Bridge structure load sensing driven photovoltaic energy supply method and system

Abstract

The invention relates to the technical field of photovoltaic energy regulation and control, in particular to a photovoltaic energy supply method and system driven by load sensing of a bridge structure. The method comprises the following steps: acquiring load response data through sensors arranged at key parts of a bridge; carrying out load sensing feature extraction according to the load response data to obtain load sensing feature data; performing load energy regulation and control mapping on the load sensing characteristic data to obtain photovoltaic regulation and control data; acquiring historical traffic flow data, and performing digital twinborn body construction according to the historical traffic flow data and a preset bridge-vehicle coupling model to obtain bridge-vehicle twinborn body data; and performing energy regulation and control response according to the photovoltaic regulation and control data and the bridge vehicle twinborn body data to obtain photovoltaic regulation and control response data. According to the invention, response control is carried out by combining twinborn body data and photovoltaic regulation and control data, so that the timeliness and execution precision of energy allocation are effectively improved, and the operation safety of a bridge structure and the utilization efficiency of renewable energy sources are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic energy regulation, and particularly to a photovoltaic power supply method and system driven by bridge structure load perception. Background Art

[0002] With the integrated development of the intelligentization of transportation infrastructure and green energy technology, the state perception and energy supply regulation of bridge structures have increasingly become the focus of research. Most existing bridge energy supply systems adopt a fixed power strategy, lacking the dynamic adaptability to the structural operation state, which easily leads to a decline in energy supply efficiency or a lag in energy dispatching during periods of severe load fluctuations or sudden impacts. At the same time, traditional photovoltaic regulation systems generally rely on weather forecasts or external environmental information for output adjustment and are difficult to perceive the load state inside the bridge in real time. Therefore, how to achieve a deep coupling of structural response and energy management has become a problem. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a photovoltaic power supply method and system driven by bridge structure load perception to solve at least one of the above technical problems.

[0004] The present application provides a photovoltaic power supply method driven by bridge structure load perception, and the method includes: S1. Collect load response data through sensors arranged at key parts of the bridge; extract load perception feature data according to the load response data; S2. Perform load energy regulation mapping on the load perception feature data to obtain photovoltaic regulation data; S3. Obtain historical traffic flow data, and construct a digital twin based on the historical traffic flow data and a preset bridge-vehicle coupling model to obtain bridge-vehicle twin data; S4. Perform energy regulation response according to the photovoltaic regulation data and the bridge-vehicle twin data to obtain photovoltaic regulation response data.

[0005] In the present invention, real-time acquisition of structural response is achieved through multi-source sensors arranged at key parts of the bridge, and high-dimensional load state information is obtained based on perception feature extraction, which can enhance the recognition ability of abnormal states such as impact loads and modal drift; through the regulation mapping mechanism of load characteristics and energy strategies, the photovoltaic power supply system is enabled to have the ability to dynamically respond to changes in the bridge state, improving the structural adaptability of energy distribution; introducing historical traffic flow and a bridge-vehicle coupling model to construct a digital twin can realize the prediction of future short-term structural response trends, providing a feedforward basis for energy dispatching; by combining the results of twin deduction and regulation instructions, an executable photovoltaic response plan is output, improving the collaborative efficiency of the adaptive adjustment ability of the power supply system and the structural safety.

[0006] Optionally, S1 includes: Collect load response data through sensors arranged at key parts of the bridge; Conduct bionic dynamic expression construction based on the load response data to obtain bionic expression data; Construct a modal energy spectrum for the bionic expression data to obtain modal energy spectrum data; Perform graph embedding convolution on the modal energy spectrum data to obtain structure-aware data; Identify structure impacts on the structure-aware data to obtain structure impact data; Attribute the load behavior based on the structure impact data to obtain load perception feature data.

[0007] In the present invention, through bionic dynamic expression construction, the structural response data is made to have time-coding characteristics similar to biological nerve impulses, enhancing the ability to capture sudden load impacts and time-varying modal responses; through modal energy spectrum construction and graph embedding convolution, the propagation relationship of multi-frequency modal energy among spatial structure nodes is modeled, enhancing the ability to identify the coupling of local modal excitation and overall vibration modes; combining the structure impact identification and load behavior attribution processes can not only accurately locate the influence path of impact events in the structural system but also extract the corresponding load inducement types, obtaining load perception feature data with temporality, spatiality, and causal semantics, providing a data basis with in-depth understanding of the structural state for the generation of photovoltaic regulation strategies, and significantly improving the decision-making accuracy and response reliability of the bridge intelligent energy supply system.

[0008] Optionally, the bionic dynamic expression construction includes: Perform time-based synchronous conversion on the load response data to obtain response conversion data; Conduct neuron-like excitation on the response conversion data to obtain pulse sequence data; Perform nerve synapse structure mapping on the pulse sequence data to obtain synaptic activation weight map data; Extract bionic dynamic expression features based on the synaptic activation weight map data to obtain bionic expression data.

[0009] In the present invention, through time-based synchronous conversion, the time expression density of different types of sensor data can be unified, making the response process have physiological rhythm perception characteristics; through the neuron-like excitation mechanism, continuous response signals are encoded into pulse sequences with event-driven characteristics, enhancing the temporal resolution ability for key dynamic features such as impacts and mutations; introducing nerve synapse structure mapping can construct activation correlation relationships between nodes and simulate the load conduction paths between structural regions; the extracted bionic dynamic expression features have sparsity, temporality, and compressibility, and can more effectively represent the dynamic response behavior of the structure under complex loads.

[0010] Optionally, the modal energy spectrum construction includes: Perform periodic structure reconstruction on the bionic expression data to obtain periodic response data; Perform frequency-domain analysis on the periodic response data to obtain modal frequency-domain data; Extract the modal energy spectrum from the modal frequency-domain data to obtain modal energy spectrum data; Construct a modal excitation map based on the modal energy spectrum data to obtain modal energy map data.

[0011] In the present invention, by performing periodic structure reconstruction on the bionic expression data, the approximate periodic response process of the structure under load excitation can be effectively restored, providing a stable signal basis for frequency-domain analysis; the frequency-domain analysis step can identify the frequency components of each order of the modal in the structural response, realizing the joint perception of the low-order modes and high-frequency disturbances; the extraction of the modal energy spectrum can quantify the energy distribution of different modes in the current structural state, thereby revealing the excitation intensity and coupling effect of the structure in a specific frequency band; by constructing the modal excitation map, not only can the distribution characteristics of the modal energy among the spatial nodes of the bridge structure be clearly presented, but it can also be used to identify potential instability risk areas such as modal local concentration areas and abnormal high-energy areas.

[0012] Optionally, the structural impact identification includes: Perform structural disturbance screening on the structural perception data to obtain structural disturbance data; Perform vibration energy analysis on the structural disturbance data to obtain vibration energy data; Perform impact event matching based on the vibration energy data to obtain impact event data; Perform impulse response fitting based on the impact event data to obtain structural impact data.

[0013] In the present invention, through structural disturbance screening, the low-amplitude stable segments can be removed from the continuous perception data, and the disturbance segments with significant dynamic characteristics can be retained, improving the computational efficiency and data validity of subsequent processing; based on vibration energy analysis, the energy mutation points can be identified, accurately capturing the spatio-temporal positions where the structure undergoes severe disturbances, providing a basis for the accurate positioning of impact events; the impact event matching process is based on a preset template or feature comparison mechanism, which can determine whether there are typical impact events and eliminate false detection interference; through impulse response fitting, the response mode characteristics of the structure under impact (such as response amplitude, duration, main frequency component, etc.) can be extracted, thereby obtaining structural impact data with physical explanatory power.

[0014] Optionally, S2 includes: Select the regulation factor for the load perception feature data to obtain regulation factor data; Construct the load state space based on the regulation factor data to obtain load state space data; Grey decision-making is carried out based on the load state space data to obtain the photovoltaic strategy matching data; Photovoltaic regulation network output is carried out according to the photovoltaic strategy matching data to obtain the photovoltaic regulation data.

[0015] In the present invention, by selecting regulation factors for the load perception feature data, key indicators highly correlated with energy response (such as impact intensity, modal drift rate, strain growth trend, etc.) can be extracted from multi-dimensional structural information, realizing feature dimension compression and decision variable focusing; the load state space construction step establishes a multi-dimensional state expression of the structural working conditions based on the selected factors, providing a unified expression format for the decision-making model; the system conducts grey decision-making, which can, under the conditions of incomplete data and uncertain information, evaluate the similarity between the current state and historical strategies through grey correlation degree, realizing strategy matching driven by experience; based on the photovoltaic strategy matching result, combined with the neural network regulation module, accurate photovoltaic power adjustment parameters and control instructions are output, enabling the system to have the ability to quickly adapt to changes in the structural state.

[0016] Optionally, the grey decision-making includes: Modal change trend processing is carried out on the load state space data to obtain the load change space data; Grey correlation degree calculation is carried out according to the load change space data and the preset historical regulation library to obtain the grey correlation degree data; Strategy matching is carried out according to the grey correlation degree data to obtain the photovoltaic strategy matching data.

[0017] In the present invention, by carrying out modal change trend processing on the load state space data, the regularity and mutability of the structural modal frequency evolving with time can be effectively captured, enhancing the ability to describe the long-term and short-term trends of load disturbances; calculating the grey correlation degree between the extracted load change space data and the preset historical regulation library can establish the similarity correlation between the current structural state and historical effective strategies under the condition of limited sample size or incomplete data, with good robustness and adaptability; the system realizes strategy matching through grey correlation coefficient sorting, effectively avoiding the problems of being sensitive to outliers and lacking fuzzy fault tolerance ability in the selection of traditional nearest neighbor strategies. The obtained photovoltaic strategy matching data can provide an accurate and reliable strategy basis for the energy control system, thereby realizing flexible photovoltaic regulation driven by the bridge structure state and improving the system operation efficiency and load response coupling ability.

[0018] Optionally, S3 includes: Obtain historical traffic flow data; Extract traffic semantic features according to the historical vehicle data to obtain the traffic semantic feature data; Parameter recalibration is carried out according to the traffic semantic feature data and the preset bridge-vehicle coupling model to obtain the bridge-vehicle update model; Capture the impact propagation path of the updated car model to obtain impact propagation path data; Perform evolutionary simulation based on the impact propagation path data to obtain car twin data.

[0019] In the present invention, by obtaining historical traffic flow data and extracting traffic semantic features, semantic parameters such as vehicle flow patterns, load combination forms, and vehicle speed distributions can be constructed, providing support for coupled modeling at the behavioral level; based on traffic semantic features, parameter recalibration of the car-bridge coupling model can improve the adaptability and timeliness of the model to the actual load environment, making its dynamic response simulation closer to the real working conditions; through impact propagation path capture, the excitation path, conduction channel, and modal coupling key points of traffic flow loads in the bridge structure can be identified, strengthening the coupled modeling of local impact sources and the entire bridge response section; by generating car twin data through evolutionary simulation, it not only has temporal continuity and spatial resolution, but can also be used to predict the short-term structural state trend in the future.

[0020] Optionally, S4 includes: Construct a state trigger matrix based on the photovoltaic regulation data and the car twin data to obtain state trigger data; Perform logistic regression processing on the state trigger data to obtain response condition data; Generate a response path based on the response condition data to obtain response path data; Perform redundant screening on the response path data to obtain path screening data; Perform execution level control on the path screening data to obtain photovoltaic regulation response data.

[0021] In the present invention, by combining photovoltaic regulation data and car twin data to construct a state trigger matrix, the coupled modeling between the predicted structural state and energy demand can be realized, and the response trigger conditions at critical moments can be identified; using logistic regression to process the state trigger data can extract response conditions with high correlation and discriminative power from multi-dimensional variables, constructing a conditional judgment mechanism with statistical learning ability; through response path generation, various regulation strategy combination schemes can be formed, improving the strategy flexibility of the system; subsequently, a redundant screening mechanism is introduced to eliminate paths with low performance or high risk, retaining the optimal control channel, significantly enhancing the regulation robustness of the system; by performing execution level control on the response path for instruction output and grading execution according to the matching degree between the response intensity and the structural state, it is ensured that the regulation process has the ability of flexible switching and gradual adjustment, so as to achieve the efficient response and precise control of the photovoltaic system under complex structural states.

[0022] Optionally, the present application also provides a photovoltaic power supply system driven by bridge structure load perception, which is used to execute the photovoltaic power supply method driven by bridge structure load perception as described above. The photovoltaic power supply system driven by bridge structure load perception includes: A structure load perception module, which is used to collect load response data through sensors arranged at key parts of the bridge; extract load perception features from the load response data to obtain load perception feature data; A load-driven energy regulation mapping module, which is used to perform load energy regulation mapping on the load perception feature data to obtain photovoltaic regulation data; A bridge-vehicle coupling digital twin prediction module, which is used to obtain historical traffic flow data and construct a digital twin based on the historical traffic flow data and a preset bridge-vehicle coupling model to obtain bridge-vehicle twin data; An intelligent energy response control module, which is used to perform energy regulation response based on the photovoltaic regulation data and the bridge-vehicle twin data to obtain photovoltaic regulation response data.

[0023] The purpose of the present invention is to be able to collect structural load response data in real time by arranging multiple types of sensors at key parts of the bridge, and to identify key load behaviors through feature extraction, so as to achieve precise perception of the structural state; through the mapping model between load characteristics and photovoltaic regulation logic, the photovoltaic system is enabled to have the ability to dynamically adjust the output power and energy storage strategy according to the structural state, improving the response sensitivity of the system; constructing a digital twin based on historical traffic flow data and a bridge-vehicle coupling model can realize the prediction and deduction of the propagation path of future short-term structural impacts, providing feedforward support for the regulation strategy; combining twin data and photovoltaic regulation data for response control effectively improves the timeliness and execution accuracy of energy allocation, thereby enhancing the operational safety of the bridge structure and the utilization efficiency of renewable energy, and realizing the deep coupling of structural monitoring and green energy supply systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, objects and advantages of the present application will become more obvious: Figure 1 Shows the step flow chart of a photovoltaic power supply method driven by bridge structure load perception in an embodiment; Figure 2 Shows the step flow chart of a structure load perception method in an embodiment; Figure 3 Shows the step flow chart of a load-driven energy regulation mapping method in an embodiment; Figure 4 Shows the step flow chart of a bridge-vehicle coupling digital twin prediction method in an embodiment; The realization, functional features and advantages of the present invention will be further described in conjunction with embodiments with reference to the accompanying drawings. Specific Embodiments

[0025] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0026] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0027] It should be understood that although the terms "first", "second", etc. may be used here to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly, the second unit may be referred to as the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed related items.

[0028] Please refer to Figures 1 to 4 , this application provides a photovoltaic energy supply method driven by bridge structure load perception, and the method includes: S1. Collect load response data through sensors arranged at key parts of the bridge; extract load perception feature data according to the load response data. In one embodiment, the load response data of the bridge structure is collected by deploying multiple types of sensors at key parts of the bridge. Strain gauges, accelerometers, and displacement sensors are deployed in areas where structures such as the main girder, web, bridge deck pavement, bearings, and suspension cables are prone to stress deformation; the above sensors are connected to the edge acquisition terminal module through a wired network or a low-power wide-area wireless communication network (such as LoRa). The sensor sampling frequency is set to not less than 200 Hz. All the collected data is attached with unified timestamp information, and time synchronization and cache update are completed in the local cache node. In terms of load perception feature extraction, the system adopts a time window processing mechanism. The time window length is set to 5 seconds, and the time window step is 0.5 seconds. In each sliding window, the system extracts the following typical load perception feature indicators, including peak acceleration (the maximum vibration response amplitude), which represents the maximum value of the acceleration data recorded within the window and is used to measure the current impact response intensity of the bridge; the main modal frequency change rate, which calculates the difference between the current main frequency modal value and the main frequency of the previous time window; the instantaneous strain growth rate, which is calculated through the derivative form of the strain time series; the vibration energy density per unit time, which calculates the energy index by integrating the square of the acceleration within the window, that is , is the vibration energy density per unit time, is the initial time data, is the time window length data, is the acceleration. The above feature data constitutes a load perception feature data structure, including but not limited to the following fields, such as the start time of the current time window, the maximum acceleration value, the modal frequency change rate, the strain growth rate, the vibration energy density, etc.

[0029] S2. Perform load energy regulation mapping on the load perception feature data to obtain photovoltaic regulation data; In one embodiment, the system screens the regulation factors for the key variables in the load perception feature data. Select two feature variables that have the most influence on energy regulation, namely the main modal frequency change rate ( ) and the vibration energy density per unit time (E), and combine them to form a state space vector . The system performs a normalization operation on each dimension of the feature variables in , using the minimum-maximum scaling rule based on historical samples, that is, scaling the current value of each variable to the interval [0,1], so that the state space data has consistent dimensional characteristics and a comparison basis. The normalized state space set is defined as: , where is the index of the current sliding window moment. The system performs a grey decision-making process on the state space data to match the most suitable current photovoltaic control strategy from a preset historical energy regulation strategy library. A correlation coefficient calculation model based on grey system theory is used to evaluate the similarity between each historical strategy sample and the current state space vector. For any th historical strategy sample, the th eigenvalue is denoted as , then its grey correlation coefficient with the current state is calculated as follows: , where is the grey correlation coefficient between the th historical strategy sample and the current state, is the minimum value function, is the th eigenvalue in the current state vector, is the th eigenvalue of the th historical strategy sample, is the resolution coefficient, set to 0.5. According to all values, select the top K strategy samples with the highest correlation. Weighted voting is performed on the above K associated strategy samples according to their grey correlation coefficients to determine the photovoltaic energy control strategy label (such as strategy ID or control group number) that best suits the current structural state. According to the above selected strategy label, the system maps the corresponding control parameter set, including but not limited to the following: Photovoltaic output power upper limit: , used to limit the maximum output power in the current period; Energy storage system charge rate: , indicating the rate of the current charging stage; Discharge start minimum power threshold: , defining the lower limit of the energy storage that allows the discharge behavior to be triggered. The above output parameter combination forms a complete photovoltaic regulation data structure , denoted as .

[0030] S3. Obtain historical traffic flow data, and construct a digital twin based on the historical traffic flow data and a preset bridge-vehicle coupling model to obtain bridge-vehicle twin data; In one embodiment, historical traffic flow data processing and traffic semantic feature extraction are performed. The format of the collected original traffic flow data includes vehicle passing timestamp, driving speed , vehicle type , and the lane number to which it belongs. Based on this original data, the system performs statistics in a fixed time window (such as 5 minutes) and extracts the following traffic semantic feature indicators: Traffic flow density, indicating the number of vehicles passing through the monitoring area per unit time, is , where is the traffic flow density, is the number of vehicles recorded within a unit time , is the time window length. The traffic flow impact index , and the calculation method is , where is the traffic flow impact index, is the estimated mass of the vehicle, and is the corresponding speed. The proportion of heavy vehicles refers to the ratio of heavy vehicles in the current window. Heavy vehicles are classified and identified according to the vehicle type code . The above features constitute the traffic semantic vector of the current time period, which is used to reflect the spatial and dynamic effects of vehicle loads on the bridge deck. The system performs parameter recalibration of the bridge-vehicle coupling model. Based on the synchronous recording of historical bridge structure response data (such as vibration mode monitoring values) and traffic flow characteristics, an inverse fitting method is used to establish a functional mapping relationship between vehicle excitation (speed and mass) and structural response (frequency change). The system reversely corrects the key physical parameters in the bridge-vehicle coupling model according to this relationship, including but not limited to the equivalent stiffness coefficient, damping ratio parameter, etc. In the bridge-vehicle coupling model after parameter calibration, the system introduces historical traffic flow excitation to simulate the structural impact propagation path. The modeling methods include a continuous structure model constructed based on the finite element method (FEM), or a node-edge representation graph structure established based on the graph neural network (GNN). During the simulation process, the system tracks the propagation process of excitation between different structural regions of the bridge (such as the main girder, bridge tower, and bearing). The system outputs the following two key data results, including the activated mode sequence, which records the sequence of structural mode numbers that are activated or significantly responded in the simulation, and is used to depict the dynamic evolution trajectory of the response frequency; the impact path matrix is defined as the impact response intensity between any two nodes and

[0031] S4. Perform energy regulation response according to the photovoltaic regulation data and the bridge-vehicle twin data to obtain the photovoltaic regulation response data.

[0032] In one embodiment, the system performs the process of generating an energy regulation response strategy according to the previously generated photovoltaic regulation data and the bridge-vehicle twin data, for realizing the response control of structural safety and energy coordination. The system constructs a state trigger matrix. The photovoltaic regulation parameter data set ( , including the maximum output power, energy storage charge and discharge rate, minimum available charge state threshold, etc.) and the impact path matrix output from the bridge-vehicle twin ( , which represents the impact energy propagation intensity between different regions of the bridge structure), are fused to construct the structural - energy joint state matrix at the current moment . Based on the pre - set logical rules, the system defines the state trigger conditions. For example, if the impact intensity in the main span region is higher than the warning threshold (such as ), and at the same time the current state of charge (SoC) of the energy storage module is higher than the safety lower limit (such as SoC > 0.6), then the "early discharge" instruction is triggered; if the impact in the side span region is weak and the SoC is insufficient, the energy output is suspended. This process generates a set of formal conditional logics. The system performs response path generation and redundant path screening. For each satisfied state trigger rule, a corresponding regulation path is generated. This path is a structured instruction sequence including control objectives, regulation quantities, action timings, and target values, describing the response behavior that the system should take in the current state. The system forms a set of all response paths. The system performs redundant screening on the response path set, mainly based on the following criteria: if the estimated benefit value of a certain regulation path (such as impact mitigation rate or energy utilization efficiency) is lower than the set threshold, it is excluded; those paths with short response times, compact action links, and significant compensation effects on structural disturbances are preferentially retained. The system performs execution - level control and scheduling output for each instruction in the screened paths. Specifically, the system assigns execution labels to each control action within each path, such as P1 level: indicating a high - priority operation that needs to be executed immediately; P2 level: indicating a medium - to - low - priority operation that can be delayed. A standardized control instruction set is formed, including regulation objectives (such as energy storage units, inverters, power - limiting devices); action types (such as discharging, load reduction, pausing); execution priorities (such as P1, P2); delay times (if applicable); and the control instruction set is sent to the edge execution terminals of the photovoltaic controller and the energy storage module to perform the regulation task of coupling structural safety and energy efficiency, realizing intelligent response control of photovoltaic energy based on the bridge load state. The photovoltaic regulation data refers to the operating parameters, power status, voltage stability, etc. recorded or fed back when the photovoltaic system provides electrical energy for data acquisition devices (such as edge nodes or sensor arrays), which is used to assist in judging whether the system power supply is stable or whether it is necessary to switch to an alternative power supply.

[0033] Optionally, S1 includes: S11. Collect load response data through sensors arranged at key parts of the bridge; In one embodiment, the selection of sensor layout points is based on the finite element modal analysis results of the bridge structure. Priority is given to selecting key structural parts with larger modal participation factors (i.e., the influence degree of a certain element on a specific mode in the structural vibration response) as the layout areas. Specifically, areas sensitive to structural stress changes, such as the mid-span position of the main girder, bridge bearing nodes, and diaphragm connections, can be selected for sensor layout. The types of sensors used include acceleration sensors for collecting impact load and vibration response signals. The sampling frequency is set in the range of 200 to 500 Hertz (Hz). Strain sensors: including strain gauges (resistance strain gauge bridge configuration) or fiber Bragg grating (FBG) sensors, mainly used to monitor changes in the structural stress state and capture the stress change trends of components such as the main girder and web. Displacement sensors, such as laser displacement meters or resistive displacement meters, are used to measure the deflection deformation of long-span structures under load. The edge acquisition module performs data caching, and sensor data is uniformly aligned by timestamp into a structural response matrix.

[0034] S12. Construct a bionic dynamic expression based on the load response data to obtain bionic expression data; In one embodiment, the structural response signal is converted into a non-uniform biological time axis , to simulate the neural time series of high-frequency impact fast sampling and low-disturbance slow response; using the instantaneous acceleration energy (where is the instantaneous acceleration energy, is the acceleration value of the structural response signal) as the dynamic time base weight; the time axis is non-linearly weighted and integrated according to this energy, thereby forming a bionic time base, that is, the growth rate of bionic time changes with the strength of energy, and the stronger the energy, the faster the time advances, that is , where is the bionic time, is the original time, is the bionic time adjustment coefficient, with a value range of 0.5 to 2.0, is the instantaneous acceleration energy. The system performs pulse neural coding on the synchronized structural response data, initializing the membrane potential of each neuron channel, that is , where is the membrane potential at the current moment, is the membrane potential at the previous moment, is the membrane resistance, is the input electricity at the current moment, is the membrane potential decay time constant, with a value ranging from 20 to 30 milliseconds; in each time step, the system updates the membrane potential according to the input stimulus intensity (such as the current response amplitude), and performs natural decay (i.e., leakage) processing with time constant control; when the membrane potential exceeds the set firing threshold, the neuron generates a pulse event and resets the membrane potential. This neural coding process is independently executed for each sensor channel, generating multiple corresponding pulse train sequences, that is, the pulse output stream of each channel on the bionic time axis, for expressing the neural coding result of the structural response.

[0035] S13. Construct a modal energy spectrum for the bionic expression data to obtain modal energy spectrum data; In one embodiment, the system performs periodic structure reconstruction processing based on the bionic expression data obtained by pulse coding to identify and unify the main periodic features in the response sequence. The system identifies the main periodic pattern with significant repeatability by statistically analyzing the time intervals between pulses in the pulse sequence, determines the most significant periodic behavior in the structural response, corresponding to the excitation period of certain main modal frequencies. Using a moving average filter combined with the maximum cross-correlation method, the time axes of all pulse clusters are aligned so that each period segment has consistency in time sequence, reducing the error accumulation caused by phase perturbation. After completing the period alignment, the system performs frequency domain analysis on the pulse response data within each period segment. The specific operation is to perform time-frequency joint analysis on each period segment using continuous wavelet transform to extract the energy characteristics of the frequency distribution changing with time; extract the main frequency components (i.e., the center frequencies of frequency bands) and their corresponding amplitude spectra (i.e., the response amplitudes at different frequencies) from the wavelet transform results, and form the periodic modal frequency domain expression data. Based on the frequency domain energy data, the system calculates the integral of the modal energy of each frequency component to obtain the excitation intensity in the frequency dimension, that is , where is the frequency is the modal excitation energy at frequency is the start time of the period segment, is the end time of the period segment, is the wavelet amplitude spectrum of frequency f at time t, is the modal frequency variable, is the time variable. The system constructs a modal energy spectrum tensor, whose dimension is the structural node number modal frequency. Each element in the spectrum represents the excitation energy intensity of a certain structural node at a certain modal frequency, thus forming a two-dimensional energy spectrum matrix.

[0036] S14. Perform graph embedding convolution on the modal energy spectrum data to obtain structure-aware data; In one embodiment, each sensor installation position is defined as a node in the graph. Each node has its energy vector at the modal frequency as the feature input, and the dimension of this vector is equal to the number of modal frequencies, representing the modal energy intensity at each frequency component. The edges between the nodes in the graph include two types of connection relationships. One is the structural connection edge, which reflects the connection relationship of the physical components between the sensors. The other is the geometric adjacency edge, which is constructed based on spatial proximity. For example, when the distance between two nodes in the vertical direction (Z-axis direction) is less than 1 meter, they are regarded as an adjacent relationship and an edge connection is established. Each node is assigned its corresponding modal energy feature vector, and each dimension in the vector corresponds to the energy value at a modal frequency. After the graph structure is established, the system uses a graph convolutional neural network to update the node feature embedding, that is , where is the node feature matrix after the -th layer of graph convolution, is a non-linear activation function (such as ReLU, tanh, etc.), is the normalized graph adjacency matrix, is the node feature input matrix of the -th layer, is the trainable weight matrix of the -th layer. After multiple iterations of graph convolution update, the output node feature matrix represents the structure-aware feature data. This matrix integrates the energy distribution information at the modal frequency and the structure topology propagation characteristics.

[0037] S15. Identify the structural impact on the structure-aware data to obtain the structural impact data; In one embodiment, the average acceleration energy within a continuous time window is calculated as the energy discrimination index. That is, within each sliding time period, the squared acceleration values at all times within this period are statistically calculated and their mean value is obtained. If the energy value of this window is significantly higher than the mean value of the system's historical energy distribution (exceeding the mean value plus twice the standard deviation), then this time period is marked as a suspected disturbance segment and used as a candidate segment for impact identification. For each marked disturbance segment, the system extracts its short-time energy features, including the short-time root mean square value (RMS), which represents the average vibration intensity of the response signal of this segment; the main frequency offset, which analyzes whether there is an obvious offset of the energy peak in the frequency spectrum compared to the normal structural vibration state; and the energy mutation rate, which evaluates the energy growth rate per unit time. The system performs dynamic time warping (DTW) matching between the vibration response of the current disturbance segment and a pre-constructed typical impact template to calculate their similarity to measure the morphological consistency on the time axis. If the matching similarity is higher than a set threshold (such as 85%), then it is determined that the current disturbance segment is an actual structural impact event. For the identified impact event, the system uses a damped vibration model to perform function fitting on the acceleration response of this event, that is , where The acceleration response function of the structure at the moment of the impact event , is the initial amplitude of the impact response, is the natural exponential term, is the damping ratio, indicating the energy dissipation capacity of the structure, is the angular frequency of system vibration, is the time variable, is the cosine function, is the main frequency of damped vibration, is the initial phase. The structural impact data output by the system includes the time position of the impact event, the corresponding channel, the similarity score, and the impact physical parameters obtained by fitting.

[0038] S16. Attribute the load behavior based on the structural impact data to obtain the load perception feature data.

[0039] In one embodiment, the system establishes a set of load behavior attribution label systems, specifically including type A: periodic light impact, corresponding to regular slight disturbance behaviors such as daily traffic flow; type B: heavy load sudden impact, such as the short-term strong impact caused by the passing of large vehicles (such as trucks, construction equipment); type C: multi-vehicle synchronous resonance, referring to the structural resonance phenomenon caused by multiple vehicles passing through at the same frequency or specific mode; type D: abnormal structural vibration, which is the excitation of an atypical frequency band or long-term vibration persistence, indicating a structural problem and treated as an abnormal warning. The attribution analysis method uses a multi-layer perceptron model (obtained by training the multi-layer perceptron based on preset data) for multi-class classification training. The feature vectors input to the model are extracted from the structural impact data, mainly including impact amplitude: reflecting the initial intensity of load excitation; duration: the time length from the occurrence to the end of the impact event; modal frequency distribution characteristics: representing the main frequency and its energy distribution excited during the impact process; spatial diffusion rate: measuring the speed of impact response propagation between structural nodes. After the above features are standardized, they are input into the neural network model, and the behavior type recognition of different impact events is realized through model training. The system output results include the load behavior label corresponding to each impact event, that is, its attribution category (A / B / C / D); the embedded representation of the structural response mode of this event, that is, the structural perception feature vector, used to characterize its representation in dimensions such as mode, space, and energy.

[0040] Optionally, the bionic dynamic expression construction includes: Performing time-based synchronous conversion on the load response data to obtain response conversion data; In one embodiment, the structural response sequence with a fixed sampling interval is converted into a "non-uniform time base" expression with biological rhythm, simulating the reaction density of the nervous system to different intensity stimuli. The input signal is a multi-channel structural response data matrix For each channel, calculate the local activation intensity per unit time (e.g., the instantaneous energy of the acceleration signal): , where is the channel at time local activation energy, is the current time, is the energy calculation window length, is the channel at time acceleration signal, is the integration variable, representing the time element; according to the relative ratio between the current channel energy and the maximum energy, dynamically adjust the advancement rate of the time axis, that is, when the local energy is high, increase the sampling density for this time period; otherwise, decrease the sampling density to achieve compression processing of the time axis. A regulation coefficient is introduced in this compression rule to control the non - linear degree of time compression. This coefficient can be adjusted according to the response characteristics, and the value range is 0.5 to 2.0, that is , where is the integration variable, representing the time element, is the current time, is the biological response regulation coefficient, where is the biological response regulation coefficient (such as 0.5 - 2.0), is the channel at time local activation energy, is the maximum activation energy within the current time window, is the maximum energy of the current window. Through the above - mentioned time - base conversion mechanism, map the original response data sequence with equally - spaced sampling into a new sequence with non - linear time intervals.

[0041] Perform neural - like excitation on the response - converted data to obtain pulse - sequence data; In one embodiment, for each response channel, the system initializes its corresponding membrane potential to zero. Subsequently, at each moment, the membrane potential is iteratively updated, and its update mechanism consists of the following two elements, including an input current term, which is determined by the normalized response intensity data of the current channel; a decay term, which reflects the natural decay behavior of the neuron membrane potential without continuous input, and the decay rate is determined by the membrane constant , and its typical value range is 20 to 30 milliseconds. After the membrane potential update is completed, the system determines whether it exceeds the preset threshold . If the membrane potential at the current moment exceeds this threshold, the system immediately triggers a pulse signal (i.e., regarded as neuron firing), and resets the current membrane potential to zero. This process is repeated until all time series are traversed, that is , where is the current moment under the channel membrane potential, for the previous moment lower channel membrane potential, is the time step, for the current moment lower channel input current (normalized response intensity), is the membrane constant, representing the time scale of potential decay. The system output is a pulse event matrix composed of the excitation states of each channel, where the element at each time point represents whether the corresponding neuron triggers an action potential. This pulse sequence encodes continuous structural response information in a sparse and discrete form.

[0042] Perform neural synapse structure mapping on the pulse sequence data to obtain synaptic activation weight map data; In one embodiment, the system constructs a sensing node graph structure , where each node corresponds to a response sensing channel (such as an accelerometer, strain gauge, etc.) deployed in the bridge structure. The edges in the graph are used to represent that there is a potential synaptic connection relationship between node and . The connection rules are mainly defined based on the following two types of prior knowledge, including physical proximity. If two sensors are adjacent in the deployment location in the bridge structure or in the same component area, they are considered to have the possibility of connection; modal coupling. If historical modal analysis shows that there is a resonance behavior between the two channels within the main modal frequency range, the connection probability is enhanced. The initial synaptic weight can be initialized by the structural connection strength, sensor layout density, or a unified set constant. The system calculates the time difference between the action potentials of each pair of connected nodes (i, j), where and are the most recent pulse activation times of nodes and respectively. According to the positive and negative relationship of the time difference, the following two weight adjustment strategies are executed, including when triggers before (i.e., ): The synaptic weight is enhanced exponentially, and the update amount is , triggers before (i.e., ): The synaptic weight is weakened exponentially, and the update amount is , where is the synaptic activation weight between nodes and , is the positive - direction synaptic enhancement learning rate parameter, with a value of 0.01, is the base of the natural logarithm, is the time difference between two spike - discharges, is the time constant (time window) of positive - direction synaptic enhancement, with a value of 30ms, is the negative - direction synaptic weakening learning rate parameter, with a value of 0.012, is the time constant (time window) of negative - direction synaptic weakening, with a value of 50ms. After all spike - time comparisons and weight adjustments are completed, the system outputs the synaptic activation weight map matrix , where each element represents the co - activation intensity between the corresponding channels under temporal - sequence drive.

[0043] According to the synaptic activation weight map data, bionic dynamic expression feature extraction is performed to obtain bionic expression data.

[0044] In one embodiment, the system constructs the input graph structure of the graph neural network model based on the synaptic activation weight matrix , where each node corresponds to a sensor channel, and the weight of each edge represents the co - activation intensity between node and under temporal - sequence spike drive. For each node , the system maps its corresponding impulse response information to a one - dimensional or multi - dimensional initial feature vector , which includes feature dimensions such as spike frequency, time - interval statistical value, and the amplitude of the most recent spike. The system uses the graph attention mechanism to perform weighted fusion on the association relationships between nodes, and the update method is as , where is the feature representation of node in the th layer of the graph neural network, represents the feature representation of node in the th layer of the graph neural network; is the attention weight of node to node , and its calculation is based on the synaptic weight and neighborhood relationship; W is a shared linear transformation weight matrix; represents a non - linear activation function (such as ReLU or tanh); represents the neighbor set of node . After completing the multi - layer graph neural network feature aggregation, the system outputs the embedded feature vector for each node i, where is the number of layers of the graph neural network. All node vectors are concatenated in the order of their numbers to form a structural bionic expression data structure: .

[0045] Optionally, the construction of the modal energy spectrum includes: Performing periodic structure reconstruction on the bionic expression data to obtain periodic response data; In one embodiment, the input is a sequence of bionic expression data, denoted as a bionic expression vector, which contains pulse activation information from multiple channels. The data of each channel can be represented as a pulse emission vector or a time series expression sequence generated by a neural embedding model. For each sensor channel, the system extracts the time interval sequence between its pulse events. The system calculates the time interval between every two consecutive pulses to form an activation interval sequence. The system calculates the autocorrelation function of this sequence within a specified sliding time window , where is the value of the autocorrelation function at a delay , is the total time length of the bionic expression sequence, is the time delay (the lag parameter of autocorrelation), is the time step index, is the pulse activation value at time t (an item in the activation interval sequence), is the time pulse activation value. In a signal with a repetitive structure, the autocorrelation function will have local maxima at certain time delays. The system identifies the position of the strongest main peak in the autocorrelation function, and this time delay value represents the dominant period length of the channel. The dominant period represents the period of the most significant repetitive pulse pattern within the channel. After obtaining the dominant period length, the system performs reconstruction processing on the pulse activation sequence within the channel. The sequence is folded according to the dominant period length to construct a two-dimensional periodic matrix. Each row of this matrix corresponds to the pulse activation pattern of a period segment, and each column corresponds to the time points within the period. The original one-dimensional time series is reorganized into a periodic two-dimensional structure, which can intuitively present the periodically repetitive pulse pattern and strengthen the modal consistency and rhythm structure in the signal. The output is the periodic response data for each channel.

[0046] Performing frequency domain analysis on the periodic response data to obtain modal frequency domain data; In one embodiment, a frequency domain transformation operation is performed on each periodic segment in the periodic response data. The system performs a fast Fourier transform on each row in the periodic matrix, that is, the time series segment of each independent period, to obtain the frequency components corresponding to that period segment. The output includes an amplitude spectrum, representing the energy intensity of each frequency component in the current period segment; and a phase spectrum, representing the displacement information of each frequency component relative to time. The system integrates the spectral data extracted from all periodic segments to form a spectral set, and performs frequency pattern clustering on it. An unsupervised clustering algorithm (such as K-Means) is used to cluster the frequency peaks that appear in the amplitude spectrum to extract the frequently occurring dominant modal frequencies; or a frequency alignment technique is adopted to statistically analyze the energy peak positions of multiple period segments in the spectrum to identify the consistent peak positions in the frequency dimension as the candidate set of modal frequencies. A set of representative main modal frequency sets is obtained, and each frequency represents the typical response of the structure in a specific mode. The modal frequency domain data output by the system includes the modal frequency value, that is, the identified main frequency; the amplitude corresponding to the frequency, representing the energy intensity at that modal frequency; and the phase corresponding to the frequency, representing the phase difference of that frequency in different period segments.

[0047] Modal energy spectrum extraction is performed on the modal frequency domain data to obtain modal energy spectrum data; In one embodiment, for each main modal frequency , the local energy spectrum is calculated within the adjacent bandwidth , where is the local energy spectrum of the -th order mode, is the -th order main modal frequency, is the frequency bandwidth (for example, 0.1 Hz), is the amplitude-frequency response function, is the frequency variable. If there are multiple sensing channels, weighted average or maximum aggregation is performed, where is the channel aggregated energy spectrum (average or maximum) of the -th order mode, is the number of sensing channels, is the channel index, is the -th order mode on the -th channel energy spectrum, and the full-bridge average excitation energy spectrum of each order mode is obtained. The output forms modal energy spectrum data , where is the modal energy spectrum data, is the -th order modal frequency, is the -th order aggregated energy spectrum of the mode, is the modal index, with values of 1, 2…k.

[0048] Construct a modal excitation map based on the modal energy spectrum data to obtain modal energy map data.

[0049] In one embodiment, the system constructs a two-dimensional modal excitation map based on the aforementioned modal energy spectrum data to characterize the modal response distribution characteristics of the structure in the spatial dimension and the frequency dimension. The X-axis is the structural sensing channel or position index i; the Y-axis is the main modal frequency index ; the value range is the modal energy , that is, the response intensity of the th channel to the th frequency; the system traverses all sensing channel nodes in the structure and sequentially extracts the energy values of each node at all dominant modal frequencies. Specifically, for each node-frequency combination (the i-th node, the k-th frequency), its corresponding energy value is read from the frequency-domain analysis result. All energy values are arranged in the order of nodes and frequencies to form a two-dimensional energy matrix, the number of rows of which is equal to the number of sensor nodes, and the number of columns is equal to the number of modal frequencies. Each element of the matrix represents the response intensity under the corresponding node and modal frequency combination. The output is the modal energy map data matrix. This matrix can be represented as a two-dimensional tensor, and its dimension is "number of nodes number of modal frequencies". Each row represents the modal frequency energy distribution of a node, and each column represents the response intensity of a modal frequency at different nodes, , where is the modal energy map matrix, is the modal energy of the th node at the th modal frequency, is the node number, with values of 1, 2,..n, is the modal frequency number, with values of 1, 2,..k.

[0050] Optionally, the structure impact identification includes: Screen the structure perception data for structure perturbations to obtain structure perturbation data; In one embodiment, the structure perception feature data is defined as a multi-channel time series matrix , where each column corresponds to the perception feature vector of a structure node at time , such as modal frequency drift, instantaneous acceleration value, synaptic activation density, etc. The system calculates the short-time mean square energy of the structure perception value of each node channel using a sliding time window (window length W, range 1–2 seconds) as follows: , where represents the perception value of the current channel at time ; represents the time The energy level within the sliding window starting from is the sliding window length, and is the time index within the sliding window. Calculate the RMS energy mean of each channel over the entire sequence and the standard deviation , and set the perturbation judgment threshold accordingly , where is the perturbation judgment threshold, is the adjustment factor, and the empirical value range is from 1.5 to 3.0. If a certain channel satisfies in multiple consecutive time windows, then the corresponding time interval is determined as the structural perturbation segment. The system supports setting the minimum number of continuous windows (such as 3 sliding windows) to filter out occasional spike noises. The system outputs a set of structural perturbation data segments, defined as , where respectively represent the start and end times of the th perturbation segment; is the total number of identified perturbation segments.

[0051] Conduct vibration energy analysis on the structural perturbation data to obtain vibration energy data; In one embodiment, the system input is a dataset of structural perturbation time periods . For the structural response sequence within each perturbation segment, the system calculates the total vibration energy within this time period, defined as , where is the total vibration energy within the perturbation segment, is the time index, is the start time of the perturbation segment, is the end time of the perturbation segment, represents the instantaneous response value (such as acceleration, strain, etc.) of the perturbation channel at time . Identify the maximum term of the squared instantaneous response value in the perturbation segment as the peak energy index: , where is the peak energy index of the perturbation segment, is the maximum operator, is the time index, is the start time of the perturbation segment, is the end time of the perturbation segment, is the instantaneous response value (such as acceleration or strain) of the structure at time . The system performs frequency domain analysis processing on the perturbation response sequence to extract the main excitation frequency band , where is the frequency interval where the energy of this segment is most concentratedly distributed, is the frequency corresponding to the maximum frequency domain response, Represents the frequency-domain response. The system calculates the peak-to-average power ratio , defined as , where represents the duration of the disturbance segment, is the average energy level of this segment. The system outputs a vibration energy feature structure corresponding to each disturbance segment, denoted as , characterizing information such as the energy amplitude, energy concentration, main frequency position, and time boundary of the th disturbance segment.

[0052] Match impact events based on the vibration energy data to obtain impact event data; In one embodiment, the system presets or learns from historical data to define multiple typical impact event templates, including Type A: vehicle flow impact type: characterized by a short disturbance duration, a high peak-to-average power ratio, and a relatively broad frequency distribution; Type B: heavy load concentration type: characterized by a medium disturbance duration, a relatively high total energy, and a low main frequency; Type C: modal resonance type: characterized by a relatively long disturbance duration and energy concentrated within a single main frequency range. Each template includes reference feature indicators in multiple dimensions, such as , , , , the disturbance duration, etc. The system uses a feature distance metric method to match and evaluate the vibration energy data with the impact template , that is , where is the feature distance between the disturbance sample and the template, represents the th feature dimension; is the weight coefficient of this dimension, set according to feature sensitivity; and represent the values of the disturbance sample and the template in this dimension respectively. The distance metric method can be Euclidean distance, cosine similarity, or the normalized weighted Manhattan distance. The system calculates the matching distance between the current disturbance segment and all impact templates, and extracts the minimum distance value: , if the determination condition is satisfied, where is the preset similarity threshold, then it is considered that the current disturbance belongs to a certain type of impact event, and its type is labeled as the category of the corresponding template . Otherwise, this disturbance segment is regarded as an atypical disturbance and is not temporarily included in the impact event set. For the disturbance segments that match successfully, the system outputs an impact event annotation structure, denoted as: , where represents the identified impact event category (such as Type A / B / C); Indicates the start and end times of the impact event; and are the main energy features; score is the similarity score or distance value.

[0053] Perform impulse response fitting based on the impact event data to obtain the structural impact data.

[0054] In one embodiment, the system performs impulse response fitting processing based on the identified impact event data, , where is the structural response function, is the initial response amplitude, is the damping ratio, is the undamped circular frequency, is the time variable, is the damped natural frequency, is the initial phase. For each impact event segment (defined by the start and end times ), the system constructs the least squares fitting objective function based on the actual observed response signal , is the least squares fitting loss function, is the time variable, is the observed response signal, is the fitting response function; output the impact description data , where is the initial amplitude obtained by fitting, is the damping ratio, reflecting the energy dissipation characteristics of the structure, is the damped natural frequency, indicating the main response frequency of the structure, is the duration of the impact event, is the total vibration energy associated with the impact event, is the corresponding sensor number or spatial position coordinate for locating the impact occurrence position.

[0055] Optionally, S2 includes: S21. Select the regulation factor for the load perception feature data to obtain the regulation factor data; In one embodiment, the system input is a set of load perception feature sets, and each feature vector represents the structural response characteristics corresponding to a load event. This set contains multiple feature dimensions, and each feature item includes but is not limited to modal amplitude, which represents the energy excitation intensity at each modal frequency of the structure under the action of the load; impact frequency, which is the dominant frequency at which the impact event occurs; duration, which is the duration of the action of the impact event; structural node centrality, which is the degree of aggregation of response energy in the node space distribution; vibration common mode ratio, which is an index to measure the response consistency between channels and represents the resonance intensity. The system performs regulatory factor screening, calculates the variation degree of each feature in the sample set (such as coefficient of variation or information entropy); evaluates the correlation between each feature and the target variable, and at the same time punishes the redundancy with other features to select a feature subset that is both representative and highly complementary; based on the aforementioned weighted summation, that is where is the score, is the weight term of the coefficient of variation, with a value of 0.7, is the coefficient of variation, is the weight term of the correlation, with a value of 0.3, is the correlation, to assign scores to all features, sort them according to the scores, and retain the top 3 to 5 features with the highest scores as representative regulatory factors. The system output is a set of regulatory factor feature sets, that is, several most representative feature items selected from the original feature set.

[0056] S22. Construct the load state space according to the regulatory factor data to obtain the load state space data; In one embodiment, the system is based on a time series and organizes and constructs the regulatory factors within a continuous time window. Specifically, at each time point, the system extracts the regulatory factor sequence at that moment and within the previous T - 1 time steps to form a feature vector in the time dimension. Each vector contains the continuous observed values of several currently selected key regulatory factors during this time period to form the state input variables. The system first normalizes the original time series features using the Z-score standardization method, that is, converting each feature value into a standard normal distribution form with a mean of zero and a standard deviation of one. The system uses the principal component analysis method to reduce the dimension of the normalized feature vector. By sorting the cumulative contribution rates of the eigenvalues, the first several principal components with a cumulative contribution rate greater than 90% are selected as the compressed state space expression vectors. The system combines the values of the above compressed state vectors at different time points to form a complete set of load state space trajectories. This trajectory set is used to reflect the evolution path of the load state of the structure in different time periods, specifically represented as a set of compressed state vectors at multiple continuous time points. Each state vector represents the state description of the structure under the current load condition, and the entire trajectory reflects the dynamic behavior of the load response of the structure during the observation period.

[0057] S23. Perform grey decision-making based on the load state space data to obtain the photovoltaic strategy matching data; In one embodiment, the system pre-constructs a corresponding set of a group of historical load states and photovoltaic strategies, denoted as the historical strategy library. Each record includes a historical state trajectory vector for describing the structural load state at a certain moment; a photovoltaic regulation strategy corresponding to this state, including but not limited to power output level setting, charge and discharge switching strategy of the energy storage system, load transfer scheduling time point, etc. Calculate the grey correlation degree: , where is the grey correlation degree between the th historical record and the current state, is the total number of regulation factors, is the regulation factor index, is the minimum value in all samples, is the normalized value of the th load state factor at the current moment, is the normalized value of the th th state factor in the th historical record,

[0058] S24. Perform photovoltaic regulation network output based on the photovoltaic strategy matching data to obtain the photovoltaic regulation data.

[0059] In one embodiment, the system constructs a feedforward neural network model as the photovoltaic regulation network. The model structure is as follows: an input layer that receives the currently matched photovoltaic policy data as the neural network input vector; a hidden layer with three layers of neurons, using the rectified linear unit (ReLU) as the activation function to enhance the model's ability to express non-linear regulation relationships; an output layer that selects different output forms according to the type of regulation target: if the output is a continuous value (such as power ratio, charge-discharge ratio), the Sigmoid activation function is used; if the output is a discrete classification signal (such as a mode switching signal), the Softmax activation function is used for multi-class or binary classification output. The output of the model includes the following three key photovoltaic regulation parameters: the output power ratio, which represents the ratio of the power that should be output by the photovoltaic system at the current moment to the rated power; the energy storage mode switching signal, which is a binary control signal used to indicate the switching of the working state of the energy storage system, such as switching from the charging state to the discharging state or entering the standby mode; the charge-discharge ratio, which represents the power rate of the current energy storage device during the charging or discharging process. The neural network model is trained through supervised learning, and the training data comes from historical policy execution records and their corresponding regulation feedback data. The input is the historical photovoltaic policy; the output is the actual regulation parameters at the corresponding moment; the loss function uses the mean squared error function to measure the deviation between the model output and the historical feedback, serving as the goal for network training optimization. The output is a set of photovoltaic regulation parameter results, including the power output ratio, energy storage control instructions, and charge-discharge ratio. This data can be directly sent to the execution layer of the photovoltaic system as the control input for driving current scheduling, voltage regulation, and energy storage switching, realizing the real-time intelligent regulation of the photovoltaic energy system.

[0060] Optionally, the grey decision includes: Performing modal change trend processing on the load state space data to obtain the load change space data; In one embodiment, for the input state data structure, the load state space data is: , where each represents the standardized state value of the i-th regulation factor, and the value of is from 1 to k. For the modal change trend processing, calculate the trend increment within its local time window for each factor: , where is the trend increment of the -th regulation factor, is the standardized state value of the i-th regulation factor, is the state value of the -th factor time steps ago, and is the time step (such as min); the trend direction can be encoded into three values: , where is the trend direction encoding result of the th factor, is the trend increment of the th regulatory factor, is the trend discrimination threshold, set to 0.05 - 0.1. Combine the trend direction encoding results of all regulatory factors to form the modal trend vector at the current moment.

[0061] Calculate the grey relational grade based on the load change spatial data and the preset historical regulation library to obtain the grey relational grade data; In one embodiment, construct the historical regulation library structure, the historical library , where is the th historical modal change vector; is the associated photovoltaic strategy tuple (such as power distribution, energy storage switching, etc.). Calculate the grey relational grade. For the current and each historical calculate the grey absolute relational grade: , where is the grey absolute relational grade between the th historical modal vector and the current modal vector, is the dimension of the regulatory factor, is the regulatory factor index, is the minimum absolute difference among all factor pairs, is the change amount of the th factor in the current modal change vector, is the th historical modal change vector's th factor's change amount, is the grey decision resolution coefficient (with a value of 0.5). Output the grey relational grade list.

[0062] Perform strategy matching based on the grey relational grade data to obtain the photovoltaic strategy matching data.

[0063] In one embodiment, set the matching strategy as if , where is the maximum value in the grey relational grade, is the grey relational grade set at the current moment, is the grey matching determination threshold, then directly select the corresponding strategy : , where is the maximum grey relational grade value, is the strategy corresponding to the maximum relational grade, is the index corresponding to the maximum grey relational grade, For the sake of taking the maximum index operation, For the grey relational degree between the i-th historical strategy and the current state, otherwise perform policy weighted fusion , where is the photovoltaic policy obtained by matching under the current load trend, is the historical policy index, is the number of historical policies, is the weight after grey relational degree normalization, For the i-th historical photovoltaic regulation policy, For the grey relational degree between the i-th historical strategy and the current state, is the summation index variable in the denominator of the normalization weight, is the grey relational degree used for the normalization denominator; each policy can be expressed as , is the output power adjustment coefficient; is the energy storage current ratio control factor; is the energy storage mode switch signal (0 for off, 1 for parallel). The output result obtains the optimal photovoltaic policy matching data under the current load trend .

[0064] Optionally, S3 includes: S31. Obtain historical traffic flow data; In one embodiment, the system relies on traffic perception devices preset on the bridge structure to collect real-time vehicle passing data. The perception devices used may include, but are not limited to, millimeter-wave radars for high-precision monitoring of vehicle speed and size; video recognition systems that use computer vision technology to identify traffic flow and vehicle types; and geomagnetic induction devices, which are embedded sensors for detecting magnetic field changes when vehicles pass through to determine passing time and position. The recorded information of each passing vehicle constitutes a data item, including passing time, which represents the timestamp when the vehicle passes through the bridge monitoring area; driving speed, which is the average speed at which the vehicle moves on the bridge per unit time; vehicle length, which is an estimate of the vehicle's length on the bridge for identifying vehicle types or body load distributions; single-axle axle weight, which is the load weight corresponding to a certain axle of the vehicle for calculating the total weight and the impact of local bridge loads; and lane position number, which records the specific lane position number where the vehicle is passing, supporting lane distribution statistics. The continuous collection period is set to be not less than 30 days. The data is sliced and sorted at a time granularity of 5 minutes, that is, a data packet for each 5-minute time period is formed.

[0065] S32. Extract traffic semantic features based on the historical vehicle data to obtain traffic semantic feature data; In one embodiment, the system performs structured processing on the traffic flow information within each time segment to construct a traffic semantic feature vector. Each feature vector includes traffic flow density, which represents the number of vehicles passing through the monitored section of the bridge per unit time, with the unit of "vehicles per minute"; average vehicle speed, which is the arithmetic mean of the driving speeds of all vehicles during the current time period and reflects the traffic flow operation efficiency; speed standard deviation, which statistically measures the degree of dispersion of vehicle speeds and serves as an indicator of traffic flow stability or volatility; proportion of large vehicles, which refers to the proportion of vehicles with a body length greater than 6 meters in the total number of passing vehicles during the current time period and is used to evaluate the heavy traffic flow pressure; and two-vehicle parallel probability, which is based on the simultaneous passing situation of multiple lanes and statistically calculates the proportion of parallel vehicles (i.e., different driving trajectories overlap and pass side by side) within the same time period to reflect congestion and concurrency.

[0066] S33. Perform parameter back-calibration based on the traffic semantic feature data and the preset bridge-vehicle coupling model to obtain an updated bridge-vehicle model; In one embodiment, based on the preset bridge-vehicle coupling dynamics model, the system completes the reverse calibration of the model parameters through the deviation between the traffic semantic feature data and the bridge structure response data. The system sets the initial bridge-vehicle coupling model as a dynamic response system. This model takes the traffic semantic feature vector as the input and the bridge structure response quantity (such as vibration amplitude, acceleration, or deflection, etc.) as the output. The model internally contains a set of parameters to be optimized, including but not limited to the modal stiffness coefficient of the bridge; damping factor or attenuation coefficient; and weighting factors for the influence of different types of traffic loads on the structure. The input data is the constructed traffic semantic feature sequence; the target output data is the structure response data observed at the corresponding moment, which is sourced from devices such as vibration sensors and accelerometers deployed on the bridge. The system performs reverse solution of the model parameters through an optimization method to make the model output as close as possible to the actual observed results. The back-calibration objective is to find a set of optimal parameters to minimize the sum of the squares of the errors between the model predicted response and the actual response, that is , where is the set of optimal bridge-vehicle coupling model parameters, is the parameter search objective for minimizing the loss function, is the time index (representing each time point in the time series), is the parameter when the model's predicted structural response to the traffic semantic feature , is the actually observed structural response value (such as vibration, acceleration, deflection, etc.). Methods such as the least squares method; genetic algorithm; and Bayesian optimization method can be used. After completing the parameter back-calibration and substitution, the system outputs an updated bridge-vehicle coupling model, whose parameters have been adjusted from the original set values to an optimized version that highly matches the actual data.

[0067] S34. Capture the impact propagation path of the updated car model to obtain impact propagation path data; In one embodiment, in the updated car-bridge coupling model, the system defines the vehicle impact load as an external input excitation. This excitation is modeled as a spatially distributed load where the instantaneous axle weights of multiple vehicles act on the surface of the bridge structure at different time points. , where is the total impact load on the bridge structure at time , is the index of the th vehicle, is the total number of vehicles, is the th vehicle's axle weight (impact excitation amplitude) at time , is the unit impulse function, representing the position matching degree between the vehicle position and the bridge structure position , is the th vehicle's traveling position at time , is the position of the response observation point on the bridge structure. The axle weight of each vehicle at any time is input as a function of time; the position of each load point corresponds to the bridge structure coordinates; the local load indicates that the load is concentratedly applied at specific position nodes, thus forming a time-space coupled impact input in the model. Based on the modal response of the bridge structure, the system constructs a response propagation matrix. This matrix describes the response energy propagation relationship between different nodes within the structure, where each matrix element represents the probability or weight of a certain modal response energy transferring from the source node to the target node; it can be weighted and constructed using structure mode participation factors, modal energy distribution, and adjacency topological relationships. Based on the propagation matrix, the system uses graph structure propagation to trace the structure response path, including PageRank, which measures the propagation importance and path preference between structure nodes; extended breadth-first search (BFS), which is used for layer-by-layer backtracking of multi-level propagation paths; it can combine the adjacency matrix of the structure and modal response characteristics to determine the response path caused by each impact. Generate one or more impact propagation path chains within the structure, and each path chain represents a directed path for the response signal to propagate from the initial excitation point along the structure to other key nodes. The impact propagation path data output by the system is represented in the form of a path chain set, and each path chain consists of several structure nodes in sequence, representing the propagation process from the initial impact application point to the distal response node. For example, a typical path can be expressed as "node propagates to node , and then to node , and so on", forming a complete path chain.

[0068] S35. Perform an evolutionary simulation based on the impact propagation path data to obtain the data of the bridge vehicle twin.

[0069] In one embodiment, the previously calibrated bridge vehicle coupling model is used as the basic model of system dynamics, and the impact propagation path data is used as an external disturbance input to initialize the boundary conditions and initial state of the simulation system. The impact path data may include physical characteristic parameters such as the starting position of the impact event, propagation direction, propagation speed, and energy intensity distribution. The simulation system is constructed as a time-stepping structural evolution sequence, denoted as a set of continuous time segments. Each time step represents the state field response of the bridge structure at that moment. The simulation results of each time step include structural state field data, including stress distribution, strain response, and displacement changes of key nodes; an energy dissipation map, which characterizes the internal energy transfer and dissipation process of the structure under impact or vehicle load excitation; and a fatigue cumulative index distribution, which is used to infer the development trend and risk level of potential fatigue areas under the cumulative action of impacts. The simulation framework allows setting various external control strategies as scenario inputs, such as different energy storage system regulation modes and different photovoltaic output power levels. The system compares the response trajectory differences under different control schemes. After each simulation task is completed, the system outputs multiple performance evaluation indicators, including but not limited to the response delay time, the shortest propagation time delay from the impact occurrence to the structural response of the system; the stress / strain peak value, the maximum load response of the key component during the simulation period; the structural fluctuation degree, which refers to the degree of variation of the state quantity during the time evolution process; and the change trend of fatigue hot spots, which is used to predict the drift trajectory of the long-term cumulative risk position.

[0070] Optionally, S4 includes: S41. Construct a state trigger matrix based on the photovoltaic regulation data and the data of the bridge vehicle twin to obtain state trigger data; In one embodiment, define the state trigger matrix , where is the number of enumerations of the bridge modal state, denoted as , and each represents a different bridge condition, such as the main span modal frequency drift, the impact energy concentration state, etc.; is the number of the photovoltaic strategy action set, denoted as , and each corresponds to a set of regulation actions (such as power increase, energy storage mode switching, etc.). The elements in this matrix are defined as: , by setting a series of composite trigger logic rules, the bridge condition and the regulation strategy are coupled. For example, if the main frequency drift is detected in a certain bridge modal state, is the critical value of modal frequency drift, and if the current wind speed is greater than 3 m / s, it is determined that there is a potential risk of wind-induced vibration. At this time, strategy mode A (efficient energy feedback) should be triggered; if the predicted traffic impact energy , and the state of charge SOC of the current energy storage system is lower than 40%, then strategy mode B (low-power buffered power supply) is triggered; if the micro-vibration response of a certain span of the bridge is concentrated in the low-frequency band and the amplitude exceeds a certain threshold, and the current photovoltaic module is in the high-irradiance power range, the system activates mode C (constant-power peak shaving + energy storage switching), etc. The system extracts all non-zero element positions in the state trigger matrix , combines the corresponding bridge state with the trigger strategy , and generates a state trigger data set in the form of . Each record indicates that under the state , the strategy should be activated, and the system generates a control instruction stream accordingly.

[0071] S42. Perform logistic regression processing on the state trigger data to obtain response condition data; In one embodiment, the system uses logistic regression to discriminate the state vector , and its output is the probability of entering the response state, , where is the response state probability, is the natural exponential function, is the state combination feature vector, is the discriminant variable, 1 indicates that it can enter the response path, and 0 indicates no response; is the logistic regression weight vector, which can be trained through the state label samples of "response success" and "response failure" in the historical control samples; is the currently input state combination feature vector. The system compares the probability value output by the model with the preset discrimination threshold (for example ). The specific rule is that if , it is considered that this state combination has a sufficient probability of successful control and can enter the response path generation process; otherwise, it is considered that this state does not have a reliable response basis and is excluded. All state vector combinations x that meet the response condition determination threshold are recorded to form the response condition data set , where is the response condition data set, is a single state vector, is the state trigger data set, and each entry represents the probability condition that the current system state meets the execution of the control instruction, which is a key screening node in the control logic chain.

[0072] S43. Generate a response path based on the response condition data to obtain response path data; In one embodiment, the response path refers to an ordered operation chain that starts from the current system state, sequentially passes through multiple logical decision nodes (i.e., control gating conditions), and reaches the photovoltaic execution command. The path can be expressed as , where is the response path, represents the current load state (including the degree of structural disturbance, energy storage state, environmental conditions, etc.); represents the intermediate control gating node, such as the operation determination conditions for fine-tuning the charging voltage, correcting the power factor, switching the inversion mode, and whether grid connection is allowed, with a value range of 1..k; is the output photovoltaic execution control instruction, such as switching to the maximum power point tracking (MPPT), starting power curtailment, starting and stopping the energy storage unit discharge, etc. The system uses a finite state machine (FSM) model or a control strategy diagram to construct the response path space: The FSM model formalizes each state and transfer condition into a state transition diagram and sets legal gating transfer rules; the strategy diagram model maps all response conditions to the nodes and edges in the diagram, where the nodes represent the system states and the edges represent the control actions, forming a traversable graph structure. On the premise of meeting the current response conditions, the system performs path search in the FSM or the strategy diagram, and can adopt depth-first (DFS), breadth-first (BFS), or heuristic search algorithms to find one or more effective paths that meet the response conditions. At the same time, the following preferred metrics are introduced to sort the candidate paths, such as the minimum response delay, the sum of the times required for node transitions in the path; the maximum disturbance absorption effect, the compensation efficiency of the control action on the structural impact; the optimal power utilization efficiency, the policy benefit brought by unit energy control. Select the path that meets the optimal criterion as the output path. The system represents the generated response path as a structured data set: , each path includes a path number, a starting state, a sequence of each gating node, a final execution instruction, and associated metrics (such as time consumption, expected control effect, etc.).

[0073] S44. Perform redundant screening on the response path data to obtain path screening data; In one embodiment, for the situation where there are operation conflicts for the same control actuator (such as energy storage access module, inverter switch, etc.) in different paths, the system performs conflict identification and elimination. For example, if an actuator is set to "start" in path A and "off" in path B, then compare the path priorities of the two (which can be based on indicators such as response timeliness, prediction success rate, historical return effect, etc.); only retain the one with the higher priority and eliminate the conflicting path. For all the remaining candidate paths, calculate the output power benefit ratio corresponding to the unit input energy, that is, the path energy efficiency index: , where is the expected regulated output energy corresponding to this path (such as photovoltaic grid-connected feedback power, energy storage release amount, etc.), is the energy consumed required for regulation corresponding to this path (such as system energy required for control communication, state conversion). The system sets an energy efficiency threshold (for example, 0.5) and eliminates the low-efficiency paths below this value. For the situation where there are duplicate sub-path segments in the path set (that is, a certain continuous gated logic operation is shared among multiple paths), the system performs a strategy merging operation on this segment, including identifying the duplicate sub-path segment; merging it into a unified strategy node structure; introducing a "shared node" or "reference node" mechanism in the path description structure to reduce redundant strategy records.

[0074] S45. Perform execution level control on the path screening data to obtain photovoltaic regulation response data.

[0075] In one embodiment, the system converts each response path into a set of actual photovoltaic control instructions that can be issued according to the regulation nodes and target execution strategies involved. Each path is mapped to the corresponding regulation action , which includes the following instruction parameter fields: tracking rate adjustment command, such as "adjust MPPT (maximum power point tracking) rate = δ"; energy storage state scheduling parameter, such as "set SOC (state of charge of energy storage) discharge threshold = 45%"; network scheduling behavior, such as "delay grid connection switching time = 10 seconds". Each generated control instruction will be prioritized in combination with the current state scenario. If the traffic impact prediction index exceeds the set threshold , then the corresponding instruction set is marked as "emergency response level"; if the remaining energy storage capacity in the system is greater than 80% and the bridge vibration fluctuation is in a stable range, it is marked as "conventional scheduling level"; if it is at the critical edge of the system state (such as frequency perturbation but insufficient energy storage), it can be set to "medium priority" for pre-debugging or pre-startup. The system assembles all the mapped control action instructions, their corresponding execution levels, and recommended durations into the form of a response data triple to construct the final photovoltaic regulation response structure: , where represents the controls behaviors such as power regulation, switching commands, etc.; represents the execution priority, and the values include "high", "medium", "emergency", etc.; represents the recommended duration of the instruction, which is used to define the action duration window of the controller. This photovoltaic regulation response data structure is sent to the photovoltaic control module to implement dynamic response execution based on the load state and structural behavior, and is used to support the real-time deployment and feedback regulation of the actual control layer.

[0076] Optionally, the present application also provides a photovoltaic power supply system driven by bridge structure load perception for executing the photovoltaic power supply method driven by bridge structure load perception as described above. The photovoltaic power supply system driven by bridge structure load perception includes: a structural load perception module, which is used to collect load response data through sensors arranged at key parts of the bridge; extract load perception feature data according to the load response data; a load-driven energy regulation mapping module, which is used to perform load energy regulation mapping on the load perception feature data to obtain photovoltaic regulation data; a bridge-vehicle coupling digital twin prediction module, which is used to obtain historical traffic flow data and construct a digital twin based on the historical traffic flow data and a preset bridge-vehicle coupling model to obtain bridge-vehicle twin data; an intelligent energy response control module, which is used to perform energy regulation response according to the photovoltaic regulation data and the bridge-vehicle twin data to obtain photovoltaic regulation response data.

[0077] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended application documents rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be included in the present invention.

[0078] The above are only specific embodiments of the present invention, which enable those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A photovoltaic power supply method driven by bridge structure load perception, characterized in that, The method includes: S1. Collect load response data through sensors arranged at key parts of the bridge; extract load perception features based on the load response data to obtain load perception feature data; S2. Perform load energy regulation mapping on the load perception feature data to obtain photovoltaic regulation data; S3. Obtain historical traffic flow data, and construct a digital twin based on the historical traffic flow data and a preset bridge-vehicle coupling model to obtain bridge-vehicle twin data; S4. Perform energy regulation response based on the photovoltaic regulation data and the bridge-vehicle twin data to obtain photovoltaic regulation response data.

2. The method according to claim 1, characterized in that S1 includes: Collect load response data through sensors arranged at key parts of the bridge; Construct a bionic dynamic expression based on the load response data to obtain bionic expression data; Construct a modal energy spectrum map for the bionic expression data to obtain modal energy spectrum map data; Perform graph embedding convolution on the modal energy spectrum map data to obtain structure perception data; Identify structure impacts on the structure perception data to obtain structure impact data; Attribute the load behavior based on the structure impact data to obtain load perception feature data.

3. The method according to claim 2, wherein The bionic dynamic expression construction includes: Perform time-based synchronous conversion on the load response data to obtain response conversion data; Perform neural-like excitation on the response conversion data to obtain pulse sequence data; Perform neural synapse structure mapping on the pulse sequence data to obtain synaptic activation weight map data; Extract bionic dynamic expression features based on the synaptic activation weight map data to obtain bionic expression data.

4. The method according to claim 2, wherein The modal energy spectrum map construction includes: Reconstruct the periodic structure of the bionic expression data to obtain periodic response data; Perform frequency-domain analysis on the periodic response data to obtain modal frequency-domain data; Extract the modal energy spectrum from the modal frequency-domain data to obtain modal energy spectrum data; Construct a modal excitation spectrum map based on the modal energy spectrum data to obtain modal energy spectrum map data.

5. The method according to claim 2, characterized in that: The structure impact identification includes: Screen structure perturbations on the structure perception data to obtain structure perturbation data; Perform vibration energy analysis on the structure perturbation data to obtain vibration energy data; Match impact events based on the vibration energy data to obtain impact event data; Fit the impulse response based on the impact event data to obtain structure impact data.

6. The method according to claim 1, characterized in that S2 Includes: Select regulation factors for the load perception feature data to obtain regulation factor data; Construct a load state space based on the regulation factor data to obtain load state space data; Perform grey decision-making based on the load state space data to obtain photovoltaic strategy matching data; Output a photovoltaic regulation network based on the photovoltaic strategy matching data to obtain photovoltaic regulation data.

7. The method according to claim 6, characterized in that The grey decision-making includes: Process the modal change trend of the load state space data to obtain load change space data; Calculate the grey correlation degree based on the load change space data and a preset historical regulation library to obtain grey correlation degree data; Match strategies based on the grey correlation degree data to obtain photovoltaic strategy matching data.

8. The method according to claim 1, wherein S3 Includes: Obtain historical traffic flow data; Extract traffic semantic features based on the historical vehicle data to obtain traffic semantic feature data; Parameter back-calibration is performed according to traffic semantic feature data and a preset bridge-vehicle coupling model to obtain an updated bridge-vehicle model; The impact propagation path of the updated bridge-vehicle model is captured to obtain impact propagation path data; Evolution simulation is performed based on the impact propagation path data to obtain bridge-vehicle twin data.

9. The method according to claim 1, wherein S4 It includes: A state trigger matrix is constructed according to the photovoltaic regulation data and the bridge-vehicle twin data to obtain state trigger data; Logical regression processing is performed on the state trigger data to obtain response condition data; A response path is generated according to the response condition data to obtain response path data; Redundancy screening is performed on the response path data to obtain path screening data; Execution level control is performed on the path screening data to obtain photovoltaic regulation response data.

10. A photovoltaic energy supply system driven by bridge structure load sensing, characterized in that: For implementing the photovoltaic energy supply method driven by bridge structure load perception as described in claim 1, the bridge structure load perception-driven photovoltaic energy supply system includes: A structure load perception module, configured to collect load response data through sensors arranged at key parts of the bridge; perform load perception feature extraction according to the load response data to obtain load perception feature data; A load-driven energy regulation mapping module, configured to perform load energy regulation mapping on the load perception feature data to obtain photovoltaic regulation data; A bridge-vehicle coupling digital twin prediction module, configured to obtain historical traffic flow data, and construct a digital twin according to the historical traffic flow data and a preset bridge-vehicle coupling model to obtain bridge-vehicle twin data; An intelligent energy response control module, configured to perform energy regulation response according to the photovoltaic regulation data and the bridge-vehicle twin data to obtain photovoltaic regulation response data.

Images