Dangerous chemical wharf breakwater monitoring system and method based on multi-source data integration

Abstract

The application discloses a dangerous chemical wharf breakwater monitoring system and method based on multi-source data integration, which first acquires multi-modal monitoring data of a breakwater first surface structure, an environmental load, a breakwater underwater structure and a breakwater second surface structure through a multi-source sensor; then a model containing a space-time feature extraction, dynamic feature analysis and multi-source data fusion module is used to calculate four safety monitoring coefficients of the first surface structure, the environment, the underwater structure and the second surface structure; an eight-layer fine-grained cross-fusion mechanism is innovatively introduced, multi-modal feature deep interaction and fusion are realized through cross-modal attention, gate aggregation and other methods; a dynamic threshold is obtained by establishing a digital twin model, and adaptive adjustment is carried out based on fine-grained feature distribution; finally, through multi-level early warning logic, precise early warning from a single index to a multi-modal associated risk is realized. The application improves the state perception and risk early warning capability of a dangerous chemical wharf complex system.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for breakwaters at hazardous chemical terminals, and more specifically, to a monitoring system and method for breakwaters at hazardous chemical terminals based on multi-source data integration. Background Technology

[0002] As the frontal barrier of a wharf, breakwaters directly bear complex environmental loads such as waves, currents, and ship impacts, and their structural health directly affects the safety of the entire wharf. However, traditional breakwater monitoring methods have significant limitations: on the one hand, they often use single sensors or isolated systems for monitoring, such as focusing only on structural displacement or corrosion rates, lacking collaborative perception and fusion analysis of multi-source information from the structure, environment, loads, and the tank area behind, making it difficult to comprehensively grasp the overall safety status of the system; on the other hand, data analysis methods are relatively simple, relying mostly on threshold alarms or simple statistics, failing to deeply mine the spatiotemporal correlations, multimodal coupling relationships, and early hidden fault characteristics contained in the data, resulting in insufficient early warning capabilities and a high rate of false alarms and missed alarms.

[0003] In recent years, digital twin technology has provided new insights into infrastructure health management, enabling deep data fusion and state simulation prediction by constructing virtual mappings of physical entities. Meanwhile, artificial intelligence technologies such as multimodal data analysis and deep learning have demonstrated strong potential in structural health monitoring, capable of processing heterogeneous, high-dimensional, and time-series monitoring data and extracting deep features. However, existing research largely focuses on the analysis of single structures or single data modalities. For complex systems like hazardous chemical terminals, which incorporate multiple high-risk elements such as breakwater structures and complex marine environments, a comprehensive solution is still lacking that can achieve deep integration of multi-source data, cross-fusion of multimodal features, intelligent adjustment of dynamic thresholds, and accurate early warning of systemic risks. In particular, how to quantify the dynamic interaction between environmental loads and structural responses, and how to achieve the leap from single-parameter over-limit alarms to multi-factor coupled risk early warning, remain prominent challenges for current technology.

[0004] Therefore, existing technologies lack accurate perception and intelligent early warning of the entire life cycle, all elements, and all risks of breakwaters and related facilities at hazardous chemical terminals, and lack the ability to accurately perceive the status and provide risk warnings for the complex systems of hazardous chemical terminals. Summary of the Invention

[0005] The purpose of this invention is to provide a monitoring system and method for breakwaters at hazardous chemical terminals based on multi-source data integration, in order to solve the aforementioned problems existing in the prior art.

[0006] The application is as follows:

[0007] A monitoring method for breakwaters at hazardous chemical terminals based on multi-source data integration includes:

[0008] S1. Obtain the current structural settlement value and structural displacement value to obtain the surface structure monitoring data of the first breakwater. Obtain the real-time wave intensity value, the total number of hazardous chemical vessels and the number of hazardous chemical vessels preparing to dock to obtain the basic environmental monitoring data. Obtain the breakwater scour value and collapse value to obtain the underwater structure monitoring data of the breakwater. Obtain the cracks and block displacements on the surface of the breakwater to obtain the surface structure monitoring data of the second breakwater. Define the surface structure monitoring data of the first breakwater, the basic environmental monitoring data, the underwater structure monitoring data of the breakwater, and the surface structure monitoring data of the second breakwater as the basic health monitoring data of the breakwater.

[0009] S2. Spatiotemporal features are extracted from the surface structure monitoring data of the first breakwater using a multimodal feature analysis model to calculate the first structural safety monitoring coefficient. Dynamic features are analyzed from the environmental monitoring data using a multimodal feature analysis model to calculate the environmental risk monitoring coefficient. Underwater performance structural features are analyzed from the underwater structure monitoring data of the breakwater using a multimodal feature analysis model to calculate the underwater safety structure monitoring coefficient of the breakwater. Multi-source data fusion analysis is performed from the surface structure monitoring data of the second breakwater using a multimodal feature analysis model to calculate the second structural safety monitoring coefficient. The first structural safety monitoring coefficient, the environmental risk monitoring coefficient, the underwater safety structure monitoring coefficient of the breakwater, and the second structural safety monitoring coefficient are defined as breakwater health monitoring analysis data.

[0010] S3. Establish a digital twin model of the breakwater, and obtain the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater through the digital twin model of the breakwater. Define the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater as the breakwater health monitoring threshold data.

[0011] S4. Conduct safety risk warnings based on breakwater health monitoring analysis data and breakwater health monitoring threshold data.

[0012] Furthermore, in step S1, acquiring monitoring data on the surface structure of the first breakwater includes:

[0013] By deploying N monitoring points on the breakwater, each equipped with a GNSS receiver, satellite signals are received in real time, the three-dimensional coordinates of each monitoring point are calculated, and the displacement values ​​of the breakwater monitoring points are obtained based on the three-dimensional coordinates, thus obtaining the structural displacement values. Within the current monitoring cycle, k monitoring time nodes are determined for the N monitoring points, with a standard monitoring duration between each two consecutive monitoring time nodes. The structural settlement values ​​corresponding to the k monitoring time nodes are obtained through the GNSS receiver and marked as the first settlement rate to the kth settlement rate. A weighted average is calculated for the first settlement rate to the kth settlement rate to obtain the structural settlement value. The current structural settlement value and structural displacement value are defined as the surface structural monitoring data of the first breakwater.

[0014] Furthermore, in step S1, acquiring basic environmental monitoring data includes:

[0015] The wave intensity values ​​corresponding to the breakwater are obtained in real time by wave sensors to obtain real-time wave intensity values; the total number of hazardous chemical vessels currently docked at the wharf and the number of hazardous chemical vessels that have not yet docked at the wharf are obtained by camera image recognition algorithms. The total number of hazardous chemical vessels currently docked at the wharf is marked as the total number of hazardous chemical vessels, and the number of hazardous chemical vessels that have not yet docked at the wharf is marked as the number of hazardous chemical vessels that are expected to dock; the real-time wave intensity values, the total number of hazardous chemical vessels, and the number of hazardous chemical vessels that are expected to dock are defined as basic environmental monitoring data.

[0016] Furthermore, in step S1, acquiring monitoring data on the surface structure of the second breakwater specifically includes:

[0017] Based on navigation technology, the initial navigation path of the drone is determined, and a high-definition camera and a 3D LiDAR are configured on the drone.

[0018] The drone flew along the initial navigation path and collected basic information and initial image data of the breakwater surface structure using its onboard high-definition camera.

[0019] Based on the basic information, the key points of the breakwater surface structure were collected in the region to obtain the coordinates of several key points in the region.

[0020] The initial image data, the basic information, and the coordinates of key points in several regions are input into the navigation parameter planning model, and the navigation path trajectory set of the UAV is output.

[0021] The drone periodically acquires images according to the navigation path trajectory set. The images captured by the drone are analyzed using image recognition algorithms to identify the crack values ​​and block displacement values ​​on the surface of the breakwater. The crack values ​​and block displacement values ​​are defined as the second breakwater surface structure monitoring data.

[0022] Furthermore, S2 includes:

[0023] A database is created, storing data including allowable settlement thresholds and allowable structural displacement thresholds; a multimodal feature analysis model is constructed, comprising a spatiotemporal feature extraction module, a dynamic feature analysis module, an underwater performance structure analysis module, a multi-source data fusion module, and a fine-grained cross-fusion layer, with a total of 8 fine-grained cross-fusion layers.

[0024] S21. Obtain basic data for breakwater health monitoring, and based on the basic data for breakwater health monitoring, obtain the surface structure monitoring data of the first breakwater, the basic environmental monitoring data, the underwater structure monitoring data of the breakwater, and the surface structure monitoring data of the second breakwater.

[0025] S22. Analyze the monitoring data of the surface structure of the first breakwater through the spatiotemporal feature extraction module to obtain the safety monitoring coefficient Ks of the first structure;

[0026] S23. Analyze the basic environmental monitoring data through the dynamic feature analysis module to obtain the environmental risk monitoring coefficient Ke;

[0027] S24. Analyze the underwater structure monitoring data of the breakwater using the underwater performance structure analysis module to obtain the underwater safety structure monitoring coefficient Km of the breakwater.

[0028] S25. The monitoring data of the surface structure of the second breakwater is analyzed by the multi-source data fusion module to obtain the safety monitoring coefficient Kt of the second structure;

[0029] S26. Intramodal multi-scale feature enhancement is achieved through the first fine-grained cross-fusion layer for the first fusion analysis:

[0030] Multi-scale spatiotemporal characteristics were extracted from the monitoring data of the surface structure of the first breakwater:

[0031] Fs = [ , , ..., = MLP(CNN(LSTM( ))),in, The original data sequence for monitoring the breakwater structure, The dimensions are [T1, Ds], where T1 is the time step and Ds is the structural feature dimension; LSTM represents Long Short-Term Memory Network, used to capture the time dependence of the monitoring data of the surface structure of the first breakwater; CNN represents a one-dimensional convolutional neural network, used to extract local patterns of the monitoring data of the surface structure of the first breakwater; MLP represents Multilayer Perceptron, used for feature transformation and dimensionality reduction; Fs represents multi-scale spatiotemporal feature vector. This represents the nth element of the multi-scale spatiotemporal eigenvector;

[0032] Extract multi-resolution dynamic features from basic environmental monitoring data:

[0033] Fe = [ , , ..., ] = Wavelet(GRU( )),in, The environmental monitoring basic data sequence has the dimension [T, De], where De is the environmental feature dimension; GRU is a gated recurrent unit neural network; Wavelet is a wavelet transform used to extract time-frequency features; Fe represents a multi-resolution dynamic feature vector. This represents the m-th element of the multi-resolution dynamic feature vector;

[0034] S27. Cross-modal attention feature interaction is achieved through a second fusion analysis using a second fine-grained cross-fusion layer. The input of the second fine-grained cross-fusion layer is the output of the first fine-grained cross-fusion layer.

[0035] Construct the structure-environment cross-modal attention matrix Ase to calculate fine-grained structural features under the influence of environmental conditions. :

[0036] Ase = softmax((Wq*Fs)*(Wk*Fe) T / qqq);

[0037] = Fs + Ase * (Wv * Fe);

[0038] Where, Wq ∈ , Wk ∈ , Wv ∈ Wq, Wk, and Wv represent the learnable weight matrices, d represents the hidden layer dimension, qqq represents the scaling factor, T represents the matrix transpose, and R represents the set of real numbers.

[0039] Calculate the fine-grained environmental characteristics under the influence of structural state. :

[0040] Aes = , where Aes represents the transpose of Ase;

[0041] = Fe + Aes * (Wv * Fs);

[0042] S28, Based on cross-fusion fine-grained structural features and fine-grained environmental characteristics The monitoring coefficients are reconstructed through a third fine-grained cross-fusion layer, and a third fusion analysis is performed to achieve gated multimodal feature aggregation.

[0043] = σ( Ws * pool( ) + bs );

[0044] = σ( We * pool( ) + be );

[0045] gate = sigmoid( Wg * [pool( ), pool( )] + bg);

[0046] = Km + λ * gate * ( + );

[0047] align = cosine_similarity( pool( ), pool( ));

[0048] = Kt + μ * align * ( + );

[0049] Where Ws represents the pooled structural feature pool ( Mapped to the first structural safety monitoring coefficient The weight matrix; We represents the pooled environmental features pooled together. Mapped to environmental risk monitoring coefficient The weight matrix; Wg represents the weight matrix used to combine the spliced ​​structure with environmental features [pool( pool( The weight matrix mapped to the gate signal; bs, be, bg represent the bias vectors of the learnable model; gate represents the gate signal; pool(·) represents the pooling function; σ(·) represents the activation function; λ represents the fusion intensity coefficient; μ represents the alignment gain coefficient; align represents the feature alignment degree; This represents the refined first structural safety monitoring coefficient; This represents the refined environmental risk monitoring coefficient; This represents the modified underwater safety structure monitoring coefficient of the dike. This represents the corrected safety monitoring coefficient for the second structure; cosine_similarity represents the cosine similarity function.

[0050] S29. Monitoring coefficient of underwater safety structure of the dike Environmental risk monitoring coefficient First structural safety monitoring coefficient Second structural safety monitoring coefficient and all fine-grained structural features and fine-grained environmental characteristics They are collectively defined as breakwater health monitoring and analysis data.

[0051] Further, S22 includes:

[0052] S221. Spatiotemporal correlation analysis is performed on the monitoring data of the surface structure of the first breakwater using the spatiotemporal feature extraction module.

[0053] S222. Based on the analysis results, obtain the spatiotemporal characteristic vectors of the current structural settlement and displacement values;

[0054] S223. Obtain the allowable structural settlement threshold Dd0 and the allowable structural displacement threshold Vf0 from the database;

[0055] S224. The first structural safety monitoring coefficient Ks is obtained by multimodal calculation using the spatiotemporal feature vector and the corresponding threshold:

[0056] Ks = β * (Dd / Dd0) + γ * (Vf / Vf0), where α, β, γ are weighting coefficients, and β+γ=1, Dd represents the structural settlement characteristic value, and Vf represents the structural displacement characteristic value;

[0057] S225. A fine-grained contextual awareness is introduced through the fourth fine-grained cross-fusion layer to perform the fourth fusion analysis and calculate the contextual vector:

[0058] Extracting short-term fluctuation characteristics (Eshort) and long-term trend characteristics (Elong) from environmental data:

[0059] Eshort = TCN( ), where TCN stands for Temporal Convolutional Network, used to process recent data; This represents a short-term environmental data sequence;

[0060] Elong = Transformer( Here, Transformer represents the Transformer neural network, which is used to process long-term historical data; Represents a long-term environmental data sequence;

[0061] Calculate the environment context vector Cenv:

[0062] Cenv = Attention([Eshort, Elong]), where Attention represents the attention function;

[0063] S226. Generate the fine-grained first-structure safety monitoring coefficient Ksfine based on the environmental context vector. Perform a fifth fusion analysis on the fine-grained first-structure safety monitoring coefficient through the fifth fine-grained cross-fusion layer to achieve adaptive weighting, and obtain the final first-structure safety monitoring coefficient:

[0064] Ksfine = LSTM([ ,Cenv]);

[0065] = sigmoid([ , Cenv]), where Indicates adaptive weights;

[0066] Ks = * Ks + (1 - Ksfine, where Ks represents the final first structural safety monitoring coefficient.

[0067] Furthermore, S25 includes:

[0068] S253. Multimodal features are extracted from the surface structure monitoring data of the second breakwater through the multi-source data fusion module to generate multimodal features, including crack values ​​and block displacement values.

[0069] S254. Obtain the crack threshold Tt0 and block displacement threshold Pp0 from the database;

[0070] S255. The initial value of the second structural safety monitoring coefficient Kt is obtained by calculating the multimodal features and corresponding safety thresholds using a BP deep neural network:

[0071] Kt = δ1 * (Tt / Tt0) + δ2 * (Pp / Pp0), where δ1 and δ2 are weights, Tt represents the crack characteristic value, and Pp represents the block displacement characteristic value;

[0072] S256. Set up multimodal fine-grained regions of interest (ROIs) through the sixth fine-grained cross-fusion layer, and perform a sixth fusion analysis using the multimodal fine-grained ROIs:

[0073] For n monitoring points on the breakwater, based on their spatial location ,crack Block displacement Calculate attention weights:

[0074] = exp(MLP([ , , ])) / ,in This represents the attention weight corresponding to the i-th monitoring point;

[0075] S257. Seventh fusion analysis was performed using the seventh fine-grained cross-fusion layer:

[0076] The structural settlement characteristics Xvib from the surface structure monitoring of the first breakwater and the crack development characteristics Pfluct from the surface structure monitoring of the second breakwater are cross-modal aligned, and the alignment matrix Avibp is calculated:

[0077] Avibp = crossattention(Xvib, Pfluct), where crossattention represents the cross attention function;

[0078] Calculate the surface safety factor Ktfine of the fine-grained second breakwater under physical constraints:

[0079] = + bf1 * MLP( ), where bf1 is the physical influence coefficient, and i represents the i-th monitoring point, Ktfine represents the fine-grained surface safety factor of the second breakwater at the i-th monitoring point; This represents the initial value of the second structural safety monitoring coefficient Kt at the i-th monitoring point; Avibp represents the alignment matrix of the i-th monitoring point;

[0080] S258. Eighth fusion analysis is performed through the eighth fine-grained cross-fusion layer:

[0081] = ,in This represents the final safety monitoring coefficient for the second structure.

[0082] Further, S3 includes:

[0083] S31. Obtain historical operational data of the breakwater and use the historical operational data of the breakwater to establish a digital twin model of the breakwater through BIM modeling tools;

[0084] S32. Obtain the following thresholds through digital twin models: allowable displacement threshold, crack propagation threshold, wave intensity threshold, hazardous chemical vessel capacity threshold, number of vessels to be moored, allowable settlement threshold, crack threshold, and block displacement threshold.

[0085] S33. Obtain the allowable settlement threshold based on the basic data of breakwater health monitoring;

[0086] S34. The structural settlement threshold and structural displacement threshold are calculated using multimodal methods to obtain the basic threshold Ks0base of the first structural safety monitoring coefficient.

[0087] S35. Dynamic feature calculations are performed using wave intensity threshold, hazardous chemical vessel capacity threshold, and pre-berth vessel number threshold to obtain the basic threshold for environmental risk monitoring coefficient Ke0base.

[0088] S36. Calculate the underwater safety performance characteristics of the scour threshold and collapse threshold of the embankment to obtain the basic threshold Km0base of the underwater safety structure monitoring coefficient of the embankment.

[0089] S37. The crack threshold and block displacement threshold are calculated by multi-source data fusion to obtain the basic threshold Kt0base of the second structural safety monitoring coefficient.

[0090] S38. Adjust the first structural safety monitoring coefficient threshold and the environmental risk monitoring coefficient threshold based on the fine-grained structural features and the fine-grained environmental features, respectively, to generate the first dynamic first structural safety monitoring coefficient threshold and the first environmental risk monitoring threshold.

[0091] Calculate fine-grained structural features Historical distribution statistics are used to dynamically adjust the threshold of the first structural safety monitoring coefficient.

[0092] Ks0 = Ks0base * (1 + ηs * (|| - μs|| / σs)), where ηs is the sensitivity coefficient, μs represents the mean, σs represents the standard deviation, and Ks0 represents the threshold of the first dynamic first structural safety monitoring coefficient;

[0093] Calculate fine-grained environmental features The real-time fluctuation index dynamically adjusts the environmental risk monitoring coefficient threshold.

[0094] Ke0 = Ke0base * (1 + ηe * (std( ) / mean( ))), where ηe represents the environmental fluctuation coefficient, Ke0 represents the first environmental risk monitoring coefficient threshold; std represents the standard deviation function, and mean represents the mean function;

[0095] S39. Generate the structure-environment coupling threshold by coupling the cross-modal attention matrix Ase from S27 and the first dynamic first structural safety monitoring coefficient threshold:

[0096] Kse0 = (Ase * Ke0) + ((1 - Ase) * Ks0);

[0097] Using the gate signal (gate), alignment (align), and first environmental risk monitoring coefficient threshold of S28, a levee-environment association threshold is generated:

[0098] Kmt0 = gate * (Km0base + Kt0base) + (1 - gate) * align * max(Km0base,Kt0base);

[0099] S40. Define the basic thresholds Ks0base, Ke0base, Km0base, Kt0base, the dynamically adjusted thresholds Ks0, Ke0, and the coupling thresholds Kse0, Kmt0 together as the breakwater health monitoring threshold data.

[0100] A monitoring system for breakwaters at hazardous chemical terminals based on multi-source data integration, the system being used to execute any of the methods described in the "Methods for Monitoring Breakwaters at Hazardous Chemical Terminals Based on Multi-Source Data Integration", the system comprising a monitoring data acquisition module, a multimodal analysis module, a threshold analysis module, and an early warning analysis module;

[0101] The monitoring data acquisition module acquires the current structural settlement value and structural displacement value to obtain the surface structure monitoring data of the first breakwater, acquires the real-time wave intensity value, the total number of hazardous chemical vessels and the number of hazardous chemical vessels preparing to dock, acquires the basic environmental monitoring data, acquires the breakwater scour value and collapse value to obtain the underwater structure monitoring data of the breakwater, acquires the cracks and block displacements on the surface of the breakwater, and acquires the surface structure monitoring data of the second breakwater. The surface structure monitoring data of the first breakwater, the basic environmental monitoring data, the underwater structure monitoring data of the breakwater, and the surface structure monitoring data of the second breakwater are defined as the basic health monitoring data of the breakwater.

[0102] The multimodal analysis module extracts spatiotemporal features from the surface structure monitoring data of the first breakwater and calculates the first structural safety monitoring coefficient. It then performs dynamic feature analysis on the environmental monitoring data using a multimodal feature analysis model to calculate the environmental risk monitoring coefficient. Finally, it performs underwater performance structural feature analysis on the underwater structure monitoring data of the breakwater using the same model to calculate the underwater safety structural monitoring coefficient. Finally, it performs multi-source data fusion analysis on the surface structure monitoring data of the second breakwater using the same model to calculate the second structural safety monitoring coefficient. The first structural safety monitoring coefficient, the environmental risk monitoring coefficient, the underwater safety structural monitoring coefficient, and the second structural safety monitoring coefficient are defined as breakwater health monitoring analysis data.

[0103] The threshold analysis module establishes a digital twin model of the breakwater, and obtains the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater through the digital twin model. The underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold are defined as breakwater health monitoring threshold data.

[0104] The early warning analysis module provides safety risk warnings based on breakwater health monitoring analysis data and breakwater health monitoring threshold data.

[0105] Compared with the prior art, the present invention achieves the following beneficial effects:

[0106] This invention acquires multimodal monitoring data on the first surface structure, environmental load, underwater structure, and second surface structure of a breakwater using multi-source sensors. It then utilizes a model incorporating spatiotemporal feature extraction, dynamic feature analysis, and multi-source data fusion modules to calculate four safety monitoring coefficients for the first surface structure, environment, underwater structure, and second surface structure. It innovatively introduces an eight-layer fine-grained cross-fusion mechanism, achieving deep interaction and fusion of multimodal features through cross-modal attention and gated aggregation. Dynamic thresholds are obtained by establishing a digital twin model and adaptively adjusted based on fine-grained feature distribution. Finally, through multi-level early warning logic, it achieves accurate early warning of risks from single indicators to multimodal correlations. This invention enhances the state perception and risk early warning capabilities of complex systems at hazardous chemical terminals. Attached Figure Description

[0107] Figure 1 This is a flowchart illustrating the method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration provided in this embodiment of the invention.

[0108] Figure 2This is an architecture diagram of the hazardous chemical terminal breakwater monitoring system based on multi-source data integration provided in an embodiment of the present invention;

[0109] Figure 3 This is a breakwater monitoring network topology diagram of the hazardous chemical terminal breakwater monitoring system based on multi-source data integration provided in an embodiment of the present invention. Detailed Implementation

[0110] The present invention will now be described in detail with reference to the accompanying drawings.

[0111] Example 1

[0112] This invention provides a method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration, such as... Figure 1 As shown, the method includes:

[0113] S1. Obtain current structural settlement and displacement values ​​to obtain surface structure monitoring data for the first breakwater; obtain real-time wave intensity values, total number of hazardous chemical vessels, and number of hazardous chemical vessels preparing to dock to obtain basic environmental monitoring data; obtain breakwater scour and collapse values ​​to obtain underwater structure monitoring data for the breakwater; obtain surface cracks and block displacements of the breakwater to obtain surface structure monitoring data for the second breakwater; define the surface structure monitoring data for the first breakwater, basic environmental monitoring data, underwater structure monitoring data, and surface structure monitoring data for the second breakwater as basic health monitoring data for the breakwater; the structural displacement values ​​represent the overall structural displacement of the breakwater; the block displacements represent the local structural displacements of the breakwater.

[0114] S2. Spatiotemporal features are extracted from the surface structure monitoring data of the first breakwater using a multimodal feature analysis model to calculate the first structural safety monitoring coefficient. Dynamic features are analyzed from the environmental monitoring data using a multimodal feature analysis model to calculate the environmental risk monitoring coefficient. Underwater performance structural features are analyzed from the underwater structure monitoring data of the breakwater using a multimodal feature analysis model to calculate the underwater safety structure monitoring coefficient of the breakwater. Multi-source data fusion analysis is performed from the surface structure monitoring data of the second breakwater using a multimodal feature analysis model to calculate the second structural safety monitoring coefficient. The first structural safety monitoring coefficient, the environmental risk monitoring coefficient, the underwater safety structure monitoring coefficient of the breakwater, and the second structural safety monitoring coefficient are defined as breakwater health monitoring analysis data.

[0115] S3. Establish a digital twin model of the breakwater, and obtain the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater through the digital twin model of the breakwater. Define the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater as the breakwater health monitoring threshold data.

[0116] S4. Conduct safety risk warnings based on breakwater health monitoring analysis data and breakwater health monitoring threshold data.

[0117] Specifically, this invention first acquires multimodal monitoring data on the first surface structure, environmental load, underwater structure, and second surface structure of the breakwater using multi-source sensors. Then, using a model incorporating spatiotemporal feature extraction, dynamic feature analysis, and multi-source data fusion modules, it calculates four safety monitoring coefficients for the first surface structure, environment, underwater structure, and second surface structure. An innovative eight-layer fine-grained cross-fusion mechanism is introduced, achieving deep interaction and fusion of multimodal features through cross-modal attention and gating aggregation. Dynamic thresholds are obtained by establishing a digital twin model and adaptively adjusted based on fine-grained feature distribution. Finally, through multi-level early warning logic, it achieves accurate early warning of risks from single indicators to multimodal correlation risks. This invention enhances the state perception and risk early warning capabilities of complex systems at hazardous chemical terminals.

[0118] In the above embodiments, specifically, in step S1, acquiring monitoring data on the surface structure of the first breakwater includes:

[0119] like Figure 3 As shown, N monitoring points are set up on the breakwater, each equipped with a GNSS receiver to receive satellite signals in real time, calculate the three-dimensional coordinates of each monitoring point, and obtain the displacement values ​​of the breakwater monitoring points based on the three-dimensional coordinates to obtain the structural displacement values. Within the current monitoring cycle, k monitoring time nodes are determined for the N monitoring points, with a standard monitoring duration between each two consecutive monitoring time nodes. The structural settlement values ​​corresponding to the k monitoring time nodes are obtained through the GNSS receiver and marked as the first settlement rate to the kth settlement rate. A weighted average is calculated for the first settlement rate to the kth settlement rate to obtain the structural settlement value. The current structural settlement value and structural displacement value are defined as the monitoring data of the first breakwater surface structure. The monitoring frequency is set to 1 time / hour for each GNSS receiver.

[0120] It should be noted that the principle of GNSS automated monitoring is as follows: Utilizing GNSS (Global Navigation Satellite System) technology, multiple monitoring points are deployed on the breakwater to receive satellite signals in real time. The system then calculates the three-dimensional coordinate changes of each monitoring point, achieving high-precision monitoring of the breakwater's surface settlement and displacement. The GNSS monitoring system features high precision, high frequency, and a high degree of automation, and can provide real-time displacement data of the monitoring points.

[0121] Monitoring point deployment and monitoring frequency: 25 GNSS monitoring points are evenly deployed along the breakwater axis, each equipped with a high-precision GNSS receiver. The monitoring frequency is once per hour, and the data is transmitted to the monitoring center in real time via a wireless network.

[0122] In step S1, acquiring basic environmental monitoring data includes:

[0123] In the above embodiments, specifically, wave intensity values ​​corresponding to the breakwater are obtained in real time through wave sensors to obtain real-time wave intensity values; the total number of hazardous chemical vessels currently docked at the wharf and the number of hazardous chemical vessels that have not yet docked at the wharf are obtained through camera image recognition algorithms, the total number of hazardous chemical vessels currently docked at the wharf is marked as the total number of hazardous chemical vessels, and the number of hazardous chemical vessels that have not yet docked at the wharf is marked as the number of hazardous chemical vessels that are expected to dock; the real-time wave intensity values, the total number of hazardous chemical vessels, and the number of hazardous chemical vessels that are expected to dock are defined as basic environmental monitoring data.

[0124] It should be noted that camera image recognition algorithms require the installation and deployment of a monitoring system to acquire image data. The installation and layout of the monitoring system are as follows: Figure 3 As shown, five high-definition cameras are deployed at key locations on the breakwater. Two are installed at the LNG terminal, one at the breakwater head, and the other two on both sides of the breakwater to achieve real-time monitoring of the breakwater. The camera image recognition algorithm obtains the total number of hazardous chemical vessels currently docked at the terminal and the number of hazardous chemical vessels that have not yet docked. The total number of hazardous chemical vessels currently docked at the terminal is marked as the total number of hazardous chemical vessels, and the number of hazardous chemical vessels that have not yet docked at the terminal is marked as the number of hazardous chemical vessels that are expected to dock.

[0125] In the above embodiments, specifically, in step S1, acquiring monitoring data on the surface structure of the second breakwater includes:

[0126] Based on navigation technology, the initial navigation path of the drone is determined, and a high-definition camera and a 3D LiDAR are configured on the drone.

[0127] The drone flew along the initial navigation path and collected basic information and initial image data of the breakwater surface structure using its onboard high-definition camera.

[0128] Based on the basic information, the key points of the breakwater surface structure were collected in the region to obtain the coordinates of several key points in the region.

[0129] The initial image data, the basic information, and the coordinates of key points in several regions are input into the navigation parameter planning model, and the navigation path trajectory set of the UAV is output.

[0130] The drone periodically acquires images according to the navigation path trajectory set. The images captured by the drone are analyzed using image recognition algorithms to identify the crack values ​​and block displacement values ​​on the surface of the breakwater. The crack values ​​and block displacement values ​​are defined as the second breakwater surface structure monitoring data.

[0131] It should be noted that, in addition to image data collected by high-definition cameras on drones, data from other sensors, such as LiDAR, can also be combined. LiDAR can acquire high-precision 3D point cloud data of the terrain. After being fused with image data, it can more accurately reflect the topographic features of the breakwater and help to more accurately identify cracks and block displacements.

[0132] For example, in breakwater areas with complex terrain, LiDAR data can supplement the lack of depth information in image data, better identifying subtle displacement changes. Multi-temporal data comparison: Regularly collected data should cover different time periods. By comparing image data from different time phases, the development trend of cracks and the direction and velocity of block displacement can be observed more clearly. For instance, comparing this year's aerial photography data with last year's can visually show whether cracks are expanding or shrinking, the direction in which the blocks are moving, and the distance they have moved. This provides a stronger basis for breakwater maintenance and repair, and further allows for the calculation of crack values ​​and block displacement values ​​on the breakwater surface.

[0133] Advanced deep learning algorithms, such as Convolutional Neural Networks (CNNs) or Generative Adversarial Networks (GANs), are employed to process and identify image data. The algorithms are trained using a large amount of labeled data, enabling them to more accurately identify cracks and block displacement features. For example, CNNs can automatically extract features such as texture and shape of cracks from image data, while GANs can generate more realistic crack and displacement samples for training, improving the robustness and accuracy of the algorithm.

[0134] Multi-scale analysis of image data enables the identification of cracks and block displacements at different scales, allowing for the calculation of crack and block displacement values ​​on the breakwater surface. At a large scale, potentially problematic areas can be quickly located, while at a small scale, more detailed analysis of crack features and minute block displacements is possible. For example, large-scale analysis of the overall breakwater image data can first identify suspicious areas, followed by detailed small-scale analysis of these areas, improving identification efficiency and accuracy. Image processing techniques, such as edge enhancement and filtering, can be used to preprocess image data to highlight crack and block displacement characteristics, improving the effectiveness of the identification algorithm. For instance, edge enhancement algorithms can make crack edges clearer, facilitating subsequent identification and measurement; filtering can remove noise from image data, reducing false identifications.

[0135] In GIS data processing, the spatial topological relationships of the breakwater surface are constructed to analyze the spatial correlation between cracks and block displacement. For example, if multiple cracks are spatially interconnected or the direction of block displacement is consistent with the direction of crack extension, it may indicate an increased risk of collapse. Based on this, collapse points can be more reasonably marked, and then the crack values ​​and block displacement values ​​of the breakwater surface can be calculated.

[0136] By combining expert knowledge and experience, a series of rules are formulated to automatically identify collapse points through a rule engine. For example, when the crack width exceeds a certain threshold and the block displacement distance reaches a certain range, it is automatically marked as a high-risk collapse point. At the same time, machine learning algorithms are used to optimize and adjust the rules so that they can better adapt to different breakwater conditions and environmental changes, and obtain the crack values ​​and block displacement values ​​of the breakwater surface.

[0137] Machine learning algorithms possess the ability to provide real-time feedback and dynamic updates. When new image data is input, they can promptly update the marking results of landslide points and adjust and optimize the algorithm model based on the new data. For example, if significant changes are detected in the cracks and block displacement in a certain area, the algorithm can quickly reassess the landslide risk in that area and update the marking results, providing support for the real-time monitoring and maintenance of breakwaters.

[0138] A dedicated database is established to uniformly store and manage the collected image data, GIS data, and identification and labeling results. Distributed storage and efficient data indexing technologies are employed to ensure fast data read / write and retrieval. For example, cloud databases can conveniently store large amounts of data and can be elastically expanded as needed. Furthermore, spatial indexing allows for rapid location of data in specific areas.

[0139] 3D visualization technology is used to visually display image data of breakwaters, identification results of cracks and block displacements, and marking results of collapse points. Users can interact with the breakwaters by rotating, zooming, and panning to view their condition from different angles, gaining a clearer understanding of the distribution and severity of cracks and block displacements. For example, 3D visualization software can generate a 3D model of the breakwater, marking the locations of cracks and collapse points on the model, and then calculating the crack and block displacement values ​​on the breakwater surface.

[0140] It should be noted that the process of obtaining monitoring data for the underwater structure of the breakwater includes:

[0141] The underwater measurements in this embodiment are mainly divided into underwater topographic measurements of the surrounding area on both sides of the breakwater below the water surface and side-scan sonar measurements. The measurement area is within 50m on both sides of the breakwater centerline. According to the actual situation in this embodiment, the measurement lines are laid out parallel to the breakwater slope during multibeam measurement, with a measurement line spacing of 10 meters. The total length of the measurement lines is about 20km. In areas where the water depth is less than 5m, only the center beam data of the track is used. The measurement line layout of the side-scan sonar is the same as that of the multibeam measurement line layout.

[0142] The depth sounding system and side-scan sonar system use the 3DSS-iDX side-scan sonar, the Trimble SPS356 as the navigation and positioning device, the Ping DSP series software built into the 3DSS-iDX, the 3DSS-DX Sonar Contro as the water depth data and side-scan sonar data acquisition software, Cairs as the water depth data processing software, SonarWiz as the side-scan sonar data post-processing software, and Yidiantong measurement software as the navigation and positioning software.

[0143] The sounding system measures underwater topography and geomorphology with accuracy meeting IHO's top-class measurement standards. High-resolution images of underwater structures are acquired using side-scan sonar to detect underwater scour and collapse, thereby generating scour and collapse data for the dike.

[0144] In the above embodiments, specifically, step S2 includes:

[0145] A database is created, storing data including allowable settlement thresholds and allowable structural displacement thresholds; a multimodal feature analysis model is constructed, comprising a spatiotemporal feature extraction module, a dynamic feature analysis module, an underwater performance structure analysis module, a multi-source data fusion module, and a fine-grained cross-fusion layer, with a total of 8 fine-grained cross-fusion layers.

[0146] S21. Obtain basic data for breakwater health monitoring, and based on the basic data for breakwater health monitoring, obtain the surface structure monitoring data of the first breakwater, the basic environmental monitoring data, the underwater structure monitoring data of the breakwater, and the surface structure monitoring data of the second breakwater.

[0147] S22. Analyze the monitoring data of the surface structure of the first breakwater through the spatiotemporal feature extraction module to obtain the safety monitoring coefficient Ks of the first structure;

[0148] S23. Analyze the basic environmental monitoring data through the dynamic feature analysis module to obtain the environmental risk monitoring coefficient Ke;

[0149] S24. Analyze the underwater structure monitoring data of the breakwater using the underwater performance structure analysis module to obtain the underwater safety structure monitoring coefficient Km of the breakwater.

[0150] S25. The monitoring data of the surface structure of the second breakwater is analyzed by the multi-source data fusion module to obtain the safety monitoring coefficient Kt of the second structure;

[0151] S26. Intramodal multi-scale feature enhancement is achieved through the first fine-grained cross-fusion layer for the first fusion analysis:

[0152] Multi-scale spatiotemporal characteristics were extracted from the monitoring data of the surface structure of the first breakwater:

[0153] Fs = [ , , ..., = MLP(CNN(LSTM( ))),in, The original data sequence for monitoring the breakwater structure, The dimensions are [T1, Ds], where T1 is the time step and Ds is the structural feature dimension; LSTM represents Long Short-Term Memory Network, used to capture the time dependence of the monitoring data of the surface structure of the first breakwater; CNN represents a one-dimensional convolutional neural network, used to extract local patterns of the monitoring data of the surface structure of the first breakwater; MLP represents Multilayer Perceptron, used for feature transformation and dimensionality reduction; Fs represents multi-scale spatiotemporal feature vector. This represents the nth element of the multi-scale spatiotemporal eigenvector;

[0154] Extract multi-resolution dynamic features from basic environmental monitoring data:

[0155] Fe = [ , , ..., ] = Wavelet(GRU( )),in, The environmental monitoring basic data sequence has the dimension [T, De], where De is the environmental feature dimension; GRU is a gated recurrent unit neural network; Wavelet is a wavelet transform used to extract time-frequency features; Fe represents a multi-resolution dynamic feature vector. This represents the m-th element of the multi-resolution dynamic feature vector;

[0156] S27. Cross-modal attention feature interaction is achieved through a second fusion analysis using a second fine-grained cross-fusion layer. The input of the second fine-grained cross-fusion layer is the output of the first fine-grained cross-fusion layer.

[0157] Construct the structure-environment cross-modal attention matrix Ase to calculate fine-grained structural features under the influence of environmental conditions. :

[0158] Ase = softmax((Wq*Fs)*(Wk*Fe) T / qqq);

[0159] = Fs + Ase * (Wv * Fe);

[0160] Where, Wq ∈ , Wk ∈ , Wv ∈ Wq, Wk, and Wv represent the learnable weight matrices, d represents the hidden layer dimension, qqq represents the scaling factor, T represents the matrix transpose, and R represents the set of real numbers.

[0161] Calculate the fine-grained environmental characteristics under the influence of structural state. :

[0162] Aes = , where Aes represents the transpose of Ase;

[0163] = Fe + Aes * (Wv * Fs);

[0164] S28, Based on cross-fusion fine-grained structural features and fine-grained environmental characteristics The monitoring coefficients are reconstructed through a third fine-grained cross-fusion layer, and a third fusion analysis is performed to achieve gated multimodal feature aggregation.

[0165] = σ( Ws * pool( ) + bs );

[0166] = σ( We * pool( ) + be );

[0167] gate = sigmoid( Wg * [pool( ), pool( )] + bg);

[0168] = Km + λ * gate * ( + );

[0169] align = cosine_similarity( pool( ), pool( ));

[0170] = Kt + μ * align * ( + );

[0171] Where Ws represents the pooled structural feature pool ( Mapped to the first structural safety monitoring coefficient The weight matrix; We represents the pooled environmental features pooled together. Mapped to environmental risk monitoring coefficient The weight matrix; Wg represents the weight matrix used to combine the spliced ​​structure with environmental features [pool( pool( The weight matrix mapped to the gate signal; bs, be, bg represent the bias vectors of the learnable model; gate represents the gate signal; pool(·) represents the pooling function; σ(·) represents the activation function; λ represents the fusion intensity coefficient; μ represents the alignment gain coefficient; align represents the feature alignment degree; This represents the refined first structural safety monitoring coefficient; This represents the refined environmental risk monitoring coefficient; This represents the modified underwater safety structure monitoring coefficient of the dike. This represents the corrected safety monitoring coefficient for the second structure; cosine_similarity represents the cosine similarity function.

[0172] S29. Monitoring coefficient of underwater safety structure of the dike Environmental risk monitoring coefficient First structural safety monitoring coefficient Second structural safety monitoring coefficient and all fine-grained structural features and fine-grained environmental characteristics They are collectively defined as breakwater health monitoring and analysis data.

[0173] It should be noted that this is the core of this embodiment: building a database to store thresholds, baseline values, and historical data;

[0174] Develop a multimodal feature analysis model: Use LSTM+CNN+MLP to process structural data and generate a spatiotemporal feature extraction module; use GRU+Wavelet transform to process environmental data and generate a dynamic feature analysis module; generate a cross-modal attention layer based on the Transformer attention mechanism; use a gated fusion layer to implement gating signals, cosine similarity calculation, and other calculations; train model weights (Wq, Wk, Wv, Ws, We, Wg, etc.) using historical data, and then deploy the model on a server or edge computing device for real-time inference to complete the training and deployment of the model;

[0175] Beneficial effects: Through advanced algorithms such as LSTM, CNN, and wavelet transform, deep spatiotemporal patterns and dynamic features that are difficult for the human eye to detect are extracted from the raw data; the cross-modal attention mechanism and gating fusion mechanism are no longer simple data superposition, but allow data from different modalities to "dialogue" and intelligently weigh their influence, which greatly improves the accuracy of state assessment. This is the most significant technical highlight of this invention compared with traditional methods.

[0176] It should be noted that the fine-grained cross-fusion layer is a hierarchical, serial-parallel combined deep neural network architecture. Its core idea is to progressively refine the feature extraction, interaction, and fusion of multi-source heterogeneous data from shallow to deep and from the inside out. The 8-layer structure is not a simple stacking, but is designed according to the data modality and processing objectives, forming a complete structure of "intra-modal enhancement → cross-modal interaction → inter-modal aggregation → context modulation → physical constraint fusion".

[0177] The 8-layer connection relationship and specific composition:

[0178] Overall connectivity: The 8-layer fusion layer adopts a data flow approach that is mainly serial and partially parallel. The output of the previous layer serves as one of the inputs of the next layer, progressing layer by layer and gradually refining the feature expression. At the same time, at specific levels (such as the fourth and fifth layers), new external data sources (such as original data and intermediate features from other modalities) will be introduced for parallel fusion.

[0179] Layer 1: Intramodal multi-scale feature enhancement layer

[0180] Connection relationship: This layer is the starting layer, and it directly receives the original monitoring data sequence of the first breakwater surface structure in parallel. and environmental monitoring basic data sequence ;

[0181] Specific components and functions: It consists of three sub-modules: LSTM, CNN, and MLP, connected in series.

[0182] LSTM: Captures long-term temporal dependencies in structural data (corrosion, displacement, cracks);

[0183] CNN: Extracting local spatial correlations and short-term patterns between signals from different sensors;

[0184] MLP: Performs nonlinear transformation and dimensionality reduction on the fused spatiotemporal features to output a multi-scale spatiotemporal feature vector F_s of the structure;

[0185] Environment branch: Consists of two sub-modules, GRU and Wavelet, connected in parallel or in a lightweight series;

[0186] GRU: Captures dynamic temporal changes in environmental data (waves, ships);

[0187] Wavelet (wavelet transform): provides time-frequency domain analysis, accurately extracts the frequency domain features and transient change features of environmental signals, and outputs the dynamic feature vector F_e of the environment;

[0188] Objective: To elevate raw data into more expressive, fine-grained feature vectors within each modality, in preparation for subsequent cross-modal interactions;

[0189] Layer 2: Cross-modal attention interaction layer

[0190] Connection relationship: This layer receives the outputs Fs and Fe of the first layer as inputs;

[0191] Specific components and functions:

[0192] At its core is a cross-modal attention mechanism.

[0193] It contains three sets of learnable weight matrices: Wq, Wk, Wv;

[0194] Calculation process:

[0195] Using structural features Fs as the query and environmental features Fe as the key and value, the attention matrix Ase is calculated to obtain fine-grained structural features. ;

[0196] Using environmental features Fe as the query and structural features Fs as the key and value (using a different set of weights), the attention matrix Aes is calculated to obtain fine-grained environmental features. ;

[0197] Objective: To enable a "two-way dialogue" between the characteristics of the structure and the environment, and to uncover the deep correlation between them; for example, to identify "which wave mode has the greatest impact on structural vibration";

[0198] Layer 3: Gated Multimodal Aggregation Layer

[0199] Connection relationship: This layer receives the output of the second layer. and It also receives the initial monitoring coefficients Km and Kt, which are preliminarily calculated by other modules.

[0200] Specific components and functions:

[0201] Pooling layer (Pool) and Perform global pooling to obtain a generalized feature representation;

[0202] Fully connected layer + activation function: calculate the refinement coefficients and ;

[0203] The gate mechanism consists of a small neural network (sigmoid(W_g * [pool( pool( This is achieved by generating a gating value between 0 and 1 to weigh the relative impacts of structure and environment on the levee system number.

[0204] Alignment calculation (align), using cosine similarity to calculate pool( ) and pool( Consistency in direction;

[0205] Aggregate calculation, using gating signals and alignment, merges the primary coefficients Km and Kt with the refined coefficients. and Perform adaptive fusion and output corrected high-level coefficients. and ;

[0206] The function of this layer is to intelligently integrate multimodal information and refine the initially calculated safety factor.

[0207] Layer 4: Context-aware layer

[0208] The connection relationships are such that this layer reintroduces the original short-term and long-term environmental data sequences. , Its output acts on the structural feature processing path after the first layer;

[0209] Specific components and functions:

[0210] TCN (Temporal Convolutional Network) is specifically designed to process short-term, high-resolution environmental sequences and capture recent fluctuations.

[0211] Transformer processes long-term, low-resolution environmental sequences to uncover historical trends and periodic patterns.

[0212] Attention fusion combines the outputs of TCN and Transformer to generate a comprehensive context encoding vector Cenv.

[0213] The purpose of this layer is to provide richer environmental background information across multiple time scales, thus providing contextual support for structural safety analysis.

[0214] Layer 5: Adaptive Weighted Fusion Layer

[0215] The connection relationship is such that this layer receives the output Cenv from layer 4 and the original structure data. And the structural features / coefficients after the first layer;

[0216] Specific components and functions:

[0217] LSTM converts the original structural data By fusing with the environmental context Cenv, Ksfine, an environmentally-aware structural feature is generated;

[0218] Adaptive weight generation, which calculates a weight using the sigmoid function. This is used to balance the importance of the original structural coefficient Ks and the environmental perception coefficient Ksfine;

[0219] The weighted output is Ks;

[0220] The function of this layer is to enable dynamic and adaptive modulation of environmental information for structural safety assessment.

[0221] Layer 6: Spatial Attention (ROI) Layer

[0222] The connection relationship is that this layer acts on the surface structure area of ​​the breakwater monitored by the UAV, and receives the original characteristics (location L, crack T, block displacement P) of all monitoring points.

[0223] Specific components and functions:

[0224] A feature vector is calculated for each monitoring unit using MLP;

[0225] The feature vectors of all units are normalized by Softmax to generate a spatial attention weight distribution. Units with higher weight values ​​represent higher risks and require more attention.

[0226] The purpose of this layer is to enable the model to automatically focus on the areas with the highest structural risks on the breakwater surface, thereby achieving optimal allocation of monitoring resources.

[0227] Layer 7: Physical Constraint Fusion Layer

[0228] This layer introduces the vibration characteristics Xvib from the surface structure monitoring of the first breakwater and interacts with the wave characteristics PfluctPfluct from the surface structure monitoring of the second breakwater.

[0229] Specific components and functions:

[0230] The correlation matrix Avibp between the vibration monitored on the surface structure of the first breakwater and the fluctuation monitored on the surface structure of the second breakwater was calculated using cross-attention.

[0231] The correlation matrix is ​​mapped to a physical influence increment through MLP.

[0232] This increment is added to the initial safety factor of each monitoring unit through additive fusion to obtain... .

[0233] The function of this layer is to embed the causal relationships of the physical world into the model, so that the evaluation results conform to physical laws.

[0234] Layer 8: Global Aggregation Output Layer

[0235] This layer receives the spatial attention weights from layer 6 and the fine-grained security factor for each cell in layer 7.

[0236] Specific components and functions:

[0237] The fine-grained safety factor for each unit is calculated by weighted summation. Its spatial attention weight Multiplying and summing the results yields the final safety monitoring coefficient for the entire surface structural area of ​​the breakwater. ;

[0238] The role of this layer is to comprehensively assess the risk status of all monitoring nodes and output a final quantitative indicator that represents the overall safety status of the breakwater, taking into account both spatial importance and physical constraints.

[0239] Through these eight layers of progressive processing, the model transforms raw data into highly refined and deeply integrated intelligent assessment results, greatly improving the accuracy and reliability of monitoring. This is one of the technical highlights of this invention.

[0240] In the above embodiments, specifically, S22 includes:

[0241] S221. Spatiotemporal correlation analysis is performed on the monitoring data of the surface structure of the first breakwater using the spatiotemporal feature extraction module.

[0242] S222. Based on the analysis results, obtain the spatiotemporal characteristic vectors of the current structural settlement and displacement values;

[0243] S223. Obtain the allowable structural settlement threshold Dd0 and the allowable structural displacement threshold Vf0 from the database;

[0244] S224. The first structural safety monitoring coefficient Ks is obtained by multimodal calculation using the spatiotemporal feature vector and the corresponding threshold:

[0245] Ks = β * (Dd / Dd0) + γ * (Vf / Vf0), where β and γ are weighting coefficients, and β+γ=1, Dd represents the structural settlement characteristic value, and Vf represents the structural displacement characteristic value;

[0246] S225. A fine-grained contextual awareness is introduced through the fourth fine-grained cross-fusion layer to perform the fourth fusion analysis and calculate the contextual vector:

[0247] Extracting short-term fluctuation characteristics (Eshort) and long-term trend characteristics (Elong) from environmental data:

[0248] Eshort = TCN( ), where TCN stands for Temporal Convolutional Network, used to process recent data; This represents a short-term environmental data sequence;

[0249] Elong = Transformer( Here, Transformer represents the Transformer neural network, which is used to process long-term historical data; Represents a long-term environmental data sequence;

[0250] Calculate the environment context vector Cenv:

[0251] Cenv = Attention([Eshort, Elong]), where Attention represents the attention function;

[0252] S226. Generate the fine-grained first-structure safety monitoring coefficient Ksfine based on the environmental context vector. Perform a fifth fusion analysis on the fine-grained first-structure safety monitoring coefficient through the fifth fine-grained cross-fusion layer to achieve adaptive weighting, and obtain the final first-structure safety monitoring coefficient:

[0253] Ksfine = LSTM([ ,Cenv]);

[0254] = sigmoid([ , Cenv]), where Indicates adaptive weights;

[0255] Ks = * Ks + (1 - Ksfine, where Ks represents the final first structural safety monitoring coefficient.

[0256] It should be noted that environmental context awareness is introduced into structural safety monitoring. TCN is used to process environmental data from the past few hours to capture short-term shocks; Transformer is used to process data from the past few weeks or months to grasp long-term patterns and cycles; attention mechanism is used to fuse long-term and short-term environmental features to generate a comprehensive "environmental context vector" Cenv; LSTM is used to fuse structural data with Cenv to generate environmentally modulated structural safety coefficients, and adaptive weights are used to fuse them with the original coefficients.

[0257] Through the above steps, structural safety assessments are no longer isolated but take into account the real-world environment in which they occur. For example, the same structural displacement carries completely different risk levels in calm weather and in stormy conditions; adaptive weighting allows the system to intelligently determine whether to trust the results of the traditional model or the results of the environment-aware model in the current environment, leading to more scientific decision-making.

[0258] In the above embodiments, specifically, step S25 includes:

[0259] S253. Multimodal feature extraction is performed on the monitoring data of the surface structure of the second breakwater using a multi-source data fusion module to generate multimodal features; crack values ​​and block displacement values.

[0260] S254. Obtain the crack threshold Tt0 and block displacement threshold Pp0 from the database;

[0261] S255. The initial value of the second structural safety monitoring coefficient Kt is obtained by calculating the multimodal features and corresponding safety thresholds using a BP deep neural network:

[0262] Kt = δ1 * (Tt / Tt0) + δ2 * (Pp / Pp0), where δ1 and δ2 are weights, Tt represents the crack characteristic value, and Pp represents the block displacement characteristic value;

[0263] It should be noted that Tt includes a crack risk quantification value that incorporates spatial distribution, temporal trend, and cross-modal relationships. It is a value that has undergone multimodal fusion and feature engineering, rather than the original crack data.

[0264] Pp contains a quantitative value of block displacement risk based on dynamic characteristics and causal relationships, which is a value obtained through multimodal fusion and feature engineering, rather than the original pressure data;

[0265] The two characteristic risk values, after intelligent fusion analysis, are compared with their corresponding safety thresholds to obtain two normalized risk indicators (ratios). These are then fused into a comprehensive second structural safety monitoring coefficient Kt through a weighted average. This process deeply explores the intrinsic relationship between data and is far more advanced and reliable than simply setting two independent over-limit alarms.

[0266] S256. Set up multimodal fine-grained regions of interest (ROIs) through the sixth fine-grained cross-fusion layer, and perform a sixth fusion analysis using the multimodal fine-grained ROIs:

[0267] For n monitoring points on the breakwater, based on their spatial location ,crack Block displacement Calculate attention weights:

[0268] = exp(MLP([ , , ])) / ,in This represents the attention weight corresponding to the i-th monitoring point;

[0269] S257. Seventh fusion analysis was performed using the seventh fine-grained cross-fusion layer:

[0270] The structural settlement characteristics Xvib from the surface structure monitoring of the first breakwater and the crack development characteristics Pfluct from the surface structure monitoring of the second breakwater are cross-modal aligned, and the alignment matrix Avibp is calculated:

[0271] Avibp = crossattention(Xvib, Pfluct), where crossattention represents the cross attention function;

[0272] Calculate the surface safety factor Ktfine of the fine-grained second breakwater under physical constraints:

[0273] = + bf1 * MLP( ), where bf1 is the physical influence coefficient, and i represents the i-th monitoring point, Ktfine represents the fine-grained surface safety factor of the second breakwater at the i-th monitoring point; This represents the initial value of the second structural safety monitoring coefficient Kt at the i-th monitoring point; Avibp represents the alignment matrix of the i-th monitoring point;

[0274] S258. Eighth fusion analysis is performed through the eighth fine-grained cross-fusion layer:

[0275] = ,in This represents the final second structural safety monitoring coefficient.

[0276] In the above embodiments, specifically, S3 includes:

[0277] S31. Obtain historical operational data of the breakwater and use the historical operational data of the breakwater to establish a digital twin model of the breakwater through BIM modeling tools;

[0278] S32. Obtain the following thresholds through digital twin models: allowable displacement threshold, crack propagation threshold, wave intensity threshold, hazardous chemical vessel capacity threshold, number of vessels to be moored, allowable settlement threshold, crack threshold, and block displacement threshold.

[0279] S33. Obtain the allowable settlement threshold based on the basic data of breakwater health monitoring;

[0280] S34. The structural settlement threshold and structural displacement threshold are calculated using multimodal methods to obtain the basic threshold Ks0base of the first structural safety monitoring coefficient.

[0281] S35. Dynamic feature calculations are performed using wave intensity threshold, hazardous chemical vessel capacity threshold, and pre-berth vessel number threshold to obtain the basic threshold for environmental risk monitoring coefficient Ke0base.

[0282] S36. Calculate the underwater safety performance characteristics of the scour threshold and collapse threshold of the embankment to obtain the basic threshold Km0base of the underwater safety structure monitoring coefficient of the embankment.

[0283] S37. The crack threshold and block displacement threshold are calculated by multi-source data fusion to obtain the basic threshold Kt0base of the second structural safety monitoring coefficient.

[0284] S38. Adjust the first structural safety monitoring coefficient threshold and the environmental risk monitoring coefficient threshold based on the fine-grained structural features and the fine-grained environmental features, respectively, to generate the first dynamic first structural safety monitoring coefficient threshold and the first environmental risk monitoring threshold.

[0285] Calculate fine-grained structural features Historical distribution statistics are used to dynamically adjust the threshold of the first structural safety monitoring coefficient.

[0286] Ks0 = Ks0base * (1 + ηs * (|| - μs|| / σs)), where ηs is the sensitivity coefficient, μs represents the mean, σs represents the standard deviation, and Ks0 represents the threshold of the first dynamic first structural safety monitoring coefficient;

[0287] Calculate fine-grained environmental features The real-time fluctuation index dynamically adjusts the environmental risk monitoring coefficient threshold.

[0288] Ke0 = Ke0base * (1 + ηe * (std( ) / mean( ))), where ηe represents the environmental fluctuation coefficient, Ke0 represents the first environmental risk monitoring coefficient threshold; std represents the standard deviation function, and mean represents the mean function;

[0289] S39. Generate the structure-environment coupling threshold by coupling the cross-modal attention matrix Ase from S27 and the first dynamic first structural safety monitoring coefficient threshold:

[0290] Kse0 = (Ase * Ke0) + ((1 - Ase) * Ks0);

[0291] Using the gate signal (gate), alignment (align), and first environmental risk monitoring coefficient threshold of S28, a levee-environment association threshold is generated:

[0292] Kmt0 = gate * (Km0base + Kt0base) + (1 - gate) * align * max(Km0base,Kt0base);

[0293] S40. Define the basic thresholds Ks0base, Ke0base, Km0base, Kt0base, the dynamically adjusted thresholds Ks0, Ke0, and the coupling thresholds Kse0, Kmt0 together as the breakwater health monitoring threshold data.

[0294] It should be noted that, in step S4, the specific steps for issuing a safety risk warning based on the breakwater health monitoring analysis data and the breakwater health monitoring threshold data are as follows:

[0295] S41. Obtain the breakwater health monitoring analysis data and breakwater health monitoring threshold data respectively, and obtain the underwater safety structure monitoring coefficient of the breakwater body based on the breakwater health monitoring analysis data. Environmental risk monitoring coefficient First structural safety monitoring coefficient Second structural safety monitoring coefficient and fine-grained structural features Based on the breakwater health monitoring threshold data, the following thresholds were obtained: underwater safety structure monitoring coefficient threshold Km0, environmental risk monitoring coefficient threshold Ke0, first structural safety monitoring coefficient threshold Ks0, second structural safety monitoring coefficient threshold Kt0, and coupling thresholds Kse0 and Kmt0.

[0296] S42. Primary early warning based on coefficient threshold:

[0297] When the first structural safety monitoring coefficient If the breakwater structure safety warning is issued when the value is greater than or equal to the first structural safety monitoring coefficient threshold Ks0;

[0298] When environmental risk monitoring coefficient If the environmental risk monitoring coefficient threshold Ke0 is greater than or equal to the environmental risk monitoring coefficient threshold, an environmental risk warning will be issued.

[0299] When the underwater safety structure monitoring coefficient of the dike If the underwater safety structure monitoring coefficient Km0 of the dike is greater than or equal to the threshold value of the dike body, a safety warning for the dike body will be issued.

[0300] When the second structural safety monitoring coefficient If the value is greater than or equal to the threshold Kt0 of the second structural safety monitoring coefficient, a safety warning for the surface structure of the breakwater will be issued.

[0301] S43. Intermediate early warning based on fine-grained feature anomaly patterns:

[0302] Calculate fine-grained structural features The real-time Mahalanobis distance Ds = ( - μs)^T Σs^{-1} ( - μs), where Σs is the mean and covariance of the historical characteristic distribution;

[0303] When Ds > χ^2(p) (p is the feature dimension, χ^2 is the chi-square critical value), even if < Ks0 also issues early warnings of structural anomalies;

[0304] Calculate fine-grained environmental features The mutation index Ce = ||d / dt||;

[0305] When Ce > Me (preset mutation threshold), even <Ke0, also issued an environmental mutation warning;

[0306] S44: Advanced early warning (correlation warning) based on cross-fusion coupling relationship:

[0307] when / Ks0 + When / Ke0 > Kse0, issue a warning about the risk of structure-environment coupling.

[0308] A monitoring system for breakwaters at hazardous chemical terminals based on multi-source data integration, such as Figure 2 As shown, the system includes a monitoring data acquisition module, a multimodal analysis module, a threshold analysis module, and an early warning analysis module;

[0309] The monitoring data acquisition module acquires the current structural settlement value and structural displacement value to obtain the surface structure monitoring data of the first breakwater, acquires the real-time wave intensity value, the total number of hazardous chemical vessels and the number of hazardous chemical vessels preparing to dock, acquires the basic environmental monitoring data, acquires the breakwater scour value and collapse value to obtain the underwater structure monitoring data of the breakwater, acquires the cracks and block displacements on the surface of the breakwater, and acquires the surface structure monitoring data of the second breakwater. The surface structure monitoring data of the first breakwater, the basic environmental monitoring data, the underwater structure monitoring data of the breakwater, and the surface structure monitoring data of the second breakwater are defined as the basic health monitoring data of the breakwater.

[0310] The multimodal analysis module extracts spatiotemporal features from the surface structure monitoring data of the first breakwater and calculates the first structural safety monitoring coefficient. It then performs dynamic feature analysis on the environmental monitoring data using a multimodal feature analysis model to calculate the environmental risk monitoring coefficient. Finally, it performs underwater performance structural feature analysis on the underwater structure monitoring data of the breakwater using the same model to calculate the underwater safety structural monitoring coefficient. Finally, it performs multi-source data fusion analysis on the surface structure monitoring data of the second breakwater using the same model to calculate the second structural safety monitoring coefficient. The first structural safety monitoring coefficient, the environmental risk monitoring coefficient, the underwater safety structural monitoring coefficient, and the second structural safety monitoring coefficient are defined as breakwater health monitoring analysis data.

[0311] The threshold analysis module establishes a digital twin model of the breakwater, and obtains the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater through the digital twin model. The underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold are defined as breakwater health monitoring threshold data.

[0312] The early warning analysis module provides safety risk warnings based on breakwater health monitoring analysis data and breakwater health monitoring threshold data.

[0313] It should be understood that the above embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations based on the present invention; any variations and modifications made by those skilled in the art through the present invention without making pioneering innovations are all within the protection scope of the present invention.

Claims

1. A monitoring method for breakwaters at hazardous chemical terminals based on multi-source data integration, characterized in that, include: S1. Obtain the current structural settlement value and structural displacement value to obtain the surface structure monitoring data of the first breakwater, obtain the real-time wave intensity value, the total number of hazardous chemical vessels and the number of hazardous chemical vessels preparing to dock, obtain the basic environmental monitoring data, obtain the breakwater scour value and collapse value to obtain the underwater structure monitoring data of the breakwater, obtain the cracks and block displacements on the surface of the breakwater, and obtain the surface structure monitoring data of the second breakwater. S2. Spatiotemporal features are extracted from the surface structure monitoring data of the first breakwater using a multimodal feature analysis model to calculate the first structural safety monitoring coefficient. Dynamic features are analyzed from the environmental monitoring data using a multimodal feature analysis model to calculate the environmental risk monitoring coefficient. Underwater performance structural features are analyzed from the underwater structure monitoring data of the breakwater using a multimodal feature analysis model to calculate the underwater safety structure monitoring coefficient of the breakwater. Multi-source data fusion analysis is performed from the surface structure monitoring data of the second breakwater using a multimodal feature analysis model to calculate the second structural safety monitoring coefficient. The first structural safety monitoring coefficient, the environmental risk monitoring coefficient, the underwater safety structure monitoring coefficient of the breakwater, and the second structural safety monitoring coefficient are defined as breakwater health monitoring analysis data. S3. Establish a digital twin model of the breakwater, and obtain the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater through the digital twin model of the breakwater. Define the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater as the breakwater health monitoring threshold data. S4. Conduct safety risk warnings based on breakwater health monitoring analysis data and breakwater health monitoring threshold data.

2. The method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration according to claim 1, characterized in that, In step S1, acquiring monitoring data on the surface structure of the first breakwater includes: By deploying N monitoring points on the breakwater, each equipped with a GNSS receiver, satellite signals are received in real time, the three-dimensional coordinates of each monitoring point are calculated, and the displacement values ​​of the breakwater monitoring points are obtained based on the three-dimensional coordinates, thus obtaining the structural displacement values. Within the current monitoring cycle, k monitoring time nodes are determined for the N monitoring points, with a standard monitoring duration between each two consecutive monitoring time nodes. The structural settlement values ​​corresponding to the k monitoring time nodes are obtained through the GNSS receiver and marked as the first settlement rate to the kth settlement rate. A weighted average is calculated for the first settlement rate to the kth settlement rate to obtain the structural settlement value. The current structural settlement value and structural displacement value are defined as the surface structural monitoring data of the first breakwater.

3. The method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration according to claim 1, characterized in that, In step S1, acquiring basic environmental monitoring data includes: The wave intensity values ​​corresponding to the breakwater are obtained in real time by wave sensors to obtain real-time wave intensity values; the total number of hazardous chemical vessels currently docked at the wharf and the number of hazardous chemical vessels that have not yet docked at the wharf are obtained by camera image recognition algorithms. The total number of hazardous chemical vessels currently docked at the wharf is marked as the total number of hazardous chemical vessels, and the number of hazardous chemical vessels that have not yet docked at the wharf is marked as the number of hazardous chemical vessels that are expected to dock; the real-time wave intensity values, the total number of hazardous chemical vessels, and the number of hazardous chemical vessels that are expected to dock are defined as basic environmental monitoring data.

4. The method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration according to claim 1, characterized in that, In step S1, the acquisition of monitoring data on the surface structure of the second breakwater specifically includes: Based on navigation technology, the initial navigation path of the drone is determined, and a high-definition camera and a 3D LiDAR are configured on the drone. The drone flew along the initial navigation path and collected basic information and initial image data of the breakwater surface structure using its onboard high-definition camera. Based on the basic information, the key points of the breakwater surface structure were collected in the region to obtain the coordinates of several key points in the region. The initial image data, the basic information, and the coordinates of key points in several regions are input into the navigation parameter planning model, and the navigation path trajectory set of the UAV is output. The drone periodically acquires images according to the navigation path trajectory set. The images captured by the drone are analyzed using image recognition algorithms to identify the crack values ​​and block displacement values ​​on the surface of the breakwater. The crack values ​​and block displacement values ​​are defined as the second breakwater surface structure monitoring data.

5. The method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration according to claim 1, characterized in that, S2 includes: A database is created, storing data including allowable settlement thresholds and allowable structural displacement thresholds; a multimodal feature analysis model is constructed, comprising a spatiotemporal feature extraction module, a dynamic feature analysis module, an underwater performance structure analysis module, a multi-source data fusion module, and a fine-grained cross-fusion layer, with a total of 8 fine-grained cross-fusion layers. S21. Acquire monitoring data of the surface structure of the first breakwater, basic environmental monitoring data, monitoring data of the underwater structure of the breakwater, and monitoring data of the surface structure of the second breakwater. S22. Analyze the monitoring data of the surface structure of the first breakwater through the spatiotemporal feature extraction module to obtain the safety monitoring coefficient Ks of the first structure; S23. Analyze the basic environmental monitoring data through the dynamic feature analysis module to obtain the environmental risk monitoring coefficient Ke; S24. Analyze the underwater structure monitoring data of the breakwater using the underwater performance structure analysis module to obtain the underwater safety structure monitoring coefficient Km of the breakwater. S25. The monitoring data of the surface structure of the second breakwater is analyzed by the multi-source data fusion module to obtain the safety monitoring coefficient Kt of the second structure; S26. Intramodal multi-scale feature enhancement is achieved through the first fine-grained cross-fusion layer for the first fusion analysis: Multi-scale spatiotemporal characteristics were extracted from the monitoring data of the surface structure of the first breakwater: Fs = [ , , ..., = MLP(CNN(LSTM( ))),in, The original data sequence for monitoring the breakwater structure, The dimensions are [T1, Ds], where T1 is the time step and Ds is the structural feature dimension; LSTM represents Long Short-Term Memory Network, used to capture the time dependence of the monitoring data of the surface structure of the first breakwater; CNN represents a one-dimensional convolutional neural network, used to extract local patterns of the monitoring data of the surface structure of the first breakwater; MLP represents Multilayer Perceptron, used for feature transformation and dimensionality reduction; Fs represents multi-scale spatiotemporal feature vector. This represents the nth element of the multi-scale spatiotemporal eigenvector; Extract multi-resolution dynamic features from basic environmental monitoring data: Fe = [ , , ..., ] = Wavelet(GRU( )),in, The environmental monitoring basic data sequence has the dimension [T, De], where De is the environmental feature dimension; GRU is a gated recurrent unit neural network; Wavelet is a wavelet transform used to extract time-frequency features; Fe represents a multi-resolution dynamic feature vector. This represents the m-th element of the multi-resolution dynamic feature vector; S27. Cross-modal attention feature interaction is achieved through a second fusion analysis using a second fine-grained cross-fusion layer. The input of the second fine-grained cross-fusion layer is the output of the first fine-grained cross-fusion layer. Construct the structure-environment cross-modal attention matrix Ase to calculate fine-grained structural features under the influence of environmental conditions. : Ase = softmax((Wq*Fs)*(Wk*Fe) T / qqq); = Fs + Ase * (Wv * Fe); Where, Wq ∈ , Wk ∈ , Wv ∈ Wq, Wk, and Wv represent the learnable weight matrices, d represents the hidden layer dimension, qqq represents the scaling factor, T represents the matrix transpose, and R represents the set of real numbers. Calculate the fine-grained environmental characteristics under the influence of structural state. : Aes = , where Aes represents the transpose of Ase; = Fe + Aes * (Wv * Fs); S28, Based on cross-fusion fine-grained structural features and fine-grained environmental characteristics The monitoring coefficients are reconstructed through a third fine-grained cross-fusion layer, and a third fusion analysis is performed to achieve gated multimodal feature aggregation. = σ( Ws * pool( ) + bs ); = σ( We * pool( ) + be ); gate = sigmoid( Wg * [pool( ), pool( )] + bg); = Km + λ * gate * ( + ); align = cosine_similarity( pool( ), pool( )); = Kt + μ * align * ( + ); Where Ws represents the pooled structural feature pool ( Mapped to the first structural safety monitoring coefficient The weight matrix; We represents the pooled environmental features pooled together. Mapped to environmental risk monitoring coefficient The weight matrix; Wg represents the weight matrix used to combine the spliced ​​structure with environmental features [pool( pool( The weight matrix mapped to the gate signal; bs, be, bg represent the bias vectors of the learnable model; gate represents the gate signal; pool(·) represents the pooling function; σ(·) represents the activation function; λ represents the fusion intensity coefficient; μ represents the alignment gain coefficient; align represents the feature alignment degree; This represents the refined first structural safety monitoring coefficient; This represents the refined environmental risk monitoring coefficient; This represents the modified underwater safety structure monitoring coefficient of the dike. This represents the corrected safety monitoring coefficient for the second structure; cosine_similarity represents the cosine similarity function. S29. Modify the underwater safety structure monitoring coefficient of the dike. Detailed environmental risk monitoring coefficients The refined first structural safety monitoring coefficient Corrected second structural safety monitoring coefficient and all fine-grained structural features and fine-grained environmental characteristics They are collectively defined as breakwater health monitoring and analysis data.

6. The method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration according to claim 5, characterized in that, S22 includes: S221. Spatiotemporal correlation analysis is performed on the monitoring data of the surface structure of the first breakwater using the spatiotemporal feature extraction module. S222. Based on the analysis results, obtain the spatiotemporal characteristic vectors of the current structural settlement and displacement values; S223. Obtain the allowable structural settlement threshold Dd0 and the allowable structural displacement threshold Vf0 from the database; S224. Obtain the initial value Ks0 of the first structural safety monitoring coefficient by multimodal calculation using the spatiotemporal feature vector and the corresponding threshold: Ks0 = β * (Dd / Dd0) + γ * (Vf / Vf0), where β and γ are weighting coefficients, and β+γ=1, Dd represents the structural settlement characteristic value, and Vf represents the structural displacement characteristic value; S225. A fine-grained contextual awareness is introduced through the fourth fine-grained cross-fusion layer to perform the fourth fusion analysis and calculate the contextual vector: Extracting short-term fluctuation characteristics (Eshort) and long-term trend characteristics (Elong) from environmental data: Eshort = TCN( ), where TCN stands for Temporal Convolutional Network, used to process recent data; This represents a short-term environmental data sequence; Elong = Transformer( Here, Transformer represents the Transformer neural network, which is used to process long-term historical data; Represents a long-term environmental data sequence; Calculate the environment context vector Cenv: Cenv = Attention([Eshort, Elong]), where Attention represents the attention function; S226. Generate the fine-grained first-structure safety monitoring coefficient Ksfine based on the environmental context vector. Perform a fifth fusion analysis on the fine-grained first-structure safety monitoring coefficient through the fifth fine-grained cross-fusion layer to achieve adaptive weighting, and obtain the final first-structure safety monitoring coefficient: Ksfine = LSTM([ ,Cenv]); = sigmoid([ , Cenv]), where Indicates adaptive weights; Ks = * Ks0 + (1 - Ksfine, where Ks represents the first structural safety monitoring coefficient.

7. The method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration according to claim 5, characterized in that, S25 includes: S253. Multimodal features are extracted from the monitoring data of the surface structure of the second breakwater through the multi-source data fusion module to generate multimodal features; S254. Obtain the crack threshold Tt0 and block displacement threshold Pp0 from the database; S255. The initial value Kt0 of the second structural safety monitoring coefficient is obtained by calculating the multimodal features and corresponding safety thresholds using a BP deep neural network: Kt0 = δ1 * (Tt / Tt0) + δ2 * (Pp / Pp0), where δ1 and δ2 are weights, Tt represents the crack characteristic value, and Pp represents the block displacement characteristic value; S256. Set up multimodal fine-grained regions of interest (ROIs) through the sixth fine-grained cross-fusion layer, and perform a sixth fusion analysis using the multimodal fine-grained ROIs: For n monitoring points on the breakwater, based on their spatial location ,crack Block displacement Calculate attention weights: = exp(MLP([ , , ])) / ,in This represents the attention weight corresponding to the i-th monitoring point; S257. Seventh fusion analysis was performed using the seventh fine-grained cross-fusion layer: The structural settlement characteristics Xvib from the surface structure monitoring of the first breakwater and the crack development characteristics Pfluct from the surface structure monitoring of the second breakwater are cross-modal aligned, and the alignment matrix Avibp is calculated: Avibp = crossattention(Xvib, Pfluct), where crossattention represents the cross attention function; Calculate the surface safety factor Ktfine of the fine-grained second breakwater under physical constraints: = + bf1 * MLP( ), where bf1 is the physical influence coefficient, and i represents the i-th monitoring point, Ktfine represents the fine-grained surface safety factor of the second breakwater at the i-th monitoring point; Kt0 represents the initial value of the second structural safety monitoring coefficient at the i-th monitoring point; Avibp represents the alignment matrix of the i-th monitoring point; S258. Eighth fusion analysis is performed through the eighth fine-grained cross-fusion layer: = , where Kt represents the second structural safety monitoring coefficient.

8. The method for monitoring breakwaters at hazardous chemical terminals based on multi-source data integration according to claim 5, characterized in that, The S3 includes: S31. Obtain historical operational data of the breakwater and use the historical operational data of the breakwater to establish a digital twin model of the breakwater through BIM modeling tools; S32. Obtain the following thresholds through digital twin models: allowable displacement threshold, crack propagation threshold, wave intensity threshold, hazardous chemical vessel capacity threshold, number of vessels to be moored, allowable settlement threshold, crack threshold, and block displacement threshold. S34. The structural settlement threshold and structural displacement threshold are calculated using multimodal methods to obtain the basic threshold Ks0base of the first structural safety monitoring coefficient. S35. Dynamic feature calculations are performed using wave intensity threshold, hazardous chemical vessel capacity threshold, and pre-berth vessel number threshold to obtain the basic threshold for environmental risk monitoring coefficient Ke0base. S36. Calculate the underwater safety performance characteristics of the scour threshold and collapse threshold of the embankment to obtain the basic threshold Km0base of the underwater safety structure monitoring coefficient of the embankment. S37. The crack threshold and block displacement threshold are calculated by multi-source data fusion to obtain the basic threshold Kt0base of the second structural safety monitoring coefficient. S38. Adjust the first structural safety monitoring coefficient threshold and the environmental risk monitoring coefficient threshold based on the fine-grained structural features and the fine-grained environmental features, respectively, to generate the first dynamic first structural safety monitoring coefficient threshold and the first environmental risk monitoring threshold. Calculate fine-grained structural features Historical distribution statistics are used to dynamically adjust the threshold of the first structural safety monitoring coefficient. Ks0 = Ks0base * (1 + ηs * (|| - μs|| / σs)), where ηs is the sensitivity coefficient, μs represents the mean, σs represents the standard deviation, and Ks0 represents the threshold of the first dynamic first structural safety monitoring coefficient; Calculate fine-grained environmental features The real-time fluctuation index dynamically adjusts the environmental risk monitoring coefficient threshold. Ke0 = Ke0base * (1 + ηe * (std( ) / mean( ))), where ηe represents the environmental fluctuation coefficient, Ke0 represents the first environmental risk monitoring coefficient threshold; std represents the standard deviation function, and mean represents the mean function; S39. Generate the structure-environment coupling threshold by coupling the cross-modal attention matrix Ase from S27 and the first dynamic first structural safety monitoring coefficient threshold: Kse0 = (Ase * Ke0) + ((1 - Ase) * Ks0); Using the gate signal (gate), alignment (align), and first environmental risk monitoring coefficient threshold of S28, a levee-environment association threshold is generated: Kmt0 = gate * (Km0base + Kt0base) + (1 - gate) * align * max(Km0base,Kt0base); S40. The basic thresholds Ks0base, Ke0base, Km0base, Kt0base, the dynamically adjusted thresholds Ks0, Ke0, and the coupling thresholds Kse0, Kmt0 are collectively defined as the breakwater health monitoring threshold data.

9. A monitoring system for breakwaters at hazardous chemical terminals based on multi-source data integration, characterized in that, The system is used to perform the method according to any one of claims 1-8, and the system includes a monitoring data acquisition module, a multimodal analysis module, a threshold analysis module, and an early warning analysis module; The monitoring data acquisition module acquires the current structural settlement value and structural displacement value to obtain the surface structure monitoring data of the first breakwater, acquires the real-time wave intensity value, the total number of hazardous chemical vessels and the number of hazardous chemical vessels preparing to dock, obtains the basic environmental monitoring data, acquires the breakwater scour value and collapse value to obtain the underwater structure monitoring data of the breakwater, acquires the cracks and block displacements on the surface of the breakwater, and obtains the surface structure monitoring data of the second breakwater. The multimodal analysis module extracts spatiotemporal features from the surface structure monitoring data of the first breakwater and calculates the first structural safety monitoring coefficient. It then performs dynamic feature analysis on the environmental monitoring data using a multimodal feature analysis model to calculate the environmental risk monitoring coefficient. Finally, it performs underwater performance structural feature analysis on the underwater structure monitoring data of the breakwater using the same model to calculate the underwater safety structural monitoring coefficient. Finally, it performs multi-source data fusion analysis on the surface structure monitoring data of the second breakwater using the same model to calculate the second structural safety monitoring coefficient. The first structural safety monitoring coefficient, the environmental risk monitoring coefficient, the underwater safety structural monitoring coefficient, and the second structural safety monitoring coefficient are defined as breakwater health monitoring analysis data. The threshold analysis module establishes a digital twin model of the breakwater, and obtains the underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold of the breakwater through the digital twin model. The underwater safety structure monitoring coefficient threshold, environmental risk monitoring coefficient threshold, first structural safety monitoring coefficient threshold, and second structural safety monitoring coefficient threshold are defined as breakwater health monitoring threshold data. The early warning analysis module provides safety risk warnings based on breakwater health monitoring analysis data and breakwater health monitoring threshold data.

Images