Ship block manufacturing time prediction method under small sample data

Abstract

The invention relates to a ship block manufacturing time prediction method under small sample data, and aims to solve the problem that the accuracy of ship block manufacturing time prediction under the small sample data is difficult to guarantee, the method takes a block with a large amount of historical data as a source field, and takes a block under a small sample data set as a target field; a prediction model of source domain segment manufacturing time is constructed through a machine learning technology, a network weight and an offset value are extracted and migrated to a prediction model of a target domain segment, similarity and difference characteristics of a source domain and the target domain segment are analyzed, and according to an analysis result, a prediction model of source domain segment manufacturing time is established. And freezing the model parameters of the similarity features and finely adjusting the model parameters of the difference features to complete the construction of the target domain segmented manufacturing time prediction model. The time prediction method provided by the invention not only effectively solves the problem of ship block manufacturing time prediction under small sample data, but also shows remarkable advantages in the aspects of precision and stability. The invention belongs to the technical field of time prediction.

Description

Technical Field

[0001] The invention belongs to the technical field of time prediction, and in particular relates to a method for predicting the ship section manufacturing time under small sample data. Background Art

[0002] In the modern shipbuilding model, ship sections are important intermediate products. The accurate prediction of their manufacturing time is crucial to the resource allocation, material procurement and overall production plan of the shipyard, which directly affects the efficiency and cost control of ship construction, and determines the progress and stability of ship delivery. However, since ship manufacturing is usually based on orders, each order requirement is highly personalized, and different ship sections vary significantly, the manufacturing workload is heavy, which makes the prediction of ship section manufacturing time particularly difficult, especially in the absence of sufficient historical data for certain ship types, the accuracy of the prediction is even more difficult to guarantee. Therefore, in the case of small sample data, improving the accuracy of ship section manufacturing time prediction has become one of the key issues in the current shipbuilding industry.

[0003] Prior art document 1 "A knowledge discovery and reuse method for time estimation in ship block manufacturing planning using DEA", Li et al., Advanced Engineering Informatics, 2019, 39: 25-40, proposed a knowledge discovery and reuse method for ship block manufacturing time prediction, but this method did not consider the case of small sample data sets. Technical document 2 "Gear contact fatigue life prediction based on transfer learning", Li et al., International Journal of Fatigue, 2023, 173: 12, proposed a gear contact fatigue life prediction method based on transfer learning and machine learning, but this method did not consider the similarity and difference characteristics between the source domain data set and the target domain data set. Therefore, it is still necessary to further study methods suitable for ship block manufacturing time prediction under small sample data. Summary of the invention

[0004] In order to solve the problem that it is difficult to predict the manufacturing time of ship sections and difficult to ensure accuracy under small sample data, the present invention further proposes a method for predicting the manufacturing time of ship sections under small sample data.

[0005] The technical solution of the present invention is a method for predicting the manufacturing time of ship segments under small sample data. A prediction model for the manufacturing time of source domain segments is constructed through machine learning technology, and its network weights and bias values ​​are extracted. The weights and bias values ​​of the manufacturing time of source domain segments are migrated to the prediction model of target domain segments, and the similarity and difference characteristics of the source domain and target domain segments are analyzed. According to the analysis results, the network parameters of the similarity characteristics are frozen to prevent them from changing in subsequent training, and the network parameters of the difference characteristics are fine-tuned to adapt to the characteristics of the target domain data. On the above basis, the construction of the target domain segment manufacturing time prediction model is completed, so that it has the ability to accurately predict the target domain segment manufacturing time under small sample data. The specific steps of the method are as follows:

[0006] Step 1: Construct source domain segmentation time prediction model based on machine learning;

[0007] First, determine the methodological selection of the model. A neural network model based on particle swarm algorithm (PSOBPNN) is selected for segmented manufacturing time prediction. Each particle represents a set of parameters of the neural network model, namely the weight and bias value of the model. The dimension of the particle is determined by the structure of the model, which depends on the number of nodes in the input layer of the model, the number of nodes in each hidden layer, and the number of nodes in the output layer. The particle dimension calculation formula is:

[0008]

[0009] Among them, Dim is the particle dimension, i represents the hidden layer index value, L represents the maximum hidden layer index value, n in is the number of nodes in the input layer, is the number of nodes in the ith hidden layer, n out is the number of nodes in the output layer.

[0010] Secondly, determine the input and output variables of the model. The segment weight, projection area, number of intermediate groups, weld length and steel plate thickness are selected as the input features of PSOBPNN, and the segment manufacturing time is the output. In order to improve the training efficiency and stability of the model, all input and output data are normalized by decimal scaling. The decimal scaling normalization formula is as follows:

[0011]

[0012] k = log 10 (max(x)) (3)

[0013] Wherein, x' is the normalized data value, x is the data value before normalization, and k is the logarithmic function value with base 10 corresponding to the maximum value of x.

[0014] Thirdly, determine the prediction accuracy index. The present invention uses the mean absolute percentage error (MAPE) as a relevant accuracy index, and the MAPE value can show the average error degree of the prediction.

[0015]

[0016] Among them, j represents the index value of the sample, y j represents the actual operation time of the jth sample, represents the prediction operation time of the jth sample, and n represents the maximum index value of the sample.

[0017] Next, determine the hyperparameter configuration of the neural network model. The hyperparameters of the neural network model include the number of hidden layers, the number of hidden layer nodes, the hidden layer activation function, the output layer activation function, the loss function, the number of training times, and the training target error value. Experiments are conducted under different hyperparameter combinations to evaluate the impact of each set of hyperparameters on the model's prediction performance, and finally the best hyperparameter configuration is selected.

[0018] Finally, determine the hyperparameter configuration of the particle swarm algorithm. The hyperparameters of the particle swarm algorithm include individual and population learning factors, maximum number of iterations, number of particles, minimum inertia weight value, maximum inertia weight value, particle velocity range, and particle position range. By systematically adjusting and testing these hyperparameters and analyzing their impact on the model prediction results, we finally select the hyperparameter configuration with the best performance.

[0019] Based on the above steps, the source domain segmented manufacturing time prediction model is constructed and trained, and its network weights and bias values ​​are extracted. This model will serve as the basis for transfer learning and will be further applied to the segmented manufacturing time prediction in the target domain.

[0020] Step 2: Build a segmented time prediction model for the target domain based on transfer learning and machine learning;

[0021] First, a migration strategy for weights and bias values ​​is established. The weights and bias values ​​of the source domain segmented manufacturing time prediction model are migrated to the target domain segmented time prediction model (TL-PSOBPNN). By analyzing the similarity and difference characteristics between the source domain segmentation and the target domain segmentation, the following strategy is determined: for the weights from the input layer to the hidden layer of the similarity characteristics between the source domain and the target domain, a freezing action is taken; for other weights and bias values, a fine-tuning action is taken.

[0022] Secondly, train the target domain segmentation time prediction model. Use the target domain segmentation data to train the TL-PSOBPNN, optimize the weights and biases to be fine-tuned in the model to ensure that the model adapts to the characteristics of the target domain segmentation, and thus form a TL-PSOBPNN suitable for target domain segmentation time prediction.

[0023] Finally, the effectiveness of the method is verified. By selecting actual production cases and using the mean absolute percentage error (MAPE) as the accuracy indicator, the prediction effects with and without the transfer learning model are compared to verify the effectiveness of the present invention.

[0024] The beneficial effects of the present invention are:

[0025] 1. The present invention constructs a prediction model for the manufacturing time of source domain segments through machine learning technology, mines the associated knowledge between source domain segment features and manufacturing time, and applies this knowledge to target domain segments through transfer learning technology. The time prediction method of the present invention not only effectively solves the problem of predicting the manufacturing time of ship segments under small sample data, but also shows significant advantages in accuracy and stability;

[0026] 2. The present invention proposes a method for predicting the manufacturing time of ship sections under small sample data. The average MAPE of the test results obtained on the test set is 8.62%, which can more accurately predict the manufacturing time of ship sections under small sample data, improve the efficiency and cost control of ship construction, and improve the reliability and stability of ship delivery schedule. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of the method of the present invention;

[0028] Figure 2 This is a comparison chart of the MAPE values ​​of the target domain segmented time prediction results with and without transfer learning;

[0029] A-axis is the predicted operation time of ship section manufacturing, and B-axis is the actual operation time of ship section manufacturing. DETAILED DESCRIPTION

[0030] Combination Figure 1 and Figure 2 The specific implementation of the present invention is described. The present implementation is a method for predicting the manufacturing time of ship sections under small sample data. Figure 1 As shown, the following steps are included:

[0031] Step 1: Construct source domain segmentation time prediction model based on machine learning;

[0032] The specific steps of the implementation process of step one are as follows:

[0033] First, determine the methodological selection of the model. The segmented manufacturing time prediction uses a neural network model based on a particle swarm algorithm (PSOBPNN). Each particle represents a set of parameters of the neural network model, namely the weight and bias value of the model. The dimension of the particle is determined by the structure of the model, which depends on the number of nodes in the input layer of the model, the number of nodes in each hidden layer, and the number of nodes in the output layer. The particle dimension calculation formula is:

[0034]

[0035] Among them, Dim is the particle dimension, i represents the hidden layer index value, L represents the maximum hidden layer index value, n in is the number of nodes in the input layer, is the number of nodes in the ith hidden layer, n out is the number of nodes in the output layer.

[0036] The neural network model structure adopted in the present invention includes 1 hidden layer, 5 input layer nodes, 4 hidden layer nodes, and 1 output layer node, so the particle dimension is 83.

[0037] Secondly, determine the input and output variables of the model. The segment weight, projection area, number of intermediate groups, weld length and steel plate thickness are selected as the input features of PSOBPNN, and the segment manufacturing time is the output. In order to improve the training efficiency and stability of the model, all input and output data are normalized by decimal scaling. The decimal scaling normalization formula is as follows:

[0038]

[0039] k = log 10 (max(x)) (3)

[0040] Wherein, x' is the normalized data value, x is the data value before normalization, and k is the logarithmic function value with base 10 corresponding to the maximum value of x.

[0041] Thirdly, determine the prediction accuracy index. The present invention uses the mean absolute percentage error (MAPE) as a relevant accuracy index, and the MAPE value can show the average error degree of the prediction.

[0042]

[0043] Among them, j represents the index value of the sample, y j represents the actual operation time of the jth sample, represents the prediction operation time of the jth sample, and n represents the maximum index value of the sample.

[0044] Next, determine the hyperparameter configuration of the neural network model, including the number of hidden layers, the number of hidden layer nodes, the hidden layer activation function, the output layer activation function, the loss function, the number of training times, and the training target error value. The hyperparameters of the neural network model used in the present invention are selected as follows:

[0045]

[0046] Finally, the hyperparameter settings of the particle swarm algorithm are determined, including individual and population learning factors, maximum number of iterations, number of particles, minimum inertia weight value, maximum inertia weight value, particle velocity range, and particle position range. The hyperparameters of the particle swarm algorithm used in the present invention are selected as follows:

[0047]

[0048] Based on the above steps, the source domain segmented manufacturing time prediction model is constructed and trained, and its network weights and bias values ​​are extracted. This model will serve as the basis for transfer learning and will be further applied to the segmented manufacturing time prediction in the target domain.

[0049] Step 2: Build a segmented time prediction model for the target domain based on transfer learning and machine learning;

[0050] In step 2, first, a migration strategy for weights and bias values ​​is established. The weights and bias values ​​of the source domain segmented manufacturing time prediction model are migrated to the target domain segmented time prediction model (TL-PSOBPNN). By analyzing the similarity and difference characteristics between the source domain segmentation and the target domain segmentation, the following strategy is determined: for the weights from the input layer to the hidden layer of the similarity characteristics between the source domain and the target domain, a freezing action is taken; for other weights and bias values, a fine-tuning action is taken.

[0051] Secondly, the target domain segmentation time prediction model is trained. The constructed TL-PSOBPNN is trained using the target domain segmentation data, and the weights and biases to be fine-tuned in the model are optimized to ensure that the model adapts to the characteristics of the target domain segmentation, thereby forming a TL-PSOBPNN suitable for target domain segmentation time prediction.

[0052] Finally, the effectiveness of the method is verified. By selecting actual production cases, bulk carrier segments are used as source domain segments, and cruise ship segments are used as target domain segments. The main characteristic difference between the two types of segments lies in the thickness of the steel plate. The mean absolute percentage error (MAPE) is used as the accuracy indicator to compare the prediction effects of the models with and without transfer learning, thereby verifying the effectiveness of the present invention.

[0053] Attached Figure 2The following is a comparison chart of the MAPE values ​​of the segmented time prediction results of the target domain with and without transfer learning. The figure describes the predicted time and actual operation time of 40 test set samples. Each data point in the figure is composed of an interval, which represents the mean, minimum and maximum values ​​of the results of 100 training times for each data point. The closer the data point is to the 45° symmetry line, the better the match between the model's predicted time and the actual operation time. The mean values ​​of the MAPE of the TR model and the non-TR model are 8.62% and 14.97% respectively. At the same time, the mean values ​​of 80% of the test samples under the TR model are controlled within the 1.5-fold dispersion band, while only 70% of the test samples under the non-TR model are controlled within the 1.5-fold dispersion band. In comparison, the error bars of the non-TR model are larger than those of the TR model. Therefore, this shows that the TR model has better prediction accuracy and prediction stability than the non-TR model.

[0054] In summary, the method for predicting the manufacturing time of ship sections under small sample data proposed in the present invention has a test result MAPE average of 8.62% on the test set, which can more accurately predict the manufacturing time of ship sections under small sample data, improve the efficiency and cost control of ship construction, and improve the reliability and stability of ship delivery schedule.

[0055] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.

Claims

1. A method for predicting ship section manufacturing time under small sample data, characterized in that: The prediction method comprises the following steps: Step 1: Construct source domain segmentation time prediction model based on machine learning; Step 2: Construct a segmented time prediction model for the target domain based on transfer learning and machine learning.

2. The method for predicting ship section manufacturing time under small sample data according to claim 1 is characterized in that: Step 1 specifically includes: Step 1.1 Determine the methodological selection of the model; Step 1.2 Determine the input and output variables of the model; Step 1.3: Determine the prediction accuracy index; Step 1.4: Determine the hyperparameter configuration of the neural network model; Step 1.5 determines the hyperparameter configuration of the particle swarm algorithm.

3. The method for predicting ship section manufacturing time under small sample data according to claim 2 is characterized in that: The step 1.1 specifically includes: A neural network model based on particle swarm algorithm is used for segmented manufacturing time prediction. Each particle represents a set of parameters of the neural network model, namely the weight and bias value of the model. The dimension of the particle is determined by the structure of the model, which depends on the number of nodes in the input layer, the number of nodes in each hidden layer, and the number of nodes in the output layer. The particle dimension calculation formula is: Among them, Dim is the particle dimension, i represents the hidden layer index value, L represents the maximum hidden layer index value, nin is the number of nodes in the input layer, is the number of nodes in the ith hidden layer, n out is the number of nodes in the output layer.

4. The method for predicting ship section manufacturing time under small sample data according to claim 2 is characterized in that: The step 1.2 specifically includes: The segment weight, projection area, number of middle groups, weld length and steel plate thickness are selected as the input features of the neural network model, and the segment manufacturing time is the output. All input and output data are normalized by decimal scaling. The decimal scaling normalization formula is as follows: k = log 10 (max(x))(3)wherein x′ is the normalized data value, x is the data value before normalization, and k is the logarithmic function value with base 10 corresponding to the maximum value of x.

5. The method for predicting ship section manufacturing time under small sample data according to claim 2 is characterized in that: The step 1.3 specifically includes: The mean absolute percentage error is used as a relevant accuracy indicator. The mean absolute percentage error value can show the average error degree of the prediction; Among them, j represents the index value of the sample, y j represents the actual operation time of the jth sample, represents the prediction operation time of the jth sample, and n represents the maximum index value of the sample.

6. The method for predicting ship section manufacturing time under small sample data according to claim 2 is characterized in that: The step 1.4 specifically includes: The hyperparameters of the neural network model include the number of hidden layers, the number of hidden layer nodes, the hidden layer activation function, the output layer activation function, the loss function, the number of training times, and the training target error value. By conducting experiments under different hyperparameter combinations, the impact of each set of hyperparameters on the model's prediction performance is evaluated, and finally the hyperparameter configuration with the best performance is selected.

7. The method for predicting ship section manufacturing time under small sample data according to claim 2 is characterized in that: The step 1.5 specifically includes: The hyperparameters of the particle swarm algorithm include individual and population learning factors, maximum number of iterations, number of particles, minimum inertia weight value, maximum inertia weight value, particle velocity range, and particle position range. By systematically adjusting and testing these hyperparameters and analyzing their impact on the model prediction results, the best hyperparameter configuration is finally selected.

8. The method for predicting ship section manufacturing time under small sample data according to claim 1 is characterized in that: The step 2 specifically includes the following steps: Step 2.1 Establish a migration strategy for weights and bias values; Step 2.2: training the target domain segment time prediction model; Step 2.3: Verify the effectiveness of the method.

9. The method for predicting ship section manufacturing time under small sample data according to claim 8 is characterized in that: The step 2.1 specifically includes: The weights and bias values ​​of the source domain segmented manufacturing time prediction model are transferred to the target domain segmented time prediction model. By analyzing the similarity and difference characteristics between the source domain segmentation and the target domain segmentation, the following strategy is determined: for the weights from the input layer to the hidden layer of the similarity characteristics between the source domain and the target domain, freeze the action; for other weights and bias values, take fine-tuning actions.

10. The method for predicting ship section manufacturing time under small sample data according to claim 8 is characterized in that: The step 2.2 specifically includes: Use the target domain segmentation data to train the target domain segmentation time prediction model and optimize the weights and biases to be fine-tuned in the model; The step 2.3 specifically includes: By selecting actual production cases and using the mean absolute percentage error as the accuracy indicator, the prediction effects of the transfer learning model and the one with and without the transfer learning model are compared to verify the effectiveness of the present invention.

Images