Construction method of ship underwater radiation noise online prediction model

Abstract

The invention discloses a construction method of a ship underwater radiation noise online prediction model. The method comprises the following steps: collecting ship body vibration signals and far-field radiation noise signals as a sample database of a radiation noise prediction model; on the basis of a differential evolution feature optimization algorithm, two dimensions of a vibration sensor and a 1 / 3 octave frequency band of prediction model input features are optimized, and an optimal feature subset is obtained to serve as input of a prediction model; establishing an integrated learning prediction model, taking a ridge regression model as a base learning device, and determining optimal parameters of each base learning device by adopting grid search and cross validation to obtain n different base learning devices; and finally, taking a mean value of prediction results of the base learners as a final prediction result of the radiation noise prediction model, and realizing online prediction of the ship radiation noise. According to the method, the utilization rate of existing test data can be improved, the degree of dependence of a prediction model on specific distribution type data is reduced, the sensitivity of the model to small changes of data distribution is relieved, and the model robustness and prediction precision are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of ship noise monitoring and evaluation, and in particular, relates to a method for constructing an online prediction model for ship underwater radiated noise. Background Art

[0002] When a ship is sailing at sea, the operation of mechanical equipment such as the power unit will cause the hull to vibrate, and the vibration energy will propagate to the far field through the water medium, thereby generating underwater radiation noise. Excessive radiation noise not only affects the acoustic stealth performance of the ship, but also causes underwater environmental noise pollution. Therefore, real-time monitoring and forecasting of underwater radiation noise of ships is of great significance for timely grasping and maintaining the acoustic performance of ships, guiding the work of reducing vibration and noise of ships, and controlling the underwater radiation noise generated by ships.

[0003] For the prediction method of underwater radiated noise of ships, the experimental transfer function method is currently widely used in engineering. The surface vibration of the ship is taken as the reference point response, and the far-field radiated noise is taken as the target point response. The reference point response and the target point response under different working conditions of the ship are used to construct the vibration-radiated noise transfer function of "response-transfer path-response". Combined with the real-time monitoring data of the hull vibration, the real-time prediction of the ship's radiated noise during navigation can be achieved. However, in actual engineering, the number of ship radiated noise test conditions is less than the number of vibration monitoring sensors, resulting in underdetermined linear equations for solving the transfer function, which may produce large calculation errors under small disturbances and poor model robustness. Summary of the invention

[0004] In view of the above problems existing in the prior art, the present invention provides a method for constructing an online prediction model for underwater radiated noise of ships, which is used to solve the ill-posed problem of the traditional test transfer function method when the number of test condition samples is less than the number of sensors. The "black box" model is used to characterize the complex mapping relationship between vibration and radiated noise, and the differential evolution algorithm is used to optimize the input features of sample data to improve the utilization rate of limited data; and multiple base learners are fused through an integrated learning algorithm to reduce the sensitivity of the prediction model to data disturbances and improve the prediction accuracy of the model.

[0005] To achieve the above object, the present invention provides a method for constructing an online prediction model of underwater radiated noise of a ship, which is used to realize real-time prediction of underwater radiated noise of a ship according to hull vibration monitoring data, comprising: S1: Using the ship's vibration and acoustic radiation test, vibration sensors are arranged at the noise source locations of the ship's main engine, generator and propeller to measure the hull vibration, and hydrophones are arranged in the far field to measure the ship's underwater radiated noise. The hull vibration and radiated noise time domain signals under the ship's stable navigation conditions are collected at a certain sampling rate.

[0006] S2: Preprocessing the time domain signals of the hull vibration and radiated noise, performing Fourier transform on the time domain signals, converting the time domain signals into 1 / 3 octave band form in the frequency domain, and performing environmental background noise correction on the 1 / 3 octave band sound pressure level of the radiated noise.

[0007] S3: Based on the preprocessing results of hull vibration and radiated noise signals, a ship radiated noise prediction model based on differential evolution ensemble learning is constructed.

[0008] S4: While the ship is underway, the hull vibration monitoring data is processed according to the preprocessing method described in step S2, and the processed data is input into the ship radiated noise prediction model described in step S3.

[0009] S5: At regular time intervals, the ship radiation noise prediction model is used to calculate the real-time underwater radiation noise level of the ship.

[0010] Specifically, step S3 constructs a ship radiated noise prediction model based on differential evolution ensemble learning based on the preprocessing results of the hull vibration and radiated noise signals, and includes the following steps: S31: Establish a ridge regression model to calculate the individual fitness function value in the differential evolution algorithm.

[0011] S32: Iteratively optimize the input features of the preprocessed vibration signal data according to the differential evolution algorithm steps, determine the optimal vibration sensor and the optimal 1 / 3 octave frequency band corresponding to each radiation noise prediction frequency band, and form a new sample data set D.

[0012] S33: Using the ridge regression model as the base learner of the ensemble learning model, the sample data set D of step S32 is randomly divided into n data sets of the same size. D 1 ,D 2 , … ,D n , each time a data set is selected D j Do not participate in training, the remaining data set D -j As the sample data of the base learner of the ensemble learning model, grid search and cross validation are used to obtain n base learners with different hyperparameters.

[0013] S34: The output results of n base learners are integrated by using the averaging ensemble learning method to form a ship underwater radiated noise prediction model based on differential evolution ensemble learning.

[0014] Specifically, step S31 includes the following steps: S311: The preprocessed vibration and radiation noise signal data are used as original data set training samples, and the samples are normalized. Grid search and cross-validation are used to determine the optimal hyperparameters of the ridge regression model.

[0015] S312: Using the ridge regression model to calculate the individual fitness function in the differential evolution process, The calculation method of individual fitness function can be expressed as: ; in, h ( x ) is the ridge regression model; ( x,y )∈ D train and( x,y )∈ D val They are training set data and validation set data respectively; L 1 、L 2 They are the sensor and frequency band indexes after iteration respectively; x ( L 1 ,L 2 ) is the sample feature in the iterative process; α,β are the weights of the training set error and the validation set error, respectively.

[0016] Specifically, step S32 includes the following steps: S321: Define the initial population, determine the population size and the number of algorithm iterations.

[0017] S322: Use random initialization method to obtain the initial solution of individuals in the population.

[0018] S323: Calculate the fitness value of each generation of individuals according to the individual fitness function described in step 312.

[0019] S324: Perform mutation, crossover and selection operations.

[0020] S325: Repeat steps S323 and S324 until a preset maximum number of iterations is met.

[0021] S326: Obtain an optimal vibration sensor and an optimal frequency band.

[0022] Specifically, in the method of step S34, the prediction result calculation formula of the ship radiated noise prediction model is: ;

[0023] in, y predis the final radiation noise prediction result of the prediction model, n is the number of base learners in the ensemble learning model, is the radiation noise prediction result of the i-th base learner.

[0024] Furthermore, when predicting the radiated noise of a certain 1 / 3 octave frequency band of the ship, the hull vibration data of the corresponding frequency band and the hull vibration data of the non-prediction frequency band are used.

[0025] Furthermore, when predicting different 1 / 3 octave frequency bands of ship radiated noise, a differential evolution ensemble learning model with different parameters is used to adapt to the changes in the optimal vibration sensors and optimal frequency bands corresponding to different 1 / 3 octave frequency bands of ship radiated noise, and the changes in the input characteristics of the sample data used to construct the model.

[0026] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects: 1. The present invention provides a method for constructing an online prediction model for underwater radiated noise of ships. Through vibration sensors arranged on the hull, the underwater radiated noise of ships under current working conditions can be calculated in real time. Compared with traditional radiated noise prediction methods, it has the advantages of rapidity and real-time performance.

[0027] 2. Furthermore, the method for constructing an online prediction model of underwater radiated noise of ships provided by the present invention adopts a differential evolution algorithm to optimize the features of the input data of the prediction model, and uses the weighted average of the training set error and the validation set error of the model as the fitness function, which can more efficiently search for the optimal individual globally and improve the generalization performance of the model.

[0028] 3. Furthermore, the method for constructing an online prediction model for underwater radiated noise of ships provided by the present invention makes full use of limited vibration sensor data, optimizes sensors and frequency bands at the same time, and screens out key features that have a greater impact on radiated noise. On the one hand, it reduces redundant feature interference, and on the other hand, it improves the utilization rate of existing test data, effectively controls model overfitting, and alleviates the problem of insufficient data utilization in current radiated noise prediction.

[0029] 4. Furthermore, the method for constructing an online prediction model for underwater radiated noise of ships provided by the present invention adopts an integrated learning model based on ridge regression, divides the sample data into n parts on average, selects any n-1 parts as sub-sample data, and uses grid search and cross-validation to determine the optimal parameters of the ridge regression model, and makes full use of multiple groups of sub-sample data to achieve the diversity of base learners, reduce the dependence of base learners on specific types of data, alleviate the sensitivity of the prediction model to slight changes in data distribution, and improve the prediction accuracy and robustness of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1A flow chart of a method for constructing an online prediction model for underwater radiated noise of a ship provided in an embodiment of the present invention.

[0031] Figure 2 A schematic diagram of the principle of an integrated learning algorithm provided in an embodiment of the present invention.

[0032] Figure 3 A flow chart of a differential evolution algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described implementation is a part of the implementation of the present invention, not all of the implementation. Based on the implementation of the present invention, all other implementations obtained by ordinary technicians in this field without making creative work results belong to the protection scope of the present invention.

[0034] For better understanding, Figure 1 The flowchart of the construction method of the online prediction model of the underwater radiated noise of ships is as follows; Figure 1 As shown in the figure, the online prediction of underwater radiated noise of ships includes the following steps: S1: Using the ship's vibration and acoustic radiation test, vibration sensors are evenly arranged at important parts of the hull, such as the main engine, generator, propeller and other noise source locations to measure the hull vibration, and hydrophones are arranged in the far field to measure the ship's underwater radiated noise. The hull vibration and radiated noise time domain signals under the ship's stable navigation conditions are collected at a certain sampling rate; S2: Perform Fourier transform on the vibration and radiation noise time domain signals to convert the time domain signals into 1 / 3 octave band form in the frequency domain; meanwhile, the 1 / 3 octave band sound pressure level of the radiation noise needs to be corrected for the environmental background noise; S3: Based on the preprocessing results of hull vibration and radiated noise signals, a ship radiated noise prediction model based on differential evolution ensemble learning is constructed; S4: While the ship is underway, the hull vibration monitoring data is processed according to the preprocessing method described in step S2, and the processed data is input into the radiation noise prediction model; S5: At regular time intervals, the model is used to calculate the real-time underwater radiation noise level of the ship.

[0035] The above embodiments constitute a complete technical solution of the present invention. Different from the prior art, the above embodiments adopt a machine learning method, utilize a differential evolution algorithm and an integrated learning model to construct a ship underwater radiated noise prediction model, which can effectively improve the prediction accuracy of the prediction model.

[0036] In this embodiment, in order to obtain the 1 / 3 octave form in the frequency domain described in step S2, it is necessary to design a center frequency of f c The upper and lower limits of the 1 / 3 octave filter are f up and f low .

[0037] ; ; . ;

[0038] in, H ( f ) is the frequency response of the filter.

[0039] In this embodiment, in step S2, the environmental background noise correction needs to first calculate the arithmetic mean of the environmental background noise before and after the vibration sound radiation test. Ln ;

[0040] L n1 , L n2 The 1 / 3 octave band sound pressure level of the background noise measured by the hydrophone before and after the measurement. The original underwater noise sound pressure level (including background noise) L np With background noise L n The difference, that is, the signal-to-noise ratio of the underwater noise signal is: ;

[0041] If the signal-to-noise ratio is greater than a dB, no correction is required; if the signal-to-noise ratio is less than b dB, the data does not meet the measurement requirements and should be discarded. b≤ Δ ≤a , then the correction is made according to the following formula, where b < a .

[0042] ;

[0043] L p It is the 1 / 3 octave band sound pressure level of underwater radiated noise after correction for background noise.

[0044] In this embodiment, the ship radiated noise prediction model based on differential evolution ensemble learning described in step S3 includes the following steps: S31: Establish a ridge regression model to calculate the individual fitness function value in the differential evolution algorithm; S32: performing iterative optimization of input features of the preprocessed vibration signal data according to the differential evolution algorithm steps, determining the optimal vibration sensor and the optimal 1 / 3 octave frequency band corresponding to each radiation noise prediction frequency band, and forming a new sample data set D; S33: Using the ridge regression model as the base learner of the ensemble learning model, the sample data set D in step S32 is randomly divided into n data sets of the same size. D 1 , D 2 , … , D n , each time a data set is selected D j Do not participate in training, the remaining data set D -j As sample data of the base learner of the ensemble learning model, grid search and cross validation are used to obtain n base learners with different hyperparameters; S34: The output results of n base learners are fused using the ensemble learning method of averaging, and finally the prediction result of ship underwater radiated noise based on differential evolution ensemble learning is obtained.

[0045] In this embodiment, the schematic diagram of the integrated learning algorithm principle of steps S33 and S34 is as follows: Figure 2 shown.

[0046] In this embodiment, step S31 includes the following steps: S311: using the preprocessed vibration and radiation noise signal data as original data set training samples, performing sample normalization, and using grid search and cross validation to determine the optimal hyperparameters of the ridge regression model; S312: Calculate the individual fitness function in the differential evolution process using the ridge regression model.

[0047] The calculation method of individual fitness function can be expressed as:

[0048] in, h ( x ) is the ridge regression model; ( x,y )∈ D train and( x,y )∈ D valThey are training set data and validation set data respectively; L 1 、L 2 They are the sensor and frequency band indexes after iteration respectively; x ( L 1 ,L 2 ) is the sample feature in the iterative process; α,β are the weights of the training set error and the validation set error, respectively.

[0049] In this embodiment, the differential evolution algorithm flow chart described in step S32 is as follows: Figure 3 As shown, where P is the number of individuals in the population, p is the number of sensors, and q represents the number of 1 / 3 octave frequency bands, the following steps are included: S321: define the initial population, determine the population size and the number of algorithm iterations; S322: Using random initialization method to obtain initial solutions of individuals in the population; S323: Calculate the fitness value of each generation of individuals according to the fitness function described in step 312; S324: performing mutation, crossover, and selection operations; S325: Repeat steps S323 and S324 until a preset maximum number of iterations is met; S326: Get the best vibration sensor and 1 / 3 octave frequency band.

[0050] In this embodiment, the base learner of the integrated learning model in step S33 is a ridge regression model.

[0051] In the method described in step S34, the final prediction result calculation formula of the ship underwater radiated noise is: ; in, y pred is the final radiation noise prediction result of the prediction model, n is the number of base learners in the ensemble learning model, For the Radiated noise prediction results of a basis learner.

[0052] The present invention can improve the utilization rate of existing test data, reduce the dependence of the prediction model on data of a specific distribution type, alleviate the sensitivity of the model to slight changes in data distribution, and improve the robustness of the model and the prediction accuracy. By arranging vibration sensors, the underwater radiated noise of the ship under the current working conditions can be calculated in real time, which has the advantages of rapidity and real-time performance compared with traditional radiated noise prediction methods.

[0053] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for constructing an online prediction model for underwater radiated noise of a ship, characterized in that: It includes the following steps: S1: Using the ship's vibration and acoustic radiation test, vibration sensors are arranged at the noise source locations of the ship's main engine, generator and propeller to measure the hull vibration, and hydrophones are arranged in the far field to measure the ship's underwater radiated noise. The hull vibration and radiated noise time domain signals under the ship's stable navigation conditions are collected at a certain sampling rate; S2: Preprocessing the time domain signals of the hull vibration and radiated noise, performing Fourier transform on the time domain signals, converting the time domain signals into 1 / 3 octave band form in the frequency domain, and performing environmental background noise correction on the 1 / 3 octave band sound pressure level of the radiated noise; S3: Based on the preprocessing results of hull vibration and radiated noise signals, a ship radiated noise prediction model based on differential evolution ensemble learning is constructed; S4: While the ship is underway, the hull vibration monitoring data is processed according to the preprocessing method described in step S2, and the processed data is input into the ship radiated noise prediction model described in step S3; S5: At regular time intervals, the real-time underwater radiated noise level of the ship is calculated using the ship radiated noise prediction model.

2. The method for constructing an online prediction model for underwater radiated noise of a ship according to claim 1, characterized in that: The step S3 constructs a ship radiated noise prediction model based on differential evolution ensemble learning based on the preprocessing results of the hull vibration and radiated noise signals, and includes the following steps: S31: Establish a ridge regression model to calculate the individual fitness function value in the differential evolution algorithm; S32: performing iterative optimization of input features of the preprocessed vibration signal data according to the differential evolution algorithm steps, determining the optimal vibration sensor and the optimal 1 / 3 octave frequency band corresponding to each radiation noise prediction frequency band, and forming a new sample data set D; S33: Using the ridge regression model as the base learner of the ensemble learning model, the sample data set D of step S32 is randomly divided into n data sets of the same size. D 1 ,D 2 , … ,D n , each time a data set is selected D j Do not participate in training, the remaining data set D -j As sample data of the base learner of the ensemble learning model, grid search and cross validation are used to obtain n base learners with different hyperparameters; S34: The output results of n base learners are fused using the ensemble learning method of averaging to form a ship radiated noise prediction model based on differential evolution ensemble learning.

3. The method for constructing an online prediction model for underwater radiated noise of a ship according to claim 2, characterized in that: The step S31 comprises the following steps: S311: using the preprocessed vibration and radiation noise signal data as original data set training samples, performing sample normalization, and using grid search and cross validation to determine the optimal hyperparameters of the ridge regression model; S312: Using the ridge regression model to calculate the individual fitness function in the differential evolution process, The individual fitness function calculation method can be expressed as: ； in, h ( x ) is the ridge regression model; ( x,y )∈ D train and( x,y )∈ D val They are training set data and validation set data respectively; L 1 、L 2 They are the sensor and frequency band indexes after iteration respectively; x ( L 1 ,L 2 ) is the sample feature in the iterative process; α , β are the weights of the training set error and the validation set error, respectively.

4. The method for constructing an online prediction model for underwater ship radiated noise according to claim 3, characterized in that: The step S32 comprises the following steps: S321: define the initial population, determine the population size and the number of algorithm iterations; S322: Using random initialization method to obtain initial solutions of individuals in the population; S323: Calculate the fitness value of each generation of individuals according to the individual fitness function described in step 312; S324: performing mutation, crossover and selection operations; S325: Repeat steps S323 and S324 until a preset maximum number of iterations is met; S326: Get the best vibration sensor and 1 / 3 octave frequency band.

5. The method for constructing an online prediction model for underwater radiated noise of a ship according to claim 2, characterized in that: The prediction result calculation formula of the ship radiated noise prediction model in step S34 is: ； in, y pred is the final radiation noise prediction result of the prediction model, n is the number of base learners in the ensemble learning model, For the i Radiated noise prediction results of a basis learner.

6. The method for constructing an online prediction model for underwater radiated noise of a ship according to any one of claims 1 to 5, characterized in that: When predicting the radiated noise of a certain 1 / 3 octave frequency band of a ship, the hull vibration data of the corresponding frequency band and the hull vibration data of the non-prediction frequency band are used.

7. The method for constructing an online prediction model for underwater radiated noise of a ship according to any one of claims 1 to 5, characterized in that: When predicting different 1 / 3 octave frequency bands of ship radiated noise, a differential evolution ensemble learning model with different parameters is used to adapt to the changes in the optimal vibration sensors and optimal frequency bands corresponding to different 1 / 3 octave frequency bands of ship radiated noise, and the changes in the input characteristics of the sample data used to build the model.

Images