Medium wave transmitter fault prediction and adaptive operation and maintenance method based on artificial intelligence

Abstract

The invention relates to a medium wave transmitter fault prediction and adaptive operation and maintenance method based on artificial intelligence, and the method comprises the steps: constructing an electromagnetic coupling topological graph through analyzing a circuit board design file, and quantifying a fault propagation path between power amplifier modules; fault current signals are collected in real time, and risk quantitative indexes are generated in combination with topological characteristics; modeling and predicting a cascading risk probability based on the graph structure; dynamically adjusting a radio frequency power path and a carrier frequency according to a topological position evaluation result; and utilizing an execution feedback closed loop to optimize the prediction model. The method solves the problems that a traditional scheme cannot quantify an electromagnetic coupling propagation path, prediction and hardware control are separated, and a dynamic optimization mechanism is lacked, achieves accurate blocking of a fault propagation path and adaptive evolution of a prediction model, and remarkably improves the operation stability and fault response efficiency of a medium-wave transmitter.

Description

Technical Field

[0001] This invention relates to the field of medium-wave transmitter fault prediction technology, and specifically to an artificial intelligence-based method for medium-wave transmitter fault prediction and adaptive operation and maintenance. Background Technology

[0002] As a core component of broadcast communication systems, the stability of the power amplifier module in a medium-wave transmitter directly impacts signal coverage quality. In complex electromagnetic environments, a failure in a single power amplifier module can trigger a cascading failure through electromagnetic coupling, leading to complete system malfunction. Current methods for predicting such failures primarily rely on monitoring thresholds for operating parameters such as temperature and current over-limit alarms, or on statistical model analysis based on historical data. While these methods attempt to achieve early warning by capturing abnormal signal characteristics, they still face significant limitations in real-world industrial scenarios.

[0003] The existing technology has the following key defects: First, traditional methods are difficult to quantify the fault propagation path caused by electromagnetic coupling, focusing only on single-point faults and ignoring mutual interference between modules, causing the prediction results to deviate from the actual propagation risk; Second, the prediction model is disconnected from operation and maintenance control, and cannot dynamically adjust hardware parameters according to real-time risks, resulting in a lag in early warning response; Finally, the static model lacks a closed-loop optimization mechanism and cannot adapt to dynamic changes such as power amplifier module aging and environmental interference. Summary of the Invention

[0004] Based on this, the purpose of this invention is to provide an artificial intelligence-based method for mid-wave transmitter fault prediction and adaptive operation and maintenance that can accurately predict cascaded faults, block propagation paths in real time, and dynamically optimize prediction models.

[0005] The objective of this invention is achieved through the following solution:

[0006] In a first aspect, the present invention provides a method for fault prediction and adaptive operation and maintenance of medium-wave transmitters based on artificial intelligence, comprising the following steps:

[0007] S1: Perform physical parameter analysis on the circuit board design files and electromagnetic interference data of the medium wave transmitter, extract the physical coordinates and electrical connection paths of all power amplifier modules, calculate the electromagnetic coupling strength between modules, construct a topology structure with nodes representing modules and edges representing coupling paths, and generate a weighted topology graph containing electromagnetic coupling strength and electromagnetic strength information.

[0008] S2: Real-time acquisition of current signals from faulty modules, extraction of time-frequency characteristic parameters, and analysis of the correlation between signal distortion and coupling strength of adjacent modules in conjunction with electromagnetic strength information to generate risk quantification indicators;

[0009] S3: Perform graph structure modeling on the weighted topology graph and risk quantification indicators, call the preset risk prediction model to calculate the cascade risk probability of each power amplifier module, and generate cascade risk probability data for each module.

[0010] S4: Perform topology position evaluation on cascaded risk probability data and weighted topology graph, adjust the RF power output path and carrier signal frequency according to the centrality weight of the module in the topology, generate hardware control instruction set and send it to the execution terminal of the medium wave transmitter;

[0011] S5: Process the execution results fed back from the execution terminal, optimize the risk prediction model by combining the cascaded risk probability data, dynamically adjust the topology weights and prediction parameters according to the actual control effect, and generate an updated risk prediction model.

[0012] In one embodiment, S1 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0013] S11: Perform coordinate analysis on the circuit board design file of the medium wave transmitter, extract the physical location coordinates of the power amplifier module and the power bus connection relationship, and generate module layout data;

[0014] S12: Perform spacing calculation on the module layout data, analyze the straight-line distance between adjacent modules based on the Euclidean distance algorithm, and generate a module distance matrix;

[0015] S13: Based on the vector network analyzer, the electromagnetic interference data of the medium wave transmitter is measured in real time. Combined with the module distance matrix, the weight calculation is performed to calculate the electromagnetic field interaction strength between the power amplifier modules and generate electromagnetic strength information characterizing the electromagnetic coupling relationship.

[0016] S14: Perform topology construction processing on the power amplifier modules, power bus connection relationships and electromagnetic intensity information of the circuit board in the module layout data. Use the power amplifier modules as topology nodes and the power bus connection paths as topology edges to generate a weighted topology graph. The weighted topology graph is used to indicate the fault propagation path and electromagnetic coupling intensity distribution.

[0017] In one embodiment, S2 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0018] S21: Real-time acquisition of the current signal of the faulty module, capturing pulse waveforms at a preset sampling rate through a high-speed sensor to generate real-time current data;

[0019] S22: Perform time-frequency analysis on real-time current data, extract pulse amplitude, rise time and fundamental frequency offset, and generate a three-dimensional feature vector containing amplitude-frequency characteristics;

[0020] S23: Perform distortion correlation processing on the three-dimensional feature vector and electromagnetic intensity information, calculate the correlation coefficient between the signal difference and coupling strength of adjacent modules, and generate risk quantification index. The risk quantification index is used to indicate the risk level of cascaded fault propagation between adjacent modules.

[0021] In one embodiment, S3 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0022] S31: Extract the adjacency relationship from the weighted topology graph, analyze the connection paths and coupling strength between each power amplifier module, and generate a topology adjacency matrix;

[0023] S32: Based on the preset risk prediction model, the risk quantification indicators are fused with features and combined with the preset module basic parameters to form node feature vectors, generating a multi-dimensional feature set;

[0024] S33: Perform graph neural network modeling on the topological adjacency matrix and multidimensional feature set, update the node state representation through a three-layer message passing mechanism, and generate a feature embedding matrix containing high-order abstract features;

[0025] S34: Perform probability mapping processing on the feature embedding matrix, call the risk classifier of the trained risk prediction model to calculate the cascading probability of each power amplifier module, and generate cascading risk probability data containing probability values ​​and confidence levels.

[0026] In one embodiment, S4 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0027] S41: Perform centrality calculation on the weighted topology graph, calculate the influence of the module in the topology based on the sum of the outgoing edge weights of each node, and generate the topology centrality weight;

[0028] S42: Perform strategy matching processing on cascaded risk probability data and topology centrality weights, switch the power output path of the power amplifier module according to the risk probability level and centrality weight range, calculate the carrier frequency compensation value, and generate adjustment strategy parameters containing path switching instructions and frequency adjustment values.

[0029] S43: Perform instruction conversion processing on the adjustment strategy parameters, convert the power path switching strategy into an RF switch control instruction sequence, convert the carrier frequency compensation value into a frequency modulation instruction, generate a hardware control instruction set, and send it to the execution terminal of the medium wave transmitter.

[0030] In one embodiment, S5 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0031] S51: Perform effect analysis on the execution results fed back by the execution terminal, extract the actual blocking time, signal stability index and frequency control accuracy data, and generate a control effect dataset;

[0032] S52: Perform deviation calculation processing on the cascaded risk probability data and control effect dataset, analyze the difference between the predicted risk probability and the actual blocking effect, and generate a time deviation vector and a frequency offset matrix.

[0033] S53: Perform dynamic decay processing on the topology weights and time deviation vectors of the weighted topology graph, reduce the weight values ​​of the faulty paths proportionally according to the degree of deviation, and generate updated topology weights.

[0034] S54: Optimize the model parameters by updating the topological weights and frequency offset matrix, and adjust the convolution kernel weights of the graph neural network through the backpropagation algorithm to generate an optimized risk prediction model.

[0035] In one embodiment, the formula for calculating the updated topology weights in the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention is as follows:

[0036]

[0037] in, To update the topological weights, w ij Let λ be the original topological weight from module i to module j, and t be the decay factor. p Let t be the predicted blocking time in the time deviation vector. a This represents the actual blocking time.

[0038] Secondly, this invention provides an artificial intelligence-based fault prediction and adaptive operation and maintenance system for medium-wave transmitters, which is configured with the following modules:

[0039] The topology construction module is used to analyze the physical parameters of the circuit board design files and electromagnetic interference data of the medium wave transmitter, extract the physical coordinates and electrical connection paths of all power amplifier modules, calculate the electromagnetic coupling strength between modules, construct a topology structure in which nodes represent modules and edges represent coupling paths, and generate a weighted topology graph containing electromagnetic coupling strength and electromagnetic strength information.

[0040] The risk quantification module is used to collect the current signal of the faulty module in real time, extract time-frequency characteristic parameters, and combine electromagnetic strength information to analyze the correlation between signal distortion and coupling strength of adjacent modules, and generate risk quantification indicators.

[0041] The cascaded probability module is used to perform graph structure modeling on the weighted topology graph and risk quantification indicators, call the preset risk prediction model to calculate the cascaded risk probability of each power amplifier module, and generate cascaded risk probability data for each module.

[0042] The control instruction module is used to evaluate the topology position of the cascaded risk probability data and weighted topology map, adjust the radio frequency power output path and carrier signal frequency according to the module's centrality weight in the topology, generate a hardware control instruction set and send it to the execution terminal of the medium wave transmitter.

[0043] The model optimization module is used to process the execution results fed back from the execution terminal, optimize the risk prediction model by combining cascaded risk probability data, dynamically adjust the topology weights and prediction parameters according to the actual control effect, and generate an updated risk prediction model.

[0044] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-mentioned artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance methods.

[0045] Fourthly, this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-mentioned artificial intelligence-based methods for medium-wave transmitter fault prediction and adaptive operation and maintenance.

[0046] In summary, the artificial intelligence-based fault prediction and adaptive operation and maintenance method for medium-wave transmitters provided in this application accurately quantifies the fault propagation path between power amplifier modules through electromagnetic coupling topology modeling, solving the problem that traditional solutions cannot track electromagnetic coupling effects; it generates dynamic risk quantification indicators by combining real-time current signals and topology features, enabling accurate capture of early cascade propagation risks; it integrates topology relationships and real-time signal features based on graph structure modeling to achieve high-confidence cascade risk probability prediction; it dynamically adjusts the RF power path and carrier frequency based on topology location assessment, enabling real-time matching of hardware parameters and risk levels, effectively blocking the physical path of fault propagation; and it utilizes closed-loop optimization of model parameters and topology weights using execution feedback data, enabling the prediction system to continuously adapt to dynamic factors such as equipment aging and environmental interference.

[0047] The method provided in this application can achieve full-chain prevention and control of cascaded faults in medium-wave transmitters, fundamentally solving the core defects of the separation between prediction models and hardware control and the lack of dynamic optimization mechanisms. It significantly improves the fault blocking success rate and system stability of medium-wave transmitters, while reducing operation and maintenance response delay and false alarm rate, providing full life cycle protection for key equipment in complex electromagnetic environments.

[0048] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0049] Figure 1 A flowchart illustrating an artificial intelligence-based method for fault prediction and adaptive operation and maintenance of a medium-wave transmitter, provided in an embodiment of this application;

[0050] Figure 2 A flowchart illustrating the updated risk prediction model provided in this application embodiment;

[0051] Figure 3 This is a schematic diagram of the structure of an artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance system, which is provided as another embodiment of this application. Detailed Implementation

[0052] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Preferred embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a thorough and complete understanding of the disclosure of the invention.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0054] In one embodiment, such as Figure 1 As shown, an artificial intelligence-based method for fault prediction and adaptive operation and maintenance of medium-wave transmitters is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0055] S1: Perform physical parameter analysis on the circuit board design documents and electromagnetic interference data of the medium-wave transmitter, extract the physical coordinates and electrical connection paths of all power amplifier modules, calculate the electromagnetic coupling strength between modules, construct a topology structure with nodes representing modules and edges representing coupling paths, and generate a weighted topology graph containing electromagnetic coupling strength and electromagnetic strength information.

[0056] Specifically, the system imports the circuit board design file of the medium-wave transmitter through a dedicated data interface. This design file details the layout and electrical connections of all components on the circuit board. The system performs in-depth analysis of the design file, identifying and extracting the physical coordinate information of the power amplifier modules, including the specific location of each power amplifier module on the circuit board and its relative position to surrounding components.

[0057] Simultaneously, the system also obtains detailed information on electrical connection paths from the design documents, covering the direction, width, length, and start and end points of the wires. Furthermore, the system connects to electromagnetic interference testing equipment to collect electromagnetic interference data generated by the medium-wave transmitter under different operating conditions and environmental conditions. This data includes key information such as the intensity, frequency range, and location of the interference source.

[0058] For example, the system integrates the collected data along a timeline to form a dataset containing multi-dimensional parameters, with each data point accompanied by a precise timestamp and a corresponding transmitter operating status identifier. When parsing the circuit board design file, the system extracts the physical coordinates of each component. The coordinate data includes XYZ axis parameters in three-dimensional space. Based on the wiring diagram, the system determines the start and end points and path parameters of the wire connections, generating an initial topology framework. The system uses electromagnetic field numerical calculation methods, combined with parameters such as the dielectric constant and conductor conductivity of the circuit board material, to construct a simulation model. It calculates the electromagnetic coupling parameters between components, converts them into weight values, marks high-risk paths, and then uses graph theory tools to generate a weighted topology graph, storing the real-time electromagnetic intensity information of each node.

[0059] S2: Real-time acquisition of current signals from faulty modules, extraction of time-frequency characteristic parameters, and analysis of the correlation between signal distortion and coupling strength of adjacent modules in conjunction with electromagnetic strength information to generate risk quantification indicators.

[0060] Specifically, the system integrates current sensors in the power supply circuit of each power amplifier module in the medium-wave transmitter. These sensors acquire current signals in real time and convert them into digital signals for transmission to the data acquisition system. The data acquisition system performs filtering, noise reduction, and normalization preprocessing on the signals. It then uses time-frequency analysis to extract features from the preprocessed current signals and employs a Fast Fourier Transform (FFT) algorithm to convert the current signals from the time domain to the frequency domain, obtaining their spectral characteristics. Simultaneously, it calculates statistical characteristic parameters such as the mean, variance, and peak value of the current signals in the time domain. These time-frequency characteristic parameters reflect the current variation characteristics of the power amplifier modules under different states. The system combines electromagnetic coupling strength information to analyze the correlation between signal distortion and coupling strength between adjacent modules. By establishing a mathematical model, this correlation is quantified, and the time-frequency characteristic parameters are fused with the electromagnetic coupling strength to generate a risk quantification index, which numerically characterizes the degree of fault risk of the power amplifier modules.

[0061] S3: Perform graph structure modeling on the weighted topology graph and risk quantification indicators, call the preset risk prediction model to calculate the cascade risk probability of each power amplifier module, and generate cascade risk probability data for each module.

[0062] Specifically, the system models a graph structure based on a weighted topological graph and risk quantification indicators, using a graph convolutional network as its basic framework. The weighted topological graph is taken as input, where node feature vectors contain operating parameters such as temperature and voltage, as well as risk quantification indicators, and edge features are coupling coefficients. The graph convolutional network consists of multiple convolutional layers and an output layer, using activation functions to handle nonlinear relationships.

[0063] For example, the pre-defined risk prediction model is trained using historical failure data. The system divides the historical failure data into training and validation sets, and adjusts the model parameters using the cross-entropy loss function as the optimization objective. During training, the system iteratively updates the model weights and monitors model performance through the validation set to ensure that the model output represents the probability values ​​of cascading failures occurring in each part within a set time period, forming a cascading risk probability dataset. During the modeling process, the system normalizes the input data to avoid the impact of extreme values ​​on model training.

[0064] S4: Perform topology position evaluation on cascaded risk probability data and weighted topology graph, adjust the RF power output path and carrier signal frequency according to the centrality weight of the module in the topology, generate hardware control instruction set and send it to the execution terminal of the medium wave transmitter.

[0065] Specifically, based on a weighted topology graph, the system uses graph theory's centrality measurement methods to evaluate the centrality of power amplifier modules within the topology, including indicators such as degree centrality, proximity centrality, and betweenness centrality, to determine the criticality of each module in the topology. Based on the centrality weights and cascading risk probability data, the system adjusts the RF power output path and carrier signal frequency through an optimization algorithm. For modules with high centrality and a high cascading risk probability, the system reduces their RF power output or switches the carrier signal frequency to other frequencies with less interference to prevent fault propagation.

[0066] Simultaneously, the system re-plans the RF power allocation and transmission path based on the network topology and electromagnetic coupling relationships to ensure stable system operation. The system converts the adjusted RF power output path and carrier signal frequency information into a hardware control instruction set, represented in binary code or a specific control protocol format, and sends it to the execution terminal of the medium-wave transmitter through the communication interface to achieve real-time control of the hardware devices.

[0067] S5: Process the execution results fed back from the execution terminal, optimize the risk prediction model by combining the cascaded risk probability data, dynamically adjust the topology weights and prediction parameters according to the actual control effect, and generate an updated risk prediction model.

[0068] Specifically, the system receives execution results from the execution terminal, including indicators such as the standing wave ratio (VSWR) after power adjustment, temperature change rate, and fault blocking status. The feedback data is preprocessed to remove outliers and ensure data validity. The system uses an incremental learning algorithm to optimize the risk prediction model, retaining important parameters from historical training and updating only weights relevant to new data. Each iteration uses a set amount of the latest data.

[0069] For example, based on the actual control effect, the system dynamically adjusts the topology weights. When the deviation between the actual fault propagation probability and the predicted value of a certain coupled path reaches a set condition, the weight value of the corresponding edge is corrected through Bayesian estimation. Simultaneously, the system optimizes prediction parameters, including the kernel size of the graph convolutional network and the risk threshold. The risk threshold is dynamically calibrated using a ROC curve. The optimized risk prediction model is saved to the local storage of the edge computing node. The system confirms that the updated model performance meets the requirements through a model validation process, ensuring that it can be used in subsequent fault prediction processes.

[0070] In summary, the artificial intelligence-based fault prediction and adaptive operation and maintenance method for medium-wave transmitters provided in this application accurately quantifies the fault propagation path between power amplifier modules through electromagnetic coupling topology modeling, solving the problem that traditional solutions cannot track electromagnetic coupling effects; it generates dynamic risk quantification indicators by combining real-time current signals and topology features, enabling accurate capture of early cascade propagation risks; it integrates topology relationships and real-time signal features based on graph structure modeling to achieve high-confidence cascade risk probability prediction; it dynamically adjusts the RF power path and carrier frequency based on topology location assessment, enabling real-time matching of hardware parameters and risk levels, effectively blocking the physical path of fault propagation; and it utilizes closed-loop optimization of model parameters and topology weights using execution feedback data, enabling the prediction system to continuously adapt to dynamic factors such as equipment aging and environmental interference.

[0071] The method provided in this application can achieve full-chain prevention and control of cascaded faults in medium-wave transmitters, fundamentally solving the core defects of the separation between prediction models and hardware control and the lack of dynamic optimization mechanisms. It significantly improves the fault blocking success rate and system stability of medium-wave transmitters, while reducing operation and maintenance response delay and false alarm rate, providing full life cycle protection for key equipment in complex electromagnetic environments.

[0072] In one embodiment, S1 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0073] S11: Perform coordinate analysis on the circuit board design file of the medium-wave transmitter, extract the physical location coordinates of the power amplifier module and the power bus connection relationship, and generate module layout data.

[0074] Specifically, when the system performs coordinate analysis on the circuit board design file of the medium-wave transmitter, it reads the Gerber file, PCB design drawing, and component list of the circuit board through a dedicated file analysis interface. During the analysis process, the system first identifies the layer information in the file, distinguishing the top layer, bottom layer, and internal wiring layer containing the component locations. Then, it uses a coordinate transformation algorithm to convert the relative coordinates in the design file into three-dimensional absolute coordinates, with the coordinate reference point set as the geometric center of the circuit board.

[0075] For each component of the power amplifier, the system extracts the vertex coordinates of its package boundary and determines its physical location coordinates through polygon fitting. The coordinate data includes X-axis and Y-axis planar positions and Z-axis height parameters. Simultaneously, the system analyzes the power bus wiring diagram of the circuit board, identifies the start, end, and branch nodes of the power bus, and determines the connection points between each power amplifier component and the power bus by tracing the wire connection paths. The system records the pin numbers and wire specifications of the connection points, forming a power bus connection relationship table. The system associates the extracted physical location coordinates with the power bus connection relationships, stores them according to circuit board areas, and generates module layout data. This data includes the unique identifier of each component, three-dimensional coordinate parameters, power bus connection path, and corresponding connection point information.

[0076] S12: Perform spacing calculation on the module layout data, analyze the straight-line distance between adjacent modules based on the Euclidean distance algorithm, and generate a module distance matrix.

[0077] Specifically, when the system processes the spacing calculation of the module layout data, it first extracts the three-dimensional coordinate parameters of each power amplifier component from the module layout data, and then selects the X-axis and Y-axis planar coordinates for distance calculation. Preferably, the system uses the Euclidean distance algorithm, taking the geometric center of each component as the calculation point, and calculates the straight-line distance in pairs, ignoring the influence of Z-axis height differences on the planar distance during the calculation process.

[0078] To determine adjacent modules, the system uses a preset spatial range threshold to filter out component pairs whose distance falls within the threshold range and marks them as adjacent. After calculation, the system sorts all components by unique identifiers and constructs a module distance matrix. The rows and columns of the matrix correspond to each component, and the matrix element values ​​are the straight-line distances between corresponding components. Self-distances within the same component are filled with zero values, and component pairs without adjacent relationships are also filled with zero values. After the matrix is ​​generated, the system uses a consistency check procedure to check the symmetry of the matrix and the rationality of the distance values, ensuring that the distance value between any two components in the matrix is ​​consistent with the reverse distance value. After the check passes, the module distance matrix is ​​stored in a data buffer. The buffer uses time-series tagging management to ensure that the data matches the version of the module layout data.

[0079] S13: Based on the vector network analyzer, the electromagnetic interference data of the medium-wave transmitter is measured in real time. Combined with the module distance matrix, the weight calculation is performed to calculate the electromagnetic field interaction strength between power amplifier modules and generate electromagnetic strength information characterizing the electromagnetic coupling relationship.

[0080] Specifically, when the system performs real-time measurement of electromagnetic interference data of a medium-wave transmitter using a vector network analyzer, the test port of the vector network analyzer is first connected to the transmitter's test interface via a coaxial cable. The test interfaces are located around the transmitter's power input, signal output, and various power amplifier components. During the measurement process, the system controls the vector network analyzer to perform frequency sweep measurements within a preset frequency range, recording the electromagnetic signal amplitude and phase information corresponding to each frequency point. The measurement interval is dynamically adjusted according to the frequency band range to ensure coverage of the transmitter's operating frequency band and potentially interfering frequency bands.

[0081] After the measurement data is transmitted to the system, the system filters the data to remove environmental noise and interference signals introduced by the measuring equipment itself, retaining only valid electromagnetic interference data. Combining the module distance matrix, the system uses an electromagnetic field coupling model for weight calculation, taking the straight-line distance as an influence factor and substituting it into the coupling strength calculation formula to calculate the electromagnetic field interaction strength between the relevant components of each power amplifier. The interaction strength is inversely proportional to the distance. The system integrates the calculation results into electromagnetic strength information, which includes a unique identifier combination for each component pair, the corresponding electromagnetic field interaction strength value, and the frequency parameters used during measurement.

[0082] S14: Perform topology construction processing on the power amplifier modules, power bus connection relationships and electromagnetic intensity information of the circuit board in the module layout data. Use the power amplifier modules as topology nodes and the power bus connection paths as topology edges to generate a weighted topology graph. The weighted topology graph is used to indicate the fault propagation path and electromagnetic coupling intensity distribution.

[0083] Specifically, when the system performs topology construction processing on the circuit board power amplifier-related components, power bus connection relationships, and electromagnetic intensity information in the module layout data, it first determines the basic elements of the topology structure, taking each power amplifier-related component as a topology node, with node attributes including the unique identifier of the component and its physical location coordinates; and taking the power bus connection path as a topology edge, with edge attributes including the start and end identifiers of the connection path and wire specification parameters.

[0084] For example, the system uses the electromagnetic field interaction strength value from the electromagnetic intensity information as the weight value of the topology edge and assigns it to the corresponding power bus connection path to form a weighted topology edge. For component pairs that have electromagnetic coupling but no direct power bus connection, the system adds virtual topology edges based on the interaction strength value in the electromagnetic intensity information. The weight value of the virtual edge is also the corresponding electromagnetic field interaction strength value, used to characterize the indirect electromagnetic coupling relationship. After the topology is constructed, the system generates a weighted topology graph containing all topology nodes and topology edges, stored in a graphical data format. The stored content includes a node list, an edge list, and an edge weight list, with the node list and edge list linked by a unique identifier.

[0085] In one embodiment, S2 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0086] S21: Real-time acquisition of the current signal of the faulty module, capturing pulse waveforms at a preset sampling rate through a high-speed sensor to generate real-time current data.

[0087] Specifically, when the system acquires the current signal of the faulty module in real time, it first determines the location of the current signal acquisition point. The acquisition point is set at the power input and signal output terminals of the faulty module to ensure complete capture of current changes during module operation. A high-speed sensor is mounted at the acquisition point using a dedicated mounting bracket, and an insulating gasket isolates the sensor from the module to avoid electrical interference. The sensor's signal output terminal is connected to the data acquisition unit via a shielded cable. The cable connector uses a threaded locking structure to prevent loosening and signal loss. The system controls the sensor to start acquisition according to a preset sampling rate. The sampling rate is set based on the operating frequency of the medium-wave transmitter and the characteristic frequency of the fault signal to ensure capture of detailed changes in the pulse waveform. During acquisition, the system performs analog-to-digital conversion on the analog signal output by the sensor. The converted digital signal is processed with a checksum and then stored in a temporary buffer in chronological order. The temporary buffer uses a cyclic overwrite mechanism; when the storage space reaches its limit, the earliest acquired data is automatically replaced. The real-time current data includes a sampling timestamp, current value sequence, and sensor identification information. The data format uses binary encoding to reduce storage usage.

[0088] S22: Perform time-frequency analysis on real-time current data, extract pulse amplitude, rise time and fundamental frequency offset, and generate a three-dimensional feature vector containing amplitude-frequency characteristics.

[0089] Specifically, when performing time-frequency analysis on real-time current data, the system first extracts the current value sequence from the real-time current data, removes outliers from the sequence, and determines outliers based on a preset threshold value range. Values ​​exceeding the threshold are replaced by interpolation of adjacent valid values. Preferably, the system can use a time-frequency analysis algorithm to analyze the processed current value sequence. The algorithm segments the sequence using a sliding window, with the window length determined based on the periodic characteristics of the current signal, and the overlap ratio between windows set according to signal continuity requirements. A Fourier transform is performed on the signal within each window to obtain the corresponding frequency domain features. Combined with the time domain waveform, pulse amplitude, rise time, and fundamental frequency offset are extracted. The pulse amplitude is the maximum peak value of the current signal within the window, the rise time is the time interval between the signal rising from the reference value to the peak value, and the fundamental frequency offset is the difference between the actual signal fundamental frequency and the nominal fundamental frequency. The system combines the three extracted parameters in the order of pulse amplitude, rise time, and fundamental frequency offset to form a three-dimensional feature vector. The three-dimensional feature vector contains parameter identifiers and corresponding values, retaining the original measurement accuracy without normalization.

[0090] S23: Perform distortion correlation processing on the three-dimensional feature vector and electromagnetic intensity information, calculate the correlation coefficient between the signal difference and coupling strength of adjacent modules, and generate risk quantification index. The risk quantification index is used to indicate the risk level of cascaded fault propagation between adjacent modules.

[0091] Specifically, when the system performs distortion correlation processing on the three-dimensional feature vector and electromagnetic intensity information, it first extracts the pulse amplitude, rise time, and fundamental frequency offset of adjacent modules from the three-dimensional feature vector, and calculates the signal difference between adjacent modules. Preferably, the signal difference can be calculated using a vector differential mode algorithm, which treats the three-dimensional feature vectors of adjacent modules as spatial vectors, calculates the magnitude between the vectors, and the magnitude value reflects the degree of signal distortion.

[0092] The system retrieves the electromagnetic field interaction strength value from the electromagnetic intensity information and pairs it with the signal difference degree to form a correlated data pair. Each correlated data pair includes a module identifier pair, the signal difference degree, and the corresponding electromagnetic field interaction strength value. The system uses a statistical analysis algorithm to process the correlated data pairs, calculating the correlation coefficient between the signal difference degree and the electromagnetic field interaction strength. The correlation coefficient is calculated based on the ratio of the sample covariance to the standard deviation. Based on the magnitude of the correlation coefficient, the system generates a risk quantification index through a mapping algorithm. This algorithm converts the correlation coefficient value into an index value within a specific range, and the magnitude of the index value corresponds to the risk level of cascaded fault propagation between adjacent modules. The risk quantification index includes the module identifier pair, the index value, and the calculation timestamp.

[0093] In one embodiment, S3 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0094] S31: Perform adjacency relationship extraction on the weighted topology graph, analyze the connection paths and coupling strength between each power amplifier module, and generate a topology adjacency matrix.

[0095] Specifically, the system performs adjacency relationship extraction processing on the weighted topology graph, reads the association data of nodes and edges in the weighted topology graph, and parses the connection paths between each power amplification section, including power bus connections and indirect connection paths formed by electromagnetic coupling. For each connection path, the system extracts the corresponding coupling strength value, numbers each power amplification section according to a preset node numbering rule, and constructs a two-dimensional matrix using the numbers as indexes, with the rows and columns of the matrix corresponding to different power amplification sections.

[0096] The element values ​​in the matrix are determined based on the connectivity and coupling strength. If two power amplification sections have a connection path, the element value is the corresponding coupling strength; otherwise, the element value is zero. In this way, the system generates a topological adjacency matrix, which is stored as a two-dimensional array. This matrix fully records the connectivity and coupling strength between each power amplification section, providing a topological foundation for subsequent graph neural network modeling.

[0097] S32: Based on the preset risk prediction model, the risk quantification indicators are fused with features and combined with the preset module basic parameters to form node feature vectors, generating a multi-dimensional feature set.

[0098] Specifically, the system performs feature fusion processing on risk quantification indicators based on a preset risk prediction model. The system inputs the risk quantification indicators into the feature fusion module, which expands and reorganizes the indicators using a preset fusion algorithm to eliminate redundant information. At the same time, the system calls preset module basic parameters, including the rated voltage, operating temperature range, and design life parameters of the power amplification section. These parameters are extracted from the equipment's factory configuration data and stored in a local database.

[0099] The system combines the fused risk quantification indicators with the module's basic parameters according to feature dimensions. Each power amplification section corresponds to a set of feature parameters, forming a node feature vector. All node feature vectors are arranged in the order of the power amplification section's number, constituting a multi-dimensional feature set. The multi-dimensional feature set is stored in tabular form, with each row corresponding to a feature vector of a power amplification section, containing risk-related features and basic parameter features, providing node attribute input for the graph neural network.

[0100] S33: Perform graph neural network modeling on the topological adjacency matrix and multidimensional feature set, update the node state representation through a three-layer message passing mechanism, and generate a feature embedding matrix containing high-order abstract features.

[0101] Specifically, the system performs graph neural network modeling on the topological adjacency matrix and the multidimensional feature set, initializing the graph neural network model. This model contains three message-passing layers, each consisting of a feature transformation unit and an aggregation unit. First, the first message-passing layer receives the topological adjacency matrix and the multidimensional feature set. The feature transformation unit performs linear transformations on the features of each node, and the aggregation unit aggregates the feature information of adjacent nodes using a weighted summation method to generate the first-layer node states. The second message-passing layer takes the first-layer node states as input and repeats the feature transformation and aggregation operations, incorporating the feature influence of more distant nodes. The third message-passing layer further processes the node states output from the second layer, forming higher-order abstract features.

[0102] After three layers of message passing, the system outputs the state representation of each node, arranges these state representations by node number, and generates a feature embedding matrix. The feature embedding matrix is ​​stored in matrix form, where each row vector represents the high-order abstract feature of the corresponding node, including the node's own attributes and the association information of its multi-order neighbors.

[0103] S34: Perform probability mapping processing on the feature embedding matrix, call the risk classifier of the trained risk prediction model to calculate the cascading probability of each power amplifier module, and generate cascading risk probability data containing probability values ​​and confidence levels.

[0104] Specifically, the system performs probability mapping processing on the feature embedding matrix and calls the risk classifier in the trained risk prediction model. This classifier consists of a fully connected layer and an activation function. First, each row vector in the feature embedding matrix is ​​input into the fully connected layer, and the feature space is transformed through matrix multiplication and bias adjustment. The transformed feature vectors are then processed by the activation function to output the probability value of cascade failure in each power amplification section. Simultaneously, the system determines the confidence level by calculating the entropy value of the predicted probability distribution; a lower entropy value indicates a higher confidence level. The system associates the probability value of each power amplification section with its corresponding confidence level, arranging them in node number order to generate cascade risk probability data. The cascade risk probability data is stored in list form, with each record containing a power amplification section identifier, a cascade failure probability value, and a confidence level.

[0105] In one embodiment, S4 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0106] S41: Perform centrality calculation on the weighted topology graph, calculate the influence of the module in the topology based on the sum of the outgoing edge weights of each node, and generate the topology centrality weight.

[0107] Specifically, the system performs centrality calculation on the weighted topology graph, reads the weight data of nodes and edges in the graph, and determines the set of outgoing edges for each node. Outgoing edges are connection paths from the current node to other nodes. For each node, the system extracts the weight values ​​of all its outgoing edges, which correspond to the electromagnetic coupling strength between nodes. The system sums the outgoing edge weight values ​​for each node to quantify its influence on the topology. The calculation formula is as follows:

[0108]

[0109] Among them, C i Let A represent the topological centrality weight of the i-th node. ij This represents the element in the i-th row and j-th column of the topological adjacency matrix, i.e., the outgoing edge weight from the i-th node to the j-th node. The system arranges the calculation results in node number order to generate a set of topological centrality weights. This set is stored in array form, with each element corresponding to the centrality weight of a node. The larger the value, the higher the influence of the node on other nodes, providing a quantitative basis for node importance in subsequent policy matching.

[0110] S42: Perform strategy matching processing on cascaded risk probability data and topology centrality weights, switch the power output path of the power amplifier module according to the risk probability level and centrality weight range, calculate the carrier frequency compensation value, and generate adjustment strategy parameters containing path switching instructions and frequency adjustment values.

[0111] Specifically, the system performs strategy matching processing on the cascaded risk probability data and topology centrality weights, extracts the probability values ​​of each power amplification part from the cascaded risk probability data, and divides them into different risk probability levels according to preset thresholds; at the same time, the topology centrality weights are divided into multiple intervals according to their numerical range.

[0112] Preferably, the system can call a preset strategy matching rule base, which contains processing strategies corresponding to different combinations of risk levels and centrality intervals. When the risk probability level and centrality weight interval meet a certain rule condition, the system activates the corresponding power output path switching mechanism, selecting an alternative path based on the connection relationships in the topological adjacency matrix to ensure the continuity of power transmission. For the carrier frequency compensation value, the system calculates it based on the deviation between the electromagnetic coupling strength and the current frequency, using the following formula:

[0113]

[0114] Where, Δf ik is the carrier frequency compensation value for the i-th power amplification section. f Here, f is the frequency compensation coefficient, f is the current carrier frequency of the adjacent node, and f0 is the nominal carrier frequency. The system combines the path switching command with the frequency adjustment value to generate adjustment strategy parameters. The parameters are stored in structured data format, including information such as target node identifier, path number, and frequency compensation value.

[0115] S43: Perform instruction conversion processing on the adjustment strategy parameters, convert the power path switching strategy into an RF switch control instruction sequence, convert the carrier frequency compensation value into a frequency modulation instruction, generate a hardware control instruction set, and send it to the execution terminal of the medium wave transmitter.

[0116] Specifically, the system performs instruction conversion processing on the adjustment strategy parameters. For the power path switching strategy, the system queries a preset RF switch correspondence table based on the path number to determine the RF switch number and action sequence that needs to be activated, and generates a control instruction sequence. The instruction format follows a preset switch control protocol and includes switch identifier, action type (closed / open), and execution timestamp. For the carrier frequency compensation value, the system converts it into a frequency modulation instruction and calculates the frequency adjustment voltage value using a frequency control algorithm. The formula is:

[0117] V f =k v ·Δf i

[0118] Among them, V f k is the voltage value corresponding to the frequency modulation command. v Δf is the voltage conversion coefficient. i This is the carrier frequency compensation value. The system sorts the RF switch control command sequence and frequency modulation command according to execution priority and encapsulates them into a hardware control command set. The command set adopts a frame structure, including a frame header, command content, and check bits. The system sends the command set to the execution terminal of the medium wave transmitter through a communication interface (such as RS485). After receiving the command set, the execution terminal performs verification. If the verification is successful, the corresponding operation is executed according to the command to ensure the accurate implementation of power path switching and frequency adjustment.

[0119] In one embodiment, such as Figure 2 As shown, S5 of the artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method provided by the present invention specifically includes the following steps:

[0120] S51: Perform effect analysis on the execution results fed back by the execution terminal, extract the actual blocking time, signal stability index and frequency control accuracy data, and generate a control effect dataset.

[0121] Specifically, the system analyzes and processes the execution results fed back by the execution terminal. The execution terminal records the time interval from the start of instruction execution to the interruption of fault propagation through its built-in timestamp recording module, and the system extracts this time interval as the actual blocking time. Simultaneously, the system acquires the signal waveform after blocking through the signal monitoring unit, calculates the amplitude fluctuation range and phase jitter value of the signal, and uses the weighted sum of these two as the signal stability index. For frequency control accuracy, the system measures the adjusted actual carrier frequency using a frequency meter and calculates the difference between it and the target frequency in the instruction to obtain frequency control accuracy data. The system correlates the actual blocking time, signal stability index, and frequency control accuracy data in a time sequence. Each record contains the corresponding power amplification section identifier, the execution instruction number, and the values ​​of the three parameters mentioned above, generating a control effect dataset.

[0122] S52: Perform deviation calculation processing on the cascaded risk probability data and control effect dataset, analyze the difference between the predicted risk probability and the actual blocking effect, and generate a time deviation vector and a frequency offset matrix.

[0123] Specifically, the system performs deviation calculation processing on the cascaded risk probability data and the control effect dataset. It reads the predicted blocking time from the cascaded risk probability data and the actual blocking time from the control effect dataset, calculates the difference between the two, and obtains a time deviation vector. Specifically, the formula for calculating the time deviation vector is:

[0124] t dev =t pred -t act

[0125] Among them, t dev Let t be the time deviation vector. pred To predict the blocking time vector, t act This represents the actual blocking time vector. Simultaneously, by comparing the predicted frequency adjustment values ​​with the actual frequency control accuracy data, the frequency offset is analyzed, and a frequency offset matrix is ​​generated. The formula for calculating the frequency offset matrix is:

[0126] F dev =F pred -F act

[0127] Among them, F dev F is the frequency offset matrix. pred For the predicted frequency adjustment matrix, F act This represents the actual frequency control accuracy data matrix. The time deviation vector and frequency offset matrix can intuitively reflect the difference between the predicted risk probability and the actual blocking effect, providing a quantitative basis for subsequent model optimization. Through these deviation calculations, the system can assess the accuracy of the current model's predictions and provide direction for subsequent model adjustments.

[0128] S53: Perform dynamic decay processing on the topology weights and time deviation vectors of the weighted topology graph, reduce the weight values ​​of the faulty paths proportionally according to the degree of deviation, and generate updated topology weights.

[0129] Specifically, the system dynamically attenuates the topology weights and time deviation vectors of the weighted topology graph. Based on the difference between the predicted blocking time and the actual blocking time in the time deviation vector, and combined with a preset attenuation factor, the system proportionally reduces the weight value of the fault path. Preferably, the calculation formula for updating the topology weights in the medium-wave transmitter fault prediction and adaptive operation and maintenance method is as follows:

[0130]

[0131] in, To update the topological weights, w ij Let λ be the original topological weight from module i to module j, and t be the decay factor. p Let t be the predicted blocking time in the time deviation vector. a This represents the actual blocking time. The system updates the weighted topology graph based on this formula, generating updated topology weights to more accurately reflect the current actual operating status of the network.

[0132] S54: Optimize the model parameters by updating the topological weights and frequency offset matrix, and adjust the convolution kernel weights of the graph neural network through the backpropagation algorithm to generate an optimized risk prediction model.

[0133] Specifically, the system optimizes the model parameters of the updated topological weights and frequency offset matrices. The updated topological weights and frequency offset matrices are then used as new training data and input into the graph neural network. Preferably, using the backpropagation algorithm, the system calculates the gradient of the loss function with respect to the convolutional kernel weights of the graph neural network and adjusts the convolutional kernel weights based on the gradient information. The formula for calculating the loss function is:

[0134]

[0135] Where L is the loss function value, n is the number of samples, and y i For the true value, This is the predicted value. The core of the backpropagation algorithm is calculating the gradient, that is:

[0136]

[0137] in, Let W be the gradient of the loss function with respect to the convolutional kernel weights. The system iteratively updates the convolutional kernel weights, continuously reducing the value of the loss function until the deviation between the model's predictions and the actual data is minimized, ultimately generating an optimized risk prediction model. The optimized model can more accurately predict cascading failure risks, providing more reliable decision support for fault prediction and adaptive operation and maintenance of medium-wave transmitters. During model optimization, the system automatically adjusts hyperparameters such as the learning rate to ensure the model's convergence speed and stability. Simultaneously, the system monitors the model's performance in real time, evaluating its generalization ability through validation and test sets to ensure the model's effectiveness and reliability in practical applications.

[0138] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0139] Based on the same inventive concept, this application also provides an AI-based medium-wave transmitter fault prediction and adaptive operation and maintenance system for implementing the aforementioned AI-based medium-wave transmitter fault prediction and adaptive operation and maintenance method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more AI-based medium-wave transmitter fault prediction and adaptive operation and maintenance system embodiments provided below can be found in the limitations of the AI-based medium-wave transmitter fault prediction and adaptive operation and maintenance method described above, and will not be repeated here.

[0140] Preferably, such as Figure 3 As shown, this invention provides an artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance system 600, which is configured with the following modules:

[0141] Topology building module 610 is used to analyze the physical parameters of the circuit board design file and electromagnetic interference data of the medium wave transmitter, extract the physical coordinates and electrical connection paths of all power amplifier modules, calculate the electromagnetic coupling strength between modules and construct a topology structure in which nodes represent modules and edges represent coupling paths, and generate a weighted topology graph containing electromagnetic coupling strength and electromagnetic strength information.

[0142] The risk quantification module 620 is used to collect the current signal of the faulty module in real time, extract time-frequency characteristic parameters, and combine electromagnetic strength information to analyze the correlation between signal distortion and coupling strength of adjacent modules, and generate risk quantification indicators.

[0143] The cascaded probability module 630 is used to perform graph structure modeling on the weighted topology graph and risk quantification indicators, call the preset risk prediction model to calculate the cascaded risk probability of each power amplifier module, and generate cascaded risk probability data for each module.

[0144] The control instruction module 640 is used to evaluate the topology position of the cascaded risk probability data and the weighted topology map, adjust the radio frequency power output path and carrier signal frequency according to the centrality weight of the module in the topology, generate a hardware control instruction set and send it to the execution terminal of the medium wave transmitter.

[0145] The model optimization module 650 is used to process the execution results fed back by the execution terminal, optimize the risk prediction model by combining the cascaded risk probability data, dynamically adjust the topology weights and prediction parameters according to the actual control effect, and generate an updated risk prediction model.

[0146] Preferably, the topology building module 610 provided in this application is configured with the following units:

[0147] The coordinate analysis unit is used to perform coordinate analysis on the circuit board design file of the medium wave transmitter, extract the physical location coordinates of the power amplifier module and the power bus connection relationship, and generate module layout data.

[0148] The spacing calculation unit is used to perform spacing calculation on the module layout data, analyze the straight-line distance between adjacent modules based on the Euclidean distance algorithm, and generate a module distance matrix.

[0149] The electromagnetic strength unit is used to measure the electromagnetic interference data of the medium-wave transmitter in real time based on the vector network analyzer. It combines the module distance matrix to perform weight calculation processing, calculates the electromagnetic field interaction strength between power amplifier modules, and generates electromagnetic strength information characterizing the electromagnetic coupling relationship.

[0150] The topology building unit is used to perform topology building processing on the power amplifier modules, power bus connection relationships and electromagnetic strength information of the circuit board in the module layout data. With the power amplifier modules as topology nodes and the power bus connection paths as topology edges, a weighted topology graph is generated. The weighted topology graph is used to indicate the fault propagation path and electromagnetic coupling strength distribution.

[0151] Preferably, the risk quantification module 620 provided in this application is configured with the following units:

[0152] The current acquisition unit is used to acquire the current signal of the fault module in real time. It captures the pulse waveform at a preset sampling rate through a high-speed sensor and generates real-time current data.

[0153] The time-frequency feature unit is used to perform time-frequency analysis on real-time current data, extract pulse amplitude, rise time and fundamental frequency offset, and generate a three-dimensional feature vector containing amplitude-frequency characteristics.

[0154] The distortion correlation unit is used to perform distortion correlation processing on the three-dimensional feature vector and electromagnetic intensity information, calculate the correlation coefficient between the signal difference and coupling strength of adjacent modules, and generate risk quantification index. The risk quantification index is used to indicate the risk level of cascaded fault propagation between adjacent modules.

[0155] Preferably, the cascaded probability module 630 provided in this application is configured with the following units:

[0156] The adjacency extraction unit is used to extract adjacency relationships from the weighted topology graph, analyze the connection paths and coupling strengths between each power amplifier module, and generate a topology adjacency matrix.

[0157] The feature fusion unit is used to perform feature fusion on risk quantification indicators based on a preset risk prediction model, and combine them with preset module basic parameters to form node feature vectors, generating a multi-dimensional feature set.

[0158] The graph network modeling unit is used to model graph neural networks on topological adjacency matrices and multidimensional feature sets. It updates node state representations through a three-layer message passing mechanism and generates a feature embedding matrix containing high-order abstract features.

[0159] The probability mapping unit is used to perform probability mapping processing on the feature embedding matrix, call the risk classifier of the trained risk prediction model to calculate the cascading probability of each power amplifier module, and generate cascading risk probability data containing probability values ​​and confidence levels.

[0160] Preferably, the control command module 640 provided in this application is configured with the following units:

[0161] The centrality calculation unit is used to perform centrality calculation on the weighted topology graph. It calculates the influence of the outgoing edge weights of each node on the topology and generates the topology centrality weight.

[0162] The strategy matching unit is used to perform strategy matching processing on cascaded risk probability data and topology centrality weights, switch the power output path of the power amplifier module according to the risk probability level and centrality weight range, calculate the carrier frequency compensation value, and generate adjustment strategy parameters containing path switching instructions and frequency adjustment values.

[0163] The instruction conversion unit is used to perform instruction conversion processing on the adjustment strategy parameters, convert the power path switching strategy into a sequence of radio frequency switch control instructions, convert the carrier frequency compensation value into a frequency modulation instruction, generate a hardware control instruction set, and send it to the execution terminal of the medium wave transmitter.

[0164] Preferably, the model optimization module 650 provided in this application is configured with the following units:

[0165] The effect analysis unit is used to analyze the execution results fed back by the execution terminal, extract the actual blocking time, signal stability index and frequency control accuracy data, and generate control effect dataset;

[0166] The deviation calculation unit is used to perform deviation calculation processing on the cascaded risk probability data and control effect dataset, analyze the difference between the predicted risk probability and the actual blocking effect, and generate a time deviation vector and a frequency offset matrix.

[0167] The topology weight update unit is used to dynamically decay the topology weights and time deviation vectors of the weighted topology graph, and proportionally reduce the weight values ​​of the faulty paths according to the degree of deviation to generate updated topology weights.

[0168] The model parameter optimization unit is used to optimize the model parameters of the updated topological weights and frequency offset matrix. It adjusts the convolution kernel weights of the graph neural network through the backpropagation algorithm to generate an optimized risk prediction model.

[0169] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method.

[0170] In one embodiment, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described artificial intelligence-based medium-wave transmitter fault prediction and adaptive operation and maintenance method.

[0171] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0172] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0173] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all 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. An artificial intelligence-based medium wave transmitter fault prediction and adaptive operation and maintenance method, characterized in that, The method comprises the following steps: S1: physical parameter analysis is performed on the circuit board design file and electromagnetic interference data of the medium wave transmitter, physical coordinates and electrical connection paths of all power amplifier modules are extracted, electromagnetic coupling strength between the modules is calculated, a topology structure in which nodes represent modules and edges represent coupling paths is constructed, and a weighted topology graph containing electromagnetic coupling strength and electromagnetic intensity information is generated; S2: real-time acquisition is performed on the current signal of the faulty module, time-frequency characteristic parameters are extracted, and the correlation between signal distortion of adjacent modules and coupling strength is analyzed in combination with the electromagnetic intensity information, thereby generating a risk quantization index; S3: graph structure modeling is performed on the weighted topology graph and the risk quantization index, a preset risk prediction model is called to calculate the cascading risk probability of each power amplifier module, and cascading risk probability data of each module is generated; S4: topology position evaluation is performed on the cascading risk probability data and the weighted topology graph, the center degree weight of the module in the topology is used to adjust the radio frequency power output path and the carrier signal frequency, a hardware control instruction set is generated, and the hardware control instruction set is sent to an execution terminal of the medium wave transmitter; S5: the execution result fed back by the execution terminal is processed, the risk prediction model is optimized in combination with the cascading risk probability data, the topology weight and the prediction parameter are dynamically adjusted according to the actual control effect, and an updated risk prediction model is generated.

2. The method of claim 1, wherein, The S1 comprises: S11: coordinate analysis processing is performed on the circuit board design file of the medium wave transmitter, the physical position coordinates of the power amplifier module and the power bus connection relationship are extracted, and module layout data is generated; S12: distance calculation processing is performed on the module layout data, the straight-line distance between adjacent modules is analyzed based on the Euclidean distance algorithm, and a module distance matrix is generated; S13: real-time measurement is performed on the electromagnetic interference data of the medium wave transmitter based on a vector network analyzer, weight calculation processing is performed in combination with the module distance matrix, the electromagnetic field interaction strength between the power amplifier modules is calculated, and electromagnetic intensity information representing the electromagnetic coupling relationship is generated; S14: topology construction processing is performed on the power amplifier module, the power bus connection relationship of the circuit board of the module layout data, and the electromagnetic intensity information, the power amplifier module is taken as a topology node, and the power bus connection path is taken as a topology edge, thereby generating a weighted topology graph, and the weighted topology graph is used to indicate the fault propagation path and the electromagnetic coupling strength distribution.

3. The method of claim 1, wherein, The S2 comprises: S21: real-time acquisition is performed on the current signal of the faulty module, a high-speed sensor is used to capture a pulse waveform at a preset sampling rate, and real-time current data is generated; S22: time-frequency analysis is performed on the real-time current data, pulse amplitude, rise time, and frequency offset are extracted, and a three-dimensional feature vector containing amplitude-frequency characteristics is generated; S23: distortion correlation processing is performed on the three-dimensional feature vector and the electromagnetic intensity information, the correlation coefficient of the signal difference degree of adjacent modules and the coupling strength is calculated, a risk quantization index is generated, and the risk quantization index is used to indicate the propagation risk level of cascading faults between adjacent modules.

4. The method of claim 1, wherein, The S3 comprises: S31: Adjacency relationship extraction processing is performed on the weighted topology graph, connection paths and coupling strength between the power amplifier modules are analyzed, and a topology adjacency matrix is generated; S32: Based on the preset risk prediction model, the risk quantitative indicators are fused, combined with the preset module basic parameters to form a node feature vector, and a multi-dimensional feature set is generated; S33: The topology adjacency matrix and the multi-dimensional feature set are modeled by a graph neural network, the node state representation is updated through a three-layer message passing mechanism, and a feature embedding matrix containing high-order abstract features is generated; S34: The feature embedding matrix is subjected to probability mapping processing, the risk classifier of the trained risk prediction model is called to calculate the cascade occurrence probability of each power amplifier module, and cascade risk probability data containing probability values and confidence levels are generated.

5. The method of claim 1, wherein, The S4 includes: S41: Center degree calculation processing is performed on the weighted topology graph, the influence degree of the module in the topology is calculated based on the sum of the out-edge weight values of each node, and a topology center degree weight is generated; S42: Strategy matching processing is performed on the cascade risk probability data and the topology center degree weight, the power output path of the power amplifier module is switched according to the risk probability level and the center degree weight interval, and the carrier frequency compensation value is calculated, and adjustment strategy parameters containing path switching instructions and frequency adjustment values are generated; S43: Instruction conversion processing is performed on the adjustment strategy parameters, the power path switching strategy is converted into a radio frequency switch control instruction sequence, the carrier frequency compensation value is converted into a frequency modulation instruction, a hardware control instruction set is generated and sent to the execution terminal of the medium wave transmitter.

6. The method according to any one of claims 1 to 5, characterized in that, The S5 includes: S51: Effect analysis processing is performed on the execution results fed back by the execution terminal, actual blocking time, signal stability indicators and frequency control accuracy data are extracted, and a control effect data set is generated; S52: Deviation calculation processing is performed on the cascade risk probability data and the control effect data set, the difference between the predicted risk probability and the actual blocking effect is analyzed, and a time deviation vector and a frequency offset matrix are generated; S53: Dynamic attenuation processing is performed on the topology weight of the weighted topology graph and the time deviation vector, the weight value of the fault path is reduced in proportion according to the deviation degree, and an updated topology weight is generated; S54: Model parameter optimization processing is performed on the updated topology weight and the frequency offset matrix, the convolution kernel weight of the graph neural network is adjusted through the back propagation algorithm, and an optimized risk prediction model is generated.

7. The method of claim 6, wherein, The calculation formula of the updated topology weight is: wherein, is the updated topology weight, w ij is the original topology weight from module i to module j, λ is a decay factor, t p is the predicted outage time in the time bias vector, t a is the actual outage time.

8. An artificial intelligence-based medium wave transmitter fault prediction and adaptive operation and maintenance system, characterized in that, The system includes: A topology construction module is configured to analyze physical parameters of a circuit board design file and electromagnetic interference data of a medium wave transmitter, extract physical coordinates and electrical connection paths of all power amplifier modules, calculate electromagnetic coupling strength between the modules, and construct a topology structure of node representation modules and edge representation coupling paths, to generate a weighted topology graph containing electromagnetic coupling strength and electromagnetic intensity information. A risk quantification module is configured to collect current signals of fault modules in real time, extract time-frequency feature parameters, and analyze the correlation between signal distortion of adjacent modules and coupling strength in combination with the electromagnetic intensity information, to generate risk quantitative indicators. A cascading probability module is configured to model a graph structure of the weighted topology graph and the risk quantification index, call a preset risk prediction model to calculate a cascading risk probability of each power amplifier module, and generate cascading risk probability data of each module; A control instruction module is configured to perform topology position evaluation on the cascading risk probability data and the weighted topology graph, adjust a radio frequency power output path and a carrier signal frequency according to a centrality weight of a module in a topology, generate a hardware control instruction set, and send the hardware control instruction set to an execution terminal of the medium wave transmitter; A model optimization module is configured to process an execution result fed back by the execution terminal, perform model optimization on the risk prediction model in combination with the cascading risk probability data, dynamically adjust a topology weight and a prediction parameter according to an actual control effect, and generate an updated risk prediction model. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the method in any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the method in any one of claims 1 to 7.

Images