Semi-physical network equivalent simulation method and system for multi-mode communication

Abstract

The invention discloses a semi-physical network equivalent simulation method and system oriented to multi-mode communication, and relates to the technical field of semi-physical simulation. The method comprises the following steps: collecting a multi-mode communication data flow through a probe deployed in a physical communication network; a network behavior phase space is reconstructed through Takens embedding, and topological and dynamic invariant core features are extracted; simulation platform resources are abstracted into a virtual power architecture, communication behavior equivalent mapping is carried out with core features as constraints, and an equivalent simulation communication network is generated. The technical problem that a simulation result is dynamically mismatched with an actual network due to the fact that cross-plane interaction behaviors in a multi-modal communication network are difficult to accurately capture by an existing semi-physical simulation method is solved, and the purposes that topology and dynamic invariants are extracted through phase-space reconstruction to achieve cross-scale behavior equivalent mapping are achieved. And elastic resource scheduling is completed in combination with a virtual resource power architecture abstracted by a difference equation, so that the technical effects of high-precision network behavior reproduction and efficient resource utilization are achieved.

Description

Technical Field

[0001] This invention relates to the field of hardware-in-the-loop simulation technology, specifically to a hardware-in-the-loop network equivalent simulation method and system for multimodal communication. Background Technology

[0002] With the development of technologies such as 5G / 6G, satellite communication, the Internet of Things, and cloud-network convergence, communication networks are exhibiting multimodal, cross-plane, strongly coupled, and highly dynamic characteristics. Signaling planes, user planes, and control planes coexist within these networks, and heterogeneous data from multiple sources are highly intertwined across spatiotemporal scales, resulting in network behavior with significant nonlinear and complex dynamic characteristics. High-fidelity modeling, verification, and risk assessment of these complex communication networks have become critical issues in network planning, operation and maintenance, and security assurance.

[0003] Existing communication network simulation methods mainly include pure software simulation, mathematical modeling simulation, and hardware-in-the-loop (HIL) simulation. Pure software simulation relies on idealized models, making it difficult to reflect the actual characteristics of real networks, such as cross-plane coupling, time-series dependencies, and resource constraints. Traditional HIL simulations are typically based on topology or traffic-level mappings, focusing more on the consistency of link, node, or protocol layer parameters, and failing to characterize and maintain the core dynamics of real networks from the perspective of overall network behavior. When network scale increases or communication modes diversify, these methods often face problems such as high simulation resource consumption, insufficient equivalent accuracy, and limited scalability. On the other hand, full physical reproduction of actual communication networks is costly, and it is difficult to flexibly reproduce extreme conditions, abnormal evolution, and risk propagation processes in experimental environments, which also limits the ability to assess network evolution and verify risk control. Summary of the Invention

[0004] This application provides a semi-physical network equivalent simulation method and system for multimodal communication, which solves the technical problem that existing semi-physical simulation methods are unable to accurately capture cross-plane interaction behavior in multimodal communication networks, resulting in a dynamic mismatch between simulation results and actual networks.

[0005] The first aspect of this application provides a semi-physical network equivalent simulation method for multimodal communication, the method comprising: Multimodal communication data streams are collected using probes deployed in the physical communication network. These multimodal communication data streams include time-series data from the signaling plane, user plane, and control plane. The multimodal communication data streams are then reconstructed into a high-dimensional network behavior phase space using Takens embedding. Core features are extracted from this network behavior phase space, where the core features are topological and dynamic invariants that characterize the core network behavior and do not change with the observation scale. By abstracting the resource structure of the hardware-in-the-loop simulation platform into a virtual resource dynamic architecture, and using the core features as constraints, an equivalent mapping of communication behavior is performed to generate an equivalent simulated communication network. This is described using difference equations, and the equivalent mapping includes behavior-resource phase space mapping and elastic resource allocation scheduling.

[0006] A second aspect of this application provides a semi-physical network equivalent simulation system for multimodal communication, the system comprising: Data Acquisition Module: Acquires multimodal communication data streams through probes deployed in the physical communication network. These multimodal data streams include time-series data from the signaling plane, user plane, and control plane. Feature Extraction Module: Reconstructs the multimodal communication data streams into a high-dimensional network behavior phase space using Takens embedding. Extracts core features from this phase space, where the core features are topological and dynamic invariants characterizing core network behaviors and remaining constant regardless of the observation scale. Behavior Mapping Module: Abstracts the resource structure of the hardware-in-the-loop simulation platform into a virtual resource dynamic architecture. Using the core features as constraints, performs an equivalent mapping of communication behaviors to generate an equivalent simulated communication network. This is described using difference equations, and the equivalent mapping includes behavior-resource phase space mapping and flexible resource allocation scheduling.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, probes are deployed in a real communication network to collect communication time-series data across multiple planes, including signaling, user, and control. Then, based on the collected multimodal data, the Takens embedding method is used to reconstruct the network's high-dimensional behavioral phase space, extracting key topological and dynamic features that stably reflect the network's essential operational state. Next, the computing, link, and other resources in the hardware-in-the-loop simulation platform are abstracted into an evolvable virtual resource dynamic architecture. While maintaining the aforementioned core features, the communication behavior of the real network is equivalently mapped to the simulation environment. The evolutionary relationship between behavior and resources is described using difference equations. Combined with behavior-resource phase space mapping and elastic resource scheduling, an equivalent simulated communication network is constructed to support high-precision simulation verification and resource optimization. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of the process for a semi-physical network equivalent simulation method for multimodal communication provided in an embodiment of this application.

[0010] Figure 2 This is a schematic diagram of the equivalent simulation system structure of a semi-physical network for multimodal communication provided in an embodiment of this application.

[0011] Figure labeling: Data acquisition module 11, feature extraction module 12, behavior mapping module 13. Detailed Implementation

[0012] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0013] Example 1, as Figure 1 As shown, this application provides a semi-physical network equivalent simulation method for multimodal communication, wherein the method includes: Multimodal communication data streams are collected by probes deployed in physical communication networks. These multimodal communication data streams include timing data from the signaling plane, user plane, and control plane.

[0014] In this embodiment, probe devices or probe programs are deployed at key locations in the physical communication network to collect multimodal communication data online during network operation. These probes can access the core network, bearer network, edge gateway, base station side interface, network element mirror port, or observation interface of virtualized network functions, and acquire data through port mirroring, bypass splitting, log subscription, protocol stack interface capture, flow table export, or telemetry. The collected data streams cover at least the signaling plane, user plane, and control plane simultaneously. The signaling plane time-series data includes, but is not limited to, signaling messages such as access authentication, session establishment / release, handover, and route update, as well as their key fields, message direction, associated identifiers, and triggering reasons. The user plane time-series data includes, but is not limited to, traffic and performance indicators such as arrival time, packet length, 5-tuple, throughput, latency, jitter, packet loss, retransmission, and queue length of service data packets. The control plane time-series data includes, but is not limited to, control commands and status feedback such as resource orchestration and scheduling events, policy issuance, QoS configuration, slice / bearer establishment parameters, network element status changes, alarms, and counter changes. To ensure cross-plane data correlation, the probe uniformly timestamps and synchronizes the collected data with clocks based on NTP or PTP, and extracts key identifiers for correlation, such as session ID, tunnel identifier, and user identifier, and then forms continuous time-series records according to a preset sampling period. Finally, the collected communication data is encapsulated into a multimodal communication data stream according to a unified data structure to characterize the dynamic evolution process of the physical communication network on different planes. This provides a complete and synchronous observational basis for subsequent in-depth analysis of the network's intrinsic dynamic behavior, ensuring the accuracy and reliability of the simulation.

[0015] For the multimodal communication data stream, it is reconstructed into a high-dimensional network behavior phase space through Takens embedding, and the core features in the network behavior phase space are extracted. The core features are topological and dynamic invariants that characterize the core network behavior and do not change with the observation scale.

[0016] In one embodiment, for the acquired multimodal communication data stream, the time-series data of the signaling plane, user plane, and control plane are first uniformly aligned and preprocessed. The data from each plane are synchronized by timestamp, missing, abnormal spikes, and duplicate records are removed, and indices of different dimensions are normalized, such as by max-min normalization. Subsequently, an observation sequence characterizing the network's operational state is constructed from the multimodal data. This observation sequence can consist of a single comprehensive scalar sequence or a multivariate sequence. The comprehensive scalar sequence can be obtained by fusing throughput, latency, packet loss, signaling interaction frequency, and control event intensity according to preset weights. The multivariate sequence can be a vector time series composed of the aforementioned indices. Afterward, Takens embedding reconstruction is performed. During this process, the time delay parameter and embedding dimension are determined. The time delay parameter is determined by the first zero-crossing of the autocorrelation function or the first minimum value of the mutual information function, and the embedding dimension is determined by the proportion of false nearest neighbors falling below a threshold. After obtaining the time delay parameters and embedding dimensions, the observation sequence is mapped to a high-dimensional point set using the delay coordinate method. Specifically, an embedding vector is constructed for any given time step, and all embedding vectors form the network behavior phase space. For multivariate sequences, delay components are constructed for each variable and concatenated into a unified embedding vector. Then, to verify the effectiveness of the phase space reconstruction, a consistency check is performed on the network behavior phase space based on the geometric structure and key dynamic indicators. After passing the check, core features are extracted from this high-dimensional network behavior phase space. These core features are defined as topological and dynamic invariants that remain stable under different sampling frequencies, observation windows, or indicator scaling conditions, including at least the correlation dimension, Kolmogorov entropy, and principal component vectors. Finally, these core features are used as constraint inputs for subsequent equivalent mappings, ensuring that the equivalent network generated on the simulation side maintains an intrinsic dynamic structure consistent with the real physical network at the behavioral level.

[0017] Furthermore, the network behavior phase space is reconstructed through Takens embedding, including: For the multimodal communication data stream, time delay parameters and state space dimensions are determined; based on the time delay parameters and state space dimensions, the original variable time series is transformed into a set of points in a high-dimensional space using delay coordinates, forming the network behavior phase space, where each point represents the complete communication state at any given time; the network behavior phase space is verified based on the geometric structure and key dynamic indicators, with the intrinsic dynamic characteristics of the equivalent physical communication network as the verification target.

[0018] Preferably, for the obtained multimodal communication data stream, the timing data from the signaling plane, user plane, and control plane are first aligned to a unified time reference during the preprocessing stage. This alignment includes resampling based on a unified network clock or logical timestamp, ensuring a one-to-one correspondence between data from different planes under the same sampling sequence. Simultaneously, outlier removal, missing value compensation, and normalization are performed on the original data to eliminate the influence of different dimensions and sampling densities on subsequent phase space reconstruction. Based on this, an observation sequence characterizing the overall operating state of the communication system is constructed from the multimodal communication data stream. This observation sequence can be a single comprehensive state variable or a multivariate time series composed of multiple key state variables. Subsequently, a time delay parameter and state space dimension are determined for the observation sequence. This time delay parameter is used to capture the effective dependencies between data points before and after the communication system state evolves over time. Its selection principle is to reflect system state changes between adjacent delay components without introducing redundant information. Specifically, by calculating the autocorrelation function or average mutual information function of the observed sequence, the time interval corresponding to the first significant decay of the autocorrelation function or the first local minimum of the average mutual information function is selected as the time delay parameter. This state space dimension is used to determine the minimum embedding dimension that can fully unfold the dynamic characteristics of the communication system. Its selection aims to avoid overlap between different system states in the embedding space. Specifically, a pseudo-nearest neighbor analysis method can be used to gradually increase the embedding dimension and calculate the pseudo-nearest neighbor ratio. When the ratio is lower than a preset threshold, the current dimension is determined as the state space dimension.

[0019] After obtaining the time delay parameters and state space dimensions, the original variable time series is reconstructed using a delay coordinate representation method. In this process, for any time t, the current observation and its preceding delayed observations are selected sequentially according to the time delay parameters to construct an embedding vector containing the components corresponding to each state space dimension. For multivariate observation sequences, delay components are constructed for each variable and concatenated in a predetermined order to form a unified high-dimensional embedding vector. Then, the embedding vectors corresponding to all times constitute a point set in the high-dimensional space. This point set evolves over time to form a continuous trajectory, thus constituting the network behavior phase space. Each point in the network behavior phase space corresponds to the complete communication state of the physical communication network at that time, under the combined action of signaling, user, and control multi-plane information.

[0020] After constructing the phase space, the validity of the network behavior phase space is verified to confirm whether it truly reflects the inherent dynamic characteristics of the equivalent physical communication network. The verification process includes two levels: geometric structure analysis and key dynamic index evaluation. At the geometric structure level, the phase space trajectory is visualized and quantitatively analyzed to determine whether it exhibits a clear, stable, and repeatable attractor pattern. Specifically, several typical two-dimensional or three-dimensional subspaces are selected from the constructed high-dimensional network behavior phase space for projection display. The subspace can be composed of adjacent embedded dimensional components, principal axis directions obtained from principal component analysis, or randomly selected dimensional combinations. By comparing the trajectory distribution patterns under different projection perspectives, it is determined whether the phase space trajectory is concentrated within a finite region and whether it exhibits a continuous, non-divergent, and non-randomly scattered geometric structure, in order to preliminarily determine whether an attractor structure exists. Based on this, the spatial envelope volume, point density distribution, and their changes over time of the phase space trajectory point cloud are statistically analyzed to determine whether the trajectory remains bounded and structurally stable during long-term evolution. Then, the degree of overlap or similarity of the trajectory in phase space within different time periods is calculated. For example, by calculating the differences in distance distribution, Hausdorff distance, or occupancy probability distribution based on grid partitioning between trajectory point sets, it is determined whether the trajectory repeatedly returns to the same geometric region under the same or similar operating conditions, thereby verifying the repeatability of the attractor structure. Furthermore, to verify the consistency of the attractor structure under different sampling windows or observation scales, the network behavior phase space is reconstructed for the same multimodal communication data stream using different time window lengths, different sampling intervals, or different index scaling methods. The distribution characteristics and main topological structures of the attractors reconstructed under different conditions after morphological and scale normalization are compared to see if they remain consistent. For example, the similarity of the attractor projection shape, the degree of overlap of the main point density distribution region, or the range of changes in key geometric statistics are compared. When the attractor geometry remains stable in its overall shape and main topological relationships under the different observation conditions mentioned above, it is determined that the constructed network behavior phase space has scale invariance and structural consistency, and can truly reflect the intrinsic dynamic behavior of the equivalent physical communication network.

[0021] At the level of key dynamic indicators, based on the constructed network behavior phase space, a phase space point corresponding to any given moment is selected as the initial state. Within its neighborhood, several nearest neighbors with the smallest distance are chosen; these nearest neighbors correspond to states in the communication system that have similar historical evolution characteristics in the state space. Then, starting from the initial state, a short-term prediction method based on the local evolution law of the phase space is used to predict the state for several subsequent time steps. This prediction method can be a local linear approximation, a mapping based on the average evolution of nearest neighbors. By comparing the predicted phase space trajectory with the actual trajectory reconstructed from the multimodal communication data stream, the prediction error index under different prediction step lengths is calculated using Euclidean distance. Then, by statistically analyzing the trend of prediction error as the prediction step length increases, the predictability decay rate of the communication system behavior is obtained. Finally, this predictability decay rate is compared with the behavioral characteristics of the physical communication network under the corresponding operating state. When the geometric structural features and key dynamic indicators are consistent with the physical communication network in terms of morphology, trend, or magnitude, it is determined that the reconstructed network behavior phase space can effectively characterize the real dynamic behavior of the communication system, providing a reliable behavioral basis for subsequent core feature extraction and semi-physical equivalent simulation mapping.

[0022] Furthermore, the core features include at least the correlation dimension, Kolmogorov entropy, and principal component vectors; wherein, the correlation dimension is used to quantify the complexity and degree of freedom of network communication behavior, the Kolmogorov entropy characterizes the rate of unpredictability of network information generation or behavior, and the principal component vectors identify the main direction of behavior evolution.

[0023] Preferably, after obtaining and verifying the network behavior phase space, core features for characterizing the core operational characteristics of the communication network are extracted from the network behavior phase space. These core features include at least the correlation dimension, Kolmogorov entropy, and principal component vectors, used to characterize network communication behavior from different aspects such as complexity, unpredictability, and dominant causation direction. Among them, the correlation dimension is used to quantify the complexity and effective degrees of freedom of network communication behavior, and its calculation is based on the spatial distribution characteristics of the trajectory point set in the network behavior phase space. Specifically, in the network behavior phase space, a certain number of phase space points are selected as reference points, and the number of neighborhood points of each reference point under different scale radius conditions is counted. The proportion of point pairs in the phase space with a distance less than a given scale is calculated. By analyzing the power-law relationship of this proportion with scale, the fractal dimension of the phase space trajectory, i.e., the correlation dimension, is estimated. This correlation dimension reflects the minimum number of state variables required by the communication system at the behavioral level. When the correlation dimension is large, it indicates that the network behavior is complex and has high degrees of freedom; when the correlation dimension is small, it indicates that the network behavior is constrained by fewer dominant mechanisms. To ensure the robustness of the results, the correlation dimension can be repeatedly calculated under different sampling windows and different embedding parameters, and the results can be statistically averaged or interval analyzed.

[0024] Kolmogorov entropy is used to characterize the rate of information generation or the unpredictability of system state evolution in network communication behavior. In the network behavior phase space, by tracking the separation of adjacent trajectories over time, and statistically estimating the rate of increase in state uncertainty over time in the phase space, the average information generation rate of the system can be calculated. This estimate of the average information generation rate can be approximated based on the local divergence characteristics of trajectories in the phase space, symbolic state transition sequences, or the trend of the increase in nearest neighbor prediction errors over time. A high Kolmogorov entropy indicates rapid changes in communication network behavior and difficulty in long-term prediction; a low value indicates relatively stable network operation and strong predictability.

[0025] Principal component vectors (PCVs) are used to identify the main direction of network behavior evolution. They are obtained based on statistical analysis of the set of embedded vectors in the network behavior phase space. Specifically, PCV analysis is performed on the dataset consisting of all embedded vectors in the network behavior phase space, and then eigenvalue decomposition is performed on the covariance matrix to obtain a set of PCVs sorted by the magnitude of explained variance. Next, a set of PCVs whose cumulative explained variance reaches a preset threshold is selected as the feature subspace describing the main direction of network behavior evolution. This set of PCVs reflects the combination patterns with the largest changes and the greatest contribution to overall behavior in multimodal communication data, and can be used to identify key factors dominating network evolution, such as the impact of changes in service load, signaling interaction intensity, or control strategy adjustments on system behavior.

[0026] Finally, the correlation dimension, Kolmogorov entropy, and principal component vectors are used as the core feature outputs of the network behavior phase space to constrain the equivalent mapping process between subsequent communication behaviors and hardware-in-the-loop simulation resources, thereby ensuring that the simulated network is consistent with the physical communication network in key behavioral characteristics such as complexity level, dynamic uncertainty, and main evolution modes.

[0027] By abstracting the resource structure of the hardware-in-the-loop simulation platform into a virtual resource dynamic architecture, and using the core features as constraints to perform equivalent mapping of communication behavior, an equivalent simulation communication network is generated. The network is described by difference equations, and the equivalent mapping includes behavior-resource phase space mapping and elastic resource allocation and scheduling.

[0028] In one embodiment, after obtaining the network behavior phase space and its core features, the resources available for network simulation in the hardware-in-the-loop simulation platform are first modeled. These resources include at least computing resources, storage resources, link bandwidth resources, forwarding and processing capabilities, protocol stack instances, and virtualized network function instances. By abstracting these into state variables that evolve over time, resource occupancy, available capacity, scheduling priority, and coupling relationships between resources are characterized, thus constituting a virtual resource dynamic architecture and its corresponding resource phase space. In this virtual resource dynamic architecture, a difference equation is used to abstractly describe the evolution process of resource states. This difference equation takes the current resource state, scheduling decisions, and external load as inputs and the resource state at the next time step as output, and is used to characterize the dynamic changes of resources during the simulation process. Subsequently, using the aforementioned extracted core features of the network behavior phase space as constraints, an equivalent mapping relationship between the network behavior phase space and the resource phase space is established, ensuring that the evolution trajectory in the resource phase space is consistent with the behavior phase space of the real communication network in terms of topology, complexity level, and dynamic characteristics. Next, based on the target baseline obtained from the behavior-resource phase space mapping, the resources in the hardware-in-the-loop simulation platform are dynamically scheduled and configured, including starting, stopping, scaling up, or adjusting parameters of computing instances, link bandwidth, processing queues, and protocol function modules. By iteratively executing resource state updates and scheduling decisions, the evolution trajectory of the resource phase space gradually approaches the target baseline. Thus, under limited hardware-in-the-loop resource conditions, a simulated communication network equivalent to the physical communication network at the behavioral level is constructed, enabling it to realistically reflect the operational characteristics of the actual communication network in terms of complexity, dynamic evolution characteristics, and main behavioral patterns.

[0029] Furthermore, in the equivalent mapping of communication behaviors constrained by the aforementioned core features, the behavior-resource phase space mapping includes: By abstracting the resource structure of the hardware-in-the-loop simulation platform into a virtual resource dynamic architecture, a resource phase space is constructed. With the key topological and dynamic invariants as constraints, points in the network behavior phase space are mapped to the resource phase space to determine the network equivalent phase space.

[0030] Preferably, after constructing the network behavior phase space and extracting core features, the resource structures of the hardware-in-the-loop simulation platform, such as computing, storage, links, interfaces, and scheduling capabilities, are uniformly abstracted to construct a virtual resource dynamic architecture for resource regulation, and a resource phase space is built on this basis. Specifically, various physical or virtual resources in the hardware-in-the-loop simulation platform are regarded as state variables. These state variables include at least processing capacity utilization, link bandwidth utilization, queue length, forwarding table entry size, number of virtual network function instances, and scheduling cycle parameters. Resource state vectors are formed by combining the above resource states, and the resource phase space is composed of all possible resource state vectors. Each point in the resource phase space represents the overall resource configuration and operating state of the hardware-in-the-loop simulation platform at a certain moment. Based on this, while maintaining key topological and dynamic invariants, behavioral feature points that can represent typical communication states are selected from the constructed network behavior phase space in the physical communication network. These behavioral feature points are used to characterize the core behavioral patterns of the communication system under different load levels, control strategies, or operating stages. Subsequently, based on the pre-established behavior-resource correlation, points in the behavior phase space are projected onto the phase space of the virtual resource dynamic architecture. This ensures that the corresponding states in the resource phase space maintain consistency with the original network behavior in terms of topology and dynamic characteristics. This projection process is constrained by preserving key invariants in the behavior phase space. These invariants include at least the correlation dimension, Kolmogorov entropy, and the complexity level, unpredictability, and main evolution direction represented by the principal component vectors. This ensures that the virtual resource dynamic architecture can equivalently reproduce the core operational characteristics of the physical communication network at the behavioral level. In this way, key behavioral features in the physical communication network are abstracted into points in the behavior phase space and further mapped onto the virtual resource dynamic architecture constructed for resource regulation, forming a network equivalent phase space. This network equivalent phase space serves as a unified description basis for communication behavior and resource scheduling in the hardware-in-the-loop simulation environment, providing a controllable, adjustable, and dynamically consistent state representation for subsequent elastic resource allocation scheduling and equivalent simulation operation.

[0031] Furthermore, in the communication behavior equivalence mapping constrained by the aforementioned core features, elastic resource allocation and scheduling includes: Obtain the logical entity of the physical communication network as the target baseline for resource scheduling; to minimize the deviation between the evolution trajectory of the network equivalent space and the target baseline, perform iterative optimization of resource allocation and scheduling in the network equivalent space under the physical resource state based on a semi-physical simulation platform, and generate the equivalent simulated communication network.

[0032] Optionally, after constructing the network equivalent phase space, logical entities directly related to communication behavior are obtained from the physical communication network and equivalently represented as desired trajectory states or desired trajectories in the network equivalent phase space, serving as the target baseline for resource allocation and scheduling. These logical entities include at least one or more of communication sessions, service flows, and Virtual Network Function (VNF) instances. A communication session represents a continuous communication relationship formed by a user or terminal within a certain time period; a service flow represents a set of messages with unified forwarding and quality of service characteristics; and a Virtual Network Function represents a schedulable logical unit carrying a specific network function. For different types of logical entities, key operational characteristics in the physical communication network are extracted, including but not limited to session duration, service flow bandwidth and latency requirements, VNF processing load, and instance status. These characteristics are mapped to target state points or time-evolving target trajectories in the network equivalent phase space to characterize the behavioral evolution path of the logical entities under ideal or desired operating conditions. Based on this, the desired trajectory state or desired trajectory is used as the target baseline for resource scheduling. The deviation of the current evolution trajectory in the network equivalent phase space is measured. This deviation measure is used to characterize the difference between the equivalent network behavior and the target baseline under the current resource configuration of the hardware-in-the-loop simulation platform. It can be quantified by calculating the distance, direction deviation or cumulative error between the actual trajectory and the desired trajectory in the equivalent phase space at the corresponding time step.

[0033] Subsequently, with minimizing the deviation as the optimization objective, the resource states in the resource phase space are dynamically adjusted while satisfying the physical resource constraints of the hardware-in-the-loop simulation platform. Specifically, based on the currently available computing resources, link bandwidth, storage capacity, and scheduling capabilities of the hardware-in-the-loop simulation platform, the resource state vector is iteratively updated. This update process can be implemented using a difference equation strategy. That is, in each scheduling cycle, based on the deviation between the network behavior state of the previous scheduling cycle and the corresponding expected trajectory state, the direction and magnitude of the resource configuration change are determined. This change is then used as an incremental correction to the current resource state, making the new resource configuration state equal to the resource configuration state of the previous scheduling cycle plus this incremental change, thus obtaining the resource configuration state for the next scheduling cycle. In this way, the resource state is gradually updated with each scheduling cycle, forming a discrete-time evolution trajectory in the resource phase space. After the resource state completes one incremental update, the updated resource configuration state drives the network equivalent phase space to enter the behavior state of the next scheduling cycle. This behavior state is recursively evolved from the behavior state of the previous scheduling cycle under the current resource conditions. Through the above method, the network behavior state continuously evolves in discrete time according to the recursive relationship of "the current state determines the next state", forming a discrete trajectory in the network equivalent phase space. The above resource state recursion and behavior state recursion are executed alternately in each scheduling cycle, forming a closed-loop iterative process based on discrete-time differential updates. As the scheduling cycle progresses, if the behavior trajectory in the network equivalent phase space gradually approaches the desired trajectory state or the desired trajectory, and the amplitude of state changes between adjacent scheduling cycles continues to decrease, then the differential update process is judged to be stable. When the behavior deviation remains within a preset threshold range for multiple consecutive scheduling cycles, the differential equation is considered to have reached a convergence state, and the final resource configuration of the generated equivalent simulated communication network is determined accordingly. This ensures that the generated equivalent simulated communication network maintains the same dynamic evolution characteristics as the physical communication network at the behavior level, which is used for subsequent communication simulation operation, behavior analysis, and risk inference.

[0034] Furthermore, after generating the equivalent simulated communication network, it includes: Based on the equivalent simulated communication network, a simulation of the physical communication network is performed; during the simulation, samples are taken from the physical communication network and the equivalent simulated communication network at a preset period to determine the short-term behavior phase space; distance measurement and core feature difference measurement of the short-term behavior phase space are performed to determine the equivalent error; and the equivalent simulated communication network is adjusted based on the equivalent error.

[0035] Optionally, after generating the equivalent simulated communication network, the topology, logical entity relationships, service load patterns, and control strategies of the physical communication network are mapped to the equivalent simulated communication network. The corresponding communication simulation task is then initiated on the hardware-in-the-loop simulation platform, allowing the equivalent simulated communication network to run under a given resource configuration and scheduling strategy, and outputting multimodal communication time-series data corresponding to the operation of the physical communication network. During the simulation, the physical communication network and the equivalent simulated communication network are sampled at preset periods to determine the corresponding short-term behavioral phase spaces. This preset period can be configured according to the network's dynamic change rate, for example, by sampling using a fixed time window or event-triggered methods. Within each sampling period, short-term time-series data for the signaling plane, user plane, and control plane are obtained from both the physical and equivalent simulated communication networks. Using a method consistent with the aforementioned phase space reconstruction, the short-term time-series data is reconstructed using Takens embedding to form a short-term behavioral phase space corresponding to the current sampling period. Each short-term behavioral phase space is used to characterize the local dynamic evolution characteristics of the network within that time window. After obtaining the short-term behavioral phase spaces of the physical communication network and the equivalent simulated communication network, distance and core feature difference measures are applied to determine the equivalence error. The distance measure assesses the geometrical differences between the two short-term behavioral phase spaces, which can be characterized by calculating the average point distance, distribution distance, or trajectory similarity index between the phase space trajectories. The core feature difference measure assesses the differences at the level of dynamic invariants, specifically comparing the differences in core features such as correlation dimension, Kolmogorov entropy, and principal component vectors in the short-term behavioral phase spaces. These differences are then normalized and weighted to quantify a uniformly measurable error index. By weightedly fusing the geometric distance measurement results and the core feature difference results, the equivalence error, representing the degree of equivalence under the current simulation state, is obtained. After determining the equivalence error, the equivalent simulated communication network is adjusted based on it. Specifically, when the equivalent error exceeds a preset threshold, it is determined that there is a deviation between the current simulation state and the behavior of the physical communication network, and the parameter correction process of the equivalent simulation communication network is triggered. This correction process includes adjusting the behavior-resource phase space mapping parameters, the state update rules in the virtual resource dynamic architecture, and the elastic resource allocation and scheduling strategy.

[0036] When correcting the mapping parameters at the behavior-resource phase space mapping level, the distribution of equivalent errors in geometric structure differences and core feature differences is analyzed. The direction and magnitude of deviation between the behavior phase space and resource phase space in the current mapping result are analyzed. For error terms corresponding to core features such as correlation dimension, Kolmogorov entropy, or principal component direction, the weight coefficients, mapping scale factor, or nonlinear mapping parameters in the behavior-resource mapping function are adjusted to make the mapped resource phase space more closely resemble the behavioral characteristics of the physical communication network in terms of complexity level, divergence rate, and principal evolution direction. After completing the mapping parameter correction, the initial mapping state of the network's equivalent phase space is recalculated, providing a corrected starting point for subsequent evolution.

[0037] When modifying resource state update rules at the virtual resource dynamic architecture level, the difference equations are adjusted based on the deviation trend between the current resource phase space evolution trajectory and the expected trajectory. This includes, but is not limited to, resource growth or release rates, state coupling strength, decay coefficients, and constraint boundary conditions. Through these adjustments, the response speed and magnitude of resource states to behavioral changes are made consistent with the dynamic characteristics of resource use and release in physical communication networks, thereby improving the consistency of resource phase space trajectory evolution over time.

[0038] When modifying the scheduling strategy at the elastic resource allocation and scheduling level, the equivalent error is decomposed to the specific logical entity level. The resource configuration corresponding to sessions, service flows, or virtual network functions is analyzed. For logical entities with large deviations, the allocation priority of their corresponding resources is increased, the scheduling cycle is adjusted, or the resource reclamation strategy is modified to enhance their ability to track the target trajectory. At the same time, for entities with small deviations or excessively high resource utilization, resource allocation is appropriately tightened or resource expansion is delayed, thereby achieving a dynamic balance of resource allocation under the condition of overall resource constraints.

[0039] After adjustment, the simulation is run again and the next sampling cycle begins. The sampling, measurement, and adjustment process is repeated, forming a closed-loop optimization mechanism based on equivalent error feedback. This allows the equivalent simulated communication network to gradually approximate the real behavior of the physical communication network during continuous operation. Through these steps, online correction and adaptive adjustment of the equivalent simulated communication network are achieved, ensuring that under different operating stages and network conditions, the equivalent simulated communication network maintains consistent behavioral characteristics with the physical communication network in terms of both geometry and dynamics.

[0040] Furthermore, the equivalent simulation communication network adopts a master-slave collaborative control architecture, in which the master node performs mapping management based on behavior-resource phase space mapping, and the slave nodes perform resource structure configuration and simulation tasks locally.

[0041] Optionally, this equivalent simulation communication network adopts a master-slave collaborative control architecture to decouple and coordinate behavior-resource phase space mapping management and distributed resource execution. In this architecture, the master node communicates with multiple slave nodes through a control interface. The master node is responsible for global behavior modeling and mapping management, while the slave nodes are responsible for executing specific resource configuration and simulation tasks within their respective resource domains. Specifically, the master node centrally maintains the network behavior phase space, resource phase space, and the behavior-resource phase space mapping relationship between them, and manages the behavioral characteristics of the physical communication network in a unified manner based on the mapping relationship. The master node periodically or event-triggeredly receives status feedback information from the physical communication network and each slave node, evaluates the evolution trajectory of the network's equivalent phase space, and executes mapping parameter updates, resource dynamic rule adjustments, and scheduling policy generation when equivalent errors or changes in operating status are detected. The master node also generates resource scheduling and simulation execution strategies based on the above analysis results. This strategy includes at least the target resource state, resource adjustment magnitude, scheduling priority, and simulation task parameters, and distributes the strategy to the corresponding slave nodes through control signaling. The slave nodes are deployed in different resource domains of the hardware-in-the-loop simulation platform, with each slave node corresponding to a relatively independent set of computing, link, or functional resources. After receiving the policy from the master node, each slave node completes the specific resource structure configuration and simulation task execution within its local resource domain. This resource structure configuration includes adjusting the allocation of computing resources, link bandwidth limits, queue and cache parameters, the number of virtual network function instances, and their running status according to the policy requirements. This simulation task execution includes starting, pausing, or adjusting the corresponding communication simulation modules to ensure that the local simulation behavior conforms to the target resource state and evolution direction specified by the master node.

[0042] During simulation, slave nodes continuously monitor their local resource status and simulation results, feeding back information such as resource utilization, state changes, and local behavioral characteristics to the master node. The master node updates and evaluates the global network equivalent phase space based on the feedback from each slave node, and regenerates and distributes adjustment strategies when necessary, thus forming a collaborative control closed loop with unified decision-making by the master node, distributed execution by slave nodes, and continuous state feedback. Through this master-slave collaborative control architecture, the equivalent simulation communication network achieves efficient distributed implementation of resource scheduling and simulation execution while ensuring global behavioral consistency, improving the system's scalability and real-time performance.

[0043] Furthermore, after generating the equivalent simulated communication network, the method further includes: A feedback controller is constructed, and an interaction is established between the feedback controller and the equivalent simulation communication network. The feedback controller takes the equivalent error as input and the adjustable mapping parameter from the network behavior phase space to the resource phase space as output. If the equivalent error meets a preset threshold, the feedback controller is activated to execute an error adjustment decision and adjust the equivalent simulation communication network.

[0044] Preferably, after the equivalent simulation communication network is constructed and put into operation, a feedback controller is constructed, and an interaction mechanism is established between the feedback controller and the equivalent simulation communication network to dynamically adjust the equivalent deviation during the simulation process. This feedback controller is constructed using an LSTM (Long Short-Term Memory) network, taking the equivalent error calculated during the simulation as input and the adjustable mapping parameters in the mapping process from network behavior phase space to resource phase space as output. These adjustable mapping parameters include at least mapping weight coefficients, time lag parameters, scale adjustment parameters, and nonlinear mapping strength parameters, which are used to adjust the simulation system's ability to approximate the real network behavior. During the simulation, the equivalent error between the equivalent simulation communication network and the physical communication network is periodically or event-triggered and input to the feedback controller. When the equivalent error is less than a preset threshold, the current system is determined to be in a stable equivalent state, and the feedback controller remains on standby. When the equivalent error exceeds the preset threshold, the feedback controller is activated and enters the error adjustment decision process. During the error adjustment decision-making process, the feedback controller, based on the magnitude of the equivalent error and the learned mapping relationship, generates adjustments to the adjustable mapping parameters. For example, when the equivalent error mainly manifests as deviations in behavioral complexity or degrees of freedom, the weight parameters in the behavior-resource mapping are adjusted to change the response strength of the resource phase space to behavioral changes. When the equivalent error mainly manifests as behavioral evolution rhythm or phase differences, the time lag parameters in the mapping process are adjusted to realign the system's state update sequence with the evolution rhythm of the physical communication network. When the equivalent error manifests as divergent behavioral trajectories or inconsistent convergence speeds, the sensitivity of state evolution in the resource phase space is altered by adjusting the scale or nonlinear mapping parameters. After determining the adjustment results of the adjustable mapping parameters, the feedback controller outputs the adjustment results and drives the equivalent simulated communication network to re-map and evolve the resource states based on the updated mapping parameters. Finally, it enters a new operating cycle and continuously receives input from the feedback controller during subsequent sampling and error assessment, realizing a closed-loop control mechanism of "error detection, parameter adjustment, and behavior re-evolution", so that the behavior trajectory of the equivalent simulated communication network gradually converges to the real behavior trajectory of the physical communication network.

[0045] Before constructing the feedback controller, a sample dataset for training is acquired. This dataset originates from the operation of the communication network under different network loads, topologies, and control strategies. During the model training phase, a sliding time window of fixed length L is used to divide the historical equivalent error sequence, constructing training samples. Each training sample consists of an equivalent error vector within L consecutive sampling periods, serving as the model input. The corresponding training label represents the historical adjustment amount of the mapping adjustable parameter under the influence of that error sequence. This mapping adjustable parameter includes at least the mapping weight coefficient, time lag parameter, scaling parameter, and nonlinear mapping intensity parameter. By progressively shifting the sliding window, a large number of input-output sample pairs are constructed, thus forming the training dataset for the feedback controller.

[0046] Subsequently, the neural network structure of the feedback controller was designed. This feedback controller adopts an LSTM-based temporal modeling structure, including an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer receives error sequence data of shape (L,D), where L represents the time window length and D represents the dimension of the error vector. The LSTM layer consists of multiple stacked LSTM units; for example, a two-layer LSTM structure is used, with each layer containing several hidden units, to learn the evolution of the equivalent error and its long-term dependencies step-by-step. In the initial model design, a Dropout mechanism is introduced between the LSTM layers to randomly mask the output of some neurons, preventing the model from overfitting to local error patterns during training.

[0047] During the model parameter initialization phase, the weight parameters in the LSTM layer are initialized using a random initialization method based on Xavier or an equivalent strategy to ensure the stability of signal propagation between different layers, and the bias terms are uniformly initialized to zero. After initialization, the training samples are input into the feedback controller to execute the forward propagation process. During the forward propagation process, the first-layer LSTM sequentially reads the error vectors of L time steps. In each time step, the LSTM unit combines the current error input with the hidden state and cell state of the previous time step, and automatically learns which historical error information needs to be retained, which information can be forgotten, and the degree of influence of the current error on subsequent parameter adjustment through input gate, forget gate, and output gate mechanisms. After L time steps of iterative calculation, the first-layer LSTM outputs a set of hidden state sequences containing complete historical error information.

[0048] The second LSTM layer takes the hidden state sequence output from the first layer as input to model the error signal at a higher level of temporal features. This allows it to learn more abstract evolutionary patterns behind error changes, such as continuous error accumulation, periodic oscillations, or rapid divergence. The Dropout mechanism is activated in this layer to enhance the model's generalization ability to different network operating scenarios. In the final time step, the second LSTM layer outputs the final hidden state vector, which is a highly condensed representation of the equivalent error evolution information over the past L sampling periods.

[0049] Next, the final hidden state vector is input into the fully connected layer. The fully connected layer learns the nonlinear mapping relationship between error evolution features and mapping parameter adjustment. Its output is the adjustment vector of the mappable parameters. The output layer then uses this adjustment vector as the final output of the controller to guide parameter correction in subsequent simulation systems. During training, the mapping parameter adjustment output by the model is compared dimension-by-dimensionally with the corresponding true adjustment in the training samples. The error between the two is calculated, and the mean squared error is used as the loss function to measure the predictive performance of the current model on the sample. This loss value is then propagated layer by layer in the network through the backpropagation algorithm, sequentially calculating the gradient information corresponding to the fully connected layer weights, the gate weights of each LSTM unit, and the bias parameters. Then, the Adam optimization algorithm is used to update the model parameters. This optimization algorithm combines the estimation of the first and second moments of the gradient to adaptively adjust the update step size for different parameters, thereby improving the stability and convergence speed of the training process.

[0050] The above training process is repeated across multiple training batches until the preset maximum number of training epochs or the loss function convergence condition is reached. During training, an independent validation set is simultaneously used to evaluate model performance. When the equivalent error on the validation set no longer decreases significantly over multiple training epochs, it is determined that the model may be overfitting. In this case, the model is regularized by increasing the Dropout ratio or decreasing the learning rate, and training continues. If the validation set error still does not reach the preset threshold after multiple learning rate decays, the training batch size or time window length is adjusted to enhance the randomness of gradient updates and the model's generalization ability. When the equivalent error on the validation set decreases to within the preset threshold range, and the adjusted output mapping parameters can stably guide the behavior trajectory of the equivalent simulated communication network to converge to the real behavior trajectory of the physical communication network, the training process is terminated. The currently trained feedback controller is stored and deployed for online error adjustment and parameter correction during the operation of the equivalent simulated communication network.

[0051] Furthermore, after generating the equivalent simulated communication network, the method further includes: Based on the equivalent simulation communication network, a communication network simulation is performed. If the simulation trajectory shows an abnormal divergence trend based on the threshold of key risk indicators, a risk control instruction is generated. The source of the risk is traced in the equivalent simulation data based on the risk control instruction to determine the cause of the risk, and a risk intervention verification based on the risk plan is performed to determine the risk control strategy. Based on the cause of the risk and the risk control strategy, the physical communication network is managed under risk control.

[0052] Preferably, after generating the equivalent simulated communication network and completing parameter correction, communication network simulation is performed based on the equivalent simulated communication network. Specifically, on a semi-physical simulation platform, the equivalent simulated communication network is driven to run multiple rounds of simulation according to preset scenarios of service load changes, topology evolution, control strategy adjustments, and anomaly injection. The evolution trajectory of the network behavior phase space and the network equivalent phase space is continuously recorded during the simulation process, and key risk indicators related to network operation security, stability, and service quality are calculated simultaneously. These key risk indicators include at least network congestion intensity, latency surge magnitude, and packet loss rate trend. During the simulation, the key risk indicators are compared with preset risk thresholds in real time or periodically. When an abnormal divergence exceeding the risk threshold is detected in the simulation trajectory at the behavioral or indicator level, it is determined that there is a potential network risk in the current simulation scenario, and a corresponding risk control instruction is generated. This risk control instruction is used to identify the time interval of the risk occurrence, the scope of the logical entities involved, and the risk level. After generating risk control instructions, based on the risk time window and logical entities indicated by the instructions, the behavioral trajectory, resource phase space evolution process, and resource scheduling logs of the equivalent simulated communication network are retrospectively analyzed to examine the behavioral changes before and after the abnormal divergence. By using an LSTM-based causal localization model to compare the behavioral phase space structure, core feature changes, and resource allocation status under normal and abnormal operating conditions, the key triggering factors for risk generation are identified. These risk causes may include, but are not limited to, sudden increases in business load, resource allocation imbalances, unreasonable control strategies, virtual network functional bottlenecks, or multi-plane coupling amplification effects. The construction process of the causal localization model is similar to that described above. After determining the risk causes, corresponding intervention strategies are selected from a pre-built risk contingency plan library. Examples include adjusting resource allocation priorities, restricting high-risk business flows, reconfiguring control strategy parameters, adding key virtual network function instances, or introducing rate limiting and isolation mechanisms. The selected intervention strategies are then injected into the equivalent simulated communication network to re-execute the simulation and evaluate the changes in key risk indicators and behavioral phase space evolution trajectories before and after the intervention. When the simulation results after intervention show that the risk indicators have fallen back to the safe threshold range and the network behavior trajectory has converged back to the stable region, the intervention plan is deemed effective and identified as the risk control strategy for the corresponding risk scenario. Finally, based on the risk causes and the determined risk control strategy, risk control management is implemented on the physical communication network. In this process, the risk control strategy, which has been verified as effective through simulation, is transformed into configuration instructions executable by the physical communication network and implemented as needed. At the same time, the effectiveness of risk control execution is continuously monitored, and the actual operation results are compared with the simulation verification results to verify the effectiveness and stability of the risk control strategy in the real network. This achieves a closed-loop mechanism that guides the risk prevention and operation management of the physical communication network based on equivalent simulation results.

[0053] In summary, the embodiments of this application have at least the following technical effects: First, multimodal communication data streams are collected using probes deployed in the physical communication network. These multimodal data streams include time-series data from the signaling plane, user plane, and control plane. Next, the multimodal communication data streams are reconstructed into a high-dimensional network behavior phase space using Takens embedding. Core features are extracted from this phase space, where the core features are topological and dynamic invariants that characterize the core network behavior and do not change with the observation scale. Finally, the resource structure of the hardware-in-the-loop simulation platform is abstracted into a virtual resource dynamic architecture. An equivalent mapping of communication behavior is performed using these core features as constraints, generating an equivalent simulated communication network. This is described using difference equations, and the equivalent mapping includes behavior-resource phase space mapping and elastic resource allocation scheduling. This invention solves the technical problem that existing hardware-in-the-loop simulation methods cannot accurately capture cross-plane interaction behavior in multimodal communication networks, leading to a mismatch between simulation results and actual network dynamics. It achieves the technical effect of high-precision network behavior reproduction and efficient resource utilization by extracting topological and dynamic invariants through phase space reconstruction, and completing elastic resource scheduling by combining the virtual resource dynamic architecture abstracted by difference equations.

[0054] Example 2, based on the same inventive concept as the semi-physical network equivalent simulation method for multimodal communication in the previous examples, such as... Figure 2 As shown, this application provides a semi-physical network equivalent simulation system for multimodal communication, wherein the system includes: Data acquisition module 11: Collects multimodal communication data streams through probes deployed in the physical communication network, wherein the multimodal communication data streams include time-series data of the signaling plane, user plane, and control plane; Feature extraction module 12: For the multimodal communication data streams, reconstructs them into a high-dimensional network behavior phase space through Takens embedding, and extracts the core features in the network behavior phase space, wherein the core features are topological and dynamic invariants that characterize the core network behavior and do not change with the observation scale; Behavior mapping module 13: By abstracting the resource structure of the semi-physical simulation platform into a virtual resource dynamic architecture, and performing equivalent mapping of communication behavior with the core features as constraints, an equivalent simulated communication network is generated, wherein the abstract description is performed using difference equations, and the equivalent mapping includes behavior-resource phase space mapping and elastic resource allocation scheduling.

[0055] Furthermore, the feature extraction module 12 is used to perform the following method: For the multimodal communication data stream, time delay parameters and state space dimensions are determined; based on the time delay parameters and state space dimensions, the original variable time series is transformed into a set of points in a high-dimensional space using delay coordinates, forming the network behavior phase space, where each point represents the complete communication state at any given time; the network behavior phase space is verified based on the geometric structure and key dynamic indicators, with the intrinsic dynamic characteristics of the equivalent physical communication network as the verification target.

[0056] Furthermore, the feature extraction module 12 is used to perform the following method: The core features include at least the correlation dimension, Kolmogorov entropy, and principal component vectors; wherein, the correlation dimension is used to quantify the complexity and degree of freedom of network communication behavior, the Kolmogorov entropy characterizes the rate of unpredictability of network information generation or behavior, and the principal component vectors identify the main direction of behavior evolution.

[0057] Furthermore, the behavior mapping module 13 is used to perform the following method: By abstracting the resource structure of the hardware-in-the-loop simulation platform into a virtual resource dynamic architecture, a resource phase space is constructed. With the key topological and dynamic invariants as constraints, points in the network behavior phase space are mapped to the resource phase space to determine the network equivalent phase space.

[0058] Furthermore, the behavior mapping module 13 is used to perform the following method: Obtain the logical entity of the physical communication network as the target baseline for resource scheduling; to minimize the deviation between the evolution trajectory of the network equivalent space and the target baseline, perform iterative optimization of resource allocation and scheduling in the network equivalent space under the physical resource state based on a semi-physical simulation platform, and generate the equivalent simulated communication network.

[0059] Furthermore, the behavior mapping module 13 is used to perform the following method: Based on the equivalent simulated communication network, a simulation of the physical communication network is performed; during the simulation, samples are taken from the physical communication network and the equivalent simulated communication network at a preset period to determine the short-term behavior phase space; distance measurement and core feature difference measurement of the short-term behavior phase space are performed to determine the equivalent error; and the equivalent simulated communication network is adjusted based on the equivalent error.

[0060] Furthermore, the behavior mapping module 13 is used to perform the following method: The equivalent simulation communication network adopts a master-slave collaborative control architecture, in which the master node performs mapping management based on behavior-resource phase space mapping, and the slave nodes perform resource structure configuration and simulation tasks locally.

[0061] Furthermore, the behavior mapping module 13 is used to perform the following method: A feedback controller is constructed, and an interaction is established between the feedback controller and the equivalent simulation communication network. The feedback controller takes the equivalent error as input and the adjustable mapping parameter from the network behavior phase space to the resource phase space as output. If the equivalent error meets a preset threshold, the feedback controller is activated to execute an error adjustment decision and adjust the equivalent simulation communication network.

[0062] Furthermore, the behavior mapping module 13 is used to perform the following method: Based on the equivalent simulation communication network, a communication network simulation is performed. If the simulation trajectory shows an abnormal divergence trend based on the threshold of key risk indicators, a risk control instruction is generated. The source of the risk is traced in the equivalent simulation data based on the risk control instruction to determine the cause of the risk, and a risk intervention verification based on the risk plan is performed to determine the risk control strategy. Based on the cause of the risk and the risk control strategy, the physical communication network is managed under risk control.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A semi-physical network equivalent simulation method for multimodal communication, characterized in that, The method includes: Multimodal communication data streams are collected by probes deployed in physical communication networks, wherein the multimodal communication data streams include timing data from the signaling plane, user plane, and control plane. For the multimodal communication data stream, it is reconstructed into a high-dimensional network behavior phase space through Takens embedding, and the core features in the network behavior phase space are extracted. The core features are topological and dynamic invariants that characterize the core network behavior and do not change with the observation scale. By abstracting the resource structure of the hardware-in-the-loop simulation platform into a virtual resource dynamic architecture, and using the core features as constraints to perform equivalent mapping of communication behavior, an equivalent simulation communication network is generated. The network is described by difference equations, and the equivalent mapping includes behavior-resource phase space mapping and elastic resource allocation and scheduling.

2. The semi-physical network equivalent simulation method for multimodal communication as described in claim 1, characterized in that, Reconstructing the network behavior phase space through Takens embeddings includes: For the multimodal communication data stream, determine the time delay parameter and the state space dimension; Based on the time delay parameter and the state space dimension, the original variable time series is transformed into a set of points in a high-dimensional space through delay coordinate representation, which constitutes the network behavior phase space, where each point represents the complete communication state at any given time. The network behavior phase space is verified based on the geometric structure and key dynamic indicators, with the intrinsic dynamic characteristics of equivalent physical communication networks as the verification target.

3. The semi-physical network equivalent simulation method for multimodal communication as described in claim 1, characterized in that, The core features include at least the correlation dimension, Kolmogorov entropy, and principal component vector; The correlation dimension is used to quantify the complexity and degree of freedom of network communication behavior, the Kolmogorov entropy represents the rate of unpredictability of network information generation or behavior, and the principal component vector identifies the main direction of behavior evolution.

4. The semi-physical network equivalent simulation method for multimodal communication as described in claim 1, characterized in that, In the equivalent mapping of communication behaviors constrained by the aforementioned core features, the behavior-resource phase space mapping includes: By abstracting the resource structure of the hardware-in-the-loop simulation platform into a virtual resource dynamic architecture, a resource phase space is constructed; By maintaining key topological and dynamic invariants as constraints, points in the network behavior phase space are mapped to the resource phase space to determine the network equivalent phase space.

5. The semi-physical network equivalent simulation method for multimodal communication as described in claim 4, characterized in that, In the communication behavior equivalence mapping constrained by the aforementioned core features, elastic resource allocation and scheduling includes: Obtain the logical entities of the physical communication network as the target baseline for resource scheduling; To minimize the deviation between the evolution trajectory of the network equivalent space and the target baseline, the network equivalent space is iteratively optimized for resource allocation and scheduling under physical resource conditions based on a hardware-in-the-loop simulation platform, thereby generating the equivalent simulated communication network.

6. The semi-physical network equivalent simulation method for multimodal communication as described in claim 5, characterized in that, After generating the equivalent simulated communication network, the following steps are included: Based on the equivalent simulated communication network, perform a simulation of the physical communication network; During the simulation, samples are taken from the physical communication network and the equivalent simulated communication network at preset intervals to determine the short-term behavior phase space. Perform distance measurement and core feature difference measurement in short-term behavior phase space to determine equivalent error; The equivalent simulation communication network is adjusted based on the equivalent error.

7. The semi-physical network equivalent simulation method for multimodal communication as described in claim 6, characterized in that, The equivalent simulation communication network adopts a master-slave collaborative control architecture, in which the master node performs mapping management based on behavior-resource phase space mapping, and the slave nodes perform resource structure configuration and simulation tasks locally.

8. The semi-physical network equivalent simulation method for multimodal communication as described in claim 1, characterized in that, After generating the equivalent simulated communication network, the method further includes: Construct a feedback controller and establish the interaction between the feedback controller and the equivalent simulation communication network, wherein the feedback controller takes the equivalent error as input and the adjustable mapping parameter from the network behavior phase space to the resource phase space as output; If the equivalent error meets a preset threshold, the feedback controller is activated to execute an error adjustment decision and adjust the equivalent simulation communication network.

9. The semi-physical network equivalent simulation method for multimodal communication as described in claim 1, characterized in that, After generating the equivalent simulated communication network, the method further includes: Based on the equivalent simulated communication network, a communication network simulation is performed. If the simulation trajectory shows an abnormal divergence trend based on the threshold of key risk indicators, a risk control instruction is generated. In equivalent simulation data, risk control instructions are used to trace the source of risks, identify the causes of risks, and risk intervention verification based on risk plans is performed to determine risk control strategies. Based on the aforementioned risk causes and risk control strategies, risk control management is implemented for the physical communication network.

10. A semi-physical network equivalent simulation system for multimodal communication, characterized in that, The system is used to implement the semi-physical network equivalent simulation method for multimodal communication according to any one of claims 1-9, the system comprising: Data acquisition module: Collects multimodal communication data streams through probes deployed in the physical communication network, wherein the multimodal communication data streams include timing data of the signaling plane, user plane and control plane; Feature extraction module: For the multimodal communication data stream, the network behavior phase space is reconstructed through Takens embedding, and the core features in the network behavior phase space are extracted. The core features are topological and dynamic invariants that characterize the core network behavior and do not change with the observation scale. Behavior mapping module: By abstracting the resource structure of the hardware-in-the-loop simulation platform into a virtual resource dynamic architecture, and using the core features as constraints to perform equivalent mapping of communication behaviors, an equivalent simulation communication network is generated. The equivalent mapping is described by difference equations and includes behavior-resource phase space mapping and elastic resource allocation and scheduling.

Images