Heuristic-based two-stage selective integration approach for maritime accident risk prediction

Abstract

The application discloses a shipping accident risk prediction method based on a two-stage selective integration of heuristics, comprising the following steps: 1, obtaining a shipping accident data set and dividing it into a training set G1 and a verification set G2, thereby constructing an original base classification pool; 2, preselecting an ELM classifier by using a K-means++ algorithm to obtain n' shipping accident risk prediction base models with large differences; 3, improving a Harris eagle optimization algorithm and performing secondary selection to obtain a combination scheme of m' shipping accident risk prediction base models with relatively optimal comprehensive performance; and 4, performing shipping accident risk integrated prediction. The application introduces a chaotic initialization strategy and a reverse refraction learning mechanism to improve the standard Harris eagle optimization algorithm, improves the shipping accident risk prediction accuracy and prediction efficiency, and provides support for risk decision-making of shipping practitioners.

Description

Technical Field

[0001] This invention relates to the field of shipping accident risk analysis, and specifically to a shipping accident risk prediction method based on heuristic two-stage selective integration. Background Technology

[0002] With the rapid development of economic globalization, the scale of maritime transportation is constantly expanding, and the risk of shipping accidents is increasing accordingly. At the same time, maritime transportation is affected by a variety of complex and uncertain risk factors, such as severe weather conditions and force majeure events like natural disasters. These factors can easily lead to shipping accidents, causing serious losses to the economy, personnel safety, and the marine environment. Therefore, it is urgent to study accurate and efficient methods for predicting shipping accident risks, providing auxiliary support for improving shipping accident early warning mechanisms and enhancing shipping safety.

[0003] Machine learning technology is widely used in risk prediction due to its adaptability, multi-feature comprehensive analysis advantages, and ability to handle nonlinear relationships. Because shipping data is complex, high-dimensional, massive in scale, and highly time-sensitive, and the risk factors involved are characterized by dynamic uncertainty, the classification and prediction performance of single machine learning methods is unstable and the prediction accuracy is insufficient. Therefore, based on single machine learning methods, an ensemble learning method is proposed to achieve the final shipping accident risk decision by combining multiple base classifiers. Although ensemble learning has more stable classification and prediction capabilities compared to single classifiers, as the number of base classifiers and the data scale increase in ensemble learning, it requires significant computational overhead, thus affecting the computational efficiency of the entire shipping accident risk prediction process. Summary of the Invention

[0004] This invention aims to overcome the shortcomings of existing technologies by proposing a heuristic two-stage selective integration method for predicting shipping accident risks. This method aims to effectively utilize complex and high-dimensional shipping data, adapt to rapidly changing new data, and thus improve the accuracy and efficiency of integrated prediction of shipping accident risks, providing support for risk decision-making by shipping professionals.

[0005] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0006] The present invention provides a heuristic two-stage selective integration-based method for predicting shipping accident risks, characterized by the following steps:

[0007] Step 1: Obtain the shipping accident dataset and divide it into a training set G1 and a validation set G2;

[0008] The training set G1 is sampled n times using the Bootstrap method to obtain n sampled datasets, which are then used to train n ELM classifiers to obtain n shipping accident risk prediction base models, which together form the original base classification pool.

[0009] Step 2: Based on the K-means++ algorithm, pre-select the original base classification pool to obtain n' shipping accident risk prediction base models with significant differences;

[0010] Step 3: Use the improved Harris Eagle optimization algorithm to perform a secondary selection on the n' shipping accident risk prediction base models until the optimal combination of base models is found and used for shipping accident risk prediction:

[0011] Step 3.1: Define the current iteration number as t and initialize t = 1; define the maximum iteration number as T; and the population size as pop; randomly initialize the position set of the Harris Eagle population in the t-th generation within the search space. This represents the position of the o-th Harris Eagle in generation t, and signifies the selection of different shipping accident risk prediction base models from n' shipping accident risk prediction base models to form a model combination scheme; and lb and ub represent the lower and upper bounds of the search space, respectively;

[0012] Step 3.2: Introduce a chaotic initialization mechanism into the n' shipping accident risk prediction base models to update the position set of the t-th generation Harris Eagle population and obtain the updated position set;

[0013] Step 3.3: Select the position of the Harris Eagle with the best fitness value from the updated position set as the position X of the globally optimal individual. best ;

[0014] Step 3.4: Calculate the escape energy E of the prey in generation t according to equation (4). t :

[0015]

[0016] In equation (4), E0 represents the prey's initial escape energy;

[0017] Step 3.5: Update the updated position set again using the improved position update strategy, and after obtaining the (t+1)th generation Harris eagle population, select the position of the Harris eagle with the best fitness value as the position X of the globally optimal individual. best ;

[0018] Step 3.6: Using the reverse refraction learning mechanism, search for reverse individuals in the (t+1)th generation Harris eagle population within the search space to obtain the updated position set of the (t+1)th generation Harris eagle population, and select the position of the Harris eagle with the best fitness value from it as the position X of the optimal individual. best ;

[0019] Step 3.7: Determine whether t = T holds true. If it does, use the Harris Eagle individual corresponding to the global optimal fitness value as the selection scheme for the combination of shipping accident risk prediction base models and use it to construct the integrated prediction model for shipping accident risk. Otherwise, assign t+1 to t and return to step 3.4 to execute sequentially.

[0020] Step 4: Use the majority voting method to integrate the optimal shipping accident risk prediction base model combination scheme obtained in the above steps, thereby constructing a shipping accident risk integrated prediction model, which is used to predict shipping accident risks for input high-dimensional, complex, and time-sensitive shipping data.

[0021] The shipping accident risk prediction method based on heuristic two-stage selective integration described in this invention is also characterized in that step 2 includes:

[0022] Step 2.1: The validation set G2 is input into n shipping accident risk prediction base models for processing to obtain n shipping accident risk prediction values, and the prediction accuracy of the n shipping accident risk prediction base models is calculated.

[0023] Define a variable z and initialize z = 1;

[0024] Randomly select a shipping accident risk prediction value as the z-th cluster center c z And form a cluster center set c = {c z Define the number of cluster centers in c as α, and initialize α = 1;

[0025] Step 2.2: Calculate the remaining n-α shipping accident risk prediction values ​​and their correlation with the z-th cluster center c. z The Euclidean distance between them forms the z-th distance set D. z ={d z,1 ,d z,2 ,...,d z,i ,...,d z,n-α}, where d z,i This represents the predicted value of the i-th shipping accident risk and the cluster center c of the z-th cluster. z The Euclidean distance between them;

[0026] Step 2.3: Calculate the i-th accumulated distance. Where, d z,i′This represents the predicted shipping accident risk value of the i'th shipping accident and the cluster center c of the zth cluster. z The Euclidean distance is obtained; thus, the z-th cumulative distance set D is obtained. z,s ={v z,1 ,v z,2 ,...,v z,i ,v z,n-α};

[0027] Step 2.4, from D z,s Randomly select an accumulated distance v z Then, subtract set D in sequence. z Each element in the list, and determine v. z -d z,i Is ≤0 true? If true, then set d... z,i Add the corresponding shipping risk forecast value to set c; otherwise, proceed directly to step 2.5.

[0028] Step 2.5: Calculate the minimum Euclidean distance d from the i-th predicted shipping accident risk value to each cluster center in c. i,min Thus, the set of minimum Euclidean distances D is obtained. min ={d 1,min ,d 2,min ,...,d i,min ,...,d n-α,min};

[0029] Step 2.6: Calculate the minimum Euclidean distance set D min The i-th minimum Euclidean distance d i,min Probability of being selected Therefore, based on the probability of each minimum Euclidean distance being selected, the roulette wheel algorithm is used to select from D. min The minimum Euclidean distance d is determined in the final selection. last,min , thus making d last,min Add the corresponding shipping accident risk prediction value to set c; otherwise, proceed directly to step 2.7.

[0030] Step 2.7: Determine whether the number of cluster centers in set c has reached k. If it has, proceed to step 2.8; otherwise, assign z+1 to z and return to step 2.2.

[0031] Step 2.8: Calculate the Euclidean distance between the remaining nk shipping accident risk prediction values ​​and the k cluster centers, and then assign the nk shipping accident risk prediction values ​​to the clusters to which the nearest cluster centers belong based on the Euclidean distance;

[0032] Step 2.9: Select the base models corresponding to the top w shipping accident risk prediction values ​​with high prediction accuracy from each cluster as the n' shipping accident risk prediction base models with greater differences.

[0033] Step 3.2 includes:

[0034] Step 3.2.1: Use equation (1) to determine the position of the o-th Harris Eagle in the t-th generation. Perform an update to obtain the updated location.

[0035]

[0036] In equation (1), δ is the chaos factor, and mod represents the remainder;

[0037] Step 3.2.2, if If all elements in the initialization list are 0, then return to step 3.2.1 to re-initialize.

[0038] Step 3.3 includes:

[0039] The fitness function value of the o-th Harris Eagle in the t-th generation is calculated using equation (2). Thus, the fitness function value of the t-th generation Harris Eagle population is obtained. Set the position of the Harris Eagle with the best fitness value as the position of the prey in generation t. And the position X of the globally optimal individual. best ;

[0040]

[0041] In equation (3), Let λ represent the prediction accuracy of the position of the o-th Harris Hawk in the t-th generation, and let λ represent the total number of data in the validation set G2. and y h This represents the output and actual results for the h-th shipping accident data in G2; and we have:

[0042]

[0043] Step 3.5 includes:

[0044] Step 3.5.1: Generate a random number rand in the range (0,1), and obtain the position of the o-th Harris Eagle in the (t+1)-th generation according to equation (5). This yields the location set of the Harris Eagle population in the (t+1)th generation;

[0045]

[0046] Step 3.5.2, if If all elements in the array are 0, then return to step 3.2.1 to re-initialize and assign values ​​to the array.

[0047] Step 3.5.3: Calculate the fitness function value set F of the (t+1)th generation Harris eagle population using equation (2). t+1 The position of the Harris Eagle with the best fitness value is selected and set as the position of the prey in generation t+1.

[0048] Step 3.5.4, from X best and The position of the Harris Eagle with the best fitness value is selected as the position X of the globally optimal individual. best .

[0049] Step 3.6 includes:

[0050] Step 3.6.1: Use equation (6) to obtain the updated position of the o-th Harris Eagle in the (t+1)-th generation.

[0051]

[0052] In equation (6), f represents the refractive index, and σ represents the length ratio of the incident ray to the refracted ray;

[0053] Step 3.6.2, if If all elements in the array are 0, then return to step 3.2.1 to re-initialize and assign values ​​to the array.

[0054] Step 3.6.3: Calculate using equation (2) fitness function value Thus, the fitness function value set F of the (t+1)th generation Harris eagle population is obtained. ′t+1 The position of the Harris Eagle with the best fitness value is selected and set as the position of the prey in generation t+1.

[0055] Step 3.6.4, from X best and The position of the Harris Eagle with the best fitness value is selected as the position X of the globally optimal individual. best .

[0056] The present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a program that supports the processor in executing the shipping accident risk prediction method, and the processor is configured to execute the program stored in the memory.

[0057] The present invention discloses a computer-readable storage medium on which a computer program is stored, wherein the computer program, when executed by a processor, performs the steps of the shipping accident risk prediction method.

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0059] 1. This invention proposes a selective ensemble method that integrates K-means++ and the Harris Eagle optimization algorithm based on back-refractive learning for complex, high-dimensional, large-scale, time-sensitive, and dynamically changing shipping data. This method achieves a balance between the diversity of base models and average accuracy, and innovatively combines clustering algorithms and heuristic algorithms to solve the practical problem of shipping accident risk prediction.

[0060] 2. This invention employs the K-means++ algorithm to detect abnormal data in shipping information and to select diverse base models for shipping accident risk prediction. Compared to the K-means algorithm, K-means++ is highly sensitive to the selection of initial cluster centers, which can better disperse the initial positions of cluster centers, eliminate redundant base models, and provide a smaller and more diverse set of shipping accident risk prediction base models for the second stage of accurate selection. Simultaneously, it avoids the problem of the K-means algorithm easily getting trapped in local optima, thus improving the diversity of the set of shipping accident risk prediction base models.

[0061] 3. This invention addresses the limitations of the standard Harris Eagle algorithm in solving discrete optimization problems and those with specific constraints. It innovatively proposes a novel Harris Eagle optimization algorithm based on back-refractive learning. First, a chaotic initialization mechanism is designed to ensure that Harris Eagle individuals are uniformly distributed in the feasible solution space. Second, a novel energy escape strategy based on biological principles is proposed, using a sine function to simulate the escape behavior of prey, avoiding the algorithm from getting trapped in local optima and accelerating convergence. Finally, the back-refractive learning mechanism is used to search for reverse individuals in the population within the solution space, thereby improving the accuracy of integrated prediction of shipping accident risks. This method can discover the correlation and dynamics of shipping accident data and select the base model with superior overall performance, thus improving the accuracy of integrated prediction of shipping accident risks and effectively saving computational resources. Simultaneously, it enhances the diversity among base models in the integrated prediction, avoiding the problem of getting trapped in local optima and ensuring the independence of base models. Attached Figure Description

[0062] Figure 1 This is a flowchart of the K-means++ algorithm of this invention;

[0063] Figure 2 This is a schematic diagram of the reverse refraction learning mechanism of the present invention;

[0064] Figure 3 This is a flowchart of the method of the present invention. Detailed Implementation

[0065] In this embodiment, as Figure 3As shown, a heuristic two-stage selective integration-based shipping accident risk prediction method proceeds according to the following steps:

[0066] Step 1: We reviewed shipping accident investigation reports from Shanghai, Liaoning and Guangdong between 2005 and 2014 to obtain a dataset G containing s shipping accidents. This dataset includes 20 risk factors such as accident type, accident location and time, weather and sea conditions, visibility, ship attributes and accident causes. The dataset is divided into a training set G1 and a validation set G2 in a 3:2 ratio.

[0067] The Bootstrap technique is used to repeatedly sample the shipping accident training set G1 n times to obtain n sampled datasets. These datasets are then used to train n ELM classifiers on the training set G1, resulting in n base models for shipping accident risk prediction, which together form the original base classification pool. The input weights and biases of the ELM models are randomly given. While preserving the characteristics of the neural network, the output weights are calculated and obtained using regularization principles.

[0068] Step 2: Based on the K-means++ algorithm, calculate the distance between the predicted shipping accident risk values ​​output by each shipping accident risk prediction base model, thereby pre-selecting the original base classification pool to obtain n' shipping accident risk prediction base models with significant differences; for example... Figure 1 As shown, the K-means++ algorithm can effectively handle complex, high-dimensional, time-sensitive, and nonlinear data in shipping accidents. By clustering the features of the output distance, it can discover the features of correlation and dynamics, which helps to obtain base models with large differences.

[0069] Step 2.1: Input the validation set G2 into the n shipping accident risk prediction base models respectively for processing to obtain n shipping accident risk prediction values, and calculate the prediction accuracy of the n shipping accident risk prediction base models;

[0070] Define a variable z and initialize z = 1;

[0071] Randomly select a shipping accident risk prediction value as the z-th cluster center c z And form a cluster center set c = {c z Define the number of cluster centers in c as α, and initialize α = 1;

[0072] Step 2.2: Calculate the remaining n-α shipping accident risk prediction values ​​and their correlation with the z-th cluster center c. z The Euclidean distance between them forms the z-th distance set D. z ={d z,1 ,d z,2 ,...,d z,i ,...,dz,n-α}, where d z,i This represents the predicted value of the i-th shipping accident risk and the cluster center c of the z-th cluster. z The Euclidean distance between them;

[0073] Step 2.3: Calculate the i-th accumulated distance. Where, d z,i′ This represents the predicted shipping accident risk value of the i'th shipping accident and the cluster center c of the zth cluster. z The Euclidean distance is obtained; thus, the z-th cumulative distance set D is obtained. z,s ={v z,1 ,v z,2 ,...,v z,i ,v z,n-α};

[0074] Step 2.4, from D z,s Randomly select an accumulated distance v z Then, subtract set D in sequence. z Each element in the list, and determine v. z -d z,i Is ≤0 true? If true, then set d... z,i Add the corresponding shipping risk forecast value to set c; otherwise, proceed directly to step 2.5.

[0075] Step 2.5: Calculate the minimum Euclidean distance d from the i-th predicted shipping accident risk value to each cluster center in c. i,min Thus, the set of minimum Euclidean distances D is obtained. min ={d 1,min ,d 2,min ,...,d i,min ,...,d n-α,min};

[0076] Step 2.6: Calculate the minimum Euclidean distance set D min The i-th minimum Euclidean distance d i,min Probability of being selected Therefore, based on the probability of each minimum Euclidean distance being selected, the roulette wheel algorithm is used to select from D. min The minimum Euclidean distance d is determined in the final selection. last,min , thus making d last,min Add the corresponding shipping accident risk prediction value to set c; otherwise, proceed directly to step 2.7.

[0077] Step 2.7: Determine whether the number of cluster centers in set c has reached k. If it has, proceed to step 2.8; otherwise, return to step 2.4.

[0078] Step 2.8: Calculate the Euclidean distance between the remaining nk shipping accident risk prediction values ​​and the k cluster centers, and then assign the nk shipping accident risk prediction values ​​to the clusters to which the nearest cluster centers belong based on the Euclidean distance;

[0079] Step 2.9: Select the base models corresponding to the top w shipping accident risk prediction values ​​with high prediction accuracy from each cluster as the n' shipping accident risk prediction base models with greater differences.

[0080] Step 3: Use the improved Harris Eagle optimization algorithm to perform a secondary selection on the n' shipping accident risk prediction base models until the optimal combination of base models is found, and this combination is used as the shipping accident risk prediction model base combination scheme for prediction of shipping accident risks. The Harris Eagle algorithm can handle discretization problems and avoids the situation of easily getting trapped in local optima due to uneven initial population distribution. The steps are as follows:

[0081] Step 3.1: Define the current iteration number as t and initialize t = 1; define the maximum iteration number as T; and the population size as pop; randomly initialize the position set of the Harris Eagle population in the t-th generation within the search space. This represents the position of the o-th Harris Eagle in generation t, and signifies the selection of different shipping accident risk prediction base models from n' shipping accident risk prediction base models to form a model combination scheme; and lb and ub represent the lower and upper bounds of the search space, respectively;

[0082] Step 3.2: During the initialization phase, the positions of individual Harris Eagles in each dimension are randomly generated, leading to clustering of initial solutions and low diversity among individuals. To improve the problem and increase the diversity of the initial population, a chaotic initialization mechanism is introduced into the n' shipping accident risk prediction base models to initialize the population and update the position set of the t-th generation Harris Eagle population. The steps are as follows:

[0083] Step 3.2.1: Use equation (1) to determine the position of the o-th Harris Eagle in the t-th generation. Perform an update to obtain the updated location.

[0084]

[0085] In equation (1), δ is the chaos factor, and mod represents the remainder;

[0086] Step 3.2.2: Repeat pop 10 times, if... If all elements in the initialization list are 0, then return to step 3.2.1 to re-initialize.

[0087] Step 3.3: Calculate the fitness function value of the o-th Harris Eagle in the t-th generation using equation (2). Thus, the fitness function value of the t-th generation Harris Eagle population is obtained. Set the position of the Harris Eagle with the best fitness value as the position of the prey in generation t. And the position X of the globally optimal individual. best ;

[0088]

[0089] In equation (3), Let λ represent the prediction accuracy of the position of the o-th Harris Hawk in the t-th generation, and let λ represent the total number of data in the validation set G2. and y h This represents the output and actual results for the h-th shipping accident data in G2; and we have:

[0090]

[0091] Step 3.4: Calculate the escape energy of the prey in the t-th iteration according to equation (4) to update the prey escape energy value. The escape energy update formula proposed in this invention takes into account the reality of multiple rounds of interaction between the eagle and the prey. In the early stage, the prey has abundant escape energy. As the escape distance increases, the escape energy value decreases rapidly, as shown in the following formula:

[0092]

[0093] In equation (4), E0 represents the initial escape energy of the prey.

[0094] Step 3.5: Update with the improved location update strategy It can improve the algorithm's optimization performance and search efficiency:

[0095] Step 3.5.1: Generate a random number rand in the range (0,1), and obtain the position of the o-th Harris Eagle in the (t+1)-th generation according to equation (5). This yields the location set of the Harris Eagle population in the (t+1)th generation;

[0096]

[0097] Step 3.5.2, if If all elements in the array are 0, then return to step 3.2.1 to re-initialize and assign values ​​to the array.

[0098] Step 3.5.3: Calculate the fitness function set F of the (t+1)th generation Harris eagle population based on step 3.3. t +1The position of the Harris Eagle with the best fitness value is selected and set as the position of the prey in generation t+1.

[0099] Step 3.5.4, from X best and The position of the Harris Eagle with the best fitness value is selected as the position X of the globally optimal individual. best .

[0100] Step 3.6: Utilize the back-refractive learning mechanism to find the reverse individuals of the population within the search space. Back-refractive learning is an algorithm that combines the phenomenon of light refraction with back-learning. In physics, when light travels from one medium to another, the angle of incidence and the angle of refraction change. Figure 3 As shown, the search interval for the solution on the x-axis is [lc, le], representing the boundary of the Harris Hawk search space; the y-axis represents the normal, l and l * θ1 and θ2 represent the lengths of the incident and refracted rays, respectively, and the angles of incidence and refraction are θ1 and θ2, respectively; X represents the current solution to the problem. * The reversed individual obtained through the reverse refraction mechanism has the following sine values ​​for the incident angle and the reflection angle:

[0101]

[0102] like Figure 2 As shown, σ is the incident ray l and the refracted ray l * The length ratio, where f is the refractive index of the light, is shown in the following formula:

[0103]

[0104]

[0105] Based on the law of refraction, we can deduce that:

[0106]

[0107] Step 3.6.1: Use equation (6) to obtain the updated position of the o-th Harris Eagle in the (t+1)-th generation.

[0108]

[0109] Step 3.6.2, if If all elements in the array are 0, then return to step 3.2.1 to re-initialize and assign values ​​to the array.

[0110] Step 3.6.3: Calculate the fitness function set F' of the (t+1)th generation Harris eagle population based on step 3.3. t+1 The position of the Harris Eagle with the best fitness value is selected and set as the position of the prey in generation t+1.

[0111] Step 3.6.4, from X best and The position of the Harris Eagle with the best fitness value is selected as the position X of the globally optimal individual. best ;

[0112] Step 3.7: Determine whether t = T holds true. If it does, use the Harris Eagle individual corresponding to the global optimal fitness value as the selection scheme for the combination of shipping accident risk prediction base models to construct the integrated prediction model for shipping accident risk. Otherwise, assign t+1 to t and return to step 3.3 to execute sequentially.

[0113] Step 4: Integrate the optimal shipping accident risk prediction base model combination scheme obtained in the above steps using the majority voting method, thereby constructing a shipping accident risk integrated prediction model. This model is used to predict shipping accident risks from high-dimensional, complex, and time-sensitive input shipping data, which helps to improve the shipping accident early warning mechanism and enhance shipping safety.

[0114] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0115] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. A heuristic-based two-stage selective integration method for predicting shipping accident risks, characterized in that, The procedure is as follows: Step 1: Obtain the shipping accident dataset and divide it into a training set G1 and a validation set G2; The training set G1 is sampled n times using the Bootstrap method to obtain n sampled datasets, which are then used to train n ELM classifiers to obtain n shipping accident risk prediction base models, which together form the original base classification pool. Step 2: Based on the K-means++ algorithm, pre-select the original base classification pool to obtain n' shipping accident risk prediction base models with significant differences; Step 3: Use the improved Harris Eagle optimization algorithm to perform a secondary selection on the n' shipping accident risk prediction base models until the optimal combination of base models is found and used for shipping accident risk prediction: Step 3.1: Define the current iteration number as t and initialize t = 1; define the maximum iteration number as T; and the population size as pop; randomly initialize the position set of the Harris Eagle population in the t-th generation within the search space. This represents the position of the o-th Harris Eagle in generation t, and signifies the selection of different shipping accident risk prediction base models from n' shipping accident risk prediction base models to form a model combination scheme; and lb and ub represent the lower and upper bounds of the search space, respectively; Step 3.2: Introduce a chaotic initialization mechanism into the n' shipping accident risk prediction base models to update the position set of the t-th generation Harris Eagle population and obtain the updated position set; Step 3.3: Select the position of the Harris Eagle with the best fitness value from the updated position set as the position X of the globally optimal individual. best ; Step 3.4: Calculate the escape energy E of the prey in generation t according to equation (4). t : In equation (4), E0 represents the prey's initial escape energy; Step 3.5: Update the updated position set again using the improved position update strategy, and after obtaining the (t+1)th generation Harris eagle population, select the position of the Harris eagle with the best fitness value as the position X of the globally optimal individual. best ; Step 3.6: Utilize the back-refractive learning mechanism to search for the back-refractive individuals of the (t+1)th generation Harris eagle population within the search space, obtaining the updated position set of the (t+1)th generation Harris eagle population, and selecting the position of the Harris eagle with the best fitness value from this set as the position X of the optimal individual. best ; Step 3.7: Determine whether t = T holds true. If it does, use the Harris Eagle individual corresponding to the global optimal fitness value as the selection scheme for the combination of shipping accident risk prediction base models and use it to construct the optimal shipping accident risk integrated prediction model. Otherwise, assign t+1 to t and return to step 3.4 to execute sequentially. Step 4: Use the majority voting method to integrate the combination schemes of the optimal shipping accident risk prediction base model, thereby constructing a shipping accident risk integrated prediction model, which is used to predict shipping accident risks based on input shipping data.

2. The shipping accident risk prediction method based on heuristic two-stage selective integration according to claim 1, characterized in that, Step 2 includes: Step 2.1: The validation set G2 is input into n shipping accident risk prediction base models for processing to obtain n shipping accident risk prediction values, and the prediction accuracy of the n shipping accident risk prediction base models is calculated. Define a variable z and initialize z = 1; Randomly select a shipping accident risk prediction value as the z-th cluster center c z And form a cluster center set c = {c z Define the number of cluster centers in c as α, and initialize α = 1; Step 2.2: Calculate the remaining n-α shipping accident risk prediction values ​​and their correlation with the z-th cluster center c. z The Euclidean distances between them form the z-th distance set D. z ={d z,1 ,d z,2 ,...,d z,i ,...,d z,n-α }, where d z,i This represents the predicted shipping accident risk value of the i-th shipping accident and the cluster center c of the z-th cluster. z The Euclidean distance between them; Step 2.3: Calculate the i-th accumulated distance. Where, d z,i′ This represents the predicted shipping accident risk value of the i'th shipping accident and the cluster center c of the zth cluster. z The Euclidean distance is obtained; thus, the z-th cumulative distance set D is obtained. z,s ={v z,1 ,v z,2 ,...,v z,i ,v z,n-α }; Step 2.4, from D z,s Randomly select an accumulated distance v z Then, subtract set D in sequence. z Each element in the list, and determine v. z -d z,i Is ≤0 true? If true, then set d... z,i Add the corresponding shipping risk forecast value to set c; otherwise, proceed directly to step 2.

5. Step 2.5: Calculate the minimum Euclidean distance d from the i-th predicted shipping accident risk value to each cluster center in c. i,min Thus, the set of minimum Euclidean distances D is obtained. min ={d 1,min ,d 2,min ,...,d i,min ,...,d n-α,min }; Step 2.6: Calculate the minimum Euclidean distance set D min The i-th minimum Euclidean distance d i,min Probability of being selected Therefore, based on the probability of each minimum Euclidean distance being selected, the roulette wheel algorithm is used to select from D. min The minimum Euclidean distance d is determined in the final selection. last,min , thus making d last,min Add the corresponding shipping accident risk prediction value to set c; otherwise, proceed directly to step 2.

7. Step 2.7: Determine whether the number of cluster centers in set c has reached k. If it has, proceed to step 2.8; otherwise, assign z+1 to z and return to step 2.

2. Step 2.8: Calculate the Euclidean distance between the remaining nk shipping accident risk prediction values ​​and the k cluster centers, and then assign the nk shipping accident risk prediction values ​​to the clusters to which the nearest cluster centers belong based on the Euclidean distance; Step 2.9: Select the base models corresponding to the top w shipping accident risk prediction values ​​with high prediction accuracy from each cluster as the n' shipping accident risk prediction base models with greater differences.

3. The shipping accident risk prediction method based on heuristic two-stage selective integration according to claim 2, characterized in that, Step 3.2 includes: Step 3.2.1: Use equation (1) to determine the position of the o-th Harris Eagle in the t-th generation. Perform an update to obtain the updated location. In equation (1), δ is the chaos factor, and mod represents the remainder; Step 3.2.2, if If all elements in the initialization list are 0, then return to step 3.2.1 to re-initialize.

4. The shipping accident risk prediction method based on heuristic two-stage selective integration according to claim 3, characterized in that, Step 3.3 includes: The fitness function value of the o-th Harris Eagle in the t-th generation is calculated using equation (2). Thus, the fitness function value of the t-th generation Harris Eagle population is obtained. Set the position of the Harris Eagle with the best fitness value as the position of the prey in generation t. And the position X of the globally optimal individual. best ; In equation (2), Let λ represent the prediction accuracy of the position of the o-th Harris Hawk in the t-th generation, and let λ represent the total number of data in the validation set G2. and y h This represents the output and actual results for the h-th shipping accident data in G2; and we have:

5. The shipping accident risk prediction method based on heuristic two-stage selective integration according to claim 4, characterized in that, Step 3.5 includes: Step 3.5.1: Generate a random number rand in the range (0,1), and obtain the position of the o-th Harris Eagle in the (t+1)-th generation according to equation (5). This yields the location set of the Harris Eagle population in the (t+1)th generation; Step 3.5.2, if If all elements in the array are 0, then return to step 3.2.1 to re-initialize and assign values ​​to the array. Step 3.5.3: Calculate the fitness function value set F of the (t+1)th generation Harris eagle population using equation (2). t+1 The position of the Harris Eagle with the best fitness value is selected and set as the position of the prey in generation t+1. Step 3.5.4, from X best and The position of the Harris Eagle with the best fitness value is selected as the position X of the globally optimal individual. best .

6. The shipping accident risk prediction method based on heuristic two-stage selective integration according to claim 5, characterized in that, Step 3.6 includes: Step 3.6.1: Use equation (6) to obtain the updated position of the o-th Harris Eagle in the (t+1)-th generation. In equation (6), f represents the refractive index, and σ represents the length ratio of the incident ray to the refracted ray; Step 3.6.2, if If all elements in the array are 0, then return to step 3.2.1 to re-initialize and assign values ​​to the array. Step 3.6.3: Calculate using equation (2) fitness function value Thus, the fitness function value set F′ of the (t+1)th generation Harris eagle population is obtained. t+1 The position of the Harris Eagle with the best fitness value is selected and set as the position of the prey in generation t+1. Step 3.6.4, from X best and The position of the Harris Eagle with the best fitness value is selected as the position X of the globally optimal individual. best .

7. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store programs that support the processor in executing any of the shipping accident risk prediction methods of claims 1-6, and the processor is configured to execute the programs stored in the memory.

8. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is run by the processor, it performs the steps of the shipping accident risk prediction method according to any one of claims 1-6.

Images