Communication iron tower and monitoring method and system thereof

Abstract

The invention relates to the technical field of communication iron towers, in particular to a communication iron tower and a monitoring method and system.The communication iron tower comprises a communication iron tower body, a communication iron tower monitoring device and a far-end monitoring platform, and the communication iron tower body is used for guaranteeing normal operation of communication and comprises a tower body structure, a power system, communication equipment and a safety system; the communication iron tower monitoring device is used for monitoring the operation state of an iron tower in real time and comprises a sensor assembly and a communication module. The far-end monitoring platform is used for carrying out centralized management and intelligent analysis and classification on the operation state of the iron tower, and the intelligent analysis and classification comprises the steps of removing environmental noise from the sound and vibration signals, judging whether the communication iron tower breaks down or not according to the characteristics of the sound and vibration signals, and timely giving an early warning to operation and maintenance personnel after the iron tower fault is detected. Therefore, the problems in the prior art that early faults of a communication tower are difficult to find, and feature signals are difficult to extract due to environmental interference are solved.

Description

Technical Field

[0001] This application relates to the technical field of communication towers, and particularly to a communication tower and its monitoring method and system. Background Art

[0002] As a key supporting structure of the wireless network, the safety and stability of communication towers have been increasingly concerned. The monitoring and maintenance of traditional communication towers mainly rely on manual inspections. Traditional manual inspection methods have disadvantages such as low efficiency, high cost, and untimely detection, and it is difficult to meet the requirements of real-time and accurate monitoring.

[0003] In Related Art 1, CN115134345B discloses an inclination monitoring system and data transmission method for a tower. By installing inclination sensors on the communication tower to monitor the state of the tower, there are obvious technical limitations, such as a single detection method and the inability to comprehensively capture the complex characteristics of the tower operation state; affected by environmental interference, sensor data is easily affected by environmental noises such as strong winds, rainfall, and electromagnetic interference.

[0004] In Related Art 2, CN111238446A discloses a communication tower inclination monitoring system. By installing a camera device on the communication tower and using the method of taking real-time images to detect the communication tower, and by analyzing the taken images, it is judged whether the structure of the tower is inclined. This method has a high cost, is affected by light conditions, and it is difficult to detect early faults of the tower. For example, the tower needs to be inclined by more than a certain deformation to be recognized. Summary of the Invention

[0005] This application provides a communication tower and its monitoring method and system to solve problems such as the difficulty in detecting early faults of communication towers and the difficulty in extracting characteristic signals due to environmental interference in the prior art.

[0006] The first aspect of the present application provides a communication tower, including: a communication tower body, a communication tower monitoring device, and a remote monitoring platform. Among them, the communication tower body is used to ensure the normal operation of communication, including a tower body structure, a power system, communication equipment, and a safety system. Among them, the tower body structure adopts a bionic honeycomb structure, including a tower body, a foundation, a hexagonal antenna platform, a ladder, etc. The power equipment includes a power supply line, a solar panel, a distribution box, etc. The communication equipment includes an antenna, a feeder, a base station device, etc. The safety system includes lightning protection, grounding, and safety protection facilities; the communication tower monitoring device is used to monitor the operation status of the tower in real time, including a sensor component and a communication module. The sensor component is used to obtain the sound, vibration, and environmental noise data during the operation of the tower. The communication module is used to upload the data obtained by the sensor component to the remote monitoring platform, supporting 4G, 5G wireless transmission, and fiber optic wired backhaul; the remote monitoring platform is used to centrally manage and intelligently analyze and classify the operation status of the tower. The intelligent analysis and classification include removing environmental noise from the acoustic vibration signal according to the WaveNet noise reduction network, analyzing and classifying the characteristics of the acoustic vibration signal according to the Transformer-CNN hybrid model, judging whether the communication tower is faulty, and warning the operation and maintenance personnel in time after detecting the tower fault.

[0007] In a preferred embodiment, the sensor component includes: a sound sensor, a low-frequency vibration sensor, and an environmental sensor. Among them, the sound sensor is used to capture various sound signals generated during the operation of the communication tower; the low-frequency vibration sensor is used to detect the vibration information of the communication tower structure to detect silent faults; the environmental sensor is used to obtain environmental parameters such as wind speed and rain intensity in real time to provide environmental reference data for subsequent signal preprocessing.

[0008] The second aspect of the present application provides a communication tower monitoring method, including the following steps: Step A1, obtain the sound, vibration, and environmental information of the communication tower; Step A2, build and train a WaveNet noise reduction network, preprocess the obtained information, and obtain a sound signal and a vibration signal after filtering environmental noise; Step A3, extract the features of the sound signal and the vibration signal, adopt the PCA dimensionality reduction algorithm, reduce the dimensionality and fuse the features of the acoustic vibration signal to obtain a comprehensive feature vector; Step A4, input the comprehensive feature vector into the Transformer-CNN hybrid model for analysis and classification to obtain a classification result. Among them, the classification result is a normal acoustic vibration mode and a faulty acoustic vibration mode; Step A5: If the classification result is the fault acoustic-vibration mode, then compare it with the acoustic-vibration feature library, identify the fault acoustic-vibration features through pattern matching and similarity calculation methods, determine the fault type and degree, remind the operation and maintenance personnel to repair, recommend a repair plan for the operation and maintenance personnel based on historical repair data and the fault type, perform repairs according to the recommended repair plan, and update the newly occurred fault acoustic-vibration features to the acoustic-vibration feature library; if the classification result is the normal acoustic-vibration mode, then continue to run.

[0009] In a preferred embodiment, build and train a WaveNet noise reduction network to preprocess the acquired information to obtain sound signals and vibration signals after filtering environmental noise. The specific steps are as follows: Step B1: Input the acquired sound, vibration, and environmental information into the starting convolutional layer for splicing, and convert the data with different numbers of channels into data with the target number of channels to obtain preliminary data. Step B2: Input the preliminary data into a multi-level dilated convolutional layer to learn features. The dilated convolutional layer includes a filtering convolutional layer, a gated convolutional layer, a residual connection convolutional layer, and a skip connection convolutional layer. Among them, extract features of different scales through the filtering convolutional layer, selectively retain and suppress the features of different scales according to the gated convolutional layer, multiply the results of the filtering convolutional layer and the gated convolutional layer to obtain a first eigenvalue, and input the first eigenvalue into the residual connection convolutional layer and the skip connection convolutional layer for information transmission. Among them, the residual connection convolutional layer adds the input value of the current dilated convolutional layer to the first eigenvalue to obtain a target value, and accumulates the target values of each layer according to the skip connection convolutional layer to obtain feature information of different layers. Step B3: Use the ReLU activation function for the feature information of different layers, set negative values to zero to enhance the nonlinearity of the features, and obtain the final output according to two ending convolutional layers. The first ending convolutional layer transforms and integrates the features, and the second ending convolutional layer converts the target number of channels to 2 and outputs the denoised dual-channel data. The dual-channel data are respectively sound signal data and vibration signal data.

[0010] In a preferred embodiment, the formula for inputting the preliminary data into a multi-level dilated convolutional layer to learn features is as follows: The formula for the filtering convolutional layer is , where is the weight of the filtering convolutional layer, x is the input, is the bias, and tanh() is the activation function that maps the input to the interval (−1, 1); The formula for the gated convolutional layer is , where is the weight of the gated convolutional layer, x is the input, is the bias, ( ) is the activation function, which maps the input to the interval (−1, 1); The formula for the residual connection convolutional layer is , where is the weight of the residual connection convolutional layer, is the input of the current dilated convolutional layer, is the updated input, y is the output of the filtering convolutional layer and the gated convolutional layer, and its calculation formula is ; The formula for the skip connection convolutional layer is , where is the weight of the residual connection convolutional layer, and S is the sum of the skip connection outputs of all previous layers; In a preferred embodiment, the dimensionality reduction and fusion of the acoustic-vibration signal features are as follows: Step C1: Extract the features of the noise-reduced sound signal and vibration signal respectively to construct the feature matrix X; Step C2: Perform mean processing on the feature matrix X to obtain the standardized feature matrix Z; Step C3: Calculate the covariance matrix C for the standardized feature matrix Z, perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues sorted by size and the corresponding eigenvectors. Among them, the formula for calculating the covariance is: , Step C4: According to the size of the eigenvalues, select the top p largest eigenvalues and their corresponding eigenvectors, and form the projection matrix P with these p eigenvectors; Step C5: Project the standardized data matrix Z onto the projection matrix P to obtain the dimensionality-reduced matrix Y = ZP, and each row of the matrix Y is a comprehensive feature vector.

[0011] In a preferred embodiment, the input of the comprehensive feature vector into the Transformer-CNN hybrid model for analysis and classification is as follows: Step D1: Input the comprehensive feature vector into the CNN convolutional layer. The CNN convolutional layer contains multiple convolutional kernels, and captures the key feature information of the data within the local range through convolutional operations to obtain the output data; Step D2: Use the max-pooling method in the CNN pooling layer to downsample the output data of the convolutional layer, and splice the pooling outputs of all convolutional kernels to obtain the output of the CNN layer ; Step D3: Input It is input into the Transformer layer for linear transformation to obtain the embedding vector E. The position encoding PE is added to the embedding vector. The multi-head attention mechanism is used to capture information in different subspaces to obtain the token sequence. The token sequence is transformed and processed based on the feed-forward neural network, which includes two linear layers and an activation function. After being processed by the multi-head attention mechanism and the feed-forward neural network, a normalization operation is performed to obtain the output of the Transformer layer. ; Step D4: Input into the fully connected layer for classification, and use the Softmax function to convert the output into the classification probability P; Step D5: Determine the final classification result according to the classification probability P. If P is greater than or equal to the target threshold, it is determined that the classification result is the normal acoustic vibration mode; otherwise, it is determined that the classification result is the faulty acoustic vibration mode.

[0012] In a preferred embodiment, the method for identifying faulty acoustic vibration characteristics through pattern matching and similarity calculation is as follows: Step E1: Calculate the similarity between the faulty acoustic vibration feature vector to be identified and the feature vectors in the acoustic vibration feature library according to the cosine similarity algorithm. The calculation formula is: , where is the faulty acoustic vibration feature vector to be identified, is the feature vector in the feature library; Step E2: Take the highest similarity as the target feature vector, compare the similarity value with the target threshold. If the similarity value is higher than the target threshold, the corresponding fault type and degree are used as the preliminary judgment of the current fault. If the similarity value is lower than the target threshold, the current fault is marked as a new unknown fault type; Step E3: If the initially determined fault type is a known fault, according to the fault type and degree, search for the historical maintenance data of similar faults in the acoustic vibration feature library to obtain the historical maintenance plan, generate a recommended maintenance plan based on the historical maintenance plan, notify the corresponding operation and maintenance personnel to repair and recommend the recommended maintenance plan to the operation and maintenance personnel, repair according to the recommended maintenance plan, and add the newly occurred faulty acoustic vibration characteristics, fault type, fault degree, and detailed information of this repair to the acoustic vibration feature library; if the initially determined fault type is an unknown fault, directly notify the corresponding operation and maintenance personnel to repair the unknown fault, and add the newly occurred faulty acoustic vibration characteristics, fault type, fault degree, and detailed information of this repair to the acoustic vibration feature library according to the repair feedback.

[0013] In a preferred embodiment, the acoustic-vibration feature database includes the structure of acoustic-vibration feature data, new fault features, and redundant data, where The structure of the acoustic-vibration feature data includes the fault type, acoustic-vibration feature vector, fault degree label, and historical maintenance records, where the historical maintenance records include the maintenance time, maintenance personnel, maintenance method, and components used; Add the new fault features. When a new fault is detected, obtain the acoustic-vibration feature vector, and according to the maintenance feedback, add the fault type, acoustic-vibration feature vector, fault degree label of the newly occurred fault, and the detailed information of this maintenance to the acoustic-vibration feature library; Clean up the redundant data. Periodically clean the acoustic-vibration feature database. For two acoustic-vibration feature vectors with high similarity and highly similar fault types, fault degree labels, and maintenance methods, merge the maintenance records and delete one of the records.

[0014] The third aspect embodiment of this application provides a communication tower monitoring system, including: an acquisition module for acquiring the sound, vibration, and environmental information of the communication tower; a noise reduction module for preprocessing the acquired information according to the WaveNet noise reduction network to obtain a sound signal and a vibration signal after filtering out environmental noise; a dimensionality reduction and fusion module for extracting features from the sound signal and the vibration signal, using the PCA dimensionality reduction algorithm to perform dimensionality reduction and fusion on the acoustic-vibration signal features to obtain a comprehensive feature vector; an analysis and classification module for inputting the comprehensive feature vector into a Transformer-CNN hybrid model for analysis and classification to obtain a classification result, where the classification result is a normal acoustic-vibration mode and a fault acoustic-vibration mode; a maintenance module for maintaining the communication tower determined to be in the fault acoustic-vibration mode. If the classification result is the fault acoustic-vibration mode, then compare it with the acoustic-vibration feature library, identify the fault acoustic-vibration features through pattern matching and similarity calculation methods, determine the fault type and fault degree, remind the operation and maintenance personnel to perform maintenance, recommend a maintenance plan for the operation and maintenance personnel based on the historical maintenance data and the fault type, perform maintenance according to the recommended maintenance plan, and update the newly occurred fault acoustic-vibration features to the acoustic-vibration feature library.

[0015] The beneficial effects of the present invention are as follows: By using sound characteristics and vibration characteristics to detect the operating conditions of communication towers, the operating state of the towers can be reflected in real time, small changes in the tower state can be captured in a timely manner, problems can be detected at the initial stage of a fault, and thus timely and effective measures can be taken to avoid the further development and expansion of the fault; By using the WaveNet noise reduction network to process the signals collected by the sound sensor and the low-frequency vibration sensor, environmental noise interference can be effectively filtered, making the collected sound and vibration signals closer to the real operating state; By using the PCA dimensionality reduction algorithm to perform dimensionality reduction and fusion on the characteristic parameters of the sound and vibration signals, the amount of calculation can be reduced, the processing efficiency can be improved, key information can be retained, the representativeness and discrimination of the characteristics can be enhanced, and a foundation for fault mode recognition can be laid; By using the Transformer-CNN hybrid model to analyze and classify the comprehensive feature vector, while capturing the local subtle changes in the acoustic-vibration signals, the overall change trend can also be grasped, the normal acoustic-vibration mode and the fault acoustic-vibration mode can be more accurately identified, and the fault recognition accuracy can be improved. Thus, the problems in the prior art such as the difficulty in detecting early faults of communication towers and the difficulty in extracting characteristic signals due to environmental interference are solved.

[0016] Additional aspects and advantages of the present application will be given in part in the following description, become apparent in part from the following description, or be understood through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and / or additional aspects and advantages of the present application will become apparent and be readily understood from the following description of the embodiments in conjunction with the drawings, where: Figure 1 is a schematic structural diagram of a communication tower provided according to an embodiment of the present application; Figure 2 is a schematic diagram of a communication tower provided according to an embodiment of the present application; Figure 3 is a flowchart of a method for monitoring a communication tower provided according to an embodiment of the present application; Figure 4 is a schematic diagram of the composition of a communication tower monitoring system provided according to an embodiment of the present application; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The embodiments of the present application will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present application, and should not be construed as limiting the present application.

[0019] A communication tower according to an embodiment of the present application will be described below with reference to the drawings.

[0020] Specifically,Figure 1 Schematic diagram of the composition of a communication tower provided by an embodiment of the present application.

[0021] As Figure 1 shown, the communication tower 10 includes: a communication tower body 100, a communication tower monitoring device 200, and a remote monitoring platform 300.

[0022] Among them, as Figure 2 shown, the communication tower body 100 is used to ensure the normal operation of communication, including a tower body structure, a power system, communication equipment, and a safety system. Among them, the tower body structure adopts a bionic honeycomb structure, including a tower body, a foundation, a hexagonal antenna platform, a ladder, etc. The power equipment includes a power supply line, a solar panel, a distribution box, etc. The communication equipment includes an antenna, a feeder, a base station equipment, etc. The safety system includes lightning protection, grounding, and safety protection facilities; the communication tower monitoring device 200 is used to monitor the operation status of the tower in real time, including a sensor component and a communication module. The sensor component is used to obtain the sound, vibration, and environmental noise data during the operation of the tower. The communication module is used to upload the data obtained by the sensor component to the remote monitoring platform, supporting 4G and 5G wireless transmission and fiber optic wired backhaul; the remote monitoring platform 300 is used to centrally manage and intelligently analyze and classify the operation status of the tower. The intelligent analysis and classification include removing environmental noise from the acoustic vibration signal according to the WaveNet noise reduction network, analyzing and classifying the characteristics of the acoustic vibration signal according to the Transformer-CNN hybrid model, judging whether the communication tower is faulty, and timely warning the operation and maintenance personnel after detecting the tower fault.

[0023] It can be understood that by designing a communication tower with a bionic honeycomb structure in the embodiment of the present application, while ensuring the strength, the self-weight can be reduced and the material utilization rate can be improved; an intelligent monitoring system is constructed around acoustic vibration monitoring, which provides a core guarantee for the safe and stable operation of the communication tower. By focusing on the acquisition of acoustic vibration signals through the built-in sensor component, the subtle sound and vibration changes during the operation of the tower can be keenly captured. Potential faults such as loose tower bolts and component wear will cause abnormal vibration and friction sound in the initial stage. The sensor can collect these signals in real time. Compared with the traditional manual inspection that only relies on experience judgment, problems can be discovered earlier and the expansion of faults can be avoided. For example, by analyzing the vibration frequency and sound intensity, early warning can be given in time before the loose bolts cause structural instability, and the hidden danger can be nipped in the bud. The remote monitoring platform deeply mines and analyzes the long-term accumulated acoustic vibration data and fault information, and can summarize the fault occurrence rules and potential risk points under different types of towers and different environmental conditions, providing a strong basis for formulating the maintenance strategy of the communication tower.

[0024] In an embodiment of the present application, the sensor components include: a sound sensor, a low-frequency vibration sensor, and an environmental sensor, wherein the sound sensor is used to capture various sound signals generated during the operation of the communication tower; the low-frequency vibration sensor is used to detect the vibration information of the communication tower structure and discover silent faults; the environmental sensor is used to obtain environmental parameters such as wind speed and rain intensity in real time, and provide environmental reference data for subsequent signal preprocessing.

[0025] Specifically, sound sensors can be installed at the top of the communication tower antenna and feeder connection, at the platforms and connection nodes on each floor of the tower, near the equipment in the machine room, etc. Low-frequency vibration sensors can be installed at the connection between the tower base and the foundation of the communication tower, at the key stress nodes of the tower body, near the lightning rod on the top of the tower and ancillary facilities, etc. Environmental sensors can be installed at the top of the tower, in the middle of the tower body, around the tower base, etc.

[0026] It is understandable that the embodiments of the present application use sound sensors to keenly capture abnormal noises during the operation of the tower, such as the metal friction sound of loose bolts and abnormal vibrations caused by equipment failures; low-frequency vibration sensors focus on the structure itself to detect subtle vibration changes that are difficult to detect with the naked eye, such as the abnormal swing frequency of the tower body caused by wind; and environmental sensors record external factors such as wind speed and rain intensity in real time. The communication tower sensor assembly builds an all-round, multi-level monitoring system through the coordinated work of sound, low-frequency vibration and environmental sensors, providing core guarantees for the safe and stable operation of the tower.

[0027] Next, a communication tower monitoring method proposed according to an embodiment of the present application is described with reference to the accompanying drawings.

[0028] like Figure 2 As shown, the communication tower monitoring method comprises the following steps: In step A1, the sound, vibration and environmental information of the communication tower are obtained.

[0029] It can be understood that the embodiment of the present application obtains the sound, vibration, and environmental information of the communication tower in a regular manner, can determine the operating status of the tower and its ancillary equipment, realizes real-time monitoring of the communication tower, and provides data support for subsequent analysis of the tower's operating status and early warning.

[0030] In step A2, a WaveNet noise reduction network is built and trained to preprocess the acquired information to obtain the sound signal and vibration signal after filtering the environmental noise.

[0031] Specifically, the operating environment of communication towers is complex, and sound and vibration signals are easily interfered by environmental noise. Based on the time domain modeling capability of deep learning, the WaveNet noise reduction network can automatically learn and distinguish the abnormal signal characteristics caused by environmental noise and tower failures. For example, it can separate the background vibration caused by strong wind from the abnormal vibration caused by the loose tower structure, significantly reducing the misjudgment rate.

[0032] It can be understood that the embodiment of the present application pre-processes the communication tower monitoring signal by building and training the WaveNet noise reduction network. By deeply mining the signal characteristics, it can effectively enhance the signal purity, which is conducive to improving the accuracy of fault identification and can effectively avoid missed detection or false alarms caused by noise interference.

[0033] In the embodiment of the present application, a WaveNet noise reduction network is built and trained, and the acquired information is preprocessed to obtain a sound signal and a vibration signal after filtering the environmental noise. The specific steps are as follows: Step B1: Input the acquired sound, vibration, and environmental information into the starting convolutional layer for splicing, and convert the data with different numbers of channels into data with the target number of channels to obtain preliminary data.

[0034] Specifically, in a communication tower monitoring scenario, the sound signal collected by the sound sensor is 2-channel data, corresponding to sound information of different frequency bands; the vibration signal collected by the low-frequency vibration sensor is 1-channel data; the environmental sensor collects environmental data containing temperature, humidity, wind speed and other information, a total of 3 channels. At this time, the total number of input data channels of the starting convolution layer is 2+1+3=6. The target number of channels is set to 32. After processing by the starting convolution layer, the original 6-channel sound, vibration, and environmental information data are converted into 32-channel data. These data contain the fusion features of different sensor data in the same feature space, preparing for further feature extraction in the subsequent expansion convolution layer.

[0035] It is understandable that the embodiment of the present application splices sound, vibration, and environmental information through the starting convolution layer, breaking the boundaries between different types of data and realizing the fusion of multi-source information. The data collected by different sensors reflect the operating conditions of the communication tower from different angles. For example, the sound signal reflects the interaction between the tower components, the vibration signal reflects the stress state of the structure, and the environmental information affects the external conditions of the tower operation. After integration, the model can make comprehensive use of this information to explore potential connections between the data. Converting data with different numbers of channels into data with a target number of channels is conducive to the subsequent expansion of the convolution layer to uniformly process the data, avoiding calculation confusion and information loss caused by inconsistent data dimensions, and is also conducive to reducing complexity and calculation amount, and improving operation efficiency.

[0036] Step B2: Input the preliminary data into the multi-level dilated convolutional layer to learn features. The dilated convolutional layer includes a filtering convolutional layer, a gated convolutional layer, a residual connection convolutional layer, and a skip connection convolutional layer. Among them, features of different scales are extracted through the filtering convolutional layer, and selective retention and suppression are performed on the features of different scales according to the gated convolutional layer. Multiply the results of the filtering convolutional layer and the gated convolutional layer to obtain the first eigenvalue, and input the first eigenvalue into the residual connection convolutional layer and the skip connection convolutional layer for information transmission. Among them, the residual connection convolutional layer adds the input value of the current dilated convolutional layer to the first eigenvalue to obtain the target value, and accumulates the target values of each layer according to the skip connection convolutional layer to obtain the feature information of different layers.

[0037] Specifically, the formula is as follows: The formula for the filtering convolutional layer is , where is the weight of the filtering convolutional layer, x is the input, is the bias, and tanh() is the activation function that maps the input to the interval (−1, 1).

[0038] The formula for the gated convolutional layer is , where is the weight of the gated convolutional layer, x is the input, is the bias, () is the activation function that maps the input to the interval (−1, 1).

[0039] The formula for the residual connection convolutional layer is , where is the weight of the residual connection convolutional layer, is the input of the current dilated convolutional layer, is the updated input, and y is the output of the filtering convolutional layer and the gated convolutional layer, and its calculation formula is .

[0040] The formula for the skip connection convolutional layer is , where is the weight of the residual connection convolutional layer, and S is the sum of the skip connection outputs of all previous layers.

[0041] It can be understood that in the embodiments of the present application, multi-scale feature extraction is performed through a filtering convolutional layer. By different dilation rates, the input data can be sampled and feature-extracted at different time scales. This multi-scale feature extraction method allows the model to comprehensively understand the operating state of the communication tower. Different types of faults may exhibit features at different scales. Through multi-scale feature extraction, the model can better adapt to different fault modes and improve the accuracy of fault detection. By selectively retaining and suppressing the features of different scales extracted by the filtering convolutional layer through a gated convolutional layer, it helps the model remove features that may be noise or interference, only retain the important features related to the operating state of the communication tower, thereby improving the quality and representativeness of the features, and at the same time being able to better resist the influence of noise and improve the performance and robustness in complex environments.

[0042] It can be understood that in the embodiments of the present application, the input information of the current dilated convolutional layer is retained through a residual connection convolutional layer, avoiding the loss of information in the deep network. During the learning process of the model, it can combine the original information and the newly extracted features to better capture the changes and features of the data. The residual connection convolutional layer adds the input value of the current dilated convolutional layer to the first feature value, enabling the gradient to be directly transmitted from the deeper layer to the shallower layer, ensuring the smooth flow of information and effectively alleviating the problem of gradient disappearance. The skip connection convolutional layer accumulates the target values of each layer, enabling the model to integrate the feature information of different layers. By integrating the features of different layers, the model can learn more complex and rich feature representations, thereby improving the ability to understand and judge the operating state of the communication tower.

[0043] Step B3: Use the ReLU activation function for the feature information of different layers, set negative values to zero to enhance the non-linearity of the features, and obtain the final output according to the two end convolutional layers. Among them, the first end convolutional layer transforms and integrates the features, and the second end convolutional layer converts the number of target channels to 2, and outputs the denoised dual-channel data. Among them, the dual-channel data are respectively sound signal data and vibration signal data.

[0044] It can be understood that in the embodiments of the present application, the ReLU activation function is used to enhance the non-linear expression ability. The ReLU activation function is a non-linear function. In a neural network, the expression ability of a linear model is limited, while a non-linear activation function allows the model to learn more complex features and patterns. In the monitoring of communication towers, the features of sound and vibration signals are often non-linear. By using the ReLU activation function, the model can better capture these non-linear features and improve the accuracy of judging the operating state of the communication tower. The first ending convolutional layer transforms and integrates the features after ReLU activation processing, which can further extract and optimize the features. The second ending convolutional layer converts the number of target channels to 2 and outputs the denoised dual-channel data, that is, sound signal data and vibration signal data. This makes the output of the model directly correspond to the two signals of the target, facilitating subsequent analysis and processing. Through the processing of the two ending convolutional layers, the model can better remove the interference of environmental noise on the sound and vibration signals and improve the signal quality.

[0045] In step A3, feature extraction is performed on the sound signal and the vibration signal. The PCA dimensionality reduction algorithm is used to reduce the dimensionality and fuse the features of the acoustic-vibration signals to obtain a comprehensive feature vector.

[0046] Specifically, the steps are as follows: Step C1: Respectively extract the features of the denoised sound signal and vibration signal to construct a feature matrix X.

[0047] Step C2: Perform mean processing on the feature matrix X to obtain a standardized feature matrix Z.

[0048] Step C3: Calculate the covariance matrix C for the standardized feature matrix Z, and perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues sorted by size and the corresponding eigenvectors. Among them, the formula for calculating the covariance is: .

[0049] Step C4: According to the size of the eigenvalues, select the top p largest eigenvalues and their corresponding eigenvectors, and form a projection matrix P with these p eigenvectors.

[0050] Step C5: Project the standardized data matrix Z onto the projection matrix P to obtain a matrix Y = ZP after dimensionality reduction. Each row of the matrix Y is a comprehensive feature vector.

[0051] For example, the acoustic-vibration feature vector obtained by extracting and fusing the features of the sound signal and vibration signal after noise reduction is 20-dimensional. Calculate the covariance matrix of these features and find the direction with the largest variation in the data, which is the principal component. If it is calculated that the first 4 principal components can retain more than 95% of the information of the original data. Then, project the original 20-dimensional acoustic-vibration signal feature vector into the new low-dimensional space formed by these 4 principal components, and the original 20-dimensional feature vector is reduced to a 4-dimensional comprehensive feature vector. This new 4-dimensional vector is the comprehensive feature vector that integrates the key information of the sound and vibration signals.

[0052] It can be understood that the embodiment of the present application reduces the dimension and fuses the acoustic-vibration signal through the PCA dimensionality reduction algorithm, which can reduce the computational complexity. In the original high-dimensional acoustic-vibration signal feature space, subsequent processing such as model training and feature analysis has a huge amount of calculations and will consume a large amount of time and computing resources. After adopting the PCA dimensionality reduction algorithm, the high-dimensional feature vector is converted into a low-dimensional comprehensive feature vector, which can reduce the data dimension and the amount of data to be processed, so that when using the Transformer-CNN hybrid model for analysis and classification subsequently, the calculation speed is significantly improved, the training time is shortened, and it is beneficial to improve the operation efficiency of the entire monitoring system.

[0053] In step A4, input the comprehensive feature vector into the Transformer-CNN hybrid model for analysis and classification to obtain the classification result, where the classification result is the normal acoustic-vibration mode and the fault acoustic-vibration mode.

[0054] Specifically, the steps are as follows: Step D1: Input the comprehensive feature vector into the CNN convolutional layer. The CNN convolutional layer contains multiple convolutional kernels, and captures the key feature information of the data within a local range through convolutional operations to obtain the output data.

[0055] Step D2: In the CNN pooling layer, use the max pooling method to downsample the output data of the convolutional layer, and splice the pooling outputs of all convolutional kernels to obtain the output of the CNN layer 。

[0056] Step D3: Input into the Transformer layer for linear transformation to obtain the embedding vector E. Add the positional encoding PE to the embedding vector, and use the multi-head attention mechanism to capture the information in different subspaces to obtain the token sequence. Based on the feed-forward neural network, perform transformation and processing on the token sequence. The feed-forward neural network contains two linear layers and an activation function. After being processed by the multi-head attention mechanism and the feed-forward neural network, a normalization operation is performed to obtain the output of the Transformer layer 。

[0057] Step D4: Input into the fully connected layer for classification, and use the Softmax function to convert the output into classification probability P.

[0058] Step D5: Determine the final classification result according to the classification probability P. If P is greater than or equal to the target threshold, it is determined that the classification result is a normal acoustic vibration mode; otherwise, it is determined that the classification result is a faulty acoustic vibration mode.

[0059] Specifically, on a sunny and calm day, the communication tower is in a normal operating state. The comprehensive feature vector contains features such as a stable sound frequency distribution and regular vibration amplitude. Input this comprehensive feature vector into the Transformer-CNN hybrid model. The CNN first extracts the local features of the sound and vibration signals, such as stable low-frequency vibration modes and uniform sound spectrum features. The Transformer analyzes these features from a global perspective and finds that they conform to the acoustic vibration mode law under normal operating conditions. Finally, the model outputs a classification result of normal acoustic vibration mode, indicating that the tower is operating normally and no additional maintenance and inspections are required.

[0060] Specifically, when a certain bolt of the communication tower is loose, the comprehensive feature vector will show features such as high-frequency abnormal sound signals and irregular vibration amplitude fluctuations. Input this comprehensive feature vector into the hybrid model. The CNN can capture these local abnormal features, such as the sudden appearance of high-frequency noise and the mutation of vibration amplitude. The Transformer analyzes the correlation between these abnormal features from a global level and determines that this acoustic vibration mode does not conform to the normal situation. Finally, the model outputs a classification result of faulty acoustic vibration mode, reminding the operation and maintenance personnel to check and repair the tower in time.

[0061] It can be understood that in the embodiment of the present application, the CNN extracts local patterns and features in the comprehensive feature vector. The self-attention mechanism of the Transformer can handle the dependencies between any positions in the sequence and can capture the global information and long-distance dependencies in the comprehensive feature vector. Combining the two enables the model to not only pay attention to the local details of the acoustic vibration signal but also grasp its overall characteristics, thereby improving the accuracy of classification.

[0062] In step A5, if the classification result is a faulty acoustic vibration mode, then compare it with the acoustic vibration feature library, identify the faulty acoustic vibration features through pattern matching and similarity calculation methods, determine the fault type and degree of the fault, remind the operation and maintenance personnel to repair and recommend a repair plan for the operation and maintenance personnel based on historical repair data and fault types, perform repairs according to the recommended repair plan, and update the newly occurred faulty acoustic vibration features to the acoustic vibration feature library; if the classification result is a normal acoustic vibration mode, continue to run.

[0063] Specifically, the steps are as follows: Step E1: Calculate the similarity between the fault acoustic vibration feature vector to be identified and the feature vector in the acoustic vibration feature library according to the cosine similarity algorithm, wherein the calculation formula is: , in, is the acoustic vibration feature vector of the fault to be identified, is the feature vector in the feature library.

[0064] For example, when a communication tower fails, data is collected through sound and vibration sensors. After feature extraction and PCA dimension reduction, the fault sound and vibration feature vector D = [0.15, 0.75, 0.25, 0.08] to be identified is obtained. The cosine similarity algorithm is used to calculate the similarity between vector D and the existing sound and vibration feature vectors in the current sound and vibration feature library that are the same as the current communication.

[0065] It can be understood that the embodiment of the present application uses the cosine similarity algorithm to calculate the similarity between feature vectors, which is relatively simple to implement and has low computational complexity. In the communication tower monitoring scenario, it is necessary to quickly process a large number of real-time fault acoustic vibration feature vectors. The algorithm can complete the similarity calculation in a short time, buy time for fault diagnosis, and promptly notify the operation and maintenance personnel to handle the fault.

[0066] Step E2: The one with the highest similarity is used as the target feature vector, and the similarity value is compared with the target threshold. If the similarity value is higher than the target threshold, the corresponding fault type and degree are used as the preliminary judgment of the current fault. If the similarity value is lower than the target threshold, the current fault is marked as a new unknown fault type.

[0067] For example, if a communication tower is judged to have a fault, there are 4 data in the acoustic vibration feature library, namely fault 1 [bracket deformation, mild, feature vector A, historical maintenance data], fault 2 [bolt loosening, mild, feature vector B, historical maintenance data], fault 3 [antenna imbalance, moderate, feature vector C, historical maintenance data], fault 4 [feeder contact is poor, moderate, feature vector D, historical maintenance data], and the similarity between the acoustic vibration feature vector F of the current communication tower and the feature vectors A, B, C, and D is calculated in turn, and the similarities are 0.78, 0.82, 0.65, and 0.73, respectively. The highest similarity 0.82 is compared with the target threshold. Because the highest similarity 0.82 is greater than the target threshold, the preliminary judgment of the current fault is that the fault type is loose bolts and the degree label is mild.

[0068] It can be understood that by using the feature vector with the highest similarity as the preliminary judgment basis in the embodiments of the present application, the most matching fault type and degree can be quickly found from the acoustic-vibration feature library, avoiding the cumbersome process of checking all possible faults one by one, and greatly shortening the fault diagnosis time. Setting a target threshold for judgment can avoid misjudgment caused by similarity calculation errors or data fluctuations. When the similarity value is higher than the target threshold, it indicates that the fault to be identified has a high similarity with the known fault type. At this time, using the corresponding fault type and degree as the preliminary judgment has a high credibility. When the similarity value is lower than the target threshold, it is not easily classified as a known fault type, but marked as an unknown fault, which helps to discover new fault modes, avoid misdiagnosis, and improve the overall accuracy of fault diagnosis.

[0069] Step E3: If the preliminarily determined fault type is a known fault, then according to the fault type and degree, search for the historical maintenance data of similar faults from the acoustic-vibration feature library to obtain the historical maintenance plan, generate a recommended maintenance plan based on the historical maintenance plan, notify the corresponding operation and maintenance personnel to perform maintenance and recommend the recommended maintenance plan to the operation and maintenance personnel, perform maintenance according to the recommended maintenance plan, and add the newly occurred fault acoustic-vibration features, fault type, fault degree, and the detailed information of this maintenance to the acoustic-vibration feature library; if the preliminarily determined fault type is an unknown fault, directly notify the corresponding operation and maintenance personnel to repair the unknown fault, and according to the maintenance feedback situation, add the newly occurred fault acoustic-vibration features, fault type, fault degree, and the detailed information of this maintenance to the acoustic-vibration feature library.

[0070] Specifically, a communication tower has a fault. After a series of previous analyses, the fault type is preliminarily determined to be "bolt loose, mild", which is a known fault type in the acoustic-vibration feature library. The system searches for the historical maintenance data of similar faults from the acoustic-vibration feature library according to the fault type and degree. It is found that there have been many previous similar "bolt loose, mild" faults, and the historical maintenance plans are mostly to use a wrench to tighten the loose bolts and check and pre-tighten the surrounding bolts. The system generates a recommended maintenance plan based on these historical maintenance plans: first, use a torque wrench to tighten the loose bolts according to the specified torque, then check each bolt at the surrounding relevant positions one by one, and pre-tighten the bolts with insufficient torque. Then the system notifies the operation and maintenance personnel responsible for this area to go for maintenance and pushes the recommended maintenance plan to the operation and maintenance personnel. After the operation and maintenance personnel arrive at the scene, they perform maintenance according to the recommended maintenance plan. After the maintenance is completed, add the newly occurred fault acoustic-vibration features (the sound and vibration features collected during this bolt loose fault), fault type (bolt loose), fault degree (mild), and the detailed information of this maintenance (maintenance time, maintenance personnel, maintenance method, parts used) to the acoustic-vibration feature library.

[0071] It can be understood that in the embodiments of the present application, by searching for historical maintenance data of similar faults in the acoustic-vibration feature library and generating recommended maintenance solutions, maintenance personnel can quickly understand the maintenance steps and methods, avoiding the time for re-exploring and formulating maintenance plans, enabling rapid commencement of maintenance work, reducing the fault repair time, and improving the maintenance efficiency. The recommended maintenance solutions are summarized based on historical maintenance experience and have been tested through multiple practices, with high reliability and effectiveness, which is conducive to improving the maintenance quality. Adding the detailed information of each maintenance to the acoustic-vibration feature library and continuously enriching and improving the fault knowledge base contribute to the continuous optimization of the system.

[0072] In this embodiment, it is specifically necessary to explain the acoustic-vibration feature database, which specifically includes the structure of acoustic-vibration feature data, new fault features, and redundant data. Among them, The structure of the acoustic-vibration feature data includes fault type, acoustic-vibration feature vector, fault severity label, and historical maintenance records. Among them, the historical maintenance records include maintenance time, maintenance personnel, maintenance methods, and parts used.

[0073] Add new fault features. When a new fault is detected, obtain the acoustic-vibration feature vector, and according to the maintenance feedback, add the fault type, acoustic-vibration feature vector, fault severity label of the newly occurred fault, and the detailed information of this maintenance to the acoustic-vibration feature library.

[0074] Clean up redundant data. Among them, regularly clean up the acoustic-vibration feature database. For two acoustic-vibration feature vectors with high similarity and highly similar fault types, fault severity labels, and maintenance methods, merge the maintenance records and delete one of the records.

[0075] It can be understood that in the embodiments of the present application, adding new fault features can enhance the fault diagnosis ability, effectively adapt to equipment changes and environmental differences. With the continuous accumulation of new fault features, the diagnostic accuracy and comprehensiveness of the system for various complex faults will gradually improve, enabling more timely and accurate detection of problems existing in communication towers and reducing the situations of fault omission and false alarm. Cleaning up redundant data can reduce the data volume, improve the speed of data retrieval and processing, enable the system to respond more quickly to user requests and analysis tasks, and provide more timely support for fault diagnosis and decision-making.

[0076] A communication tower monitoring method is provided according to an embodiment of the present application. In this method, the operating condition of the communication tower is detected by using sound features and vibration features, which can reflect the operating state of the tower in real time, capture the minute changes in the tower state in a timely manner, and detect problems at the initial stage of a fault, so as to take timely and effective measures to avoid the further development and expansion of the fault. The WaveNet noise reduction network is used to process the signals collected by the sound sensor and the low-frequency vibration sensor, which can effectively filter out environmental noise interference and make the collected sound and vibration signals closer to the real operating state. The PCA dimensionality reduction algorithm is used to reduce the dimensionality and fuse the characteristic parameters of the sound and vibration signals, which can reduce the amount of calculation, improve the processing efficiency, retain key information, enhance the representativeness and distinguishability of the features, and lay a foundation for fault mode recognition. The Transformer-CNN hybrid model is used to analyze and classify the comprehensive feature vector, which can capture the local subtle changes of the acoustic vibration signal while grasping the overall change trend, and can more accurately identify the normal acoustic vibration mode and the fault acoustic vibration mode, improving the fault recognition accuracy. Thus, the problems in the prior art such as the difficulty in detecting early faults of communication towers and the difficulty in extracting characteristic signals due to environmental interference are solved.

[0077] A communication tower monitoring method will be specifically described through a specific embodiment as follows: In the southwestern mountainous area, a certain communication tower stands among the rolling mountains, surrounded by dense vegetation and complex terrain. Strong winds, heavy rains and other adverse weather often occur here, and geological disasters such as landslides and debris flows occur from time to time, posing a great threat to the stable operation of the communication tower.

[0078] A sound sensor, a vibration sensor and an environment sensor are installed on the communication tower, and information is collected every 5 minutes and transmitted to the remote monitoring platform. Among them, the sound sensor uses a high-sensitivity microphone array that can detect sounds in the frequency range of 20 - 20000 Hz, and is installed at the top and middle of the tower to collect the sounds emitted from various parts of the tower; the low-frequency vibration sensor selects an acceleration sensor with a measurement range of ±5g and a frequency response of 0.1 - 100 Hz, and is arranged at the tower legs, the middle of the tower body and the antenna support of the tower respectively; the environment sensor integrates an anemometer (measurement range 0 - 60 m / s, accuracy ±0.3 m / s), a rainfall sensor (resolution 0.2 mm), and a temperature and humidity sensor (temperature measurement range -40°C - 80°C, humidity measurement range 0 - 100%RH), and is installed at an open position near the tower.

[0079] According to the WaveNet noise reduction network, the real-time collected sound data and vibration data are processed for noise reduction to remove environmental noise interferences such as wind and rain sounds and wildlife activities, obtaining clear sound signals and vibration signals. Extract 10 characteristic parameters such as short-time energy, zero-crossing rate, and Mel-frequency cepstral coefficients from the sound signal; extract 8 characteristic parameters such as peak value, mean value, root mean square value, and frequency components from the vibration signal. Combine the extracted sound and vibration characteristic parameters into an 18-dimensional feature vector, and through the PCA dimensionality reduction algorithm, set to retain 90% of the information, and reduce its dimension to an 8-dimensional comprehensive feature vector.

[0080] Input the comprehensive feature vector into the Transformer-CNN hybrid model to analyze and classify the acoustic-vibration signals. After the CNN extracts local features and the Transformer analyzes global features and dependencies, it is determined that the acoustic-vibration feature vector is a fault acoustic-vibration mode. Among them, the CNN is set with 3 convolutional layers, the convolutional kernel sizes are 3x3, 3x3, and 5x5 respectively, the number of channels is 32, 64, and 128 in sequence, and the pooling layer uses the maximum pooling method. The Transformer is set with 6 attention heads, the hidden layer dimension is 256, and the number of layers is 4 layers.

[0081] Compare with the acoustic-vibration feature library and calculate the similarity using the cosine similarity algorithm. Assume that there are 10 known fault types in the feature library. It is calculated that the similarity with the fault feature vector of "loose tower leg foundation" is the highest and higher than the set threshold of 0.8, then the fault type is determined to be loose tower leg foundation, and the fault degree is the corresponding mild degree. The system recommends a maintenance plan to reinforce the tower leg foundation and use concrete to pour and reinforce the foundation. After the maintenance personnel complete the maintenance, update the fault acoustic-vibration features (the sound and vibration features collected for this loose tower leg foundation fault), fault type (loose tower leg foundation), fault degree (mild), and the detailed information of this maintenance (maintenance time, maintenance personnel, maintenance method, parts used) to the acoustic-vibration feature library.

[0082] In summary, in the mountain environment scenario, this communication tower monitoring method can comprehensively capture the sound, vibration, and environmental information of the tower under the threat of bad weather and geological disasters by accurately deploying sensors with high sensitivity and suitable for complex terrain; the WaveNet noise reduction network is trained with a large amount of data to effectively filter interferences such as wind and rain, ensuring the authenticity and reliability of the data; the PCA dimensionality reduction algorithm and the Transformer-CNN hybrid model cooperate to efficiently extract fusion features and accurately identify fault modes; combined with the fault diagnosis and maintenance recommendation of the acoustic-vibration feature library, it realizes rapid response and scientific maintenance, and can also continuously optimize the system through data update, comprehensively ensuring the stable operation of the communication tower in the complex mountain environment, reducing the fault risk and operation and maintenance costs.

[0083] Next, a communication tower monitoring system according to an embodiment of the present application will be described with reference to the accompanying drawings.

[0084] Specifically, Figure 3 is a schematic diagram of the composition of a communication tower monitoring system provided by an embodiment of the present application.

[0085] As Figure 3 shown, the communication tower monitoring system 20 includes: an acquisition module 210, a noise reduction module 220, a dimensionality reduction and fusion module 230, an analysis and classification module 240, and a maintenance module 250.

[0086] Among them, the acquisition module 210 is used to acquire the sound, vibration, and environmental information of the communication tower; the noise reduction module 220 is used to preprocess the acquired information according to the WaveNet noise reduction network to obtain the sound signal and vibration signal after filtering the environmental noise; the dimensionality reduction and fusion module 230 is used to extract the features of the sound signal and vibration signal, and use the PCA dimensionality reduction algorithm to reduce the dimensionality and fuse the acoustic and vibration signal features to obtain a comprehensive feature vector; the analysis and classification module 240 is used to input the comprehensive feature vector into the Transformer-CNN hybrid model for analysis and classification to obtain a classification result, where the classification result is a normal acoustic and vibration mode and a fault acoustic and vibration mode; the maintenance module 250 is used to repair the communication tower determined to be in the fault acoustic and vibration mode. If the classification result is the fault acoustic and vibration mode, the acoustic and vibration feature library is compared, and the fault acoustic and vibration feature is identified by the pattern matching and similarity calculation method to determine the fault type and degree, remind the operation and maintenance personnel to repair and recommend a repair plan for the operation and maintenance personnel based on the historical maintenance data and fault type, repair according to the recommended repair plan, and update the newly occurred fault acoustic and vibration feature to the acoustic and vibration feature library.

[0087] Next, a communication tower monitoring system will be specifically described through a specific embodiment as follows: A certain communication tower is located in the northeast region, where the winter is long and the temperature is extremely low, and there are often weather conditions such as heavy snow and freezing.

[0088] Acquisition Module: Install sound sensors, vibration sensors, and environmental sensors on the communication tower, collect information every 5 minutes, and transmit the data to the remote monitoring platform. Among them, the sound sensor uses a model with good low-temperature performance, the working temperature range is -50°C - 50°C, the frequency response is 25 - 18000 Hz, and it is installed near the key connection parts of the tower; the low-frequency vibration sensor selects products suitable for low temperatures, with a measuring range of ±8g and a frequency response of 0.2 - 150 Hz, distributed at the bottom, middle, and top of the tower; the environmental sensor includes a temperature and humidity sensor (temperature measurement range -60°C - 50°C, humidity measurement range 0 - 100%RH), a snow depth sensor (measurement range 0 - 5m, accuracy ±1cm), and an ice thickness sensor (measurement range 0 - 10cm, accuracy ±0.1cm), which are installed around the tower.

[0089] Noise Reduction Module: Perform noise reduction processing on the real-time collected sound data and vibration data according to the WaveNet noise reduction network to remove environmental noise interference and obtain clear sound signals and vibration signals. Among them, the number of input channels of the initial convolutional layer in the noise reduction network is 8, including 2 for sound, 3 for vibration, and 3 for environment, and the number of output channels is 56; there are 10 dilation convolutional layers, and the dilation rates are 1, 2, 4, 8, 16, 8, 4, 2, 1, 2 in sequence; the convolutional kernels of the filtering convolutional layer and the gated convolutional layer are 3; the residual connection convolutional layer and the skip connection convolutional layer both use 1x1 convolutional kernels.

[0090] Dimensionality Reduction and Fusion Module: Extract 10 characteristic parameters such as short-time energy, zero-crossing rate, and Mel frequency cepstral coefficients from the sound signal; extract 8 characteristic parameters such as peak value, mean value, root mean square value, and frequency components from the vibration signal. Combine the extracted sound and vibration characteristic parameters into an 18-dimensional feature vector, and through the PCA dimensionality reduction algorithm, reduce it to an 8-dimensional comprehensive feature vector.

[0091] Analysis and Classification Module: Input the comprehensive feature vector into the Transformer-CNN hybrid model to analyze and classify the acoustic-vibration signals. CNN quickly extracts the local features of the sound and vibration signals, such as the local vibration abnormality of a certain component of the tower caused by wind load, and Transformer analyzes the features as a whole to judge the change trend of the tower vibration over a period of time. Through the analysis and classification of the model, it is determined that the acoustic-vibration mode of the tower at this time belongs to the fault acoustic-vibration mode.

[0092] Maintenance module: Compare with the acoustic-vibration feature library, calculate the similarity using the cosine similarity algorithm. If the calculated similarity with the fault feature vector of "low-temperature brittle fracture of tower components" is the highest and higher than the set threshold of 0.8, then determine that the fault type is low-temperature brittle fracture of tower components and the fault degree is the corresponding medium degree. The system recommends the maintenance plan of replacing the damaged components and conducting a comprehensive inspection and maintenance of the tower. After the maintenance personnel complete the maintenance, update the fault acoustic-vibration features (the sound and vibration features collected during this low-temperature brittle fracture fault of tower components), fault type (low-temperature brittle fracture of tower components), fault degree (medium degree), and the detailed information of this maintenance (maintenance time, maintenance personnel, maintenance method, and parts used) to the acoustic-vibration feature library.

[0093] In summary, relying on the targeted low-temperature adaptation sensors, this communication tower monitoring method can accurately collect the sound, vibration, and environmental data of the tower under extreme climates such as heavy snow and freezing, and promptly capture the performance changes of the tower affected by low temperature. The WaveNet noise reduction network can effectively remove the interference of wind and snow noise, providing a reliable data basis for subsequent analysis. The PCA dimensionality reduction algorithm and the Transformer-CNN hybrid model are used to deeply mine data features, accurately identifying fault modes such as component brittle fracture and excessive ice and snow load caused by low temperature. Combining with the acoustic-vibration feature library for fault diagnosis and recommending maintenance plans can quickly respond to faults, and the update of new fault data enables the system to continuously adapt to new fault situations in the severe cold environment, greatly improving the stability and reliability of the communication tower operating in the severe cold conditions and reducing the operation and maintenance difficulty and cost.

[0094] In the description of this specification, the descriptions referring to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or N embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0095] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "N" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0096] Any process or method description, whether in a flowchart or otherwise described herein, can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a customized logical function or process. The scope of the preferred embodiments of the present application includes additional implementations, where functions may be performed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0097] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), and the like.

[0098] Those of ordinary skill in the art can understand that all or part of the steps carried out in implementing the above-described method embodiments can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiments.

Claims

1. A communication tower, characterized in that, Including: A communication tower body, a communication tower monitoring device, and a remote monitoring platform. Among them, The communication tower body is used to ensure the normal operation of communication, including a tower body structure, a power system, communication equipment, and a safety system. Among them, the tower body structure adopts a bionic honeycomb structure, including a tower body, a foundation, a hexagonal antenna platform, a ladder, etc. The power equipment includes a power supply line, a solar panel, a distribution box, etc. The communication equipment includes an antenna, a feeder, a base station device, etc. The safety system includes lightning protection, grounding, and safety protection facilities; The communication tower monitoring device is used to monitor the operation status of the tower in real time, including a sensor component and a communication module. The sensor component is used to obtain the sound, vibration, and environmental noise data during the operation of the tower. The communication module is used to upload the data obtained by the sensor component to the remote monitoring platform, supporting 4G and 5G wireless transmission and fiber optic wired backhaul; The remote monitoring platform is used to centrally manage and intelligently analyze and classify the operation status of the tower. The intelligent analysis and classification include removing environmental noise from the acoustic vibration signal according to the WaveNet noise reduction network, analyzing and classifying the characteristics of the acoustic vibration signal according to the Transformer-CNN hybrid model, judging whether the communication tower is faulty, and timely warning the operation and maintenance personnel after detecting the tower fault.

2. The communication tower according to claim 1, wherein The sensor component includes: a sound sensor, a low-frequency vibration sensor, and an environmental sensor. Among them, the sound sensor is used to capture various sound signals generated during the operation of the communication tower; the low-frequency vibration sensor is used to detect the vibration information of the communication tower structure and discover silent faults; the environmental sensor is used to obtain environmental parameters such as wind speed and rain intensity in real time to provide environmental reference data for subsequent signal preprocessing.

3. A communication tower monitoring method, characterized in that, Applied to the communication tower as described in claims 1-2, the method includes the following steps: Step A1, obtain the sound, vibration, and environmental information of the communication tower; Step A2, build and train a WaveNet noise reduction network, preprocess the obtained information to obtain a sound signal and a vibration signal after filtering environmental noise; Step A3, extract the features of the sound signal and the vibration signal, adopt the PCA dimensionality reduction algorithm, reduce the dimensionality and fuse the features of the acoustic vibration signal to obtain a comprehensive feature vector; Step A4, input the comprehensive feature vector into the Transformer-CNN hybrid model for analysis and classification to obtain a classification result. Among them, the classification result is a normal acoustic vibration mode and a faulty acoustic vibration mode; Step A5, if the classification result is a faulty acoustic vibration mode, then compare the acoustic vibration feature library, identify the faulty acoustic vibration features through pattern matching and similarity calculation methods, determine the fault type and degree of the fault, remind the operation and maintenance personnel to repair and recommend a repair plan for the operation and maintenance personnel based on historical repair data and fault types, repair according to the recommended repair plan, and update the newly occurred faulty acoustic vibration features to the acoustic vibration feature library; if the classification result is a normal acoustic vibration mode, then continue to run.

4. A communication tower monitoring method according to claim 3, characterized in that, Build and train a WaveNet noise reduction network to preprocess the acquired information and obtain sound signals and vibration signals after filtering out environmental noise. The specific steps are as follows: Step B1: Input the acquired sound, vibration, and environmental information into the starting convolutional layer for splicing, and convert data with different numbers of channels into data with the target number of channels to obtain preliminary data; Step B2: Input the preliminary data into a multi-level dilated convolutional layer to learn features. The dilated convolutional layer includes a filtering convolutional layer, a gated convolutional layer, a residual connection convolutional layer, and a skip connection convolutional layer. Among them, features of different scales are extracted through the filtering convolutional layer, and selective retention and suppression are performed on features of different scales according to the gated convolutional layer. Multiply the results of the filtering convolutional layer and the gated convolutional layer to obtain a first eigenvalue, and input the first eigenvalue into the residual connection convolutional layer and the skip connection convolutional layer for information transmission. Among them, the residual connection convolutional layer adds the input value of the current dilated convolutional layer to the first eigenvalue to obtain a target value, and accumulates the target values of each layer according to the skip connection convolutional layer to obtain feature information of different layers; Step B3: Use the ReLU activation function for the feature information of different layers, set negative values to zero to enhance the nonlinearity of the features, and obtain the final output according to two ending convolutional layers. The first ending convolutional layer transforms and integrates the features, and the second ending convolutional layer converts the target number of channels to 2 and outputs dual-channel data after noise reduction. The dual-channel data are sound signal data and vibration signal data respectively.

5. A communication tower monitoring method according to claim 4, characterized in that, Input the preliminary data into a multi-level dilated convolutional layer to learn features. The specific formula is as follows: The formula of the filtering convolution layer is , where is the weight of the filtering convolution layer, x is the input, is the bias, and tanh() is the activation function that maps the input to the interval (−1, 1); The formula for the gated convolutional layer is , where is the weight of the gated convolutional layer, x is the input, is the bias, () is the activation function that maps the input to the interval (−1, 1); The formula of the residual connection convolutional layer is , where are the weights of the residual connection convolutional layer, is the input of the current dilated convolutional layer, is the updated input, and y is the output of the filtering convolutional layer and the gated convolutional layer, and its calculation formula is ; The formula of the skip connection convolutional layer is , where are the weights of the residual connection convolutional layer, and S is the sum of the skip connection outputs of all previous layers.

6. The communication tower monitoring method according to claim 3, characterized in that, Reduce the dimension and fuse the acoustic-vibration signal features. The specific steps are as follows: Step C1: Extract the features of the sound signal and vibration signal after noise reduction respectively to construct a feature matrix X; Step C2: Perform mean processing on the feature matrix X to obtain a standardized feature matrix Z; Step C3: Calculate the covariance matrix C for the standardized feature matrix Z, perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues sorted by size and corresponding eigenvectors. The formula for calculating the covariance is: ， Step C4: Select the first p largest eigenvalues and their corresponding eigenvectors according to the size of the eigenvalues, and form a projection matrix P with these p eigenvectors; Step C5: Project the standardized data matrix Z onto the projection matrix P to obtain a matrix Y = ZP after dimensionality reduction. Each row of the matrix Y is a comprehensive feature vector.

7. A communication tower monitoring method according to claim 3, characterized in that, Input the comprehensive feature vector into a Transformer-CNN hybrid model for analysis and classification. The specific steps are as follows: Step D1: Input the comprehensive feature vector into the CNN convolutional layer. The CNN convolutional layer contains multiple convolutional kernels, and captures key feature information of the data within a local range through convolutional operations to obtain output data; Step D2: In the CNN pooling layer, use the max pooling method to downsample the output data of the convolutional layer, and splice the pooling outputs of all convolutional kernels to obtain the output of the CNN layer ; Step D3: Input into the Transformer layer for linear transformation to obtain the embedding vector E. Add the positional encoding PE to the embedding vector, and use the multi-head attention mechanism to capture information in different subspaces to obtain the token sequence. Perform transformation and processing on the token sequence based on the feed-forward neural network, which includes two linear layers and an activation function. After being processed by the multi-head attention mechanism and the feed-forward neural network, perform a normalization operation to obtain the output of the Transformer layer ; Step D4: Input into the fully connected layer for classification, and use the Softmax function to convert the output into the classification probability P; Step D5: Determine the final classification result according to the classification probability P. If P is greater than or equal to the target threshold, the classification result is determined to be a normal acoustic-vibration mode; otherwise, the classification result is determined to be a faulty acoustic-vibration mode.

8. A communication tower monitoring method according to claim 3, characterized in that, The fault acoustic-vibration feature recognition by the pattern matching and similarity calculation method is as follows: Step E1: Calculate the similarity between the fault acoustic-vibration feature vector to be recognized and the feature vectors in the acoustic-vibration feature library according to the cosine similarity algorithm, where the calculation formula is: ， Among them, is the acoustic and vibration feature vector of the fault to be recognized, is the feature vector in the feature library; Step E2: Take the feature vector with the highest similarity as the target feature vector, compare the similarity value with the target threshold. If the similarity value is higher than the target threshold, take the corresponding fault type and degree as the preliminary judgment of the current fault. If the similarity value is lower than the target threshold, mark the current fault as a new unknown fault type; Step E3: If the preliminarily determined fault type is a known fault, according to the fault type and degree, search the historical maintenance data of similar faults in the acoustic-vibration feature library to obtain the historical maintenance plan, generate a recommended maintenance plan according to the historical maintenance plan, notify the corresponding operation and maintenance personnel to repair and recommend the recommended maintenance plan to the operation and maintenance personnel, perform maintenance according to the recommended maintenance plan, and add the newly occurred fault acoustic-vibration features, fault types, fault degrees and the detailed information of this maintenance to the acoustic-vibration feature library; If the preliminarily determined fault type is an unknown fault, directly notify the corresponding operation and maintenance personnel to repair the unknown fault, and add the newly occurred fault acoustic-vibration features, fault types, fault degrees and the detailed information of this maintenance to the acoustic-vibration feature library according to the maintenance feedback.

9. The method for monitoring a communication tower according to claim 3, wherein The acoustic-vibration feature database includes the structure of acoustic-vibration feature data, new fault features and redundant data, where The structure of the acoustic-vibration feature data includes the fault type, acoustic-vibration feature vector, fault degree label, and historical maintenance record, where the historical maintenance record includes the maintenance time, maintenance personnel, maintenance method, and parts used; Add the new fault features. When a new fault is detected, obtain the acoustic-vibration feature vector, and add the fault type, acoustic-vibration feature vector, fault degree label of the newly occurred fault and the detailed information of this maintenance to the acoustic-vibration feature library according to the maintenance feedback; Clean the redundant data, where the acoustic-vibration feature database is regularly cleaned. For two acoustic-vibration feature vectors with high similarity and whose fault types, fault degree labels, and maintenance methods are highly similar, merge the maintenance records and delete one of the records.

10. A communication tower monitoring system, characterized in that, It includes: An acquisition module for acquiring the sound, vibration, and environmental information of the communication tower; A noise reduction module for preprocessing the acquired information according to the WaveNet noise reduction network to obtain the sound signal and vibration signal after filtering the environmental noise; A dimensionality reduction and fusion module for extracting features from the sound signal and vibration signal, using the PCA dimensionality reduction algorithm to reduce the dimensionality and fuse the acoustic-vibration signal features to obtain a comprehensive feature vector; An analysis and classification module for inputting the comprehensive feature vector into the Transformer-CNN hybrid model for analysis and classification to obtain the classification result, where the classification result is the normal acoustic-vibration mode and the fault acoustic-vibration mode; Maintenance module, which is used to repair the communication tower determined to be in the fault acoustic vibration mode. If the classification result is the fault acoustic vibration mode, then compare with the acoustic vibration feature library, identify the fault acoustic vibration features through the pattern matching and similarity calculation methods, determine the fault type and degree of fault, remind the operation and maintenance personnel to repair and recommend a repair plan for the operation and maintenance personnel based on the historical maintenance data and fault type, perform the repair according to the recommended repair plan, and update the newly occurred fault acoustic vibration features to the acoustic vibration feature library.

Images