A risk entropy perception-based SDN transmission path dynamic scheduling method and system

Abstract

This invention provides a dynamic scheduling method and system for SDN transmission paths based on risk entropy perception in the field of structural health monitoring of large-scale transportation infrastructure. The method includes: Step S1, generating network configuration parameters and flow table generation rules that can be called by the SDN controller; Step S2, aligning and fusing multimodal monitoring data, and inputting the fused features into a risk identification model to obtain a risk entropy index; Step S3, determining the risk level based on the risk entropy index, and the SDN controller generating and issuing emergency flow table strategies; Step S4, performing differentiated encapsulation and transmission of multimodal monitoring data according to the risk level; Step S5, when the monitored physical monitoring indicators return to below the safety threshold, starting an observation window based on hysteresis loop control logic; after confirming that the risk entropy index is stable at a low level and the network congestion risk is resolved, reclaiming reserved network resources. The advantages of this invention are: effectively enhancing the accuracy and timeliness of structural health assessment and safety early warning.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring technology for large-scale transportation infrastructure, and in particular to a dynamic scheduling method and system for SDN transmission paths based on risk entropy perception. Background Technology

[0002] In the operation and maintenance of large-scale transportation engineering structures such as bridges, tunnels, and subways, existing monitoring systems generally employ a multi-source sensing-based monitoring architecture to dynamically monitor the impact of surrounding construction, traffic loads, and environmental factors on structural safety. This architecture comprehensively deploys various high-frequency and low-frequency sensing units, such as accelerometers, stress-strain gauges, fiber optic dynamic strain and displacement sensors, and settlement monitoring equipment, supplemented by high-definition video monitoring and manual inspections, to acquire information on the structure's dynamic response, long-term deformation, and external operating status.

[0003] The aforementioned multi-source, heterogeneous monitoring data needs to be transmitted back to the monitoring center for analysis and decision-making in real-time or near real-time via field communication networks. However, since these engineering structures are mostly located across rivers, seas, underground, or in areas with restricted transportation, their field monitoring systems typically rely on wired dedicated links or industrial communication networks with limited bandwidth for data upload. This results in high-bitrate video streams and high-frequency structural response data (such as vibration and strain) that require extremely high real-time and continuous performance sharing the same narrow transmission channel for extended periods. In this context, the performance of the communication network has become a bottleneck restricting the effectiveness of the entire monitoring system.

[0004] Existing technologies have gradually revealed significant shortcomings in this system: under extreme conditions such as concentrated heavy vehicle traffic or strong wind-induced vibration, the structural dynamic response intensifies, and the demand for low-latency, continuous transmission of high-frequency monitoring data increases dramatically. Simultaneously, video surveillance services may experience sudden surges in traffic due to unforeseen events, easily crowding out limited transmission links and thus blocking the real-time uplink of critical structural response data. This bandwidth contention not only leads to packet loss and discontinuous time series in monitoring data, disrupting the continuous observation conditions required for structural health assessment models and affecting the accuracy of assessment results, but in severe cases, it may even trigger false alarms or failures in safety warnings.

[0005] A more fundamental problem lies in the fact that the communication networks in existing engineering monitoring systems are typically treated as general-purpose transmission platforms independent of the engineering objects. Their data scheduling and resource allocation strategies are primarily configured based on general service types, fixed priorities, or preset bandwidth parameters, failing to incorporate existing engineering monitoring and safety assessment standards (such as the "Technical Specifications for Monitoring Building and Bridge Structures" and the "Technical Specifications for Monitoring Highway Bridge Structures JT / T1037-2022") into the network control decision-making process. The monitoring indicators, graded early warning thresholds, and handling requirements clearly stipulated in these standards currently rely mainly on backend manual analysis and post-event judgment, and have not yet been transformed into scheduling rules that can be understood and executed by the network system in real time. This results in the network system's inability to dynamically adjust transmission paths and allocate channel resources based on the real-time physical state and risk level of the engineering structure.

[0006] Therefore, in critical moments when structural safety risks accumulate or sudden anomalies occur, communication networks still transmit data according to the routine "best-effort" strategy, failing to reflect the "safety first" and "critical data first" principles required in the field of engineering safety. This makes the communication network, which should be a safeguard, a key weakness restricting the improvement of safety assessment and emergency response capabilities for large-scale transportation infrastructure.

[0007] Therefore, how to provide a dynamic scheduling method and system for SDN transmission paths based on risk entropy awareness to improve the dynamic scheduling capability of transmission priority for safety monitoring data of large-scale transportation infrastructure, optimize network resource allocation, and ensure the real-time and reliable transmission of critical structural response data, thereby enhancing the accuracy and timeliness of structural health assessment and safety early warning, has become an urgent technical problem to be solved. Summary of the Invention

[0008] The technical problem to be solved by this invention is to provide a dynamic scheduling method and system for SDN transmission paths based on risk entropy awareness, which can improve the dynamic scheduling capability of transmission priority for safety monitoring data of large-scale transportation infrastructure, optimize network resource allocation, and ensure the real-time and reliable transmission of critical structural response data, thereby enhancing the accuracy and timeliness of structural health assessment and safety early warning.

[0009] In a first aspect, the present invention provides a dynamic scheduling method for SDN transmission paths based on risk entropy awareness, comprising the following steps: Step S10: Construct an engineering monitoring standard knowledge base. Store the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring services in the engineering safety standards in the engineering monitoring standard knowledge base in a structured manner. Based on the structured information in the engineering monitoring standard knowledge base, generate network configuration parameters and flow table generation rules that can be called by the SDN controller. Step S20: Perform spatiotemporal alignment and feature fusion processing on multimodal monitoring data from large-scale transportation infrastructure to obtain fused features. Input the fused features into a risk identification model driven by physical information to calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time. Step S30: The SDN controller determines the current risk level based on the risk entropy index, generates and issues an emergency flow table strategy based on the risk level, network configuration parameters and flow table generation rules, and then performs deterministic bandwidth slicing and dynamic path scheduling, including: dynamically adjusting the bandwidth weight of preset virtual slices, implementing differentiated bandwidth suppression for multimodal monitoring data of non-critical services, and reselecting transmission paths for multimodal monitoring data of critical services. Step S40: Based on the risk level, use a structure-oriented data transmission protocol to perform differentiated encapsulation and transmission control on the multimodal monitoring data; Step S50: When the physical monitoring index is detected to return to below the safety threshold, the observation window is started based on the hysteresis loop control logic; after confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated, the network resources reserved for critical services are gradually reclaimed, and the emergency flow table policy issued by the SDN controller is restored.

[0010] Secondly, the present invention provides a risk entropy-aware SDN transmission path dynamic scheduling system, comprising the following modules: The engineering monitoring standard knowledge base construction module is used to build an engineering monitoring standard knowledge base. It stores the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring services in the engineering safety standards in a structured manner in the engineering monitoring standard knowledge base. Based on the structured information in the engineering monitoring standard knowledge base, it generates network configuration parameters and flow table generation rules that can be called by the SDN controller. The risk entropy index calculation module is used to perform spatiotemporal alignment and feature fusion processing on multimodal monitoring data from large-scale transportation infrastructure to obtain fused features. The fused features are then input into a risk identification model based on physical information to calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time. The emergency flow table policy distribution module is used by the SDN controller to determine the current risk level based on the risk entropy index, generate and distribute emergency flow table policies based on the risk level, network configuration parameters and flow table generation rules, and then perform deterministic bandwidth slicing and dynamic path scheduling, including: dynamically adjusting the bandwidth weight of preset virtual slices, implementing differentiated bandwidth suppression for multimodal monitoring data of non-critical services, and reselecting transmission paths for multimodal monitoring data of critical services; The monitoring data transmission module is used to perform differentiated encapsulation and transmission control of the multimodal monitoring data according to the risk level and using a structured data-oriented transmission protocol; The emergency flow table policy restoration module is used to start an observation window based on hysteresis loop control logic when the physical monitoring index is detected to return to below the safety threshold; after confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated, it gradually reclaims the network resources reserved for critical services and restores the emergency flow table policy issued by the SDN controller.

[0011] The advantages of this invention are: 1. By structuring and internalizing engineering safety standards (such as early warning thresholds and response timeliness) into executable policies of the SDN controller, the basis for scheduling decisions is first established. Then, a risk entropy index of the structure is calculated in real time using a risk identification model based on physical information to achieve quantitative perception of security risks. With this as the core driver, the SDN controller dynamically triggers emergency flow table policies, executing dynamic scheduling including deterministic bandwidth guarantees for critical structure response data, bandwidth suppression for non-critical services, and reselection of high-quality paths for them. At the same time, data is encapsulated using appropriate transmission protocols according to the risk level. This series of measures ensures that when the risk increases, network resources can be allocated to the most critical monitoring data streams with priority and reliability. When the risk is eliminated, resources are intelligently recovered and normal operation is restored through hysteresis loop control logic. Thus, a closed loop of "perception and identification - decision-making and scheduling - transmission guarantee - elastic recovery" is constructed as a whole. Ultimately, the transmission priority is dynamically adjusted according to the risk, and network resources are optimized and allocated on demand, ensuring the real-time and reliable transmission of critical data and significantly improving the accuracy and timeliness of structural health assessment and security early warning.

[0012] 2. By constructing an engineering monitoring standard knowledge base, safety specifications are transformed into network rules. A risk identification model driven by physical information is used to calculate the "risk entropy index" of the structure in real time, which serves as the core driving signal. When the risk increases, the SDN controller dynamically implements differentiated bandwidth scheduling based on this index (such as suppressing high-bandwidth non-critical business video streams) and reselects high-reliability paths for critical structural response data (such as vibration and strain) based on a "communication-physical" dual-domain cost function that incorporates risk entropy. At the same time, emergency mechanisms such as fast forwarding and sensing compression are activated at the transmission protocol layer according to the risk level to ensure low-latency and reliable transmission of critical data. After the structural risk stabilizes, resources are gradually reclaimed. Overall, the network transmission is transformed from static configuration to dynamic intelligent scheduling driven by the real-time risk of the engineering entity, thereby prioritizing the real-time and continuous nature of critical monitoring data under bandwidth-constrained conditions, ultimately enhancing the accuracy and timeliness of structural health assessment and safety early warning.

[0013] 3. Traditional methods typically treat network status (such as bandwidth and latency) and engineering safety as two independent systems. This invention utilizes the cross-domain quantitative indicator of "risk entropy index" as the core of network scheduling decisions. By constructing a risk identification model driven by physical information, it deeply integrates multimodal physical monitoring data (such as stress and displacement) and calculates a unified "risk entropy." This enables the SDN controller to "understand" the safety status of the engineering structure, achieving a paradigm shift from "passive response based on network status" to "proactive and predictive scheduling based on engineering risks." This allows for deep integration and intelligent linkage between network resource scheduling and engineering safety requirements.

[0014] 4. The network scheduling strategy is not static or based on simple priorities, but is driven by the dynamic risk level determined by "risk entropy" to execute precise "deterministic bandwidth slicing" and "dynamic path scheduling". Its "communication-physical" dual-domain cost function can automatically adjust the weights when the risk increases, so that the path selection prioritizes the critical monitoring data flow that is sensitive to packet loss. At the same time, by implementing "differentiated bandwidth suppression" for non-critical services, it can quickly and directionally release and guarantee bandwidth resources for critical services in emergency situations, realizing the optimal elastic allocation of network resources in the spatiotemporal dimension, and ensuring the absolute priority and determinism of critical monitoring information transmission in critical situations.

[0015] 5. Through deep optimization at the transport layer, the protocol can adaptively select encapsulation, error correction, and transmission modes based on the physical properties of the data (such as pulse-type damaged signals and slowly varying signals) and the real-time risk level. Especially in emergency situations, the protocol can temporarily simplify security verification, open fast channels, and bypass verification. This design greatly reduces transmission latency while ensuring basic security, meeting the extreme timeliness requirements of engineering safety emergency response. At the same time, the perceptual compression coding and core parameter priority transmission mechanism ensure that the most critical engineering status information can still be transmitted when the network is congested, maximizing the information value under limited bandwidth.

[0016] 6. By establishing a complete "perception-decision-execution-recovery" intelligent loop, from knowledge base and rule pre-setting, real-time risk perception and prediction, network policy generation and distribution, and differentiated data transmission, to the gradual recovery of resources and policy restoration after risk resolution, the entire process can automatically complete the optimization of network resource allocation, emergency response and smooth recovery based on the dynamic changes in the project status, significantly improving the overall autonomy and operation and maintenance efficiency of the large-scale infrastructure monitoring system.

[0017] 7. During the recovery phase after risk mitigation, a "hysteresis loop control logic" and a "gradual recovery" strategy were adopted. This does not mean immediately withdrawing all emergency measures as soon as the indicators return to the threshold, but rather opening an "observation window" to confirm that the risk has been stabilized and resolved. This design can effectively avoid frequent and drastic policy switching (i.e., "system oscillation") caused by short-term fluctuations in monitoring data or instantaneous improvement in network status, thereby ensuring the stability of network policies and the overall state of the project, preventing secondary impacts on the network and services caused by policy switching, and demonstrating foresight and robustness. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] Figure 1 This is a flowchart of a dynamic scheduling method for SDN transmission paths based on risk entropy awareness, according to the present invention.

[0020] Figure 2 This is a schematic diagram of the structure of a dynamic scheduling system for SDN transmission paths based on risk entropy awareness, according to the present invention. Detailed Implementation

[0021] The technical solution in this application embodiment follows the general idea as follows: A closed-loop intelligent system integrating "perception and identification - decision-making and scheduling - transmission assurance - elastic recovery" is constructed to address the disconnect between network transmission and engineering safety requirements in the monitoring of large-scale transportation infrastructure. Specifically, firstly, an engineering monitoring standard knowledge base is built to structure safety specifications such as physical monitoring indicators, early warning thresholds, and response requirements in the engineering field. These are then compiled into network policy rules that the software-defined network (SDN) controller can understand and execute, providing a basis for intelligent scheduling. Secondly, multimodal monitoring data from the site is fused and processed, and input into a risk identification model driven by physical information. A "risk entropy index" characterizing the structural safety status is calculated in real time, enabling quantitative perception of engineering risks. This risk entropy index serves as the core driving signal, triggering dynamic scheduling by the SDN controller. Based on the risk level, the controller issues emergency flow table policies, executes "deterministic bandwidth slicing" (such as dynamically adjusting virtual slice bandwidth and suppressing non-critical business traffic) and "dynamic path reselection" based on a "communication-physical" dual-domain cost function, ensuring that critical monitoring data streams receive priority and reliable transmission resources. Simultaneously, at the transport layer, protocol behavior is adaptively adjusted based on real-time risk levels. For example, fast forwarding channels are enabled for critical data, and perceptual compression coding is implemented to achieve differentiated data transmission protection. Finally, when monitoring indicators show that the risk has been eliminated, all emergency measures are not immediately withdrawn. Instead, an observation window is activated based on the "hysteresis loop control logic." After confirming that the risk has been stably eliminated, network resources and configurations are restored in an orderly manner using a "gradual recovery" strategy. This avoids system oscillations caused by state fluctuations or frequent policy switching, ensuring a smooth transition of the entire system from emergency state back to normal. Through this closed loop, a fundamental transformation is achieved in network resource scheduling from static configuration to dynamic intelligent scheduling driven by real-time risks of engineering entities.

[0022] Please refer to Figures 1 to 2 As shown, a preferred embodiment of the SDN transmission path dynamic scheduling method based on risk entropy awareness of the present invention includes the following steps: Step S10: Construct an engineering monitoring standard knowledge base. Store the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring services in the engineering safety standards in the engineering monitoring standard knowledge base in a structured manner. Based on the structured information in the engineering monitoring standard knowledge base, generate network configuration parameters and flow table generation rules that can be called by the SDN controller. Step S20: Perform spatiotemporal alignment and feature fusion processing on multimodal monitoring data from large-scale transportation infrastructure to obtain fused features. Input the fused features into a risk identification model driven by physical information to calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time. Step S30: The SDN controller determines the current risk level based on the risk entropy index, generates and issues an emergency flow table strategy based on the risk level, network configuration parameters and flow table generation rules, and then performs deterministic bandwidth slicing and dynamic path scheduling, including: dynamically adjusting the bandwidth weight of preset virtual slices, implementing differentiated bandwidth suppression for multimodal monitoring data of non-critical services, and reselecting transmission paths for multimodal monitoring data of critical services. Step S40: Based on the risk level, use the Structure-Oriented Data Transfer Protocol (SDDP) to perform differentiated encapsulation and transmission control on the multimodal monitoring data; Step S50: When the physical monitoring index is detected to return to below the safety threshold, the observation window is started based on the hysteresis loop control logic; after confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated, the network resources reserved for critical services are gradually reclaimed, and the emergency flow table policy issued by the SDN controller is restored.

[0023] The core logic of this invention lies in breaking the traditional "best-effort" general transmission mode of networks and constructing a communication resource on-demand slicing and dynamic scheduling system guided by engineering physical risks.

[0024] First, a leap from "static routing" to "semantic-driven path orchestration" is achieved in transmission path partitioning. Specifically, by constructing an engineering monitoring standard knowledge base, the real-time identified structural risk entropy (risk entropy index) is directly mapped to the optimization cost function of the underlying network. During the routine monitoring phase, data flows are distributed across multiple available physical links according to the principle of balance; however, once a structural anomaly is identified (such as bridge vibration or tunnel convergence), the SDN controller immediately initiates a path reselection algorithm to dynamically open a low-latency, high-reliability transmission path for critical high-frequency monitoring data. This path can avoid high-traffic congestion areas and has near-physical isolation transmission guarantees. This path partitioning mechanism not only achieves logical isolation between critical data and high-bandwidth services (such as high-definition video streams) but also ensures the priority passage of critical security data in extreme environments.

[0025] Secondly, at the channel resource allocation level, a sub-channel adaptive allocation mechanism based on multi-dimensional feature perception is introduced. By performing fine-grained logical slicing of the communication link, the bandwidth weight and scheduling priority of the sub-channels are dynamically adjusted according to the physical attributes of the monitoring data (such as transient damage pulses and quasi-static displacements) and security and timeliness requirements. In emergency situations, through the coordinated control logic of "bandwidth suppression and redundancy compensation," the sub-channel quotas of non-critical service flows are forcibly compressed, and the released frequency domain and time resources are compensated to high-frequency sensing sequences. Combined with a deterministic transmission protocol for structured data, efficient and lossless backhaul of critical monitoring data within the sub-channels is achieved.

[0026] Finally, a flexible balance logic between security strategy and transmission performance is achieved. When an extremely high risk level is identified, a "security load reduction mode" is proactively triggered on the transmission link, stripping away time-consuming encryption and verification mechanisms to maximize channel capacity and convert it into effective data payload, thereby eliminating processing fluctuations at the transmission layer. This closed-loop transmission system, which uses physical structural risks as its core instruction and consists of precise path partitioning, dynamic sub-channel suppression, and adaptive protocol degradation, significantly improves the anti-interference capability and emergency assessment accuracy of traffic engineering monitoring systems, achieving a technological leap from a "general transmission platform" to a "security-customized infrastructure."

[0027] Step S10 specifically includes: Step S11: Construct an engineering monitoring standard knowledge base, and structurally store the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring operations in the engineering safety standards into the engineering monitoring standard knowledge base; the physical monitoring indicators include at least bridge deflection, tunnel settlement, and subway vibration; Step S12: Deploy a dynamic compilation engine on the SDN controller. The dynamic compilation engine generates network configuration parameters and flow table generation rules that can be called by the SDN controller based on structured information from the engineering monitoring standard knowledge base.

[0028] The design of the structured mapping between engineering monitoring standards and flow table generation rules needs to start with the semantic reconstruction of the underlying protocol, transforming the "physical requirements" of engineering safety into "execution instructions" for network communication. This involves establishing a protocol parsing matrix based on an engineering standard knowledge graph. By extracting quantified thresholds for key indicators such as bridge deflection, tunnel settlement, or subway vibration from industry standards, these thresholds are abstracted into logical judgment benchmarks with engineering semantics. This step is no longer simple text storage; instead, it utilizes structured modeling techniques to compile parameters such as the hazard level of physical indicators, response time requirements, and data sampling frequency into a set of configuration parameters that the network controller can call in real time. This imbues the network system with a "decision-making brain" capable of understanding engineering risks.

[0029] Step S20 specifically includes: Step S21: Use high-precision timestamps to perform spatiotemporal alignment on multimodal monitoring data from large-scale transportation infrastructure. Use a normalized spatiotemporal mapping model based on non-uniform sampling reconstruction to adaptively upsample the low-frequency signals in the multimodal monitoring data. Calculate the cross-correlation matrix between different modal data to perform feature fusion processing and obtain fused features. The multimodal monitoring data includes multiple types of data such as accelerometer data, stress-strain data, fiber optic grating dynamic strain and displacement data, settlement monitoring data, and video surveillance data. Step S22: Input the fused features into the risk identification model driven by physical information to calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time; the risk identification model is constructed based on a deep residual neural network, combined with dynamic weight risk entropy, and embeds partial differential equation constraints based on structural mechanical properties as physical constraint operators in the loss function; Step S23: Convert the risk entropy index into a network scheduling feedforward signal and output a predictive resource reservation request to the SDN controller.

[0030] In the logical construction of the perception and decision-making layer, the deep fusion of multimodal monitoring data is a prerequisite for achieving accurate risk identification. By deploying a spatiotemporal alignment engine at the edge or in the cloud, high-frequency vibration acceleration, low-frequency static leveling data, fiber optic strain signals, and traffic load flows captured by video are timestamped at the millisecond level and correlated with spatial coordinates. This process is not merely a physical stacking of data, but rather utilizes heterogeneous data association feature extraction technology to eliminate phase delays caused by differences in sensor physical characteristics, transforming isolated physical quantities into feature tensors reflecting structural evolution. This full-dimensional fusion of multi-source heterogeneous data provides a joint input with physical and logical support for subsequent highly reliable risk assessment.

[0031] To address the trade-off between accuracy and real-time performance in risk identification, a risk identification model (algorithm) combining a "physically driven deep residual neural network" and "dynamic weighted risk entropy" is introduced. This model compares multimodal input features (fusion features) in real time and uses residual mapping to capture nonlinear abnormal fluctuations in structural responses. The innovation lies in the embedded physical constraint operator, which automatically filters environmental noise interference and accurately identifies sudden safety events such as resonance in heavy-load convoys or brittle structural damage. Compared to traditional threshold alarms, this method significantly improves the accuracy of capturing weak damage signals while maintaining low computational load, elevating identification accuracy and false alarm suppression capabilities to engineering practice levels.

[0032] Once the risk entropy index is determined to overflow, it will quickly enter a "linked trigger mode." The core of this mode is to convert the risk entropy index into a predictive network scheduling feedforward signal in real time. The system can not only identify the current structural state but also predict potential transmission demand peaks within the next few seconds based on the evolution trend of multimodal monitoring data. This feedforward mechanism can output precise resource reservation requests to the SDN controller before actual network congestion occurs, ensuring that the risk identification module is not only an "observer" of the security status but also a "commander" of the underlying communication resources. This closed-loop logic from multimodal perception to proactive decision-making truly demonstrates the performance advantages of perception-driven architecture under extreme conditions in large-scale transportation engineering projects.

[0033] In the spatiotemporal alignment logic of multimodal data, the traditional simple interpolation method was abandoned, and a normalized spatiotemporal mapping model based on non-uniform sampling reconstruction was adopted instead. Due to the order-of-magnitude difference in sampling frequencies between the accelerometer (high frequency) and the hydrostatic level (low frequency), a time-domain synchronization operator Γ(t) based on high-precision GPS / PTP timestamps was established. This operator adaptively upsampled the low-frequency signal using Lagrange polynomials and, in conjunction with the cross-attention mechanism in a convolutional neural network, calculated the cross-correlation matrix between different modalities. This model can automatically compensate for the physical transmission delay offset Γi caused by the geographical distribution of the sensors, ensuring that the instantaneous load excitation value captured by the video stream and the strain response measured by the sensors are strictly spatiotemporally aligned on the order of 10^−3 seconds, providing a deterministic data basis for subsequent feature fusion.

[0034] The introduction of a Physics-Informed Operator is crucial for improving model accuracy and robustness. Partial Differential Equation (PDE) constraints based on the dynamic characteristics of bridges or tunnels are embedded in the loss function of the deep residual network. Specifically, the energy conservation terms and stiffness matrix degradation model in structural mechanics are transformed into a penalty term ℒphys in the algorithm, ensuring that feature extraction follows the physical evolution of the structure. When the model's predicted structural response deviates from physical reality (e.g., spurious features caused by environmental noise), this operator generates a large penalty value, forcing the neural network to correct the deviation. This design solves the "spurious correlation" problem that easily arises in purely data-driven models under small sample or high-noise environments, enabling the system to achieve an order-of-magnitude improvement in robustness to structural anomaly identification while maintaining low computational consumption.

[0035] From a performance optimization perspective, this approach achieves dimensionality reduction guidance for the probabilistic model through "physical logic." By directly integrating physical constraint operators into the gradient update process of the neural network, the model can locate the feature space reflecting structural damage with extremely fast convergence speed, avoiding the latency caused by large-scale black-box computation. In terms of practicality, it can filter out interference from conventional environmental vibrations caused by subway operation, accurately identifying truly engineering-significant risk features such as segment instability or main beam cracks. This efficient and accurate risk identification logic directly provides a high-quality trigger decision source for the SDN controller, ensuring that every bit of network bandwidth accurately serves real safety risks, significantly improving the system's real-time early warning performance.

[0036] In step S30, the dynamic path scheduling specifically includes path reselection, which employs an optimization algorithm based on a dual-domain cost function of "communication-physical". The expression of the dual-domain cost function is as follows: C_path=α*Σ(D_i / D_max+J_i / J_max+1 / (1-U_i))+β*exp(R*θ); Where C_path represents the comprehensive cost of the path; α represents the dynamically adjusted communication weight coefficient; β represents the dynamically adjusted engineering risk weight coefficient; D_i, J_i, and U_i represent the real-time latency, jitter, and bandwidth utilization of link i, respectively; D_max represents the maximum allowable latency threshold; J_max represents the maximum allowable jitter threshold; R represents the risk entropy index; and θ represents the link's sensitivity factor to packet loss of high-frequency data. When the risk entropy index R increases, the weight of β is increased, making path selection prioritize links with low packet loss rate and high stability. By introducing the nonlinear term (1-U_i), when the link is close to full load, the cost will explode exponentially, forcing data flow to avoid congested nodes in advance or generating a sharp increase in cost when the link load approaches saturation, forcibly triggering route reselection. Under normal monitoring, α dominates, pursuing global network throughput. After the structure enters the warning state (such as triggering a threshold above level 2), β grows exponentially, forcing the network to enter the "security guarantee mode".

[0037] The differentiated bandwidth suppression of multimodal monitoring data for non-critical services includes: when physical monitoring indicators trigger the warning threshold, the SDN controller locates and dynamically limits the transmission rate of multimodal monitoring data of non-critical services that occupy the most bandwidth, and redistributes the bandwidth resources released therefrom to multimodal monitoring data of critical services.

[0038] In the logical design of the transmission execution layer, perception-driven deterministic bandwidth slicing technology serves as the last line of defense for ensuring engineering security. Through the global view of the SDN controller, physical link resources are logically decoupled into multiple independent virtual slices, such as "core engineering sampling," "multi-dimensional environmental perception," and "routine operation and maintenance video." When the perception decision layer issues a high-risk entropy warning, the system no longer uses traditional static QoS configuration but instead initiates a dynamic slice resource reconstruction strategy. By calling the underlying OpenFlow protocol's Meter table and Queue mapping mechanism, the redistribution of slice weights can be completed in microseconds, instantly allocating limited bandwidth reserves to critical high-frequency monitoring data. This dynamic slicing technology ensures that, under extreme traffic loads or sudden disasters, the transmission path of core monitoring data has independent, undisturbed resource quotas, completely eliminating "channel preemption" of precision pulse signals by large-scale video traffic.

[0039] To address transmission conflicts under bandwidth-constrained conditions, a risk-linked, differentiated bandwidth suppression logic was designed. The core lies in proactive "on-demand degradation": once structural monitoring indicators trigger level one or two thresholds, the SDN controller precisely identifies the non-critical service flow (such as high-definition surveillance footage) with the highest occupancy rate in the current link and implements dynamic rate limiting. By adaptively adjusting the bandwidth limit of the video encoding stream, redundant bandwidth is forcibly redistributed to structural dynamic response signals, which are extremely sensitive to packet loss rates. This suppression logic is not a blind traffic cutoff, but rather a tiered backoff mechanism established based on engineering risk levels, maximizing the critical data carrying potential of narrowband communication networks while ensuring overall monitoring visibility.

[0040] In implementing path re-selection, this approach breaks away from traditional shortest path (Dijkstra's algorithm) or equivalent multipath (ECMP) algorithms, introducing a novel optimization algorithm based on a dual-domain cost function of "communication-physical". This algorithm jointly models real-time monitored network latency jitter with the risk sensitivity of the physical structure to calculate an optimal path for the current risk conditions. The innovation lies in the fact that when a predetermined path experiences momentary congestion or link quality degradation, the SDN controller can predictively switch critical monitoring data streams to "backup links" or "low-interference channels" with higher deterministic guarantees, rather than simply searching for the shortest physical distance with the fewest hops. Through this perception-driven path reconstruction, the end-to-end latency of critical data under extreme conditions can be reduced by approximately 35% to 50%, and the data transmission continuity guarantee rate can be improved to over 99.99%. This innovative design, which guides network flow based on physical risks, significantly improves the emergency response efficiency and structural assessment reliability of large-scale infrastructure in complex environments.

[0041] In path reselection decisions, traditional algorithms only focus on link bandwidth or hop count. This invention constructs a composite cost function C_path to calculate the cost of all available paths in real time and dynamically selects the path with the lowest cost for switching. This achieves deep coupling between the engineering physical security state and the underlying network communication environment. The core logic of the algorithm lies in using the real-time monitored risk entropy index as a "feedforward signal" for network path selection, dynamically adjusting the weight of each link in the network topology, thereby forcibly avoiding high-risk, high-congestion transmission paths under extreme conditions. This function consists of two core components: "network communication overhead" and "engineering risk sensitivity".

[0042] Step S40 specifically includes: Step S41: Based on the risk level, encapsulate the multimodal monitoring data using a structure-oriented data transmission protocol, embed engineering feature identifier bits in the data packet header, and mark the transmission priority and signal type according to the physical attributes of the multimodal monitoring data; the signal type includes at least high transient, high-frequency pulse-type damage signals and slowly varying signals; Step S42: For the pulse-type damage signal, enable forward error correction and dual-path redundant concurrent transmission mode; for the slowly varying signal, enable verification-based reliable transmission mode. Step S43: When network congestion is detected and the structure is at the risk threshold edge, the protocol layer triggers perceptual compression coding to prioritize the transmission of core dynamic parameters in the multimodal monitoring data, including the peak value, valley value and spectral feature value of the waveform. When the risk entropy index exceeds a preset threshold and enters an emergency state, the transmission protocol temporarily removes or simplifies the security verification process, opens a fast forwarding channel based on hardware address binding for key monitoring nodes, and bypasses the security verification process to the cloud for processing. Step S44: Establish an end-to-end timing synchronization domain based on a precision clock protocol, and control the latency jitter of data transmission through pre-allocated deterministic time slots; Physical semantic tags (transmission priority and signal type) are embedded in the packet header, and the transmission mode is switched in real time according to the risk level: in the case of extremely high risk (pulse-type damage signal), dual-path redundant concurrent mode is enabled, and forward error correction (FEC) mechanism is used to replace the traditional retransmission mechanism to eliminate delay jitter caused by packet loss and retransmission.

[0043] The SDDP protocol completely changes the traditional "black box transmission" mode of protocols, realizing a deterministic transmission mechanism that is deeply coupled with the dynamic response characteristics of engineering structures.

[0044] First, implement message differentiation encapsulation based on engineering semantics.

[0045] The SDDP protocol embeds an "engineering feature identifier" in the data frame header, enabling it to adjust transmission strategies in real time based on the dynamic features extracted by the sensing layer. For pulse-type damage signals with high transient and high-frequency characteristics, such as bridge cable vibration and sudden displacement of tunnel segments, it automatically identifies their physical meaning, assigns them a "real-time level" weight, and forcibly activates the "FEC error correction + dual-path spatial redundancy" mode. This means that even in environments with strong electromagnetic interference, such as subway tunnels, if a wireless or wired link experiences more than 10% random packet loss, the protocol can use redundant copies to perfectly reconstruct the original waveform of the structural damage at the receiving end without triggering a "retransmission request." For slowly changing tunnel convergence or static temperature-compensated displacement of long-span bridges, it switches to "reliable level" transmission, prioritizing the use of verification mechanisms to ensure the absolute accuracy of the data. This transmission logic, based on physical event attributes rather than simply data volume, ensures that critical damage data is never lost due to "head-of-line congestion" at the protocol layer, even under extremely complex operating conditions.

[0046] Second, design an adaptive protocol control logic that is linked to the physical sampling frequency.

[0047] To cope with the sudden surge in data traffic generated by strong winds or heavy vehicle traffic on bridges, the SDDP protocol incorporates a sampling rate-bandwidth coupling adjustment algorithm. When network path congestion (RTT fluctuations) is detected and the structure is nearing a safety risk threshold, the protocol layer no longer blindly discards packets but instead triggers a "physical feature point priority" transmission strategy. The protocol collaborates with sensor terminals to directly perform perceptual compression coding at the network layer: prioritizing the transmission of core dynamic parameters such as waveform peaks, valleys, zero-crossings, and spectral characteristic values, while downsampling redundant samples in stable regions as needed. This mechanism ensures that even under extreme bandwidth constraints, the structural evaluation algorithm at the analysis center can still acquire key features for support mode identification and stiffness assessment, completely avoiding the delay or "deadlock" of early warning information caused by the slow start mechanism of the traditional TCP protocol during congestion.

[0048] Third, ensure end-to-end timing determinism.

[0049] For multi-point synchronous vibration analysis, which is crucial in bridge health monitoring, the SDDP protocol establishes a unified physical clock synchronization domain based on PTP precision at the transport layer, strictly controlling the jitter of data encapsulation at each monitoring point to the microsecond level. Whether the monitoring points are distributed at both ends of a long-span bridge (kilometers or more) or in a tunnel section several kilometers long, SDDP ensures consistent physical timing of all modal data upon arrival at the central station through pre-allocated deterministic time slots. Experiments show that this transmission scheme, tailored for structural monitoring scenarios, strictly suppresses the end-to-end deterministic latency to within 20ms, providing extremely accurate "high-fidelity" time-domain input for the backend structural damage identification algorithm, significantly improving the accuracy of determining minute cracks or fatigue damage in the structure.

[0050] During the routine operation of the traffic engineering monitoring system, the SDDP protocol maintains a high level of security protection, including full message encryption, complex authentication, and integrity checks to defend against potential network attacks. However, once the physical constraint operator determines that the risk entropy index has overflowed and enters an emergency warning state, an adaptive degradation strategy prioritizing timeliness will be activated. Its innovation lies in the fact that the protocol layer automatically strips away non-core security overhead, temporarily disabling high-latency encryption algorithms and cumbersome handshake confirmation mechanisms, and allocating all computing resources and bandwidth to the monitoring data stream. This strategy effectively solves the latency jitter caused by encryption and decryption operations during emergencies, ensuring that the physical link from "emergency occurrence" to "instruction arrival" is in a simplified, high-speed direct connection state.

[0051] To ensure the self-healing and reliability of the system after degradation, this mechanism does not blindly abandon security, but rather compensates for security vulnerabilities through "trust pre-stored" and "post-audit" logic. Upon detecting an emergency trigger signal, the SDN controller opens a fast, green channel based on hardware MAC address binding for critical monitoring nodes, enabling line-speed forwarding without verification. Simultaneously, the relevant security verification steps are "stripped" from the transmission path and deployed in the cloud. This means that while ensuring the front-end monitoring curve is transmitted back with minimal latency (reduced by approximately 30% or more), the system can still maintain a basic security defense posture. Through this innovative "risk-awareness-guided security degradation," this invention achieves dynamic optimization of security resource utilization and emergency response efficiency while ensuring the deterministic transmission of "life-saving data" for engineering structures.

[0052] Step S50 specifically includes: Step S51: When the physical monitoring index is detected to return to below the safety threshold, a dynamic observation window is started based on the hysteresis loop control logic to continuously evaluate the structural status. Step S52: After confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated, the SDN controller issues a policy restoration command. Step S53: Based on the policy restoration instruction, adopt a progressive bandwidth compensation strategy to release the network resources of suppressed non-critical services in an orderly manner and cancel redundant backup paths to avoid secondary network impact, thereby recovering the network resources reserved for critical services and restoring the emergency flow table policy issued by the SDN controller.

[0053] In the design of the closed-loop control layer, closed-loop feedback and flow table failure / restoration mechanisms are core components for ensuring a smooth return to normal operation from an emergency state and maximizing resource efficiency. At this stage, a hysteresis loop control logic based on an "observation window" is introduced to address the frequent network policy jumps (i.e., the "ping-pong effect") that may be caused by structural oscillations or sensor noise. When physical monitoring indicators return to below a safe threshold, high-priority dedicated slices are not immediately revoked; instead, an observation window linked to engineering dynamic characteristics is automatically opened. During this window, the attenuation trend of residual structural vibration is continuously assessed. Only when the physical risk entropy index remains stable at a low level and the network congestion risk is resolved will the SDN controller issue a policy restoration command. This design simulates the prudent decision-making logic of human experts during on-site handling, ensuring the self-healing capability and business continuity of the monitoring system during complex operating condition switching processes.

[0054] To optimize resource utilization, the system employs flow table lifecycle management (Soft / Hard Timeout) and refined resource reclamation algorithms to perform real-time "slimming down" of transmission links. During the policy restoration phase, the system releases the bandwidth cap of suppressed non-critical traffic (such as high-definition video streams) in an orderly manner according to the priority of business relevance and cancels redundant backup paths. The innovation lies in the use of a gradual bandwidth compensation strategy during the restoration process, rather than instantaneous release, thus avoiding the secondary impact on the network caused by a large backlog of data streams upon link recovery. Through this refined resource scheduling of "peak shaving and valley filling," the system can ensure that the effective utilization rate of network bandwidth remains above 95% during non-warning periods, maximizing the utilization of expensive dedicated link resources.

[0055] Through the implementation of the aforementioned closed-loop feedback mechanism, this invention has achieved a significant breakthrough in system operation and maintenance efficiency and resource protection efficiency. Experimental data shows that compared with traditional static configuration or manual switching modes, this self-healing mechanism improves the efficiency of automated network resource recovery by approximately 70%. Simultaneously, by avoiding ineffective policy reconstruction and network jitter, the effective uptime in high-frequency continuous monitoring scenarios is increased to over 99.95%. This intelligent closed loop of "triggering upon risk perception and recovering upon risk resolution" truly endows large-scale traffic engineering monitoring systems with practical engineering value in dealing with extremely complex environments, significantly enhancing the real-time assurance capability of structural safety assessment while reducing operation and maintenance costs.

[0056] A preferred embodiment of the SDN transmission path dynamic scheduling system based on risk entropy awareness of the present invention includes the following modules: The engineering monitoring standard knowledge base construction module is used to build an engineering monitoring standard knowledge base. It stores the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring services in the engineering safety standards in a structured manner in the engineering monitoring standard knowledge base. Based on the structured information in the engineering monitoring standard knowledge base, it generates network configuration parameters and flow table generation rules that can be called by the SDN controller. The risk entropy index calculation module is used to perform spatiotemporal alignment and feature fusion processing on multimodal monitoring data from large-scale transportation infrastructure to obtain fused features. The fused features are then input into a risk identification model based on physical information to calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time. The emergency flow table policy distribution module is used by the SDN controller to determine the current risk level based on the risk entropy index, generate and distribute emergency flow table policies based on the risk level, network configuration parameters and flow table generation rules, and then perform deterministic bandwidth slicing and dynamic path scheduling, including: dynamically adjusting the bandwidth weight of preset virtual slices, implementing differentiated bandwidth suppression for multimodal monitoring data of non-critical services, and reselecting transmission paths for multimodal monitoring data of critical services; The monitoring data transmission module is used to perform differentiated encapsulation and transmission control of the multimodal monitoring data according to the risk level using the Structured Data-Oriented Transmission Protocol (SDDP). The emergency flow table policy restoration module is used to start an observation window based on hysteresis loop control logic when the physical monitoring index is detected to return to below the safety threshold; after confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated, it gradually reclaims the network resources reserved for critical services and restores the emergency flow table policy issued by the SDN controller.

[0057] The core logic of this invention lies in breaking the traditional "best-effort" general transmission mode of networks and constructing a communication resource on-demand slicing and dynamic scheduling system guided by engineering physical risks.

[0058] First, a leap from "static routing" to "semantic-driven path orchestration" is achieved in transmission path partitioning. Specifically, by constructing an engineering monitoring standard knowledge base, the real-time identified structural risk entropy (risk entropy index) is directly mapped to the optimization cost function of the underlying network. During the routine monitoring phase, data flows are distributed across multiple available physical links according to the principle of balance; however, once a structural anomaly is identified (such as bridge vibration or tunnel convergence), the SDN controller immediately initiates a path reselection algorithm to dynamically open a low-latency, high-reliability transmission path for critical high-frequency monitoring data. This path can avoid high-traffic congestion areas and has near-physical isolation transmission guarantees. This path partitioning mechanism not only achieves logical isolation between critical data and high-bandwidth services (such as high-definition video streams) but also ensures the priority passage of critical security data in extreme environments.

[0059] Secondly, at the channel resource allocation level, a sub-channel adaptive allocation mechanism based on multi-dimensional feature perception is introduced. By performing fine-grained logical slicing of the communication link, the bandwidth weight and scheduling priority of the sub-channels are dynamically adjusted according to the physical attributes of the monitoring data (such as transient damage pulses and quasi-static displacements) and security and timeliness requirements. In emergency situations, through the coordinated control logic of "bandwidth suppression and redundancy compensation," the sub-channel quotas of non-critical service flows are forcibly compressed, and the released frequency domain and time resources are compensated to high-frequency sensing sequences. Combined with a deterministic transmission protocol for structured data, efficient and lossless backhaul of critical monitoring data within the sub-channels is achieved.

[0060] Finally, a flexible balance logic between security strategy and transmission performance is achieved. When an extremely high risk level is identified, a "security load reduction mode" is proactively triggered on the transmission link, stripping away time-consuming encryption and verification mechanisms to maximize channel capacity and convert it into effective data payload, thereby eliminating processing fluctuations at the transmission layer. This closed-loop transmission system, which uses physical structural risks as its core instruction and consists of precise path partitioning, dynamic sub-channel suppression, and adaptive protocol degradation, significantly improves the anti-interference capability and emergency assessment accuracy of traffic engineering monitoring systems, achieving a technological leap from a "general transmission platform" to a "security-customized infrastructure."

[0061] The engineering monitoring standard knowledge base construction module specifically includes: The knowledge structured storage unit is used to construct an engineering monitoring standard knowledge base, which stores the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring operations in the engineering safety standards in a structured manner; the physical monitoring indicators include at least bridge deflection, tunnel settlement, and subway vibration. The dynamic compilation unit is used to deploy a dynamic compilation engine on the SDN controller. The dynamic compilation engine generates network configuration parameters and flow table generation rules that can be called by the SDN controller based on structured information from the engineering monitoring standard knowledge base.

[0062] The design of the structured mapping between engineering monitoring standards and flow table generation rules needs to start with the semantic reconstruction of the underlying protocol, transforming the "physical requirements" of engineering safety into "execution instructions" for network communication. This involves establishing a protocol parsing matrix based on an engineering standard knowledge graph. By extracting quantified thresholds for key indicators such as bridge deflection, tunnel settlement, or subway vibration from industry standards, these thresholds are abstracted into logical judgment benchmarks with engineering semantics. This step is no longer simple text storage; instead, it utilizes structured modeling techniques to compile parameters such as the hazard level of physical indicators, response time requirements, and data sampling frequency into a set of configuration parameters that the network controller can call in real time. This imbues the network system with a "decision-making brain" capable of understanding engineering risks.

[0063] The risk entropy index calculation module specifically includes: The fusion feature generation unit is used to perform spatiotemporal alignment of multimodal monitoring data from large-scale transportation infrastructure using high-precision timestamps. It adopts a normalized spatiotemporal mapping model based on non-uniform sampling reconstruction to adaptively upsample the low-frequency signals in the multimodal monitoring data and calculate the cross-correlation matrix between different modal data to perform feature fusion processing to obtain fused features. The multimodal monitoring data includes multiple types of data such as accelerometer data, stress-strain data, fiber optic grating dynamic strain and displacement data, settlement monitoring data, and video surveillance data. The model calculation unit is used to input the fused features into the risk identification model driven by physical information and calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time. The risk identification model is constructed based on a deep residual neural network, combined with dynamic weight risk entropy, and embeds partial differential equation constraints based on structural mechanical properties as physical constraint operators in the loss function. The resource reservation request sending unit is used to convert the risk entropy index into a network scheduling feedforward signal and output a predictive resource reservation request to the SDN controller.

[0064] In the logical construction of the perception and decision-making layer, the deep fusion of multimodal monitoring data is a prerequisite for achieving accurate risk identification. By deploying a spatiotemporal alignment engine at the edge or in the cloud, high-frequency vibration acceleration, low-frequency static leveling data, fiber optic strain signals, and traffic load flows captured by video are timestamped at the millisecond level and correlated with spatial coordinates. This process is not merely a physical stacking of data, but rather utilizes heterogeneous data association feature extraction technology to eliminate phase delays caused by differences in sensor physical characteristics, transforming isolated physical quantities into feature tensors reflecting structural evolution. This full-dimensional fusion of multi-source heterogeneous data provides a joint input with physical and logical support for subsequent highly reliable risk assessment.

[0065] To address the trade-off between accuracy and real-time performance in risk identification, a risk identification model (algorithm) combining a "physically driven deep residual neural network" and "dynamic weighted risk entropy" is introduced. This model compares multimodal input features (fusion features) in real time and uses residual mapping to capture nonlinear abnormal fluctuations in structural responses. The innovation lies in the embedded physical constraint operator, which automatically filters environmental noise interference and accurately identifies sudden safety events such as resonance in heavy-load convoys or brittle structural damage. Compared to traditional threshold alarms, this method significantly improves the accuracy of capturing weak damage signals while maintaining low computational load, elevating identification accuracy and false alarm suppression capabilities to engineering practice levels.

[0066] Once the risk entropy index is determined to overflow, it will quickly enter a "linked trigger mode." The core of this mode is to convert the risk entropy index into a predictive network scheduling feedforward signal in real time. The system can not only identify the current structural state but also predict potential transmission demand peaks within the next few seconds based on the evolution trend of multimodal monitoring data. This feedforward mechanism can output precise resource reservation requests to the SDN controller before actual network congestion occurs, ensuring that the risk identification module is not only an "observer" of the security status but also a "commander" of the underlying communication resources. This closed-loop logic from multimodal perception to proactive decision-making truly demonstrates the performance advantages of perception-driven architecture under extreme conditions in large-scale transportation engineering projects.

[0067] In the spatiotemporal alignment logic of multimodal data, the traditional simple interpolation method was abandoned, and a normalized spatiotemporal mapping model based on non-uniform sampling reconstruction was adopted instead. Due to the order-of-magnitude difference in sampling frequencies between the accelerometer (high frequency) and the hydrostatic level (low frequency), a time-domain synchronization operator Γ(t) based on high-precision GPS / PTP timestamps was established. This operator adaptively upsampled the low-frequency signal using Lagrange polynomials and, in conjunction with the cross-attention mechanism in a convolutional neural network, calculated the cross-correlation matrix between different modalities. This model can automatically compensate for the physical transmission delay offset Γi caused by the geographical distribution of the sensors, ensuring that the instantaneous load excitation value captured by the video stream and the strain response measured by the sensors are strictly spatiotemporally aligned on the order of 10^−3 seconds, providing a deterministic data basis for subsequent feature fusion.

[0068] The introduction of a Physics-Informed Operator is crucial for improving model accuracy and robustness. Partial Differential Equation (PDE) constraints based on the dynamic characteristics of bridges or tunnels are embedded in the loss function of the deep residual network. Specifically, the energy conservation terms and stiffness matrix degradation model in structural mechanics are transformed into a penalty term ℒphys in the algorithm, ensuring that feature extraction follows the physical evolution of the structure. When the model's predicted structural response deviates from physical reality (e.g., spurious features caused by environmental noise), this operator generates a large penalty value, forcing the neural network to correct the deviation. This design solves the "spurious correlation" problem that easily arises in purely data-driven models under small sample or high-noise environments, enabling the system to achieve an order-of-magnitude improvement in robustness to structural anomaly identification while maintaining low computational consumption.

[0069] From a performance optimization perspective, this approach achieves dimensionality reduction guidance for the probabilistic model through "physical logic." By directly integrating physical constraint operators into the gradient update process of the neural network, the model can locate the feature space reflecting structural damage with extremely fast convergence speed, avoiding the latency caused by large-scale black-box computation. In terms of practicality, it can filter out interference from conventional environmental vibrations caused by subway operation, accurately identifying truly engineering-significant risk features such as segment instability or main beam cracks. This efficient and accurate risk identification logic directly provides a high-quality trigger decision source for the SDN controller, ensuring that every bit of network bandwidth accurately serves real safety risks, significantly improving the system's real-time early warning performance.

[0070] In the emergency flow table policy distribution module, the dynamic path scheduling specifically includes path reselection, which employs an optimization algorithm based on a "communication-physical" dual-domain cost function. The expression of the dual-domain cost function is as follows: C_path=α*Σ(D_i / D_max+J_i / J_max+1 / (1-U_i))+β*exp(R*θ); Where C_path represents the comprehensive cost of the path; α represents the dynamically adjusted communication weight coefficient; β represents the dynamically adjusted engineering risk weight coefficient; D_i, J_i, and U_i represent the real-time latency, jitter, and bandwidth utilization of link i, respectively; D_max represents the maximum allowable latency threshold; J_max represents the maximum allowable jitter threshold; R represents the risk entropy index; and θ represents the link's sensitivity factor to packet loss of high-frequency data. When the risk entropy index R increases, the weight of β is increased, making path selection prioritize links with low packet loss rate and high stability. By introducing the nonlinear term (1-U_i), when the link is close to full load, the cost will explode exponentially, forcing data flow to avoid congested nodes in advance or generating a sharp increase in cost when the link load approaches saturation, forcibly triggering route reselection. Under normal monitoring, α dominates, pursuing global network throughput. After the structure enters the warning state (such as triggering a threshold above level 2), β grows exponentially, forcing the network to enter the "security guarantee mode".

[0071] The differentiated bandwidth suppression of multimodal monitoring data for non-critical services includes: when physical monitoring indicators trigger the warning threshold, the SDN controller locates and dynamically limits the transmission rate of multimodal monitoring data of non-critical services that occupy the most bandwidth, and redistributes the bandwidth resources released therefrom to multimodal monitoring data of critical services.

[0072] In the logical design of the transmission execution layer, perception-driven deterministic bandwidth slicing technology serves as the last line of defense for ensuring engineering security. Through the global view of the SDN controller, physical link resources are logically decoupled into multiple independent virtual slices, such as "core engineering sampling," "multi-dimensional environmental perception," and "routine operation and maintenance video." When the perception decision layer issues a high-risk entropy warning, the system no longer uses traditional static QoS configuration but instead initiates a dynamic slice resource reconstruction strategy. By calling the underlying OpenFlow protocol's Meter table and Queue mapping mechanism, the redistribution of slice weights can be completed in microseconds, instantly allocating limited bandwidth reserves to critical high-frequency monitoring data. This dynamic slicing technology ensures that, under extreme traffic loads or sudden disasters, the transmission path of core monitoring data has independent, undisturbed resource quotas, completely eliminating "channel preemption" of precision pulse signals by large-scale video traffic.

[0073] To address transmission conflicts under bandwidth-constrained conditions, a risk-linked, differentiated bandwidth suppression logic was designed. The core lies in proactive "on-demand degradation": once structural monitoring indicators trigger level one or two thresholds, the SDN controller precisely identifies the non-critical service flow (such as high-definition surveillance footage) with the highest occupancy rate in the current link and implements dynamic rate limiting. By adaptively adjusting the bandwidth limit of the video encoding stream, redundant bandwidth is forcibly redistributed to structural dynamic response signals, which are extremely sensitive to packet loss rates. This suppression logic is not a blind traffic cutoff, but rather a tiered backoff mechanism established based on engineering risk levels, maximizing the critical data carrying potential of narrowband communication networks while ensuring overall monitoring visibility.

[0074] In implementing path re-selection, this approach breaks away from traditional shortest path (Dijkstra's algorithm) or equivalent multipath (ECMP) algorithms, introducing a novel optimization algorithm based on a dual-domain cost function of "communication-physical". This algorithm jointly models real-time monitored network latency jitter with the risk sensitivity of the physical structure to calculate an optimal path for the current risk conditions. The innovation lies in the fact that when a predetermined path experiences momentary congestion or link quality degradation, the SDN controller can predictively switch critical monitoring data streams to "backup links" or "low-interference channels" with higher deterministic guarantees, rather than simply searching for the shortest physical distance with the fewest hops. Through this perception-driven path reconstruction, the end-to-end latency of critical data under extreme conditions can be reduced by approximately 35% to 50%, and the data transmission continuity guarantee rate can be improved to over 99.99%. This innovative design, which guides network flow based on physical risks, significantly improves the emergency response efficiency and structural assessment reliability of large-scale infrastructure in complex environments.

[0075] In path reselection decisions, traditional algorithms only focus on link bandwidth or hop count. This invention constructs a composite cost function C_path to calculate the cost of all available paths in real time and dynamically selects the path with the lowest cost for switching. This achieves deep coupling between the engineering physical security state and the underlying network communication environment. The core logic of the algorithm lies in using the real-time monitored risk entropy index as a "feedforward signal" for network path selection, dynamically adjusting the weight of each link in the network topology, thereby forcibly avoiding high-risk, high-congestion transmission paths under extreme conditions. This function consists of two core components: "network communication overhead" and "engineering risk sensitivity".

[0076] The monitoring data transmission module specifically includes: The data encapsulation unit is used to encapsulate the multimodal monitoring data according to the risk level using a structure-oriented data transmission protocol, embed engineering feature identifier bits in the data packet header, and mark the transmission priority and signal type according to the physical attributes of the multimodal monitoring data; the signal type includes at least high transient, high-frequency pulse-type damage signals and slowly varying signals; A differentiated transmission unit is used to enable forward error correction and dual-path redundant concurrent transmission mode for the pulse-type damaged signal; and to enable a verification-based reliable transmission mode for the slowly varying signal. The perceptual compression coding unit is used to trigger perceptual compression coding at the protocol layer when network congestion is detected and the structure is at the risk threshold edge, and to prioritize the transmission of core dynamic parameters in the multimodal monitoring data, including the peak value, valley value and spectral characteristic value of the waveform. When the risk entropy index exceeds a preset threshold and enters an emergency state, the transmission protocol temporarily removes or simplifies the security verification process, opens a fast forwarding channel based on hardware address binding for key monitoring nodes, and bypasses the security verification process to the cloud for processing. The delay jitter control unit is used to establish an end-to-end timing synchronization domain based on a precision clock protocol and control the delay jitter of data transmission through pre-allocated deterministic time slots; Physical semantic tags (transmission priority and signal type) are embedded in the packet header, and the transmission mode is switched in real time according to the risk level: in the case of extremely high risk (pulse-type damage signal), dual-path redundant concurrent mode is enabled, and forward error correction (FEC) mechanism is used to replace the traditional retransmission mechanism to eliminate delay jitter caused by packet loss and retransmission.

[0077] The SDDP protocol completely changes the traditional "black box transmission" mode of protocols, realizing a deterministic transmission mechanism that is deeply coupled with the dynamic response characteristics of engineering structures.

[0078] First, implement message differentiation encapsulation based on engineering semantics.

[0079] The SDDP protocol embeds an "engineering feature identifier" in the data frame header, enabling it to adjust transmission strategies in real time based on the dynamic features extracted by the sensing layer. For pulse-type damage signals with high transient and high-frequency characteristics, such as bridge cable vibration and sudden displacement of tunnel segments, it automatically identifies their physical meaning, assigns them a "real-time level" weight, and forcibly activates the "FEC error correction + dual-path spatial redundancy" mode. This means that even in environments with strong electromagnetic interference, such as subway tunnels, if a wireless or wired link experiences more than 10% random packet loss, the protocol can use redundant copies to perfectly reconstruct the original waveform of the structural damage at the receiving end without triggering a "retransmission request." For slowly changing tunnel convergence or static temperature-compensated displacement of long-span bridges, it switches to "reliable level" transmission, prioritizing the use of verification mechanisms to ensure the absolute accuracy of the data. This transmission logic, based on physical event attributes rather than simply data volume, ensures that critical damage data is never lost due to "head-of-line congestion" at the protocol layer, even under extremely complex operating conditions.

[0080] Second, design an adaptive protocol control logic that is linked to the physical sampling frequency.

[0081] To cope with the sudden surge in data traffic generated by strong winds or heavy vehicle traffic on bridges, the SDDP protocol incorporates a sampling rate-bandwidth coupling adjustment algorithm. When network path congestion (RTT fluctuations) is detected and the structure is nearing a safety risk threshold, the protocol layer no longer blindly discards packets but instead triggers a "physical feature point priority" transmission strategy. The protocol collaborates with sensor terminals to directly perform perceptual compression coding at the network layer: prioritizing the transmission of core dynamic parameters such as waveform peaks, valleys, zero-crossings, and spectral characteristic values, while downsampling redundant samples in stable regions as needed. This mechanism ensures that even under extreme bandwidth constraints, the structural evaluation algorithm at the analysis center can still acquire key features for support mode identification and stiffness assessment, completely avoiding the delay or "deadlock" of early warning information caused by the slow start mechanism of the traditional TCP protocol during congestion.

[0082] Third, ensure end-to-end timing determinism.

[0083] For multi-point synchronous vibration analysis, which is crucial in bridge health monitoring, the SDDP protocol establishes a unified physical clock synchronization domain based on PTP precision at the transport layer, strictly controlling the jitter of data encapsulation at each monitoring point to the microsecond level. Whether the monitoring points are distributed at both ends of a long-span bridge (kilometers or more) or in a tunnel section several kilometers long, SDDP ensures consistent physical timing of all modal data upon arrival at the central station through pre-allocated deterministic time slots. Experiments show that this transmission scheme, tailored for structural monitoring scenarios, strictly suppresses the end-to-end deterministic latency to within 20ms, providing extremely accurate "high-fidelity" time-domain input for the backend structural damage identification algorithm, significantly improving the accuracy of determining minute cracks or fatigue damage in the structure.

[0084] During the routine operation of the traffic engineering monitoring system, the SDDP protocol maintains a high level of security protection, including full message encryption, complex authentication, and integrity checks to defend against potential network attacks. However, once the physical constraint operator determines that the risk entropy index has overflowed and enters an emergency warning state, an adaptive degradation strategy prioritizing timeliness will be activated. Its innovation lies in the fact that the protocol layer automatically strips away non-core security overhead, temporarily disabling high-latency encryption algorithms and cumbersome handshake confirmation mechanisms, and allocating all computing resources and bandwidth to the monitoring data stream. This strategy effectively solves the latency jitter caused by encryption and decryption operations during emergencies, ensuring that the physical link from "emergency occurrence" to "instruction arrival" is in a simplified, high-speed direct connection state.

[0085] To ensure the self-healing and reliability of the system after degradation, this mechanism does not blindly abandon security, but rather compensates for security vulnerabilities through "trust pre-stored" and "post-audit" logic. Upon detecting an emergency trigger signal, the SDN controller opens a fast, green channel based on hardware MAC address binding for critical monitoring nodes, enabling line-speed forwarding without verification. Simultaneously, the relevant security verification steps are "stripped" from the transmission path and deployed in the cloud. This means that while ensuring the front-end monitoring curve is transmitted back with minimal latency (reduced by approximately 30% or more), the system can still maintain a basic security defense posture. Through this innovative "risk-awareness-guided security degradation," this invention achieves dynamic optimization of security resource utilization and emergency response efficiency while ensuring the deterministic transmission of "life-saving data" for engineering structures.

[0086] The emergency flow table strategy restoration module specifically includes: The observation window activation unit is used to activate a dynamic observation window based on hysteresis loop control logic when the physical monitoring index is detected to return to below the safety threshold, so as to continuously evaluate the structural status. The policy restoration instruction issuing unit is used to issue a policy restoration instruction to the SDN controller after confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated. The policy restoration instruction execution unit is used to, based on the policy restoration instruction, adopt a progressive bandwidth compensation strategy to orderly release network resources of suppressed non-critical services and cancel redundant backup paths to avoid secondary network impact, thereby recovering network resources reserved for critical services and restoring the emergency flow table policy issued by the SDN controller.

[0087] In the design of the closed-loop control layer, closed-loop feedback and flow table failure / restoration mechanisms are core components for ensuring a smooth return to normal operation from an emergency state and maximizing resource efficiency. At this stage, a hysteresis loop control logic based on an "observation window" is introduced to address the frequent network policy jumps (i.e., the "ping-pong effect") that may be caused by structural oscillations or sensor noise. When physical monitoring indicators return to below a safe threshold, high-priority dedicated slices are not immediately revoked; instead, an observation window linked to engineering dynamic characteristics is automatically opened. During this window, the attenuation trend of residual structural vibration is continuously assessed. Only when the physical risk entropy index remains stable at a low level and the network congestion risk is resolved will the SDN controller issue a policy restoration command. This design simulates the prudent decision-making logic of human experts during on-site handling, ensuring the self-healing capability and business continuity of the monitoring system during complex operating condition switching processes.

[0088] To optimize resource utilization, the system employs flow table lifecycle management (Soft / Hard Timeout) and refined resource reclamation algorithms to perform real-time "slimming down" of transmission links. During the policy restoration phase, the system releases the bandwidth cap of suppressed non-critical traffic (such as high-definition video streams) in an orderly manner according to the priority of business relevance and cancels redundant backup paths. The innovation lies in the use of a gradual bandwidth compensation strategy during the restoration process, rather than instantaneous release, thus avoiding the secondary impact on the network caused by a large backlog of data streams upon link recovery. Through this refined resource scheduling of "peak shaving and valley filling," the system can ensure that the effective utilization rate of network bandwidth remains above 95% during non-warning periods, maximizing the utilization of expensive dedicated link resources.

[0089] Through the implementation of the aforementioned closed-loop feedback mechanism, this invention has achieved a significant breakthrough in system operation and maintenance efficiency and resource protection efficiency. Experimental data shows that compared with traditional static configuration or manual switching modes, this self-healing mechanism improves the efficiency of automated network resource recovery by approximately 70%. Simultaneously, by avoiding ineffective policy reconstruction and network jitter, the effective uptime in high-frequency continuous monitoring scenarios is increased to over 99.95%. This intelligent closed loop of "triggering upon risk perception and recovering upon risk resolution" truly endows large-scale traffic engineering monitoring systems with practical engineering value in dealing with extremely complex environments, significantly enhancing the real-time assurance capability of structural safety assessment while reducing operation and maintenance costs.

[0090] In summary, the advantages of this invention are: 1. By structuring and internalizing engineering safety standards (such as early warning thresholds and response timeliness) into executable policies of the SDN controller, the basis for scheduling decisions is first established. Then, a risk entropy index of the structure is calculated in real time using a risk identification model based on physical information to achieve quantitative perception of security risks. With this as the core driver, the SDN controller dynamically triggers emergency flow table policies, executing dynamic scheduling including deterministic bandwidth guarantees for critical structure response data, bandwidth suppression for non-critical services, and reselection of high-quality paths for them. At the same time, data is encapsulated using appropriate transmission protocols according to the risk level. This series of measures ensures that when the risk increases, network resources can be allocated to the most critical monitoring data streams with priority and reliability. When the risk is eliminated, resources are intelligently recovered and normal operation is restored through hysteresis loop control logic. Thus, a closed loop of "perception and identification - decision-making and scheduling - transmission guarantee - elastic recovery" is constructed as a whole. Ultimately, the transmission priority is dynamically adjusted according to the risk, and network resources are optimized and allocated on demand, ensuring the real-time and reliable transmission of critical data and significantly improving the accuracy and timeliness of structural health assessment and security early warning.

[0091] 2. By constructing an engineering monitoring standard knowledge base, safety specifications are transformed into network rules. A risk identification model driven by physical information is used to calculate the "risk entropy index" of the structure in real time, which serves as the core driving signal. When the risk increases, the SDN controller dynamically implements differentiated bandwidth scheduling based on this index (such as suppressing high-bandwidth non-critical business video streams) and reselects high-reliability paths for critical structural response data (such as vibration and strain) based on a "communication-physical" dual-domain cost function that incorporates risk entropy. At the same time, emergency mechanisms such as fast forwarding and sensing compression are activated at the transmission protocol layer according to the risk level to ensure low-latency and reliable transmission of critical data. After the structural risk stabilizes, resources are gradually reclaimed. Overall, the network transmission is transformed from static configuration to dynamic intelligent scheduling driven by the real-time risk of the engineering entity, thereby prioritizing the real-time and continuous nature of critical monitoring data under bandwidth-constrained conditions, ultimately enhancing the accuracy and timeliness of structural health assessment and safety early warning.

[0092] 3. Traditional methods typically treat network status (such as bandwidth and latency) and engineering safety as two independent systems. This invention utilizes the cross-domain quantitative indicator of "risk entropy index" as the core of network scheduling decisions. By constructing a risk identification model driven by physical information, it deeply integrates multimodal physical monitoring data (such as stress and displacement) and calculates a unified "risk entropy." This enables the SDN controller to "understand" the safety status of the engineering structure, achieving a paradigm shift from "passive response based on network status" to "proactive and predictive scheduling based on engineering risks." This allows for deep integration and intelligent linkage between network resource scheduling and engineering safety requirements.

[0093] 4. The network scheduling strategy is not static or based on simple priorities, but is driven by the dynamic risk level determined by "risk entropy" to execute precise "deterministic bandwidth slicing" and "dynamic path scheduling". Its "communication-physical" dual-domain cost function can automatically adjust the weights when the risk increases, so that the path selection prioritizes the critical monitoring data flow that is sensitive to packet loss. At the same time, by implementing "differentiated bandwidth suppression" for non-critical services, it can quickly and directionally release and guarantee bandwidth resources for critical services in emergency situations, realizing the optimal elastic allocation of network resources in the spatiotemporal dimension, and ensuring the absolute priority and determinism of critical monitoring information transmission in critical situations.

[0094] 5. Through deep optimization at the transport layer, the protocol can adaptively select encapsulation, error correction, and transmission modes based on the physical properties of the data (such as pulse-type damaged signals and slowly varying signals) and the real-time risk level. Especially in emergency situations, the protocol can temporarily simplify security verification, open fast channels, and bypass verification. This design greatly reduces transmission latency while ensuring basic security, meeting the extreme timeliness requirements of engineering safety emergency response. At the same time, the perceptual compression coding and core parameter priority transmission mechanism ensure that the most critical engineering status information can still be transmitted when the network is congested, maximizing the information value under limited bandwidth.

[0095] 6. By establishing a complete "perception-decision-execution-recovery" intelligent loop, from knowledge base and rule pre-setting, real-time risk perception and prediction, network policy generation and distribution, and differentiated data transmission, to the gradual recovery of resources and policy restoration after risk resolution, the entire process can automatically complete the optimization of network resource allocation, emergency response and smooth recovery based on the dynamic changes in the project status, significantly improving the overall autonomy and operation and maintenance efficiency of the large-scale infrastructure monitoring system.

[0096] 7. During the recovery phase after risk mitigation, a "hysteresis loop control logic" and a "gradual recovery" strategy were adopted. This does not mean immediately withdrawing all emergency measures as soon as the indicators return to the threshold, but rather opening an "observation window" to confirm that the risk has been stabilized and resolved. This design can effectively avoid frequent and drastic policy switching (i.e., "system oscillation") caused by short-term fluctuations in monitoring data or instantaneous improvement in network status, thereby ensuring the stability of network policies and the overall state of the project, preventing secondary impacts on the network and services caused by policy switching, and demonstrating foresight and robustness.

[0097] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the present invention. Equivalent modifications and variations made by those skilled in the art in accordance with the spirit of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A dynamic scheduling method for SDN transmission paths based on risk entropy awareness, characterized in that: Includes the following steps: Step S10: Construct an engineering monitoring standard knowledge base. Store the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring services in the engineering safety standards in the engineering monitoring standard knowledge base in a structured manner. Based on the structured information in the engineering monitoring standard knowledge base, generate network configuration parameters and flow table generation rules that can be called by the SDN controller. Step S20: Perform spatiotemporal alignment and feature fusion processing on multimodal monitoring data from large-scale transportation infrastructure to obtain fused features. Input the fused features into a risk identification model driven by physical information to calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time. Step S30: The SDN controller determines the current risk level based on the risk entropy index, generates and issues an emergency flow table strategy based on the risk level, network configuration parameters and flow table generation rules, and then performs deterministic bandwidth slicing and dynamic path scheduling, including: dynamically adjusting the bandwidth weight of preset virtual slices, implementing differentiated bandwidth suppression for multimodal monitoring data of non-critical services, and reselecting transmission paths for multimodal monitoring data of critical services. Step S40: Based on the risk level, use a structure-oriented data transmission protocol to perform differentiated encapsulation and transmission control on the multimodal monitoring data; Step S50: When the physical monitoring index is detected to return to below the safety threshold, the observation window is started based on the hysteresis loop control logic; after confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated, the network resources reserved for critical services are gradually reclaimed, and the emergency flow table policy issued by the SDN controller is restored. 2.The risk entropy-aware SDN transmission path dynamic scheduling method of claim 1, wherein: Step S10 specifically includes: Step S11: Construct an engineering monitoring standard knowledge base, and structurally store the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring operations in the engineering safety standards into the engineering monitoring standard knowledge base; the physical monitoring indicators include at least bridge deflection, tunnel settlement, and subway vibration; Step S12: Deploy a dynamic compilation engine on the SDN controller. The dynamic compilation engine generates network configuration parameters and flow table generation rules that can be called by the SDN controller based on structured information from the engineering monitoring standard knowledge base.

3. The SDN transmission path dynamic scheduling method based on risk entropy awareness as described in claim 1, characterized in that: Step S20 specifically includes: Step S21: Use high-precision timestamps to perform spatiotemporal alignment on multimodal monitoring data from large-scale transportation infrastructure. Use a normalized spatiotemporal mapping model based on non-uniform sampling reconstruction to adaptively upsample the low-frequency signals in the multimodal monitoring data. Calculate the cross-correlation matrix between different modal data to perform feature fusion processing and obtain fused features. The multimodal monitoring data includes multiple types of data such as accelerometer data, stress-strain data, fiber optic grating dynamic strain and displacement data, settlement monitoring data, and video surveillance data. Step S22: Input the fused features into the risk identification model driven by physical information to calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time; the risk identification model is constructed based on a deep residual neural network, combined with dynamic weight risk entropy, and embeds partial differential equation constraints based on structural mechanical properties as physical constraint operators in the loss function; Step S23: Convert the risk entropy index into a network scheduling feedforward signal and output a predictive resource reservation request to the SDN controller.

4. The SDN transmission path dynamic scheduling method based on risk entropy awareness as described in claim 1, characterized in that: In step S30, the dynamic path scheduling specifically includes path reselection, which employs an optimization algorithm based on a "communication-physical" dual-domain cost function. The expression of the dual-domain cost function is as follows: C_path=α*Σ(D_i / D_max+J_i / J_max+1 / (1-U_i))+β*exp(R*θ); Where C_path represents the comprehensive cost of the path; α represents the dynamically adjusted communication weight coefficient; β represents the dynamically adjusted engineering risk weight coefficient; D_i, J_i, and U_i represent the real-time latency, jitter, and bandwidth utilization of link i, respectively; D_max represents the maximum allowable latency threshold; J_max represents the maximum allowable jitter threshold; R represents the risk entropy index; and θ represents the link's sensitivity factor to packet loss of high-frequency data. When the risk entropy index R increases, the weight of β is increased, so that path selection prioritizes links with low packet loss rate and high stability. The differentiated bandwidth suppression of multimodal monitoring data for non-critical services includes: when physical monitoring indicators trigger the warning threshold, the SDN controller locates and dynamically limits the transmission rate of multimodal monitoring data of non-critical services that occupy the most bandwidth, and redistributes the bandwidth resources released therefrom to multimodal monitoring data of critical services.

5. The SDN transmission path dynamic scheduling method based on risk entropy awareness as described in claim 1, characterized in that: Step S40 specifically includes: Step S41: Based on the risk level, encapsulate the multimodal monitoring data using a structure-oriented data transmission protocol, embed engineering feature identifier bits in the data packet header, and mark the transmission priority and signal type according to the physical attributes of the multimodal monitoring data; the signal type includes at least high transient, high-frequency pulse-type damage signals and slowly varying signals; Step S42: For the pulse-type damage signal, enable forward error correction and dual-path redundant concurrent transmission mode; for the slowly varying signal, enable verification-based reliable transmission mode. Step S43: When network congestion is detected and the structure is at the risk threshold edge, the protocol layer triggers perceptual compression coding to prioritize the transmission of core dynamic parameters in the multimodal monitoring data, including the peak value, valley value and spectral feature value of the waveform. When the risk entropy index exceeds a preset threshold and enters an emergency state, the transmission protocol temporarily removes or simplifies the security verification process, opens a fast forwarding channel based on hardware address binding for key monitoring nodes, and bypasses the security verification process to the cloud for processing. Step S44: Establish an end-to-end timing synchronization domain based on a precision clock protocol, and control the latency jitter of data transmission through pre-allocated deterministic time slots; Step S50 specifically includes: Step S51: When the physical monitoring index is detected to return to below the safety threshold, a dynamic observation window is started based on the hysteresis loop control logic to continuously evaluate the structural status. Step S52: After confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated, the SDN controller issues a policy restoration command. Step S53: Based on the policy restoration instruction, adopt a progressive bandwidth compensation strategy to release the network resources of suppressed non-critical services in an orderly manner and cancel redundant backup paths to avoid secondary network impact, thereby recovering the network resources reserved for critical services and restoring the emergency flow table policy issued by the SDN controller.

6. A risk entropy-aware SDN transmission path dynamic scheduling system, characterized in that: Includes the following modules: The engineering monitoring standard knowledge base construction module is used to build an engineering monitoring standard knowledge base. It stores the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring services in the engineering safety standards in a structured manner in the engineering monitoring standard knowledge base. Based on the structured information in the engineering monitoring standard knowledge base, it generates network configuration parameters and flow table generation rules that can be called by the SDN controller. The risk entropy index calculation module is used to perform spatiotemporal alignment and feature fusion processing on multimodal monitoring data from large-scale transportation infrastructure to obtain fused features. The fused features are then input into a risk identification model based on physical information to calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time. The emergency flow table policy distribution module is used by the SDN controller to determine the current risk level based on the risk entropy index, generate and distribute emergency flow table policies based on the risk level, network configuration parameters and flow table generation rules, and then perform deterministic bandwidth slicing and dynamic path scheduling, including: dynamically adjusting the bandwidth weight of preset virtual slices, implementing differentiated bandwidth suppression for multimodal monitoring data of non-critical services, and reselecting transmission paths for multimodal monitoring data of critical services; The monitoring data transmission module is used to perform differentiated encapsulation and transmission control of the multimodal monitoring data according to the risk level and using a structured data-oriented transmission protocol; The emergency flow table policy restoration module is used to start an observation window based on hysteresis loop control logic when the physical monitoring index is detected to return to below the safety threshold; after confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated, it gradually reclaims the network resources reserved for critical services and restores the emergency flow table policy issued by the SDN controller.

7. The SDN transmission path dynamic scheduling system based on risk entropy awareness as described in claim 6, characterized in that: The engineering monitoring standard knowledge base construction module specifically includes: The knowledge structured storage unit is used to construct an engineering monitoring standard knowledge base, which stores the physical monitoring indicators, early warning thresholds, response time requirements, and key levels of monitoring operations in the engineering safety standards in a structured manner; the physical monitoring indicators include at least bridge deflection, tunnel settlement, and subway vibration. The dynamic compilation unit is used to deploy a dynamic compilation engine on the SDN controller. The dynamic compilation engine generates network configuration parameters and flow table generation rules that can be called by the SDN controller based on structured information from the engineering monitoring standard knowledge base.

8. The SDN transmission path dynamic scheduling system based on risk entropy awareness as described in claim 6, characterized in that: The risk entropy index calculation module specifically includes: The fusion feature generation unit is used to perform spatiotemporal alignment of multimodal monitoring data from large-scale transportation infrastructure using high-precision timestamps. It adopts a normalized spatiotemporal mapping model based on non-uniform sampling reconstruction to adaptively upsample the low-frequency signals in the multimodal monitoring data and calculate the cross-correlation matrix between different modal data to perform feature fusion processing to obtain fused features. The multimodal monitoring data includes multiple types of data such as accelerometer data, stress-strain data, fiber optic grating dynamic strain and displacement data, settlement monitoring data, and video surveillance data. The model calculation unit is used to input the fused features into the risk identification model driven by physical information and calculate the risk entropy index of the structure of large-scale transportation infrastructure in real time. The risk identification model is constructed based on a deep residual neural network, combined with dynamic weight risk entropy, and embeds partial differential equation constraints based on structural mechanical properties as physical constraint operators in the loss function. The resource reservation request sending unit is used to convert the risk entropy index into a network scheduling feedforward signal and output a predictive resource reservation request to the SDN controller.

9. A risk entropy-aware SDN transmission path dynamic scheduling system as described in claim 6, characterized in that: In the emergency flow table policy distribution module, the dynamic path scheduling specifically includes path reselection, which employs an optimization algorithm based on a "communication-physical" dual-domain cost function. The expression of the dual-domain cost function is as follows: C_path=α*Σ(D_i / D_max+J_i / J_max+1 / (1-U_i))+β*exp(R*θ); Where C_path represents the comprehensive cost of the path; α represents the dynamically adjusted communication weight coefficient; β represents the dynamically adjusted engineering risk weight coefficient; D_i, J_i, and U_i represent the real-time latency, jitter, and bandwidth utilization of link i, respectively; D_max represents the maximum allowable latency threshold; J_max represents the maximum allowable jitter threshold; R represents the risk entropy index; and θ represents the link's sensitivity factor to packet loss of high-frequency data. When the risk entropy index R increases, the weight of β is increased, so that path selection prioritizes links with low packet loss rate and high stability. The differentiated bandwidth suppression of multimodal monitoring data for non-critical services includes: when physical monitoring indicators trigger the warning threshold, the SDN controller locates and dynamically limits the transmission rate of multimodal monitoring data of non-critical services that occupy the most bandwidth, and redistributes the bandwidth resources released therefrom to multimodal monitoring data of critical services.

10. A risk entropy-aware SDN transmission path dynamic scheduling system as described in claim 6, characterized in that: The monitoring data transmission module specifically includes: The data encapsulation unit is used to encapsulate the multimodal monitoring data according to the risk level using a structure-oriented data transmission protocol, embed engineering feature identifier bits in the data packet header, and mark the transmission priority and signal type according to the physical attributes of the multimodal monitoring data; the signal type includes at least high transient, high-frequency pulse-type damage signals and slowly varying signals; A differentiated transmission unit is used to enable forward error correction and dual-path redundant concurrent transmission mode for the pulse-type damaged signal; and to enable a verification-based reliable transmission mode for the slowly varying signal. The perceptual compression coding unit is used to trigger perceptual compression coding at the protocol layer when network congestion is detected and the structure is at the risk threshold edge, and to prioritize the transmission of core dynamic parameters in the multimodal monitoring data, including the peak value, valley value and spectral characteristic value of the waveform. When the risk entropy index exceeds a preset threshold and enters an emergency state, the transmission protocol temporarily removes or simplifies the security verification process, opens a fast forwarding channel based on hardware address binding for key monitoring nodes, and bypasses the security verification process to the cloud for processing. The delay jitter control unit is used to establish an end-to-end timing synchronization domain based on a precision clock protocol and control the delay jitter of data transmission through pre-allocated deterministic time slots; The emergency flow table strategy restoration module specifically includes: The observation window activation unit is used to activate a dynamic observation window based on hysteresis loop control logic when the physical monitoring index is detected to return to below the safety threshold, so as to continuously evaluate the structural status. The policy restoration instruction issuing unit is used to issue a policy restoration instruction to the SDN controller after confirming that the risk entropy index is stable at a low level and the network congestion risk is eliminated. The policy restoration instruction execution unit is used to, based on the policy restoration instruction, adopt a progressive bandwidth compensation strategy to orderly release network resources of suppressed non-critical services and cancel redundant backup paths to avoid secondary network impact, thereby recovering network resources reserved for critical services and restoring the emergency flow table policy issued by the SDN controller.

Images