CN-121984846-A - Indoor distribution system fault detection method

CN121984846ACN 121984846 ACN121984846 ACN 121984846ACN-121984846-A

Abstract

The invention belongs to the technical field of indoor mobile communication, and particularly relates to an indoor distribution system fault detection method which comprises the following steps of S1, constructing an indoor distribution system fault data acquisition system, collecting multidimensional original data, S2, preprocessing the original data to obtain target fault characteristics, S3, training a built training model through the target fault characteristics to obtain a fault detection model, S4, preprocessing the multidimensional data collected in real time, inputting the preprocessed multidimensional data into the fault detection model to obtain a real-time fault type and a real-time fault position coordinate, S5, building a high-precision three-dimensional building indoor site distribution model, S6, mapping the real-time fault position coordinate into the built model to obtain a fault detection result and a three-dimensional fault display diagram, S7, generating fault early warning information according to the fault detection result, and pushing the fault early warning information to an operation and maintenance management platform. The invention can solve the problems of thick positioning, low efficiency and more missed detection in the fault detection and positioning of the indoor distribution system in the prior art.

Inventors

SHAO LEI
ZENG CHENG
HUANG HAI
TANG JIANFENG
LUO YIWEI
GU MIN
FU ZIJIAN
GUO SHIWU
LI PU

Assignees

中国铁塔股份有限公司广西壮族自治区分公司

Dates

Publication Date: 20260505
Application Date: 20260312

Claims (10)

1. The fault detection method for the indoor distribution system is characterized by comprising the following steps of: S1, constructing an indoor distribution system fault data acquisition system, and acquiring multidimensional original data in the running process of the system; s2, preprocessing the collected original data to obtain target fault characteristics; s3, constructing a training model, and training the training model through target fault characteristics to obtain a fault detection model; s4, preprocessing the multidimensional data acquired in real time through the S2, and inputting the preprocessed multidimensional data into a fault detection model to obtain a real-time fault type and a real-time fault position coordinate; s5, building structure data and system equipment data are collected, and a high-precision three-dimensional building indoor station distribution model is built through the building structure data and the system equipment data; S6, mapping the real-time fault position coordinates into a high-precision three-dimensional building indoor site distribution model, and combining the real-time fault types to obtain a fault detection result and a three-dimensional fault display diagram so as to realize the visual presentation of the fault position; And S7, generating fault early warning information according to a fault detection result, and pushing the fault early warning information and the three-dimensional fault display diagram to the operation and maintenance management platform.
2. The indoor distribution system fault detection method according to claim 1, wherein the data acquisition in S1 is realized through sensors, intelligent acquisition terminals and a network management system which are deployed at key nodes of the indoor distribution system, the acquisition frequency is configurable, the multi-dimensional original data acquired in S1 comprises active equipment operation parameters, antenna feeder parameters, signal coverage parameters and environment parameters, wherein the active equipment operation parameters comprise equipment input power, output power, working current, working voltage and temperature, the antenna feeder parameters comprise standing wave ratio, feeder loss and joint attenuation, the signal coverage parameters comprise signal strength, signal to noise ratio and error rate of different indoor areas, and the environment parameters comprise indoor temperature, humidity and electromagnetic interference strength; The fault early warning information in the S7 comprises fault occurrence time, fault type, fault position, fault severity and recommended repair scheme, wherein the fault severity is divided into three grades of slight, general and serious according to the degree that fault parameters deviate from a normal range, and the fault early warning information and the three-dimensional fault display diagram are pushed to an operation and maintenance management platform through a wireless communication network and notify relevant operation and maintenance personnel.
3. The indoor distribution system fault detection method according to claim 1, wherein the preprocessing of the collected raw data in S2 specifically includes the following steps: s21, carrying out data cleaning on the collected multi-dimensional original data to obtain multi-dimensional data after data cleaning, wherein the data cleaning comprises the following specific steps: S211, eliminating abnormal values of the collected multi-dimensional original data by adopting a box line graph method to obtain eliminated multi-dimensional data; s212, filling the missing values of the removed multidimensional data by using a Lagrange interpolation method to obtain the filled multidimensional data; S213, filtering the filled multidimensional data by a wavelet threshold denoising method to obtain the multidimensional data after data cleaning; S22, carrying out normalization processing on the multi-dimensional data after data cleaning by adopting a maximum-minimum normalization method to obtain normalized multi-dimensional data, wherein the normalization formula is as follows: ; Wherein, the For multi-dimensional data after data cleansing, As a minimum value for this dimension data, For the maximum value of the dimension data, Is normalized multidimensional data; s23, performing feature analysis on the normalized multidimensional data to obtain target fault features, wherein the target fault features are specifically as follows: S231, performing dimension reduction processing on the normalized multidimensional data by using a PCA principal component analysis algorithm to obtain preliminary fault characteristics, and reserving principal components of which the accumulated contribution rate reaches a threshold value; S232, calculating mutual information entropy between each preliminary fault feature and a preset fault type, and screening out the preliminary fault features with the mutual information entropy larger than a threshold value theta as target fault features.
4. The indoor distribution system fault detection method according to claim 1, wherein the step of constructing a training model in S3, training the training model through target fault characteristics to obtain a fault detection model, comprises the following steps: S31, dividing all target fault characteristic data sets into a training set, a verification set and a test set according to a proportion, wherein the training set is used for learning training model parameters, the verification set is used for adjusting super parameters of the training model, and the test set is used for final evaluation of performance of the training model; The SMOTE is adopted to expand a few types of fault samples in the training set, so that the quantity proportion of the various types of fault samples is ensured to be balanced; S32, constructing a training model, wherein the training model comprises an input layer, a spatial feature extraction module, a time sequence feature extraction module, a fusion module and a classification and positioning output layer; S33, performing dimension conversion on the training set through the input layer to obtain a converted training set; s34, extracting spatial features of the converted training set through a spatial feature extraction module to obtain spatial features; S35, extracting time sequence features from the space features through a time sequence feature extraction module to obtain the time sequence features; s36, fusing the spatial features and the time sequence features through a fusion module to obtain fused features; S37, processing the fused features through a classification and positioning output layer to obtain a predicted fault type and a predicted fault position coordinate; S38, calculating a loss function of the predicted fault type, the predicted fault position coordinate, the real fault type and the real fault position coordinate to obtain a total loss function; S39, optimizing the training model by adopting an Adam optimizer and a total loss function, and monitoring the performance of the training model in real time through a verification set to prevent the training model from being fitted; S310, fine tuning is carried out on the optimized training model according to the test set, key super parameters of the training model are optimized through a grid search method, and detection performance of the optimized training model is further improved, so that a fault detection model is obtained.
5. The indoor distribution system fault detection method according to claim 4, wherein the spatial feature extraction module comprises a first convolution layer, a SENet channel attention mechanism layer, a first BN batch normalization layer, a first ReLU activation function, a second convolution layer, a second BN batch normalization layer, a second ReLU activation function, a third convolution layer, a third BN batch normalization layer, a third ReLU activation function and a first global average pooling layer, and the step S34 of extracting spatial features of the converted training set by the spatial feature extraction module to obtain the spatial features specifically comprises the following steps: s341, performing feature extraction on the converted training set through a first convolution layer to obtain a first feature extracted training set; s342, carrying out weighting treatment on the training set after the first feature extraction through a SENet channel attention mechanism layer to obtain a weighted training set; s343, normalizing the weighted training set through a first BN batch normalization layer to obtain a first normalized training set; s344, mapping the first normalized training set through a first ReLU activation function to obtain a second mapped training set; s345, performing feature extraction on the second mapped training set through the second convolution layer to obtain a second feature extracted training set; s346, normalizing the training set after the second feature extraction through a second BN batch normalization layer to obtain a second normalized training set; S347, mapping the second normalized training set through a second ReLU activation function to obtain a third mapped training set; S348, performing feature extraction on the third mapped training set through a third convolution layer to obtain a third feature extracted training set; s349, normalizing the training set after the third feature extraction through a third BN batch normalization layer to obtain a third normalized training set; s3410, mapping the third normalized training set through a third ReLU activation function to obtain a fourth mapped training set; S3411, pooling the fourth mapped training set through the first global average pooling layer to obtain spatial features.
6. The indoor distributed system fault detection method according to claim 5, wherein the SENet channel attention mechanism layer includes a second global average pooling layer, two full-connection layers and a Sigmoid activation function, and the step S342 is to perform weighting processing on the first feature extracted training set through the SENet channel attention mechanism layer to obtain a weighted training set, and specifically includes the following steps: s3421, pooling the training set after the first feature extraction through a second global average pooling layer to obtain a training set after the first pooling; s3422, mapping the first pooled training set through two full-connection layers to obtain a first mapped training set; S3423, calculating a weight coefficient of the first mapped training set through a Sigmoid activation function; S3424, multiplying the weight coefficient and the training set after the first feature extraction channel by channel to obtain a weighted training set.
7. The indoor distribution system fault detection method according to claim 4, wherein the timing sequence feature extraction module comprises a first Bi-LSTM unit, a first dropout layer, a second Bi-LSTM unit and a second dropout layer, and the step S35 of extracting the timing sequence feature from the spatial feature by the timing sequence feature extraction module to obtain the timing sequence feature specifically comprises the following steps: S351, carrying out feature extraction on the spatial features through a first Bi-LSTM unit to obtain the spatial features after the first feature extraction; S352, performing feature optimization on the first feature extracted spatial features through the first dropout layer to obtain first feature optimized spatial features; s353, performing feature extraction on the spatial features through a second Bi-LSTM unit to obtain second feature extracted spatial features; s354, performing feature optimization on the second feature extracted spatial features through a second dropout layer to obtain second feature optimized spatial features; S355, splicing the space features after the first feature optimization and the space features after the second feature optimization to obtain the time sequence features.
8. The indoor distribution system fault detection method according to claim 4, wherein in S36, the spatial feature and the time sequence feature are fused by a fusion module to obtain a fused feature, and the method specifically comprises the following steps: S361, calculating a space feature weight coefficient, wherein the formula is as follows: ; Wherein, the As the weight coefficient of the spatial feature, For the calculation of the variance of the values, As a feature of the space it is possible to provide, Is a time sequence feature; s362, calculating a time sequence feature weight coefficient, wherein the formula is as follows: ; Wherein, the As the weight coefficient of the time sequence characteristic, For the calculation of the variance of the values, As a feature of the space it is possible to provide, For the time sequence characteristics, satisfy ; S363, calculating the fused characteristics, wherein the formula is as follows: ; Wherein, the Is a fused feature.
9. The method for detecting faults of indoor distribution system according to claim 4, wherein the classification and positioning output layer comprises a ReLU activation function, a Softmax activation function and a linear activation function, and the step S37 is to process the fused features through the classification and positioning output layer to obtain the predicted fault type and the predicted fault position coordinates, and specifically comprises the following steps: S371, mapping the fused features through a ReLU activation function; s372, predicting the fault type of the mapped feature through a Softmax activation function to obtain a predicted fault type; s373, predicting the fault location coordinates of the mapped features through the linear activation function to obtain predicted fault location coordinates.
10. The indoor distribution system fault detection method according to claim 4, wherein the step of calculating the loss function of the predicted fault type, the predicted fault location coordinate, the true fault type, and the true fault location coordinate in S38 to obtain a total loss function specifically includes the steps of: S381, calculating a cross entropy loss function of the predicted fault type and the real fault type to obtain the cross entropy loss function; S382, calculating a mean square error loss function of the predicted fault position coordinate and the real fault position coordinate to obtain the mean square error loss function; S383, calculating a total loss function, wherein the formula is as follows: ; Wherein, the As a total loss function, weight coefficient In order to be able to configure the parameters, In order to cross-entropy loss function, Is a mean square error loss function.

Description

Indoor distribution system fault detection method Technical Field The invention belongs to the technical field of indoor mobile communication, and particularly relates to a fault detection method of an indoor distribution system. Background The indoor distribution system is a core infrastructure for improving indoor mobile communication environment, and ensures the communication quality of users by uniformly distributing base station signals in an indoor area through equipment such as a feeder line, a coupler, an antenna and the like. With the popularization of communication technologies such as 5G, the complexity of an indoor distribution system is greatly improved, the number of devices is increased, the wiring concealment is enhanced, the probability of occurrence of faults is remarkably increased, and common faults comprise uplink interference, weak coverage, equipment starting control, parameter configuration errors, cross coverage Shi Yancha exceeding and the like. The existing indoor distribution system fault detection and positioning method has a plurality of defects: Firstly, relying on a threshold alarming mechanism of a network management system, the system only can identify the abnormality of an overall signal, can not accurately locate hidden faults of floors and regional levels, is difficult to capture actively on a network management side, and often depends on user complaints or manual inspection trigger investigation, so that fault discovery is delayed, false alarm and missing alarm risks are increased; secondly, fault positioning relies on manual point-by-point test, all equipment such as RRU, feeder line, antenna and the like need to be checked, positioning time delay is large, site resource occupation is high, and continuous monitoring closed loop treatment is difficult to form; Thirdly, the existing detection method based on the model (such as three-dimensional modeling combined with MR data) has the problems of poor model generalization capability and low fault recognition accuracy, is difficult to adapt to different building structures and different equipment types of different scenes, has insufficient recognition capability on complex faults such as high-order intermodulation interference, equipment recessive attenuation and the like, and is difficult to stably extract distinguishable features under noise disturbance and scene difference. Aiming at the defects of the prior art (especially the fault detection method based on the model network) in terms of fault recognition accuracy, scene generalization capability and positioning granularity, the multi-dimensional degree of structural optimization, feature enhancement, training strategy improvement and the like of the model network is required to be improved so as to improve the precision, efficiency and scene adaptability of fault detection, and the core problems of thick positioning, low efficiency and more missed detection in the prior art are solved. Therefore, the application provides a fault detection method of an indoor distribution system, which aims at solving the technical problems of fault detection and classification, fault position (floor/region/coordinate or associated equipment) determination, three-dimensional visual display of positioning results, alarm pushing closed loop and the like under the condition of multisource operation data in an intelligent operation and maintenance scene of the indoor distribution system, and overcomes the defects of coarse positioning granularity, large positioning delay, insufficient complex fault recognition capability, poor cross-scene suitability and the like in the prior art. The information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person of ordinary skill in the art. Disclosure of Invention The invention aims to provide a fault detection method of an indoor distribution system, which aims to solve the technical problems of fault detection and classification, fault position determination, three-dimensional visual display of positioning results, alarm pushing closed loop and the like under the condition of multi-source operation data. In order to achieve the above object, the present invention provides the following technical solutions: the fault detection method of the indoor distribution system comprises the following steps: S1, constructing an indoor distribution system fault data acquisition system, and acquiring multidimensional original data in the running process of the system; s2, preprocessing the collected original data to obtain target fault characteristics; s3, constructing a training model, and training the training model through target fault characteristics to obtain a fault detection model; s4, preprocessing the multidimensional data acquired in real time throu