A ship transverse section structure optimization method based on improved bee colony algorithm

Abstract

The application discloses a ship cross section structure optimization method based on an improved bee colony algorithm. First, the discrete structure model is converted into a continuous optimization model based on a component library index, thereby laying a foundation for application of the bee colony algorithm. Secondly, in order to solve the latitude explosion caused by high-dimensional optimization, a uniformized Logistic mapping is designed for component section size initialization, so as to improve the randomness and ergodicity of the decision space. Since the bee colony algorithm has the problem of being good at exploration but poor at exploitation, the individual goodness and algorithm degree are considered, and the dynamic leading bee search and follower following strategy are respectively reconstructed, so as to balance the search ability of the algorithm and enhance the optimization quality. Finally, based on the section modulus sensitivity analysis, the continuous value obtained through iteration is normalized, so as to meet the actual engineering requirements. The method can adaptively adjust the global search and local search tendency of each individual, improves the efficiency and result precision of the ship structure optimization, and has practical value.

Description

Technical Field

[0001] This invention relates to the field of ship structure optimization, and more specifically, to a method for optimizing the cross-sectional structural dimensions of inland waterway vessels based on an improved bee colony algorithm. Background Technology

[0002] Inland waterway transportation has been a vital medium for the exchange of goods and economic cooperation since ancient times. With the development of global trade and multimodal transport technologies, water transport, compared to land transport, has attracted increasing attention due to its significant advantages of lower energy consumption, larger capacity, smaller footprint, and lower pollution. In 2020, inland waterway freight volume reached 3.8 billion tons, exceeding that of rail freight and reaching a ratio of 13:1 with ocean freight, while competition in the water transport market has returned to its peak. Meanwhile, the China Classification Society (CCS) has been continuously issuing new standards in recent years, imposing higher requirements on various indicators of inland waterway vessels. Faced with stricter regulations and a more severe competitive environment, improving the design technology of inland waterway vessels, accelerating design efficiency, and refining design methods are crucial to further promoting the high-quality development of inland waterway transportation and are also important directions for the reform of the shipbuilding industry in the new era.

[0003] To meet the diverse performance requirements of ships, structural optimization design has emerged, leveraging the material foundation provided by computational mathematics and computing power. Ship structural optimization seeks a more rational arrangement, form, and size of the ship's structure, enabling it to meet various requirements such as strength, stability, and natural hull frequency, while simultaneously possessing ideal economic, mechanical, or maneuvering performance.

[0004] In existing technologies, whether the standard design method is used directly or the margin interpolation method is used indirectly, the result is only a general feasible solution for the structural dimensions, not the optimal solution.

[0005] Research on optimization based on population intelligence algorithms such as genetic algorithms has yielded considerable results. Current technologies tend to focus on the application and transformation of actual ship structures into optimization models, such as considering discrete structural variables or constructing algorithm boundaries through actual constraints. However, research on the algorithm's optimization mechanism is somewhat lacking. In fact, as the main body of the optimization method, the algorithm's own optimization mechanism is also the key to determining the final optimization quality. Especially when facing ship structure optimization problems where the dimensions and objective function convexity and concavity of various models are very different, in addition to modifying the optimization model of the algorithm to match the ship structure, improving the optimization strategy of the optimization algorithm itself should be a more universally applicable solution. Summary of the Invention

[0006] To address the aforementioned issues, this invention establishes a method for optimizing ship cross-sectional structures based on an improved bee colony algorithm. The key improvement lies in the algorithm's optimization strategy, thereby enhancing the efficiency and accuracy of ship structure optimization.

[0007] A method for optimizing the cross-sectional structure of a ship based on an improved bee colony algorithm includes the following steps: Based on the component library index, the discrete structural model is converted into a continuous optimization model; Determine the design variables, constraints, and objective function; Design a uniform Logistic mapping to initialize design variables; Reconstruct the dynamic search strategies for leading bees and the following strategies for trailing bees, respectively; The continuous eigenvalues ​​obtained through iterative solutions are converted into usable structural models.

[0008] Preferably, the design variables are the thickness of all longitudinal plates within the section and the type of longitudinal stiffeners on the plates; the objective function is the total section area or the section orthocentric height; the constraints are determined by ship specifications, and the constraints are set to be within 20% above and below the original design values ​​of the design variables.

[0009] The design uniformity logistic mapping used for design variable initialization can be expressed as follows:

[0010] In the formula, for Random initial values ​​between; To finally initialize the generated decision variables, and These represent the upper and lower boundaries of the variables in the actual optimization model. Iterating over the expression within parentheses Second-rate, Take 150-250.

[0011] The method of converting discrete structural models into continuous optimization models based on component library indexes specifically involves:

[0012] in, and These refer to the longitudinal plate thickness and the longitudinal aggregate type, respectively.

[0013] The optimization state variables that correspond one-to-one with the design variables are:

[0014] in, The index representing the plate thickness in the plate library; The index of the standard aggregate model in the profile library; The section modulus is for non-standard structures.

[0015] The component library is a standard component knowledge base that includes the characteristics of sheet metal and profiles. The sheet metal characteristic is the sheet thickness. A series of sheet metal thicknesses are entered into the sheet metal library with 0.5mm as the smallest unit. The profile characteristics are cross-sectional attributes including cross-sectional dimensions and section modulus.

[0016] The process of converting the continuous eigenvalues ​​obtained through iterative solutions into usable structural models includes standard structural standardization and sensitivity-based custom structural standardization. The standard structural standardization strategy is to traverse the component library and select the structure in the library whose eigenvalues ​​are closest to the final solution. For plates, the required eigenvalue is the plate thickness, and for profiles, the required eigenvalue is the section modulus. The custom structural standardization strategy is to seek the eigenvalue, i.e., the optimization variable, that has the greatest impact on the section dimensions of the custom structure.

[0017] Compared with the prior art, the present invention has the following beneficial effects: First, to address the dimensional explosion caused by high-dimensional optimization, a homogenized Logistic mapping is designed for component profile size initialization to improve the randomness and ergodicity of the decision space.

[0018] Secondly, since the bee colony algorithm is good at exploration but poor at mining, considering the individual quality and the algorithm's progress, we reconstruct the dynamic leader bee search strategy and follower bee follow strategy respectively, so as to balance the algorithm's search ability and enhance its ability to find quality.

[0019] Finally, based on the profile modulus sensitivity analysis, the continuous values ​​obtained by the iterative solution are normalized to meet the actual engineering requirements.

[0020] In summary, this method can adaptively adjust the global and local search tendencies of each individual, improving the efficiency and accuracy of ship structure optimization, and has practical value. Attached Figure Description

[0021] Figure 1 This is a flowchart of the optimization method of the present invention; Figure 2 This is a schematic diagram of the ship profile library in this invention; Figure 3 This is a schematic diagram of the uniform Logistic mapping particle distribution in the bee colony algorithm of this invention; Figure 4 This is a schematic diagram illustrating the changing trend of the global search capability of the bee colony algorithm in this invention; Figure 5 This is a schematic diagram illustrating the changing trend of the nectar source selection probability in the bee colony algorithm of this invention; Figure 6 This is a flowchart of the recommended T-profile dimensions in this invention; Figure 7 It is the convergence curve of the test function optimization; Figure 8 It is an instance optimization object graph; Figure 9 It is the convergence curve of the actual ship optimization. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described clearly and completely below with reference to the accompanying drawings and embodiments. It should be understood that the described embodiments are merely some embodiments of a method for optimizing the cross-sectional structural dimensions of inland waterway vessels according to this invention, and not all embodiments, and are not intended to limit the invention.

[0023] Figure 1 This is a flowchart of the optimization method of the present invention. The following is an exemplary description of the structural optimization method of the present invention.

[0024] Step 1: Based on the component library index, the discrete structural model is converted into a continuous optimization model, thus laying the foundation for the application of the bee colony algorithm.

[0025] The specific implementation process is as follows: To enable the Bee Colony Algorithm (ABC) to be used for structural specification optimization, a library of standard ship components is first established.

[0026] The knowledge stored and managed in the sheet metal library is the thickness of the hull plates, as shown in Table 1. Plate thickness is limited by process conditions; based on construction experience, information on a series of plates is entered into the sheet metal library in units of 0.5mm thickness.

[0027] Table 1 Shipboard Plate Inventory

[0028] The profile library offers a variety of standard and custom profiles. Its stored feature data consists of the cross-sectional properties of each profile, which are associated with its model number, such as... Figure 2 As shown. Sectional parameters calculated from the cross-sectional dimensions, such as cross-sectional area, section modulus, and radius of inertia, are included in the library and extracted by model number during calculation.

[0029] As shown in Table 2, the profile library classifies, stores, and manages profile information according to Chinese national standards (GB) and cross-sectional shapes.

[0030] Table 2 Profile Information

[0031] Based on the above component library, an optimization model including design variables, constraints, and objective functions is established: (1) Design variables Since transverse members are not included in the overall longitudinal strength calculation, the optimization targets are the thickness of all longitudinal plates within the section and the type of longitudinal stiffeners on the plates. For standard structural components such as plates and angle steel, their specifications are mapped to library indexes through the component library. For custom profiles, their section modulus is directly optimized, and then converted to the corresponding model based on structural recommendations. After processing by the component library, discrete dimensions are transformed into continuous positive integer state quantities to address the challenging variable-step discrete optimization problem.

[0032] Conversion Method: The XML-formatted component knowledge base is deserialized into a data structure supporting key-value lookup. Structural information in the plate and profile libraries is then sorted in ascending order based on plate thickness and section modulus, respectively. In this invention, the plate thickness is the variable for plate materials, and the section modulus is the variable for profile materials. This allows the use of positive integer library indices corresponding to the initial structural variables to replace the variable-step-size normalized variables for optimization iteration.

[0033] Finally, the design variables are as follows:

[0034] in, and These refer to the longitudinal plate thickness and the longitudinal aggregate type, respectively.

[0035] The optimization state variables that correspond one-to-one with the design variables are:

[0036] in, The index representing the plate thickness in the plate library; The index of the standard aggregate model in the profile library; and For non-standard structures such as T-profiles, the section modulus is used.

[0037] (2) Constraints In addition to the overall longitudinal strength and local strength requirements specified in the standards, from the perspective of margin preservation and optimization efficiency, the range of variation of variables is limited to within 20% above and below the original design value. Furthermore, as a strip plate, the plate thickness is restricted by the aggregate specifications. The comprehensive constraints are as follows:

[0038] In the formula, For the first The original design thickness of the plate; For the first Index of plate thickness for plate number; and These are the upper and lower limits of the plate thickness mapped to the plate library index, respectively. The minimum thickness of the strip plate is calculated using the standard module of the aggregate. For the plate thickness library index of plate j, similarly, for national standard aggregates, their section modulus limits need to be converted into index boundaries. ; For the first Design section modulus of non-standard aggregates and The first Standard bone material and the first Customize the section modulus of the aggregate. , For the minimum modulus and minimum moment of inertia of the section, and , , , These are the longitudinal and local strength requirements in the "Code for Construction of Inland Waterway Vessels 2016".

[0039] (3) Objective function The initial optimization objective of this method is to minimize the total cross-sectional area, aiming at achieving lightweight structural design. The objective function is specifically expressed as:

[0040] in The total cross-sectional area, ; This represents the number of longitudinal plates. The number of longitudinal stiffeners; the remaining variables are plate width. Ship length dimensions All units are Profile cross-sectional area Calculated by model number.

[0041] Furthermore, the present invention can also set other objective functions according to actual needs, such as setting the objective function as the orthocenter in order to improve stability.

[0042] Step 2: Design a homogenized Logistic mapping for component profile size initialization to improve the randomness and ergodicity of the decision space.

[0043] This method uses a Logistic chaotic mapping instead of a pseudo-random number generator for component profile size initialization. The Logistic mapping model is shown below:

[0044] In the formula, To control the amount, when At this point, the Logistic mapping results in a completely chaotic state. Take... .

[0045] Because the Logistic mapping is not uniform and exhibits clustering at the boundaries, and for the optimization function, solutions located at the domain boundaries are often inferior solutions, far from the optimum, a universal probabilistic homogenization method is used to ensure that the random numbers generated by the mapping follow a uniform distribution, as shown in the following equation:

[0046] In the formula, for Random initial values ​​between; perform the same number of iterations on the above equation, and the particles... Value distribution as Figure 3 As shown, the Logistic mapping at this time is... It exhibits good uniformity within the interval.

[0047] Finally, the component section size initialization strategy is as follows:

[0048] In the formula, The generated decision variables are used for final initialization; and These are the upper and lower boundaries of the variables in the actual optimization model. For chaotic mapping iteration Secondly, its value is 150-250. In a preferred embodiment, Take 200.

[0049] Step 3: Considering individual quality and algorithm progress, reconstruct the dynamic leader bee search and follower bee follow strategies respectively, thereby balancing the algorithm's search ability and enhancing its ability to find quality.

[0050] Adaptive search strategy: To improve the local search capability of ABC, for example, a Q-ABC algorithm guided by the optimal nectar source, the bee search behavior is as follows:

[0051] in for The optimal nectar source in the current generation of the population. Experiments showed that the strategy of directly jumping to the optimal point easily leads to local optima. Therefore, the optimal solution information is combined with the original algorithm's random search.

[0052] First, define the algorithm's degree of progress. :

[0053] In the formula, , for The upper and lower limits, taking values ​​of between; This represents the current iteration number of the algorithm. This represents the maximum number of iterations. It's evident that towards the end of the algorithm... The larger.

[0054] Define individual rank Alternative fitness This is to give greater distinction between different solutions. For the first The ranking of an individual's fitness within the population. Clearly, .right Normalization process:

[0055] visible It can characterize the degree of excellence of an individual, and is related to They are of the same order of magnitude.

[0056] The fitness calculation remains the same as the original algorithm, as shown below:

[0057] In the formula, For fitness; This represents the objective function value.

[0058] At this point, an adaptive factor can be defined based on the algorithm's progress and the individual's excellence. :

[0059] use By adjusting the weights of global and local search capabilities, the new solution search equation is as follows:

[0060] when hour, The provided positive disturbance makes Towards the current best point Convergence, and When the above formula is used, it is equivalent to a random search, that is, it follows the ABC search strategy.

[0061] Figure 4The changing trend of the global search ability weight is presented. It can be seen that in the early stages of the algorithm, the population tends to perform global searches to provide more potential high-quality directions for subsequent optimization. Gradually, as the algorithm progresses into its later stages, the population's local search ability rapidly increases, leading to a quick convergence to the optimal solution. Inferior individuals in the population tend to approach the current optimal solution at any stage of the algorithm, while superior individuals are more likely to perform random searches to avoid remaining in local optima for extended periods.

[0062] Adaptive following strategy: Using rank And the degree of algorithm The selection pressure of nectar sources is adaptively adjusted. The probability function is shown below:

[0063] when Take 0.2, When the value is 0.8, that is... The trend of probability change of choosing to follow is as follows Figure 5 As shown: It is evident that for honey sources with similar honey yields, the probability difference increases significantly with the number of algorithm iterations, enhancing the greediness of the following behavior. In the early stages of the algorithm, regardless of fitness, the probability of selecting all solutions is distributed almost evenly, ensuring that even inferior solutions are fully exploited. This aligns with the premise of the aforementioned adaptive search mechanism.

[0064] Step 4: Based on the profile modulus sensitivity analysis, normalize the continuous values ​​obtained from the iterative solution to meet the actual engineering requirements.

[0065] The structural recommendation algorithm transforms continuous feature values ​​into reasonable and usable structural models. For standard profiles, the available specifications are limited, and the thickness of the sheet metal determines its model number. Therefore, for both, the library component that best matches the required parameters can be directly selected. Recommending custom profiles is slightly more complex because a profile model represents multiple dimensional parameters. These dimensional parameters, used to calculate the characteristic values, are defined by the designer. When a required value is received, custom profiles in the library often have excessive strength, leading to material waste, or even no suitable model exists.

[0066] consider The minimum section modulus of the T-profile with connecting plate of the model. The following can be calculated:

[0067] in , , , These are the web thickness, web height, face thickness, and face width of the T-section. , These are the thickness and width of the plate attached to the profile, in units of... The bottom of the panel is the reference axis. , where is the cross-sectional area of ​​the web; , where is the cross-sectional area of ​​the panel; , where is the cross-sectional area with plate. All units are... .

[0068] Let the dimensionless coefficient ,because The variation range is small, and the strip is usually not changed when solving for the recommended dimensions of the profile. Therefore, from the perspective of saving material, increasing the web height is... It is to improve The most effective method is to increase the panel cross-sectional area. Secondly, considering that the panel width is much greater than its thickness, then... The increase can be converted into increased panel thickness. .

[0069] The normalization of continuous values ​​obtained through iterative solutions includes standard structure normalization and sensitivity-based custom structure normalization.

[0070] The standardization strategy for standard structures is as follows: traverse the component library and select the structure in the library whose feature quantity is closest to the final solution. For plates, the required feature quantity is the plate thickness, and for profiles, the required feature quantity is the section modulus.

[0071] The standardization strategy for custom structures is as follows: Taking a T-profile as an example, when the calculated section modulus value is greater than its own modulus, firstly, the most sensitive parameter, web height, is increased in increments of 10mm. Once the percentage difference in modulus is less than 20%, the second most sensitive parameter, panel thickness, is increased in minimum increments of 5mm until the requirements are met. The output model number is then assigned to the structure and written into the profile library.

[0072] In summary, when the required module of the T-profile... Greater than its own modulus At that time, its size generation strategy is as follows Figure 6 As shown.

[0073] Implementation Case: To verify the performance of the improved bee colony algorithm I-ABC in this method, six typical single-objective test functions were selected for experimentation.

[0074] To demonstrate the superiority of I-ABC, the test function was optimized and analyzed using the original ABC algorithm, the local optimization algorithm Q-ABC, and the I-ABC algorithm simultaneously. The parameter settings for the three algorithms were identical: population size... Variable dimensions Number of loops Control parameters New control parameters for I-ABC , All algorithms were run independently 60 times on each function, and the convergence curve of the average optimal value was as follows: Figure 7 As shown.

[0075] As can be seen from the convergence curves above, ABC or Q-ABC only slightly outperforms I-ABC on Schwefel and Sphere, while the I-ABC algorithm achieves the best optimization results on the other four functions. This verifies the superiority of the I-ABC adaptive mechanism and chaotic mapping initialization.

[0076] Because Q-ABC has a strong ability to find new solutions in the neighborhood of the current optimal solution, it is very suitable for handling flat and convex optimization problems like Sphere, but it is prone to getting trapped in local optima on Ackley. ABC, on the other hand, exhibits a completely different performance. Its excessive randomness causes ABC to perform significantly worse than the comparison algorithms on both unimodal functions, but it performs better on multimodal functions, especially on the complex multimodal function Schwefel, where ABC converges to the optimum faster than I-ABC. This shows that balancing the exploration and exploitation capabilities of the algorithm is the correct strategy.

[0077] To verify the reliability and advanced nature of I-ABC in practical engineering, a semi-mid-transverse model of a real ship was selected as the optimization object. This ship is a 112m chemical / liquid oil tanker. Figure 8 As shown, the model has a total of 10 profile variables and 13 plate variables, among which S2, S5, and S10 are T-shaped longitudinal girder variables, which are continuous variables.

[0078] Optimization tests were performed using ABC, Q-ABC, and I-ABC methods respectively. Parameters: , , The number of iterations is set to 300, and the number of variables is determined by the actual model. The optimization curve is shown below. Figure 9 As shown: It is evident that I-ABC maintains its superiority over ABC and Q-ABC in practical ship optimization problems. However, it is worth considering that the ABC algorithm, renowned for its global optimization capabilities, performs almost ineffectively, while Q-ABC, which is prone to getting trapped in local optima, actually performs better. Although ship structure optimization is a multimodal problem, the examples demonstrate that local search capabilities need to be emphasized when dealing with discrete variables with narrow boundaries.

[0079] Table 3 Overall Optimization Results and Comparison

[0080] The overall optimization results are shown in Table 3. It can be seen that, due to the larger size of the plate material, the absolute value of the reduction in its cross-sectional area is significantly greater than that of the aggregate, while the reduction rate is relatively smaller. Ultimately, after optimization, the cross-sectional area of ​​the entire profile decreased by 5.08%.

Claims

1. A method for optimizing the cross-sectional structure of a ship based on an improved bee colony algorithm, characterized in that, Includes the following steps: Determine the design variables, constraints, and objective function; Design a uniform Logistic mapping to initialize design variables; The dynamic leader bee search and follower bee following strategies are reconstructed separately; specifically, the dynamic leader bee search is reconstructed as follows: Define the degree of the algorithm : In the formula, , for The upper and lower limits, taking values ​​of between; This represents the current iteration number of the algorithm. This represents the maximum number of iterations. Define individual rank Alternative fitness This is to give greater distinguishability between different solutions. For the first The ranking of individual fitness within the population. ;right Normalization process: in, Used to characterize the excellence of an individual, and with They are of the same order of magnitude; The fitness calculation is as follows: In the formula, For fitness; The objective function value; An adaptive factor is defined based on the algorithm's progress and the individual's excellence. : use By adjusting the weights of global and local search capabilities, the reconstructed solution search equation is shown below: in, Indicates the current optimal point; The continuous eigenvalues ​​obtained through iterative solutions are converted into usable structural models.

2. The method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in claim 1, characterized in that, The step "Determine design variables, constraints and objective function" also includes converting the discrete structural model into a continuous optimization model based on the component library index.

3. A method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in claim 1 or 2, characterized in that, The design variables are the thickness of all longitudinal plates in the section and the type of longitudinal reinforcement on the plates.

4. A method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in claim 1 or 2, characterized in that, The objective function is the total cross-sectional area or the vertical height of the cross-section.

5. A method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in claim 1 or 2, characterized in that, The constraints are determined by ship specifications.

6. The method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in claim 5, characterized in that, The constraint is set to be within 20% above and below the original design value of the design variable.

7. The method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in claim 1, characterized in that, The design uniformity logistic mapping used for design variable initialization can be expressed as follows: In the formula, for Random initial values ​​between; To finally initialize the generated decision variables, and These represent the upper and lower boundaries of the variables in the actual optimization model. Iterating over the expression within parentheses Second-rate, Take 150-250.

8. The method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in claim 2, characterized in that, The method of converting discrete structural models into continuous optimization models based on component library indexes specifically involves: in, and These refer to the longitudinal plate thickness and the longitudinal aggregate type, respectively. The optimization state variables that correspond one-to-one with the design variables are: in, The index representing the plate thickness in the plate library; The index of the standard aggregate model in the profile library; The section modulus is for non-standard structures.

9. The method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in claim 2, characterized in that, The component library is a standard component knowledge base that includes the characteristics of sheet metal and profiles. The sheet metal characteristic is the sheet thickness, and a series of sheet metal thicknesses are entered into the sheet metal library. The profile characteristic is the cross-sectional attribute that includes cross-sectional dimensions and section modulus.

10. A method for optimizing ship cross-sectional structure based on an improved bee colony algorithm as described in any one of claims 2, 8, or 9, characterized in that, The process of converting the continuous eigenvalues ​​obtained through iterative solutions into usable structural models includes standard structural standardization and sensitivity-based custom structural standardization. The standard structural standardization strategy is to traverse the component library and select the structure in the library whose eigenvalues ​​are closest to the final solution. For plates, the required eigenvalue is the plate thickness, and for profiles, the required eigenvalue is the section modulus. The custom structural standardization strategy is to seek the eigenvalue, i.e., the optimization variable, that has the greatest impact on the section dimensions of the custom structure.

Images