A Method for Deformation Prediction of Line Heating Based on ISSA-ELM

Abstract

The present invention relates to a method for predicting deformation of water-fire bending plates based on ISSA-ELM, which comprises the following steps: collecting historical processing data of water-fire bending plates and constructing a water-fire bending plate processing training set; normalizing the acquired training set samples; determining the input layer, output layer, and hidden layer parameters of the ELM network based on the collected data samples, and establishing an ELM model; optimizing the initial weights and thresholds of the original ELM network using an improved Salp Optimization Algorithm (ISSA); training the ELM network using the training set to complete the final construction of the ELM network; inputting actual processing data into the ELM network, and denormalizing the output results; and finally obtaining a prediction result for water-fire bending plate processing deformation. The present invention can be effectively applied to water-fire bending plate processing deformation prediction, and has good prediction accuracy and stability.

Description

Technical Field

[0001] The present invention relates to a method for predicting deformation of a water-fire bending plate, and more particularly to a method for predicting deformation of a water-fire bending plate processing based on ISSA-ELM. Background Art

[0002] The hull plating is a crucial component of a ship's main structure. Its functions include ensuring watertightness, enabling the ship to float and carry cargo, ensuring the hull's strength and rigidity, and withstanding external loads during service. There are two main methods for manufacturing curved ship plating: cold forming and hot forming. In cold forming, loads are typically applied through hydraulic pressure or roller pressing to achieve the desired deformation. In hot forming, heat is applied to the steel plate surface, creating a temperature gradient and causing localized deformation. Currently, hot forming is the primary method for fabricating non-developable surfaces. Water-heat bending is a key technology for fabricating large hulls and complex curved surfaces, offering advantages such as wide application, low cost, flexibility, and high adaptability. However, the process is often performed manually by on-site workers based on their own experience. This manual process results in inconsistent sheet quality and low efficiency. Therefore, it is necessary to study the relationship between water-heat bending parameters and deformation results. Predicting the impact of these parameters on forming results can provide a theoretical basis for technical personnel and is of great significance for on-site processing. Summary of the Invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a method for predicting deformation of water-fire bending plate processing with high prediction accuracy and high prediction efficiency.

[0004] The ISSA-ELM-based method for predicting deformation of water-fire bending plates of the present invention comprises the following steps:

[0005] S1 collects historical data of water and fire bending plate processing to build a training set;

[0006] S2 normalizes the collected data;

[0007] S3 determines the input layer, output layer and hidden layer parameters of the ELM network based on the collected processing data;

[0008] S4 uses the improved Salp algorithm to optimize the initial weights and thresholds of the ELM network;

[0009] S5 uses the processed historical data as the training set of the optimized ELM network to train the ELM network;

[0010] S6 inputs the actual processing parameters into the trained ELM network to obtain the output results, and denormalizes the results;

[0011] S7 obtains deformation prediction results under corresponding processing parameters.

[0012] The historical processing data of the water-fire bending plate described in S1 include heating speed, heat input, plate thickness, number of heating times, heating line length, heating line width, distance between the heating line and the plate width, water-fire distance, flame center distance, lateral shrinkage and angular deformation.

[0013] The data in S2 are normalized using the following formula:

[0014]

[0015] Where, T is the target data; T max is the maximum value of the target data; T min is the minimum value of the target data; X max is the maximum value of the sample data; X min is the minimum value of the sample data; X is the sample data.

[0016] In the ELM network described in S3, heating rate, heat input, plate thickness, number of heating times, heating line length, heating line width, distance between heating line and plate width, water-fire distance and flame center distance are used as input layer parameters, and lateral shrinkage and angular deformation are used as output layer parameters.

[0017] In S4, the improved salp algorithm is called to determine the initial weight w and threshold b of the ELM model, including the steps of: using the initial weight w and threshold b in the ELM network model as a salp in the salp group search space, using the salp algorithm to iteratively search for the best, setting the initial parameters of the salp algorithm and then updating them according to the position update formula of the salp leader and follower in the ISSA algorithm, until the iteration conditions are met, and outputting the optimal solution as the initial weight w and threshold b in the ELM network model;

[0018] The position update formulas of the leader salp and follower salp are as follows:

[0019]

[0020]

[0021]

[0022]

[0023] Where, Indicates the leader position; ub d , lb dThey represent the upper and lower limits of the leader's search in the d dimension, respectively; c1 and c2 are random numbers, ranging from [0,1]. c1 is used to control the search and development capabilities of the entire group, and c2 determines the length of the movement; c3 is the search balance factor, which determines the positive and negative directions of the movement, and is used to balance the global and local search capabilities, thereby enhancing the randomness and diversity of the leader; a represents acceleration; v0 represents the initial velocity; t a represents the iteration step size; R represents the movement distance; They represent the d-th dimension position of the m-th follower after and before the update, respectively.

[0024] In S4, the present invention proposes an improved salp optimization algorithm, including: adding a decay factor A(l) to gradually reduce the leader position update range with the increase of the number of iterations, avoiding falling into local extreme values in the early stage of convergence, and getting closer and closer to the optimal value in the later stage of convergence, thereby achieving higher solution accuracy; adding an adaptive inertia weight ω, so that the weight in the early stage of the algorithm search is larger, which can enhance the global search capability, and the weight in the later stage of the search is smaller, which can enhance the local search capability.

[0025] The attenuation factor A(l) and the adaptive inertia weight ω are given by the following formulas:

[0026]

[0027] ω=2 -(4 / 3)*l / T *cos(πl)

[0028] Where l represents the current number of iterations, and T represents the maximum number of iterations.

[0029] In S5, the network model is built based on ISSA-ELM. The ISSA-ELM-based water-fire bending plate deformation prediction calculation model is trained to meet the requirements, including the following steps:

[0030] S51 determines that the input layer parameters of the ELM network model are heating speed, heat input, plate thickness, heating times, heating line length, heating line width, distance between heating line and plate width, water-fire distance and flame center distance;

[0031] S52 Initialize the ELM network model: determine the number of hidden layer neurons, initialize the weight w and threshold b;

[0032] S53 processes the ship plate according to various parameters, records the experimental values, and forms a training sample with the heating rate, heat input, plate thickness, number of heating times, heating line length, heating line width, distance between the heating line and the plate width, water-fire distance, and flame center distance as inputs, and transverse shrinkage and angular deformation as outputs;

[0033] S54 calculates the output layer weights using the generalized inverse matrix theory, and uses the sum of the squares of the differences between the ideal output and the actual output as the error function to calculate the error.

[0034] In S6, the denormalization formula is as follows:

[0035]

[0036] In S7, the deformation results are the lateral shrinkage Δl and the angular deformation α, and the measurement formula is as follows:

[0037]

[0038]

[0039] Among them, Δl1 represents the linear deformation of the upper surface in the direction perpendicular to the heating line; Δl2 represents the linear deformation of the lower surface in the direction perpendicular to the heating line; and t represents the thickness of the plate.

[0040] A device for predicting deformation of water-fire bending plates based on ISSA-ELM comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the method for predicting deformation of water-fire bending plates based on ISSA-ELM is implemented.

[0041] Beneficial effects: The deformation effect of water and fire bending can be predicted accurately and quickly, which can provide certain guiding significance for processing and production; the neural network model implemented by the extreme learning machine algorithm avoids the repeated iterative process of the traditional training method and reduces the training time; the ISSA algorithm is used to optimally solve the input layer weights and hidden layer thresholds of the ELM model to improve the stability of the prediction results; a method of adding attenuation factors and inertia weights to the traditional SSA algorithm is proposed to enhance the algorithm's optimization ability and avoid falling into local extreme values, resulting in higher solution accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the process of the ISSA-ELM-based water-fire bending plate deformation prediction method of the present invention;

[0043] Figure 2 This is a flow chart of the ISSA-ELM algorithm of the present invention;

[0044] Figure 3 The fitness curves of SOA-ELM, PSO-ELM, SSA-ELM and ISSA-ELM according to the embodiment of the present invention are as follows;

[0045] Figure 4This is a comparison chart of the deformation prediction results of water-fire bending plates using SOA-ELM, PSO-ELM, SSA-ELM and ISSA-ELM according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0047] like Figure 1 As shown, this embodiment provides a method for predicting deformation of water-fire bending plates based on ISSA-ELM, which specifically includes the following steps:

[0048] S1 obtains sample data through water and fire bending plate deformation experiments and constructs a water and fire bending plate processing data training set;

[0049] In the embodiment of the present invention, a total of 63 sets of data are used as training data, and the field data include: heating speed, heat input, plate thickness, heating times, heating line length, heating line width, distance between heating line and plate width, water-fire distance, flame center distance, lateral shrinkage and angular deformation;

[0050] S2 training set data normalization processing;

[0051] In the embodiment of the present invention, the normalization process is performed using the following formula:

[0052]

[0053] Where, T is the target data; T max is the maximum value of the target data; T min is the minimum value of the target data; X max is the maximum value of the sample data; X min is the minimum value of the sample data; X is the sample data;

[0054] S3 determines the network parameters of the ELM based on the collected processing data;

[0055] In the embodiment of the present invention, the nine collected data are used as inputs of the ELM network, and the lateral shrinkage and angular deformation are used as outputs;

[0056] S4 optimizes the initial weight w and threshold b of the ELM network using the improved salp optimization algorithm proposed in the present invention;

[0057] In an embodiment of the present invention, the improved salp optimization algorithm comprises the following specific steps:

[0058] The attenuation factor A(l) is introduced into SSA, so that the leader position update range gradually decreases with the increase of the number of iterations, avoiding falling into local extreme values in the early convergence stage and getting closer and closer to the optimal value in the later convergence stage, making the algorithm more accurate.

[0059] Introducing the adaptive inertia weight ω into SSA to enhance the global search capability and local optimization capability;

[0060] The attenuation factor A(l) and the adaptive inertia weight ω are respectively expressed as follows:

[0061]

[0062] ω=2 -(4 / 3)*l / T *cos(πl)

[0063] Where l represents the current number of iterations, and T represents the maximum number of iterations;

[0064] The method adopts the ISSA algorithm to determine the connection weight w from the input layer to the hidden layer and the threshold b of the hidden layer in the ELM network model, including the following steps: taking the connection weight w from the input layer to the hidden layer and the threshold b of the hidden layer in the ELM network model as an individual in the search space of the salp population, adopting the ISSA algorithm to iteratively optimize the individual, setting the initial parameters of the ISSA algorithm and then updating the positions of the leader salp and the follower salp in the ISSA algorithm according to the position update formula of the leader salp and the follower salp until the iteration meets the conditions, and outputting the optimal solution as the connection weight w from the input layer to the hidden layer and the threshold b of the hidden layer in the ELM network model;

[0065] ISSA is used to optimize the initial weight w and threshold b of the ELM network, and the objective function is the fitness function E, which is:

[0066]

[0067] Where, y d * (n) represents the network training output, y d (n) represents the expected output, D is the length of the training data;

[0068] S5 uses the training set data to train the ELM network;

[0069] S6 inputs the actual processed data into the trained ELM network to obtain output data and denormalizes the result;

[0070] In the embodiment of the present invention, a total of 27 groups of data are used as test data, and the output results are denormalized. The formula is as follows:

[0071]

[0072] S7 outputs the prediction result.

[0073] In order to analyze the prediction results of the model more intuitively, in addition to establishing the ISSA-ELM water-fire bending plate deformation prediction model, the Seagull algorithm optimized ELM model (SOA-ELM), the particle swarm algorithm optimized ELM model (PSO-ELM) and the original Salp Ascidian algorithm optimized ELM model (SSA-ELM) were also established. The fitness curves of the four algorithm prediction models are shown in Figure 2. Figure 3 As shown in the figure; Based on the training results of the four models, the water-fire bending plate deformation test samples are input into the five prediction models (including the original ELM model), and the prediction results of the five models are as follows: Figure 4 shown.

[0074] This embodiment uses three model accuracy evaluation indicators, namely fitness value E, mean square error MSE and determination coefficient R 2 , and its performance indicators are defined as follows:

[0075]

[0076]

[0077]

[0078] Where: y d * (n) represents the network training output, y d (n) represents the expected output, and D is the length of the training data. The fitness value and determination coefficient calculation results are shown below:

[0079] Table 1 Performance index calculation results

[0080] Model E <![CDATA[R 2 ]]> SOA-ELM 0.0058 0.9988 PSO-ELM 0.0127 0.9881 SSA-ELM 0.0022 0.9991 ISSA-ELM 0.0017 0.9997

[0081] The performance indicator calculation results in the table above show that the ISSA-ELM model proposed in this embodiment performs better than the other three prediction models. Analysis of the three performance indicator calculation results for the four models shows that the ISSA-ELM model proposed in this embodiment effectively improves the model's anti-interference ability, eliminates prediction randomness, reduces prediction value deviation, and significantly improves prediction reliability. In summary, the ISSA-ELM water-fire bending plate deformation prediction model in this embodiment has higher prediction accuracy and stability, and its generalization performance has been significantly improved, making it more suitable for water-fire bending plate deformation prediction.

[0082] Figure 3 is the mean square error calculation result of each model. Figure 3 It can be seen that these algorithm models tend to be stable after fifty iterations, which shows the feasibility of using this algorithm for nonlinear modeling. Figure 3 and Figure 4 It can be seen that the ISSA-ELM proposed in the present invention has faster optimization speed and higher convergence accuracy, and the ISSA algorithm significantly improves the learning efficiency of the ELM model.

[0083] Obviously, the specific implementation of the present invention is not limited to the above-mentioned methods. As long as various non-substantial improvements are made using the method concept and technical solution of the present invention, they are all within the scope of protection of the present invention.

Claims

1. A method for predicting deformation of water-fire bending plate based on ISSA-ELM, characterized in that: The following steps are involved: S1: Collect water and fire bending plate processing data and build a training set; S2: normalize the collected water and fire bending plate processing data; S3: Determine the input layer, output layer and hidden layer parameters of the ELM network based on the collected water and fire bending processing data; S4: Optimize the initial weights and thresholds of the ELM network using the improved Salp algorithm, including the following steps: The initial weight w and threshold b in the ELM network model are used as a salp in the salp group search space, and the salp algorithm is used for iterative optimization. After setting the initial parameters of the salp algorithm, the position update formula of the salp leader and follower in the ISSA algorithm is used to update the salp until the iteration meets the conditions and the optimal solution is output as the initial weight w and threshold b in the ELM network model. The position update formulas of the leader salp and follower salp are as follows: Where, F d represents the position of the leader salp in the d-dimensional space; Indicates the leader position; ub d , lb d They represent the upper and lower limits of the leader's search in the d dimension, respectively; c1 and c2 are random numbers, ranging from [0,1]. c1 is used to control the search and development capabilities of the entire group, and c2 determines the length of the movement; c3 is the search balance factor, which determines the positive and negative directions of the movement, and is used to balance the global and local search capabilities, thereby enhancing the randomness and diversity of the leader; a represents acceleration; v0 represents the initial velocity; t a represents the iteration step; R represents the movement distance; denote the d-th dimension position of the m-th follower after and before the update, respectively; The improved salp algorithm described in step S4 includes the following steps: adding a decay factor A(l) so that the leader position update range gradually decreases with the increase of the number of iterations, avoiding falling into local extreme values in the early stage of convergence, and getting closer and closer to the optimal value in the later stage of convergence, thereby achieving higher solution accuracy; adding an adaptive inertia weight ω so that the weight in the early stage of the algorithm search is greater than the weight in the later stage of the search; The attenuation factor A(l) and the adaptive inertia weight ω are given by the following formulas: ω=2 -(4 / 3)*l / T *cos(πl) Where l represents the current number of iterations, and T represents the maximum number of iterations; Step S4 uses the improved salp algorithm to optimize the initial weight w and threshold b of the ELM network, including calling the ISSA optimization ELM model, and for ELM models with different initial weights w and thresholds b, the initial weights w and thresholds b in each ELM model are used as independent salp individuals in the ISSA algorithm population, and the fitness value of each individual is calculated using the following formula. The minimum fitness value is iteratively optimized as the optimal initial weight w and threshold b: In the formula, E represents the individual fitness value, y d * (n) represents the network training output, y d (n) represents the expected output, D is the length of the training data; S5: Use the processed data as the training set of the optimized ELM network to train the ELM network; S6: Input the actual processing parameters into the trained ELM network to obtain the output results, and perform denormalization on the results; S7: Obtain deformation prediction results under corresponding processing parameters.

2. The ISSA-ELM-based water-fire bending plate deformation prediction method according to claim 1 is characterized in that: The water-fire bending plate processing data described in step S1 include heating speed, heat input, plate thickness, heating times, heating line length, heating line width, distance between heating line and plate width, water-fire distance, flame center distance, lateral shrinkage and angular deformation.

3. The ISSA-ELM-based water-fire bending plate deformation prediction method according to claim 1 is characterized in that: According to the collected water-fire bending plate processing data described in step S3, the input layer, output layer and hidden layer parameters of the ELM network are determined. Specifically, the collected heating speed, heat input, plate thickness, number of heating times, heating line length, heating line width, heating line distance from the plate width, water-fire distance and flame center distance are used as the ELM network input layer parameters, and the lateral shrinkage and angular deformation are used as the ELM network output layer parameters, and the number of hidden layers is set accordingly.

4. The ISSA-ELM-based water-fire bending plate deformation prediction method according to claim 1 is characterized in that: Step S5 is specifically as follows: S51 determines that the input layer parameters of the ELM network model are heating speed, heat input, plate thickness, heating times, heating line length, heating line width, distance between heating line and plate width, water-fire distance and flame center distance, and the output layer parameters are lateral shrinkage and angular deformation; S52 Initialize the ELM network model: determine the number of hidden layer neurons, initial weights w and threshold b; S53 processes the ship plate according to the parameters in S51, records the experimental values, and forms a training sample with the heating rate, heat input, plate thickness, heating times, heating line length, heating line width, distance between heating line and plate width, water-fire distance, and flame center distance as inputs, and transverse shrinkage and angular deformation as outputs; S54 calculates the output layer weights using the generalized inverse matrix theory, and uses the sum of the squares of the differences between the ideal output and the actual output as the error function to calculate the error.

5. The ISSA-ELM-based water-fire bending plate deformation prediction method according to claim 1 is characterized in that: The mathematical formulas for the normalization and denormalization processes in steps S2 and S6 are as follows: Normalization processing: Denormalization: Where, T is the target data; T max is the maximum value of the target data; T min is the minimum value of the target data; X max is the maximum value of the sample data; X min is the minimum value of the sample data; X is the sample data.

6. The ISSA-ELM-based water-fire bending plate deformation prediction method according to claim 1 is characterized in that: The deformation prediction results in S7 are the lateral shrinkage Δl and the angular deformation α, and the measurement formula is as follows: Among them, Δl1 represents the linear deformation of the upper surface in the direction perpendicular to the heating line; Δl2 represents the linear deformation of the lower surface in the direction perpendicular to the heating line; and t represents the thickness of the plate.

7. A device for predicting deformation of water-fire bending plate based on ISSA-ELM, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, the ISSA-ELM-based water-fire bending plate deformation prediction method according to any one of claims 1 to 6 is implemented.

Images