Communication equipment fault prediction method and system based on big data and optical fiber sensor

Abstract

The invention provides a communication equipment fault prediction method and system based on big data and an optical fiber sensor. The method comprises the following steps: generating a dynamic physical topology model according to a multi-source heterogeneous data matrix in combination with a light attenuation characteristic value and a signal transmission direction identifier of a distributed optical fiber sensor; based on a spatial constraint condition of the dynamic physical topology model, separating an abnormal fluctuation component associated with at least one of temperature abnormal data and vibration abnormal data of the communication equipment from the optical signal data to generate a cross-modal abnormal associated feature set; and through a collaborative constraint rule, analyzing the cross-modal abnormal association feature set, and generating a communication equipment fault prediction result for early warning in advance. According to the technical scheme provided by the invention, the multi-source heterogeneous matrix and the dynamic topology model are constructed based on the distributed optical fiber sensing data, and accurate early warning is carried out on the fault position and type of the communication equipment in advance through cross-modal abnormal association feature analysis.

Description

Technical Field

[0001] This application relates to the technical field of communication equipment fault prediction, and particularly to a communication equipment fault prediction method and system based on big data and fiber optic sensors. Background Art

[0002] In the operation and maintenance of modern large-scale communication equipment, the real-time, accurate, and spatially localized monitoring of the health status of communication equipment and early fault warning are crucial for ensuring the high reliability and security of communication.

[0003] The existing solution is to analyze based on a single-modal or simply fused machine learning model. This solution uses fiber optic sensor data, and may combine static network topology information to train a model to identify abnormal points or patterns that exceed a preset threshold, so as to perform fault warning.

[0004] However, the main defects of the existing solution are that it is difficult to effectively model and does not consider the spatio-temporal correlation between multi-modal anomalies; at the same time, the static topology model cannot fully reflect the dynamic changes of the physical state of the fiber optic network, resulting in insufficient spatial positioning accuracy of the abnormal source in a complex environment, and weak ability to separate and correlate cross-modal abnormal features, which is prone to false alarms or missed alarms and difficult to meet the requirements of accurately positioning the fault point and type. Summary of the Invention

[0005] This application provides a communication equipment fault prediction method and system based on big data and fiber optic sensors to solve the problem of low accuracy of communication equipment fault prediction existing in the prior art.

[0006] In a first aspect, this application provides a communication equipment fault prediction method based on big data and fiber optic sensors, including: Integrate the temperature data, vibration data, and optical signal data of the communication equipment collected by the distributed fiber optic sensor into a multi-source heterogeneous data matrix according to the time series and spatial position; Generate a dynamic physical topology model based on the multi-source heterogeneous data matrix, in combination with the optical attenuation eigenvalue and signal transmission direction identifier of the distributed fiber optic sensor; Based on the spatial constraint conditions of the dynamic physical topology model, separate the abnormal fluctuation components associated with at least one of the temperature abnormal data and vibration abnormal data of the communication equipment from the optical signal data to generate a cross-modal abnormal association feature set; Generate a communication equipment fault prediction result by analyzing the cross-modal abnormal association feature set through a collaborative constraint rule, and the collaborative constraint rule is jointly generated by a support vector machine and an extreme learning machine. The communication equipment fault prediction result includes potential fault positions and fault types in the communication equipment, and is used for early warning.

[0007] Optionally, based on the spatial constraint conditions of the dynamic physical topology model, separating an abnormal fluctuation component associated with at least one of the temperature abnormal data and vibration abnormal data of the communication device from the optical signal data to generate a cross-modal abnormal association feature set, including: Based on the spatial constraint conditions of the dynamic physical topology model, decomposing the optical signal data into independent spatial signal components corresponding to the communication device or connection segments; Performing time-frequency domain analysis on each of the independent spatial signal components, extracting abnormal fluctuation components matching the fluctuation patterns of historical abnormal temperature or vibration data, and obtaining the temperature abnormal data and vibration abnormal data of the communication device; Calculating the association weights between the temperature abnormal data and vibration abnormal data of the communication device through multi-scale correlation analysis; Based on the association weights, performing weighted superposition on the abnormal fluctuation components to generate a weighted abnormal fluctuation component set; Pruning components conflicting with the device location or connection path from the weighted abnormal fluctuation component set to generate a cross-modal association feature set.

[0008] Optionally, the performing weighted superposition on the abnormal fluctuation components based on the association weights to generate a weighted abnormal fluctuation component set includes: Taking the association weights as amplitude scaling factors, performing an element-wise multiplication operation on each of the abnormal fluctuation components to generate weighted abnormal fluctuation components; Based on the spatial constraint conditions of the dynamic physical topology model, assigning a unique spatial identifier to each of the weighted abnormal fluctuation components; Storing the weighted abnormal fluctuation components and the corresponding spatial identifiers as key-value pairs into a hash map data structure; Based on the complete storage state of the hash map data structure, outputting the weighted abnormal fluctuation component set.

[0009] Optionally, the generating a communication device fault prediction result by analyzing the cross-modal abnormal association feature set through collaborative constraint rules includes: Based on the cross-modal abnormal association feature set, extracting the spatial identifiers and corresponding time-frequency domain feature vectors of each cross-modal abnormal association feature; Based on the time-frequency domain feature vectors, performing a feature compression operation using a first feature selector to generate a dimensionality-reduced feature vector, and performing a spatial mapping operation using a second feature selector to generate a projected feature vector; Performing a feature fusion operation on the dimensionality-reduced feature vector and the projected feature vector to generate a fusion decision vector; Retrieving the physical attributes of the corresponding communication device or connection segment in the dynamic physical topology model based on the spatial identifier, and generating a fault type label in combination with the fusion decision vector; The communication equipment fault prediction result is generated by taking the spatial identifier as the fault location index and the fault type label as an element.

[0010] Optionally, the retrieving physical attributes of a corresponding communication device or connection segment in the dynamic physical topology model based on the spatial identifier and generating a fault type label in combination with the fusion decision vector includes: Converting the physical attributes into a structured attribute vector, and generating a physical constraint rule library for the fusion decision vector based on the device material type, the connection segment signal transmission threshold, and the environmental tolerance threshold in the structured attribute vector; Under the boundary conditions of the physical constraint rule library, performing a fault mode matching operation on the fusion decision vector to generate a candidate fault mode set; When there are fault modes with conflicting physical properties in the candidate fault mode set, conflict resolution is performed based on the priority weights in the structured attribute vector, and a valid fault mode set is output; The valid fault mode set is converted into a fault type label that is compatible with the structured attribute vector.

[0011] Optionally, the step of performing a feature compression operation based on the time-frequency domain feature vector using a first feature selector to generate a dimension-reduced feature vector, and performing a space mapping operation based on the second feature selector to generate a projected feature vector includes: Based on the frequency band priority in the coordination constraint rule, assigning importance weights to each frequency band component in the time-frequency domain feature vector to generate a weighted time-frequency domain feature vector; Inputting the weighted time-frequency domain feature vector into a first feature selector, removing redundant frequency band components and retaining key frequency band features according to a preset communication feature retention rule, and generating a reduced-dimensionality feature vector; Constructing a spatial correlation index for the weighted time-frequency domain feature vector according to a mapping relationship between device ports and transmission paths in the dynamic physical topology model; The weighted time-frequency domain feature vector and the spatial correlation index are synchronously input into a second feature selector, and feature components across ports are aggregated according to a physical connection topology to generate a projected feature vector.

[0012] Optionally, generating a dynamic physical topology model based on the multi-source heterogeneous data matrix and combining optical attenuation characteristic values and signal transmission direction identifiers of distributed optical fiber sensors includes: Extract the spatial coordinates of the optical signal event points from the multi-source heterogeneous data matrix, and bind the spatial coordinates to the corresponding communication device ports based on a preset device port mapping table to generate a set of port spatial coordinates; Analyze the optical transmission connection relationship between adjacent communication device ports according to the optical attenuation eigenvalue and signal transmission direction identifier of the distributed optical fiber sensor to generate a set of port connection relationships; Construct an optical transmission connection relationship graph between ports based on the set of port connection relationships and transmission stability indicators; Bind the set of port spatial coordinates, the optical transmission connection relationship graph and a preset device physical attribute library, where the device physical attributes include port material type, optical fiber length and environmental anti-interference level, to generate a weighted dynamic physical topology model.

[0013] In a second aspect, the present application provides a communication device fault prediction system based on big data and optical fiber sensors, including: An acquisition module, configured to integrate the temperature data, vibration data and optical signal data of a communication device collected by a distributed optical fiber sensor into a multi-source heterogeneous data matrix according to time series and spatial position; A generation module, configured to generate a dynamic physical topology model according to the multi-source heterogeneous data matrix, in combination with the optical attenuation eigenvalue and signal transmission direction identifier of the distributed optical fiber sensor; A separation module, configured to separate, based on the spatial constraint conditions of the dynamic physical topology model, an abnormal fluctuation component associated with at least one of the temperature abnormal data and vibration abnormal data of the communication device from the optical signal data, so as to generate a cross-modal abnormal association feature set; An analysis module, configured to analyze the cross-modal abnormal association feature set through collaborative constraint rules to generate a communication device fault prediction result. The collaborative constraint rules are jointly generated by a support vector machine and an extreme learning machine. The communication device fault prediction result includes potential fault positions and fault types in the communication device, and is used for early warning.

[0014] In a third aspect, the present application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a communication device fault prediction method based on big data and optical fiber sensors as described in any item of the first aspect.

[0015] In a fourth aspect, the present application provides a computer storage medium storing a computer program, and when the computer program is executed by a computer, it implements a communication device fault prediction method based on big data and optical fiber sensors as described in any item of the first aspect.

[0016] In this application, a method for predicting communication device failures based on big data and fiber optic sensors is provided. The method includes: integrating the temperature data, vibration data, and optical signal data of the communication device collected by the distributed fiber optic sensor into a multi-source heterogeneous data matrix according to the time series and spatial position; generating a dynamic physical topology model based on the multi-source heterogeneous data matrix in combination with the optical attenuation eigenvalue and signal transmission direction identifier of the distributed fiber optic sensor; based on the spatial constraint conditions of the dynamic physical topology model, separating from the optical signal data an abnormal fluctuation component associated with at least one of the temperature abnormal data and vibration abnormal data of the communication device to generate a cross-modal abnormal association feature set; analyzing the cross-modal abnormal association feature set through a collaborative constraint rule to generate a communication device failure prediction result, where the collaborative constraint rule is jointly generated by a support vector machine and an extreme learning machine, and the communication device failure prediction result includes potential failure positions and failure types in the communication device, which are used for early warning.

[0017] This application integrates the temperature, vibration, and optical signal data collected by the distributed fiber optic sensor into a multi-source heterogeneous data matrix in the time and space dimensions, realizing the unified structured representation of multi-modal data; the dynamic physical topology model generated based on the multi-source heterogeneous data matrix in combination with the optical attenuation feature and signal direction identifier can map the physical state changes of the fiber optic network in real time; using the spatial constraint conditions of this model to accurately separate the abnormal fluctuation component associated with temperature or vibration anomalies from the optical signal, constructing a cross-modal abnormal association feature set, effectively suppressing environmental noise interference; finally, analyzing this feature set through the collaborative constraint rule jointly driven by the support vector machine and the extreme learning machine to generate a failure prediction result with high-precision spatial positioning, improving the accuracy and reliability of early failure warning of communication devices.

[0018] Furthermore, based on the spatial constraints of the dynamic physical topology model, this application first decomposes the optical signal data into independent spatial signal components corresponding to devices or connection segments; then performs time-frequency domain analysis on each component to extract abnormal fluctuation components that match the historical temperature and vibration anomaly patterns; next, calculates the correlation weight between the temperature and vibration anomaly data through multi-scale correlation analysis; uses this weight as an amplitude scaling factor to perform element-wise multiplication on each abnormal fluctuation component to generate weighted components, and assigns a unique spatial identifier to them, stores them in a hash map data structure in the form of key-value pairs to form a set of weighted abnormal fluctuation components; finally, prunes the components that conflict with the device location or connection path from this set to generate a cross-modal correlation feature set. This method ensures the precise positioning of anomaly analysis through spatially constrained signal decomposition; time-frequency domain analysis combined with historical pattern matching effectively extracts abnormal fluctuations strongly related to device failures; weighted superposition based on the weights calculated by multi-scale correlation strengthens the correlation characteristics between cross-modal anomalies; uses the hash map data structure to manage the weighted components with spatial identifiers, improving the efficiency of feature storage and retrieval; finally, through spatial conflict pruning, significantly filters out environmental noise and irrelevant interference components, making the generated cross-modal correlation feature set have high credibility, strong spatial directivity, and fault characterization ability, laying a solid foundation for subsequent fault prediction.

[0019] These aspects or other aspects of this application will be more clearly understood in the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0021] Figure 1 It is a flowchart of a communication device fault prediction method based on big data and fiber optic sensors provided by an embodiment of this application; Figure 2 It is a schematic structural diagram of a communication device fault prediction system based on big data and fiber optic sensors provided by an embodiment of this application; Figure 3 It is a schematic structural diagram of a computing device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to enable those skilled in the art to better understand the solutions of this application, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the drawings in the embodiments of this application.

[0023] In some of the processes described in the specification, claims, and the above-mentioned drawings of this application, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear herein or may be executed in parallel. The serial numbers of the operations, such as 11, 12, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions such as "first" and "second" in this article are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit that "first" and "second" are of different types.

[0024] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0025] To solve the problem of low accuracy in predicting communication device failures in the prior art, the embodiments of the present application provide a communication device failure prediction method based on big data and fiber optic sensors. The method adopts the following concept: to solve the problem of collaborative analysis and early failure warning of multi-source heterogeneous sensing data of communication devices, first, a structured data matrix is constructed by integrating temperature, vibration, and optical signals through spatio-temporal alignment, laying the foundation for multi-modal fusion; then, the fiber optic physical topology is dynamically modeled by combining the optical attenuation characteristics and signal direction identification, breaking through the bottleneck of the static topology in characterizing state evolution; based on the spatial constraints of this model, the fluctuation components strongly correlated with temperature and vibration anomalies in the optical signal are accurately separated, and a cross-modal anomaly feature set is constructed to suppress environmental noise interference; finally, a collaborative constraint rule jointly driven by a support vector machine and an extreme learning machine is designed to deeply fuse the spatio-temporal correlation of cross-modal features, realizing accurate positioning and early warning of the failure location and type, and improving the reliability of the system.

[0026] Figure 1 The flowchart of a communication device failure prediction method based on big data and fiber optic sensors provided by the embodiments of the present application is as Figure 1 shown, and the method includes: S11. Integrate the temperature data, vibration data, and optical signal data of the communication device collected by the distributed fiber optic sensor into a multi-source heterogeneous data matrix according to the time series and spatial position.

[0027] Among them, the distributed optical fiber sensor can refer to a sensing device continuously distributed along the optical fiber laying path, and obtains the physical parameters of the monitored object by measuring the change in the intensity of Rayleigh scattered light. In the embodiment of the present application, the monitored object is a communication device, and the communication device includes network infrastructures such as an optical transmission cabinet and a base station radio frequency unit. Temperature data can be understood as a sequence of physical quantities used to reflect the thermal distribution state on the surface of the device. Vibration data can be a sequence of acceleration quantities characterizing the mechanical vibration intensity of the device. Optical signal data refers to a sequence in which the optical power value transmitted in the optical fiber changes over time. A time series refers to a set of data arranged at a fixed sampling interval. The spatial position refers to the physical coordinate points based on distance on the optical fiber path. The multi-source heterogeneous data matrix refers to a three-dimensional data structure integrating temperature, vibration, and optical signals.

[0028] In the embodiment of the present application, first, the temperature data, vibration data, and optical signal data of the communication device are synchronously collected through a distributed optical fiber sensor; secondly, the three types of data are aligned according to the millisecond-level time stamp to form a time series, and at the same time, the spatial position is mapped according to the distance coordinates of the sensor on the optical fiber path; then, the time series and the spatial position information are fused through a multi-dimensional array splicing technology, and finally, a multi-source heterogeneous data matrix including three dimensions of temperature, vibration, and optical signal is generated.

[0029] S12. According to the multi-source heterogeneous data matrix, a dynamic physical topology model is generated in combination with the optical attenuation eigenvalue and the signal transmission direction identifier of the distributed optical fiber sensor.

[0030] Among them, the optical attenuation eigenvalue refers to the optical power loss coefficient per unit length of the optical fiber, which is calculated by taking the logarithm after dividing the input optical power value by the output optical power value. The signal transmission direction identifier is a logical variable indicating the propagation direction of the optical signal in the optical fiber. The dynamic physical topology model can be used to establish a multi-dimensional mapping relationship between the potential fault area and the sensing data acquisition point.

[0031] In the embodiment of the present application, first, the optical attenuation eigenvalue is extracted from the optical signal dimension of the multi-source heterogeneous data matrix, and this value represents the optical power loss amount per unit length of the optical fiber; secondly, the propagation path of the optical signal in the optical fiber is determined in combination with the signal transmission direction identifier; then, a network skeleton with the communication device as the node and the optical fiber link as the edge is constructed by using a graph theory algorithm; finally, the real-time calculated optical attenuation eigenvalue is injected into the network skeleton as a dynamic edge weight to generate a dynamic physical topology model reflecting the change of the physical state.

[0032] S13. Based on the spatial constraint conditions of the dynamic physical topology model, the abnormal fluctuation components associated with at least one of the temperature abnormal data and the vibration abnormal data of the communication device are separated from the optical signal data to generate a cross-modal abnormal association feature set.

[0033] Among them, the spatial constraint condition refers to the geometric rule that limits the signal analysis range, and an effective radius is set centered on the device position. The specific content of the implementation of the spatial constraint condition in this application is not specifically limited. The temperature anomaly data can be a set of continuous data points that exceed the normal fluctuation range of historical temperatures. The vibration anomaly data refers to a set of frequency bands whose vibration energy exceeds a preset threshold. The cross-modal anomaly correlation feature set refers to a set of feature vectors that includes spatial position, temperature anomaly intensity, and vibration anomaly intensity. It should be noted that the numerical values of the thresholds of various types are not specifically limited in the embodiments of this application.

[0034] In the embodiments of this application, first, a signal analysis area around the communication device is delimited based on the spatial constraint condition of the dynamic physical topology model; secondly, wavelet packet decomposition is performed on the optical signal data within this area; subsequently, the sub-band components after decomposition are subjected to similarity matching with the spectral characteristics of historical temperature anomaly data and vibration anomaly data. The formula used for the similarity matching is not specifically limited in the embodiments of this application; finally, the sub-band fluctuation components with a matching degree exceeding the preset threshold are extracted and aggregated to generate a cross-modal anomaly correlation feature set.

[0035] S14. Through the collaborative constraint rule, the cross-modal anomaly correlation feature set is analyzed to generate a communication device fault prediction result. The collaborative constraint rule is jointly generated by using a support vector machine and an extreme learning machine. The communication device fault prediction result includes potential fault positions and fault types in the communication device, and is used for early warning.

[0036] Among them, the collaborative constraint rule can be obtained through machine learning training on the correlation relationship between the historical multi-source heterogeneous data matrix and the dynamic physical topology model based on a big data analysis platform. The communication device fault prediction result refers to the output data that includes the device position coordinates and the fault type identifier. The support vector machine refers to a classification algorithm based on the principle of structural risk minimization. The extreme learning machine refers to a single-hidden-layer feedforward neural network algorithm. The fault position refers to the spatial coordinates of the device where the anomaly occurs on the optical fiber path. The fault type refers to the pre-defined device anomaly category identifier.

[0037] In the embodiments of this application, first, a collaborative decision-making framework of a support vector machine and an extreme learning machine is constructed: the support vector machine is responsible for spatially classifying the cross-modal anomaly correlation feature set, and the extreme learning machine is responsible for calculating the probability distribution of the fault type; secondly, the output results of the two types of models are integrated through a weighted fusion algorithm; subsequently, the fault device position is mapped according to the spatial identifier of the feature set; finally, a prediction result including the communication device fault position coordinates and the fault type code is output and the early warning protocol is triggered.

[0038] The following is a specific example: first, the original data of temperature, vibration and optical signals in the communication device area are collected in real time through a distributed optical fiber sensor; secondly, the three types of data are aligned according to a unified timestamp and mapped to the optical fiber space coordinates to form a multi-source heterogeneous data matrix; then, the optical attenuation eigenvalue is extracted from the matrix and fused with the signal direction identifier to construct a dynamic physical topology model; subsequently, based on the space constraint of the model, the optical signal data is segmented, and the components associated with temperature and vibration anomalies are extracted through spectrum matching to generate a cross-modal feature set; finally, a support vector machine and an extreme learning machine are used to jointly analyze the feature set, and the fault location coordinates and fault type identifier of the communication device are output to trigger the warning process of the operation and maintenance system.

[0039] By executing S11~S14, the embodiment of the present application constructs a unified analysis basis through multi-source spatio-temporal data fusion, accurately describes the change of the physical state of the optical fiber by using dynamic topology modeling, effectively enhances the abnormal signal identification ability based on the cross-modal feature separation of space constraints, and combines the collaborative decision-making mechanism of the dual machine learning model to achieve high-precision prediction of the fault location and type of the communication device, and improve the warning reliability.

[0040] In a possible embodiment, S13. Based on the space constraint conditions of the dynamic physical topology model, separate the abnormal fluctuation components associated with at least one of the temperature abnormal data and vibration abnormal data of the communication device from the optical signal data to generate a cross-modal abnormal association feature set, including: Step 131. Based on the space constraint conditions of the dynamic physical topology model, decompose the optical signal data into independent space signal components corresponding to the communication device or connection segment.

[0041] Among them, the connection segment may refer to the optical fiber physical section between adjacent communication devices, for example, the length range is 10-500 meters. The independent space signal component refers to the discretized representation of the optical signal within the space constraint range, including three elements: starting point coordinates, ending point coordinates, and signal curve.

[0042] In the embodiment of the present application, first, the communication device position and the optical fiber connection segment range are delimited based on the space constraint conditions of the dynamic physical topology model; secondly, the original optical signal data is segmented according to the space coordinates into independent signal segments corresponding one by one to the device or connection segment; then, the inter-segment crosstalk is eliminated through a digital filter; finally, a set of independent space signal components of each device or connection segment is output.

[0043] Step 132. Perform time-frequency domain analysis on each independent space signal component, and extract the abnormal fluctuation components matching the fluctuation patterns of historical abnormal temperature or vibration data to obtain the temperature abnormal data and vibration abnormal data of the communication device.

[0044] Among them, historical abnormal temperature refers to the set of temperature abnormal data marked as fault-related within the past monitoring period, and the storage format is a two-dimensional time-amplitude array. Vibration data fluctuation refers to the envelope characteristics of the time-domain waveform of the vibration acceleration signal, which is extracted through Hilbert transform. The abnormal fluctuation component refers to the signal segment in the time-frequency domain that matches the historical abnormal form, and includes three-dimensional characteristics of central frequency, duration, and energy intensity.

[0045] In the embodiment of the present application, first, continuous wavelet transform is performed on each independent spatial signal component; secondly, the time-frequency distribution characteristics of the wavelet coefficients are extracted; then, the characteristics are subjected to similarity matching with the fluctuation form library of historical abnormal temperature data and the fluctuation form library of historical vibration data; finally, the component with the highest similarity is selected and marked as the abnormal fluctuation component, and the temperature abnormal data and the vibration abnormal data set are generated.

[0046] Step 133: Calculate the correlation weight between the temperature abnormal data and the vibration abnormal data of the communication device through multi-scale correlation analysis.

[0047] Among them, multi-scale correlation analysis calculates the data correlation in three dimensions: time domain, frequency domain, and time-frequency domain.

[0048] In the embodiment of the present application, first, the temperature abnormal data and the vibration abnormal data are aligned according to the time window; secondly, the Pearson correlation coefficients are calculated respectively on three scales of the time domain, frequency domain, and time-frequency domain; then, the multi-scale correlation coefficients are weighted and averaged; finally, normalization processing is performed to obtain the correlation weight value of the temperature and vibration abnormal data. The correlation weight refers to the quantization value of the correlation between the temperature and vibration abnormal data, with a range of 0-1, and the calculation formula can be the arithmetic average of the correlation coefficients of each scale.

[0049] Step 134: Perform weighted superposition on the abnormal fluctuation components based on the correlation weight to generate a weighted abnormal fluctuation component set.

[0050] Among them, the weighted abnormal fluctuation component set refers to the set of abnormal fluctuation components after amplitude scaling, and the data structure is a key-value pair composed of a space identifier and a weighted signal vector.

[0051] In the embodiment of the present application, first, the correlation weight is converted into an amplitude scaling coefficient; secondly, coefficient multiplication operation is performed on each abnormal fluctuation component; then, a space identifier is assigned to the weighted component; finally, it is stored in the hash table according to the device location index to form a weighted abnormal fluctuation component set.

[0052] Step 135: Prune the components that conflict with the device location or connection path from the weighted abnormal fluctuation component set to generate a cross-modal correlation feature set.

[0053] Among them, the component with a connection path conflict refers to the component with a deviation in the distance between the space identifier and the actual connection path of the topological model.

[0054] In the embodiment of the present application, first, the device location and connection path data of the dynamic physical topology model are loaded; second, the spatial identifiers of the weighted abnormal fluctuation component set are traversed; then, the geometric deviation between the identifier and the actual device location is detected; finally, the components with deviations exceeding the threshold are deleted to generate a cross-modal association feature set.

[0055] The following is a specific example: First, based on the spatial constraints of the dynamic topology model, the optical signal is segmented into independent spatial signal components associated with the device; second, the time-frequency characteristics of the components are extracted by wavelet transform and matched with the historical temperature vibration abnormal morphology to generate abnormal fluctuation components; then, the multi-scale association weights of the temperature vibration abnormal data are calculated; subsequently, the abnormal fluctuation components are amplitude-weighted according to the weights and stored as a set; finally, the spatial offset components are pruned according to the actual device location to generate a cross-modal association feature set.

[0056] By executing steps 131 to 135, the embodiment of the present application ensures the accuracy of abnormal positioning through signal decomposition with spatial constraints, enhances the fault feature extraction ability through time-frequency domain feature matching, strengthens the cross-modal correlation through multi-scale correlation weighting, and effectively filters out interference components through spatial conflict pruning, and finally generates a cross-modal association feature set with high credibility.

[0057] In a possible embodiment, step 134, weighted superposition is performed on the abnormal fluctuation components based on the association weights to generate a weighted abnormal fluctuation component set, including: Step a1, taking the association weight as the amplitude scaling factor, and performing an element-wise multiplication operation on each abnormal fluctuation component to generate a weighted abnormal fluctuation component.

[0058] Among them, the amplitude scaling factor refers to the scaling coefficient used to adjust the signal amplitude, including the value calculated based on the association weight, which is used to amplify or reduce the intensity of the abnormal fluctuation component. Element-wise multiplication refers to the multiplication operation on the corresponding elements in two arrays or matrices, including the operation of multiplying each element one by one, which is used to generate a new array. The weighted abnormal fluctuation component refers to the abnormal fluctuation component after being weighted by the amplitude scaling factor, including the result data reflecting the product of the original fluctuation and the weight.

[0059] Step a2, based on the spatial constraint conditions of the dynamic physical topology model, assign a unique spatial identifier to each weighted abnormal fluctuation component.

[0060] Among them, the spatial identifier refers to the identifier used to uniquely identify the physical space location, including the unique code generated based on the spatial constraint conditions of the dynamic physical topology model.

[0061] Step a3, store the weighted abnormal fluctuation component and the corresponding spatial identifier as a key-value pair into the hash map data structure.

[0062] Among them, a key-value pair refers to a data structure element that includes two parts: a key and a value. The key is used to uniquely identify an entity, and the value stores associated data content. A hash map data structure refers to a data storage structure based on a hash table, including a storage mechanism for key-value pairs, which is used to efficiently search for and access data.

[0063] Step a4: Output a weighted abnormal fluctuation component set based on the complete storage state of the hash map data structure.

[0064] Among them, the complete storage state refers to the state in which all data in the hash map data structure has been successfully stored, including a confirmation state that reflects the omission-free preservation of all key-value pairs.

[0065] The following is a specific example: First, obtain the abnormal fluctuation component data of each node in the power grid through a sensor network. At the same time, extract the associated weight as an amplitude scaling factor, and perform an element-wise multiplication operation on each component to generate a weighted abnormal fluctuation component. Second, query the spatial constraint conditions based on the dynamic physical topology model of the power grid, and assign a unique spatial identifier to each weighted component. Then, combine the weighted component and the spatial identifier into a key-value pair and store it in the hash map data structure. Finally, after verifying the complete storage state of the hash map, output the weighted abnormal fluctuation component set.

[0066] By performing Steps a1 to a4, the embodiment of the present application enhances the recognizability of abnormal fluctuations through weighted processing, realizes the accurate association between components and physical positions by combining spatial constraint conditions, and improves the data storage and access efficiency by using the hash map data structure. Finally, a complete weighted set is output, thereby improving the processing accuracy and response speed of the system for abnormal fluctuations.

[0067] In a possible embodiment, S14: Generate a communication device fault prediction result by analyzing the cross-modal abnormal association feature set using collaborative constraint rules, including: Step 141: Based on the cross-modal abnormal association feature set, extract the spatial identifier and the corresponding time-frequency domain feature vector of each cross-modal abnormal association feature.

[0068] Among them, the time-frequency domain feature vector refers to a feature vector that simultaneously includes time variation and frequency distribution, including the time-frequency characteristics of a signal extracted through short-time Fourier transform or wavelet transform.

[0069] In the embodiment of the present application, first, based on the cross-modal abnormal association feature set, extract the spatial identifier and the corresponding time-frequency domain feature vector of each cross-modal abnormal association feature through a feature extraction algorithm, where the spatial identifier is used to locate the physical position, and the time-frequency domain feature vector contains feature information in the time and frequency dimensions.

[0070] Step 142: Based on the time-frequency domain feature vector, perform a feature compression operation using the first feature selector to generate a dimensionality-reduced feature vector, and perform a spatial mapping operation using the second feature selector to generate a projected feature vector.

[0071] Among them, the first feature selector refers to a feature processing module for dimensionality reduction, including algorithms based on principal component analysis or autoencoders. The feature compression operation refers to the process of reducing the feature dimension, including removing redundant information through linear transformation or neural networks. The dimensionality-reduced feature vector refers to the low-dimensional feature vector after the compression operation, including a simplified feature representation that retains key information. The second feature selector refers to a feature processing module for spatial mapping. The spatial mapping operation refers to the transformation process of mapping features to a new spatial coordinate system, including realizing feature reconstruction through matrix multiplication. The projected feature vector refers to the new feature vector generated after spatial mapping, including feature data reflecting the distribution in the low-dimensional space.

[0072] In the embodiment of the present application, first, based on the time-frequency domain feature vector, use the first feature selector to perform a feature compression operation through the principal component analysis or autoencoder algorithm, and generate a dimensionality-reduced feature vector after removing redundant information; secondly, use the second feature selector to perform a spatial mapping operation through the spatial transformation algorithm, and project the features to the low-dimensional space to generate a projected feature vector.

[0073] Step 143: Perform a feature fusion operation on the dimensionality-reduced feature vector and the projected feature vector to generate a fusion decision vector.

[0074] Among them, the fusion decision vector refers to a composite vector integrating multiple features, including decision input data generated by splicing or weighted fusion.

[0075] In the embodiment of the present application, first perform a feature fusion operation on the dimensionality-reduced feature vector and the projected feature vector, integrate the two types of features through the weighted splicing or attention mechanism algorithm, and finally generate a fusion decision vector as the input basis for fault judgment.

[0076] Step 144: Retrieve the physical attributes of the corresponding communication device or connection segment in the dynamic physical topology model based on the spatial identifier, and combine with the fusion decision vector to generate a fault type label.

[0077] Among them, the fault type label is a label indicating the fault category of the device, including the classification result based on the physical attributes and the fusion decision vector.

[0078] In the embodiment of the present application, the specific implementation process of step 144 is as follows: First, perform a retrieval operation in the dynamic physical topology model according to the spatial identifier to locate the communication device entity or network connection segment corresponding to the identifier; Second, extract the physical attribute data of the target device or connection segment, including device model, material parameters, topological connection relationship, and historical operation thresholds; Subsequently, input the fusion decision vector and the physical attribute data into the pre-trained fault classification model, and calculate the probability distribution of the fault mode by matching the vector features and physical constraint rules; Finally, determine the dominant fault type based on the peak value of the probability distribution and generate the corresponding fault type label.

[0079] Step 145: Generate a communication device fault prediction result with the spatial identifier as the fault location index and the fault type label as the element.

[0080] Among them, in the embodiment of the present application, first, with the spatial identifier as the fault location index and the fault type label as the element, integrate the spatial and type information through the data encapsulation algorithm, and finally generate a communication device fault prediction result.

[0081] The following is a specific example: First, extract the spatial identifier and time-frequency domain feature vector of the vibration and temperature anomaly features of the communication base station based on the cross-modal anomaly correlation feature set. Second, use the principal component analysis algorithm to perform feature compression on the time-frequency domain feature vector to generate a reduced-dimensional feature vector, and at the same time perform spatial mapping through the manifold learning algorithm to generate a projection feature vector. Then, perform weighted splicing fusion on the two types of feature vectors to generate a fusion decision vector. Subsequently, retrieve the heat dissipation attributes and connection relationships of the base station equipment in the dynamic physical topology model according to the spatial identifier, and combine the fusion decision vector to output an overload or short circuit fault type label. Finally, generate a base station fault prediction result with the spatial identifier locating the fault base station position and the fault type label as the element.

[0082] By executing steps 141 to 145, the embodiment of the present application realizes the deep compression and fusion of fault features through the spatial-time-frequency joint analysis of cross-modal features, accurately associates device attributes in combination with the physical topology model, generates a prediction result integrating spatial position and fault type, and improves the accuracy and positioning efficiency of communication network fault diagnosis.

[0083] In a possible embodiment, step 144: Retrieve the physical attributes of the corresponding communication device or connection segment in the dynamic physical topology model based on the spatial identifier, and combine the fusion decision vector to generate a fault type label, including: Step b1: Convert the physical attributes into a structured attribute vector, and generate a physical constraint rule library for the fusion decision vector according to the device material type, connection segment signal transmission threshold value, and environmental tolerance threshold value in the structured attribute vector.

[0084] Among them, physical properties refer to the inherent characteristics of communication devices or connection segments, including the conductivity coefficient determined by the device material type, the maximum carrying capacity specified by the signal transmission threshold value of the connection segment, the temperature and humidity adaptation range defined by the environmental tolerance threshold value, etc. The structured attribute vector refers to a mathematical vector obtained by encoding discrete physical properties by dimension, including the standardized feature representation generated through one-hot encoding or normalization processing. The device material type refers to the material category that makes up the communication device, including the core material properties affecting electrical performance such as metal conductors, optical fiber media, or semiconductor components. The signal transmission threshold value of the connection segment refers to the maximum signal strength threshold value allowed for network connection segments, including the theoretical safety boundary value calculated based on the signal-to-noise ratio and attenuation coefficient. The environmental tolerance threshold value refers to the extreme environmental parameter limit that the device can withstand, including the highest operating temperature, the maximum humidity range, and the electromagnetic interference resistance level. The physical constraint rule base refers to a database that stores the correlation between device physical characteristics and faults, including the mapping rules between the thermal expansion coefficient of the material and temperature faults, and the causal logic between the transmission threshold value and overload faults.

[0085] Step b2: Under the boundary conditions of the physical constraint rule base, perform a fault mode matching operation on the fusion decision vector to generate a set of candidate fault modes.

[0086] Among them, the fault mode matching operation refers to the process of comparing physical properties with preset fault rules, including identifying potential fault types through decision tree traversal or graph neural network inference. The set of candidate fault modes refers to the list of fault types generated by preliminary matching, including fault hypotheses that satisfy some physical constraints but may have conflicts.

[0087] Step b3: When there are fault modes with physical property conflicts in the set of candidate fault modes, perform conflict resolution based on the priority weights in the structured attribute vector, and output a set of valid fault modes.

[0088] Among them, physical property conflict refers to the phenomenon that the candidate fault mode conflicts with the actual characteristics of the device, including the logical error of requiring a "fiber break" fault in copper cable equipment. Conflict resolution refers to the process of resolving contradictions through priority rules, including sorting the fault probabilities according to the material tolerance threshold and eliminating low-probability items.

[0089] Step b4: Convert the set of valid fault modes into fault type labels compatible with the structured attribute vector.

[0090] The following is a specific example: Under the boundary conditions of the physical constraint rule base, perform a fault mode matching operation on the fusion decision vector to generate a set of candidate fault modes.

[0091] Fault pattern matching is the process of comparing physical properties with pre-set fault rules, including identifying potential fault types through decision tree traversal or graph neural network reasoning. The candidate fault pattern set is a list of fault types generated by preliminary matching, including fault hypotheses that satisfy some physical constraints but may conflict.

[0092] Step b3: When there are fault modes with conflicting physical properties in the candidate fault mode set, conflict resolution is performed based on the priority weights in the structured attribute vector, and a valid fault mode set is output.

[0093] Physical property conflicts occur when candidate failure modes conflict with the device's actual characteristics, including the logical error of requiring a "fiber break" failure in copper cable equipment. Conflict resolution is the process of resolving conflicts through priority rules, including sorting failure possibilities based on material tolerance thresholds and eliminating low-probability options.

[0094] Step b4: Convert the valid fault mode set into a fault type label that is compatible with the structured attribute vector.

[0095] The following is a specific example: First, the dynamic physical topology data of a Fifth Generation Mobile Communication Systems (5G) base station is retrieved based on the spatial identifier. The aluminum alloy chassis material type, the transmission threshold of the optical fiber connection segment, and the environmental tolerance threshold are extracted and integrated into a structured attribute vector. This vector is then input into the physical constraint rule library to match the candidate fault mode sets of "overheating and frequency reduction" and "optical failure interruption." A conflict between "optical failure interruption" and the aluminum alloy material is detected. Based on the material's thermal conductivity, the conflict is resolved, retaining the valid "overheating and frequency reduction" fault mode. Finally, a similarity calculation is performed between the fused decision vector and the valid fault mode to generate a fault type label of "overheating and frequency reduction due to heat dissipation failure."

[0096] By executing steps b1 to b4, the embodiment of the present application generates fault diagnosis results that conform to the actual characteristics of the equipment under the protection of the conflict resolution mechanism through deep binding of physical properties and fault rules, avoiding misjudgments that violate physical laws, and improving the engineering applicability and reliability of communication network fault location.

[0097] In a possible embodiment, step 142, based on the time-frequency domain feature vector, using a first feature selector to perform a feature compression operation to generate a dimension-reduced feature vector, and using a second feature selector to perform a spatial mapping operation to generate a projected feature vector, includes: Step c1: Based on the frequency band priority in the coordination constraint rule, assign importance weights to the frequency band components in the time-frequency domain feature vector to generate a weighted time-frequency domain feature vector.

[0098] Among them, the band priority refers to the importance level of bands divided according to communication protocols, including weight coefficients that are dynamically adjusted based on service types or channel quality. The weighted time-frequency domain feature vector refers to the time-frequency feature data scaled according to the band priority, including numerical representations used to reflect the relative importance of each frequency band.

[0099] Step c2: Input the weighted time-frequency domain feature vector into the first feature selector. According to the preset communication feature retention rules, eliminate redundant band components and retain key band features to generate a dimensionality-reduced feature vector.

[0100] Among them, the preset communication feature retention rules refer to predefined band screening criteria, including logical conditions that require the signal-to-noise ratio to be higher than a set threshold or the energy ratio to exceed a minimum retention ratio. The band component refers to the sub-vector in the time-frequency domain feature vector that represents a specific frequency interval, including discrete spectral units obtained by Fourier transform decomposition. The key band feature refers to the core band components retained after screening, including effective spectral components carrying main fault information.

[0101] Step c3: According to the mapping relationship between device ports and transmission paths in the dynamic physical topology model, construct a spatial association index for the weighted time-frequency domain feature vector.

[0102] Among them, the spatial association index is an index structure that identifies the relationship between features and physical locations, including encoded data of device port numbers, transmission path identifiers, and topological connection directions.

[0103] Step c4: Synchronously input the weighted time-frequency domain feature vector and the spatial association index into the second feature selector, aggregate the feature components across ports according to the physical connection topology, and generate a projection feature vector.

[0104] Among them, the physical connection topology refers to the actual connection architecture between communication devices, including logical models of port cascading relationships, transmission path orientations, and signal flow directions.

[0105] The following is a specific example: First, assign high weights to the high-frequency band components in the time-frequency domain feature vector based on the band priority of a 5G base station to generate a weighted time-frequency domain feature vector. Secondly, eliminate redundant band components with a signal-to-noise ratio lower than the threshold value through the first feature selector and retain the key features of the high-frequency band to generate a dimensionality-reduced feature vector. Subsequently, according to the mapping relationship between the device ports of the base station remote radio unit and the antenna transmission path, construct a spatial association index containing port identifiers for the weighted features. Finally, input the weighted features and the index into the second feature selector, aggregate the cross-port features according to the physical connection topology of the antenna cluster, and generate a projection feature vector representing signal coverage quality.

[0106] By performing steps c1 to c4, the embodiments of the present application enhance key fault features through band priority weighting, combine physical topology constraints to achieve feature dimensionality reduction and spatial aggregation, generate projection features that not only retain core band information but also conform to the device connection relationship, and improve the recognition and spatial interpretability of fault features in the communication system.

[0107] In a possible embodiment, S12. According to the multi-source heterogeneous data matrix, a dynamic physical topology model is generated by combining the optical attenuation eigenvalue of the distributed optical fiber sensor and the signal transmission direction identifier, including: Step 121. Extract the spatial coordinates of the optical signal event points from the multi-source heterogeneous data matrix, and bind the spatial coordinates to the corresponding communication device ports based on the preset device port mapping table to generate a set of port spatial coordinates.

[0108] Among them, the optical signal event point refers to the abnormal optical signal position point in the optical fiber caused by bending, breaking or joint loss, including the spatial coordinate data detected by the sudden change of the scattered signal intensity. The preset device port mapping table refers to the configuration table storing the corresponding relationship between the logical number of the communication device port and the actual physical position, including the mapping rules of the computer room number, rack position and port sequence. The set of port spatial coordinates refers to the three-dimensional position data set of the bound device ports, including the port geographical coordinate group composed of longitude, latitude and elevation information.

[0109] Step 122. Analyze the optical transmission connection relationship between adjacent communication device ports according to the optical attenuation eigenvalue of the distributed optical fiber sensor and the signal transmission direction identifier, and generate a set of port connection relationships.

[0110] Among them, the optical attenuation eigenvalue refers to the quantization value of the signal attenuation intensity on the optical fiber transmission path, including the unit distance loss coefficient calculated based on the backscattered optical power. The set of port connection relationships refers to the data set describing the physical connection between communication device ports, including the connection records of the source port, target port and transmission direction.

[0111] Step 123. Based on the set of port connection relationships and the transmission stability index, construct an optical transmission connection relationship graph between ports.

[0112] Among them, the transmission stability index refers to the parameter measuring the reliability of the optical transmission link, including the comprehensive score calculated based on the historical bit error rate, jitter rate and interruption frequency. The optical transmission connection relationship graph refers to the topological graph with ports as nodes and connection relationships as edges, including the graph structure with edge weights reflecting the transmission stability. The acquisition method of the transmission stability index is calculated based on the comprehensive score calculated from the historical bit error rate, jitter rate and interruption frequency.

[0113] Step 124: Bind the port space coordinate set, the optical transmission connection relationship diagram with the preset device physical property library. The device physical properties include port material type, optical fiber length, and environmental anti-interference level, and generate a weighted dynamic physical topology model.

[0114] Among them, the port material type refers to the physical material category of the optical fiber port interface, including characteristics such as ceramic ferrule, metal shell, or plastic connector that affect signal loss. The optical fiber length refers to the physical length value of the optical fiber between the ports of two communication devices, including accurate distance data measured based on an optical time domain reflectometer. The environmental anti-interference level refers to the tolerance rating of the port in a complex environment, including anti-electromagnetic interference intensity, waterproof and dustproof level, and temperature adaptation range. The weight calculation in the weighted dynamic physical topology model is obtained by fusing the transmission stability index and key physical property factors. The formula for calculating the weight W of the weight is: , where is the transmission stability index, is the optical fiber length factor, is the environmental anti-interference factor, is the port material factor.

[0115] The following is a specific example: First, extract the spatial coordinates of the optical cable bending event points from the optical fiber vibration monitoring data matrix, and bind the coordinates to the optical line terminal ports according to the core computer room device port mapping table to generate a port space coordinate set. Secondly, based on the optical attenuation eigenvalue and transmission direction of the distributed sensor, analyze the optical transmission connection relationship set between adjacent optical lines and optical network unit ports. Subsequently, construct a weighted backbone optical cable connection relationship diagram in combination with the transmission stability index. Finally, associate the port space coordinate set, the connection relationship diagram, and the ceramic material type, optical fiber length, and anti-interference level in the device physical property library to generate a dynamic physical topology model.

[0116] By executing Steps 121 to 124, the embodiment of the present application constructs a dynamic topology model that reflects the real-time connection status and reliability by fusing optical signal spatial positioning and device physical properties, providing an accurate physical layer basic framework for optical fiber network fault diagnosis.

[0117] Figure 2 FIG. is a schematic structural diagram of a communication device fault prediction system based on big data and optical fiber sensors provided by an embodiment of the present application. As Figure 2 shown, the system includes: An acquisition module 21, configured to integrate the temperature data, vibration data, and optical signal data of the communication device collected by the distributed optical fiber sensor into a multi-source heterogeneous data matrix according to the time series and spatial position.

[0118] A generation module 22, configured to generate a dynamic physical topology model according to a multi-source heterogeneous data matrix in combination with the optical attenuation eigenvalue and the signal transmission direction identifier of a distributed optical fiber sensor.

[0119] A separation module 23, configured to separate, based on the spatial constraint conditions of the dynamic physical topology model, an abnormal fluctuation component associated with at least one of the temperature abnormal data and the vibration abnormal data of a communication device from the optical signal data, so as to generate a cross-modal abnormal association feature set.

[0120] An analysis module 24, configured to analyze the cross-modal abnormal association feature set through a collaborative constraint rule to generate a communication device fault prediction result. The collaborative constraint rule is jointly generated by using a support vector machine and an extreme learning machine. The communication device fault prediction result includes potential fault positions and fault types in the communication device and is used for early warning.

[0121] Figure 2 The described communication device fault prediction system based on big data and an optical fiber sensor can execute Figure 1 The described communication device fault prediction method based on big data and an optical fiber sensor in the illustrated embodiment, and its implementation principle and technical effects will not be elaborated further. For the communication device fault prediction system based on big data and an optical fiber sensor in the above embodiment, the specific manners in which each module and unit perform operations have been described in detail in the embodiment related to the method, and will not be elaborated herein.

[0122] In a possible design, Figure 2 The communication device fault prediction system based on big data and an optical fiber sensor in the illustrated embodiment can be implemented as a computing device, such as Figure 3 as shown, the computing device may include a storage component 31 and a processing component 32.

[0123] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32.

[0124] The processing component 32 is used to execute the following processes: integrating the temperature data, vibration data, and optical signal data of the communication device collected by the distributed optical fiber sensor into a multi-source heterogeneous data matrix according to the time series and spatial position; generating a dynamic physical topology model based on the multi-source heterogeneous data matrix in combination with the optical attenuation eigenvalue and signal transmission direction identifier of the distributed optical fiber sensor; separating, based on the spatial constraint conditions of the dynamic physical topology model, an abnormal fluctuation component associated with at least one of the temperature abnormal data and vibration abnormal data of the communication device from the optical signal data to generate a cross-modal abnormal association feature set; generating a communication device fault prediction result by analyzing the cross-modal abnormal association feature set through a collaborative constraint rule, where the collaborative constraint rule is jointly generated by using a support vector machine and an extreme learning machine, and the communication device fault prediction result includes potential fault locations and fault types in the communication device, and is used for early warning.

[0125] Among them, the processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components for executing the above method.

[0126] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as random access memory (RAM), static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0127] Of course, the computing device may necessarily further include other components, such as an input / output interface, a display component, a communication component, etc.

[0128] The input / output interface provides an interface between the processing component and the peripheral interface module, and the peripheral interface module may be an output device, an input device, etc.

[0129] The communication component is configured to facilitate communication between the computing device and other devices in a wired or wireless manner, etc.

[0130] Among them, the computing device may be a physical device or a flexible computing host provided by a cloud computing platform, etc. At this time, the computing device may refer to a cloud server, and the above-mentioned processing component, storage component, etc. may be basic server resources leased or purchased from the cloud computing platform.

[0131] The embodiment of the present application also provides a computer storage medium storing a computer program, and when the computer program is executed by the computer, it can implement the above-mentioned Figure 1 A communication device fault prediction method based on big data and fiber optic sensors shown in the embodiment.

[0132] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0133] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0134] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A communication device fault prediction method based on big data and fiber optic sensors, characterized in that, Including: Integrating the temperature data, vibration data, and optical signal data of the communication device collected by the distributed fiber optic sensor into a multi-source heterogeneous data matrix according to the time series and spatial position; Generating a dynamic physical topology model based on the multi-source heterogeneous data matrix, in combination with the optical attenuation eigenvalue and signal transmission direction identifier of the distributed fiber optic sensor; Based on the spatial constraint conditions of the dynamic physical topology model, separating the abnormal fluctuation components associated with at least one of the abnormal temperature data and abnormal vibration data of the communication device from the optical signal data to generate a cross-modal abnormal association feature set; Through the collaborative constraint rule, analyzing the cross-modal abnormal association feature set to generate a communication device fault prediction result. The collaborative constraint rule is jointly generated by a support vector machine and an extreme learning machine. The communication device fault prediction result includes the potential fault location and fault type in the communication device, which is used for early warning.

2. The method according to claim 1, characterized in that The step of separating the abnormal fluctuation components associated with at least one of the abnormal temperature data and abnormal vibration data of the communication device from the optical signal data based on the spatial constraint conditions of the dynamic physical topology model to generate a cross-modal abnormal association feature set includes: Based on the spatial constraint conditions of the dynamic physical topology model, decomposing the optical signal data into independent spatial signal components corresponding to the communication device or connection segment; Performing time-frequency domain analysis on each of the independent spatial signal components, extracting the abnormal fluctuation components matching the fluctuation patterns of historical abnormal temperature or vibration data, to obtain the abnormal temperature data and abnormal vibration data of the communication device; Calculating the association weight between the abnormal temperature data and abnormal vibration data of the communication device through multi-scale correlation analysis; Based on the association weight, performing weighted superposition on the abnormal fluctuation components to generate a weighted abnormal fluctuation component set; Pruning the components conflicting with the device position or connection path from the weighted abnormal fluctuation component set to generate a cross-modal association feature set.

3. The method according to claim 2, characterized in that, The step of performing weighted superposition on the abnormal fluctuation components based on the association weight to generate a weighted abnormal fluctuation component set includes: Taking the association weight as the amplitude scaling factor, performing an element-wise multiplication operation on each of the abnormal fluctuation components to generate a weighted abnormal fluctuation component; Based on the spatial constraint conditions of the dynamic physical topology model, assigning a unique spatial identifier to each of the weighted abnormal fluctuation components; Storing the weighted abnormal fluctuation component and the corresponding spatial identifier as a key-value pair into a hash map data structure; Based on the complete storage state of the hash map data structure, outputting the weighted abnormal fluctuation component set.

4. The method according to claim 1, wherein The step of analyzing the cross-modal abnormal association feature set through the collaborative constraint rule to generate a communication device fault prediction result includes: Based on the cross-modal abnormal association feature set, extracting the spatial identifier and the corresponding time-frequency domain feature vector of each cross-modal abnormal association feature; Based on the time-frequency domain feature vector, performing a feature compression operation using the first feature selector to generate a dimensionality-reduced feature vector, and performing a spatial mapping operation using the second feature selector to generate a projection feature vector; Perform a feature fusion operation on the dimensionality-reduced feature vector and the projected feature vector to generate a fused decision vector; Retrieve the physical attributes of the corresponding communication device or connection segment in the dynamic physical topology model based on the spatial identifier, and combine with the fused decision vector to generate a fault type label; Generate a communication device fault prediction result with the spatial identifier as the fault location index and the fault type label as the element.

5. The method according to claim 4, wherein The retrieving the physical attributes of the corresponding communication device or connection segment in the dynamic physical topology model based on the spatial identifier, and combining with the fused decision vector to generate a fault type label includes: Convert the physical attributes into a structured attribute vector, and generate a physical constraint rule library for the fused decision vector according to the device material type, connection segment signal transmission threshold value, and environmental tolerance threshold value in the structured attribute vector; Perform a fault mode matching operation on the fused decision vector under the boundary conditions of the physical constraint rule library to generate a candidate fault mode set; When there are fault modes with physical attribute conflicts in the candidate fault mode set, perform conflict resolution based on the priority weights in the structured attribute vector, and output an effective fault mode set; Convert the effective fault mode set into a fault type label compatible with the structured attribute vector.

6. The method according to claim 4, characterized in that, The generating a dimensionality-reduced feature vector by performing a feature compression operation on the time-frequency domain feature vector using a first feature selector, and generating a projected feature vector by performing a spatial mapping operation using a second feature selector includes: Assign importance weights to each frequency band component in the time-frequency domain feature vector based on the frequency band priority in the collaborative constraint rule to generate a weighted time-frequency domain feature vector; Input the weighted time-frequency domain feature vector into the first feature selector, and eliminate the redundant frequency band components according to the preset communication feature retention rule, and retain the key frequency band features to generate a dimensionality-reduced feature vector; Construct a spatial association index for the weighted time-frequency domain feature vector according to the mapping relationship between device ports and transmission paths in the dynamic physical topology model; Synchronously input the weighted time-frequency domain feature vector and the spatial association index into the second feature selector, and aggregate the feature components across ports according to the physical connection topology to generate a projected feature vector.

7. The method according to claim 1, wherein The generating a dynamic physical topology model according to the multi-source heterogeneous data matrix, in combination with the optical attenuation eigenvalue and signal transmission direction identifier of the distributed optical fiber sensor includes: Extract the spatial coordinates of the optical signal event points from the multi-source heterogeneous data matrix, and bind the spatial coordinates to the corresponding communication device ports based on the preset device port mapping table to generate a set of port spatial coordinates; Analyze the optical transmission connection relationship between adjacent communication device ports according to the optical attenuation eigenvalue and signal transmission direction identifier of the distributed optical fiber sensor to generate a set of port connection relationships; Construct a graph of optical transmission connection relationships between ports based on the set of port connection relationships and the transmission stability index; Bind the set of port space coordinates, the optical transmission connection relationship diagram to a preset device physical attribute library, where the device physical attributes include port material type, optical fiber length, and environmental anti-interference level, to generate a weighted dynamic physical topology model.

8. A communication device fault prediction system based on big data and fiber optic sensors, characterized in that, It includes: An acquisition module, configured to integrate the temperature data, vibration data, and optical signal data of a communication device collected by a distributed optical fiber sensor into a multi-source heterogeneous data matrix according to time series and spatial position; A generation module, configured to generate a dynamic physical topology model according to the multi-source heterogeneous data matrix, in combination with the optical attenuation eigenvalue and signal transmission direction identifier of the distributed optical fiber sensor; A separation module, configured to separate, based on the spatial constraint conditions of the dynamic physical topology model, an abnormal fluctuation component associated with at least one of the temperature abnormal data and vibration abnormal data of the communication device from the optical signal data, so as to generate a cross-modal abnormal association feature set; An analysis module, configured to analyze the cross-modal abnormal association feature set through a collaborative constraint rule to generate a communication device fault prediction result. The collaborative constraint rule is jointly generated by a support vector machine and an extreme learning machine. The communication device fault prediction result includes potential fault positions and fault types in the communication device, and is used for early warning.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a communication device fault prediction method based on big data and optical fiber sensors according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that, A computer program is stored, and when the computer program is executed by a computer, it implements a communication device fault prediction method based on big data and optical fiber sensors according to any one of claims 1 to 7.

Images