Method and apparatus for maintaining a relay communication device of a mobile communication system

Abstract

The application belongs to the cross field of mobile communication technology and artificial intelligence predictive maintenance, and discloses a maintenance method and device of a relay communication equipment of a mobile communication system, the device comprises a cross-domain data sensing module, which is used for synchronously collecting physical state data and communication service quality data; a context self-adaptive analysis module, which is used for generating an environmental interference context factor based on the communication service quality data, and dynamically adjusting the analysis algorithm parameters of the physical state data by using the factor; a fusion prediction and decision module, which is used for combining the context self-adaptive physical state characteristics and the communication service quality data to perform fault prediction; and a closed-loop self-optimization module, which is used for quantifying the feedback of the execution effect of the maintenance task to automatically optimize the prediction model. The application helps to significantly improve the signal-to-noise ratio and accuracy of fault prediction under complex environments, realizes the evolution from passive prediction to intelligent self-adaptation, and guarantees the high reliability of the communication system under extreme working conditions.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of mobile communication technology and artificial intelligence predictive maintenance, specifically relating to a mobile communication system relay communication device and its communication method that can achieve high-precision fault prediction in a dynamic environment with strong interference. Background Technology

[0002] With the rapid deployment of next-generation mobile communication technologies such as 5G and 6G, communication networks are becoming increasingly complex. As a key node to ensure continuous signal coverage, the stable operation of relay communication equipment is of paramount importance. Traditional equipment maintenance mainly relies on regular inspections and post-event repairs, which have significant drawbacks such as delayed response, waste of resources, and inability to prevent sudden failures.

[0003] To address this, the industry has introduced predictive maintenance (PdM) technology based on sensors and machine learning. PdM technology monitors the physical state parameters of equipment, such as temperature, vibration, and voltage, in real time and uses algorithm models to predict the future failure risks of the equipment, thereby enabling early maintenance and significantly improving the reliability of the system.

[0004] However, existing predictive maintenance solutions have revealed serious inherent defects when applied to certain special and critical scenarios. In particular, in complex working conditions with strong periodicity and high instantaneous noise interference, such as subway tunnels, high-speed rail lines, and large factories, relay communication equipment is subject to severe vibrations and strong electromagnetic interference caused by factors such as passing trains and the start and stop of large equipment.

[0005] Existing PdM models are unable to effectively distinguish between normal extreme data caused by the external environment and abnormal fault precursor signals caused by wear and aging of the equipment itself.

[0006] This technical dilemma leads to two extreme problems: on the one hand, in order to capture all possible anomalies, the model will misjudge a large number of environmental disturbances as faults, resulting in a high false alarm rate, which greatly wastes maintenance resources and reduces the trust of maintenance personnel in the system; on the other hand, if preprocessing methods such as data smoothing and noise reduction are used to suppress environmental disturbances, it is easy to filter out the key feature information that represents early minor faults, resulting in a surge in the false alarm rate and losing the fundamental meaning of predictive maintenance.

[0007] The root cause lies in the fact that existing technologies separate the monitoring of the physical state of equipment from the actual communication services it undertakes, lacking a mechanism that allows predictive models to understand the real working environment in which the equipment is currently located, thereby intelligently and adaptively adjusting their analysis and judgment strategies. Summary of the Invention

[0008] The purpose of this invention is to overcome the above-mentioned defects of the prior art and provide a relay communication device and communication method for a mobile communication system, so as to solve the problems of high false alarm rate and high missed alarm rate of existing predictive maintenance technology under strong interference and high dynamic conditions.

[0009] To achieve the above objectives, the present invention provides the following technical solution: like Figure 1 As shown, the maintenance method for a relay communication device in a mobile communication system according to the present invention specifically includes the following steps: Step S1, perform cross-domain data acquisition: synchronously acquire the time series of physical status data and communication service quality data of the device; Step S2, Execute context generation: Based on the communication service quality data, generate an environmental interference context factor in real time to indicate the current working environment interference level; Step S3, perform adaptive feature extraction: dynamically adjust the key parameters of a feature extraction algorithm according to the environmental interference context factor, and use the dynamically adjusted feature extraction algorithm to process the physical state data to extract context-adaptive physical state features. Step S4, perform fusion prediction: input the context-adaptive physical state features and the communication service quality data into a preset fault prediction model to generate equipment fault prediction results; Step S5, Decision generation: Based on the fault prediction results, generate a maintenance task plan.

[0010] Preferably, the method further includes: step S6, performing closed-loop self-optimization: after the maintenance task plan is executed, the execution effect of the maintenance task is quantitatively evaluated to obtain a maintenance effect evaluation score; when the score is lower than a preset threshold, the parameters of the fault prediction model are automatically fine-tuned and optimized based on the fault prediction results and status data corresponding to the maintenance task.

[0011] Preferably, in step S6, when the score is lower than a preset threshold, the parameters of the fault prediction model are automatically fine-tuned and optimized based on the fault prediction results and status data corresponding to the maintenance task. This further includes: when the score is lower than the preset threshold, the maintenance task is determined to be a false alarm maintenance, and the physical status features and communication service quality data that caused the false alarm prediction are marked as negative samples and used to reduce the sensitivity of the fault prediction model to such feature combinations.

[0012] Preferably, the communication service quality data includes at least one or a combination of several of the following: bit error rate, signal-to-noise ratio, and data retransmission rate; the physical status data includes at least one or a combination of several of the following: equipment operating temperature, internal component vibration amplitude, and circuit voltage and current.

[0013] Preferably, the feature extraction algorithm is a dynamic time warping algorithm, and the key parameter for dynamic adjustment is the warping path window size of the dynamic time warping algorithm.

[0014] Preferably, step S3, the adaptive feature extraction step, further includes: when the communication service quality data indicates an increase in environmental interference level, relaxing the warping path window size of the dynamic time warping algorithm used to process the physical state data to suppress the interference of environmental noise on feature extraction; when the environmental interference level is reduced, tightening the warping path window size to improve the sensitivity to weak abnormal signals of the device itself.

[0015] Preferably, the fault prediction model is a cross-domain attention fusion network model, which uses the environmental interference context factor as an attention weight regulator when making fault predictions, dynamically increasing or decreasing the contribution of different physical state features to the final prediction result.

[0016] like Figure 2 As shown, the present invention also provides a maintenance device for a relay communication device in a mobile communication system, the device comprising: The cross-domain data sensing module is used to synchronously collect time series of physical status data and time series of communication service quality data of the device. The context-adaptive analysis module is used to generate an environmental interference context factor that characterizes the interference level of the current working environment in real time based on the communication service quality data; and to dynamically adjust the key parameters of the feature extraction algorithm according to the environmental interference context factor, and use the dynamically adjusted feature extraction algorithm to process the physical state data in order to extract context-adaptive physical state features. The integrated prediction and decision module is used to input the context-adaptive physical state characteristics and the communication service quality data into a preset fault prediction model, generate equipment fault prediction results, and generate a maintenance task plan based on the fault prediction results.

[0017] Specifically, physical state can be temperature, vibration, voltage, etc., while communication service quality can be signal error rate (BER), signal-to-noise ratio (SNR), etc. By introducing communication service quality data, the system can not only perceive hardware status, but also service quality.

[0018] The communication service quality data can specifically be data such as the instantaneous fluctuation of BER. Subsequently, the module uses this context factor to dynamically and in real time adjust the key parameters of the feature extraction algorithm, such as the DTW algorithm, used to process physical state data. Specifically, it can be a key parameter such as the normalized path window size. The operating mechanism of the context generation module and the adaptive feature extraction module is that when the system senses severe environmental interference such as a BER spike, it intelligently relaxes the tolerance of the analysis algorithm to filter out environmental noise; when the environment returns to calm, it tightens the tolerance to focus on the weak anomalies of the device itself. This makes the feature extraction process environmentally adaptive.

[0019] Preferably, the device further includes: The closed-loop self-optimization module is signal-connected to the fusion prediction and decision module. It is used to quantitatively evaluate the execution effect of the maintenance task after the maintenance task plan is executed, and obtain the maintenance effect evaluation score. When the maintenance effect evaluation score is lower than a preset threshold, it automatically triggers the fine-tuning and optimization of the parameters of the fault prediction model based on the fault prediction results and status data corresponding to the maintenance task.

[0020] This mechanism constructs an intelligent closed loop of "prediction-execution-evaluation-feedback-re-optimization," enabling the system to continuously learn and evolve.

[0021] Compared with the prior art, the present invention has the following advantages: This invention introduces communication service quality data as environmental context, which can effectively distinguish between environmental noise and real fault signals. While maintaining a high detection rate, it significantly reduces false alarms caused by environmental interference, solves industry pain points, and significantly reduces false alarm and missed alarm rates.

[0022] This invention introduces a closed-loop self-optimization module, enabling the system to learn from each maintenance practice and continuously optimize its prediction model. This allows the system to maintain a high level of prediction accuracy over a long period of time in constantly changing environments and equipment conditions, achieving system self-adaptation and self-evolution.

[0023] This invention, through the fusion analysis of cross-domain data on physical status and communication services, can reveal the deep intrinsic relationship between hardware health status and communication service quality, achieving a technological leap from single hardware fault prediction to comprehensive communication service quality risk early warning, and improving the depth and breadth of fault prediction. Attached Figure Description

[0024] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1This is a flowchart of a maintenance method for a relay communication device in a mobile communication system according to the present invention; Figure 2 This is a schematic block diagram of a maintenance device for a relay communication equipment in a mobile communication system according to the present invention. Figure 3 This is a schematic diagram illustrating the steps of context generation and adaptive feature extraction according to the present invention; Figure 4 This is a schematic diagram of a closed-loop self-optimization mechanism of the present invention; Figure 5 This is a graphical representation of the test results comparing the fault prediction accuracy of the embodiments of the present invention with that of traditional predictive maintenance methods in a simulated subway environment. Figure 6 This is a schematic diagram of the internal structure of a cross-domain attention fusion network model in one embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The present invention provides a maintenance method for a relay communication device in a mobile communication system, which specifically includes the following steps: Step S1, perform cross-domain data acquisition: synchronously acquire the time series of physical status data and communication service quality data of the device; Step S2, Execute context generation: Based on the communication service quality data, generate an environmental interference context factor in real time to indicate the current working environment interference level; Step S3, perform adaptive feature extraction: dynamically adjust the key parameters of a feature extraction algorithm according to the environmental interference context factor, and use the dynamically adjusted feature extraction algorithm to process the physical state data to extract context-adaptive physical state features. Step S4, perform fusion prediction: input the context-adaptive physical state features and the communication service quality data into a preset fault prediction model to generate equipment fault prediction results; Step S5, Decision generation: Based on the fault prediction results, generate a maintenance task plan.

[0027] Preferably, the method further includes: step S6, performing closed-loop self-optimization: after the maintenance task plan is executed, the execution effect of the maintenance task is quantitatively evaluated to obtain a maintenance effect evaluation score; when the score is lower than a preset threshold, the parameters of the fault prediction model are automatically fine-tuned and optimized based on the fault prediction results and status data corresponding to the maintenance task.

[0028] Preferably, in step S6, when the score is lower than a preset threshold, the parameters of the fault prediction model are automatically fine-tuned and optimized based on the fault prediction results and status data corresponding to the maintenance task. This further includes: when the score is lower than the preset threshold, the maintenance task is determined to be a false alarm maintenance, and the physical status features and communication service quality data that caused the false alarm prediction are marked as negative samples and used to reduce the sensitivity of the fault prediction model to such feature combinations.

[0029] Preferably, the communication service quality data includes at least one or a combination of several of the following: bit error rate, signal-to-noise ratio, and data retransmission rate; the physical status data includes at least one or a combination of several of the following: equipment operating temperature, internal component vibration amplitude, and circuit voltage and current.

[0030] Preferably, the feature extraction algorithm is a dynamic time warping algorithm, and the key parameter for dynamic adjustment is the warping path window size of the dynamic time warping algorithm.

[0031] Preferably, step S3, the adaptive feature extraction step, further includes: when the communication service quality data indicates an increase in environmental interference level, relaxing the warping path window size of the dynamic time warping algorithm used to process the physical state data to suppress the interference of environmental noise on feature extraction; when the environmental interference level is reduced, tightening the warping path window size to improve the sensitivity to weak abnormal signals of the device itself.

[0032] Preferably, the fault prediction model is a cross-domain attention fusion network model, which uses the environmental interference context factor as an attention weight regulator when making fault predictions, dynamically increasing or decreasing the contribution of different physical state features to the final prediction result.

[0033] This invention also provides a maintenance device for relay communication equipment in a mobile communication system, the device comprising: The cross-domain data sensing module is used to synchronously collect time series of physical status data and time series of communication service quality data of the device. The context-adaptive analysis module is used to generate an environmental interference context factor that characterizes the interference level of the current working environment in real time based on the communication service quality data; and to dynamically adjust the key parameters of a feature extraction algorithm according to the environmental interference context factor, and to process the physical state data using the dynamically adjusted feature extraction algorithm to extract context-adaptive physical state features. The integrated prediction and decision module is used to input the context-adaptive physical state characteristics and the communication service quality data into a preset fault prediction model, generate equipment fault prediction results, and generate a maintenance task plan based on the fault prediction results.

[0034] Specifically, physical state can be temperature, vibration, voltage, etc., while communication service quality can be signal error rate (BER), signal-to-noise ratio (SNR), etc. By introducing communication service quality data, the system can not only perceive hardware status, but also service quality.

[0035] The communication service quality data can specifically be data such as the instantaneous fluctuation of BER. Subsequently, the module uses this context factor to dynamically and in real time adjust the key parameters of the feature extraction algorithm, such as the DTW algorithm, used to process physical state data. Specifically, it can be a key parameter such as the normalized path window size. The operating mechanism of the context generation module and the adaptive feature extraction module is that when the system senses severe environmental interference such as a BER spike, it intelligently relaxes the tolerance of the analysis algorithm to filter out environmental noise; when the environment returns to calm, it tightens the tolerance to focus on the weak anomalies of the device itself. This makes the feature extraction process environmentally adaptive.

[0036] Preferably, the device further includes: The closed-loop self-optimization module is signal-connected to the fusion prediction and decision module. It is used to quantitatively evaluate the execution effect of the maintenance task after the maintenance task plan is executed, and obtain the maintenance effect evaluation score. When the maintenance effect evaluation score is lower than a preset threshold, it automatically triggers the fine-tuning and optimization of the parameters of the fault prediction model based on the fault prediction results and status data corresponding to the maintenance task.

[0037] Example 1 This embodiment provides a maintenance method and device for relay communication equipment in a mobile communication system applied in an urban subway environment. The characteristics of the subway environment are: trains pass by at regular intervals and at high speeds, generating severe, periodic vibrations and strong electromagnetic interference.

[0038] like Figure 2As shown, its system architecture includes a cross-domain data perception module 101, a context adaptive analysis module 102, a fusion prediction and decision module 103, and a closed-loop self-optimization module 104. The modules are connected by signals through an internal bus.

[0039] In a preferred embodiment, the cross-domain data sensing module 101 acquires physical state data by: a built-in high-precision triaxial accelerometer for acquiring equipment vibration data with a sampling frequency of 2kHz; a built-in high-precision thermocouple sensor for acquiring the operating temperature of the core chip of the equipment; and a built-in Hall effect sensor for monitoring the voltage and current of the power supply circuit.

[0040] In a preferred embodiment, the communication service quality data acquisition integrates a signal quality monitoring unit, which demodulates the downlink pilot signal in real time to obtain the bit error rate (BER) and signal-to-noise ratio (SNR), with the sampling frequency set to 10Hz. All acquired data is timestamped with high precision to ensure time continuity and synchronization, forming a complete BER time series to ensure time synchronization between different data sources, and is then transmitted to the context adaptive analysis module 102.

[0041] Context-adaptive analysis module 102, such as Figure 2 As shown, in a preferred embodiment, the acquisition logic of this module is as follows: S201. Perform sliding window calculation on the input BER time series to obtain its statistical characteristics within the short-term time window, including the mean and BER variance. Specifically, the short-term time window can be set to 1 second.

[0042] S202. The calculated BER variance is compared with a preset normal fluctuation threshold. For comparison, when the BER variance exceeds When a strong disturbance is detected in the current environment, such as a train approaching or leaving, the system generates a high environmental disturbance context factor. For example, specific settings If the BER variance is >0.8, the environment is considered stable, and a lower BER is generated. For example, specific settings <0.2.

[0043] S203. The synchronously transmitted vibration data time series is processed using the Dynamic Time Warping (DTW) algorithm to identify weak abnormal vibration modes that may indicate a fault. A key parameter of the DTW algorithm is the warping path window size. ,Depend on For dynamic control, the functional relationship can be set as follows: ,in Base window size, The adjustment coefficient is a positive value.

[0044] S204. The effect of this dynamic adjustment mechanism is that it adjusts automatically when a train passes by. When the value is high, The restrictions were temporarily relaxed, allowing the DTW algorithm to tolerate severe but normal vibrations and deformations caused by the train, avoiding misinterpretation as equipment malfunctions; during the stable period when the train is far away... When the value is low, This tightens the loop, thereby improving the algorithm's sensitivity to detecting subtle abnormal vibrations caused by loose components in the device itself.

[0045] S205. The module finally outputs the vibration characteristics after this adaptive processing, and sends them together with the original temperature, voltage, BER, SNR and other data as a multi-dimensional feature vector to the fusion prediction and decision module 103.

[0046] In a preferred embodiment, the prediction and decision-making module 103 employs a pre-trained cross-domain attention fusion network model, which receives multi-dimensional feature vectors from module 102; in another preferred embodiment, the attention mechanism within the model... The factor acts as an external regulatory signal, in When the value is high, the model will automatically reduce the attention weight of vibration-related features and instead focus more on analyzing features that are relatively unaffected by the environment, such as temperature and voltage.

[0047] The model outputs the probability value of equipment failure within a certain period of time, and can further output the most important contributing features that lead to the prediction result, such as the specific features of continuous temperature increase and slight decrease in SNR. When the failure probability exceeds the preset alarm threshold, the system automatically generates a maintenance task plan, which is sent to the terminal device of the operation and maintenance personnel through the network. The plan includes detailed information such as the geographical location of the faulty equipment, the predicted root cause, the suggested maintenance window time, and the required spare parts.

[0048] To enable those skilled in the art to more clearly understand the core working mechanism of the fusion prediction and decision module 103 described in this invention, the cross-domain attention fusion network model used therein will be further described in detail below with reference to the accompanying drawings.

[0049] Reference Figure 6 In its specific implementation, the cross-domain attention fusion network model may include an input and embedding module 501, a temporal feature extraction module 502, a cross-domain attention adjustment module 503, a feature fusion module 504, and a fault prediction output module 505.

[0050] The input and embedding module 501 is responsible for receiving multi-source heterogeneous data from upstream modules. Specifically, it includes physical state features processed by the context adaptive analysis module, as well as original communication service quality data and other physical state data. Specifically, the physical state features may be vibration feature sequences, the communication service quality data may be BER and SNR time series, and the physical state data may be temperature and voltage time series. The input and embedding module 501 maps data of different types and dimensions to a unified high-dimensional feature space through different embedding layers, which facilitates subsequent processing.

[0051] The time-series feature extraction module 502 preferably adopts a recurrent neural network structure such as Long Short-Term Memory Network (LSTM) or Gated Recurrent Unit (GRU). It processes various embedded feature sequences to capture the time dependence and dynamic change trend within each data sequence. For example, it can learn two different time-series patterns: a continuous and slow temperature increase and a sudden and sharp increase.

[0052] The cross-domain attention adjustment module 503 is a key module for realizing the core idea of ​​this invention. This module receives various features output by the temporal feature extraction module 502 and environmental interference context factors generated by the context adaptive analysis module 102. As an external, dynamic adjustment signal, the internal attention mechanism of the module executes the following logic when calculating the weights of each feature: First, like a traditional attention mechanism, it calculates a basic attention score based on the importance of the feature itself to the prediction target; then, it utilizes context factors. The score is dynamically adjusted; specifically, for features susceptible to environmental disturbances, such as vibration, the final attention weight is adjusted. The calculation formula can be illustrated as follows: ,in It is a with The value is a negatively correlated function, which means that when environmental disturbances are large... When the value is high, The value decreases, thus actively suppressing the weight of vibration characteristics; conversely, when the environment is clean... When the value is low, the feature is allowed or even its weight is increased.

[0053] The feature fusion module 504 fuses the various feature vectors that have been weighted by the cross-domain attention adjustment module 503. For example, by splicing or weighted summation, it generates a comprehensive state representation vector that contains all key information and has undergone intelligent selection.

[0054] The fault prediction output module 505 typically consists of several fully connected layers and a final output layer such as a Sigmoid or Softmax function. It receives the fused state representation vector, performs a nonlinear transformation, and finally outputs the probability value of a specific type of fault occurring in the device within a future period.

[0055] Through the above structure, the prediction model of this invention is no longer a simple black box. It is endowed with the ability to focus and ignore, similar to that of a human engineer. It can dynamically adjust its focus of analysis and diagnosis based on the real-time perceived quality of the communication environment, thereby achieving higher accuracy and stronger robustness in fault prediction in complex and ever-changing real-world environments.

[0056] Closed-loop self-optimization module 104, such as Figure 4 As shown, in a preferred embodiment, after performing maintenance tasks, the maintenance personnel report the maintenance results to the system through their terminal devices, such as the operation type being "fan module replaced" or the status being "no obvious physical abnormalities found, equipment restarted".

[0057] In step S301, the closed-loop self-optimization module receives the feedback and continuously monitors key performance indicators such as BER and temperature of the device within a preset evaluation cycle after maintenance.

[0058] Step S302: If, after maintenance, the previously indicated abnormal indicators return to the normal range, the system determines that it is a valid maintenance and gives the predicted event a high evaluation score. The sample data of this prediction is marked as high-quality positive samples and stored in the training library for subsequent regular periodic optimization of the model.

[0059] Step S3303: If the maintenance personnel report that no abnormality was found, and the various performance indicators of the equipment have not been substantially improved within the evaluation period, the system will determine it as a false alarm maintenance and give a low evaluation score.

[0060] Step S304: At this point, the system automatically triggers an immediate model fine-tuning process. It retrieves the original input data and feature vector that caused the false alarm, explicitly marks them as negative samples, and performs a backpropagation update on the network parameters of the prediction model with a small learning rate. This is mainly to reduce the probability of the model's fault prediction output when it encounters similar normal but extreme data combinations in the future.

[0061] To verify the beneficial effects of this invention, in a preferred embodiment, we built a test platform simulating a subway environment and collected data for one consecutive month. Using the same dataset, we tested both the traditional predictive maintenance model and the technical solution of this invention. The predictive maintenance model used only physical state data without context adaptation or closed-loop optimization. The test results are attached. Figure 5 As shown, and summarized in the table below: Experimental data clearly show that, by introducing context-adaptive analysis and closed-loop self-optimization mechanisms, the overall performance of this invention far surpasses that of existing technical solutions under strong interference environments.

[0062] Example 2 Another preferred embodiment of the present invention provides a maintenance method and apparatus for relay communication equipment of a mobile communication system applied in a large-scale intelligent manufacturing factory environment. This environment is characterized by a wide variety of equipment, an extremely complex electromagnetic environment, and diverse and irregular vibration sources, which places extremely high demands on the stability of the relay communication equipment.

[0063] The electromagnetic environment is extremely complex because of high-power motors, welding robots, and other similar environments. The vibration sources are diverse because they are often in conditions such as stamping machines and crane operation. The force communication equipment used for docking is usually used to ensure seamless roaming communication nodes between AGVs, mobile monitoring equipment, or handheld PDA terminals.

[0064] In this embodiment, the execution process of the cross-domain data sensing module 101 is specifically as follows: physical state data acquisition: in addition to conventional vibration, temperature and voltage data, this embodiment also adds monitoring of the metal fatigue of the equipment housing. Through the built-in miniature acoustic sensor, ultrasonic signals generated by structural stress during the operation of the equipment are collected to analyze potential structural damage risks.

[0065] In the complex wireless environment of the factory, in addition to collecting the bit error rate (BER) and signal-to-noise ratio (SNR), this module also focuses on monitoring the packet retransmission rate and communication delay jitter. These two indicators are particularly critical for automated equipment that requires precise collaborative operation, and their instantaneous deterioration often indicates strong and sudden channel interference in a local area.

[0066] The execution process of the context adaptive analysis module 102, environmental interference context factors. Generation: In this embodiment, context factors The generation of context factors is multi-dimensional. When the data packet retransmission rate suddenly rises above the preset threshold within 1 second, or the communication latency jitter exceeds 5ms, the system determines that the current environment is affected by strong pulse interference caused by the start-up or shutdown of large equipment or welding operations, thereby generating a high context factor. Under normal production conditions, It then remains at a low level.

[0067] Adaptive feature extraction, the module receives high After the value is set, not only will the warping path window of the Dynamic Time Warping (DTW) algorithm be widened as in Example 1 to ignore severe vibrations caused by the environment, but it will also temporarily reduce the analysis weight of the ultrasonic signals collected by the acoustic sensors. This is because loud noise in the environment can also pollute the acoustic signals. This dynamic adjustment can effectively avoid misjudging the external press noise as the structural fatigue signal of the equipment itself.

[0068] The execution process of the fusion prediction and decision module 103 is based on fusion prediction using an attention mechanism: In this embodiment, when the preset cross-domain attention fusion network model performs fault prediction, its internal attention mechanism is affected by context factors. Direct regulation, when When the value is high, the model will automatically reduce its attention to vibration and acoustic features, and instead allocate more computational resources and judgment weights to temperature and voltage / current features that are relatively unaffected by electromagnetic interference.

[0069] For example, even if the vibration signal is severe, if the temperature and voltage remain stable, the model will likely determine that the equipment is in normal working condition; conversely, if... When the value is very low, the model will pay close attention to the weak but continuous abnormal vibration or high-frequency ultrasonic signal, thus accurately predicting early minor faults such as wear of cooling fan bearings or loosening of internal component fixing screws.

[0070] Intelligent decision-making and maintenance task generation: When the model predicts that the probability of a fault exceeds the alarm threshold, the generated maintenance task plan will be more scenario-based. For example, if the predicted root cause is unstable power supply, the maintenance task will not only suggest repairing the equipment itself, but will also be linked to the factory's power grid monitoring system to remind maintenance personnel to pay attention to the power quality in the area.

[0071] The execution process of the closed-loop self-optimization module 104 and the quantitative evaluation of maintenance effect: maintenance personnel provide feedback on maintenance results through the factory's mobile maintenance APP, such as replacing aging power modules. After receiving the feedback, the closed-loop self-optimization module will start a 24-hour observation cycle to continuously compare key indicators such as voltage fluctuation variance and data packet retransmission rate before and after maintenance.

[0072] The model undergoes targeted fine-tuning. If all indicators return to normal and remain stable after maintenance, the system determines that this is a successful maintenance and stores the complete chain of "pre-fault data features -> fault prediction -> maintenance measures -> post-recovery data features" as a high-quality positive sample in the training library for regular iterative optimization of the model. If the indicators do not improve significantly after maintenance, it is determined that the maintenance measures do not match the root cause of the fault. This event is marked as a special sample, and a small-batch, high-learning-rate emergency fine-tuning is triggered, mainly to improve the model's ability to identify similar fuzzy fault patterns in the future.

[0073] As can be seen from the above embodiments, the technical solution proposed in this invention can flexibly adapt to complex interference environments in different industrial scenarios, and significantly improve the reliability and maintainability of communication equipment under extreme working conditions through intelligent and adaptive analysis and decision-making.

[0074] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0075] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A maintenance method for a relay communication device in a mobile communication system, characterized in that, Specifically, the following steps are included: Step S1, perform cross-domain data acquisition: synchronously acquire the time series of physical status data and communication service quality data of the device; Step S2, Execute context generation: Based on the communication service quality data, generate an environmental interference context factor in real time that characterizes the interference level of the current working environment; Step S3, perform adaptive feature extraction: dynamically adjust the key parameters of the feature extraction algorithm according to the environmental interference context factor, and use the dynamically adjusted feature extraction algorithm to process the physical state data to extract context-adaptive physical state features; Step S4, perform fusion prediction: input the context-adaptive physical state features and the communication service quality data into a preset fault prediction model to generate equipment fault prediction results; Step S5, Decision generation: Based on the fault prediction results, generate a maintenance task plan.

2. The maintenance method for a relay communication device in a mobile communication system according to claim 1, characterized in that, It also includes: Step S6, performing closed-loop self-optimization: After the maintenance task plan is executed, the execution effect of the maintenance task is quantitatively evaluated to obtain a maintenance effect evaluation score; when the score is lower than a preset threshold, the parameters of the fault prediction model are automatically fine-tuned and optimized based on the fault prediction results and status data corresponding to the maintenance task.

3. The maintenance method for a relay communication device in a mobile communication system according to claim 2, characterized in that, In step S6, when the score is lower than a preset threshold, the parameters of the fault prediction model are automatically fine-tuned and optimized based on the fault prediction results and status data corresponding to the maintenance task. This further includes: when the score is lower than the preset threshold, the maintenance task is determined to be a false alarm maintenance, and the physical status features and communication service quality data that caused the false alarm prediction are marked as negative samples and used to reduce the sensitivity of the fault prediction model to such feature combinations.

4. The maintenance method for a relay communication device in a mobile communication system according to claim 1, characterized in that, The communication service quality data includes at least one or a combination of several of the following: bit error rate, signal-to-noise ratio, and data retransmission rate; the physical status data includes at least one or a combination of several of the following: equipment operating temperature, internal component vibration amplitude, and circuit voltage and current.

5. The maintenance method for a relay communication device in a mobile communication system according to claim 1, characterized in that, The feature extraction algorithm is a dynamic time warping algorithm, and the key parameter for dynamic adjustment is the warping path window size of the dynamic time warping algorithm.

6. The maintenance method for a relay communication device in a mobile communication system according to claim 5, characterized in that, Step S3, which involves performing adaptive feature extraction, further includes: when the communication service quality data indicates an increase in environmental interference level, relaxing the warping path window size of the dynamic time warping algorithm used to process the physical state data to suppress the interference of environmental noise on feature extraction; and when the environmental interference level is reduced, tightening the warping path window size to improve the sensitivity to weak abnormal signals of the device itself.

7. The maintenance method for a relay communication device in a mobile communication system according to claim 1, characterized in that, The fault prediction model is a cross-domain attention fusion network model. When making fault predictions, it uses the environmental interference context factor as an attention weight regulator to dynamically increase or decrease the contribution of different physical state features to the final prediction result.

8. A maintenance device for relay communication equipment in a mobile communication system, characterized in that, The device includes: The cross-domain data sensing module is used to synchronously collect time series of physical status data and time series of communication service quality data of the device. The context-adaptive analysis module is used to generate an environmental interference context factor that characterizes the interference level of the current working environment in real time based on the communication service quality data; and to dynamically adjust the key parameters of a feature extraction algorithm according to the environmental interference context factor, and to process the physical state data using the dynamically adjusted feature extraction algorithm to extract context-adaptive physical state features. The integrated prediction and decision module is used to input the context-adaptive physical state characteristics and the communication service quality data into a preset fault prediction model, generate equipment fault prediction results, and generate a maintenance task plan based on the fault prediction results.

9. A maintenance device for a relay communication device in a mobile communication system according to claim 8, characterized in that, The device further includes: The closed-loop self-optimization module is signal-connected to the fusion prediction and decision module. It is used to quantitatively evaluate the execution effect of the maintenance task after the maintenance task plan is executed, and obtain the maintenance effect evaluation score. When the maintenance effect evaluation score is lower than a preset threshold, it automatically triggers the fine-tuning and optimization of the parameters of the fault prediction model based on the fault prediction results and status data corresponding to the maintenance task.

Images