Port transportation network risk assessment method and system based on topology simulation

Abstract

The present invention discloses a port transportation network risk assessment method and system based on topological simulation, comprising the following steps: collecting historical data on the transportation of dangerous goods in concentrated areas of ports; constructing a dynamic port transportation network model based on transportation routes, performing topological analysis to construct a vulnerability function; fusing the characteristic data of the historical data to obtain fusion features, and obtaining accident rates based on accident records; performing topological simulation on the port transportation network model based on the fusion features and accident rates to obtain a risk function, and obtaining risk assessment data based on the risk function and the vulnerability function; constructing a risk assessment model based on the risk assessment data, optimizing the risk assessment model using a sandcat optimization algorithm, and inputting the data to be evaluated into the risk assessment model to obtain an assessment result. This method not only improves the accuracy of port transportation network risk assessment, but also has good interpretability and can be directly applied to port transportation network risk assessment systems.

Description

Technical Field

[0001] The present invention relates to the technical field of port transportation network risk assessment, and in particular to a port transportation network risk assessment method and system based on topology simulation. Background Art

[0002] With the rapid development of global trade, ports, as important logistics hubs, are responsible for an increasing number of dangerous goods transportation tasks. These dangerous goods include flammable, explosive, toxic and corrosive chemicals, and their safety management during transportation is crucial. However, due to the complexity and dynamic changes of port transportation networks, traditional risk assessment methods can no longer meet current needs. Existing assessment methods often rely on static data and empirical judgments, lack the ability to analyze real-time data and adapt to dynamic changes in the network. Therefore, there is an urgent need for a method that can evaluate the risks of port transportation networks in real time and dynamically to improve the safety and efficiency of port transportation. It is based on this background that the present invention proposes a port transportation network risk assessment method and system based on topology simulation, aiming to achieve accurate assessment of port transportation network risks through advanced data analysis technology and network topology theory. Summary of the Invention

[0003] The purpose of the present invention is to provide a port transportation network risk assessment method and system based on topology simulation.

[0004] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0005] The present invention comprises the following steps:

[0006] Collect historical data on dangerous goods transportation in the port's dangerous goods concentration area, including dangerous goods types, transportation frequency, transportation volume, transportation routes, operation data and accident records;

[0007] Constructing a dynamic port transportation network model based on the transportation routes, performing topological analysis and constructing a vulnerability function;

[0008] Fusing the feature data of the historical data to obtain a fusion feature, and obtaining an accident occurrence rate based on the accident records;

[0009] Performing a topological simulation on the port transportation network model based on the fusion characteristics and the accident incidence rate to obtain a risk function, and obtaining risk assessment data based on the risk function and the vulnerability function;

[0010] A risk assessment model is constructed based on the risk assessment data, the risk assessment model is optimized by using a sandcat optimization algorithm, and the data to be evaluated is input into the risk assessment model to obtain an evaluation result.

[0011] Furthermore, the method for collecting historical data on dangerous goods transportation in the dangerous goods concentration area of the port, including dangerous goods types, transportation frequency, transportation volume, transportation routes, operation data and accident records, includes:

[0012] Historical data is obtained through port operation records, maritime department reports and safety supervision archives. The transportation routes include the geographical location, transportation connections and transportation processes of equipment and facilities for the transportation of dangerous goods. Operational data includes cargo loading and unloading time, equipment usage time, capacity and processing speed of equipment and facilities, as well as weather data, transportation time, congestion and transportation capacity of transportation connections.

[0013] Furthermore, a method for constructing a dynamic port transportation network model based on the transportation route and performing topological analysis to construct a vulnerability function includes:

[0014] The nodes of the dynamic port transportation network model are equipment and facilities. The node attributes include the capacity and processing speed of the nodes. The edges represent the transportation connections between nodes. The attributes of the edges include transportation time and transportation capacity. The nodes, edges and node attributes of the dynamic port transportation network model are dynamically updated as the transportation routes change. The direction of the edges is the same as the flow direction of dangerous goods.

[0015] The vulnerability function of the node is obtained by analyzing the node's clustering coefficient, betweenness, degree, degree distribution and node weight:

[0016]

[0017]

[0018] Among them, V i,cr is the vulnerability of node i, δ1, δ2, δ3, δ4 and δ5 are the influence weights of clustering coefficient, betweenness, degree, degree distribution and node weight respectively, C i , BC i,v 、k i 、P(k i ) and W i are the clustering coefficient, betweenness, degree, degree distribution and node weight of node i respectively, and are the influence weights of node capacity and processing speed respectively, C i and S i are the capacity and processing speed of node i, C max and S max are the maximum capacity and maximum processing speed among all nodes respectively;

[0019] The edge vulnerability function is obtained by performing edge vulnerability analysis based on edge attributes:

[0020]

[0021] Among them, V j,cr is the vulnerability of edge j, Υ T , Υ C and Y i are the impact weights of transportation time, transportation capacity and vulnerability of adjacent nodes, T j 、C j 、V j,n1,cr and V j,n2,cr are the transportation time, transportation capacity, vulnerability of adjacent node 1, and vulnerability of adjacent node 2 of edge j respectively.

[0022] Furthermore, the method of fusing the feature data of the historical data to obtain fused features includes:

[0023] The types of dangerous goods are converted into numerical vectors, and the transport frequency, transport volume and operation data are normalized and preprocessed.

[0024] Arrange and organize the dangerous goods types, transportation frequency, transportation volume and operation data in chronological order to form time series data that can be processed by signals, and decompose the time series data:

[0025]

[0026] in is the Dirac function, u k and ω k are the kth signal component after decomposition and the corresponding center frequency, ω k and u k are the center frequency and signal component of the kth component after decomposition, K is the number of components after decomposition, is the partial derivative of the signal f with respect to the time step t, * represents the convolution operation, π is the pi, e is the base of the natural logarithm, t is the current time step, j is the imaginary unit, st represents the constraint condition, f is the signal to be decomposed, is the time series data,

[0027] By iteratively optimizing the decomposition algorithm, the components and center frequencies are continuously adjusted to minimize the augmented Lagrangian function, which is:

[0028]

[0029] in, is the quadratic penalty factor, λ is the Lagrange multiplier, λ(t) is the Lagrange multiplier of the current time step,

[0030] The alternating direction multiplier method is introduced and combined with Fourier isometric transform to find the solution of the augmented Lagrangian function until and u for the nth and n+1th iterations respectively k , update u k 、ω k and λ, and obtain the optimal solution of the signal component;

[0031] The BiLSTM algorithm is used to extract the temporal features of the signal components, and the weighted temporal features are obtained through the cross-attention mechanism. The weighted temporal features are then concatenated to obtain the fusion features.

[0032] Furthermore, the method for obtaining the accident rate based on the accident records includes:

[0033] Based on the accident records, we obtain the annual historical accident rates of nodes and edges, and use the LSTM combined with the SVR algorithm to build an accident prediction model:

[0034] Arrange the transport frequency, transport volume, equipment usage time and historical accident rate of the nodes in chronological order to obtain the input sequence. The objective function of LSTM is:

[0035]

[0036] Where Θ represents all parameters of LSTM, N is the number of feature dimensions of the input sequence, T is the length of the input sequence, and y l,t is the eigenvalue of the lth feature dimension at time step t, is the jth parameter of LSTM, For LSTM to y l,t The predicted data, α is the coefficient of L1 regularization term, ξ is a constant term used to ensure numerical stability, and ξ=1×10 -9 ;

[0037] The LSTM prediction data and the corresponding transportation frequency, transportation volume, and equipment usage time are combined into an input vector and input into the SVR model to predict the LSTM residual. The SVR decision function is:

[0038] f(x)=ω T τ(x)+b,b∈R

[0039] Among them, f(x) is the linear regression function, x is the input vector, including the transportation frequency, transportation volume and equipment usage time of the node, τ(x) is the mapping function, ω T is the transpose of ω, ω and b are the weight vector and bias to be determined, R represents a real number, slack variables and penalty factors are added to prevent overfitting, and the Lagrangian objective function of SVR is:

[0040]

[0041] in, is the regularization term, ξ and are the slack variables for the upper and lower boundaries respectively, ξ l and are the slack variables for the upper and lower boundaries corresponding to x l respectively, x l is the l-th input vector, C is the penalty factor, 0.1 ≤ C ≤ 100, M is the number of input vectors, L is the Lagrangian objective function, the Lagrange multipliers μ l , α l , μ l and are the slack multipliers related to the slack variables, α l and are the Lagrange multipliers for the upper and lower boundaries corresponding to x l respectively, f(x l ) is the predicted value for y l obtained from x l , r l is the target value corresponding to x l , is the residual, ε is the tolerance error, 0.01 ≤ ε ≤ 0.2, solving the minimum value of the Lagrangian function to obtain the optimal optimization multipliers, and obtaining the weight vector and bias according to the optimal optimization multipliers:

[0042]

[0043] where S u and S d represent the sets of support vectors violating the upper and lower boundaries respectively, |S u | and |S d | are the numbers of support vectors in S u and S d respectively, r s is the target value corresponding to the support vector s, x s is the support vector s;

[0044] Output of the accident prediction model:

[0045]

[0046] where, is the accident rate predicted by the accident prediction model, is the accident rate predicted by LSTM, f(x sgl ) is the residual of the accident rate predicted by LSTM by SVR, x sgl is The corresponding input vector takes the historical accident rate and the accident rate of the current time step predicted by the accident prediction model as the accident occurrence rate.

[0047] Furthermore, the method of performing topological simulation on the port transportation network model based on the fusion characteristics and the accident incidence rate to obtain a risk function includes:

[0048] Get the weather risk index based on weather data:

[0049]

[0050] Among them, WRI is the weather risk index, and are the influence weights of rainfall, wind speed, temperature, humidity, fog visibility and storm index respectively. The storm index is 0 and 1, indicating no storm and presence of storm respectively. and are respectively rainfall, wind speed, temperature, humidity, fog visibility and storm index, and υ is E J The adjustment parameter, E J is the impact factor of extreme weather events. When there is no extreme weather, E J =0, in case of mild extreme weather, 0 <E J <0.5, in moderate extreme weather, 0.5≤E J <1, in severe extreme weather, E=1;

[0051] The category label of whether an accident occurred is used as the target data, and the fusion features, node attributes, edge attributes, node geographic location, accident rate and weather risk index are used as input data to perform topological simulation on the topological structure of the port transportation network model and train the port transportation network topology simulation model. The port transportation network topology simulation model consists of three parts: input layer, function layer and output layer:

[0052] Adaptive adjacency matrix of the input layer:

[0053]

[0054] Among them, W1 and W2 are weight coefficient matrices, ReLU(·) and softmax(·) are activation functions, and A (0) To initialize the adjacency matrix, an adjacency matrix is established based on the correlation between the input data and the labels of accidents in the accident records.

[0055] Adaptive graph convolution expression of the input layer:

[0056]

[0057] Among them, Z is the graph matrix after adaptive update, X is the feature matrix of input data, is a diagonal matrix whose diagonal elements are The degree of each row in , W is the weight matrix used in the graph convolution operation;

[0058] The loss function of the function layer:

[0059]

[0060] Where y is the target vector, which is 0 or 1, and P is the predicted probability of an accident. is the weight matrix of the function layer, for The element in row i and column j in , λ is the regularization coefficient,

[0061] Weight update of the function layer:

[0062]

[0063] in, and are the weight matrices before and after updating in the function layer, β is the learning rate, and X is the feature matrix of the fusion feature;

[0064] The output formula of the output layer is the risk function:

[0065]

[0066] Among them, P i and P j are the predicted probabilities of accidents occurring at node i and edge j, respectively. σ(·) represents the sigmoid activation function. and are the feature vectors of node i and edge j after hidden layer processing, b′ represents the bias term,

[0067] The predicted probability of an accident occurring is output as the predicted risk probability.

[0068] Furthermore, the method for obtaining risk assessment data based on the risk function and the vulnerability function includes:

[0069] The calculation formula for risk assessment data is:

[0070]

[0071] Among them, F PJ is the risk assessment data for the dangerous goods concentration area in the port, δ6V i,cr ·P i and δ7V j,cr ·P j are the risk data of node i and edge j respectively, and are the number of nodes and edges respectively, δ6 and δ7 are the normalization coefficients of the risk assessment data of nodes and edges respectively, V j,cr is the vulnerability of edge j, P i and P j are the predicted risk probabilities of node i and edge j, respectively.

[0072] Furthermore, a method for constructing a risk assessment model based on the risk assessment data includes:

[0073] A data set was established using fusion features and risk assessment data, and the risk assessment data was predicted using fusion features. The data set was randomly sampled to form a training set and a test set with a data ratio of 4:1. The risk assessment model was trained using the training set through a neural network algorithm, and the hyperparameters of the risk assessment model were optimized using the test set.

[0074] Furthermore, the method for optimizing the risk assessment model by using the SandCat optimization algorithm includes:

[0075] Preset n sand cat populations to search for the optimal value in the hyperparameter search space of the risk assessment model, and the fitness value is the inverse of the objective function value of the risk assessment model.

[0076] Initialization of the Sand Cat optimization algorithm:

[0077]

[0078] where x ij and x ij+1 are the jth and j+1th dimensions of the i-th sand cat population, R j+1 is the j+1th chaotic parameter, and the chaotic parameter set is

[0079] The sand cat position update formula during the prey search phase is:

[0080] Pos(t+1)=e z·l ·cos(2πl)·γ·(Pos bc (t)-rand(0,1)·Pos c (t))

[0081] Where e is the base of natural logarithm, π is the ratio of circumference to circumference, γ represents the sensitivity range, l is a dynamic parameter, which is a random number between [-1, 1], Pos bc (t) is the current best candidate position, Pos c (t) is the current position, z is the spiral search factor, rand(0,1) is a random number in (0,1),

[0082] Spiral search factor:

[0083]

[0084] Among them, k is the spiral cycle parameter, t is the current time step, t max represents the maximum time step,

[0085] The sand cat position update formula in the local development stage is:

[0086]

[0087] Among them, Pos w (t) is the current worst position, Pos b (t) is the current global optimal position, β is the step size control parameter, K is a random number between [-1, 1], f i 、f g and f w Represent the current individual, global best and worst fitness values of the sand cat, ε is a constant,

[0088] When the maximum time step is reached, the hyperparameter combination corresponding to the global optimal position is output.

[0089] The port transportation network risk assessment system based on topology simulation is characterized by including:

[0090] The collection module collects historical data on dangerous goods transportation in the port's dangerous goods concentration area, including dangerous goods types, transportation frequency, transportation volume, transportation routes, operation data and accident records;

[0091] a calculation module, constructing a dynamic port transportation network model based on the transportation routes, performing topological analysis to construct a vulnerability function, fusing feature data of the historical data to obtain a fusion feature, obtaining an accident rate based on the accident records, performing topological simulation on the port transportation network model based on the fusion feature and the accident rate to obtain a risk function, obtaining risk assessment data based on the risk function and the vulnerability function, and constructing a risk assessment model based on the risk assessment data;

[0092] An optimization module, which optimizes the risk assessment model by using a sandcat optimization algorithm;

[0093] The output module inputs the data to be evaluated into the risk assessment model to obtain the evaluation results.

[0094] The beneficial effects of the present invention are:

[0095] This invention combines feature fusion, topology simulation technology, and the SandCat optimization algorithm to accurately assess port transportation network risks. This method dynamically analyzes the vulnerabilities and risks within the transportation network, providing port managers with a scientific, quantitative risk assessment tool. By optimizing the risk assessment model, this invention not only improves the accuracy of risk predictions but also enhances the port transportation network's ability to respond to emergencies, effectively reducing the probability of accidents and ensuring the safety of port operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] Figure 1 This is a flowchart of the steps of the port transportation network risk assessment method and system based on topology simulation of the present invention;

[0097] Figure 2 This is a topological diagram of the loading and unloading ship of the present invention;

[0098] Figure 3 This is a topological diagram of the transfer of goods according to the present invention;

[0099] Figure 4 This is the topological structure diagram of the inverted tank of the present invention. DETAILED DESCRIPTION

[0100] The present invention will be further described below through specific examples. The illustrative examples and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0101] The port transportation network risk assessment method and system based on topology simulation of the present invention include the following steps:

[0102] like Figure 1 As shown, in this embodiment, the following steps are included:

[0103] Collect historical data on dangerous goods transportation in the port's dangerous goods concentration area, including dangerous goods types, transportation frequency, transportation volume, transportation routes, operation data and accident records;

[0104] In the actual assessment, the type of dangerous goods is crude oil, and the transportation route is: unloading operation: A→B, loading operation: B→C→A, lightering operation: A→A, tank unloading operation: B→C→B, A is the berth where the ship docks, including A1-A6, B is the storage area for oil tanks, including B1-B6, C is the pump room for transporting oil tanks, including C1-C4, and the geographical location is the longitude and latitude coordinates;

[0105] Constructing a dynamic port transportation network model based on the transportation routes, performing topological analysis and constructing a vulnerability function;

[0106] In the actual evaluation, the topological structure diagram of the port transportation network model is as follows: Figure 2As shown, the gray edges represent the loading process, the black edges represent the unloading process, and the barge topology diagram is Figure 3 , the inverted tank topology diagram is Figure 4 , where the black edge points from the pump room to the storage area, and the gray edge is a bidirectional edge, indicating that the storage area transports crude oil to the pump room, and the crude oil is transported to other storage areas through the pump room;

[0107] Fusing the feature data of the historical data to obtain a fusion feature, and obtaining an accident occurrence rate based on the accident records;

[0108] Performing a topological simulation on the port transportation network model based on the fusion characteristics and the accident incidence rate to obtain a risk function, and obtaining risk assessment data based on the risk function and the vulnerability function;

[0109] A risk assessment model is constructed based on the risk assessment data, the risk assessment model is optimized by using a sandcat optimization algorithm, and the data to be evaluated is input into the risk assessment model to obtain an evaluation result.

[0110] In this embodiment, the method for collecting historical data on dangerous goods transportation in a dangerous goods concentration area of a port, including dangerous goods types, transportation frequency, transportation volume, transportation routes, operation data, and accident records, includes:

[0111] Historical data is obtained through port operation records, maritime department reports and safety supervision archives. The transportation routes include the geographical location, transportation connections and transportation processes of equipment and facilities for the transportation of dangerous goods. The transportation processes include loading and unloading operations, transshipment operations and tank emptying operations. Operational data includes cargo loading and unloading time, equipment usage time, capacity and processing speed of equipment and facilities, as well as weather data, transportation time, congestion and transportation capacity of transportation connections.

[0112] In this embodiment, a method for constructing a dynamic port transportation network model based on the transportation routes and performing topological analysis to construct a vulnerability function includes:

[0113] The nodes of the dynamic port transportation network model are equipment and facilities. The node attributes include the capacity and processing speed of the nodes. The edges represent the transportation connections between nodes. The attributes of the edges include transportation time and transportation capacity. The nodes, edges and node attributes of the dynamic port transportation network model are dynamically updated as the transportation routes change. The direction of the edges is the same as the flow direction of dangerous goods.

[0114] The vulnerability function of the node is obtained by analyzing the node's clustering coefficient, betweenness, degree, degree distribution and node weight:

[0115]

[0116] Among them, V i,cris the vulnerability of node i, δ1, v2, δ3, δ4 and δ5 are the influence weights of clustering coefficient, betweenness, degree, degree distribution and node weight respectively, C i , BC i,v 、k i 、P(k i ) and W i are the clustering coefficient, betweenness, degree, degree distribution and node weight of node i respectively, and are the influence weights of node capacity and processing speed respectively, C i and S i are the capacity and processing speed of node i, C max and S max are the maximum capacity and maximum processing speed among all nodes respectively;

[0117] The edge vulnerability function is obtained by performing edge vulnerability analysis based on edge attributes:

[0118]

[0119] Among them, V j,cr is the vulnerability of edge j, Υ T , Υ C and Y i are the impact weights of transportation time, transportation capacity and vulnerability of adjacent nodes, T j 、C j 、V j,n1,cr and V j,n2,cr are the transportation time, transportation capacity, vulnerability of adjacent node 1, and vulnerability of adjacent node 2 of edge j respectively.

[0120] In this embodiment, the method of fusing the feature data of the historical data to obtain fused features includes:

[0121] The types of dangerous goods are converted into numerical vectors, and the transport frequency, transport volume and operation data are normalized and preprocessed.

[0122] Arrange and organize the dangerous goods types, transportation frequency, transportation volume and operation data in chronological order to form time series data that can be processed by signals, and decompose the time series data:

[0123]

[0124] in is the Dirac function, u k and ω k are the kth signal component after decomposition and the corresponding center frequency, ω k and u kare the center frequency and signal component of the kth component after decomposition, K is the number of components after decomposition, is the partial derivative of the signal f with respect to the time step t, * represents the convolution operation, π is the pi, e is the base of the natural logarithm, t is the current time step, j is the imaginary unit, st represents the constraint condition, f is the signal to be decomposed, is the time series data,

[0125] By iteratively optimizing the decomposition algorithm, the components and center frequencies are continuously adjusted to minimize the augmented Lagrangian function, which is:

[0126]

[0127] in, is the quadratic penalty factor, λ is the Lagrange multiplier, λ(t) is the Lagrange multiplier of the current time step,

[0128] The alternating direction multiplier method is introduced and combined with Fourier isometric transform to find the solution of the augmented Lagrangian function until and u for the nth and n+1th iterations respectively k , update u k 、ω k and λ, and obtain the optimal solution of the signal component;

[0129] The BiLSTM algorithm is used to extract the temporal features of the signal components, and the weighted temporal features are obtained through the cross-attention mechanism. The weighted temporal features are then concatenated to obtain the fusion features.

[0130] In this embodiment, the method for obtaining the accident incidence rate based on the accident records includes:

[0131] Based on the accident records, we obtain the annual historical accident rates of nodes and edges, and use the LSTM combined with the SVR algorithm to build an accident prediction model:

[0132] Arrange the transport frequency, transport volume, equipment usage time and historical accident rate of the nodes in chronological order to obtain the input sequence. The objective function of LSTM is:

[0133]

[0134] Where Θ represents all parameters of LSTM, N is the number of feature dimensions of the input sequence, T is the length of the input sequence, and y l,t is the eigenvalue of the lth feature dimension at time step t, is the jth parameter of LSTM, For LSTM to y l,tThe predicted data, α is the coefficient of L1 regularization term, ξ is a constant term used to ensure numerical stability, and ξ=1×10 -9 ;

[0135] The LSTM prediction data and the corresponding transportation frequency, transportation volume, and equipment usage time are combined into an input vector and input into the SVR model to predict the LSTM residual. The SVR decision function is:

[0136] f(x)=ω T τ(x)+b,b∈R

[0137] Among them, f(x) is the linear regression function, x is the input vector, including the transportation frequency, transportation volume and equipment usage time of the node, τ(x) is the mapping function, ω T is the transpose of ω, ω and b are the weight vector and bias to be determined, R represents a real number, slack variables and penalty factors are added to prevent overfitting, and the Lagrangian objective function of SVR is:

[0138]

[0139] in, is the regularization term, ξ and are the slack variables for the upper and lower bounds, ξ l and x l The corresponding upper and lower bounds of the slack variables, x l is the lth input vector, C is the penalty factor, 0.1≤C≤100, M is the number of input vectors, L is the Lagrangian objective function, and the Lagrangian multiplier μ is l 、 α l 、 μ l and is the slack multiplier associated with the slack variable, α l and x l The corresponding Lagrange multipliers of the upper and lower boundaries, f(x l ) is based on x l The obtained l The predicted value, r l is x l The corresponding target value is the residual, ε is the tolerance error, 0.01≤ε≤0.2, and the minimum value of the Lagrangian function is solved to obtain the optimal optimization multiplier. The weight vector and bias are obtained according to the optimal optimization multiplier:

[0140]

[0141] Among them Su and S d respectively represent the sets of support vectors violating the upper and lower boundaries, |S u | and |S d | are the numbers of support vectors in S u and S d respectively, r s is the target value corresponding to the support vector s, x s is the support vector s;

[0142] Output of the accident prediction model:

[0143]

[0144] Among them, is the accident rate predicted by the accident prediction model, is the accident rate predicted by LSTM, f(x sgl ) is the residual of the prediction of the accident rate by LSTM predicted by SVR, x sgl is corresponding input vector, taking the historical accident rate and the accident rate predicted by the accident prediction model at the current time step as the accident incidence rate, and the accident incidence rate is the inherent probability of an accident occurring.

[0145] In this embodiment, the method for performing topological simulation on the port transportation network model based on the fusion feature and the accident incidence rate to obtain a risk function includes:

[0146] Obtaining a weather risk index according to weather data:

[0147]

[0148] Among them, WRI is the weather risk index, and are the influence weights of rainfall, wind speed, temperature, humidity, fog visibility and storm index respectively, and the storm index is 0 and 1, indicating no storm and having a storm, and are rainfall, wind speed, temperature, humidity, fog visibility and storm index respectively, υ is the adjustment parameter of E J , E J is the influence factor of extreme weather events. When there is no extreme weather, E J =0. When there is mild extreme weather, 0 < E J <0.5. When there is moderate extreme weather, 0.5 ≤ E J <1. When there is severe extreme weather, E = 1;

[0149] The category label of whether an accident occurred is used as the target data, and the fusion features, node attributes, edge attributes, node geographic location, accident rate and weather risk index are used as input data to perform topological simulation on the topological structure of the port transportation network model and train the port transportation network topology simulation model. The port transportation network topology simulation model consists of three parts: input layer, function layer and output layer:

[0150] Adaptive adjacency matrix of the input layer:

[0151]

[0152] Among them, W1 and W2 are weight coefficient matrices, ReLU(·) and softmax(·) are activation functions, and A (0) To initialize the adjacency matrix, an adjacency matrix is established based on the correlation between the input data and the labels of accidents in the accident records.

[0153] Adaptive graph convolution expression of the input layer:

[0154]

[0155] Among them, Z is the graph matrix after adaptive update, X is the feature matrix of input data, is a diagonal matrix whose diagonal elements are The degree of each row in , W is the weight matrix used in the graph convolution operation;

[0156] The loss function of the function layer:

[0157]

[0158] Where y is the target vector, which is 0 or 1, and P is the predicted probability of an accident. is the weight matrix of the function layer, for The element in row i and column j in , λ is the regularization coefficient,

[0159] Weight update of the function layer:

[0160]

[0161] in, and are the weight matrices before and after updating in the function layer, β is the learning rate, and X is the feature matrix of the fusion feature;

[0162] The output formula of the output layer is the risk function:

[0163]

[0164] Among them, Pi and P j are the predicted probabilities of accidents occurring at node i and edge j, respectively. σ(·) represents the sigmoid activation function. and are the feature vectors of node i and edge j after hidden layer processing, b′ represents the bias term,

[0165] Output the predicted probability of an accident as the predicted risk probability.

[0166] In this embodiment, the method for obtaining risk assessment data based on the risk function and the vulnerability function includes:

[0167] The calculation formula for risk assessment data is:

[0168]

[0169] Among them, F PJ is the risk assessment data for the dangerous goods concentration area in the port, δ6V i,cr ·P i and δ7V j,cr ·P j are the risk data of node i and edge j respectively, and are the number of nodes and edges respectively, δ6 and δ7 are the normalization coefficients of the risk assessment data of nodes and edges respectively, V j,cr is the vulnerability of edge j, P i and P j are the predicted risk probabilities of node i and edge j, respectively.

[0170] In this embodiment, the method for constructing a risk assessment model based on the risk assessment data includes:

[0171] A data set was established using fusion features and risk assessment data, and the risk assessment data was predicted using fusion features. The data set was randomly sampled to form a training set and a test set with a data ratio of 4:1. The risk assessment model was trained using the training set through a neural network algorithm, and the hyperparameters of the risk assessment model were optimized using the test set.

[0172] In this embodiment, the method for optimizing the risk assessment model using the SandCat optimization algorithm includes:

[0173] Preset n sand cat populations to search for the optimal value in the hyperparameter search space of the risk assessment model, and the fitness value is the inverse of the objective function value of the risk assessment model.

[0174] Initialization of the Sand Cat optimization algorithm:

[0175]

[0176] where x ij and x ij+1 are the jth and j+1th dimensions of the i-th sand cat population, R j+1 is the j+1th chaotic parameter, and the chaotic parameter set is

[0177] The sand cat position update formula during the prey search phase is:

[0178] Pos(t+1)=e z·l ·cos(2πl)·γ·(Pos bc (t)-rand(0,1)·Pos c (t))

[0179] Where e is the base of natural logarithm, π is the ratio of circumference to circumference, γ represents the sensitivity range, l is a dynamic parameter, which is a random number between [-1, 1], Pos bc (t) is the current best candidate position, Pos c (t) is the current position, z is the spiral search factor, rand(0,1) is a random number in (0,1),

[0180] Spiral search factor:

[0181]

[0182] Among them, k is the spiral cycle parameter, t is the current time step, t max represents the maximum time step,

[0183] The sand cat position update formula in the local development stage is:

[0184]

[0185] Among them, Pos w (t) is the current worst position, Pos b (t) is the current global optimal position, β is the step size control parameter, K is a random number between [-1, 1], f i 、f g and f w Represent the current individual, global best and worst fitness values of the sand cat, ε is a constant,

[0186] When the maximum time step is reached, the hyperparameter combination corresponding to the global optimal position is output.

[0187] The port transportation network risk assessment system based on topology simulation is characterized by including:

[0188] The collection module collects historical data on dangerous goods transportation in the port's dangerous goods concentration area, including dangerous goods types, transportation frequency, transportation volume, transportation routes, operation data and accident records;

[0189] a calculation module, constructing a dynamic port transportation network model based on the transportation routes, performing topological analysis to construct a vulnerability function, fusing feature data of the historical data to obtain a fusion feature, obtaining an accident rate based on the accident records, performing topological simulation on the port transportation network model based on the fusion feature and the accident rate to obtain a risk function, obtaining risk assessment data based on the risk function and the vulnerability function, and constructing a risk assessment model based on the risk assessment data;

[0190] An optimization module, which optimizes the risk assessment model by using a sandcat optimization algorithm;

[0191] The output module inputs the data to be evaluated into the risk assessment model to obtain the evaluation results.

[0192] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A port transportation network risk assessment method based on topology simulation is characterized by: The following steps are involved: Collect historical data on dangerous goods transportation in the port's dangerous goods concentration area, including dangerous goods types, transportation frequency, transportation volume, transportation routes, operational data, and accident records; construct a dynamic port transportation network model based on the transportation routes, and perform topological analysis to construct a vulnerability function; Fusing the feature data of the historical data to obtain a fusion feature, and obtaining an accident occurrence rate based on the accident records; Performing a topological simulation on the port transportation network model based on the fusion characteristics and the accident incidence rate to obtain a risk function, and obtaining risk assessment data based on the risk function and the vulnerability function; Constructing a risk assessment model based on the risk assessment data, optimizing the risk assessment model using a Sandcat optimization algorithm, and inputting the data to be evaluated into the risk assessment model to obtain an evaluation result; The method for constructing a dynamic port transportation network model based on the transportation route and performing topological analysis to construct a vulnerability function includes: The nodes of the dynamic port transportation network model are equipment and facilities. Node attributes include node capacity and processing speed. Edges represent transportation connections between nodes. Edge attributes include transportation time and transportation capacity. The nodes, edges, and node attributes of the dynamic port transportation network model are dynamically updated as transportation routes change. The direction of the edges is the same as the flow direction of dangerous goods. The vulnerability function of the node is obtained by performing vulnerability analysis on the node through the node attributes: Among them, V i,cr is the vulnerability of node i, δ1, δ2, δ3, δ4 and δ5 are the influence weights of clustering coefficient, betweenness, degree, degree distribution and node weight respectively, C i , BC i,v 、k i 、P(k i ) and W i are the clustering coefficient, betweenness, degree, degree distribution and node weight of node i respectively, and are the influence weights of node capacity and processing speed respectively, C i and S i are the capacity and processing speed of node i, C max and S max are the maximum capacity and maximum processing speed among all nodes respectively; The edge vulnerability function is obtained by performing edge vulnerability analysis based on edge attributes: Among them, V j,cr is the vulnerability of edge j, Υ T , Υ C and Y i are the impact weights of transportation time, transportation capacity and vulnerability of adjacent nodes, T j 、C j 、V j,n1,cr and V j,n2,cr are the transportation time, transportation capacity, vulnerability of neighboring node 1, and vulnerability of neighboring node 2 of edge j respectively; The annual historical accident rates of nodes and edges are obtained based on accident records. An accident prediction model is established using the LSTM combined with the SVR algorithm. The historical accident rates and the accident rates at the current time step predicted by the accident prediction model are used as the accident occurrence rate.

2. The port transportation network risk assessment method based on topology simulation according to claim 1 is characterized in that: Methods for collecting historical data on dangerous goods transport in concentrated areas of ports, including types of dangerous goods, transport frequency, transport volume, transport routes, operational data, and accident records, include: Historical data is obtained through port operation records, maritime department reports and safety supervision archives. The transportation routes include the geographical location, transportation connections and transportation processes of equipment and facilities for the transportation of dangerous goods. Operational data includes cargo loading and unloading time, equipment usage time, capacity and processing speed of equipment and facilities, as well as weather data, transportation time, congestion and transportation capacity of transportation connections.

3. The port transportation network risk assessment method based on topology simulation according to claim 1 is characterized in that: The method for fusing the feature data of the historical data to obtain a fused feature includes: The types of dangerous goods are converted into numerical vectors, and the transport frequency, transport volume and operation data are normalized and preprocessed. Arrange and organize the dangerous goods types, transportation frequency, transportation volume and operation data in chronological order to form time series data that can be processed by signals, and decompose the time series data: in is the Dirac function, u k and ω k are the kth signal component and the corresponding center frequency after decomposition, K is the number of components after decomposition, is the partial derivative of the signal f with respect to the time step t, * represents the convolution operation, π is the pi, e is the base of the natural logarithm, t is the current time step, j is the imaginary unit, st represents the constraint condition, f is the signal to be decomposed, is the time series data, By iteratively optimizing the decomposition algorithm, the components and center frequencies are continuously adjusted to minimize the augmented Lagrangian function, which is: in, is the quadratic penalty factor, λ is the Lagrange multiplier, λ(t) is the Lagrange multiplier of the current time step, The alternating direction multiplier method is introduced and combined with Fourier isometric transform to find the solution of the augmented Lagrangian function until and u for the nth and n+1th iterations respectively k , update u k 、ω k and λ, and obtain the optimal solution of the signal component; The BiLSTM algorithm is used to extract the temporal features of the signal components, and the weighted temporal features are obtained through the cross-attention mechanism. The weighted temporal features are then concatenated to obtain the fusion features.

4. The port transportation network risk assessment method based on topology simulation according to claim 1 is characterized in that: The method for obtaining the accident occurrence rate based on the accident record includes: Based on the accident records, we obtain the annual historical accident rates of nodes and edges, and use the LSTM combined with the SVR algorithm to build an accident prediction model: Arrange the transport frequency, transport volume, equipment usage time and historical accident rate of the nodes in chronological order to obtain the input sequence. The objective function of LSTM is: Where Θ represents all parameters of LSTM, N is the number of feature dimensions of the input sequence, and T is the length of the input sequence. For the The eigenvalue of the feature dimension at time step t, is the jth parameter of LSTM, For LSTM The predicted data, α is the coefficient of the L1 regularization term, ξ' is a constant term used to ensure numerical stability, and ξ'=1×10 -9 The LSTM prediction data and the corresponding transportation frequency, transportation volume, and equipment usage time are combined into an input vector, which is then input into the SVR model to predict the LSTM residual. The SVR decision function is: f(x)=ω T τ(x)+b,b∈R Among them, f(x) is the linear regression function, x is the input vector, including the transportation frequency, transportation volume and equipment usage time of the node, τ(x) is the mapping function, ω T is the transpose of ω, ω and b are the weight vector and bias to be determined, R represents a real number, slack variables and penalty factors are added to prevent overfitting, and the Lagrangian objective function of SVR is: in, is the regularization term, ξ and are the slack variables for the upper and lower bounds, and They are The corresponding upper and lower bounds of the slack variables, For the input vectors, C is the penalty factor, 0.1≤C≤100, M is the number of input vectors, L is the Lagrangian objective function, and the Lagrangian multiplier ≥0, and is the slack multiplier associated with the slack variable, and They are The corresponding Lagrange multipliers for the upper and lower boundaries are, Based on Obtained for The predicted value of for The corresponding target value is the residual, ε is the tolerance error, 0.01≤ε≤0.2, and the minimum value of the Lagrangian function is solved to obtain the optimal optimization multiplier. The weight vector and bias are obtained according to the optimal optimization multiplier: where S u and S d represent the sets of support vectors violating the upper and lower boundaries respectively, |S u | and |S d | are the numbers of support vectors in S u and S d respectively, r s is the target value corresponding to the support vector s, x s is the support vector s; Output of the accident prediction model: in, is the accident rate predicted by the accident prediction model, is the accident rate predicted by LSTM, f(x sgl ) is the residual of the LSTM prediction of the accident rate predicted by SVR, x sgl for The corresponding input vector takes the historical accident rate and the accident rate of the current time step predicted by the accident prediction model as the accident occurrence rate.

5. The port transportation network risk assessment method based on topology simulation according to claim 1 is characterized in that: The method for obtaining a risk function by performing topological simulation on the port transportation network model based on the fusion characteristics and the accident incidence rate includes: Get the weather risk index based on weather data: Among them, WRI is the weather risk index, and are the influence weights of rainfall, wind speed, temperature, humidity, fog visibility and storm index respectively. The storm index is 0 and 1, indicating no storm and presence of storm respectively. and are respectively rainfall, wind speed, temperature, humidity, fog visibility and storm index, and υ is E J The adjustment parameter, E J is the impact factor of extreme weather events. When there is no extreme weather, E J =0, in case of mild extreme weather, 0 <E J <0.5, in moderate extreme weather, 0.5≤E J <1, in severe extreme weather, E=1; The category label of whether an accident occurred is used as the target data, and the fusion features, node attributes, edge attributes, node geographic location, accident rate and weather risk index are used as input data to perform topological simulation on the topological structure of the port transportation network model and train the port transportation network topology simulation model. The port transportation network topology simulation model consists of three parts: input layer, function layer and output layer: Adaptive adjacency matrix of the input layer: Among them, W1 and W2 are weight coefficient matrices, ReLU(·) and softmax(·) are activation functions, and A (0) To initialize the adjacency matrix, an adjacency matrix is established based on the correlation between the input data and the labels of accidents in the accident records. Adaptive graph convolution expression of the input layer: Among them, Z is the graph matrix after adaptive update, X is the feature matrix of input data, is a diagonal matrix whose diagonal elements are The degree of each row in , W is the weight matrix used in the graph convolution operation; The loss function of the function layer: Where y is the target vector, which is 0 or 1, and P is the predicted probability of an accident. is the weight matrix of the function layer, for The element in the i-th row and j-th column, λ' is the regularization coefficient, and the weight update of the function layer is: in, and are the weight matrices before and after updating in the function layer, β is the learning rate, and X is the feature matrix of the fusion feature; The output formula of the output layer is the risk function: Among them, P i and P j are the predicted probabilities of accidents occurring at node i and edge j, respectively. σ(·) represents the sigmoid activation function. and are the feature vectors of node i and edge j after hidden layer processing, b' represents the bias term, Output the predicted probability of an accident as the predicted risk probability.

6. The port transportation network risk assessment method based on topology simulation according to claim 1 is characterized in that: The method for obtaining risk assessment data based on the risk function and the vulnerability function includes: The calculation formula for risk assessment data is: Among them, F PJ is the risk assessment data for the dangerous goods concentration area in the port, δ6V i,cr ·P i and δ7V j,cr ·P j are the risk data of node i and edge j respectively, and are the number of nodes and edges respectively, δ6 and δ7 are the normalization coefficients of the risk assessment data of nodes and edges respectively, V j,cr is the vulnerability of edge j, P i and P j are the predicted risk probabilities of node i and edge j, respectively.

7. The port transportation network risk assessment method based on topology simulation according to claim 1 is characterized in that: The method for constructing a risk assessment model based on the risk assessment data includes: using fusion features and risk assessment data to establish a data set, using the fusion features to predict the risk assessment data, randomly sampling the data set to form a training set and a test set with a data ratio of 4:1, using the training set to train the risk assessment model through a neural network algorithm, and using the test set to optimize the hyperparameters of the risk assessment model.

8. The port transportation network risk assessment method based on topology simulation according to claim 1 is characterized in that: The method for optimizing the risk assessment model by using a sand cat optimization algorithm comprises: presetting n sand cat populations to search for an optimal value in a hyperparameter search space of the risk assessment model, wherein the fitness value is the inverse of the objective function value of the risk assessment model; Initialization of the Sand Cat optimization algorithm: where x ij and x ij+1 are the jth and j+1th dimensions of the i-th sand cat population, R j+1 is the j+1th chaotic parameter, and the chaotic parameter set is The sand cat position update formula during the prey search phase is: Pos(t+1)=e z·l ·cos(2πl)·γ·(Pos bc (t)-rand(0,1)·Pos c (t)) Where e is the base of natural logarithm, π is the ratio of circumference to circumference, γ represents the sensitivity range, l is a dynamic parameter, which is a random number between [-1, 1], Pos bc (t) is the current best candidate position, Pos c (t) is the current position, z is the spiral search factor, rand(0,1) is a random number in (0,1), Spiral search factor: Among them, k is the spiral cycle parameter, t is the current time step, t max The formula for updating the position of the sand cat in the local development stage at the maximum time step is: Among them, Pos w (t) is the current worst position, Pos b (t) is the current global optimal position, β is the step size control parameter, K' is a random number between [-1, 1], f i 、f g and f w They represent the current individual, global best, and worst fitness values of the sand cat, respectively. ε” is a constant. When the maximum time step is reached, the hyperparameter combination corresponding to the global optimal position is output.

9. A port transportation network risk assessment system based on topology simulation, for executing the port transportation network risk assessment method based on topology simulation according to any one of claims 1 to 8, characterized in that: include: The collection module collects historical data on dangerous goods transportation in the port's dangerous goods concentration area, including dangerous goods types, transportation frequency, transportation volume, transportation routes, operation data and accident records; a calculation module, constructing a dynamic port transportation network model based on the transportation routes, performing topological analysis to construct a vulnerability function, fusing feature data of the historical data to obtain a fusion feature, obtaining an accident rate based on the accident records, performing topological simulation on the port transportation network model based on the fusion feature and the accident rate to obtain a risk function, obtaining risk assessment data based on the risk function and the vulnerability function, and constructing a risk assessment model based on the risk assessment data; An optimization module, which optimizes the risk assessment model by using a sandcat optimization algorithm; The output module inputs the data to be evaluated into the risk assessment model to obtain the evaluation results.

Images