Radar fault early warning self-checking method and device based on semantic driving and medium

Abstract

The invention discloses a radar fault early warning self-checking method and device based on semantic driving and a medium, and belongs to the technical field of radar fault early warning. The method comprises the steps of obtaining operation state data of a radar system based on a preset acquisition frequency; processing the operation state data based on a trained machine learning algorithm to generate early warning information; after the prediction occurrence time window ends, calling a historical record interface to query alarm information and fault list data in the time window; performing semantic analysis and matching on the early warning information, the alarm information and the fault list to determine a semantic matching result; processing the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; and if the confidence score exceeds a preset threshold, marking as high-confidence risk early warning and pushing to a preset operation and maintenance platform. According to the method, the technical effects of intelligent semantic matching and real-time self-checking of the early warning information and the fault evidence are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar fault early warning, and in particular to a radar fault early warning self-checking method based on semantic driving, a device and a medium. BACKGROUND

[0002] With the development of artificial intelligence technology, radar fault early warning systems have gradually shifted from traditional rule and threshold judgment to intelligent analysis. Currently, the field of radar fault early warning has given rise to many mature algorithms. These algorithms can achieve early warning of potential faults by analyzing radar device operating state data, environmental monitoring data and other data sources. Although these algorithms have made certain progress in generating radar fault early warning, the early warning information lacks an effective self-checking mechanism, and the correlation between the early warning information and actual faults cannot be analyzed and verified in real time. Existing self-checking methods rely on manual checking, and the checking process usually involves manually matching the alarm information with the fault report. This process is not only inefficient, but also susceptible to human factors, affecting the reliability of the radar fault early warning system.

[0003] Therefore, how to achieve intelligent semantic matching and real-time self-checking of early warning information and fault evidence has become a technical problem to be solved. SUMMARY

[0004] The embodiments of the present application provide a radar fault early warning self-checking method based on semantic driving, a device and a medium, to solve the technical problem of how to achieve intelligent semantic matching and real-time self-checking of early warning information and fault evidence.

[0005] In a first aspect, the embodiments of the present application provide a radar fault early warning self-checking method based on semantic driving, applied to a radar system, the method comprising: acquiring operating state data of the radar system based on a preset acquisition frequency; wherein the operating state data includes transmission channel parameters, reception channel parameters, antenna system parameters, environmental parameters and state parameters; processing the operating state data based on a trained machine learning algorithm to generate early warning information; wherein the early warning information includes a predicted fault category and a predicted occurrence time window; after the predicted occurrence time window ends, calling a historical record interface to query alarm information and fault report data within the time window; wherein the alarm information is generated by a radar built-in test system based on rule thresholds, and the fault report is an actual fault description filled out by station personnel; performing semantic analysis and matching on the early warning information, the alarm information and the fault report to determine a semantic matching result; wherein the semantic analysis and matching includes semantic similarity calculation and knowledge graph reasoning; processing the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; if the confidence score exceeds a preset threshold, marking it as a high-confidence risk early warning and pushing it to a preset operation and maintenance platform.

[0006] In an implementation form of the present application, the running state data of the radar system is obtained based on a preset acquisition frequency, specifically comprising: obtaining a plurality of data sources of the radar system; wherein the plurality of data sources include parameters of system control layer, signal processing layer and environment guarantee layer; deploying sensor nodes, interfaces and log monitoring components to the radar system, and collecting running key parameters at a preset time interval; collecting the transmission channel parameters, receiving channel parameters, antenna system parameters, environment parameters and state parameters; wherein the transmission channel parameters include transmission power, voltage, current and pulse width, the receiving channel parameters include receiving gain, noise coefficient and phase error, the antenna system parameters include horizontal antenna peak power, vertical antenna peak power and direction angle, the environment parameters include machine room temperature, humidity and power supply state, and the state parameters include system running state code, time stamp and fault flag; the plurality of data sources, running key parameters, transmission channel parameters, receiving channel parameters, antenna system parameters, environment parameters and state parameters are stored in a structured manner.

[0007] In an implementation form of the present application, the running state data is processed based on the trained machine learning algorithm to generate early warning information, specifically comprising: constructing a heterogeneous model cluster containing bidirectional long short-term memory network, self-attention mechanism network and graph neural network; processing time series type running state data based on the bidirectional long short-term memory network to capture long-term time series dependency; extracting key time series features based on the self-attention mechanism network to strengthen the parameter correlation of important time points; modeling the topological relationship between components of the radar system based on the graph neural network to process structured running state data; integrating the output results of the bidirectional long short-term memory network, self-attention mechanism network and graph neural network to generate structured early warning information; wherein the early warning information includes rapid temperature rise, power supply anomaly, humidity rise and power degradation.

[0008] In an implementation form of the present application, after the prediction occurrence time window ends, a historical record interface is called to query alarm information and fault single data within the time window, specifically comprising: setting the query time window of the historical record interface, and aligning it with the occurrence time window of the early warning information; calling the historical record interface when detecting the prediction occurrence time window end event; obtaining the alarm information of the radar system, and extracting fault single text from a preset radar station operation and maintenance system.

[0009] In an implementation form of the present application, the pre-warning information, the alarm information and the fault report are subjected to semantic analysis and matching to determine a semantic matching result, specifically including: performing basic semantic analysis on the pre-warning information, the alarm information and the fault report to obtain a word segmentation result, a part-of-speech tagging and a named entity; extracting a keyword set of the pre-warning information, the alarm information and the fault report based on a pre-trained language model; mapping the keyword set to a multi-dimensional topic vector based on a pre-set topic modeling algorithm to generate a semantic topic distribution; performing synonym rewriting and sentence conversion on the pre-warning information, the alarm information and the fault report to generate an enhanced text set; calculating a semantic similarity of the enhanced text set and a pre-set historical fault case, and selecting a maximum value as a text semantic matching score; performing entity alignment and path reasoning on the keyword set based on a domain knowledge graph to generate a knowledge graph reasoning similarity; and determining a semantic matching result based on the knowledge graph reasoning similarity and the text semantic matching score.

[0010] In an implementation form of the present application, the semantic similarity of the enhanced text set and the historical fault case is calculated, and a maximum value is selected as a text semantic matching score, specifically including: calling a pre-trained language model to perform vector encoding on the enhanced text set and the historical fault case to generate a text embedding representation; calculating a similarity score of each pair of text embedding representations through a cosine similarity algorithm; traversing all combinations of the enhanced text and the historical fault case to record a maximum similarity score; and outputting the maximum similarity score as a text semantic matching score.

[0011] In an implementation form of the present application, the keyword set is subjected to entity alignment and path reasoning based on a domain knowledge graph to generate a knowledge graph reasoning similarity, specifically including: constructing a domain knowledge graph covering radar components, parameter abnormalities, fault types and symptoms; wherein the node types of the domain knowledge graph include component names, operating parameters, fault types and fault symptoms, and the edge relationships include trigger, cause and attributed to; performing entity alignment on the keyword set and the knowledge graph nodes to determine standardized concept nodes; calculating a path similarity of the standardized concept nodes and historical fault case nodes based on a pre-set graph embedding model; and aggregating the path similarity scores of multi-hop paths through an entity propagation mechanism to generate a knowledge graph reasoning similarity.

[0012] In an implementation form of the present application, the semantic matching result is processed based on a pre-set comprehensive weighted scoring algorithm to determine a confidence score, specifically including: obtaining a text semantic matching score and a knowledge graph reasoning similarity output by semantic analysis and matching; processing the text semantic matching score and the knowledge graph reasoning similarity based on a pre-set weight coefficient to determine a confidence score.

[0013] In a second aspect, the embodiments of the present application also provide a radar fault early warning self-checking device based on semantic driving, the device comprising: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: acquire running state data of a radar system based on a preset acquisition frequency; wherein the running state data comprises transmission channel parameters, reception channel parameters, antenna system parameters, environmental parameters and state parameters; process the running state data based on a trained machine learning algorithm to generate early warning information; wherein the early warning information comprises a predicted fault category and a predicted occurrence time window; after the predicted occurrence time window ends, call a historical record interface to query alarm information and fault single data within the time window; wherein the alarm information is generated by a radar built-in test system based on a rule threshold, and the fault single is an actual fault description filled in by station personnel; perform semantic analysis and matching on the early warning information, the alarm information and the fault single to determine a semantic matching result; wherein the semantic analysis and matching comprise semantic similarity calculation and knowledge graph reasoning; process the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; and if the confidence score exceeds a preset threshold, mark as a high-confidence risk early warning and push to a preset operation and maintenance platform.

[0014] In a third aspect, the embodiments of the present application also provide a nonvolatile computer storage medium for radar fault early warning self-checking based on semantic driving, which stores computer executable instructions, and the computer executable instructions are configured to: acquire running state data of a radar system based on a preset acquisition frequency; wherein the running state data comprises transmission channel parameters, reception channel parameters, antenna system parameters, environmental parameters and state parameters; process the running state data based on a trained machine learning algorithm to generate early warning information; wherein the early warning information comprises a predicted fault category and a predicted occurrence time window; after the predicted occurrence time window ends, call a historical record interface to query alarm information and fault single data within the time window; wherein the alarm information is generated by a radar built-in test system based on a rule threshold, and the fault single is an actual fault description filled in by station personnel; perform semantic analysis and matching on the early warning information, the alarm information and the fault single to determine a semantic matching result; wherein the semantic analysis and matching comprise semantic similarity calculation and knowledge graph reasoning; process the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; and if the confidence score exceeds a preset threshold, mark as a high-confidence risk early warning and push to a preset operation and maintenance platform.

[0015] The embodiment of the application provides a radar fault early warning self-checking method and device based on semantic driving and a medium, and at least includes the following technical effects: The application greatly improves the matching accuracy of early warning information and actual fault events through a semantic matching algorithm and a knowledge graph reasoning technology. The traditional method relies on manual comparison of alarm information and fault sheets, which is easily affected by human factors, resulting in false positives or false negatives. However, the method uses a pre-trained language model to perform deep semantic analysis on the early warning information, alarm information and fault sheets, and combines keyword extraction, topic modeling and synonym rewriting data enhancement methods to accurately capture the semantic association between texts. At the same time, the knowledge graph reasoning module constructs a domain graph covering radar components, fault types and symptoms, and through entity alignment and path similarity calculation, the understanding of implicit logical relationships is strengthened. This makes the matching process from word similarity to semantic structure similarity, effectively reducing false positives and false negatives, and improving the reliability and accuracy of the early warning results.

[0016] By automating the manual verification process, the efficiency and quality of the fault diagnosis and processing flow are significantly optimized. The prior art relies on manual verification, which is inefficient and prone to errors. The application uses an intelligent self-checking mechanism to automatically perform fault data queries and semantic matching within a specific time window after the early warning is generated. Specifically, the system retrieves alarm information and fault sheets through a historical record interface without human intervention; then, a comprehensive weighted scoring algorithm combines text semantic matching scores and knowledge graph reasoning similarity to generate a confidence score. If the score exceeds a predetermined threshold, the system automatically marks it as a high-confidence risk warning and pushes it to the operation and maintenance platform. This reduces the time and cost of manual verification, improves the automation level of the diagnosis process, and avoids human errors, ensuring the continuity and timeliness of the processing flow to some extent.

[0017] The application significantly enhances the efficiency and overall reliability of system cooperation through a time window mechanism and intelligent verification. The traditional method lacks a real-time verification mechanism, resulting in a disconnection between early warnings and fault verification. The method sets a time window after the early warning is generated and automatically triggers queries and matching after the window ends, ensuring real-time alignment of early warning information and actual faults. This not only improves the timeliness of fault response, but also supplements the limitations of cross-semantic expressions through the knowledge graph reasoning module. The graph embedding algorithm calculates the path similarity between entities to handle low-frequency anomalies and complex fault scenarios, enhancing the generalization ability of the system. This mechanism optimizes the coordination of early warning generation and verification, improving the operational stability of the radar system and ensuring timely handling of critical faults to some extent.

[0018] The present application improves the intelligent level and decision support capability of the radar fault early warning system as a whole. The early warning is generated by integrating machine learning algorithms such as bidirectional long and short-term memory network, self-attention mechanism network and graph neural network, and coupled with semantic-driven verification to realize closed-loop optimization. The self-learning ability of the model is enhanced, at the same time, the structured output provides clear decision basis for operation and maintenance personnel, reduces invalid early warning, and improves resource utilization. BRIEF DESCRIPTION OF DRAWINGS

[0019] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flowchart of a radar fault early warning self-verification method based on semantic driving is provided for the embodiments of the present application. Figure 2 A schematic diagram of the internal structure of a radar fault early warning self-verification device based on semantic driving is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described in detail below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0021] The embodiments of the present application provide a radar fault early warning self-verification method, device and medium based on semantic driving, to solve the technical problem of how to realize intelligent semantic matching and real-time self-verification of early warning information and fault evidence.

[0022] The technical scheme of the embodiments of the present application will be described in detail below with reference to the drawings.

[0023] Figure 1 A flowchart of a radar fault early warning self-verification method based on semantic driving is provided for the embodiments of the present application. As shown in Figure 1 The radar fault early warning self-verification method based on semantic driving provided by the embodiments of the present application specifically includes the following steps: Step 1, obtaining the running state data of the radar system based on a preset acquisition frequency; wherein the running state data includes transmission channel parameters, receiving channel parameters, antenna system parameters, environmental parameters and state parameters.

[0024] This step collects key operational parameters from the radar system to ensure comprehensive coverage of system control layer (e.g., operational instructions), signal processing layer (e.g., signal processing logic), and environmental assurance layer (e.g., environmental monitoring) to support subsequent early warning generation.

[0025] The preset acquisition frequency specifies the time interval for data acquisition (e.g., every minute or every 6 minutes) to ensure real-time performance.

[0026] Operational status data refers to the collection of various parameters generated by the radar system during operation, including transmission, reception, antenna, environmental, and status-related indicators.

[0027] Step 1.1, acquire multiple data sources of the radar system; wherein the multiple data sources include parameters of system control layer, signal processing layer, and environmental assurance layer.

[0028] Multiple data sources refer to the data input sources of different functional layers of the radar system. The system control layer involves parameters related to radar operation instructions and control logic (e.g., system switch status); the signal processing layer involves parameters related to radar signal generation and processing (e.g., signal strength); and the environmental assurance layer involves parameters related to external environmental monitoring (e.g., temperature and humidity). This ensures the diversity of data sources, covering the entire life cycle of the radar.

[0029] In a specific example, when radar system A is deployed at site B, multiple data sources are acquired: system control layer data comes from operation console instruction logs (including system startup commands), signal processing layer data comes from signal processor output (including real-time signal quality indicators), and environmental assurance layer data comes from temperature and humidity sensors installed in machine room C (including temperature readings). These data sources are integrated through a unified interface to provide a foundation for subsequent steps.

[0030] Step 1.2, deploy sensor nodes, interfaces, and log monitoring components to the radar system, and collect operational key parameters at a preset time interval.

[0031] Deploying sensor nodes refers to installing physical or software sensors (e.g., temperature sensors or software probes) on the radar hardware, interfaces refer to data communication modules (e.g., API interfaces), and log monitoring components refer to software modules used to capture system logs. The preset time interval (e.g., every 1 minute or every 6 minutes) defines the data acquisition frequency, ensuring that key parameters (e.g., peak power or fault flags) are periodically acquired.

[0032] Operational key parameters refer to parameters critical to fault early warning (e.g., transmission power).

[0033] In a specific case, radar system A deploys sensor node D on the transmitting module (for monitoring the transmitting power) and connects to the central server through interface E; log monitoring component F collects status parameters (such as system running status code) every 1 minute. When the radar is running at site B, sensor node D collects transmitting power data every 6 minutes, and interface E transmits data in real time to avoid data delay.

[0034] Step 1.3, collect the transmitting channel parameters, receiving channel parameters, antenna system parameters, environmental parameters, and status parameters; wherein the transmitting channel parameters include transmitting power, voltage, current, and pulse width, the receiving channel parameters include receiving gain, noise coefficient, and phase error, the antenna system parameters include horizontal antenna peak power, vertical antenna peak power, and direction angle, the environmental parameters include machine room temperature, humidity, and power supply status, and the status parameters include system running status code, time stamp, and fault flag.

[0035] Transmitting channel parameters affect signal transmitting performance (such as transmitting power); receiving channel parameters affect signal receiving quality (such as receiving gain); antenna system parameters involve antenna directivity and intensity (such as direction angle); environmental parameters monitor external conditions (such as humidity); and status parameters record system running status (such as fault flag). These parameters together constitute the running status data, providing input for the machine learning model.

[0036] Pulse width refers to the duration of the transmitted signal, and fault flag refers to the identification of system abnormalities.

[0037] In one case, radar system A collects environmental parameters in machine room C: temperature sensor records temperature value (such as 25°C) every 1 minute, and humidity sensor records humidity value (such as 60%). At the same time, the fault flag in the status parameter is captured by the log monitoring component (such as "fault code 001"), and the antenna system parameters such as direction angle are collected every 4 minutes (such as 30 degrees), ensuring that the data covers potential fault points comprehensively.

[0038] In a specific case, the collected data can refer to the following table (radar system parameter collection table):

[0039] Step 1.4, structured storage of the multiple data sources, running key parameters, transmitting channel parameters, receiving channel parameters, antenna system parameters, environmental parameters, and status parameters.

[0040] Structured storage refers to organizing data into database or file formats (such as JSON or SQL table) for easy access later. This includes integrating all collected data sources and parameters, ensuring data standardization (such as uniform time stamp format).

[0041] Structured means data is stored in a predefined schema, avoiding disorganization.

[0042] In a specific example, radar system A stores all parameters (e.g., transmit power and temperature values) at site B as a structured database table: each row contains a timestamp, parameter type, and value. For example, transmit channel parameters are stored as "transmit power: 100W, time: 2023-01-01 10:00" for quick querying by machine learning algorithms.

[0043] Step 2, process the operating status data based on the trained machine learning algorithm to generate early warning information; wherein the early warning information includes a predicted failure category and a predicted occurrence time window.

[0044] Analyze the operating status data using the trained machine learning model to generate early warning information. The predicted failure category refers to the type of failure that may occur (such as power abnormalities), and the predicted occurrence time window refers to the time period when the failure may occur (such as the next 30 minutes).

[0045] The trained machine learning algorithm refers to a model optimized in advance using historical data, such as a neural network.

[0046] Step 2.1, build a heterogeneous model cluster containing bidirectional long short-term memory network, self-attention mechanism network, and graph neural network.

[0047] Building refers to creating a model architecture. Bidirectional long short-term memory network (Bi-LSTM) is good at processing time series data (such as capturing long-term dependencies); self-attention mechanism network (such as Transformer) strengthens key time point features; graph neural network (GNN) models the topology relationship between components (such as the connection between antennas and transmitting modules).

[0048] Heterogeneous model cluster refers to the combination of multiple models to improve robustness.

[0049] In a specific example, radar system A deploys a heterogeneous model cluster: Bi-LSTM processes transmit power time series data, Transformer extracts temperature peak features, and GNN models the relationship between antennas and receiving modules. When running at site B, the cluster integrates the output and predicts "power abnormality" failure.

[0050] Step 2.2, process time series type operating status data based on the bidirectional long short-term memory network to capture long-term time series dependencies.

[0051] Processing time series data refers to analyzing data sequences that change over time (such as power trends). Bi-LSTM captures long-term dependencies (such as patterns over the past few hours) and avoids the influence of short-term noise.

[0052] Time series dependency refers to the temporal correlation between data points.

[0053] In one specific example, radar system A uses Bi-LSTM to analyze temperature data sequence of machine room C: the model identifies a 3-hour temperature rising trend, captures long-term dependencies, and predicts a "temperature rising too fast" fault.

[0054] Step 2.3, extract key timing features based on the self-attention mechanism network, and strengthen the parameter correlation of important time points.

[0055] The self-attention mechanism network focuses on key time points (such as peak occurrence time) and strengthens parameter correlation (such as the relationship between temperature and power).

[0056] Key timing features refer to the time point features that have the greatest impact on fault prediction.

[0057] In one specific example, radar system A applies self-attention mechanism at site B: when the transmit power suddenly drops at a certain time point, the network strengthens the feature at that time, identifies the "power degradation" risk, and improves the accuracy of early warning.

[0058] Step 2.4, based on the graph neural network, model the topological relationship between the components of the radar system to process structured operational state data.

[0059] Graph neural network (GNN) treats radar components as nodes (such as antenna nodes) and edges represent relationships (such as connection relationships) to process structured data (such as component interactions).

[0060] Topological relationship refers to the physical or logical connection between components.

[0061] In one specific example, radar system A uses GNN to model the topology of the antenna module and the receiving module: when the receiving gain is abnormal, GNN infers antenna problems and processes structured data to predict "signal distortion".

[0062] Step 2.5, integrate the output results of the bidirectional long short-term memory network, self-attention mechanism network and graph neural network to generate structured early warning information; wherein the early warning information includes temperature rising too fast, power supply abnormal, humidity rising and power degradation.

[0063] Integration refers to the fusion of model outputs to generate unified early warning information. Structured format facilitates subsequent processing (such as including fault categories and time windows).

[0064] Early warning information refers to the prediction result (such as "humidity rising will occur within the next 1 hour").

[0065] In one specific example, radar system A integrates model outputs at site B: time-series analysis of Bi-LSTM, peak features of Transformer, and topology reasoning of GNN, generating structured warning "power abnormal, predicted time window: 14:00-14:30" Step 3. After the end of the predicted time window, call the historical record interface to query alarm information and fault ticket data within the time window. The alarm information is generated by the radar built-in test system based on rule thresholds, and the fault ticket is the actual fault description filled out by the station personnel.

[0066] This step retrieves historical data after the warning time window. Alarm information is generated by the radar built-in test (BIT) system (e.g., based on threshold rules), and fault tickets are manually filled out fault details (e.g., description text).

[0067] The historical record interface refers to the data query API.

[0068] Step 3.1. Set the query time window of the historical record interface and align it with the occurrence time window of the warning information.

[0069] Setting refers to configuring the query time range (e.g., the period after the end of the window), and aligning with the warning time ensures data matching (e.g., querying data from 14:00 to 14:30).

[0070] Alignment refers to the consistency of the time range.

[0071] In one specific example, radar system A sets the query window to "14:00-14:30" after the end of the warning time window "14:00-14:30", and calls the interface to query data, ensuring alignment with the warning.

[0072] Step 3.2. When the predicted occurrence time window end event is detected, call the historical record interface.

[0073] Detecting events refers to system monitoring the end of the time window (e.g., timer triggering) to automatically call the interface to query data.

[0074] End event refers to the termination signal of the time window.

[0075] In one specific example, radar system A detects the "14:30" end event at site B, automatically calls the historical record interface, and retrieves alarm information from the database, avoiding manual intervention.

[0076] Step 3.3. Obtain the alarm information of the radar system, and extract the fault ticket text from the preset radar station operation and maintenance system.

[0077] Acquiring alarm information refers to reading BIT system output (such as "high temperature alarm"), and extracting fault single text refers to obtaining manual description from an operation and maintenance system (such as "antenna fault description").

[0078] In a specific example, radar system A acquires alarm information (such as "excessive humidity") from machine room C and extracts fault single text (such as "humidity sensor failure") from station operation and maintenance system D to provide original data for semantic analysis.

[0079] Step 4, performing semantic analysis and matching on the early warning information, the alarm information, and the fault single to determine a semantic matching result; wherein the semantic analysis and matching include semantic similarity calculation and knowledge graph reasoning.

[0080] This step analyzes and matches text through natural language processing technology. Semantic analysis extracts keywords and topics and matches to calculate similarity.

[0081] The semantic matching result refers to the correlation score of the early warning and the actual fault.

[0082] Step 4.1, performing basic semantic analysis on the early warning information, the alarm information, and the fault single to obtain word segmentation results, part-of-speech tagging, and named entities.

[0083] Basic semantic analysis includes word segmentation (such as segmenting text into words), part-of-speech tagging (such as identifying nouns and verbs), and named entity recognition (such as extracting fault component names).

[0084] In a specific example, radar system A analyzes fault single text "transmitter temperature too high": the word segmentation result is "transmitter / temperature / over high", the part-of-speech tagging is "noun / noun / adjective", and the named entity recognition is "transmitter (component)".

[0085] Step 4.2, extracting a keyword set of the early warning information, the alarm information, and the fault single based on a pre-trained language model.

[0086] The pre-trained language model (such as BERT) extracts high-weight keywords (such as "power drop").

[0087] The keyword set refers to a core term list (such as "transmitter, temperature").

[0088] In a specific example, radar system A uses a pre-trained model to process alarm information "power jitter" and extracts a keyword set {"power", "jitter"} for subsequent topic mapping.

[0089] Step 4.3, mapping the keyword set to a multi-dimensional topic vector based on a pre-set topic modeling algorithm to generate a semantic topic distribution.

[0090] Topic modeling algorithms (e.g., LDA) map keywords to topic vectors (e.g., "thermal environment anomaly" topic), generating probability distributions.

[0091] Semantic topic distribution refers to the representation of text in the topic space.

[0092] In a specific example, radar system A applies a topic modeling algorithm: the keyword set {"temperature", "rise"} maps to the topic vector "thermal environment anomaly", generating a distribution like "thermal environment anomaly: 0.8, others: 0.2".

[0093] Step 4.4, synonym paraphrasing and sentence transformation are performed on the pre-warning information, the warning information, and the fault ticket to generate an enhanced text set.

[0094] Synonym paraphrasing refers to replacing synonyms (e.g., "failure" to "malfunction"), and sentence transformation refers to changing sentence structures (e.g., active to passive), enhancing text diversity and improving matching robustness.

[0095] The enhanced text set refers to the expanded text collection.

[0096] In a specific example, radar system A synonym paraphrases the fault ticket "power failure" as "power interruption" and sentence transforms it to "the interruption occurred at the power supply", generating an enhanced text set covering more semantics.

[0097] Step 4.5, calculate the semantic similarity between the enhanced text set and the preset historical fault cases, and select the maximum value as the text semantic matching score.

[0098] Semantic similarity calculation compares text embeddings (e.g., cosine similarity), and historical fault cases are known fault libraries. The maximum value is taken as the score to ensure the best match.

[0099] The text semantic matching score refers to the similarity value (0-1).

[0100] Step 4.5.1, call a pre-trained language model to vector encode the enhanced text set and historical fault cases, generating text embedding representations.

[0101] Vector encoding converts text into numerical vectors (embedding representations) for machine processing of similarity calculations. Pre-trained language models (e.g., BERT) learn text semantics, generating high-dimensional vectors (e.g., 768-dimensional vectors) that capture deep relationships between words.

[0102] In a specific example, radar system A calls a language model to encode the enhanced text "power interruption" into a vector and compare it with the historical case "power failure" vector.

[0103] Step 4.5.2, Calculate the similarity score of each pair of text embedding representations by the cosine similarity algorithm.

[0104] The cosine similarity algorithm calculates the cosine value of the angle between two vectors, outputting a similarity score (0-1). The closer the value is to 1, the more similar the semantics (e.g., 0.9 is high similarity). This algorithm efficiently handles high-dimensional data and avoids length effects.

[0105] The similarity score refers to the matching value.

[0106] In a specific example, Radar System A calculates the cosine similarity of the "power outage" and "power failure" vectors, with a score of 0.85.

[0107] Step 4.5.3, Traverse all combinations of augmented texts and historical fault cases, and record the maximum similarity score.

[0108] Traverse refers to iterating through all possible pairs (e.g., each augmented text and each historical case), comparing and recording the highest score. This ensures coverage of diverse semantic expressions and avoids missing the best match.

[0109] In a specific example, Radar System A traverses the augmented text set and 10 historical cases, recording a maximum score of 0.9 for "power anomaly".

[0110] Step 4.5.4, Output the maximum similarity score as the text semantic matching score.

[0111] The output score is part of the semantic matching result.

[0112] In a specific example, Radar System A outputs a text semantic matching score of 0.9, indicating a high match between the warning and the fault.

[0113] Step 4.6, Perform entity alignment and path reasoning on the set of keywords based on the domain knowledge graph to generate a knowledge graph reasoning similarity.

[0114] This step uses the domain knowledge graph (covering radar components, fault entities, etc.) to calculate the knowledge graph reasoning similarity. Entity alignment maps keywords to graph nodes (e.g., "power jitter" is aligned to the "parameter anomaly" node), and path reasoning calculates similarity through node relationships (e.g., "trigger" edges). This similarity complements text matching, handles implicit logical relationships, and improves robustness.

[0115] Step 4.6.1, Construct a domain knowledge graph covering radar components, parameter anomalies, fault types, and symptoms; wherein the node types of the domain knowledge graph include component names, operating parameters, fault types, and fault symptoms, and the edge relationships include trigger, cause, and attributed to.

[0116] Construction refers to creating or invoking a pre-built knowledge graph, where nodes represent entities (e.g., "antenna" as a component name) and edges represent relationships (e.g., "causes" indicating causality). This forms a structured knowledge network that supports reasoning.

[0117] In a specific example, Radar System A constructs a knowledge graph: Nodes: component names (e.g., "transmit module"), operational parameters (e.g., "power"), fault types (e.g., "power anomaly"), fault symptoms (e.g., "weak signal").

[0118] Edge relationships: transmit module "causes" power deficiency, which "leads to" weak signal.

[0119] The graph is built based on a library of fault cases in the documentation, covering common radar failure patterns.

[0120] Step 4.6.2, align the set of keywords with knowledge graph nodes to determine standardized concept nodes.

[0121] Entity alignment refers to matching keywords (e.g., "temperature rise" extracted in Step 4.2) to standardized nodes in the graph (e.g., "thermal environment anomaly"), eliminating term discrepancies. This ensures semantic consistency.

[0122] In a specific example, Radar System A performs alignment: Input keyword set { "transmitter", "temperature rise"}.

[0123] Align to nodes: transmitter corresponds to a "component name" node, and temperature rise corresponds to a "parameter anomaly" node.

[0124] Output a list of standardized nodes for reasoning.

[0125] Step 4.6.3, calculate the path similarity of the standardized concept nodes with historical fault case nodes based on a pre-set graph embedding model.

[0126] Path similarity calculates the strength of the connection between nodes in the graph (e.g., through edge path distance). The pre-set graph embedding model (e.g., ComplEx) generates node vectors, and the vector similarity is calculated.

[0127] Historical fault case nodes refer to pre-stored fault entities in the graph.

[0128] In a specific example, Radar System A calculates the similarity: the standardized node "temperature rise" is connected to the historical case node "thermal environment anomaly" through the "causes" edge, and the embedding model calculates the path similarity score of 0.75, indicating a moderate connection.

[0129] Step 4.6.4. Aggregate the scores of path similarities of multi-hop paths through entity propagation mechanism to generate knowledge graph inference similarity.

[0130] Entity propagation refers to the diffusion of similarity along the edges of the graph (e.g., from the first-hop node to the second-hop node) to aggregate the scores of multi-hop paths (e.g., two-hop or three-hop). This captures indirect associations to generate the final knowledge graph inference similarity.

[0131] Multi-hop path refers to a multi-step connection relationship.

[0132] In a specific example, Radar System A performs aggregation: Propagation from node "temperature rise" to "hot environment anomaly" (one-hop) with a score of 0.75.

[0133] Continued propagation to "cooling failure" (two-hop) with a score of 0.65.

[0134] Weighted aggregation of multi-hop scores outputs a knowledge graph inference similarity of 0.70.

[0135] Step 4.7. Determine the semantic matching result based on the knowledge graph inference similarity and the text semantic matching score.

[0136] This step combines the text semantic matching score (from Step 4.5) and the knowledge graph inference similarity (from Step 4.6) to generate the final semantic matching result. This result is a fusion value (e.g., mean or weighted value) representing the overall degree of association between the warning and the actual fault.

[0137] In a specific example, Radar System A determines the result: Input text matching score 0.9, knowledge graph similarity 0.7.

[0138] Step 5. Process the semantic matching result based on a pre-set comprehensive weighted scoring algorithm to determine the confidence score.

[0139] This step applies a weighted algorithm to fuse the semantic matching result, outputting a confidence score (0-1) to quantify the reliability of the warning. The weight coefficient (e.g., a = 0.6) adjusts the contribution of the text and the knowledge graph, and the pre-set value is optimized through cross-validation.

[0140] Step 5.1. Obtain the text semantic matching score and knowledge graph inference similarity output by semantic analysis and matching.

[0141] Obtaining refers to receiving the scores from Steps 4.5 and 4.6 as weighted inputs.

[0142] In a specific example, Radar System A obtains the scores: text semantic matching score 0.9, knowledge graph inference similarity 0.7.

[0143] Step 5.2, processing the text semantic matching score and the knowledge graph reasoning similarity based on a preset weight coefficient to determine a confidence score.

[0144] The processing refers to calculating a weighted sum.

[0145] The formula is represented by the following formula:

[0146] wherein, a is a similarity, is an input text matching score, is a knowledge graph reasoning similarity, is a confidence score The output confidence score is used for decision-making.

[0147] In one specific example, the radar system A applies a weight a = 0.6: The confidence score is 0.6 x 0.9 + 0.4 x 0.7 = 0.82.

[0148] The score indicates a high confidence risk.

[0149] Step 6, if the confidence score exceeds a preset threshold, it is marked as a high confidence risk warning and pushed to a preset operation and maintenance platform.

[0150] If the confidence score exceeds the threshold (such as 0.7), the system automatically marks the warning as "high confidence risk" and pushes it to the operation and maintenance platform (such as the monitoring system). This achieves real-time response and reduces manual intervention.

[0151] The above is a method embodiment proposed by the present application. Based on the same inventive concept, the present application embodiment also provides a radar fault warning self-checking device based on semantic driving, which has a structure as shown in Figure 2 .

[0152] Figure 2 It is an internal structure diagram of a radar fault warning self-checking device based on semantic driving provided by the present application embodiment. As shown in Figure 2 , the device comprises: at least one processor 201; and a memory 202 in communication connection with the at least one processor; wherein the memory 202 stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor 201 to enable the at least one processor 201 to: acquire running state data of the radar system based on a preset acquisition frequency; wherein the running state data comprises transmission channel parameters, reception channel parameters, antenna system parameters, environment parameters and state parameters; process the running state data based on a trained machine learning algorithm to generate early warning information; wherein the early warning information comprises a predicted fault category and a predicted occurrence time window; after the predicted occurrence time window ends, call a historical record interface to query alarm information and fault single data within the time window; wherein the alarm information is generated by a radar built-in test system based on a rule threshold, and the fault single is an actual fault description filled in by station personnel; perform semantic analysis and matching on the early warning information, the alarm information and the fault single to determine a semantic matching result; wherein the semantic analysis and matching comprises semantic similarity calculation and knowledge graph reasoning; process the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; if the confidence score exceeds a preset threshold, mark it as a high-confidence risk early warning and push it to a preset operation and maintenance platform.

[0153] Some embodiments of the present application provide a non-volatile computer storage medium based on semantic-driven radar fault early warning self-checking corresponding to Figure 1 The computer executable instructions are set to: acquire running state data of the radar system based on a preset acquisition frequency; wherein the running state data comprises transmission channel parameters, reception channel parameters, antenna system parameters, environment parameters and state parameters; process the running state data based on a trained machine learning algorithm to generate early warning information; wherein the early warning information comprises a predicted fault category and a predicted occurrence time window; after the predicted occurrence time window ends, call a historical record interface to query alarm information and fault single data within the time window; wherein the alarm information is generated by a radar built-in test system based on a rule threshold, and the fault single is an actual fault description filled in by station personnel; perform semantic analysis and matching on the early warning information, the alarm information and the fault single to determine a semantic matching result; wherein the semantic analysis and matching comprises semantic similarity calculation and knowledge graph reasoning; process the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; if the confidence score exceeds a preset threshold, mark it as a high-confidence risk early warning and push it to a preset operation and maintenance platform.

[0154] Each of the embodiments in the present application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment mainly describes the difference from other embodiments. In particular, for the Internet of Things device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiments.

[0155] The system and medium provided by the embodiments of the present application are one-to-one corresponding, and therefore, the system and medium also have similar beneficial technical effects to the method corresponding thereto. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the system and medium will not be described here again.

[0156] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. In addition, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0157] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in one or more flows and / or blocks.

[0158] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in one or more flows and / or blocks.

[0159] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in one or more flows and / or blocks.

[0160] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memories.

[0161] Memory can include non-persistent memory, Random Access Memory (RAM), and / or non-volatile memory, such as Read Only Memory (ROM) or flash memory, in computer readable media. Memory is an example of computer readable media.

[0162] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0163] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not include only those elements in the list, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0164] The above description is only some embodiments of the present application, and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the scope of claims of the present application.

Claims

1. A radar fault early warning self-checking method based on semantic driving, applied to a radar system, characterized in that, The method comprises: acquiring operating state data of a radar system based on a preset acquisition frequency; wherein the operating state data comprises transmission channel parameters, reception channel parameters, antenna system parameters, environmental parameters and state parameters; processing the operating state data based on a trained machine learning algorithm to generate early warning information; wherein the early warning information comprises a predicted failure category and a predicted occurrence time window; after the predicted occurrence time window ends, querying alarm information and fault single data within the time window by calling a historical record interface; wherein the alarm information is generated by a radar built-in test system based on a rule threshold, and the fault single is an actual failure description filled in by station personnel; performing semantic analysis and matching on the early warning information, the alarm information and the fault single to determine a semantic matching result; wherein the semantic analysis and matching comprises semantic similarity calculation and knowledge graph reasoning; processing the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; if the confidence score exceeds a preset threshold, marking it as a high-confidence risk early warning and pushing it to a preset operation and maintenance platform.

2. The method of claim 1, wherein, Acquiring operating state data of a radar system based on a preset acquisition frequency comprises: acquiring multiple data sources of the radar system; wherein the multiple data sources comprise parameters of system control layers, signal processing layers and environmental protection layers; deploying sensor nodes, interfaces and log monitoring components to the radar system and collecting operating key parameters at a preset time interval; collecting the transmission channel parameters, reception channel parameters, antenna system parameters, environmental parameters and state parameters; wherein the transmission channel parameters comprise transmission power, voltage, current and pulse width, the reception channel parameters comprise reception gain, noise coefficient and phase error, the antenna system parameters comprise horizontal antenna peak power, vertical antenna peak power and direction angle, the environmental parameters comprise machine room temperature, humidity and power supply state, and the state parameters comprise system operating state code, time stamp and fault flag; structurally storing the multiple data sources, operating key parameters, transmission channel parameters, reception channel parameters, antenna system parameters, environmental parameters and state parameters.

3. The method of claim 1, wherein the method further comprises: Processing the operating state data based on a trained machine learning algorithm to generate early warning information comprises: building a heterogeneous model cluster comprising a bidirectional long short-term memory network, a self-attention mechanism network and a graph neural network; processing time-series operating state data based on the bidirectional long short-term memory network to capture long-term time-series dependency relationships; extracting key time-series features based on the self-attention mechanism network to strengthen parameter correlations of important time points; modeling topological relationships between components of the radar system based on the graph neural network to process structured operating state data; integrating output results of the bidirectional long short-term memory network, the self-attention mechanism network and the graph neural network to generate structured early warning information; wherein the early warning information comprises rapid temperature rise, power supply anomaly, humidity rise and power degradation.

4. The method of claim 1, wherein the method further comprises: After the predicted occurrence time window ends, querying alarm information and fault single data within the time window by calling a historical record interface comprises: Set the query time window of the history record interface and align it with the occurrence time window of the early warning information; Call the history record interface when the end event of the prediction occurrence time window is detected; Obtain the alarm information of the radar system, and extract a fault single text from a preset radar station operation and maintenance system.

5. The method of claim 1, wherein, Perform semantic analysis and matching on the early warning information, the alarm information, and the fault single to determine a semantic matching result, specifically including: Perform basic semantic analysis on the early warning information, the alarm information, and the fault single to obtain a word segmentation result, a part-of-speech tagging, and a named entity; Extract a keyword set of the early warning information, the alarm information, and the fault single based on a pre-trained language model; Map the keyword set to a multi-dimensional topic vector based on a preset topic modeling algorithm to generate a semantic topic distribution; Perform synonym rewriting and sentence conversion on the early warning information, the alarm information, and the fault single to generate an enhanced text set; Calculate the semantic similarity between the enhanced text set and a preset historical fault case, and select the maximum value as a text semantic matching score; Perform entity alignment and path reasoning on the keyword set based on a domain knowledge graph to generate a knowledge graph reasoning similarity; Determine the semantic matching result based on the knowledge graph reasoning similarity and the text semantic matching score.

6. The method of claim 5, wherein the method further comprises: Calculate the semantic similarity between the enhanced text set and a historical fault case, and select the maximum value as a text semantic matching score, specifically including: Call a pre-trained language model to perform vector encoding on the enhanced text set and the historical fault case to generate a text embedding representation; Calculate the similarity score of each pair of text embedding representations through a cosine similarity algorithm; Iterate through all combinations of the enhanced text and the historical fault case, and record the maximum similarity score; Output the maximum similarity score as the text semantic matching score.

7. The method of claim 5, wherein the method further comprises: Perform entity alignment and path reasoning on the keyword set based on a domain knowledge graph to generate a knowledge graph reasoning similarity, specifically including: Construct a domain knowledge graph covering radar components, parameter abnormalities, fault types, and symptoms; wherein the node types of the domain knowledge graph include component names, operating parameters, fault types, and fault symptoms, and the edge relationships include triggering, causing, and being attributed to; Align the keyword set with the knowledge graph nodes to determine standardized concept nodes; Calculate the path similarity of the standardized concept nodes and the historical fault case nodes based on a preset graph embedding model; Aggregate the path similarity scores of multi-hop paths through an entity propagation mechanism to generate the knowledge graph reasoning similarity.

8. The method of claim 1, wherein the method further comprises: Process the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score, specifically including: Obtain the text semantic matching score and the knowledge graph reasoning similarity output by the semantic analysis and matching; Process the text semantic matching score and the knowledge graph reasoning similarity based on a preset weight coefficient to determine the confidence score.

9. A semantic-driven based radar fault early warning self-checking device, characterized in that, The device includes: at least one processor; and a memory connected in communication with the at least one processor; The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: acquire running state data of a radar system based on a preset acquisition frequency; wherein the running state data includes transmission channel parameters, reception channel parameters, antenna system parameters, environment parameters, and state parameters; process the running state data based on a trained machine learning algorithm to generate early warning information; wherein the early warning information includes a predicted fault category and a predicted occurrence time window; after the predicted occurrence time window ends, call a historical record interface to query alarm information and fault single data within the time window; wherein the alarm information is generated by a radar built-in test system based on a rule threshold, and the fault single is an actual fault description filled in by a station personnel; perform semantic analysis and matching on the early warning information, the alarm information, and the fault single to determine a semantic matching result; wherein the semantic analysis and matching include semantic similarity calculation and knowledge graph reasoning; process the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; if the confidence score exceeds a preset threshold, mark it as a high-confidence risk early warning and push it to a preset operation and maintenance platform.

10. A non-volatile computer storage medium for self-verification of radar fault early warning based on semantics, storing computer-executable instructions, characterized in that, The computer executable instructions are configured to: acquire running state data of a radar system based on a preset acquisition frequency; wherein the running state data includes transmission channel parameters, reception channel parameters, antenna system parameters, environment parameters, and state parameters; process the running state data based on a trained machine learning algorithm to generate early warning information; wherein the early warning information includes a predicted fault category and a predicted occurrence time window; after the predicted occurrence time window ends, call a historical record interface to query alarm information and fault single data within the time window; wherein the alarm information is generated by a radar built-in test system based on a rule threshold, and the fault single is an actual fault description filled in by a station personnel; perform semantic analysis and matching on the early warning information, the alarm information, and the fault single to determine a semantic matching result; wherein the semantic analysis and matching include semantic similarity calculation and knowledge graph reasoning; process the semantic matching result based on a preset comprehensive weighted scoring algorithm to determine a confidence score; if the confidence score exceeds a preset threshold, mark it as a high-confidence risk early warning and push it to a preset operation and maintenance platform.

Images