Ship cabin noise rapid forecasting method and system based on convolutional neural network

Abstract

The invention discloses a ship cabin noise rapid forecasting method and system based on a convolutional neural network, belongs to the field of ship cabin noise forecasting, and solves the precision problem existing in the prior art. The method specifically comprises the steps of collecting data related to ship cabin noise, and processing the data into a series of sample data sets; and inputting the sample set into a convolutional neural network model for training, wherein the model comprises a feature extraction module and the convolutional neural network model. Constructing an optimal training model by adopting an Adam optimization algorithm; verifying the weight and the threshold value of the neural network through a verification set, calculating the gradient through a loss function, and carrying out the back propagation and iterative updating of parameters until the network converges; and a test set is used for evaluation, and the trained convolutional neural network model is applied to actual rapid prediction of ship cabin noise. The ship cabin noise prediction method has strong feature extraction capability, can adapt to different ship cabin noise prediction requirements, can accurately predict the cabin noise level of a complex ship, and realizes higher noise prediction efficiency.

Description

Technical Field

[0001] The invention belongs to the field of ship cabin noise prediction, and in particular relates to a ship cabin noise rapid prediction method based on convolutional neural network. Background Art

[0002] With the continuous improvement of modern living standards, people's requirements for ship comfort performance are getting higher and higher. For ship cabin noise prediction, there are mainly empirical formula methods, finite element method and boundary element method, statistical energy analysis method, neural network method, etc.

[0003] Among them, the empirical formula method can quickly predict the noise level of ships of the same type or ships with similar overall layout based on the existing empirical formula for ship noise prediction. It is mainly reliable for a certain type of ship, but lacks prediction accuracy for ships with complex structures or uncommon types. At present, it is common to use statistical energy analysis methods to predict noise. The whole ship acoustic prediction model is established, noise excitation is input, SEA parameters are determined, and noise evaluation is performed with the help of VAOne acoustic software. This method can better analyze the distribution of the noise field and optimize the design at the same time, but it often requires rich engineering experience as support for subsystem division, parameter selection and correction. The boundary element method is based on the finite element method and is derived from the boundary integral equation through the Helmholtz equation. The boundary element method properly handles the shortcomings of the finite element method, and has significantly reduced the workload required for calculating ship cabin noise, reducing costs, but for large and complex structures or models with high calculation frequencies, the calculation often exceeds the capabilities of the computer.

[0004] Artificial neural network is a branch of artificial intelligence. It uses mathematical methods to abstract the human brain neural network and establish a simplified model, aiming to imitate the structure and function of the human brain. Due to the large number of mechanical vibration equipment on board and the complex structure, the self-learning, self-adaptive and nonlinear characteristics of artificial neural networks can be used to model nonlinear systems and quickly predict ship cabin noise. Through literature search of the prior art, it is found that in recent years, there are few patent documents similar to the method for rapid prediction of ship cabin noise based on convolutional neural network. The public materials similar to the present invention application include: 1. Fan Bin. Noise prediction of marine platform cabin based on intelligent algorithm [D]. Harbin Engineering University, 2018; 2. Guo Jun, Fan Bin, Wang Xingyu, etc. Cabin noise prediction method based on RBF neural network [C] / / China Vibration Engineering Society, State Key Laboratory of Mechanical Structure Mechanics and Control, Nanjing University of Aeronautics and Astronautics. Proceedings of the 12th National Conference on Vibration Theory and Application. School of Naval Architecture, Harbin Engineering University, 2017: 8; 3. Zeng Xiangyang. Cabin noise prediction based on neural network method [J]. Journal of Northwestern Polytechnical University, 2004, (04): 492-495.

[0005] Reference 1 compiled a cabin frequency band noise prediction program based on radial basis function network (RBF network), particle swarm algorithm and differential evolution method, collected data such as key excitation sources of marine platforms, constructed a noise simulation database of three-dimensional statistical energy model, and used five intelligent algorithms to predict the cabin frequency band sound pressure level based on the measured noise database and the simulation and measured joint database. It has poor applicability to different types of marine platform noise and has a large prediction error. Reference 2 uses radial basis function network (RBF network) as the framework, takes the power of the excitation source, the number of decks between the excitation source and the target cabin, the number of transverse bulkheads between the excitation source and the target cabin, the surface area of ​​the target cabin and the average thickness of the plate frame as input parameters, and takes the A-weighted noise of the target cabin as the output parameter. The algorithm model for cabin noise prediction was established by using differential evolution optimization algorithm and Mahalanobis distance theory. After training, the noise of unknown cabins under given working conditions was predicted and analyzed. The average prediction error can be controlled within 2dB(A). However, the RBF neural network is highly dependent on the selection of radial basis functions and has high requirements for training samples. It needs to fully cover the input space, otherwise the generalization will decrease. Reference 3 aims to address the shortcomings of the existing closed sound field simulation methods, such as limited applicable frequency band range and too many required parameters. Combined with the characteristics of the cabin noise field, a back propagation neural network algorithm (BP network) that can simulate the internal noise of the closed space is given, and simulation experiments are carried out. However, the structure is simple and the results are somewhat subjective, which makes it difficult to realize the cabin noise prediction of large and complex ship structures.

[0006] In summary, there are significant differences between the currently available publicly searchable literature and the present invention, which are mainly reflected in the methods and prediction accuracy. Therefore, the present invention establishes a rapid prediction method for ship cabin noise based on convolutional neural network, which can realize the rapid prediction of ship cabin noise. Summary of the invention

[0007] The purpose of the present invention is to optimize the technical problems pointed out by the background technology, and to provide a method for rapid prediction of ship cabin noise based on convolutional neural network, which performs feature processing and full connection processing through a convolutional neural network model and uses a Sigmond activation function to output the prediction result of ship cabin noise.

[0008] To achieve the above object, the present invention provides the following solutions:

[0009] The present invention provides a ship cabin noise prediction model based on convolutional neural network, the model comprises a feature extraction module and a convolutional neural network model;

[0010] The feature extraction module is used to extract features and input the extracted features into the convolutional neural network model;

[0011] The convolutional neural network model includes N convolutional modules connected in series and a fully connected layer;

[0012] The convolution module includes a convolution layer, a normalization layer and a pooling layer, which are used to extract features to reduce data size;

[0013] The fully connected layer is connected to the last convolution module, the features output by the last convolution module are fully connected, and the Sigmoid activation function is used to output the rapid prediction result of the ship cabin noise.

[0014] Furthermore, there is a preferred embodiment in which the structure of each of the above-mentioned convolution modules is consistent, and they all include a convolution layer with a convolution kernel of 3×3, a normalization layer, and a pooling layer with a size of 2×2 and a step size of 2×2, which are connected in sequence.

[0015] The convolution layer processing method of the nth convolution module in the convolutional neural network model is:

[0016] F n =S(F n-1 )

[0017] In the formula, F n represents the output of the convolutional layer in the nth convolutional module, S(F n-1 ) represents the F input to the convolutional layer n-1 Find the convolution;

[0018] The normalization layer processing method of the nth convolution module in the convolutional neural network model is:

[0019]

[0020] In the formula, C n represents the features obtained by the network through convolution operation in the nth convolution module, E(C n ) indicates that C n Find the expectation, D(C n ) indicates that C n Find the variance, G n Represents the output F of the nth convolutional module n The output after normalization layer processing, ε represents the offset;

[0021] The pooling layer processing method of the nth convolution module in the convolutional neural network model is:

[0022]

[0023] In the formula, R n represents the output of the pooling layer, G n is the output after normalization, q n represents the size of the convolution kernel, k n Indicates the step length.

[0024] The present invention also provides a ship cabin noise prediction method based on the above noise prediction model, the noise prediction method comprising the following steps:

[0025] Step S1: Determine whether the forecast object and forecast method are applicable;

[0026] Step S2: Collecting data related to ship cabin noise, including cabin structure, outfitting materials, ship structure, equipment layout and equipment data;

[0027] Step S3: preprocessing the collected data, including data cleaning, normalization, feature selection, and dividing the collected data into a training set, a validation set, and a test set;

[0028] Step S4: input the training sample set into the convolutional neural network model for training, and use the Adam optimization algorithm to build the best training model;

[0029] Step S5: Verify the weights and thresholds of the neural network in training through the validation set and calculate the gradient through the loss function and back-propagate and iteratively update the parameters until the network converges;

[0030] Step S6: Use the test set to evaluate the trained convolutional neural network model, and apply the trained convolutional neural network model to the actual rapid prediction of ship cabin noise.

[0031] Furthermore, in another preferred embodiment, the above step S2 specifically includes:

[0032] Step S21: The cabin structure data includes the shape and size of the cabin, the outfitting material data includes the material surface density and the vibration isolation coefficient, and the equipment data includes the equipment layout position and the noise data under the working state;

[0033] Step S22: Simulate the ship cabin noise and pre-process or collect the measured noise levels of various types of ships to obtain a series of data sets.

[0034] Furthermore, in another preferred embodiment, the above step S3 specifically includes:

[0035] Step S31: using a cross-validation method to divide the data set into a training set, a test set and a validation set, wherein the training set accounts for 70% of the total data set, the test set accounts for 20%, and the validation set accounts for 10%;

[0036] Step S32: During data cleaning, check whether there are missing values ​​or outliers in the data, and clean the data by filling missing values, deleting outliers, or performing data interpolation;

[0037] Step S33: during feature selection, correlation analysis is used to evaluate the degree of association between each feature and the cabin noise result;

[0038] Step S34: The data size of the unified input convolutional neural network model is a 100×100×16 image tensor. After the convolutional layer feature map processing and pooling operation of five convolutional modules, the feature map is reduced to a tensor feature map of size 8×8×25. The size of the fully connected layer is 1600×1, and the feature map is converted into a noise level result represented by a vector.

[0039] Furthermore, in another preferred embodiment, the above step S4 specifically includes:

[0040] Step S41: According to the convolutional neural network model determined in the previous steps, the Adam optimization algorithm is used as the loss function. The expression of the Adam optimization algorithm is:

[0041]

[0042] In the formula, θ t represents the model parameters at the tth iteration, η represents the learning rate, and denote the first-order moment estimate and the second-order moment estimate of the gradient, respectively. Where β1 is the exponential decay rate of the first-order moment estimate, which is set to 0.9, β2 is the exponential decay rate of the second-order moment estimate, which is set to 0.9999, and ε is the offset, which is very small and is used to prevent division by zero in the implementation;

[0043] Step S42: By solving the above equations, the updated optimized weights and biases can be obtained, thereby obtaining the optimal convolutional neural network model for predicting ship cabin noise.

[0044] Furthermore, in another preferred embodiment, the above step S5 specifically includes:

[0045] Step S51: Adopting the Adam optimization algorithm to update the learning rate and model parameters of the convolutional neural network in real time;

[0046] Step S52: setting a forecast result threshold corresponding to the ship cabin noise, and outputting an alarm signal if the forecast result is greater than or equal to the threshold.

[0047] The method for rapid prediction of ship cabin noise based on convolutional neural network described in the present invention can be fully implemented by computer software. Therefore, correspondingly, the present invention also provides a system for rapid prediction of ship cabin noise based on convolutional neural network, and the system includes a storage device, and the storage device includes:

[0048] Used to determine whether the forecast object and forecast method are applicable;

[0049] Used to collect data related to ship cabin noise, including cabin structure, outfitting materials, ship structure, equipment layout and equipment data;

[0050] Used to preprocess the collected data, including data cleaning, normalization, feature selection, and divide the collected data into training sets, validation sets, and test sets;

[0051] It is used to input the training sample set into the pre-set convolutional neural network for training, and use the Adam optimization algorithm to build the best training model;

[0052] It is used to verify the weights and thresholds of the neural network in training through the validation set, calculate the gradient through the loss function, and back-propagate and iteratively update the parameters until the network converges;

[0053] It is used to evaluate the trained convolutional neural network model using the test set, and apply the trained convolutional neural network model to the rapid prediction of actual ship cabin noise.

[0054] The present invention also provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, any one of the above-mentioned methods for rapid prediction of ship cabin noise based on a convolutional neural network is implemented.

[0055] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for rapid prediction of ship cabin noise based on a convolutional neural network described in any one of the above is implemented.

[0056] The technical solution of the present invention has the following beneficial effects:

[0057] 1. Compared with the existing ship cabin noise prediction method, the present invention provides a ship cabin noise rapid prediction method based on convolutional neural network, which can accurately predict the cabin noise level of complex ships. Since the convolutional neural network has good pattern recognition ability, it can automatically learn the complex characteristics and patterns of ship cabin noise, and has strong processing and expression capabilities when processing large-scale data, thereby providing more accurate prediction results, achieving higher prediction accuracy and efficiency, and helping to accurately evaluate the ship cabin noise level.

[0058] 2. The present invention provides a method for rapid prediction of ship cabin noise based on convolutional neural network, which has a strong feature extraction capability. The convolutional neural network can automatically extract important features from input data and can capture more important information from complex data when predicting ship cabin noise. Compared with other methods, it has the ability to adapt to various noise prediction problems.

[0059] 3. The present invention provides a method for rapid prediction of ship cabin noise based on convolutional neural network, which can adapt to different ship cabin noise prediction requirements by adjusting the number of convolutional neural network layers or the network structure. It has good scalability and wide applicability, making the prediction method based on convolutional neural network more flexible and customizable.

[0060] The invention is suitable for rapid noise prediction in ship cabins. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of a method for rapid prediction of ship cabin noise based on convolutional neural network according to the present invention;

[0062] Figure 2 It is a flow chart of the convolutional neural network model established by the present invention;

[0063] Figure 3 This is a noise prediction result of a cabin of a ship according to an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

[0065] Implementation method 1, see Figure 2 This embodiment describes a ship cabin noise prediction model based on a convolutional neural network, wherein the noise prediction model includes a feature extraction module and a convolutional neural network model;

[0066] The feature extraction module is used to extract features and input the extracted features into the convolutional neural network model;

[0067] The convolutional neural network model includes N convolutional modules connected in series and a fully connected layer;

[0068] The convolution module includes a convolution layer, a normalization layer and a pooling layer, which are used to extract features to reduce data size;

[0069] The fully connected layer is connected to the last convolution module, the features output by the last convolution module are fully connected, and the Sigmoid activation function is used to output the rapid prediction result of the ship cabin noise.

[0070] In practical application, this embodiment constructs a rapid prediction model of ship cabin noise based on convolutional neural network and performs model training, and performs feature extraction and convolutional neural network processing on the rapid prediction model of ship cabin noise based on convolutional neural network, including a feature extraction module and a convolutional neural network model; wherein, Figure 2 As shown, the convolutional neural network model includes N convolutional modules connected in series and a fully connected layer. The convolutional module includes a convolutional layer, a normalization processing layer and a pooling layer, which are used to extract features to reduce the data size. The fully connected layer is connected to the last convolutional module. The fully connected layer fully connects the features output by the last convolutional module and uses the Sigmoid activation function to output the rapid prediction result of the ship cabin noise.

[0071] Implementation method 2: This implementation method specifically describes the convolutional neural network model in the ship cabin noise prediction model based on convolutional neural network described in implementation method 1;

[0072] The convolutional neural network model consists of five convolutional modules connected in series. The structure of each convolutional module is the same, including a convolutional layer with a convolution kernel of 3×3, a normalization layer, and a pooling layer with a size of 2×2 and a stride of 2×2. The convolutional layer processing method of the nth convolutional module in the convolutional neural network model is:

[0073] F n =S(F n-1 )

[0074] Among them, F n represents the output of the convolutional layer in the nth convolutional module, S(F n-1 ) represents the F input to the convolutional layer n-1 Find the convolution;

[0075] The normalization layer processing method of the nth convolution module in the convolutional neural network model is:

[0076]

[0077] Among them, C n represents the features obtained by the network through convolution operation in the nth convolution module, E(C n ) indicates that C n Find the expectation, D(C n ) indicates that C n Find the variance, G n Represents the output F of the nth convolutional module n The output after normalization layer processing, ε represents the offset;

[0078] The pooling layer processing method of the nth convolution module in the convolutional neural network model is:

[0079]

[0080] Among them, R n represents the output of the pooling layer, G n is the output after normalization, q n represents the size of the convolution kernel, k n Indicates the step length.

[0081] Implementation method three, see Figure 1 This embodiment describes a method for rapid prediction of ship cabin noise based on a convolutional neural network. The method is implemented based on a rapid prediction model of ship cabin noise based on a convolutional neural network described in Embodiment 1. The method for rapid prediction of ship cabin noise specifically includes the following steps:

[0082] Step S1: Determine whether the forecast object and forecast method are applicable;

[0083] Step S2: Collecting data related to ship cabin noise, including cabin structure, outfitting materials, ship structure, equipment layout and equipment data;

[0084] Step S3: preprocessing the collected data, including data cleaning, normalization, feature selection, and dividing the collected data into a training set, a validation set, and a test set;

[0085] Step S4: input the training sample set into the convolutional neural network model for training, and use the Adam optimization algorithm to build the best training model;

[0086] Step S5: Verify the weights and thresholds of the neural network in training through the validation set and calculate the gradient through the loss function and back-propagate and iteratively update the parameters until the network converges;

[0087] Step S6: Use the test set to evaluate the trained convolutional neural network model, and apply the trained convolutional neural network model to the actual rapid prediction of ship cabin noise.

[0088] In practical application, if Figure 1 As shown, the ship cabin noise rapid prediction method proposed in this embodiment is applicable to marine navigation equipment such as medium and large ships, offshore platforms, and underwater submersibles that have high requirements for cabin noise levels and have the need for rapid cabin noise prediction and evaluation. The ship cabin noise is predicted by constructing a convolutional neural network, wherein, since the convolutional neural network has good pattern recognition capabilities, it can automatically learn the complex characteristics and patterns of ship cabin noise, and has strong processing and expression capabilities when processing large-scale data, thereby providing more accurate prediction results, achieving higher prediction accuracy and efficiency, and helping to accurately evaluate the ship cabin noise level.

[0089] Furthermore, convolutional neural networks can automatically extract important features from input data and capture more important information from complex data when predicting ship cabin noise. Compared with other methods, they are more adaptable to various noise prediction problems.

[0090] Furthermore, the ship cabin noise prediction method proposed in this embodiment adapts to different ship cabin noise prediction needs by increasing the number of convolutional neural network layers or adjusting the network structure, and has good scalability and wide applicability, making the prediction method based on convolutional neural network more flexible and customizable.

[0091] Implementation mode 4: This implementation mode is to illustrate the specific steps of the method for rapid prediction of ship cabin noise based on convolutional neural network described in Implementation mode 2 by way of example;

[0092] Step S1: Determine whether the forecast object and forecast method are applicable;

[0093] Preferably, step S1 specifically includes:

[0094] The forecasting method described in this embodiment is applicable to marine navigation equipment including, but not limited to, medium and large ships, offshore platforms, underwater submersibles, etc., which have high requirements on cabin noise levels and need for rapid forecasting and evaluation of cabin noise.

[0095] Step S2: Collecting data related to ship cabin noise, including cabin structure, outfitting materials, ship structure, equipment layout and equipment data;

[0096] Preferably, step S2 specifically includes:

[0097] Step S21: The cabin structure data includes the shape and size of the cabin, the outfitting material data includes the material surface density, vibration isolation coefficient, etc., and the equipment data includes the equipment layout location, noise data under working state, etc.;

[0098] Step S22: Based on the data in step S21, simulate and pre-process the ship cabin noise, or collect the measured noise levels of various ships to obtain a series of data sets.

[0099] Step S3: preprocessing the collected data, including data cleaning, normalization, feature selection, and dividing the collected data into a training set, a validation set, and a test set;

[0100] Preferably, step S3 specifically includes:

[0101] Step S31: Considering the limited sample of ship cabin noise data, in order to maximize the use of data, the cross-validation method is used to split and combine the data into more training sets, test sets and validation sets, where the training set accounts for about 70% of the total data set, the test set accounts for 20%, the validation set accounts for 10%, and the total data set is no less than 3500;

[0102] Step S32: When performing data cleaning, check whether there are missing values, outliers, etc. in the data, and clean the data by filling missing values, deleting outliers, or performing data interpolation;

[0103] Step S33: When performing feature selection, correlation analysis is used to evaluate the degree of association between each feature and the cabin noise result.

[0104] Step S34: In view of the complexity of the ship structure, there are many characteristic factors that affect the noise level of the ship cabin. The data size of the unified input convolutional neural network model is 100×100×16 image tensor. After the convolutional layer feature map processing and pooling operation of five convolution modules, the feature map is reduced to a tensor feature map of size 8×8×25. The size of the fully connected layer is 1600×1, and the feature map is converted into a noise level result represented by a vector.

[0105] Step S4: input the training sample set into the convolutional neural network model for training, and use the Adam optimization algorithm to build the best training model;

[0106] Preferably, step S4 specifically includes:

[0107] Step S41: According to the convolutional neural network model determined in the previous steps, the Adam optimization algorithm is used as the loss function. The expression of the Adam optimization algorithm is:

[0108]

[0109] Among them, θ t represents the model parameters at the tth iteration, η represents the learning rate, and denote the first-order moment estimate and the second-order moment estimate of the gradient, respectively. Where β1 is the exponential decay rate of the first-order moment estimate, which is set to 0.9, β2 is the exponential decay rate of the second-order moment estimate, which is set to 0.9999, and ε is the offset, which is very small and is used to prevent division by zero in the implementation.

[0110] Step S42: By solving the above equations, the updated optimized weights and biases can be obtained, thereby obtaining the optimal convolutional neural network model for predicting ship cabin noise.

[0111] Step S5: Verify the weights and thresholds of the neural network in training through the validation set and calculate the gradient through the loss function and back-propagate and iteratively update the parameters until the network converges;

[0112] Preferably, step S5 specifically includes:

[0113] Step S51: The learning rate is an important hyperparameter in the gradient descent algorithm. A too large learning rate may lead to an oscillating or unstable training process, while a too small learning rate may lead to slow convergence. The Adam optimization algorithm can be used to update the learning rate and model parameters in real time to obtain the optimal convolutional neural network model for predicting ship cabin noise;

[0114] Step S52: setting a threshold value of the forecast result corresponding to the ship cabin noise, and if the forecast result corresponding to the ship cabin noise is greater than or equal to the threshold value, outputting an alarm signal accordingly.

[0115] Step S6: Use the test set to evaluate the trained convolutional neural network model, and apply the trained convolutional neural network model to the actual rapid prediction of ship cabin noise.

[0116] Implementation method 5: See Figure 3 This embodiment is described. This embodiment is to illustrate a method for rapid prediction of ship cabin noise based on a convolutional neural network proposed in the above embodiment by way of example;

[0117] Step S1: Determine the applicability of the forecast object and the operating conditions. In this embodiment, the noise level of a cabin of a ship is forecasted under the normal operation of the main engine and the motor.

[0118] Step S2: According to the cabin structure, outfitting materials, ship structure, equipment layout and equipment data, the ship cabin noise is simulated and preprocessed, and the measured noise levels of various ships are collected to obtain a series of data sets. The cross-validation method is used to divide and combine the data into more training sets, test sets and validation sets, where the training set is 70% of the total data set, the test set is 20%, and the validation set is 10%. The total number of data sets is 3786;

[0119] Step S3: Construct a convolutional neural network model for rapid prediction of ship cabin noise, including five convolution modules connected in series. The structure of each convolution module is consistent, including a convolution layer with a convolution kernel of 3×3, a normalization layer, and a pooling layer with a size of 2×2 and a step size of 2×2, which are connected in sequence to extract features to reduce the data size. The fully connected layer is connected to the last convolution module, and the features output by the last convolution module are fully connected, and the Sigmoid activation function is used to output the rapid prediction result of ship cabin noise.

[0120] The convolution layer processing method of the nth convolution module in the convolutional neural network model is:

[0121] F n =S(F n-1 )

[0122] Among them, F n represents the output of the convolutional layer in the nth convolutional module, S(F n-1 ) represents the F input to the convolutional layer n-1 Find the convolution;

[0123] The normalization layer processing method of the nth convolution module in the convolutional neural network model is:

[0124]

[0125] Among them, C nrepresents the features obtained by the network through convolution operation in the nth convolution module, E(C n ) indicates that C n Find the expectation, D(C n ) indicates that C n Find the variance, G n Represents the output F of the nth convolutional module n The output after normalization layer processing, ε represents the offset;

[0126] The pooling layer processing method of the nth convolution module in the convolutional neural network model is:

[0127]

[0128] Among them, R n represents the output of the pooling layer, G n is the output after normalization, q n represents the size of the convolution kernel, k n Indicates the step length.

[0129] Step S4: The sample data size of the unified input convolutional neural network model is 100×100×16 image tensor, the training sample set is input into the convolutional neural network model for training, and the Adam optimization algorithm is used to build the best training model. The expression of the Adam optimization algorithm is:

[0130]

[0131] Among them, θ t represents the model parameters at the tth iteration, η represents the learning rate, and denote the first-order moment estimate and the second-order moment estimate of the gradient, respectively. Where β1 is the exponential decay rate of the first-order moment estimate, which is set to 0.9, β2 is the exponential decay rate of the second-order moment estimate, which is set to 0.9999, and ε is the offset, which is very small and is used to prevent division by zero in the implementation.

[0132] Step S5: Verify the weights and thresholds of the neural network in training through the validation set and calculate the gradient through the loss function and back-propagate and iteratively update the parameters until the network converges;

[0133] Step S6: Use the test set to evaluate the trained convolutional neural network model, and apply the trained convolutional neural network model to the rapid prediction of the noise in the central control room of a ship in this embodiment, and obtain the A sound level noise result under the standard octave band, such as Figure 3 shown.

[0134] Embodiment 6: The method for rapid prediction of ship cabin noise based on convolutional neural network described in the above embodiment can be implemented entirely by computer software. Therefore, correspondingly, this embodiment provides a rapid prediction system for ship cabin noise based on convolutional neural network, and the rapid prediction system for ship cabin noise includes a storage device, and the storage device includes:

[0135] Used to determine whether the forecast object and forecast method are applicable;

[0136] Used to collect data related to ship cabin noise, including cabin structure, outfitting materials, ship structure, equipment layout and equipment data;

[0137] Used to preprocess the collected data, including data cleaning, normalization, feature selection, and divide the collected data into training sets, validation sets, and test sets;

[0138] It is used to input the training sample set into the pre-set convolutional neural network for training, and use the Adam optimization algorithm to build the best training model;

[0139] It is used to verify the weights and thresholds of the neural network in training through the validation set, calculate the gradient through the loss function, and back-propagate and iteratively update the parameters until the network converges;

[0140] It is used to evaluate the trained convolutional neural network model using the test set, and apply the trained convolutional neural network model to the rapid prediction of actual ship cabin noise.

[0141] The ship cabin noise rapid prediction system proposed in this embodiment is applicable to marine navigation equipment including but not limited to medium and large ships, offshore platforms, underwater submersibles, etc., which have high requirements on cabin noise levels and need for rapid cabin noise prediction and assessment.

[0142] The ship cabin noise rapid prediction system proposed in this embodiment predicts ship cabin noise by constructing a convolutional neural network, wherein, since the convolutional neural network has good pattern recognition ability, it can automatically learn the complex characteristics and patterns of ship cabin noise, and has strong processing and expression capabilities when processing large-scale data, thereby providing more accurate prediction results, achieving higher prediction accuracy and efficiency, and helping to accurately evaluate the ship cabin noise level. Furthermore, the convolutional neural network can automatically extract important features from the input data, and can capture more important information from complex data when predicting ship cabin noise. Compared with other methods, it has the ability to adapt to various noise prediction problems.

[0143] Furthermore, the ship cabin noise prediction system proposed in this embodiment adapts to different ship cabin noise prediction needs by increasing the number of convolutional neural network layers or adjusting the network structure, and has good scalability and wide applicability, making the prediction method based on convolutional neural networks more flexible and customizable.

[0144] Embodiment 7. This embodiment provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, a method for rapid prediction of ship cabin noise based on a convolutional neural network proposed in the above embodiment is implemented.

[0145] Embodiment 8. This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a method for rapid prediction of ship cabin noise based on a convolutional neural network proposed in the above embodiment is implemented.

[0146] The ship cabin noise prediction method is:

[0147] Step S1: Determine whether the forecast object and forecast method are applicable;

[0148] Step S2: Collecting data related to ship cabin noise, including cabin structure, outfitting materials, ship structure, equipment layout and equipment data;

[0149] Step S3: preprocessing the collected data, including data cleaning, normalization, feature selection, and dividing the collected data into a training set, a validation set, and a test set;

[0150] Step S4: input the training sample set into the convolutional neural network model for training, and use the Adam optimization algorithm to build the best training model;

[0151] Step S5: Verify the weights and thresholds of the neural network in training through the validation set and calculate the gradient through the loss function and back-propagate and iteratively update the parameters until the network converges;

[0152] Step S6: Use the test set to evaluate the trained convolutional neural network model, and apply the trained convolutional neural network model to the actual rapid prediction of ship cabin noise.

[0153] A computer device provided in this embodiment, the hardware device of this part is a general model and is not shown in the form of a diagram. The system includes a processor and a memory, wherein the processor and the memory can be connected via a bus or other means. The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer executable programs and modules, and corresponding program instructions / modules. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions and modules stored in the memory, so as to realize the method and steps for rapid prediction of ship cabin noise based on convolutional neural network in the above method embodiment.

[0154] The above description is only the implementation mode of the present invention and is not limited to the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. A ship cabin noise prediction model based on convolutional neural network, characterized in that: Includes feature extraction module and convolutional neural network model; The feature extraction module is used to extract features and input the extracted features into the convolutional neural network model; The convolutional neural network model includes N convolutional modules connected in series and a fully connected layer; The convolution module includes a convolution layer, a normalization layer and a pooling layer, which are used to extract features to reduce data size; The fully connected layer is connected to the last convolution module, the features output by the last convolution module are fully connected, and the Sigmoid activation function is used to output the rapid prediction result of the ship cabin noise.

2. According to claim 1, a rapid prediction model for ship cabin noise based on convolutional neural network is characterized in that: The structure of each convolution module is consistent, including a convolution layer with a 3×3 convolution kernel, a normalization layer, and a pooling layer with a size of 2×2 and a stride of 2×2, which are connected in sequence.

3. The noise prediction method implemented by the ship cabin noise prediction model based on convolutional neural network according to any one of claims 1 to 2, characterized in that: The method is: Step S1: Determine whether the forecast object and forecast method are applicable; Step S2: Collecting data related to ship cabin noise, including cabin structure, outfitting materials, ship structure, equipment layout and equipment data; Step S3: preprocessing the collected data, including data cleaning, normalization, feature selection, and dividing the collected data into a training set, a validation set, and a test set; Step S4: input the training sample set into the convolutional neural network model for training, and use the Adam optimization algorithm to build the best training model; Step S5: Verify the weights and thresholds of the neural network in training through the validation set and calculate the gradient through the loss function and back-propagate and iteratively update the parameters until the network converges; Step S6: Use the test set to evaluate the trained convolutional neural network model, and apply the trained convolutional neural network model to the actual rapid prediction of ship cabin noise.

4. According to claim 3, a method for rapid prediction of ship cabin noise based on convolutional neural network is characterized in that: Step S2 is specifically as follows: Step S21: The cabin structure data includes the shape and size of the cabin; the outfitting material data includes the material surface density and the vibration isolation coefficient; the equipment data includes the equipment layout location and the noise data under the working state; Step S22: Simulate the ship cabin noise and pre-process or collect the measured noise levels of various types of ships to obtain a series of data sets.

5. According to claim 3, a method for rapid prediction of ship cabin noise based on convolutional neural network is characterized in that: Step S3 is specifically as follows: Step S31: using a cross-validation method to divide the data set into a training set, a test set and a validation set, wherein the training set accounts for 70% of the total data set, the test set accounts for 20%, and the validation set accounts for 10%; Step S32: During data cleaning, check whether there are missing values ​​or outliers in the data, and clean the data by filling missing values, deleting outliers, or performing data interpolation; Step S33: during feature selection, correlation analysis is used to evaluate the degree of association between each feature and the cabin noise result; Step S34: The data size of the unified input convolutional neural network model is a 100×100×16 image tensor. After the convolutional layer feature map processing and pooling operation of five convolutional modules, the feature map is reduced to a tensor feature map of size 8×8×25. The size of the fully connected layer is 1600×1, and the feature map is converted into a noise level result represented by a vector.

6. According to claim 3, a method for rapid prediction of ship cabin noise based on convolutional neural network is characterized in that: Step S4 is specifically as follows: Step S41: According to the convolutional neural network model, the Adam optimization algorithm is used as the loss function. The expression of the Adam optimization algorithm is: Among them, θ t represents the model parameters at the tth iteration, η represents the learning rate, and denote the first-order moment estimate and the second-order moment estimate of the gradient, respectively. Where β1 is the exponential decay rate of the first-order moment estimate, which is set to 0.9, β2 is the exponential decay rate of the second-order moment estimate, which is set to 0.9999, and ε is the offset; Step S42: By solving the above equations, the updated optimized weights and biases can be obtained, thereby obtaining the optimal convolutional neural network model for predicting ship cabin noise.

7. According to claim 3, a method for rapid prediction of ship cabin noise based on convolutional neural network is characterized in that: Step S5 is specifically as follows: Step S51: Adopting the Adam optimization algorithm to update the learning rate and model parameters of the convolutional neural network in real time; Step S52: setting a forecast result threshold corresponding to the ship cabin noise, and outputting an alarm signal if the forecast result is greater than or equal to the threshold.

8. A rapid prediction system for ship cabin noise based on convolutional neural network, characterized in that: The system includes a storage device, the storage device including: Used to determine whether the forecast object and forecast method are applicable; Used to collect data related to ship cabin noise, including cabin structure, outfitting materials, ship structure, equipment layout and equipment data; Used to preprocess the collected data, including data cleaning, normalization, feature selection, and divide the collected data into training sets, validation sets, and test sets; It is used to input the training sample set into the convolutional neural network model for training, and use the Adam optimization algorithm to build the best training model; It is used to verify the weights and thresholds of the neural network in training through the validation set, calculate the gradient through the loss function, and back-propagate and iteratively update the parameters until the network converges; It is used to evaluate the trained convolutional neural network model using the test set, and apply the trained convolutional neural network model to the rapid prediction of actual ship cabin noise.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, it implements the method for rapid prediction of ship cabin noise based on convolutional neural network as described in any one of claims 3 to 7.

10. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, a method for rapid prediction of ship cabin noise based on a convolutional neural network as described in any one of claims 3 to 7 is implemented.

Images