A method and system for real-time simulation of communication networks empowered by a large model of omnest

Abstract

The application discloses a communication network real-time simulation method of a large model empowering OMNEST, comprising the following steps: establishing a two-way communication connection between an OMNEST simulation core and one or more large model service modules through a two-way interface and a control module; the large model service module receives a simulation task description defined by a user or simulation state data collected in real time from the OMNEST simulation core; based on the task description or the state data, analysis, reasoning or generation are carried out, and simulation parameter configuration suggestions, simulation scene descriptions, simulation process adjustment instructions or simulation result analyses are output; through learning the behavior mode of simulation data, the results of part of event sequences in simulation are predicted, and instructions are generated to guide the OMNEST simulation core to skip corresponding detailed calculation, so that the simulation process is accelerated; and the output of the large model service module is applied to the OMNEST simulation core, so as to initialize simulation configuration, dynamically adjust a simulation process or process simulation results. Corresponding systems, electronic equipment and computer readable storage media are also disclosed.

Description

Technical Field

[0001] This invention relates to the field of communication network simulation technology, specifically to a method and system for real-time communication network simulation empowered by large models in OMNEST, and particularly to a method and system for enhancing the OMNEST simulation platform capabilities by utilizing large language models (LMs) or generalized large models (LMs) to achieve more efficient, intelligent and near real-time communication network simulation. Background Technology

[0002] Communication networks are critical infrastructure of modern information society, growing increasingly large in scale and complex in structure, and carrying increasingly diverse types of services. From traditional internet and mobile communication networks (such as 4G, 5G, and evolving towards 6G) to emerging Internet of Things (IoT), Industrial Internet, Vehicle-to-Everything (V2X), and integrated air-space-ground networks, all place extremely high demands on network performance, reliability, and security. To address these challenges, communication network simulation plays a crucial role in network planning, design, optimization, operation and maintenance, as well as the research and deployment of new technologies (such as network slicing, edge computing, Software-Defined Networking (SDN), Network Functions Virtualization (NFV), and AI-native networks). Through simulation, researchers and engineers can evaluate the performance of new protocols, algorithms, and architectures in a controlled environment, predict network behavior under different load and fault conditions, thereby reducing the risks and costs of actual deployment.

[0003] OMNEST (Objective Modular Network Testbed in C++) and its open-source version OMNeT++ are widely used, modular, discrete-event network simulation platforms based on C++. They have gained popularity in academia and industry due to their powerful flexibility, scalability, and rich model library (such as the INET framework supporting various TCP / IP protocol stacks, wireless communication protocols, and mobile network models). Users can build various simulation models based on OMNEST, ranging from simple links to complex heterogeneous networks. However, with the expansion of network scale and the exponential growth of complexity, traditional OMNEST-based simulation methods face a series of severe challenges.

[0004] First, simulation efficiency and scalability issues are becoming increasingly prominent. For large-scale networks (such as IoT networks containing tens of thousands or even millions of nodes, or complex 5G / 6G core network and wireless access network co-simulation scenarios), traditional simulation methods often require huge computing resources and long simulation times, making it difficult to meet the needs of rapid iteration and agile development. Especially when a large number of parameter scans, Monte Carlo simulations, or exploratory studies are required, simulation time becomes the main bottleneck.

[0005] Secondly, there is the complexity of parameter configuration and scenario definition. Modern communication networks involve numerous adjustable parameters (such as routing protocol parameters, MAC layer parameters, transport layer congestion control algorithm parameters, queue management strategy parameters, mobility model parameters, channel model parameters, etc.), and these parameters often have complex nonlinear coupling relationships. Manually configuring and optimizing these parameters requires deep domain knowledge and a large number of trial-and-error experiments, which is inefficient and makes it difficult to achieve global optimization. At the same time, how to generate complex simulation scenarios that can fully reflect the dynamic characteristics and sudden behaviors of real-world networks is also a challenge. Traditional methods usually rely on simplified mathematical models or historical data, which are difficult to capture the unpredictability and diversity of network behavior.

[0006] Secondly, there is insufficient capacity to model dynamic and intelligent behavior. Future networks (such as 6G) will be more intelligent and adaptive, with network elements (such as AI-driven base stations, smart terminals, and self-healing network functions) dynamically adjusting their behavior according to the environment. Traditional simulation tools struggle to accurately simulate this inherent intelligent decision-making process and the resulting complex interactions. This makes it extremely difficult to evaluate the effectiveness of AI applications in networks or to simulate AI-native networks themselves.

[0007] Furthermore, the analysis and insight extraction of simulation results are inefficient. Large-scale simulations generate massive amounts of log data and statistical results. Quickly and accurately extracting valuable information from these data, identifying performance bottlenecks, anomaly patterns, and potential risks places extremely high demands on human analytical skills, often proving to be time-consuming and labor-intensive.

[0008] In recent years, large language models (LLMs), represented by the GPT series, and even more extensive large models (LMs) have made groundbreaking progress in fields such as natural language processing, code generation, knowledge-based question answering, and logical reasoning, demonstrating powerful learning, understanding, and generative capabilities. These models, pre-trained on massive amounts of data, have acquired rich world knowledge and pattern recognition capabilities. Introducing these capabilities of large models into the field of communication network simulation holds promise for overcoming many limitations of traditional simulation methods. For example, large models hold immense potential for intelligent parameter recommendation, automatic generation of complex scenes, acceleration of the simulation process (through predictive skipping or approximate calculations), and intelligent analysis of simulation results.

[0009] Therefore, the market urgently needs innovative methods that can integrate large-scale model intelligence to improve the efficiency, intelligence level, and real-time interactive capabilities of the OMNEST simulation platform. This demand stems not only from academic research's desire for more realistic and efficient simulation tools but also from industry's practical needs to accelerate new technology development, optimize network operations, and reduce testing costs. Empowering OMNEST with the capabilities of large-scale models, enabling it to better cope with the complexity and intelligent trends of future communication networks, is the key direction this invention aims to explore and achieve. Especially in scenarios requiring near-real-time analysis and decision-making, such as digital twin networks, online network optimization verification, and attack-defense drills, the lag of traditional simulation methods becomes a major obstacle, while large-scale model empowerment holds the promise of breaking this deadlock.

[0010] The existing technical solutions that are closest to or most closely related to this invention mainly include the following categories:

[0011] 1. Traditional OMNEST / INET simulation method

[0012] This is the most fundamental and widely used method in the field of communication network simulation. Researchers or engineers manually build network topologies using the OMNEST platform and its INET framework (or other specific model libraries), configure node parameters, link characteristics, traffic models, mobility models, etc., and then run discrete event simulations to collect and analyze the simulation results. For example, when evaluating a new routing protocol, developers might implement the protocol module in C++, define the network scenario in OMNEST, run simulations, and observe metrics such as convergence speed, path cost, and packet loss rate. This method is mature and reliable, but as mentioned earlier, it faces challenges such as efficiency and complexity in parameter configuration.

[0013] 2. Simulation optimization based on simple machine learning (non-large model) assistance

[0014] Some research attempts to apply traditional machine learning algorithms (such as Support Vector Machines (SVM), decision trees, reinforcement learning Q-learning, and neural networks (ANN)) to specific stages of the network simulation process. For example:

[0015] Parameter optimization: Using heuristic algorithms such as genetic algorithms, particle swarm optimization, or Bayesian optimization, combined with a small number of simulation results, the optimal combination of simulation parameters is automatically searched.

[0016] Traffic prediction: Train a machine learning model (such as ARIMA or LSTM) to predict network traffic patterns and use the prediction results as simulation input to improve the realism of the scenario.

[0017] Simulation acceleration: For certain computationally intensive modules in the simulation process (such as complex channel model calculations), an attempt is made to replace them with a pre-trained surrogate model, sacrificing some accuracy for speed. For example, a small neural network is used to fit complex physical layer calculations.

[0018] Results analysis: Using clustering algorithms or anomaly detection algorithms to perform preliminary screening on a large amount of data from simulation output helps researchers quickly locate events or patterns of interest.

[0019] A typical such scheme might involve the following processes:

[0020] (a) Define key adjustable parameters for the OMNEST simulation model for a specific optimization objective (such as minimizing end-to-end delay).

[0021] (b) Run a series of OMNEST simulation experiments to collect performance index data under different parameter combinations.

[0022] (c) Use these data to train a predictive model (such as Gaussian process regression) that can predict the output performance metrics based on the combination of input parameters.

[0023] (d) Using this prediction model, combined with optimization algorithms (such as the acquisition function in Bayesian optimization), the next parameter combination to be simulated is intelligently selected to gradually approach the optimal solution.

[0024] This approach improves the efficiency of parameter tuning to some extent, but its "intelligence" is limited. The models used usually have poor generalization ability and are mainly for offline optimization, making it difficult to achieve real-time dynamic adjustment during the simulation process.

[0025] 3. Hardware-in-the-Loop (HIL) simulation or hybrid simulation of network simulators and real devices.

[0026] To more closely approximate real-world environments, some solutions implement some network functions using real hardware devices, while others run in simulation or modeling environments, interacting through specific interfaces. For example, when testing a real wireless router, its surrounding network environment (such as a large number of clients and background traffic) can be generated using OMNEST simulation. This method can improve the credibility of the simulation and evaluate the performance of real hardware in complex network environments. However, it is costly, complex to configure, its scalability is limited by the capabilities of the real hardware, and it is difficult to flexibly explore various network conditions and parameters.

[0027] 4. AI-based network management and control system (non-simulation)

[0028] In real-world network operations, numerous systems already utilize AI / ML technologies for network traffic scheduling, fault diagnosis, and resource allocation. These systems typically process real-time data from the actual network and make control decisions. While not directly simulation tools, the AI / ML models and methods they employ (such as deep reinforcement learning for network control) provide valuable insights for the simulation field. However, directly embedding these AI models into simulation tools while ensuring simulation efficiency and stability is itself a challenge.

[0029] While there have been attempts to apply machine learning to simulation in existing technologies, most are limited to using small models for specific tasks, lacking the general understanding, complex reasoning, and flexible generation capabilities of large models. They typically optimize isolated parts of the simulation process, failing to form a systematic approach that deeply integrates large models into the OMNEST simulation core loop to achieve comprehensive empowerment (including parameter optimization, scene generation, process adaptive adjustment, and intelligent interpretation of results). In particular, existing solutions have limited capabilities in pursuing "real-time" or "near real-time" responses.

[0030] For example, the OMNEST parameter optimization process based on simple machine learning is a typical existing technique for OMNEST parameter optimization using machine learning. Its core idea is to reduce the number of costly real-world simulations through surrogate models, including:

[0031] A: Users define the OMNEST network model, the performance metrics they are interested in (such as latency and throughput), and the optimization objectives (such as minimizing latency).

[0032] B: The user determines the range of parameters to be optimized and the initial parameter sampling points.

[0033] C: The OMNEST simulator runs a simulation based on a given combination of parameters.

[0034] D: Record the performance metrics obtained from the simulation.

[0035] E: Determine whether the collected (parameter-performance) data pairs are sufficient to train the agent model.

[0036] F: Use the collected data to train a machine learning model that can predict the parameter-to-performance mapping.

[0037] G: The optimization algorithm utilizes the predictive power of the surrogate model to efficiently select the next most promising parameter combination for simulation.

[0038] H: Determine whether the preset number of iterations, time budget, or performance convergence criteria have been met.

[0039] I: Output the best combination of parameters found.

[0040] Although the aforementioned existing technical solutions have made some progress in certain aspects, they still have the following significant drawbacks:

[0041] 1. Limited level of intelligence, lacking versatility and adaptability.

[0042] Existing simple machine learning models (such as SVM and shallow neural networks) are typically customized and trained for specific problems and network scenarios, resulting in limited learning capabilities and knowledge coverage. These models often lack a deep understanding of network protocols and behaviors, making it difficult to handle highly complex, dynamically changing, or previously unseen network scenarios. For example, a model trained to optimize parameters for a specific routing protocol is difficult to directly apply to another protocol or a completely different network topology. They lack the extensive prior knowledge and powerful zero-shot or few-shot learning capabilities of larger models, which are based on massive amounts of pre-trained data. This leads to poor generalization ability and requires significant retraining and manual intervention when adapting to new tasks or environments.

[0043] 2. The automation level of parameter optimization and scene generation is not high, and it still relies heavily on human experience.

[0044] While some automatic parameter tuning methods exist, their search spaces typically require manual pre-definition, and their search efficiency remains low for high-dimensional parameter spaces. In terms of scene generation, existing methods largely rely on parameterized traffic models or replays based on historical data, making it difficult to generate truly innovative, complex, or adversarial scenarios that can effectively test the robustness boundaries of networks. Model selection, feature engineering, and hyperparameter tuning still require expert experience, and the level of automation and intelligence needs improvement. Large-scale models hold promise for automatically inferring appropriate parameter ranges or generating scene elements from high-level requirement descriptions through natural language understanding and other methods—a feat currently difficult to achieve with existing technologies.

[0045] 2. The simulation process lacks real-time dynamic adjustment and adaptive capabilities.

[0046] Current simulations are typically run "one-off": once parameters and scenarios are set, the simulator runs from start to finish, making it difficult to intelligently adjust based on the simulation's progress. For example, if an unexpected congestion pattern occurs during the simulation, existing systems struggle to automatically adjust traffic injection rates, routing strategies, or resource allocation strategies to explore this phenomenon, or proactively guide the simulation towards a specific "state of interest." This static execution method limits the simulation's ability to explore unknown behaviors and conduct interactive experiments. Achieving such dynamic adjustments requires the model to have the ability to analyze the simulation state in real time and make rapid decisions, which places excessively high real-time requirements on traditional small models.

[0047] 4. The improvement in simulation efficiency is limited, making it difficult to handle ultra-large-scale and highly complex networks.

[0048] While techniques like surrogate models can accelerate parameter search to some extent, their speedup effect is limited when dealing with the time consumption of a single simulation run, especially in scenarios with a large number of nodes and extremely complex interactions. They typically cannot fundamentally change the simulator's event-by-event processing mechanism. Achieving greater efficiency improvements might require models capable of predictively skipping certain unimportant event sequences, or performing higher-level abstractions and approximations of the network's macroscopic behavior—all of which exceed the capabilities of traditional machine learning models.

[0049] 5. Significant gap between the current performance and the "real-time" requirement.

[0050] In communication network simulation, "real-time" can have multiple meanings: First, the simulation time is synchronized or nearly synchronized with real time, which is crucial for hardware-in-the-loop simulation or digital twin applications. Second, the simulation system can respond quickly to external inputs (such as user commands and real network data streams), enabling interactive simulation. Existing pure software simulations based on OMNEST, especially for complex models, are much slower than real time. Furthermore, auxiliary methods based on simple machine learning may introduce additional latency during model training and inference, and their decision-making capabilities are insufficient to support complex real-time interactions.

[0051] 6. The human-computer interaction is not user-friendly, and the interpretation of results relies on professional knowledge.

[0052] Configuring simulations and analyzing results typically requires users to have extensive experience with OMNEST and network expertise. The large amounts of raw data output from simulations are often difficult to understand intuitively. Current technologies lack the ability to simplify simulation task definition and parameter settings using natural language processing or other methods, or to automatically summarize complex simulation results into reports or insights that are easily understood by humans.

[0053] These shortcomings collectively limit the effectiveness of traditional simulation tools such as OMNEST in meeting the needs of future network development, making the simulation process still a relatively time-consuming, labor-intensive, and insufficiently intelligent process. Summary of the Invention

[0054] The purpose of this invention is to provide a real-time simulation method and system for communication networks powered by a large model for OMNEST, addressing these shortcomings. It aims to systematically solve these problems by introducing the powerful capabilities of large models. The core of this method lies in constructing a collaborative system comprising an OMNEST simulation core, a large model (LM) service module, a bidirectional interface and control module, a data acquisition and preprocessing module, and a knowledge base module. By building an innovative framework for deep integration and collaborative work between a large model and the OMNEST simulator, the invention aims to: enhance simulation intelligence by enabling intelligent recommendation of simulation parameters, automatic generation and dynamic adjustment of complex scenarios, and adaptive control of the simulation process; and improve simulation efficiency by exploring predictive acceleration and intelligent simulation using large models. Event management shortens the runtime of large-scale complex network simulations, enabling them to meet the needs of near real-time or interactive applications in certain scenarios; enhanced simulation capabilities allow OMNEST to better simulate future network systems with intrinsic intelligence and dynamic adaptability, supporting the evaluation of AI applications in networks and research on AI-native networks; improved user experience provides a more natural and convenient human-computer interaction method and enables intelligent analysis and presentation of simulation results; ultimately, this invention aims to transform OMNEST from a traditional simulation tool that mainly relies on manual configuration and driving into an intelligent simulation platform with higher autonomy, adaptability, and insight, thereby more effectively supporting the research, development, and optimization of next-generation communication network technologies.

[0055] The first aspect of this invention is to provide a real-time simulation method for communication networks using a large model-enabled OMNEST, comprising:

[0056] S1, establish a bidirectional communication connection between the OMNEST simulation core (101) and one or more large model service modules (102) through a bidirectional interface and control module (103);

[0057] S2, the large model service module (102) receives a user-defined simulation task description or simulation status data collected in real time from the OMNEST simulation core (101); wherein the simulation task description is in natural language form, and the large model service module (102) uses natural language understanding capabilities to convert it into structured simulation objectives and constraints;

[0058] S3, the large model service module (102) analyzes, infers, or generates based on the task description or state data, and outputs simulation parameter configuration suggestions, simulation scene descriptions, simulation process adjustment instructions, or simulation result analysis; and the large model service module (102) predicts the results of some event sequences in the simulation by learning the behavior patterns of the simulation data, and generates instructions to guide the OMNEST simulation core (101) to skip the corresponding detailed calculations in order to accelerate the simulation process; wherein, the simulation parameter configuration suggestions and simulation scene descriptions generated by the large model service module (102) are converted into configuration files or scripts readable by the OMNEST simulation core (101) for initializing the OMNEST simulation;

[0059] S4, the output of the large model service module (102) is applied to the OMNEST simulation core (101) to initialize the simulation configuration, dynamically adjust the simulation process, or process the simulation results; wherein, the large model service module (102) dynamically generates simulation process adjustment instructions based on real-time simulation status data, and the simulation process adjustment instructions are sent to the OMNEST simulation core (101) through a bidirectional interface and control module to modify network model parameters, topology, node behavior, or inject events during simulation.

[0060] Preferably, S1 includes:

[0061] S11, Simulation Task Definition and Initialization: Users define the simulation purpose, performance metrics of interest, and network type through a graphical user interface (UI) that supports natural language; system initialization and model loading;

[0062] S12, LM Auxiliary Parameters and Scene Generation: Based on user input and knowledge base information, LM generates initial OMNEST simulation parameters and scene descriptions, including the dynamic generation of the topology file .ned and the traffic script;

[0063] S13, OMNEST Simulation Startup and Run: The OMNEST simulator loads its configuration and begins executing discrete event simulations.

[0064] Preferably, S2 includes: real-time data acquisition and preprocessing, including: the data acquisition and preprocessing module (104) collecting network status data, performance indicators, and event logs according to a preset strategy or LM command during simulation operation, and performing preprocessing; and the simulation status data collected in real time from the OMNEST simulation core is processed by the data acquisition and preprocessing module, key features are extracted and formatted and then provided to the large model service module (102).

[0065] Preferably, S3 includes: a large model service module (102) for real-time analysis and decision-making, including:

[0066] The preprocessed data is transmitted to the large model service module (102) through the bidirectional interface and the control module (103).

[0067] The large model service module (102) analyzes the current simulation status to obtain analysis results, which are used to determine whether the simulation meets expectations, whether there are any anomalies, or whether a certain key node has been reached.

[0068] Preferably, S4 includes:

[0069] S41, Determine whether dynamic adjustment is needed: The large model service module (102) determines whether intervention in the simulation process is needed based on the analysis results.

[0070] S42, LM generates adjustment instructions: If adjustment is required, the large model service module (102) generates specific adjustment instructions, including modifying the parameters of a node, changing the state of a link, or injecting new traffic or events; the bidirectional interface and control module (103) translates the instructions and sends them to the OMNEST simulation core (101); the OMNEST simulation core (101) continues to run after performing the adjustment;

[0071] S43, Determine whether the simulation has ended: Determine whether the simulation has reached the preset end conditions, the end conditions include simulation time, occurrence of a specific event, or stable performance indicators;

[0072] S44, Large Model Assisted Result Analysis and Report Generation: After the simulation is completed, the large model service module (102) performs in-depth analysis on all collected simulation data, extracts key insights, generates summary reports, visualization charts and proposes optimization suggestions;

[0073] S45 stores the configuration, key results, decision-making process, and effects of this simulation into the knowledge base module; if some decisions of the large model bring good / bad results, this information is used as feedback for subsequent fine-tuning of the large model or to prompt engineering optimization.

[0074] The second aspect of the present invention provides a real-time simulation system for communication networks that enables large-scale modeling of OMNEST simulations and performs real-time simulations of communication networks, thereby implementing the method of the first aspect, comprising:

[0075] The OMNEST simulation core (101) is used to perform discrete event simulations of communication network models based on the traditional OMNEST simulation environment.

[0076] The large model service module (102) (LM_Service, LM) has one or more large models built in and hosts, which are used to receive tasks, perform intelligent analysis and decision reasoning and return results; the large models are GPT-4, LLaMA or models finely tuned for the network domain;

[0077] The bidirectional interface and control module (103) (Interface_Ctrl) is used for bidirectional communication between the OMNEST simulation core (101) and the large model service module (102) to exchange data, translate instructions and perform synchronization control, as well as parse user instructions and convert them into large model tasks; convert the output of LM into commands or parameters that can be executed by OMNEST; and manage data flow to ensure synchronization.

[0078] The data acquisition and preprocessing module (104) (Data_Collector) is used to collect and process data from the OMNEST simulation core (101), including: collecting network status data, event information and statistical indicators in real time or periodically during the OMNEST simulation process and performing preprocessing such as cleaning, transformation and feature engineering. The network status data includes node location, link load, queue length, packet loss rate and latency.

[0079] Preferably, the system further includes: a knowledge base module for storing and managing the knowledge required and generated for human-computer interaction and LM decision-making, wherein the required and generated knowledge includes historical simulation data, network domain knowledge and large model decision records and feedback data, successful parameter configurations, typical scenario definitions, network domain knowledge graphs, and feedback and learning results of the large model; when executing S3, the large model service module (102) queries the knowledge base module to obtain contextual information to enhance its analysis, reasoning or generation capabilities; after the simulation task is completed, the configuration, key results, large model decisions and their effect evaluation of this simulation are stored in the knowledge base module, and can be used for feedback learning or fine-tuning of the large model service module; and after the simulation is completed, the large model service module analyzes all or part of the simulation data generated by the OMNEST simulation core, automatically generates a natural language analysis report containing key insights, statistical summaries or visualization suggestions and stores it in the knowledge base module.

[0080] Preferably, the system also includes a graphical interface, through which users interact with the system using either a graphical interface or natural language.

[0081] A third aspect of the present invention provides an electronic device including a processor and a memory, the memory storing a plurality of instructions, the processor being configured to read the instructions and execute the method as described in the first aspect.

[0082] A fourth aspect of the present invention provides a computer-readable storage medium storing a plurality of instructions which can be read by a processor and executed as described in the first aspect.

[0083] The beneficial effects of the method and system of the present invention are as follows:

[0084] 1. System-Level Deep Integration Architecture Innovation: This invention proposes a system architecture that deeply integrates the Large Model (LM) with the OMNEST simulation platform. This architecture not only includes the OMNEST core and LM services, but also, crucially, incorporates bidirectional interface and control modules, real-time data acquisition and preprocessing modules, and a dynamic knowledge base module. These modules work collaboratively to enable LM to intelligently empower the entire OMNEST simulation lifecycle (from initial configuration to dynamic operation and result analysis), forming a tightly coupled closed-loop feedback system. This is fundamentally different from the loose integration approach in existing technologies where LM / ML is typically used as an add-on tool or only for simulating isolated aspects (such as offline parameter optimization). Technical Effects: This achieves efficient, bidirectional information and control flow between LM and the simulator, allowing LM's intelligence to permeate every key stage of simulation execution in real time, rather than just preprocessing or post-processing.

[0085] 2. Innovation in intelligent and automated simulation process based on LM:

[0086] Natural Language-Driven Intelligent Generation of Parameters and Scenes: This invention leverages the powerful natural language understanding and generation capabilities of LM (Model-Language Logic), allowing users to describe simulation requirements using high-level, fuzzy, or even natural language. LM can then automatically reason and generate specific OMNEST parameter configurations (.ini, .ned fragments) and complex dynamic scene scripts by combining a knowledge base. This surpasses existing parameter / scene generation methods based on fixed rules or simple optimization algorithms. Technical Effects: It significantly reduces the threshold and complexity of simulation configuration, improves the intelligence level of configuration, and enhances the richness and relevance of scenes, enabling even non-OMNEST experts to effectively utilize the simulation platform.

[0087] LM-driven simulation process with real-time adaptive adjustment: One of the core innovations of this invention is that LM can analyze data collected from OMNEST in real time, understand the current simulation state, and dynamically generate adjustment instructions (such as modifying parameters, injecting events, and changing topology) to feed back to OMNEST for execution. This makes the simulation process no longer a rigid execution of a preset script, but can intelligently evolve based on intermediate states. Technical effects: It endows the simulation system with unprecedented flexibility and exploratory capabilities, enabling it to proactively discover and respond to sudden behaviors, bottlenecks, or unknown phenomena in the network, and more realistically simulate the dynamic characteristics of intelligent networks (such as AI-native networks).

[0088] LM-Driven Predictive Simulation Acceleration Mechanism (Concept Proposal and Integration Framework): This invention proposes using LM to predict the results of certain predictable event sequences or subprocesses in simulations, or those with minimal impact on macroscopic outcomes. This allows OMNEST to selectively skip some detailed calculations under the guidance of LM, achieving acceleration. While the construction of specific predictive models is a future research area, this invention provides a framework for integrating such mechanisms. Technical Effects: This provides a new approach to solving the problem of time-consuming large-scale simulations, and is expected to significantly improve simulation efficiency while maintaining acceptable accuracy, bringing it closer to the needs of real-time analysis.

[0089] 3. Innovation in Dynamic Knowledge Base and LM Continuous Learning Mechanism: This invention introduces a dynamic knowledge base closely integrated with the LM (Model-Learning System). This knowledge base not only stores historical data and domain knowledge but also records the LM's decisions, intervention effects, and user feedback during the simulation process. This information provides the LM with richer real-time context (through technologies such as RAG) and serves as valuable material for the LM's continuous learning and optimization. Technical Effects: This enables the entire empowerment system to "grow," as the LM's professional capabilities and service effectiveness continuously improve with usage and data accumulation, forming a positive cycle.

[0090] Enhanced Human-Computer Interaction and Intelligent Result Interpretation Innovation: Leveraging the natural language processing capabilities of LM, this invention supports more natural human-computer interaction methods (such as simulation configuration and querying via dialogue), and can automatically extract easily understandable insights, summary reports, and visual suggestions from massive and complex OMNEST raw output data. Technical Effects: Significantly improves the usability of OMNEST simulations and the interpretability of results, enabling simulation tools to better serve a wider range of users and decision-making processes.

[0091] Compared with the closest solutions of the present invention, such as simulation optimization based on simple machine learning (non-large models), the present invention has the following significant technical advantages or effects:

[0092] 1. Significantly improved level of intelligence and versatility:

[0093] Limitations of existing technologies: Traditional ML models are usually specialized small models with narrow knowledge, limited understanding capabilities, poor generalization, and difficulty in handling complex and unseen scenarios. For example, a Bayesian optimizer optimized for a specific routing protocol cannot directly understand natural language instructions such as "simulate the impact of a DDoS attack on a large-scale IoT network," nor can it proactively suggest appropriate attack models or IoT device behavior patterns.

[0094] Advantages of this invention: This invention employs pre-trained large models (LLMs / LMs), which, through training on massive amounts of diverse data (including text, code, and even potentially network knowledge after fine-tuning), possess powerful capabilities in natural language understanding, knowledge reasoning, complex pattern recognition, and content generation. This means:

[0095] Enhanced task understanding and adaptability: The LM can directly understand the complex and ambiguous simulation requirements described by users in natural language, and can provide a more comprehensive and insightful response by combining its extensive background knowledge (or through knowledge base enhancement). For example, it can understand high-level instructions such as "evaluate the robustness of the new XX protocol in future urban intelligent transportation scenarios" and proactively consider factors that may need to be considered, such as high mobility, multipath effects, and interference, which is difficult for traditional small models to achieve.

[0096] Zero-shot / few-shot learning capability: For new network protocols, new scene types, or new optimization objectives, large models can often provide reasonable initial suggestions based on their general capabilities, without having to perform extensive retraining or feature engineering for new problems as traditional ML models do.

[0097] Cross-domain knowledge transfer: The abstract reasoning capabilities learned by large models in other domains can be applied to the field of network simulation. For example, the ability to generate code can help generate OMNEST configuration files or simple module scripts.

[0098] Results: Significantly reduced reliance on user expertise, improved the automation and intelligence of simulations, and enabled simulation tools to adapt to a wider range of more complex communication network research needs.

[0099] 2. Revolutionary improvements in parameter configuration and scene generation:

[0100] Existing technology has the following drawbacks: parameter optimization is usually limited to the search of numerical parameters, and the definition of the search space still needs to be done manually; scene generation relies on parameterized models or replay of historical data, which lacks innovation and adversarial nature.

[0101] Advantages of this invention:

[0102] Semantic-driven parameter recommendation: LM not only optimizes numerical parameters but also understands their semantic meaning and interrelationships, thus providing more logical and practical parameter combinations. For example, based on the goal of "high reliability and low latency," it can recommend a set of protocol parameters that favor conservative retransmission and fast route convergence.

[0103] Creative Scenario Generation: LM possesses generative capabilities, enabling it to create entirely new and challenging simulation scenarios based on simulation objectives (such as stress testing and security assessment), rather than simply repeating existing patterns. For example, LM can design a novel, multi-stage, collaborative attack strategy to test network defense systems, or generate complex user movement and application usage patterns that better align with human behavioral logic.

[0104] Automation from text to configuration: LM can transform high-level scene descriptions (even narrative descriptions) into specific OMNEST executable configurations, and this end-to-end automation capability far surpasses existing technologies.

[0105] Results: It greatly improves the efficiency and quality of simulation parameter configuration, enabling researchers to quickly build and test more diverse simulation scenarios that are closer to the complexity of the real world or more exploratory.

[0106] 3. Achieve real-time dynamic adaptive control during the simulation process:

[0107] Existing technology has limitations: simulation processes are typically static and pre-defined, lacking the ability to intelligently adjust based on intermediate simulation states. Even with machine learning assistance, optimization is mostly offline rather than online.

[0108] Advantages of this invention: By constructing a real-time data feedback and control loop between LM and OMNEST, LM can:

[0109] Real-time monitoring and understanding: Analyze multi-dimensional data generated during the simulation process in near real-time to understand the current operating status and dynamic evolution trend of the network.

[0110] Intelligent Decision-Making and Intervention: Based on preset goals or observed phenomena, LM can proactively make decisions and generate control commands to dynamically modify simulation parameters, network topology, node behavior, or inject new events in OMNEST. For example, when LM detects unexpected congestion in a certain area, it can instruct the simulation system to try different mitigation strategies (such as dynamically adjusting routes, activating backup links, or limiting some traffic) and observe their effects.

[0111] Effects: This transforms simulation from a passively executed tool into an intelligent system with a degree of "autonomy" and "exploration." It can more effectively study the dynamic behavior of networks, their emergency response capabilities, and the performance of AI-driven adaptive network mechanisms—areas that traditional static simulations struggle to fully cover.

[0112] 4. Potential ways to improve simulation efficiency:

[0113] Existing technological limitations: simulation acceleration methods are limited, mainly relying on hardware upgrades or proxy models targeting specific computational bottlenecks, making it difficult to fundamentally address the challenges of ultra-large-scale simulations.

[0114] Advantages of this invention: The "LM-driven predictive simulation acceleration" proposed in this invention is an innovative direction. If the LM can accurately predict the macroscopic results of certain "mediocre" or highly repeatable simulation phases by learning historical simulation data and network behavior patterns, then OMNEST can skip the detailed event simulation of these phases under the guidance of the LM and directly fast-forward to the next critical state point.

[0115] Effects: While achieving robust predictive acceleration remains challenging, success could lead to orders of magnitude reductions in simulation time, making large-scale, long-duration simulations, which were previously infeasible due to excessive time consumption, possible. This would bring us closer to the need for "real-time" analysis, and would be of great significance, especially in applications such as digital twins.

[0116] 5. Significant improvements in human-computer interaction experience and efficiency in interpreting results:

[0117] Existing technology drawbacks: Simulation tools such as OMNEST are complex to configure, output massive amounts of data, require high levels of user expertise, and result analysis is time-consuming and laborious.

[0118] Advantages of this invention:

[0119] Natural language interaction: Users can use natural language to define simulation tasks, query parameters, inquire about status, and even issue control commands, which greatly reduces the barrier to entry.

[0120] Intelligent Report Generation: LM can automatically process, analyze, and summarize massive amounts of multi-dimensional simulation raw data, generating easy-to-understand natural language reports, key charts, and valuable insights. For example, LM can summarize the latency data points of thousands of nodes as "The overall network latency increases sharply after time point X due to reason Y, mainly affecting type Z services," and provide possible optimization suggestions.

[0121] Results: The OMNEST simulation platform is made easier to use and understand, and simulation results can be transformed into actionable knowledge more quickly, improving R&D and decision-making efficiency.

[0122] 6. The ability to accumulate knowledge and continuously evolve the system:

[0123] Existing technology limitations: Most ML-assisted simulation solutions are one-off, and the learning outcomes and experience of the models are difficult to accumulate and reuse effectively.

[0124] Advantages of this invention: Through the dynamic knowledge base module, the system can record the configuration of each simulation, the decision-making process of the learning model (LM), the intervention effect, user feedback, etc. This structured empirical data not only provides more accurate context for subsequent simulations (through RAG), but also serves as valuable corpus for continuous fine-tuning and optimization of the LM.

[0125] Effect: The empowered system has the ability to "learn and grow". Its performance will continue to improve with the increase of usage time and the accumulation of data, forming a virtuous cycle, which is generally lacking in existing technical solutions.

[0126] In summary, this invention, by systematically and deeply integrating the general intelligence of large-scale models with the professional simulation capabilities of OMNEST, not only overcomes many shortcomings of existing technologies in terms of intelligence, automation, efficiency, and ease of use, but also opens up a new paradigm for the development of communication network simulation technology, enabling it to better cope with the complexity, dynamism, and intelligence requirements of future networks (such as 6G and AI-native networks). The technical effects brought about by this invention are comprehensive, from improving the efficiency and quality of individual simulation tasks to enhancing the capabilities and user experience of the entire simulation platform, and further promoting the accumulation of simulation knowledge and the evolution of intelligent systems, all demonstrating significant advantages. Attached Figure Description

[0127] To more clearly illustrate the technical solutions in the specific embodiments or related technologies of the present invention, the drawings used in the description of the specific embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0128] Figure 1 This is a flowchart of a method provided according to an embodiment of the present invention;

[0129] Figure 2 A system architecture diagram provided according to an embodiment of the present invention;

[0130] Figure 3 This is a structural diagram of an electronic device provided according to an embodiment of the present invention. Detailed Implementation

[0131] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0132] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0133] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0134] The core technical problems to be solved include:

[0135] 1. How to deeply integrate the general intelligence of large models with the professional capabilities of the OMNEST simulation platform?

[0136] In existing technologies, the integration of AI / ML with simulation is often loose and add-on-top. The first technical problem this invention aims to solve is how to design a tightly coupled system architecture and interaction mechanism that enables large models to effectively understand the state of the OMNEST simulation environment and to accurately and efficiently control or influence the simulation process, forming a closed-loop intelligent simulation system. This includes the design of data interfaces, communication protocols, model invocation methods, and synchronization mechanisms.

[0137] 2. How to use large models to improve the intelligence and automation level of OMNEST simulation?

[0138] This specifically includes:

[0139] Intelligent parameter configuration: How can we leverage the knowledge reasoning and learning capabilities of large models to automatically recommend or generate optimized OMNEST simulation parameter sets (such as protocol parameters, topology parameters, and traffic parameters) based on the user's high-level description (such as network objectives and expected performance described in natural language) or historical simulation data, thereby reducing manual trial and error and reliance on expert experience?

[0140] Automatic generation and adaptation of complex scenarios: How can large models automatically design and generate challenging, diverse, or specific simulation scenarios (such as network topology, traffic patterns, fault injection, and user behavior sequences) based on simulation objectives (such as stress testing, robustness verification, and reproduction of specific phenomena)? Furthermore, during the simulation process, how can large models dynamically adjust scenario elements according to the current simulation state to achieve adaptive evolution of the scenario?

[0141] 3. How can we significantly improve the running efficiency of OMNEST simulation by using large models, making it closer to the "real-time" requirement?

[0142] For large-scale complex network simulation, time consumption is the main bottleneck. The technical problem this invention aims to solve is:

[0143] Predictive simulation acceleration: Can large models learn network behavior patterns to predict the results of certain subprocesses or event sequences in the simulation, thereby allowing OMNEST to skip some redundant or predictable detailed calculations and achieve macro-level acceleration?

[0144] Intelligent event scheduling / filtering: Can the large model identify events that have a minor impact on the overall simulation results and guide the OMNEST simulator to perform intelligent event pruning or priority adjustment to reduce unnecessary computational overhead?

[0145] 4. How to enhance the dynamic adaptation and interactive exploration capabilities of the OMNEST simulation process?

[0146] Existing simulation processes are rigid and difficult to adjust dynamically. The technical problem this invention aims to solve is: how to enable large models to analyze intermediate data in real-time or near real-time during simulation, identify key trends or anomalies, and dynamically adjust simulation parameters, inject new events, or change model behavior based on this, thereby guiding the simulation to evolve towards a specific goal or exploring the boundary behavior and emergency response capabilities of network systems?

[0147] 5. How to improve the user-friendliness of OMNEST simulation and the efficiency of result interpretation?

[0148] The technical problem this invention aims to solve is: how to utilize the natural language processing capabilities of large models to allow users to interact with the simulation system through natural language (such as configuring simulation, querying status, and issuing commands), and how to enable large models to automatically extract key insights from massive simulation data and generate understandable analysis reports or visualization suggestions, thereby reducing the user's usage threshold and analysis burden.

[0149] Example 1

[0150] like Figure 1 As shown, this embodiment provides a real-time simulation method for communication networks using a large model-enabled OMNEST, including:

[0151] S1, establish a bidirectional communication connection between the OMNEST simulation core (101) and one or more large model service modules (102) through a bidirectional interface and control module (103);

[0152] S2, the large model service module (102) receives a user-defined simulation task description or simulation status data collected in real time from the OMNEST simulation core (101); wherein the simulation task description is in natural language form, and the large model service module (102) uses natural language understanding capabilities to convert it into structured simulation objectives and constraints;

[0153] S3, the large model service module (102) analyzes, infers, or generates based on the task description or state data, and outputs simulation parameter configuration suggestions, simulation scene descriptions, simulation process adjustment instructions, or simulation result analysis; and the large model service module (102) predicts the results of some event sequences in the simulation by learning the behavior patterns of the simulation data, and generates instructions to guide the OMNEST simulation core (101) to skip the corresponding detailed calculations in order to accelerate the simulation process; wherein, the simulation parameter configuration suggestions and simulation scene descriptions generated by the large model service module (102) are converted into configuration files or scripts readable by the OMNEST simulation core (101) for initializing the OMNEST simulation;

[0154] S4, the output of the large model service module (102) is applied to the OMNEST simulation core (101) to initialize the simulation configuration, dynamically adjust the simulation process, or process the simulation results; wherein, the large model service module (102) dynamically generates simulation process adjustment instructions based on real-time simulation status data, and the simulation process adjustment instructions are sent to the OMNEST simulation core (101) through a bidirectional interface and control module to modify network model parameters, topology, node behavior, or inject events during simulation.

[0155] In a preferred embodiment, S1 includes:

[0156] S11, Simulation Task Definition and Initialization: Users define the simulation purpose, performance metrics of interest, and network type through a graphical user interface (UI) that supports natural language; system initialization and model loading;

[0157] S12, LM Auxiliary Parameters and Scene Generation: Based on user input and knowledge base information, LM generates initial OMNEST simulation parameters (such as the contents of .ini files) and scene descriptions (which may include dynamic generation of topology files .ned, traffic scripts, etc.).

[0158] S13, OMNEST Simulation Startup and Operation: The OMNEST simulator loads its configuration and begins discrete event simulation.

[0159] In a preferred embodiment, S2 includes:

[0160] Real-time data acquisition and preprocessing: During simulation, the Data Collector module collects network status data, performance indicators, and event logs according to preset strategies (time intervals, event triggers) or LM commands, and performs preprocessing.

[0161] The simulation state data collected in real time from the OMNEST simulation core is processed by the data acquisition and preprocessing module, and key features are extracted and formatted before being provided to the large model service module.

[0162] In a preferred embodiment, S3 includes:

[0163] LM Real-time Analysis and Decision-Making: Preprocessed data is passed to LM via the Interface module. LM analyzes the current simulation state to determine whether it meets expectations, whether there are any anomalies, or whether a certain critical node has been reached.

[0164] In a preferred embodiment, S4 includes:

[0165] S41, Determine if dynamic adjustment is needed: Based on the analysis results, LM determines whether intervention in the simulation process is necessary.

[0166] S42, LM generates adjustment instructions: If adjustments are needed, LM generates specific adjustment instructions (such as modifying parameters of a node, changing the status of a link, injecting new traffic or events, etc.). The Interface module translates the instructions and sends them to OMNEST. OMNEST executes the adjustments and continues running.

[0167] S43, Determine if the simulation has ended: Determine if the simulation has met the preset end conditions (such as simulation time, occurrence of a specific event, stable performance indicators, etc.).

[0168] S44, LM-assisted result analysis and report generation: After the simulation is completed, LM performs in-depth analysis on all collected simulation data, extracts key insights, generates summary reports, visualization charts, and even proposes optimization suggestions.

[0169] S45. Store the configuration, key results, LM decision-making process, and effects of this simulation in the knowledge base. If some decisions of the LM lead to good / bad results, this information can be used as feedback for subsequent fine-tuning of the LM or to suggest engineering optimizations.

[0170] Detailed steps explained:

[0171] Step 1: Simulation Task Definition and Initialization (S11)

[0172] • Main components: user, user interface (UI), bidirectional interface and control module.

[0173] • Subject: Simulation target, network type, key performance indicators (KPIs).

[0174] • Triggering condition: The user initiates a new simulation task.

[0175] • Execution Actions: Users input simulation requirements via a UI (which can be a graphical interface, a command-line interface supporting natural language input, or a chat interface). For example: "I need to evaluate the performance of a novel wireless routing protocol in a large-scale V2X communication scenario in an urban environment, focusing on end-to-end latency and reliability, considering different vehicle densities and speeds." The bidirectional interface and control module receives these inputs and performs preliminary parsing. The system loads the OMNEST basic environment and general model library.

[0176] • Operating environment parameters: User requirements description, optional OMNEST project file.

[0177] • Operating principle / mechanism: The UI passes user input to the interface module, which may use NLU technology to initially understand the user's intent.

[0178] • Function / Significance: To clarify the simulation objectives and provide initial direction for subsequent LM intervention.

[0179] • Termination condition: The system completes basic initialization after the user has finished inputting their requirements.

[0180] • Execution results / technical effects: A structured or semi-structured description of the simulation task, ready to be submitted to LM.

[0181] Step 2: LM Auxiliary Parameters and Scene Generation (S12)

[0182] • Main components: Large model service module, bidirectional interface and control module, knowledge base module.

[0183] • Objects: OMNEST simulation configuration files (.ini, .ned), traffic scripts, and event sequences.

[0184] • Triggering condition: Step 1 is completed, and the simulation task description is obtained.

[0185] • Perform the action:

[0186] (1) The interface module encapsulates the parsed task description and related constraints (if any) into a request to the LM.

[0187] (2) The LM service module receives the request. The LM first understands the core requirements of the task (such as network type V2X, environment city, focus indicators latency / reliability, variable vehicle density / speed).

[0188] (3) LM may query the knowledge base module to obtain prior knowledge, historical success cases or parameter templates related to V2X, city channel model, common mobile model, and related protocols (such as WAVE / DSRC, C-V2X).

[0189] (4) Based on its internal knowledge and information retrieved from the knowledge base, LM generates a set of recommended OMNEST simulation parameters. This may include:

[0190] (A) Topology parameters: such as city map layout (if LM can generate or select a suitable .osm file and guide the conversion to OMNEST topology), base station / RSU deployment strategy, and range of node numbers.

[0191] (B) Channel model parameters: Recommend channel models (such as Winner+ B1) and their key parameters suitable for urban V2X.

[0192] (C) Mobility model parameters: such as the configuration of SUMO and OMNEST integration, vehicle generation rate, maximum speed, acceleration, etc.

[0193] (D) Protocol parameters: such as MAC layer parameters (CSMA / CA window size, number of retransmissions), network layer routing protocol selection, application layer message generation rate and size, etc.

[0194] (E) Scenario elements: For example, LM can design a series of test cases, including different vehicle densities (low, medium, high), different speed distributions, and can even generate specific event scripts, such as simulating sudden congestion in a certain area or the appearance of signal interference sources.

[0195] (5) The output of LM may be a structured list of parameters, a text describing the scene (which will be further processed by the interface module), or a fragment of the OMNEST configuration file directly generated.

[0196] (6) The bidirectional interface and control module translates / converts the output of LM into a configuration file that can be directly used by OMNEST (such as modifying the .ini file, selecting or generating the .ned file, and creating a traffic generation script).

[0197] • Operating environment parameters: LM model type (general or domain-specific), knowledge base content, inference parameters (temperature, top-p, etc.).

[0198] Operating principle / mechanism: The LM utilizes its pre-trained knowledge and potential fine-tuning capabilities to perform context-based reasoning and generation. If the LM has code generation capabilities, it can directly generate parts of the OMNEST configuration script. The RAG mechanism (enhanced generation from knowledge base retrieval) is crucial in this step.

[0199] • Function / Significance: Significantly reduces the complexity and workload of manual simulation configuration, and improves the initial rationality of parameter settings and the richness and relevance of scenarios.

[0200] • Termination condition: An initial configuration file and scene description that can be loaded by OMNEST have been generated.

[0201] • Execution Results / Technical Effects: One or more OMNEST simulation configuration files (for parameter scanning), and possible scenario definition files. Users do not need to start from scratch, improving configuration efficiency and quality.

[0202] Step 3: OMNEST Simulation Startup and Running (S13)

[0203] • Main body: OMNEST simulation core.

[0204] • Object: The network model instantiated based on the configuration generated in S12.

[0205] • Triggering conditions: The configuration file is ready, and the user or system issues a run command.

[0206] • Execution: The OMNEST simulator loads configuration files such as .ini and .ned files, and initializes the network topology, node modules, protocol stack, etc. Then, it begins simulation according to the discrete event scheduling mechanism. During the simulation, each network node communicates, moves, and processes data packets according to its model definition.

[0207] • Operating environment parameters: OMNEST simulation kernel parameters (simulation time limit, number of CPU cores, etc.).

[0208] Operating principle / mechanism: Standard OMNEST discrete event simulation process.

[0209] • Function / Significance: To perform actual network behavior simulation and generate simulation data.

[0210] • Termination conditions: The simulation reaches a preset breakpoint (for LM intervention), a specific event occurs, or a pause / modification command is received from the interface module.

[0211] • Results / Technical Effects: The network state evolves over time, generating various events and statistical data.

[0212] Step 4: Real-time data acquisition and preprocessing (S2)

[0213] Main body: Data acquisition and preprocessing module.

[0214] • Object: The raw data generated during the OMNEST simulation.

[0215] • Triggering conditions: periodic (e.g., every 1 simulation second), event-driven (e.g., high packet loss rate detected), or on-demand acquisition commands issued by the LM through the interface module.

[0216] • Perform the action:

[0217] (1) The module captures network state variables of interest (such as queue length of each node, link utilization, end-to-end delay sample, throughput, node location, battery power, etc.) and event information through the signal / slot mechanism, result recorders, or external script interface provided by OMNEST.

[0218] (2) Preprocess the collected raw data:

[0219] (A) Cleaning: Remove obviously erroneous or invalid data.

[0220] (B) Transformation: Transform the data into a format suitable for LM processing (such as time series data, state vectors, graph structure data). For example, represent network topology and key node states as graph embeddings.

[0221] (C) Aggregation / Sampling: For high-frequency data, it may be necessary to perform time window aggregation (such as calculating the average or maximum value) or downsampling to reduce the amount of data while retaining key information.

[0222] (D) Feature Engineering: Extracting features that characterize the current macroscopic state or specific phenomena of a network. For example, calculating the network congestion index, connection stability indicators, etc.

[0223] 3. After preprocessing, part of the data is sent directly to the interface module for LM analysis, and part of the data (especially historical trend data) can be sent to the knowledge base module for storage.

[0224] • Operating environment parameters: sampling frequency, list of data metrics of interest, and preprocessing algorithm parameters.

[0225] Operating principle / mechanism: It utilizes OMNEST's built-in data output mechanism and external data processing scripts / libraries (such as Python's Pandas, NumPy).

[0226] • Function / Significance: To provide LM with timely, accurate, and easy-to-understand simulation status information.

[0227] • Termination condition: Data acquisition and preprocessing for the current batch are complete.

[0228] • Results / Technical Effects: Structured, information-dense simulation state data packets are available for further analysis by LM.

[0229] Step 5: Real-time LM Analysis and Decision Making (S3)

[0230] Main body: Large model service module.

[0231] • Object: The preprocessed simulation state data from step 4, and contextual information from the knowledge base.

[0232] • Triggering condition: A new simulation status data packet is received.

[0233] • Perform the action:

[0234] (1) The LM receives structured current simulation state data.

[0235] (2) LM combines the initial simulation target and the relevant knowledge stored in the knowledge base (such as the expected behavior pattern in this scenario and the evolution path in similar historical situations) to analyze and evaluate the current state.

[0236] (A) State understanding: LM determines whether the network is in a normal state, a congested state, or a certain (possibly abnormal) pattern.

[0237] (B) Trend prediction: Based on the current state and historical data, LM can predict short-term network behavior trends (such as whether congestion will worsen or whether connections will be interrupted).

[0238] (C) Target compliance assessment: Determine whether the simulation process is progressing in the expected direction, or whether any phenomena worthy of further exploration have emerged.

[0239] (D) (Optional) Predictive accelerated decision-making: If the LM learns that the subsequent evolution path of the current network state is highly predictable and does not contain key information, it can propose to skip a detailed simulation period and directly set to a predicted future state (this requires great care and a verification mechanism).

[0240] (3) The LM makes a decision to determine whether the simulation needs to be intervened (such as adjusting parameters or changing the scene).

[0241] • Operating environment parameters: LM inference mode (e.g., whether to prioritize accuracy or speed), depth of analysis.

[0242] Operating principle / mechanism: LM utilizes its capabilities in pattern recognition, logical reasoning, and sequence prediction. For complex network states, it may be necessary to convert the data into a format that is easier for LM to process (such as text description or graph representation).

[0243] • Function / Significance: It endows the simulation system with the ability of "intelligent perception" and "thinking", which is the core of realizing adaptive simulation.

[0244] • Termination condition: LM completes the analysis of the current state and makes a preliminary decision.

[0245] • Execution results / technical effects: Evaluation conclusions on the current simulation status (such as "network is normal", "potential bottleneck detected", "unexpected high latency", etc.), and judgment on whether intervention is needed.

[0246] Step 6: Determine if dynamic adjustment is needed (S41)

[0247] • Main body: bidirectional interface and control module (or this judgment is directly output by LM).

[0248] • The object: the analysis results and decisions of LM.

[0249] • Triggering condition: LM completes analysis.

[0250] • Execution action: Based on the output of LM, determine whether dynamic adjustments to the OMNEST simulation process are needed.

[0251] If the LM considers the current state to be normal and no special circumstances are found, or the simulation is proceeding as expected, then no adjustments are needed.

[0252] If the LM identifies a problem (such as performance degradation or deviation from the target), an opportunity (such as the emergence of a rare phenomenon worth exploring), or a need to execute pre-defined dynamic scenario changes (such as injecting faults as planned), then adjustments are required.

[0253] Operating principle / mechanism: simple conditional judgment or confidence threshold based on LM output.

[0254] • Function / Significance: Decisions on whether to initiate the intervention process.

[0255] • Termination condition: Make a decision to "adjust" or "not adjust".

[0256] • Execution result / technical effect: A Boolean adjustment flag.

[0257] Step 7: LM generates adjustment instructions (S42) (if S41 is "Yes")

[0258] • Main components: large model service module, bidirectional interface and control module.

[0259] • Objects: OMNEST simulation parameters, module behavior, and event list.

[0260] • Triggering condition: S41 determines that adjustments are needed.

[0261] • Perform the action:

[0262] (1) Based on its analysis in S3 (such as identified root causes of problems and directions to be explored), LM generates specific adjustment strategies and instructions. For example:

[0263] (A) Parameter adjustment: Modify the protocol parameters of a node or a group of nodes (such as TCP congestion window algorithm switching, routing table update).

[0264] (B) Topology changes: Simulate link failure (disconnecting a link), node failure (shutting down a node), or adding a node / link.

[0265] (C) Traffic / load adjustment: Increase / decrease the packet sending rate of certain nodes and change the destination distribution of traffic.

[0266] (D) Behavioral intervention: forcing a module to perform a specific action or changing its internal state machine.

[0267] (E) Scenario Evolution: For example, if the task of LM is to conduct “red team” testing, it may generate new attack vectors and inject instructions based on the current network defense status.

[0268] (2) The instructions output by LM may be high-level descriptions ("Try to reduce congestion in region X") or specific parameter values ​​or operation sequences.

[0269] (3) The bidirectional interface and control module is responsible for translating LM instructions into low-level operations that OMNEST can understand and execute. This may involve:

[0270] (A) Modify the parameters of modules in a running simulation instance (through OMNEST's par() mechanism or direct memory modification, with caution).

[0271] (B) Dynamically schedule new event messages to the event queue.

[0272] (C) Call specific public methods of the OMNEST module.

[0273] (D) For more complex changes, it may be necessary to pause the simulation, modify the configuration file, and then resume or rerun the segment from a checkpoint.

[0274] • Runtime environment parameters: Granularity of LM generated instructions (macro strategy vs. micro parameters).

[0275] Operating principle / mechanism: LM generates solutions based on its "understanding". The interface module acts as both "translator" and "executor".

[0276] • Function / Significance: To realize closed-loop adaptive control in simulation, enabling the simulation to evolve dynamically according to the real-time state.

[0277] • Termination condition: An executable adjustment command for OMNEST has been generated and issued.

[0278] • Results / Technical Effects: When the state or behavior of the OMNEST simulation environment changes, the simulation will continue based on the new conditions. This makes the simulation process itself intelligent and adaptive, enabling it to explore more complex dynamics.

[0279] Step 8: Determine if the simulation has ended (S43)

[0280] • Main body: OMNEST simulation core, bidirectional interface and control module.

[0281] • Object: Simulation time and preset termination conditions.

[0282] • Triggering conditions: Completion of one simulation step (if not adjusted) or completion of one adjusted simulation step.

[0283] • Execute action: Check if the simulation termination conditions are met, for example:

[0284] o Reach the preset maximum simulation time.

[0285] The performance indicators we are interested in have reached a steady state.

[0286] A significant termination event occurs.

[0287] oLM determines that sufficient information has been collected or the exploration objective has been achieved.

[0288] • Operating principle / mechanism: Compare the current simulation time / state with the termination conditions.

[0289] • Function / Significance: To control the termination of the simulation process.

[0290] • Termination condition: Make a judgment of "end" or "not end".

[0291] • Execution result / technical effect: A Boolean end marker. If not terminated, the process returns to step 3 (if adjustments have occurred, start from the adjusted state; otherwise, continue the original trajectory) or step 4 (if no adjustments are needed, collect new data).

[0292] Step 9: LM-assisted result analysis and report generation (S44) (If S43 is "Yes")

[0293] • Main components: large model service module, bidirectional interface and control module, and data acquisition module (providing final data).

[0294] • Object: All data collected throughout the simulation process.

[0295] • Triggering condition: Step S43 determines the end of the simulation.

[0296] • Perform the action:

[0297] (1) The data acquisition module organizes and provides a complete simulation dataset (including configuration parameters, event logs, performance index time series data, LM intervention records, etc.).

[0298] (2) The interface module submits data and analysis requirements (which may come from the user's initial settings or doubts found by the LM in the simulation) to the LM.

[0299] (3) LM performs in-depth analysis of massive amounts of data:

[0300] (A) Statistical summary: Calculate the mean, variance, percentiles, etc. of key performance indicators.

[0301] (B) Pattern recognition: Discover hidden patterns, trends, inflection points, and correlations in data.

[0302] (C) Anomaly detection: Identify abnormal behaviors or performance metrics that do not conform to expectations.

[0303] (D) Causal inference (preliminary): Attempt to explain the cause of certain phenomena (e.g., "high latency is caused by congestion of the X link and insufficient processing capacity of the Y node").

[0304] (E) Comparative analysis: If multiple rounds of simulation are performed (such as parameter scanning or comparison of different strategies), LM can be used for horizontal comparison.

[0305] (4) LM generates analysis reports in a user-friendly manner. This may include:

[0306] (A) Natural language summary statement.

[0307] (B) Automatic generation of key charts (or suggestions and code for generating charts, such as Python Matplotlib scripts).

[0308] (C) Evaluate the achievement of the initial simulation objectives.

[0309] (D) Propose optimization suggestions for network design or parameter configuration.

[0310] • Operating environment parameters: the level of detail in the report and the focus of the analysis.

[0311] Operating principle / mechanism: LM leverages its powerful text generation, data summarization, and (to a certain extent) logical reasoning capabilities. It can transform structured data into natural language descriptions or convert natural language queries into operations on the data.

[0312] • Function / Significance: To efficiently extract valuable insights from complex simulation data, reducing the threshold and workload of manual analysis.

[0313] • Termination condition: Generate analysis report.

[0314] • Results / Technical Effects: A comprehensive simulation analysis report including key findings, charts, and recommendations.

[0315] Step 10: Knowledge Base Update and LM Feedback Learning (S45)

[0316] • Main components: Knowledge base module, large model service module (training / fine-tuning part), bidirectional interface and control module.

[0317] • Subject: A complete record of this simulation (configuration, data, LM decisions, results, analysis report), and the performance evaluation of the LM.

[0318] • Triggering condition: S44 completed.

[0319] • Perform the action:

[0320] (1) The important information of this simulation (such as successful parameter configuration, observed special phenomena and their triggering conditions, effect evaluation of LM intervention, and final analysis report) is structured and stored in the knowledge base module.

[0321] (2) Evaluate the performance of LM in this simulation task (such as the accuracy of parameter recommendations, the effectiveness of scene generation, the rationality of dynamic adjustments, and the depth of result analysis). These evaluations can come from user feedback or be based on certain objective indicators.

[0322] (3) If sufficient feedback data is collected, it can be used to fine-tune or optimize the LM through prompt engineering to improve its performance on similar tasks in the future. For example, if a parameter adjustment suggestion for the LM proves to be very effective, the weight of the decision model can be increased.

[0323] • Operating principle / mechanism: Persistent data storage. For LM learning, mechanisms such as online learning, offline batch fine-tuning, or reinforcement learning from human feedback (RLHF) may be involved.

[0324] • Function / Significance: Enables the accumulation of knowledge and continuous improvement of the system, making the empowering system increasingly "intelligent".

[0325] • Termination condition: The knowledge base is updated and the relevant learning / feedback process is initiated or completed.

[0326] • Results / Technical Effects: The knowledge base is enriched, and LM's professional capabilities are iteratively improved.

[0327] Explanation regarding "real-time":

[0328] In this invention, "real-time" does not refer to hard real-time in the strict sense (i.e., simulation time is completely synchronized with physical time), which is usually difficult to achieve for complex pure software simulations. Rather, it refers to one or more of the following aspects:

[0329] 1. Near real-time interaction / response: The analysis and decision-making process of LM is fast enough (e.g., in seconds or tens of seconds) to allow users or the system to observe the status, obtain the initial judgment of LM, and intervene or adjust during the simulation, thereby achieving an interactive simulation experience instead of a long "submit-wait-get result" cycle.

[0330] 2. Real-time data-driven adaptation: LM can analyze data generated "in real time" by OMNEST (i.e., at the current simulation moment) and quickly apply it to the simulation process, achieving dynamic adaptive adjustment of simulation behavior. This "real time" is relative to the simulation timeline.

[0331] 3. Significantly accelerated to near real-time: Through intelligent optimization of LM (such as parameter selection and predictive skipping of certain calculations), simulation tasks that originally took hours or even days can be completed in a shorter time (such as minutes or hours), so that simulation results can be used more quickly to guide near real-time decision-making (such as fault contingency plan verification in network operation and maintenance).

[0332] This method, through the aforementioned architecture and process, deeply integrates the cognitive, generative, and reasoning capabilities of large models into OMNEST simulation, thereby systematically addressing the shortcomings of existing technologies in terms of intelligence, efficiency, dynamism, and ease of use.

[0333] Example 2

[0334] like Figure 2 As shown, this embodiment also provides a large-model-enabled real-time simulation system for OMNEST communication networks, used to enable large-model-enabled simulation of OMNEST and real-time simulation of communication networks, thereby implementing the method of Embodiment 1, including:

[0335] The OMNEST simulation core (101) is used to perform discrete event simulations of communication network models based on the traditional OMNEST simulation environment.

[0336] The large model service module (102) (LM_Service, LM) has one or more large models built in and hosts, which are used to receive tasks, perform intelligent analysis and decision reasoning and return results; the large models are GPT-4, LLaMA or models finely tuned for the network domain;

[0337] The bidirectional interface and control module (103) (Interface_Ctrl) is used for bidirectional communication between the OMNEST simulation core (101) and the large model service module (102) to exchange data, translate instructions, and perform synchronization control, as well as to parse user instructions and convert them into large model tasks; convert the output of the LM into commands or parameters that can be executed by OMNEST; and manage data flow to ensure synchronization.

[0338] The data acquisition and preprocessing module (104) (Data_Collector) is used to collect and process data from the OMNEST simulation core (101), including: collecting network status data, event information and statistical indicators in real time or periodically during the OMNEST simulation process and performing preprocessing such as cleaning, transformation and feature engineering. The network status data includes node location, link load, queue length, packet loss rate and latency.

[0339] In a preferred embodiment, the system further includes: a knowledge base module for storing and managing the knowledge required and generated for human-computer interaction and LM decision-making, including historical simulation data, network domain knowledge, large model decision records and feedback data, successful parameter configurations, typical scenario definitions, network domain knowledge graphs, and feedback and learning results of the large model; when executing S3, the large model service module (102) queries the knowledge base module to obtain contextual information to enhance its analysis, reasoning, or generation capabilities; after the simulation task is completed, the configuration, key results, large model decisions, and their effect evaluation of this simulation are stored in the knowledge base module, and can be used for feedback learning or fine-tuning of the large model service module; and after the simulation ends, the large model service module analyzes all or part of the simulation data generated by the OMNEST simulation core, automatically generates a natural language analysis report containing key insights, statistical summaries, or visualization suggestions, and stores it in the knowledge base module. This can provide richer context for LM and be used for continuous learning and optimization of LM.

[0340] As a preferred embodiment, the system further includes a graphical interface through which the user interacts with the system, either through the graphical interface or through natural language.

[0341] The core components of this architecture and their functions are as follows:

[0342] 1. OMNEST Simulation Core:

[0343] Main components: OMNEST discrete event simulation engine and INET and other network model libraries.

[0344] Object: The communication network model being simulated (topology, nodes, protocols, traffic, etc.).

[0345] Operating environment: Standard OMNEST operating environment.

[0346] Actions performed: Simulation events are executed sequentially according to the event list, network status is updated, and performance metrics are calculated. Instructions are received from the "Bidirectional Interface and Control Module" to dynamically modify simulation parameters, network topology, module behavior, or inject events.

[0347] Function: To perform low-level, precise simulation of communication network behavior.

[0348] 2. Large Model Service Module (LM Service):

[0349] Main body: One or more pre-trained general-purpose large models (such as GPT series, Claude series, LLaMA series, etc.) or specialized large models fine-tuned with data from the communication network domain. Includes the model itself and an efficient inference service backend.

[0350] Objects: Various task requests from the "bidirectional interface and control module", such as parameter optimization suggestions, scene generation descriptions, behavior prediction requests, result analysis instructions, etc.

[0351] Operating environment: Typically a high-performance computing cluster (such as a GPU server) that provides services through an API interface.

[0352] Execution Action: Based on the input task and contextual information (which may come from a knowledge base), it utilizes its powerful understanding, reasoning, and generation capabilities to process and output decision suggestions, configuration parameters, text descriptions, code snippets, or analysis reports.

[0353] Function: To provide advanced intelligent support for the simulation process, such as decision-making, generation, prediction, and understanding.

[0354] 3. Two-way interface and control module:

[0355] Main components: a set of software interfaces, protocol converters, data formatting tools, and synchronization control logic.

[0356] Objects: OMNEST simulation core, large model service module, data acquisition module, knowledge base, and user interface.

[0357] Triggering conditions: user commands, specific OMNEST events, periodic polling, LM service return results, etc.

[0358] Execution actions include: parsing and translating high-level user requirements (potentially in natural language) into specific tasks for the LM; formatting OMNEST real-time status data obtained from the Data_Collector and submitting it to the LM_Service along with context information provided by Knowledge_Base; receiving output from the LM_Service (such as parameter values, scenario scripts, and control logic) and translating it into commands recognizable by OMNEST (such as NED file modification, INI file parameter setting, C++ module method calls, message injection, etc.); delivering analysis reports or visualization suggestions generated by the LM_Service to the user interface; and coordinating the synchronization between OMNEST simulation steps and LM inference to ensure data consistency and timely response. For example, a mechanism can be implemented to pause the simulation, wait for the LM decision, and then continue.

[0359] Operating principle: It communicates with LM_Service through well-defined APIs (such as RESTful API, gRPC); it interacts with OMNEST through external interfaces of OMNEST (such as Python script bridges, Tcl / Tk extensions, or by directly modifying the configuration file and restarting / reloading some models).

[0360] Function: To achieve decoupling and efficient collaboration between OMNEST and LM, it is the "central nervous system" of the entire empowerment system.

[0361] 4. Data Acquisition and Preprocessing Module (Data Collector & Preprocessor):

[0362] Main body: Software modules integrated within OMNEST or used as external monitoring processes.

[0363] Objects: Various raw data (such as signals, statistics, logs, and event trajectories) during the OMNEST simulation process.

[0364] Triggering conditions: OMNEST simulation event occurs, simulation time reaches the preset sampling point, or data is collected as needed according to LM command.

[0365] Actions performed: Listen to and capture various types of data output by OMNEST. Perform preprocessing operations on the raw data, such as cleaning (removing noise, handling missing values), transformation (e.g., converting enumerated values ​​to numerical values), aggregation (e.g., calculating window averages), and feature extraction (e.g., extracting statistical features and frequency domain features from time-series data), to make it structured / semi-structured data suitable for LM input or storage in the knowledge base.

[0366] Function: To provide high-quality, real-time simulation status information for LM and to accumulate data for the knowledge base.

[0367] 5. Knowledge Base Module:

[0368] Main body: Database system (such as relational database, NoSQL database, graph database, vector database).

[0369] Objects: historical simulation configurations and results, network topology templates, protocol specification documents, domain ontology knowledge, user feedback, high-quality content generated by LM, or patterns learned.

[0370] Operating environment: persistent storage server.

[0371] Execution actions include storing, retrieving, managing, and handling various types of knowledge and data related to communication network simulation. In particular, the vector database can be used to store embedded vectors of text and code snippets, supporting fast retrieval based on semantic similarity and providing LM with RAG (Retrieval Augmented Generation) capabilities.

[0372] Functions: To provide LM with rich background knowledge and context, enhancing the accuracy and relevance of its decisions; to record LM's successes and failures for its continuous learning and evolution; and to accumulate simulation knowledge assets within the organization.

[0373] The present invention also provides a memory that stores multiple instructions for implementing the method as described in Embodiment 1.

[0374] like Figure 3 As shown, the present invention also provides an electronic device, including a processor 301 and a memory 302 connected to the processor 301. The memory 302 stores a plurality of instructions, which can be loaded and executed by the processor to enable the processor to perform methods as described in Embodiments 2 and 3.

[0375] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for real-time simulation of communication networks using a large model in OMNEST, characterized in that, include: S1 establishes a bidirectional communication connection between the OMNEST simulation core and one or more large model service modules through a bidirectional interface and control module. S2, the large model service module receives user-defined simulation task descriptions or simulation status data collected in real time from the OMNEST simulation core; wherein the simulation task description is in natural language form, and the large model service module uses natural language understanding capabilities to convert it into structured simulation objectives and constraints; S3, the large model service module analyzes, infers or generates based on the task description or state data, and outputs simulation parameter configuration suggestions, simulation scenario descriptions, simulation process adjustment instructions or simulation result analysis. The large model service module learns the behavior patterns of simulation data, predicts the results of some event sequences in the simulation, and generates instructions to guide the OMNEST simulation core to skip the corresponding detailed calculations, thereby accelerating the simulation process. The simulation parameter configuration suggestions and simulation scenario descriptions generated by the large model service module are converted into configuration files or scripts that the OMNEST simulation core can read, for initializing the OMNEST simulation. S4, apply the output of the large model service module to the OMNEST simulation core to initialize the simulation configuration, dynamically adjust the simulation process, or process the simulation results; wherein, the large model service module dynamically generates simulation process adjustment instructions based on real-time simulation status data, and the simulation process adjustment instructions are sent to the OMNEST simulation core through a bidirectional interface and control module to modify network model parameters, topology, node behavior, or inject events during simulation.

2. The method for real-time simulation of communication networks using a large model-enabled OMNEST according to claim 1, characterized in that, S1 includes: S11, Simulation Task Definition and Initialization: Users define the simulation purpose, performance metrics of interest, and network type through a graphical user interface (UI) that supports natural language; system initialization and model loading; S12, Large Model Auxiliary Parameters and Scene Generation: The large model generates initial OMNEST simulation parameters and scene descriptions based on user input and knowledge base information. The initial OMNEST simulation parameters and scene descriptions include the dynamic generation of the topology file .ned and the traffic script. S13, OMNEST Simulation Startup and Run: The OMNEST simulator loads its configuration and begins executing discrete event simulations.

3. The method for real-time simulation of communication networks using a large model-enabled OMNEST according to claim 2, characterized in that, The S2 includes: real-time data acquisition and preprocessing, including: during simulation operation, the data acquisition and preprocessing module acquires network status data, performance indicators, and event logs according to a preset strategy or large model instructions, and performs preprocessing; and the simulation status data acquired in real time from the OMNEST simulation core is processed by the data acquisition and preprocessing module, key features are extracted and formatted, and then provided to the large model service module.

4. The method for real-time simulation of communication networks using a large model-enabled OMNEST according to claim 3, characterized in that, The S3 includes: real-time analysis and decision-making of the large model service module, including: The preprocessed data is transmitted to the large model service module via a bidirectional interface and control module. The large model service module analyzes the current simulation status to obtain analysis results, which are used to determine whether the simulation meets expectations, whether there are any anomalies, or whether a certain critical node has been reached.

5. The method for real-time simulation of communication networks using a large model-enabled OMNEST according to claim 4, characterized in that, S4 includes: S41, Determine if dynamic adjustment is needed: The large model service module determines whether intervention in the simulation process is needed based on the analysis results; S42, Large Model Generates Adjustment Instructions: If adjustments are needed, the large model service module generates specific adjustment instructions, including modifying the parameters of a node, changing the state of a link, or injecting new traffic or events; the bidirectional interface and control module translates the instructions and sends them to the OMNEST simulation core; the OMNEST simulation core executes the adjustments and continues to run; S43, Determine whether the simulation has ended: Determine whether the simulation has reached the preset end conditions, the end conditions include simulation time, occurrence of a specific event, or stable performance indicators; S44, Large Model-Assisted Result Analysis and Report Generation: After the simulation is completed, the large model service module performs in-depth analysis on all collected simulation data, extracts key insights, generates summary reports, visualization charts, and proposes optimization suggestions; S45 stores the configuration, key results, decision-making process and effects of this simulation into the knowledge base module; if some decisions of the large model bring good / bad results, this information is used as feedback for subsequent fine-tuning of the large model or to prompt engineering optimization.

6. A real-time simulation system for communication networks that enables large-scale model empowerment for OMNEST simulation and real-time simulation of communication networks, thereby implementing the method described in any one of claims 1-5, comprising: The OMNEST simulation core is used to perform discrete event simulations of communication network models based on the traditional OMNEST simulation environment. The large model service module has one or more large models built in and hosted, which are used to receive tasks, perform intelligent analysis and decision reasoning, and return results; the large models are GPT-4, LLaMA, or models finely tuned for the network domain. The bidirectional interface and control module is used for bidirectional communication between the OMNEST simulation core and the large model service module, including data exchange, command translation, synchronization control, parsing user commands and converting them into large model tasks; converting the output of the large model into commands or parameters executable by OMNEST; and managing data flow to ensure synchronization. The Data Acquisition and Preprocessing Module (Data_Collector) is used to collect and process data from the OMNEST simulation core. This includes: collecting network status data, event information, and statistical indicators in real time or periodically during the OMNEST simulation process and performing preprocessing such as cleaning, transformation, and feature engineering. The network status data includes node location, link load, queue length, packet loss rate, and latency.

7. The real-time simulation system for communication networks empowered by a large model for OMNEST according to claim 6, characterized in that, The system further includes: a knowledge base module for storing and managing the knowledge required and generated for human-computer interaction and large model decision-making. This knowledge includes historical simulation data, network domain knowledge, large model decision-making records and feedback data, successful parameter configurations, typical scenario definitions, network domain knowledge graphs, and feedback and learning outcomes of the large model. When executing S3, the large model service module queries the knowledge base module to obtain contextual information to enhance its analysis, reasoning, or generation capabilities. After the simulation task is completed, the configuration, key results, large model decisions, and their effectiveness evaluation for this simulation are stored in the knowledge base module, and can be used for feedback learning or fine-tuning of the large model service module. After the simulation ends, the large model service module analyzes all or part of the simulation data generated by the OMNEST simulation core, automatically generates a natural language analysis report containing key insights, statistical summaries, or visualization suggestions, and stores it in the knowledge base module.

8. A real-time simulation system for communication networks empowered by a large model for OMNEST, as described in claim 7, is characterized in that, The system also includes a graphical interface, through which users interact with the system using either a graphical interface or natural language.

9. An electronic device comprising a processor and a memory, the memory storing a plurality of instructions, the processor being configured to read the instructions and execute the method as described in any one of claims 1-5.

10. A computer-readable storage medium storing a plurality of instructions, the plurality of instructions being readable by a processor and executed as described in any one of claims 1-5.

Images