094] The perception computing resource is limited, and the deployment demand needs to be dynamically adjusted in the sparse extraction area or intensity;

[0095] The fault behavior under the same array structure presents cooperative behavior disturbance between a small number of abnormal array elements.

[0096] Referring to Figure 1 , the figure is a schematic diagram of an exemplary application scenario provided by the embodiment of the application.

[0097] Figure 1 It is shown that a certain passive plate-shaped base station antenna includes 9 array elements, wherein the array element represents an independently operable transmitting unit, and in the normal working state, the array element can emit a response signal according to the set direction, frequency point and time sequence. Figure 1 The medium gray color represents the fault array element, which is used to represent the deviation of the response characteristics from the normal array element.

[0098] Figure 1 It is shown that the passive plate-shaped base station antenna generates a transmitting response sequence in the working process, the transmitting response sequence is received by the edge device, and the transmitting response sequence is processed to determine the position of the fault array element, and the edge device can be deployed at the base station site or the near-end computing node.

[0099] It can be understood that the function of the edge device of the application is to:

[0100] Model the response behavior of the passive plate-shaped base station antenna, extract the behavior trajectory of each array element in the time-frequency direction, and construct a structure anomaly detection model based on the trajectory spread index, so as to realize the rapid determination of the local fault array element without full-array traversal analysis.

[0101] In different implementation environments, the edge device can be an embedded diagnostic module deployed on the site control unit, or an edge intelligent terminal with local computing capability, such as a small industrial computing box with GPU, an FPGA processing module or an edge inference device supporting AI acceleration, etc. The application is not limited to a specific hardware form.

[0102] Taking sparse perception as an example, it can be understood by referring to Figure 2 , Figure 2 The antenna fault detection principle diagram based on the sparse signal reconstruction algorithm of the embodiment of the application is used to assist in understanding the limitations of the existing sparse perception scheme in engineering application.

[0103] Figure 2It is shown that the current antenna fault detection by sparse sensing is generally based on the sparsity prior knowledge obtained from the running records of the antenna elements, and assumes that the faulty elements are sparsely distributed in space.

[0104] Figure 2 It is shown that the current sparse sensing usually adopts a modeling method based on historical response data of the antenna elements to construct sparsity prior knowledge, and assumes that the faulty elements are sparsely distributed in the entire spatial array, so as to realize the extraction of the fault position by solving an optimization model with a sparse constraint term.

[0105] Figure 2 It is shown that some faulty elements may not be identified or ignored in the current sparse optimization, and need to be further solved.

[0106] It can be understood that the elements of the antenna exist independently, and even if a faulty element interferes with other elements, the interference is limited to the adjacent range and does not spread widely. Generally, the fault state indeed presents a sparse distribution, which is the reason why sparse sensing is particularly suitable for antenna fault detection.

[0107] It is easy to understand that the so-called sparsity prior knowledge in sparse sensing is often to encode the statistical characteristics of the element response in the historical time series in a certain specific way, such as the position of low-frequency abnormal points, the difference of average energy distribution, or the principal component residual structure, and then use it as a structural constraint term in the antenna signal reconstruction model to strengthen the sparsity assumption of abnormal behavior in future observations.

[0108] Those skilled in the art understand that there are obvious limitations in prior knowledge:

[0109] Firstly, the modeling of the fault structure strongly depends on the historical state, and it is difficult to adapt to the real scene where the antenna operating state frequently changes and the fault distribution pattern is unstable.

[0110] Secondly, the sparse sensing based on prior knowledge still needs to model the entire array uniformly, but selectively skips some elements. As the working time of the antenna gradually increases, the prior knowledge will inevitably increase to cover all elements.

[0111] Thirdly, in the edge computing scene, the global modeling of a large number of elements will bring an unbearable computing burden and real-time bottleneck.

[0112] Next, in combination with the drawings, the method for diagnosing passive plate-shaped base station antenna faults based on sparse sensing provided by the embodiments of the present application is introduced. Figure 3 The method shown includes the following S1-S6, and the specific steps are as follows:

[0113] S1: arranging a passive plate-shaped base station antenna on a derrick tool of a vibration table;

[0114] S2: testing the passive plate-shaped base station antenna by an electrically adjustable module power supply, wherein the testing comprises applying a predetermined number of impacts in two directions of positive and negative on the passive plate-shaped base station antenna in three mutually perpendicular axial directions according to a preset test severity level;

[0115] S3: obtaining a transmission response sequence of the passive plate-shaped base station antenna in multiple directions, frequency points and time points;

[0116] In the embodiment, the transmission response sequence is formed by collecting the transmission responses of the passive plate-shaped base station antenna in different observation directions, frequency points and multiple time sampling points in the actual working environment, and is used to represent the behavior responses of each array element in the time-frequency joint space. The running data can be obtained in real time through network synchronization, field sampling, control instruction issuing and the like, and the specific data types include signal level indicators such as transmission power, echo intensity and signal-to-noise ratio, which only need to meet the minimum requirement of extracting the response state of the array element, and the present application does not make a specific limitation thereon.

[0117] It can be understood that the transmission response sequence of the present application is organized in the form of a three-dimensional tensor, and the dimensions correspond to the observation direction , frequency sampling point and time sequence step in turn, and the form can be represented as , wherein represents the transmission response sequence corresponding to the Zth array element, and the value at each position is the transmission response intensity observed at the corresponding time point in the direction and frequency.

[0118] S4: calculating an array element response behavior matrix according to the transmission response sequence and in combination with a preset array element shaping contribution model;

[0119] In the embodiment, the transmission response sequence is input into the preset array element shaping contribution model for joint processing, the array element shaping contribution model is established through historical measurement data or calibration information, and is used to express the mapping relationship from the direction response space to the array element response space. The mapping relationship can be represented by a linear projection matrix or a nonlinear neural approximation model, and through the model, the observation vector at each time point can be decomposed into the response components of the array internal array elements, and the array element response behavior matrix is reconstructed.

[0120] It can be understood that in the present application, the array element response behavior matrix is organized in the form of a two-dimensional real value matrix, and is represented as , wherein A represents the array element response behavior matrix, n represents the number of array elements, T represents the time point, and each element in the matrix is the response value of the nth array element at the Tth time point.

[0121] S5: performing sparse-aware processing on the element response behavior matrix to obtain an abnormal sparse matrix;

[0122] In this embodiment, for the element response sequence, a delay embedding is used to construct a high-dimensional behavior track, and a track unfolding degree index in the embedding space is calculated, which includes but is not limited to track envelope volume, principal component energy proportion, or manifold dimension estimation, for measuring the structural complexity of the element behavior. Then, an array behavior structure tension map is generated based on the difference in the track unfolding degree index, and the tension concentrated area is identified as a candidate sub-domain.

[0123] Further, the candidate sub-domain is configured with a perception processing parameter, including a perception window, a sparse penalty factor, and a scale extraction strategy, and an abnormal sparse matrix is obtained by solving a sparse structure modeling with a weight control term. The abnormal sparse matrix concentrates the response area where the behavior difference between the elements exists, to represent the potential fault structure.

[0124] S6: determining a fault detection result of the passive plate-shaped base station antenna according to the abnormal sparse matrix;

[0125] In this embodiment, the sparse strength of each element in the abnormal sparse matrix is counted, and an abnormal accumulation value in a preset time window is calculated, and then combined with multi-dimensional indexes such as response difference between adjacent elements and spatial abnormality proportion, a multi-factor reliability evaluation criterion is constructed. If the element meets one or more judgment threshold conditions, it can be identified as a fault element.

[0126] The threshold value used in this step can be a preset static threshold value, or can be self-adaptively adjusted by a historical observation distribution. In actual deployment, it can be dynamically modified in combination with engineering experience or an online learning mechanism, which does not constitute a limitation to the method structure.

[0127] It is easy to understand that the passive plate-shaped base station antenna fault diagnosis method based on sparse awareness proposed in the present application is based on such an engineering common sense: in a large-scale array structure, most of the elements are in a normal working state, and only a few elements show abnormal behavior in a specific direction, frequency or time. This structural sparse distribution feature provides a feasible basis for fast and reliable diagnosis.

[0128] The difficulty lies in that if the global modeling or sparse reconstruction is directly performed on the whole array response data, the high-dimensional nested evolution characteristics of the actual fault behavior in the response space will be ignored, and the difference between the behavior tracks cannot be described by a single numerical value or template, resulting in low modeling efficiency and high diagnosis delay, especially for the power limited problem in the edge device scene.

[0129] The application guides the sparse perception model to focus on modeling in the fault area through the structural index of the behavior track spread degree, thereby achieving an engineering balance between precision and efficiency, and solving the actual needs of real-time, high-density, and low-cost deployment under the array antenna scene.

[0130] Next, the part of the method of the application related to calculating the array element response behavior matrix is further expanded.

[0131] In one example, the array element shaping contribution model is constructed through historical measurement data, including a mapping relationship matrix for mapping the observed direction response to each array element response.

[0132] It can be understood that the array element shaping contribution model can be an empirical projection model trained based on multi-scene calibration data, or a directional coupling parameter model obtained through electromagnetic simulation or array calibration test, and the mapping relationship matrix is a prerequisite for reconstructing the array element response behavior matrix of the application, which is used to ensure that the response sequence of each array element in the time domain is accurately decomposed from the direction and frequency observation data.

[0133] As understood by those skilled in the art, the mapping relationship matrix can be derived from offline measurement samples under typical working conditions, feature mapping learning results of historical operation data, or a theoretical directional diagram model calculated from structural design parameters, as long as it can satisfy the spatial back-projection of multi-dimensional observation data and has engineering realizability, which is not limited by the application.

[0134] In one example, the specific steps of S2 are as follows:

[0135] S2.1: flattening the transmit response sequence according to the direction and frequency point to obtain an observation vector corresponding to each time point;

[0136] Specifically, in order to realize the mapping operation from the observation domain to the array element domain, it is necessary to perform flattening processing on the original transmit response sequence in structure, and convert its multi-dimensional format into a vector representation conforming to the matrix multiplication structure requirement. The transmit response sequence is a data structure organized in three dimensions of direction, frequency and time, and the two-dimensional slice of each time point is the signal response set under all observation directions and frequency combinations at the current time. By expanding the two-dimensional slice into a one-dimensional column vector, a standard linear multiplication relationship can be established between the vector and the preset mapping relationship matrix, thereby meeting the subsequent behavior inversion modeling requirements in the array element space.

[0137] In this embodiment, the row-first manner is adopted to perform flattening operation on the two-dimensional matrix corresponding to the direction and frequency. The response values of different frequency points under all observation directions corresponding to the current time point are sequentially spliced into a single column vector, thereby constructing an observation vector.

[0138] Furthermore, to ensure that the flattened structure of response sequences at different time points is consistent, the flattening operation relies on a fixed direction and frequency index order, and a unified index mapping table is constructed in the data preprocessing stage to ensure dimensional alignment and semantic preservation during vector construction.

[0139] S2.2: Invert the observation vector using the mapping matrix to obtain the response value of the corresponding time-point matrix element;

[0140] Specifically, the role of the mapping relationship matrix is ​​to transform the macroscopic response information received from multiple observation angles and frequency points into the independent response value of each array element at the current time point. This can be understood as a back projection process of parameter decoupling.

[0141] Furthermore, in order to avoid the high computational overhead caused by traditional matrix inversion, this embodiment preferably adopts a pseudo-inverse approximation solution strategy to avoid matrix singularity or dimension mismatch problems, and can be compatible with a certain degree of observation loss or signal occlusion.

[0142] S2.3: Arrange the response values ​​according to the array element number to obtain the array element response behavior matrix;

[0143] In this embodiment, the response values ​​are immediately aligned with their assigned numbers after collection. A fixed element number index table is used to bind the element response vector at the current time point to rows, and then insert it into the corresponding column of the element response behavior matrix. To ensure time consistency, the column numbers of the element response behavior matrix correspond one-to-one with the original sampling time axis, ensuring that the embedding dimension continuity condition is met in subsequent orbit generation operations.

[0144] It is understandable that if the response value of a certain matrix element does not exist at a certain time point or multiple time points, the matrix element at the corresponding position should be set to zero.

[0145] Next, we will further elaborate on the part of the method in this application regarding sparse perception.

[0146] Please see Figure 4 , Figure 4 This is a schematic diagram of the sparse sensing processing process in an embodiment of this application. Figure 4 The process shown can be applied to S3 above, and the specific steps are as follows:

[0147] S3.1: Perform delayed embedding processing on the response value of each element in the element response behavior matrix to generate the corresponding high-dimensional behavior trajectory;

[0148] Specifically, in this application, in order to depict the dynamic evolution law of the array element response behavior in the time dimension, the time series of each array element needs to be expanded to a high-dimensional state space to mine its hidden structural information. The original one-dimensional response sequence can be converted into a multi-dimensional trajectory point set by using delay embedding processing. In the embedding space, the array elements in the normal working state will exhibit a certain degree of trajectory expansion and evolution trend, while the array elements that have failed may form local closure, clustering or trajectory contraction and other self-winding behavior characteristics.

[0149] It can be understood that, in the normal working state, the emission behavior of the array element is controlled by the overall array scheduling and working frequency switching of the passive plate-shaped base station antenna, and the response sequence contains rich directivity modulation, frequency coupling change and time domain perturbation, which causes the array element to form a dynamic trajectory structure with good trajectory expansion and obvious direction change in the high-dimensional state space after delay embedding processing. The trajectory points are evenly distributed in multiple dimensions, showing multi-axis expansion behavior characteristics.

[0150] Further, in the case of failure, the response behavior of the array element is often affected by driving failure, coupling interruption or local circuit drift, etc., resulting in a lack of dynamic change in the response sequence, and the time response may be stable near a certain interval value for a long time, or show periodic disorder, noise dominance and other low degree of freedom states. The trajectory generated by such signals after delay embedding will gather in a local area in the embedding space, and the spatial difference between the trajectory points will be weakened, and the path will be cohesive, self-looping, kinked and other geometric shrinkage.

[0151] In this embodiment, delay embedding combines the response values of each array element at consecutive time points into a multi-dimensional vector in time sequence by setting embedding dimension and delay step, forming a corresponding trajectory point set. The point set is arranged in the embedding space to form a behavior trajectory, the embedding dimension is adjusted according to the response change speed of the array element, and the delay step can be set at equal intervals or adaptively to ensure that the trajectory form fully reflects the dynamic characteristics of the response behavior.

[0152] In one example, S3.1 specifically includes:

[0153] S3.1.1: for each array element, a response sequence of response values in the time dimension, a preset embedding dimension d and a time delay step constructing an embedding vector sequence, wherein the i th embedding vector of the embedding vector sequence is:

[0154] ,

[0155] wherein, represents the i th embedding vector, represents the response value at the i th time point;

[0156] S3.1.2: arranging the embedding vector sequence in the embedding space to generate a point set orbit, the point set orbit constituting a high-dimensional response behavior orbit of the corresponding array element, wherein represents a d-dimensional real number;

[0157] For example, assuming that the response value of a certain antenna array element at 9 consecutive time instants is as follows:

[0158] s = [2.1, 2.4, 2.7, 2.9, 3.1, 2.8, 2.5, 2.2, 2.0],

[0159] Assuming that the embedding dimension is 3 and the time delay step is 1, the embedding vector sequence can be referred to as follows:

[0160] ;

[0161] ;

[0162] ;

[0163] ;

[0164] ;

[0165] ;

[0166] ;

[0167] It can be understood that 7 three-dimensional embedding vectors can be generated, each of which represents the local behavior state of the array element at a certain time slice, and can be drawn as a behavior orbit in a three-dimensional space.

[0168] Further, in combination with Figure 5 understanding, Figure 5 is a schematic diagram of a high-dimensional response behavior orbit of an embodiment of the present application, Figure 5 The data is derived from the response values of the foregoing example, specifically a three-dimensional behavior orbit graph based on an embedding dimension of 3 and a step of 1. It can be understood that the orbit trend reflects the behavior change of the array element response in the time evolution process.

[0169] S3.2: calculating an orbit spread index according to the spread degree of the high-dimensional behavior orbit in the embedding space, wherein the orbit spread index includes at least one of a trajectory envelope volume, a principal component energy distribution, or a manifold dimension estimation;

[0170] ​Specifically, in order to extract quantifiable structural indicators from high-dimensional behavior trajectories to distinguish normal and abnormal array element response behaviors, it is necessary to model the geometric unfolding properties of trajectories in the embedding space. The trajectory unfolding degree indicator is used to measure whether the response trajectory of an array element has sufficient state divergence, i.e., whether it can unfold into a geometric trajectory with multiple degrees of freedom and directionality in high-dimensional space during time evolution. If the trajectory has significant contraction, winding or highly repetitive behavior, the geometric dimension and coverage range of the trajectory in the embedding space will be significantly reduced, which is one of the typical manifestations of fault states.

[0171] In the present embodiment, the trajectory unfolding degree indicator includes the following three types of representative methods, which can be used independently or in combination:

[0172] In one way, the envelope volume of the trajectory is calculated to represent the spatial range occupied by the trajectory point set in the embedding space.

[0173] In an optional embodiment, the envelope volume is calculated as follows:

[0174] The volume of the polyhedral volume or the boundary rectangle formed by the minimum convex hull corresponding to the trajectory point set is calculated to reflect whether the trajectory exhibits extensive divergence, changes in direction, etc.

[0175] It can be understood that the larger the envelope volume, the more state changes and higher complexity of the array element response in time series; on the contrary, if the trajectory is curled, repeated or concentrated in a certain area, the volume tends to shrink. As a geometric shape measure, the envelope volume can directly reflect the distribution scale of the response behavior in the embedding space.

[0176] In another way, the principal component energy distribution ratio of the trajectory point set is calculated to measure whether the trajectory changes are concentrated in a few dimensions.

[0177] In an optional embodiment, the principal component energy distribution ratio is calculated as follows:

[0178] The principal component analysis is performed on the trajectory point set, the proportions of the projection energies on the first and second principal components are counted, and the principal component energy distribution ratio is obtained by weighted summation.

[0179] It can be understood that if the proportions corresponding to the first and second principal components are very high, it means that the changes of the trajectory are concentrated in a single direction, and the behavior lacks variability in other dimensions, which may be an abnormal compressed trajectory; on the contrary, if the energy is uniformly distributed in multiple dimensions, it means that the behavior trajectory is multi-axis unfolded and has high complexity.

[0180] In yet another way, the manifold dimension estimation of the trajectory is performed to characterize the intrinsic dimension of the subspace to which the trajectory is attached.

[0181] In an optional embodiment, the specific way of estimating the manifold dimension is as follows:

[0182] Based on the neighborhood search and local linear fitting, the number of degrees of freedom of the trajectory point set is estimated. The lower the dimension, the more likely the trajectory is to wrap around a low-dimensional manifold, which may be a dynamic behavior reduction caused by a fault state; while the higher the dimension, the more real evolution process the trajectory has in multiple dimensions, which is a typical feature of normal state.

[0183] It can be understood that the orbit spread index in the present application is not an action attribute expressed in a literal label or an empirical classification method, but a continuous and quantifiable orbit complexity eigenvalue directly obtained by numerical calculation method based on the geometric structure and distribution characteristics of the embedded orbit point set, which has a clear mathematical meaning and engineering interpretation basis.

[0184] S3.3: calculating a sparse-aware candidate region according to the orbit spread index, performing sparse awareness on the sparse-aware candidate region, and generating an abnormal sparse matrix;

[0185] Specifically, in order to avoid the waste of computing resources and the risk of misjudgment caused by directly performing sparse reconstruction on the full array response data, the present application constructs a structural tension distribution map based on the orbit spread index of each element in the embedding space, so as to focus on the local area with larger behavior structure change, and preferably perform sparse awareness operation.

[0186] It can be understood that the present application preferentially extracts the position concentratedly embodied by the complex behavior structure as a candidate sub-domain, so as to reduce the model solving dimension and improve the recognition sensitivity to local fault features.

[0187] In an example, the specific steps of S3.3 are as follows:

[0188] S3.3.1: generating an array behavior structure tension map according to the orbit spread index, wherein the tension value of each element is calculated according to the gradient difference of the orbit spread index between the element and the adjacent element;

[0189] For example, the array behavior structure tension map can be understood with reference to Figure 6 Figure 6 The generation principle diagram of the array behavior structure tension map of the embodiment of the present application is as follows: Figure 6 ​The track spread index of a certain passive plate-shaped base station antenna corresponding to all array elements is shown, in the illustration, a central array element is selected as the current calculation object, and the track spread index of the four spatial neighborhood array elements above, below, left and right is compared to form a direction gradient calculation relationship. It should be noted that these track spread indexes are derived from the time behavior trajectory delay embedding calculation of each array element, and there is no behavior interference or coupling between the array elements, only the independent variable.

[0190] Further, according to the calculation of the direction gradient, the tension value corresponding to the array element is generated, and the corresponding array behavior structure tension diagram is displayed.

[0191] It can be understood that, Figure 6 It is only a reference example and does not represent the numerical scale or visual proportion relationship of the real array behavior structure tension diagram, and the purpose is to illustrate that the present application constructs the tension value distribution diagram by spatial gradient analysis of the track spread index of each array element at the structural level, so as to identify the region where the behavior compression, track contraction or abnormal clustering may exist.

[0192] Specifically, in order to focus the calculation resources of fault detection on the positions where the behavior anomalies are more likely to exist in the array, the present application identifies the region where the track spread index of the array element changes dramatically in the spatial distribution by constructing the array behavior structure tension diagram. It should be particularly noted that the present application assumes that each array element does not interfere with each other at the physical signal level, and the track spread index is only determined by the response behavior trajectory of each array element itself, and does not depend on the behavior state of other array elements, nor constitutes a behavior coupling. Therefore, the spatial gradient in the tension diagram does not represent the signal conduction relationship between the array elements, but a structural mutation detection method based on the independent index in the spatial array distribution.

[0193] In this embodiment, first, the track spread index value corresponding to each array element is calculated to form a two-dimensional index distribution diagram consistent with the spatial position of the array element. Then, around each array element, the track spread index values in its spatial neighborhood are selected, and the gradient changes in multiple directions are calculated, and the difference, Laplace operator or graph structure difference method is used to form the local tension value of the corresponding position in the tension diagram. The tension value essentially reflects the relative difference degree of the behavior complexity of a certain array element and the surrounding array elements, rather than whether they produce physical behavior interference.

[0194] Further, the application believes that in a real engineering scenario, most array elements in the array usually have similar response patterns under normal working conditions, and the track development degree index distribution usually belongs to the same statistical distribution section. Therefore, if the development degree index value at a certain position deviates significantly from the average value of its neighborhood, it is more likely to be caused by track abnormalities caused by faults or behavior compression. This local mutation is precisely manifested as a significant increase in local tension in the tension diagram, which can be considered as a signal focusing area of potential structural abnormalities.

[0195] S3.3.2: processing the array behavior structure tension diagram, and taking the tension concentration area as a sparse perception candidate sub-domain;

[0196] Specifically, in order to concentrate modeling resources on positions where the track behavior has structural changes, it is necessary to extract high-tension areas from the behavior structure tension diagram as candidate sub-domains for local modeling. This processing process can be considered as a pre-focusing of sparse reconstruction modeling. The tension concentration area is often formed around the behavior development abnormality, manifold dimension collapse or principal component degradation area, which is the physical mapping area of fault evolution, and has engineering rationality as a perception modeling entry.

[0197] In the embodiment, the screening process of the candidate sub-domain includes two stages: first, the local maximum value search and connected domain marking algorithm is applied on the tension diagram to identify the core pixel point area with strong tension concentration effect; second, based on each high-tension point, the area with a neighborhood tension value greater than the median value is expanded to form a sub-domain boundary. Each sub-domain can be a fixed-size sliding window, or an irregular shape structure adaptively expanded under the driving of a tension threshold, and finally a list of multiple non-overlapping or partially overlapping candidate sub-domains is formed.

[0198] Further, the candidate sub-domain will enter the sparse perception process according to the priority in subsequent modeling. The priority can be sorted according to the average tension of the sub-domain center or the internal track development degree entropy, so as to realize the modeling focus of the model on the behavior unstable area, thereby improving the calculation efficiency and fault positioning sensitivity of the overall algorithm in the engineering deployment scenario.

[0199] S3.3.3: configuring corresponding sparse perception processing parameters according to the track development degree index of the center sub-domain corresponding to the sparse perception candidate sub-domain, wherein the sparse perception processing parameters include a scale factor of sparse structure extraction, a sparse constraint strength and a region perception window, and configuring corresponding sparse perception processing parameters includes:

[0200] generating a sparse perception weight matrix according to the track development degree index of the center element in the sparse perception candidate sub-domain;

[0201] Based on the sparse sensing weight matrix and the objective function of the sparse sensing model, the size of the region sensing window is calculated using the local mean and variance of the orbital spread index, the scale factor is calculated using the gradient vector of the orbital spread index, and the sparse constraint strength is calculated using the value range of the orbital spread index.

[0202] Specifically, to adapt the sparse modeling process to local behavioral characteristics, dynamic sensing parameters need to be introduced at the subdomain level, including the penalty strength of sparse structures, extraction scale, and modeling window, thereby avoiding a one-size-fits-all uniform modeling rule for all regions. This application generates a sparse sensing weight matrix through the orbital unfolding index of the central array element, which is used to embed behavioral structure differences into the model regularization term, so that the modeling process reflects the structural guidance of real behavior.

[0203] In this embodiment, the sparse sensing weight matrix is ​​constructed as follows: using the central element of the candidate subdomain as the anchor point, its local mean, variance, and gradient direction on the orbital spread index are extracted to construct three sets of control factors. The local mean is used to calculate the sensing window size; the higher the orbital spread index, the larger the window, and the wider the range of behavioral changes it can adapt to. The local variance reflects behavioral instability and is used to control the strength of the sparse penalty; the stronger the fluctuation, the weaker the penalty, avoiding the false suppression of real fluctuation signals. The gradient direction is mapped to the structural scale factor and is used to adjust the connectivity assumption of the behavioral structure during modeling. These three types of parameters are integrated to form a multi-channel weight matrix, which is embedded into the sparse structure term of the model objective function.

[0204] Furthermore, the weight matrix not only takes effect in the initial modeling stage, but is also adaptively updated during the iterative optimization process based on the modeling residuals or model fitting entropy, achieving dynamic response adjustment of the repulsion factor. In this way, sparse modeling no longer relies on static rules, but is dynamically coupled with the actual response behavior, ensuring adaptability and structural expressiveness under different states, which is especially suitable for application scenarios in edge computing devices where both computing resources and response accuracy are important.

[0205] S3.3.4: Perform sparse sensing on the sparse sensing candidate subdomain according to the sparse sensing processing parameters to obtain sparse structure regions, summarize the array elements corresponding to the sparse structure regions to obtain an abnormally sparse matrix, wherein the sparse sensing is based on a preset sparse sensing model, the sparse sensing model includes a rank constraint term and a sparse structure term, and the sparse sensing embeds the regularization weighting factor of the sparse sensing processing parameters into the sparse structure term to solve the sparse sensing model;

[0206] Specifically, after obtaining sparse perception parameters consistent with the behavior trajectory of the candidate sub-domain, the candidate sub-domain will be an input area modeled as a local sparse structure, and a unified optimization solution model will be constructed for abnormal behavior identification within the sub-domain. The sparse perception model used in this application contains a low-rank structure term and a sparse structure term, which are used to model the common mode and local abnormal structure of element behavior, respectively. A specially designed regularization weighting factor is used to embed the trajectory expansion index into the sparse structure term weight, so that the model is more biased towards extracting contracted trajectories or abnormal accumulation areas.

[0207] In this embodiment, the sparse perception process focuses on extracting local abnormal responses by performing low-rank and sparse decomposition on the response behavior matrix within the candidate sub-domain. In the optimization process, the weighting factor of the regularization term is controlled based on the sparse perception weight matrix generated by the aforementioned trajectory expansion index. The regularization penalty coefficient of the element with low expansion degree is enhanced, which makes it more likely to be identified as an abnormal point. The element with high expansion degree is given weak sparse penalty to preserve the normal structural response space.

[0208] As can be appreciated by those skilled in the art, after the specific rank constraint term and sparse structure term are known, how to perform optimal parameter solution through sparse perception algorithm is a prior art, which will not be described here.

[0209] In one example, the specific steps of S6 are as follows:

[0210] S6.1: Perform statistical analysis on the sparse feature intensity of the remaining elements in the abnormal sparse matrix, and calculate the abnormal accumulation value of each element in the entire time window;

[0211] Specifically, in order to extract fault criteria with physical interpretation ability from the output results of sparse perception modeling, the present application performs statistical processing on the sparse feature values of each element in the abnormal sparse matrix in the time dimension, forming a cumulative abnormal intensity index with consistency measurement ability, to express the abnormal behavior intensity of the element in the entire time window. The fundamental purpose of this processing step is to eliminate the judgment fluctuations introduced by transient disturbances or local missing through time aggregation mechanism, and to realize the persistent identification of stable faults.

[0212] In this embodiment, the sparse structure region extracted in the sparse perception process generates an anomaly sparse matrix, each value in the matrix representing the degree of sparse response of an array element at a certain time. For each array element, the sparse characteristic value is summed, weighted averaged or sliding window integrated along the time axis, so as to obtain the anomaly cumulative value of the array element in the whole time interval. The cumulative value not only contains the frequency of anomaly occurrence, but also integrates the intensity factor of each anomaly response, so that the fault array element has stronger focusing in the statistical result, and the misjudgment interference caused by short-time pulse or accidental fluctuation is excluded.

[0213] S6.2: Comparing the anomaly cumulative value with a preset intensity threshold to determine whether the anomaly cumulative value meets the reliability evaluation criterion;

[0214] Specifically, in order to convert the statistical anomaly cumulative value into a fault identification result, the application sets a group of adjustable intensity judgment thresholds for grading judgment of the cumulative anomaly degree of each array element. The reliability evaluation criterion is not limited to single threshold comparison, but can also include composite conditions, multi-level strategies and spatial coordination judgment mechanism to adapt to the diagnosis needs of different working scenes and different fault types.

[0215] In this embodiment, for the anomaly cumulative value of each array element, multiple judgment paths are set: on the one hand, it is compared with the globally set intensity threshold to determine whether it exceeds the set absolute intensity limit; on the other hand, relative threshold judgment can be made based on local statistics, for example, whether the anomaly value of the array element is significantly higher than several times of the average value of its neighborhood, or whether it is in the top several percentiles of the anomaly value distribution of the whole array. In addition, the application can also introduce the continuity factor of behavior anomaly, for example, requiring the anomaly cumulative value to remain at a high level in multiple sliding time windows to be considered as a real fault.

[0216] Further, the intensity judgment threshold or other judgment condition in the above-mentioned judgment mechanism can be generated by historical data or updated and adapted by the self-learning module of the edge device, so that the fault judgment standard has environmental migratory property. Through the above comparison operation, the array elements in the high response region of the anomaly sparse matrix can be accurately extracted and used as the target object of the final judgment, improving the stability of the fault diagnosis result and the reliability of the engineering application.

[0217] S6.3: Marking the array element meeting the reliability evaluation criterion as a fault array element, and calculating the fault detection result of the passive plate-shaped base station antenna in the whole impact test process according to the cumulative characteristic value;

[0218] Specifically, after completing the statistical judgment, all the elements that meet the above reliability evaluation criteria are uniformly marked for failure, forming a structural diagnostic output result at the element level, which is used for subsequent control response, maintenance scheduling or remote trouble reporting. The marking of the faulty element not only includes its number or position, but also can be accompanied by multiple inference results, including abnormal response duration, maximum abnormal intensity, corresponding behavior track form, etc., to construct a fault description semantics for engineering systems.

[0219] In the embodiment, the marking result of the faulty element can be output in the form of a two-dimensional array mask, where the position of the faulty element can be represented by a Boolean quantity, or a sparse coordinate list can be generated to save edge storage resources. For scenarios that require multiple fault differentiation, an abnormal type classification flag can also be introduced, such as matching different feature combinations such as low track expansion degree, sparse response concentration, etc. to different fault mechanism categories to realize soft label assignment. At the same time, the marking result can be communicated with the upper layer platform or embedded in the diagnostic log for data closed-loop analysis.

[0220] The reliability evaluation criteria at least include one of:

[0221] The abnormal accumulation value of the element within a preset time window is greater than or equal to a first intensity threshold;

[0222] The abnormal response value of the element at a plurality of consecutive time points is greater than or equal to a second intensity threshold;

[0223] The difference between the abnormal response value of the element and the average response value of its adjacent elements is greater than or equal to a third spatial difference threshold;

[0224] The sparse feature distribution of the abnormal sparse matrix corresponding to the element accounts for more than or equal to a fourth regional abnormal proportion threshold in the array area to which it belongs.

[0225] In one example, the present application provides a passive plate-shaped base station antenna fault diagnosis system based on sparse perception, which comprises:

[0226] A response acquisition module for acquiring the transmission response sequence of the passive plate-shaped base station antenna in multiple directions, frequency points and time, and pre-processing the observation data;

[0227] A behavior modeling module for deconstructing the transmission response sequence in combination with a preset element shaping contribution model, generating an element response behavior matrix, and performing delay embedding processing on the response value of each element to construct a high-dimensional behavior track and calculate a track expansion degree index;

[0228] A sparse perception module for constructing an array behavior structure tension diagram based on the track expansion degree index, determining a sparse perception candidate sub-domain, configuring sparse perception processing parameters, and extracting an abnormal sparse matrix;

[0229] The fault detection module is configured to perform statistical analysis on the elements of the abnormal sparse matrix, mark the fault elements in the passive plate-shaped base station antenna according to a preset reliability evaluation criterion, and generate a fault detection result of the passive plate-shaped base station antenna.

[0230] Although the embodiments of the present application have been shown and described above, it should be understood by those skilled in the art that the above embodiments are exemplary and cannot be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for fault diagnosis of passive plate-shaped base station antennas based on sparse sensing, characterized in that, The method includes: A pole-mounted fixture is used to mount a passive plate-shaped base station antenna on a vibration table. The passive plate base station antenna is tested by the power supply of the electronically regulated module. The test includes applying a specified number of impacts to the passive plate base station antenna in both positive and negative directions on three mutually perpendicular axes according to a preset test severity level. The transmission response sequence of the passive plate base station antenna under multiple directions, frequencies, and times is obtained. Combined with the preset reliability evaluation criteria, the fault detection result of the passive plate base station antenna under the impact test is determined. The method further includes: calculating an array element response behavior matrix based on the transmit response sequence and a preset array element shaping contribution model; performing sparse sensing processing on the array element response behavior matrix to obtain an abnormally sparse matrix, wherein the sparse sensing processing identifies array element response regions with abnormally sparse structures in the array element response behavior matrix based on the behavioral characteristics of each array element; determining the fault detection result of the passive plate base station antenna based on the abnormally sparse matrix, wherein the array element shaping contribution model is constructed through historical measurement data and includes a mapping relationship matrix for mapping the observation direction response to the responses of each array element, and the array element response behavior matrix is ​​organized in the form of a two-dimensional real-valued matrix, represented as follows. Where A represents the element response behavior matrix, n represents the element number, and T represents the time point. Each element in the matrix is ​​the response value of the nth element at the Tth time point. Based on the transmission response sequence and a preset element shaping contribution model, the element response behavior matrix is ​​calculated, including: flattening the transmission response sequence according to the direction and frequency to obtain the observation vector corresponding to each time point; performing an inverse operation on the observation vector through the mapping relationship matrix to obtain the response value of the element at the corresponding time point; and arranging the response values ​​according to the element number to obtain the element response behavior matrix. The element response behavior matrix is ​​subjected to sparse sensing processing to obtain an anomalously sparse matrix, including: The response value of each element in the element response behavior matrix is ​​subjected to delayed embedding processing to generate the corresponding high-dimensional behavior trajectory; Based on the unfolding degree of the high-dimensional behavioral trajectory in the embedding space, the trajectory unfolding degree index is calculated, wherein the trajectory unfolding degree index includes at least one of trajectory envelope volume, principal component energy distribution, or manifold dimension estimation. Based on the orbital unfolding index, sparse sensing candidate regions are calculated, and sparse sensing is performed on the sparse sensing candidate regions to generate an abnormally sparse matrix. Based on the anomalous sparse matrix, the fault detection result of the passive plate base station antenna is determined, including: Statistical analysis is performed on the sparse feature strengths corresponding to the remaining array elements in the abnormal sparse matrix, and the abnormal cumulative value of each array element is calculated over the entire time window. The accumulated abnormal value is compared with a preset intensity threshold to determine whether the accumulated abnormal value meets the reliability assessment criteria. Array elements that meet the reliability assessment criteria are marked as faulty array elements, and the fault detection results of the passive plate base station antenna during the entire impact test are calculated based on the cumulative characteristic values. The reliability assessment criteria include at least one of the following: The cumulative abnormal value of the array element within a preset time window is greater than or equal to the first intensity threshold. The abnormal response value of the array element is greater than or equal to the second intensity threshold at multiple consecutive time points; The difference between the abnormal response value of the array element and the average response value of its neighboring array elements is greater than or equal to the third space difference threshold. The sparse feature distribution in the abnormal sparse matrix corresponding to the array element accounts for a proportion greater than or equal to the abnormal proportion threshold of the fourth region in its respective array region.

2. The method for fault diagnosis of passive plate-shaped base station antenna based on sparse sensing according to claim 1, characterized in that, The passive plate-shaped base station antenna is arranged at the same angle as the preset maximum tilt angle.

3. The method for fault diagnosis of passive plate-shaped base station antenna based on sparse sensing according to claim 1, characterized in that, The impact interval between two adjacent tests is adjusted according to the degree of impact response.

4. The method for fault diagnosis of passive plate-shaped base station antenna based on sparse sensing according to claim 1, characterized in that, Based on the orbital spread index, sparse sensing candidate regions are calculated, and sparse sensing is performed on the sparse sensing candidate regions to generate an anomalous sparse matrix, including: Based on the orbital spread index, an array behavior structure tension map is generated, wherein the tension value of each array element is calculated based on the gradient difference of the orbital spread index between the array element and its neighboring array elements. The tension map of the array behavior structure is processed, and the tension concentration region is used as a candidate subdomain for sparse sensing. Based on the orbital expansion index corresponding to the central subdomain of the sparse sensing candidate subdomain, the corresponding sparse sensing processing parameters are configured, wherein the sparse sensing processing parameters include the scale factor extracted from the sparse structure, the sparse constraint strength, and the region sensing window. The sparse sensing candidate subdomain is sparsely sensed according to the sparse sensing processing parameters to obtain a sparse structure region. The array elements corresponding to the sparse structure region are summarized to obtain an abnormally sparse matrix. The sparse sensing is based on a preset sparse sensing model, which includes a rank constraint term and a sparse structure term. The sparse sensing embeds the regularization weighting factor of the sparse structure term into the sparse sensing processing parameters and solves the sparse sensing model.

5. The method for fault diagnosis of passive plate-shaped base station antenna based on sparse sensing according to claim 4, characterized in that, Based on the orbital unfolding index corresponding to the central subdomain of the sparse sensing candidate subdomain, configure the corresponding sparse sensing processing parameters, including: A sparse sensing weight matrix is ​​generated based on the orbital expansion index of the central array element in the sparse sensing candidate subdomain. Based on the sparse sensing weight matrix and the objective function of the sparse sensing model, the size of the region sensing window is calculated using the local mean and variance of the orbital spread index. The scale factor is calculated using the gradient vector of the orbital spread index, and the sparse constraint strength is calculated using the value range of the orbital spread index.

6. The method for fault diagnosis of passive plate-shaped base station antenna based on sparse sensing according to claim 1, characterized in that, The array element shaping contribution model is constructed using historical measurement data and includes a mapping matrix used to map the observation direction response to the response of each array element. The computational matrix of element response behavior includes: Flatten the transmission response sequence according to the direction and frequency to obtain the observation vector corresponding to each time point; The response value of the corresponding time-point matrix element is obtained by inverting the observation vector through the mapping relationship matrix. The response values ​​are arranged according to the element numbers to obtain the element response behavior matrix.

7. A passive plate-shaped base station antenna fault diagnosis system based on sparse sensing, used to implement the passive plate-shaped base station antenna fault diagnosis method based on sparse sensing as described in any one of claims 1-6, characterized in that, The system includes: The response acquisition module is used to acquire the transmission response sequence of the passive plate base station antenna in multiple directions, frequencies and times, and to preprocess the observation data; The behavior modeling module is used to deconstruct the emission response sequence by combining the preset array element shaping contribution model, generate the array element response behavior matrix, perform delay embedding processing on the response value of each array element, construct a high-dimensional behavior orbit and calculate the orbit unfolding index. The sparse sensing module is used to construct an array behavior structure tension map based on the orbital unfolding index, determine sparse sensing candidate subdomains, configure sparse sensing processing parameters, and extract abnormal sparse matrices. The fault detection module is used to perform statistical analysis on the array elements of the abnormal sparse matrix, mark the faulty array elements according to the preset reliability evaluation criteria, and generate the fault detection results of the passive plate base station antenna.

Images