Train-ground communication quality prediction method and system based on large model and intelligent workflow

Abstract

The invention discloses a vehicle-ground communication quality prediction method and system based on a large model and intelligent workflow, and the method comprises the steps: S1, receiving an original communication message file uploaded by a user through a data access module, and extracting a structured data signal from the message file through a document analysis module; s2, searching the structured data signal from a database through a code execution module, and generating an effective abnormal event signal; s3, inputting the effective abnormal event signal and a communication quality statistical data signal derived from the structured data signal into a large language model service; s4, integrating the professional analysis text signal and the visual chart signal to a predefined report template through a report rendering module, and generating an analysis report signal; and S5, presenting the analysis report signal to a user through a user interface module. According to the train-ground communication quality prediction method and system based on the large model and the intelligent workflow, the problems that the train-ground communication quality analysis is high in false alarm rate, low in automation degree and lack of deep insight can be solved.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology for rail transit, specifically to a method and system for predicting vehicle-to-ground communication quality based on large models and intelligent workflows. Background Technology

[0002] In rail transit systems, vehicle-to-ground communication is crucial for train control, dispatching, and passenger information services. During operation, onboard equipment continuously generates communication messages containing key indicators such as reference signal received power and signal-to-noise ratio. Effective analysis of this message data is essential for assessing the health of the communication network and locating potential faults. However, current industry-standard analysis methods have significant shortcomings. First, in anomaly detection, traditional methods often rely on setting simple parameter thresholds for judgment; an alarm is triggered once the measured value exceeds the threshold. This method is highly susceptible to instantaneous signal fluctuations. For example, when a train passes through tunnels, curves, or experiences brief interference from external electromagnetic environments, numerous invalid alarms are generated, resulting in a high false alarm rate. This forces maintenance personnel to expend considerable effort to distinguish between true and false faults, leading to low work efficiency and the potential to overlook genuine, persistent problems.

[0003] Secondly, existing analysis processes have low levels of automation and intelligence, typically requiring operations personnel to manually operate multiple independent software tools to complete a series of steps such as data extraction, format conversion, chart creation, and report writing. This fragmented operation mode is not only cumbersome and time-consuming, but also heavily reliant on the personal experience of operators, making it difficult to guarantee the consistency and professionalism of the analysis results. Finally, existing technical solutions have relatively limited functionality, often outputting simple statistical tables or basic charts, lacking in-depth interpretation of the business significance behind the data. While operations personnel can see "what happened," they lack effective intelligent assistance in understanding "why it happened" and "how to respond." The system cannot provide causal inferences and operational suggestions based on domain knowledge, preventing the full realization of the value of data analysis and hindering its direct transformation into efficient operational decisions. Therefore, the industry urgently needs an intelligent analysis solution that can accurately identify anomalies, automate the entire process, and provide deep business insights. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method and system for predicting vehicle-to-ground communication quality based on a large model and intelligent workflow, to solve the problems of high false alarm rate, low automation, and lack of in-depth insight in vehicle-to-ground communication quality analysis. This invention constructs an intelligent workflow, first parsing unstructured communication messages into structured data; then, it employs an anomaly detection algorithm based on duration determination to filter transient interference and accurately capture real faults; finally, the analysis results are input into a large language model to automatically generate a comprehensive report combining data, charts, and professional causal analysis. This achieves fully automated and intelligent analysis from raw data to in-depth insights, effectively improving operational efficiency and decision support capabilities.

[0005] This invention provides a method for predicting vehicle-to-ground communication quality based on large models and intelligent workflows, including: S1: Receives raw communication message files uploaded by users through the data access module, and extracts structured data signals from the message files using the document parsing module; S2: The code execution module retrieves structured data signals from the database, runs an anomaly detection algorithm, and uses preset parameter threshold signals and continuous anomaly time window signals to traverse structured data signals to identify potential anomaly signals. Based on the persistence judgment logic, the algorithm filters potential anomaly signals and generates valid anomaly event signals. S3: Input valid abnormal event signals and communication quality statistics signals derived from structured data signals into the large language model service, generate professional analysis text signals through structured prompt words, and automatically generate visualization chart signals based on communication quality statistics signals by calling the chart library through the code execution module; S4: The report rendering module integrates professional analytical text signals and visual chart signals into a predefined report template to generate analytical report signals; S5: Presents analysis report signals to users through the user interface module.

[0006] In one embodiment of the present invention, step S2, running the anomaly detection algorithm specifically includes: S201 parameter initialization step, setting a reference signal received power threshold signal and a signal-to-noise ratio threshold signal for determining communication quality anomalies, and setting a continuous anomaly time window signal for distinguishing between transient interference and persistent faults; S202 data traversal and anomaly point identification step, traversing each data point in the structured data signal in chronological order, and identifying data points that simultaneously satisfy the condition that their reference signal received power value is lower than the reference signal received power threshold signal and their signal-to-noise ratio value is lower than the signal-to-noise ratio threshold signal as potential anomaly point signals; S203 persistence determination step, checking whether the potential anomaly point signals appear continuously in the time series, and calculating the total duration of their continuous appearance, only when the total duration exceeds the continuous anomaly time window signal, is the corresponding potential anomaly point signal sequence determined and generated as a valid anomaly event signal; S204 recording and storage step, writing the detailed information of the valid anomaly event signal into the database for persistent storage.

[0007] In one embodiment of the present invention, in step S2, the anomaly detection algorithm is executed by the code execution node called by the intelligent workflow engine. The intelligent workflow engine provides a visual configuration interface, allowing users to arrange the entire analysis process, including the code execution node, by dragging and dropping, and dynamically adjust the values ​​of the parameter threshold signal and the continuous anomaly time window signal through the graphical interface, without modifying the underlying program code of the anomaly detection algorithm. This enables flexible configuration and adaptive adjustment of the analysis process and judgment rules to meet the differentiated analysis needs of different lines or scenarios.

[0008] In one embodiment of the present invention, in step S2, filtering potential anomaly signals based on the persistence determination logic employs a sliding window technique to monitor the continuity of the abnormal state in real time. By checking whether the distribution of potential anomaly signals on the time axis constitutes a continuous segment whose duration meets the requirements of the continuous anomaly time window signal, isolated anomaly signals caused by instantaneous environmental disturbances can be effectively distinguished and filtered, ensuring that the final generated valid anomaly event signal represents a real, continuous communication failure with operational value rather than a brief signal fluctuation.

[0009] In one embodiment of the present invention, step S3, inputting the valid abnormal event signal and communication quality statistical data signal into the large language model service, specifically includes: S301, a prompt word configuration step, configuring a structured prompt word for the large language model service, which instructs the large language model to operate from the perspective and professional terminology of rail transit communication experts; S302, a signal analysis and reasoning step, in which the large language model performs a comprehensive analysis of the input valid abnormal event signal and communication quality statistical data signal based on the prompt word, including a summary of the overall communication status, pattern recognition and severity assessment of the valid abnormal event signal, and professional inference of the possible causes of the abnormality; and S303, a text generation step, ultimately generating a professional analysis text signal containing specific operation and maintenance suggestions, with standardized language and meeting the requirements of a professional report.

[0010] In one embodiment of the present invention, in step S3, the automatic generation of visual chart signals by calling the chart library through the code execution module specifically refers to the data statistics step in step S304, which calculates the distribution ratio of communication quality parameters in different numerical ranges derived from the structured data signal; and the chart rendering step in step S305, which uses the chart generation library to programmatically create a bar chart signal based on the distribution ratio statistical data signal to intuitively display the distribution of reference signal received power and signal-to-noise ratio. This chart signal reveals the overall distribution characteristics and statistical laws of communication quality in a graphical way, serving as a visual supplement and support to the professional analysis text signal.

[0011] In one embodiment of the present invention, in step S3, the system supports configuring and accessing a variety of different large language model services. Users can flexibly select and switch the large language model service instance used in the intelligent workflow engine according to their specific requirements for report generation quality or cost considerations.

[0012] In one embodiment of the present invention, step S4, which integrates the professional analysis text signal and the visualization chart signal into the predefined report template through the report rendering module, means that the system reads a report document template with a pre-designed format and placeholders, then automatically fills the corresponding text area of ​​the template with the professional analysis text signal generated by the large language model service, and embeds the generated visualization chart signal into the specified illustration position of the template, and finally automatically synthesizes a complete analysis report signal containing structured text description and matching visualization chart.

[0013] In one embodiment of the present invention, the method further includes step S101 before step S2, namely the message data uploading and structured parsing step, which specifically involves: receiving the original communication message file uploaded by the user through the user interface provided by the system, then calling the embedded document parsing module to parse the unstructured message according to the predefined communication protocol format, extracting key fields such as timestamp, train number, serving cell identifier, reference signal received power and signal-to-noise ratio, and forming structured data signals, and finally persistently storing this batch of structured data signals in a relational database for use in subsequent analysis steps.

[0014] This invention also includes a vehicle-to-ground communication quality prediction system based on large models and intelligent workflows, comprising: The parsing module receives the raw communication message file uploaded by the user through the data access module, and uses the document parsing module to extract structured data signals from the message file; The traversal module retrieves structured data signals from the database through the code execution module, runs an anomaly detection algorithm, and uses preset parameter threshold signals and continuous anomaly time window signals to traverse the structured data signals to identify potential anomaly signals. Based on the persistence judgment logic, it filters the potential anomaly signals and generates valid anomaly event signals. The execution module inputs valid abnormal event signals and communication quality statistics signals derived from structured data signals into the large language model service. It generates professional analysis text signals through structured prompt words and automatically generates visualization chart signals based on communication quality statistics signals by calling the chart library through the code execution module. The analysis module integrates professional analytical text signals and visual chart signals into a predefined report template through the report rendering module to generate analytical report signals; The presentation module presents the analysis report signals to the user through the user interface module.

[0015] The present invention provides a vehicle-to-ground communication quality prediction method and system based on a large model and intelligent workflow. By constructing an intelligent workflow, it first parses unstructured communication messages into structured data; then, it uses an anomaly detection algorithm based on duration determination to filter transient interference and accurately capture real faults; finally, it inputs the analysis results into a large language model to automatically generate a comprehensive report combining data, charts, and professional causal analysis. This achieves fully automated and intelligent analysis from raw data to in-depth insights, effectively improving operation and maintenance efficiency and decision support capabilities. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 The flowchart shows the method for predicting vehicle-to-ground communication quality based on large models and intelligent workflows. Figure 2 A flowchart illustrating the workflow of step S2; Figure 3 A flowchart illustrating the workflow of step S3; Figure 4 This is a system architecture diagram of a vehicle-to-ground communication quality prediction system based on a large model and intelligent workflow. Detailed Implementation

[0018] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0019] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0020] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0021] Please see Figure 1-4The image shows the vehicle-to-ground communication quality prediction method and system based on a large model and intelligent workflow according to the present invention. The vehicle-to-ground communication quality prediction method based on a large model and intelligent workflow of the present invention includes: S1: Receiving the original communication message file uploaded by the user through the data access module, and extracting structured data signals from the message file using the document parsing module; S2: Retrieving the structured data signals from the database through the code execution module, running an anomaly detection algorithm, wherein the anomaly detection algorithm uses preset parameter threshold signals and a continuous anomaly time window signal to traverse the structured data signals to identify potential anomaly point signals, and filters the potential anomaly point signals based on the persistence judgment logic to generate valid anomaly event signals; S3: Inputting the valid anomaly event signals and the communication quality statistical data signals derived from the structured data signals into the large language model service, generating professional analysis text signals through structured prompt word instructions, and simultaneously calling the chart library through the code execution module to automatically generate visual chart signals based on the communication quality statistical data signals; S4: Integrating the professional analysis text signals and visual chart signals into a predefined report template through the report rendering module to generate an analysis report signal; S5: Presenting the analysis report signal to the user through the user interface module.

[0022] like Figure 1As shown, this invention provides a vehicle-to-ground communication quality prediction method based on a large model and intelligent workflow. Its core lies in constructing a highly automated and intelligent analysis pipeline that transforms raw, messy communication data into an in-depth analysis report with direct operational guidance value. The entire process begins with data access and preliminary processing. Specifically, a dedicated data access module receives raw communication message files uploaded by users via a web interface or other means, generated by devices such as vehicle-mounted TAUs during operation. These files are typically text files with extensions like .log or .txt, containing massive amounts of unstructured or semi-structured data streams recorded according to specific communication protocol specifications. After successfully receiving these files, the system does not process them immediately but instead calls an embedded, specially optimized document parsing module. The core mission of this module is to act as a skilled "data translator." Pre-programmed with the specific format rules of the target communication protocol, it can automatically and accurately locate, extract, and parse the key structured data elements relied upon for subsequent analysis from the seemingly chaotic sea of ​​message fields, much like the human eye identifies specific information. These extracted data signals form the cornerstone of the entire analysis process. They include at least millisecond-accurate timestamps, recording the exact moment each data sample was generated; unique train number signals, identifying which specific operating train the data originated from; serving cell identifier signals, indicating which network base station or access point the train was communicating with at that moment; and, most critically, direct metrics of communication quality—reference signal received power and signal-to-noise ratio (SNR) signals, both core physical layer parameters for evaluating wireless link connection stability and signal purity. Once this key information is successfully captured and transformed into a unified structured format, it is systematically and securely persisted in a relational database, such as MySQL (a relational database management system). This step marks the first transformation of the raw data, from raw messages that are difficult to compute directly into structured data signals that are neat, uniform, and efficiently queried and processed by programs, providing solid "raw materials" for subsequent in-depth analysis. After obtaining high-quality structured data signals, the next step of the method is the core diagnostic phase, which involves using a code execution module to drive a sophisticated anomaly detection algorithm. This algorithm aims to automatically sift through massive amounts of data to identify anomalous segments that indicate communication quality degradation. This code execution module can be an integrated Python scripting environment or other types of sandboxed computing containers. It is responsible for retrieving all relevant structured data signals within the analyzed time period from the database as needed. Subsequently, it starts and runs the pre-built anomaly detection algorithm that encapsulates the core diagnostic logic. The execution of this algorithm is not a one-step process but a series of sophisticated steps.The first step is parameter initialization, where the system sets a series of crucial decision thresholds for the algorithm, known as threshold parameter signals. These signals are not fixed but can be flexibly configured based on different line environments, equipment types, or maintenance standards. They primarily include a reference signal received power threshold signal to determine if the signal strength has attenuated to an unacceptable level; a signal-to-noise ratio threshold signal to determine if the signal quality is severely contaminated by noise; and a continuous anomaly time window signal, which plays a key role in this method and distinguishes between transient fluctuations and actual faults. The introduction of this time window signal is one of the core design features that improves the detection accuracy of this method.

[0023] Further, the algorithm proceeds to the data traversal and anomaly identification steps. In this step, the system meticulously scans the entire sequence of retrieved structured data signals in chronological order. For each data point in the sequence, the algorithm rigorously compares its reference signal received power and signal-to-noise ratio (SNR) values ​​with the two corresponding threshold signals set in the initialization step. Only when a data point simultaneously meets the conditions of having a reference signal received power value lower than the set reference signal received power threshold signal and an SNR value lower than the set SNR threshold signal will this data point be temporarily and cautiously marked by the algorithm as a potential anomaly signal. The reason it is "potential" is that in complex real-world operating environments, there are numerous instantaneous and random interferences that may cause temporary signal degradation, but these interferences do not represent a substantial fault in the equipment or network requiring maintenance intervention. If an alarm is triggered based solely on a single data point falling below the threshold, a massive number of invalid alarms will be generated, severely interfering with the judgment of maintenance personnel. Therefore, the core continuous judgment step is then initiated. The task of this step is to "judge" all potential anomaly signals captured in the previous step. Instead of viewing each anomaly in isolation, the algorithm examines their distribution patterns over time. It checks whether these potential anomalies occur consecutively in time, forming one or more consecutive anomaly segments, and precisely calculates the total duration of each segment. This calculated total duration is then compared to a pre-defined duration window signal. Only when the duration of a consecutive anomaly segment clearly exceeds the lower limit specified by this window signal will the algorithm officially classify it as a confirmed and valid anomaly signal. This decision-making logic is crucial; it effectively filters out fleeting interference signals, ensuring that only communication quality events that persist for a certain period and are more likely to represent real equipment vulnerabilities or network problems are captured, thus significantly reducing the system's false alarm rate. Finally, in the recording and storage step, all key information about these officially confirmed valid anomaly signals is meticulously recorded by the system. This information typically includes the start and end times of the anomaly, its total duration, the associated train's identity, the worst received reference signal power, and the worst signal-to-noise ratio recorded during the event. All this detailed information is systematically written into a dedicated anomaly table in the database, forming complete historical fault records. This step not only provides direct input for current report generation but also accumulates valuable assets for future historical data queries, trend analysis, and equipment health status assessments.

[0024] In one embodiment of this invention, intelligent extraction from "data" to "events" is demonstrated. Through parameterized threshold judgment, rigorous time-series traversal, and fault-tolerant continuous logic determination, it achieves automated and precise diagnosis of massive communication data, completely freeing maintenance personnel from tedious manual data screening and significantly improving the accuracy of fault identification. The execution of this sophisticated algorithm is not fixed in rigid code; it is integrated within a larger, more flexible intelligent workflow engine framework. Specifically, the anomaly detection algorithm is carried out and run by a code execution node called by this intelligent workflow engine. This intelligent workflow engine is typically built on low-code or no-code platforms like Dify, serving as the brain and central nervous system of the entire system, presented to the end user through a visual graphical interface. On this interface, users can see nodes representing different processing functions, such as data access nodes, document parsing nodes, code execution nodes, etc. Users do not need extensive programming knowledge; they can connect these nodes through intuitive drag-and-drop, arranging them into a complete data processing pipeline, i.e., a workflow, that meets specific analytical needs. Most importantly, users can dynamically adjust key parameters of the anomaly detection algorithm embedded in the code execution node directly through this graphical interface. Whether they feel the criteria for judging the reference signal received power needs to be tightened or loosened, or that the time window for persistent anomalies needs to be modified based on new operational experience, users can directly modify the corresponding numerical configurations on the interface, and the workflow engine will automatically apply these new parameters on the next execution. All these adjustments occur at the application level, without needing to touch, modify, or redeploy the underlying program code of the anomaly detection algorithm. This design gives the system unprecedented flexibility and maintainability, enabling the analysis process and judgment rules to quickly and adaptively adjust to changes in business needs. It perfectly solves the pain point of traditional customized software systems requiring significant development resources to modify once requirements change, greatly reducing the long-term operation and maintenance costs and skill threshold of the system. From data upload to intelligent parsing, and then to configurable and accurate anomaly detection, this series of steps are interconnected, forming a powerful and flexible automated analysis foundation, laying a solid data foundation for the final generation of reports with in-depth insights. After accurately capturing communication anomalies, the core advantage of this method lies in its in-depth processing and intelligent presentation of the analysis results. Specifically, the process of filtering potential anomaly signals based on persistent judgment logic relies on a dynamic monitoring mechanism called the sliding window technique. This technique can be visualized as a fixed-length time window—its size defined by the persistent anomaly time window signal—sliding uniformly across the entire data sequence along the time axis like a moving observation mirror.The data within this window constitutes a local data segment that is analyzed in real time. The algorithm's task is to continuously examine the distribution of marked potential anomaly signals along the time axis within this sliding window. It doesn't merely count the number of anomalies within the window, but more precisely determines whether these points are temporally adjacent, thus forming a continuous, uninterrupted segment of anomalies. Simultaneously, the algorithm calculates the actual time span covered by this continuous anomaly segment in real time. Only when this calculated continuous duration completely covers and exceeds the preset time threshold represented by the sliding window—the continuous anomaly time window signal—will the system finally confirm it. This mechanism can extremely effectively distinguish and filter out isolated, scattered anomaly signals caused by trains briefly crossing electromagnetic shielding zones, encountering random co-channel interference, or experiencing momentary equipment jitter. These isolated points typically have extremely short durations and cannot fill the entire observation window, thus being intelligently ignored by the system. In this way, valid abnormal event signals are ultimately generated and transmitted to downstream processes. These signals represent real communication failures or equipment performance degradation that have lasted long enough, ruled out as accidental, and have a higher confidence level. This provides the operations and maintenance team with highly targeted and reliable alarm information, avoiding the waste of resources on investigating false alarms.

[0025] like Figure 2As shown, in step S2 of the vehicle-to-ground communication quality prediction method based on a large model and intelligent workflow, the core anomaly detection step embodies a sophisticated multi-stage processing logic. This logic aims to accurately filter out truly valuable communication fault events from massive amounts of time-series data. This step begins with a crucial parameter initialization step, where the system sets a series of key threshold values ​​for subsequent judgment procedures. These threshold values ​​include a reference signal received power threshold to measure whether the wireless signal strength has attenuated to an unacceptable level, a signal-to-noise ratio threshold to assess whether the signal quality has been severely degraded by background noise, and a sustained anomaly time window that plays a decisive role in this method, distinguishing between transient fluctuations and actual faults. These thresholds are not set arbitrarily but are based on a summary of long-term operational experience with rail transit vehicle-to-ground communication systems and a deep understanding of communication protocol standards. They collectively constitute the basis for anomaly detection criteria, ensuring that subsequent analysis can be conducted under a clear and quantifiable standard. With parameter initialization complete, the process proceeds to the data traversal and anomaly identification steps. This method requires the system to systematically examine the structured data signals retrieved from the database one by one, in chronological order. For each data point, the algorithm rigorously compares its reference signal received power and signal-to-noise ratio (SNR) values ​​with the corresponding preset thresholds from the previous step. Only when a data point simultaneously meets the conditions of having a reference signal received power value lower than the set reference signal received power threshold and an SNR value lower than the set SNR threshold will the algorithm temporarily mark this data point as a "potential anomaly." This "dual condition satisfaction" design serves as the first line of defense against a large amount of irrelevant interference. It ensures that only data exhibiting problems in both signal strength and signal quality proceeds to the next stage of in-depth analysis, avoiding misjudgments caused by accidental fluctuations in a single parameter and laying a reliable data foundation for subsequent high-precision determination.

[0026] like Figure 2As shown, identifying potential anomalies is only the beginning of diagnosis. In complex real-world operating environments, isolated anomalies that meet the above dual conditions may arise from a train briefly passing through a tunnel, experiencing transient electromagnetic interference, or sporadic equipment vibrations. However, these are usually not worthwhile for maintenance intervention. Therefore, this method introduces its most innovative core step—the persistence determination step. The task of this step is to "judge" all potential anomalies captured in the preceding steps, with temporal continuity as the core criterion. The algorithm examines the distribution of these potential anomalies along the time axis, determining whether they constitute one or more sequences that occur consecutively in time, and accurately calculates the total duration covered by each such consecutive sequence. Then, this calculated total duration is compared with a preset persistence anomaly time window. Only when the duration of a certain consecutive anomaly sequence clearly exceeds this time window will the system finally confirm and formally determine the sequence from a "potential" anomaly state and output it as a "valid anomaly event." This mechanism is crucial because it cleverly leverages the fact that real equipment failures or network bottlenecks typically lead to a continuous degradation in communication quality, while transient interference is often isolated and short-lived. This allows for efficient differentiation between the two types of situations. After successfully identifying a valid anomaly, the process proceeds to the recording and storage step. At this point, the system completely records all the key attribute information of the event. This information typically includes the precise start and end times of the anomaly, the total duration, the associated specific train number, the worst reference signal received power value, and the worst signal-to-noise ratio value recorded during the event's duration. These detailed records are systematically written into a dedicated anomaly table in the database. This step not only provides direct material for the generation of the current report but, more importantly, it establishes a valuable historical fault archive for the operations and maintenance team, providing data support for subsequent trend analysis, equipment health assessment, and the development of preventative maintenance strategies.

[0027] Specifically, the entire anomaly detection algorithm described above, from parameter initialization to the recording of valid events, does not exist as a rigid, closed software module. Its execution is invoked and managed by a higher-level intelligent workflow engine. This algorithm is encapsulated within a code execution node, which is a component of the intelligent workflow engine. This engine is typically built on a low-code platform, providing end-users with an intuitive, visual configuration interface. On this graphical interface, users can see nodes representing different functional units, such as data parsing nodes, code execution nodes, and large language model nodes. Users do not need professional programming skills; they can connect these nodes through simple drag-and-drop operations to orchestrate a complete data processing pipeline that meets specific analytical needs. Crucially, users can directly and dynamically adjust the key parameters of the anomaly detection algorithm embedded within the code execution node through this user-friendly interface. Whether adjusting the threshold for judging the reference signal received power based on the characteristics of a new line, or modifying the length of the continuous anomaly time window based on accumulated operational experience, all these adjustments can be completed at the application level, without touching, modifying, or redeploying the underlying algorithm code. This design endows the system with unprecedented flexibility and maintainability, enabling the analysis strategy to quickly respond to changes in business needs and significantly reducing long-term maintenance costs and technical barriers. Further exploration of the technical implementation of the persistence determination logic reveals that its core lies in a dynamic monitoring mechanism called sliding window technology. This technology can be figuratively understood as an "observation window" of fixed length (defined by the persistence anomaly time window) that slides uniformly across the entire data sequence along the time axis. During the sliding process, the algorithm continuously examines all potential anomaly signals within the window. Its purpose is not simply counting, but rather performing more refined morphological analysis: determining whether these points are temporally adjacent, thus forming a continuous, uninterrupted segment of anomaly. Simultaneously, the algorithm calculates the actual time span covered by this continuous anomaly segment in real time. Only when the duration of this continuous segment completely covers and exceeds the length of the observation window, thus satisfying the preset persistence condition, will the system ultimately confirm a valid anomaly event. This sliding window-based continuity check mechanism can efficiently and reliably identify and filter out isolated anomalies scattered along the timeline caused by transient environmental disturbances. These isolated interference points cannot form continuous segments that meet the duration requirements and are therefore intelligently ignored by the system. In this way, the valid anomaly signals that are ultimately captured and recorded represent genuine communication failures or equipment performance bottlenecks that have lasted for a sufficiently long time, ruled out as accidental, and have high confidence. This provides the operations and maintenance team with accurate and reliable alarm information, significantly improving the effectiveness and focus of operations and maintenance work.

[0028] like Figure 3As shown, in step S3 of the vehicle-to-ground communication quality prediction method based on large models and intelligent workflows, after the system accurately captures communication anomalies, it enters a crucial stage that imbues the data with deep insights and intuitive presentation—intelligent report generation. This process is far more than a simple data listing or template filling; it is a complex, collaborative processing flow that integrates modern artificial intelligence technology with traditional data visualization. Its goal is to produce a comprehensive analysis report that combines professional depth and visual expressiveness. The process begins with data preparation and input. The system aggregates the valid anomaly signals identified by sophisticated algorithms in the previous stage, as well as the overall communication quality statistical data signals derived from massive amounts of raw communication data, as core raw materials. These statistical data signals typically include macroscopic indicators such as the distribution of reference signal received power and signal-to-noise ratio throughout the analysis period, the proportion of each numerical interval, the maximum value, the minimum value, and the average value. Together, they depict the overall health profile of the communication network. These two types of signals—anomalies representing problems at "points" and statistical data representing the overall situation—are simultaneously input into the subsequent parallel processing channels. The following process demonstrates a high degree of intelligence and automation, specifically unfolding into two parallel sub-processes. The first sub-process is the generation of professional analysis text, primarily achieved by calling the large language model service. This process itself comprises three distinct stages. The first is the prompt word configuration stage, which is the cornerstone guiding the behavior and output quality of the large language model. The system meticulously constructs and inputs a structured prompt word for this analysis task. This prompt word is not a simple instruction, but a detailed "task specification," clearly defining the role the large language model needs to play—for example, a rail transit communication expert with years of field experience; specifying its core task—generating the "Analysis and Summary" section of the analysis report based on the provided data; and limiting the style and depth of the output—requiring the use of concise, professional engineering language, focusing on anomaly pattern recognition, root cause inference, and actionable operational recommendations. Following this is the signal analysis and reasoning stage, where the large language model is activated. Based on its vast pre-trained knowledge base and powerful semantic understanding and logical reasoning capabilities, it performs in-depth interpretation of the input list of valid anomaly events and communication quality statistics. This interpretation goes beyond superficial numerical comparisons; it attempts to construct the causal chain behind the data. For example, it provides authoritative conclusions about the overall communication status; it performs pattern recognition (is it isolated or clustered? Does it occur in a specific geographical area or is it randomly distributed?) and severity assessment on anomalies; more importantly, it combines common communication failure models with domain knowledge to make professional and logical inferences about the possible causes of anomalies, such as linking persistent signal degradation in a specific section to trackside access point equipment failure or antenna alignment problems.Finally, in the text generation stage, the large language model organizes the results of the aforementioned analysis and reasoning into a logically rigorous, professionally expressed natural language paragraph that can be directly used in the final report—the so-called professional analytical text signal. This text not only clearly states "what happened," but also profoundly explains "why it happened" and "what is recommended," thus achieving a qualitative leap from data to insights.

[0029] Furthermore, in the second sub-process, which runs parallel to the large language model's text generation, the system simultaneously performs data visualization transformation. The goal is to generate visual charts that intuitively reflect the distribution characteristics of communication quality. This task is accomplished by the code execution module calling a professional charting library (such as ECharts), and the process also involves two key steps. The first is the data statistics step. In this step, the system performs secondary processing on the massive structured data signal, not to find anomalies, but to reveal the overall distribution pattern. It categorizes and counts the entire dataset according to preset, business-meaning numerical intervals (for example, dividing the reference signal received power into intervals of 5dBm), calculating the frequency of data points appearing in each interval or the percentage of the total data, thereby generating a statistical data signal showing the distribution proportion of communication quality parameters. This step transforms continuous data into discrete, comparable statistics. The next step is the chart rendering step. The code execution module takes the above distribution statistics as input and drives the charting library to programmatically and automatically create the corresponding visualization graphics, most typically a bar chart. In this chart, the horizontal axis represents the numerical range of communication parameters, and the vertical axis represents the frequency or proportion of data appearing in that range. Through a series of bars of varying heights, the overall distribution of communication quality—for example, an ideal bell-shaped distribution or a left-skewed distribution indicating a large number of weak signal areas—is clearly presented. This chart signal serves as strong evidence for the professional analysis text, revealing the macroscopic aspects of the problem in an irrefutable graphical way, making the analysis conclusions more comprehensive and credible. The two parallel sub-processes mentioned above—one generating in-depth text, and the other generating intuitive charts—converge at one point. The professional analysis text signal and the visual chart signal they produce are sent together to the report rendering module. This module reads a pre-designed Word document template containing fixed formatting and specific placeholders, and then performs automated content filling and synthesis: accurately filling the professional analysis text generated by the large language model into the text area marked "Analysis and Summary" in the template, while embedding the program-generated charts into the illustration positions specified in the template. This integration process ultimately outputs a well-formatted, richly illustrated, and complete comprehensive analysis report document, completing the final transformation from raw data to a professional deliverable. Furthermore, the system provided by this invention offers significant flexibility in the selection of core intelligent components through its top-level design. The system architecture is designed to support the configuration and access of various large language model services with different types and capabilities. This means that the "intelligent brain" responsible for generating the core analytical text is not a fixed entity. In practical applications, users or system administrators can flexibly select the most suitable large language model instance based on the specific needs of different scenarios.For example, in formal production environments or publicly released reports where the depth of analysis, creative language, and professionalism of the report are highly demanding, a large-scale commercial model with top-performing performance and timely knowledge base updates can be used. Conversely, when higher inference speed is required, or when cost control and data security deployment within an intranet are considered, a lightweight open-source model with fewer parameters but faster response can be switched to. This flexible, pluggable design ensures the core report generation functionality is implemented while possessing strong adaptability and scalability. Users can easily switch models within the intelligent workflow engine's configuration interface based on actual business needs, budget constraints, and technical environment, achieving a good balance between technological advancement, widespread applicability, and cost control, greatly expanding the application scenarios and lifecycle of this technical solution.

[0030] like Figure 4 As shown, this invention also provides a vehicle-to-ground communication quality prediction system based on a large model and intelligent workflow, comprising: a parsing module, which receives raw communication message files uploaded by users through a data access module and extracts structured data signals from the message files using a document parsing module; a traversal module, which retrieves structured data signals from a database through a code execution module, runs an anomaly detection algorithm, wherein the anomaly detection algorithm uses preset parameter threshold signals and persistent anomaly time window signals to traverse the structured data signals to identify potential anomaly signals, and filters potential anomaly signals based on persistence judgment logic to generate valid anomaly event signals; an execution module, which inputs the valid anomaly event signals and communication quality statistical data signals derived from the structured data signals into a large language model service, generates professional analysis text signals through structured prompt word instructions, and automatically generates visual chart signals based on the communication quality statistical data signals through a chart library called by the code execution module; an analysis module, which integrates the professional analysis text signals and visual chart signals into a predefined report template through a report rendering module to generate an analysis report signal; and a presentation module, which presents the analysis report signal to the user through a user interface module.

[0031] The present invention relates to a vehicle-to-ground communication quality prediction method and system based on a large model and intelligent workflow. By constructing an intelligent workflow, it first parses unstructured communication messages into structured data; then, it employs an anomaly detection algorithm based on duration determination to filter transient interference and accurately capture real faults; finally, it inputs the analysis results into a large language model to automatically generate a comprehensive report combining data, charts, and professional causal analysis. This achieves fully automated and intelligent analysis from raw data to in-depth insights, effectively improving operational efficiency and decision support capabilities.

[0032] Therefore, the vehicle-to-ground communication quality prediction method and system based on large models and intelligent workflow of the present invention can solve the problems of high false alarm rate, low degree of automation and lack of in-depth insight in vehicle-to-ground communication quality analysis.

[0033] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for predicting vehicle-to-ground communication quality based on large models and intelligent workflows, characterized in that, include: S1: Receive the original communication message file uploaded by the user through the data access module, and extract the structured data signal from the message file using the document parsing module; S2: The structured data signal is retrieved from the database through the code execution module, and the anomaly detection algorithm is run. The anomaly detection algorithm uses a preset parameter threshold signal and a continuous anomaly time window signal to traverse the structured data signal to identify potential anomaly point signals, and filters the potential anomaly point signals based on the persistence judgment logic to generate valid anomaly event signals. S3: Input the valid abnormal event signal and the communication quality statistics signal derived from the structured data signal into the large language model service, generate professional analysis text signal through structured prompt word instructions, and automatically generate visualization chart signal based on the communication quality statistics signal by calling the chart library through the code execution module; S4: The professional analysis text signals and visualization chart signals are integrated into a predefined report template through the report rendering module to generate analysis report signals; S5: Present the analysis report signal to the user through the user interface module.

2. The vehicle-to-ground communication quality prediction method based on large model and intelligent workflow according to claim 1, characterized in that, In step S2, the anomaly detection algorithm specifically includes: S201 parameter initialization step, setting a reference signal received power threshold signal and a signal-to-noise ratio threshold signal for determining communication quality anomalies, and setting a continuous anomaly time window signal for distinguishing between transient interference and persistent faults; S202 data traversal and anomaly point identification step, traversing each data point in the structured data signal in chronological order, and identifying data points that simultaneously satisfy the condition that their reference signal received power value is lower than the reference signal received power threshold signal and their signal-to-noise ratio value is lower than the signal-to-noise ratio threshold signal as potential anomaly point signals; S203 persistence determination step, checking whether the potential anomaly point signals appear continuously in the time series, and calculating the total duration of their continuous appearance. Only when the total duration exceeds the continuous anomaly time window signal is the corresponding potential anomaly point signal sequence determined and a valid anomaly event signal generated; S204 recording and storage step, writing the detailed information of the valid anomaly event signal into the database for persistent storage.

3. The vehicle-to-ground communication quality prediction method based on large model and intelligent workflow according to claim 1, characterized in that, In step S2, the anomaly detection algorithm is executed by the code execution node called by the intelligent workflow engine. The intelligent workflow engine provides a visual configuration interface, allowing users to arrange the entire analysis process, including the code execution node, by dragging and dropping. The values ​​of the parameter threshold signal and the continuous anomaly time window signal can be dynamically adjusted through the graphical interface without modifying the underlying program code of the anomaly detection algorithm. This enables flexible configuration and adaptive adjustment of the analysis process and judgment rules to meet the differentiated analysis needs of different lines or scenarios.

4. The vehicle-to-ground communication quality prediction method based on large model and intelligent workflow according to claim 1, characterized in that, In step S2, the filtering of potential anomaly signals based on the persistence determination logic employs a sliding window technique to monitor the continuity of the abnormal state in real time. By checking whether the distribution of the potential anomaly signals on the time axis constitutes a continuous segment whose duration meets the requirements of the continuous anomaly time window signal, isolated anomaly signals caused by instantaneous environmental disturbances are effectively distinguished and filtered, ensuring that the final generated effective anomaly event signal represents a real, continuous communication failure with operational value rather than a brief signal fluctuation.

5. The vehicle-to-ground communication quality prediction method based on large model and intelligent workflow according to claim 1, characterized in that, In step S3, the input of valid abnormal event signals and communication quality statistical data signals into the large language model service specifically includes: S301, a prompt word configuration step, configuring a structured prompt word for the large language model service, which instructs the large language model to operate from the perspective and professional terminology of rail transit communication experts; S302, a signal analysis and reasoning step, in which the large language model performs a comprehensive analysis of the input valid abnormal event signals and communication quality statistical data signals based on the prompt word, the analysis including a summary of the overall communication status, pattern recognition and severity assessment of valid abnormal event signals, and professional inference of possible causes of the abnormalities; and S303, a text generation step, which finally generates a professional analysis text signal containing specific operation and maintenance suggestions, with standardized language and in accordance with professional report requirements.

6. The vehicle-to-ground communication quality prediction method based on large model and intelligent workflow according to claim 5, characterized in that, In step S3, the automatic generation of visual chart signals by calling the chart library through the code execution module specifically refers to the data statistics step in S304, which derives statistical data signals on the distribution ratio of communication quality parameters in different numerical ranges from the structured data signal; and the chart rendering step in S305, which uses the chart generation library to programmatically create a bar chart signal based on the distribution ratio statistical data signal to intuitively display the distribution of reference signal received power and signal-to-noise ratio. This chart signal reveals the overall distribution characteristics and statistical laws of communication quality in a graphical way, serving as a visual supplement and support to the professional analysis text signal.

7. The vehicle-to-ground communication quality prediction method based on large model and intelligent workflow according to claim 1, characterized in that, In step S3, the system supports configuring and accessing a variety of different large language model services. Users can flexibly select and switch the large language model service instance they are using in the intelligent workflow engine according to their specific requirements for report generation quality or cost considerations.

8. The vehicle-to-ground communication quality prediction method based on large model and intelligent workflow according to claim 1, characterized in that, The integration of professional analysis text signals and visualization chart signals into a predefined report template via the report rendering module in step S4 refers to the system reading a report document template with a pre-designed format and placeholders, then automatically filling the corresponding text area of ​​the template with the professional analysis text signals generated by the large language model service, and embedding the generated visualization chart signals into the specified illustration positions of the template, and finally automatically synthesizing a complete analysis report signal containing structured text descriptions and accompanying visualization charts.

9. The vehicle-to-ground communication quality prediction method based on large model and intelligent workflow according to claim 1, characterized in that, The method further includes step S101 before step S2, namely the message data upload and structured parsing step. Specifically, the system receives the original communication message file uploaded by the user through the user interface, then calls the embedded document parsing module to parse the unstructured message according to the predefined communication protocol format, extracts key fields such as timestamp, train number, serving cell identifier, reference signal received power and signal-to-noise ratio, and forms the structured data signal. Finally, this batch of structured data signals is persistently stored in a relational database for use in subsequent analysis steps.

10. A system using the vehicle-to-ground communication quality prediction method based on large models and intelligent workflows according to any one of claims 1-9, characterized in that, include: The parsing module receives the original communication message file uploaded by the user through the data access module, and uses the document parsing module to extract structured data signals from the message file; The traversal module retrieves the structured data signal from the database through the code execution module, runs the anomaly detection algorithm, and uses a preset parameter threshold signal and a continuous anomaly time window signal to traverse the structured data signal to identify potential anomaly point signals, and filters the potential anomaly point signals based on the persistence judgment logic to generate valid anomaly event signals. The execution module inputs the valid abnormal event signal and the communication quality statistics signal derived from the structured data signal into the large language model service, generates professional analysis text signal through structured prompt word instructions, and automatically generates visualization chart signal based on the communication quality statistics signal by calling the chart library through the code execution module. The analysis module integrates the professional analysis text signals and visualization chart signals into a predefined report template through the report rendering module to generate an analysis report signal; The presentation module presents the analysis report signal to the user through the user interface module.

Images