A method and system for predicting ship propeller efficiency

Abstract

This invention belongs to the field of ship propulsion efficiency prediction technology, and discloses a method and system for predicting ship propeller efficiency. The method includes constructing an autoencoder and pre-training it on data; using the trained autoencoder as input, constructing a physical information neural network using the relevant physical equations of propeller propulsion efficiency; introducing a regularization term to determine the final loss function; predicting propeller efficiency using the trained neural network; and iterative optimization and application of the model. This invention, by introducing an autoencoder and a regularized loss function, effectively improves the accuracy and reliability of propeller efficiency prediction under complex sea conditions, and overcomes the limitations of traditional prediction models in handling nonlinear dynamic responses.

Description

Technical Field

[0001] This invention relates to a method and system for predicting ship propeller efficiency. This invention belongs to the interdisciplinary field of applying machine learning and deep learning technologies with physical process simulation, and aims to improve the speed and accuracy of predicting propeller efficiency during the ship propeller design stage. Background Technology

[0002] Ship propeller efficiency is an important indicator for evaluating propeller performance. The main methods for evaluating propeller efficiency include: empirical graph method, numerical simulation method, and model test method.

[0003] Prior to this method, Sun Haixiao et al. from the Shanghai Ship Design and Research Institute used a second-layer evaluation method to analyze the main factors affecting efficiency evaluation results, evaluated the efficiency of propellers with different design parameters, and conducted a comparative analysis, providing a certain reference for the design and development of related ship types and for propeller design to meet efficiency evaluation requirements. Sun Mengguo, Xiang Zuquan, et al. from Wuhan University of Technology used FLUENT software to simulate the hydrodynamic performance of the P4119 propeller, obtaining 100 sets of data as training samples. They established a GA-BP neural network propeller hydrodynamic performance prediction model and compared its prediction effect with that of the traditional BP neural network model, further confirming the feasibility of the neural network method.

[0004] Neural networks, as an emerging technology, offer a solution that guarantees both prediction accuracy and significantly improves computational efficiency. The core advantage of neural networks lies in their ability, once successfully trained, to provide propeller efficiency under specific operating conditions by inputting parameters into a trained model. Neural network methods achieve prediction accuracy similar to, or even better than, traditional methods; their efficiency far surpasses several traditional methods, achieving second-level predictions with a well-trained model.

[0005] Based on the above analysis, the problems and shortcomings of the existing technology are as follows: the existing technology has low accuracy in predicting ship propulsion efficiency under complex sea conditions. Summary of the Invention

[0006] To overcome the problems existing in related technologies, this invention discloses a method and system for predicting ship propeller efficiency, specifically a method for predicting ship propeller efficiency based on a regular physical information neural network. This involves the interdisciplinary field of applying machine learning and deep learning techniques with physical process simulation, aiming to improve the speed and accuracy of ship propeller efficiency prediction.

[0007] The technical solution is as follows: A method for predicting the efficiency of a ship propeller, comprising the following steps:

[0008] S1, construct an autoencoder and pre-train the data;

[0009] S2, using the encoder part of the trained autoencoder as input, constructs a physical information neural network;

[0010] S3, Introduce a regularization term to determine the final loss function;

[0011] S4, using the Adam optimizer to train the physical information neural network model;

[0012] S5 uses a trained neural network to predict propeller efficiency.

[0013] S6, Model Iterative Optimization and Application.

[0014] In steps S1 and S2, when extracting key parameters from the data, the pre-trained autoencoder is used as the input to the subsequent network.

[0015] Constructing an autoencoder structure: First, build an autoencoder containing an encoder and a decoder. The encoder is responsible for compressing and mapping the original high-dimensional data into low-dimensional feature vectors, while the decoder is responsible for reconstructing the feature vectors back to the original data dimensions.

[0016] Pre-trained autoencoder: The autoencoder is trained with a sufficient amount of raw data. The loss function is based on the reconstruction error, and the optimization goal is to make the reconstructed data output by the decoder as consistent as possible with the original input data. After training, the encoder has learned the core rules of the data, and its output low-dimensional vector is the key parameter of the data.

[0017] Stripping the encoder as a feature extractor: Discard the trained decoder and keep only the encoder module.

[0018] Connecting to subsequent task networks: Input the raw data to be processed into the preserved encoder, and the low-dimensional feature vector output by the encoder is directly used as the input layer of the subsequent network; if the amount of data is small, the encoder parameters can be fixed and only the subsequent network can be trained; if the data is sufficient, some layers of the encoder can be unfrozen and fine-tuned together with the subsequent network to make the key parameters more suitable for the specific task.

[0019] In step S2, the output of the physical information neural network model is the propeller lift-to-drag ratio, circulation, circumferential induced velocity, and axial induced velocity. The loss function of the physical information neural network includes constraints on energy conservation, constraints on the real-time propeller propulsion efficiency output parameters, and regularization terms, given a known and determined ship type.

[0020] In step S2, the method for constructing the loss function of the physical information neural network specifically includes:

[0021] For cases where the ship's propulsion efficiency is determined, the loss between the predicted efficiency and the determined propulsion efficiency of the physical information neural network is used as a function to impose physical constraints on the network, constructing the following loss function:

[0022] ;

[0023] In the formula, This is a predicted value for the current propeller efficiency. This represents the difference between the predictive efficiency and deterministic efficiency of a neural network for physical information. For regularization constraints, These are the weights of the two loss terms and the regularization term, respectively. This is the total loss function.

[0024] Furthermore, the gap between the prediction efficiency and deterministic efficiency of physical information neural networks. The expression is:

[0025] ;

[0026] In the formula, Where is the propeller radius. The propeller hydrodynamic pitch angle, The lift-to-drag ratio of the propeller. This refers to the axial induced velocity during propeller operation. This refers to the circumferential induced velocity during propeller operation. The circulation of fluid around the propeller. The density of the fluid around the propeller. For propeller mass;

[0027] Current propeller efficiency prediction The expression is:

[0028] ;

[0029] In the formula, This represents the actual efficiency of the propeller. To improve the prediction efficiency of neural networks;

[0030] Regularization constraints The expression is:

[0031] ;

[0032] In the formula, The predicted value for propeller efficiency is given by the neural network. The average propeller efficiency in the training set

[0033] In step S4, training the physical information neural network model using the Adam optimizer includes: Adam simultaneously calculates the exponential moving averages of the first and second moments of the gradient, and performs bias correction to accelerate the initial learning rate and approach the global optimum. The Adam optimizer performs exponential moving averages of the first and second moments of the gradient, and uses a decay coefficient to smooth the historical gradient direction and fluctuation amplitude; since the initial moment estimate is biased to 0, bias correction is performed on the two moments to correct the initial error; then the corrected first moment provides momentum, and the second moment is used for adaptive learning rate scaling, combined with the base learning rate to stably update the parameters, which not only accelerates the initial convergence speed and reduces oscillations, but also improves the later optimization accuracy and more robustly approaches the global optimum.

[0034] In step S5, the trained physical information neural network model is used to predict the propeller circulation. propeller lift-to-drag ratio Axial induced velocity Circumferential induced velocity Based on the above predictions, the propeller propulsion efficiency is further calculated as follows:

[0035] The propeller's advance rate propeller radius propeller speed propeller mass As input parameters, key features are extracted using an autoencoder, and then the axial velocity of the ship's propeller at the current speed is predicted using a physical information neural network. Circumferential speed Rise-to-drag ratio Circulation The expression is:

[0036] ;

[0037] Furthermore, from the axial velocity Circumferential speed Rise-to-drag ratio Circulation The propeller efficiency was calculated. The expression is:

[0038] ;

[0039] ;

[0040] ;

[0041] ;

[0042] Where T is the propeller thrust and Q is the propeller torque. Let r be the propeller advance speed and r be the propeller radius. The propeller hydrodynamic pitch angle, The propeller angular velocity, For propeller efficiency.

[0043] Another object of the present invention is to provide a ship propeller efficiency prediction system, which implements the aforementioned ship propeller efficiency prediction method, and the system includes:

[0044] The data preprocessing module constructs an autoencoder and pre-trains the data.

[0045] The loss function construction module takes the encoder part of the trained autoencoder as input to construct a physical information neural network; it introduces a regularization term to determine the final loss function.

[0046] The model training module is used to train physical information neural network models using the Adam optimizer.

[0047] The efficiency prediction module is used to predict propeller efficiency using a trained neural network.

[0048] The iterative optimization module is used for model iterative optimization and application.

[0049] Furthermore, the aforementioned ship propeller efficiency forecasting system is mounted on a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can realize the functions of the aforementioned ship propeller efficiency forecasting system.

[0050] Furthermore, the aforementioned ship propeller efficiency forecasting system is mounted on an information data processing terminal, which provides a user input interface to implement the functions of the ship propeller efficiency forecasting system as described above when executed on an electronic device.

[0051] Combining all the above technical solutions, the beneficial effects of this invention are as follows:

[0052] This invention obtains a definite propeller propulsion efficiency by selecting fixed propeller scale parameters, compares the predicted efficiency of the network with the definite propeller efficiency as the loss function of the neural network, and uses the physical information neural network method to predict the current propeller efficiency based on the ship's sailing speed.

[0053] This invention, by introducing Physical Information Neural Networks (PINNs), is specifically designed for predicting ship propeller efficiency under known fixed parameters, demonstrating both innovation and practical value. Compared to existing technologies, this invention features a simple structure and clear algorithm, providing an accurate and reliable solution for ship efficiency prediction. Therefore, this invention not only provides an advanced technical means for ship propeller efficiency prediction systems but also opens up new paths for research and application in related fields, possessing significant theoretical importance and broad application prospects.

[0054] By accurately predicting the efficiency of ship propellers, ship designers and operators can optimize propeller design and operating parameters, thereby significantly reducing fuel consumption. This invention utilizes a physical information neural network model to perform deep learning and mining on limited historical data and simulation results, further improving the generalization ability and prediction accuracy of the prediction model. Attached Figure Description

[0055] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure;

[0056] Figure 1 This is a flowchart of the ship propeller efficiency prediction method provided in an embodiment of the present invention;

[0057] Figure 2 This is a schematic diagram of a ship propeller efficiency prediction system provided in an embodiment of the present invention;

[0058] Figure 3 This is a network framework diagram of the PINNs method provided in the embodiments of the present invention;

[0059] In the diagram: 1. Data preprocessing module; 2. Loss function construction module; 3. Model training module; 4. Efficiency prediction module; 5. Iterative optimization module. Detailed Implementation

[0060] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0061] The innovation of this invention lies in its combination of Physical Information Neural Networks (PINNs) with a physical model of a ship's propeller. By leveraging the strong approximation capability of the neural network and the accuracy of the physical model, high-precision prediction of propeller efficiency is achieved. This combination not only fully utilizes the prior knowledge of the physical model but also improves the model's generalization ability through a data-driven approach.

[0062] Example 1, such as Figure 1 As shown, the ship propeller efficiency prediction method provided in this embodiment of the invention includes:

[0063] S1. Construct an autoencoder and pre-train it on the data.

[0064] The autoencoder built in this step uses propeller speed. propeller radius propeller speed propeller mass As core input parameters of the neural network, the above parameters are key physical quantities that determine the working characteristics of the propeller and directly affect the axial induced velocity of the propeller during actual operation. With circumferential induced velocity The magnitude and distribution of the induced velocity directly determine the thrust generated by the propeller. and torque This directly impacts the final operating efficiency of the propeller. The embodiment collects relevant design and operating parameters of common standard propeller models in the field of marine engineering, constructs a standardized training sample set, and pre-trains an autoencoder based on this sample set to achieve unsupervised extraction and characterization of key propeller parameters. It is necessary to clarify that the input parameters of the physical information neural network are time and the current ship speed. These parameters directly relate to the measurement of hydrostatic resistance and the increase in resistance, thus affecting the calculation of propeller thrust, main engine torque absorbed by the propeller, and other factors. To accurately predict the propeller efficiency at the ship's current speed, a large amount of raw data is required. These datasets are obtained through computational fluid dynamics methods combined with numerical simulation techniques.

[0065] S2, using the encoder part of the trained autoencoder as input, constructs a physical information neural network.

[0066] The output of the physical information neural network model is set as the propeller lift-to-drag ratio, circulation, circumferential induced velocity, and axial induced velocity. The construction of the loss function should take into account two constraints: first, the deviation between the predicted efficiency and the theoretical efficiency given a known propeller model; and second, the constraint of the energy conservation equation.

[0067] S3, Introduce a regularization term to determine the final loss function;

[0068] In step S3, during the construction of the loss function for the propeller efficiency prediction model, the prediction error function effectively measures the deviation between the propeller efficiency output by the model and the actual measured efficiency. The physical constraint function forces the model output to meet the fluid dynamics and mechanical physical laws of the propeller operation process. The synergistic effect of the two can better ensure the model's fitting accuracy and prediction accuracy. However, in this process, the model is prone to overfitting due to excessive fitting of training sample details, resulting in a decrease in its generalization ability and inability to adapt to the efficiency prediction requirements of different working conditions and different types of propellers. Therefore, this invention introduces a regularization term into the total loss function to limit the model complexity by imposing constraints on the model network parameters, avoiding the model from having an overly complex fitting situation. The final total loss function is composed of three parts: prediction error, physical information error, and regularization constraint, which takes into account the model's prediction accuracy, physical rationality, and generalization ability, ensuring that the model can stably and accurately complete the propeller efficiency prediction in practical applications.

[0069] For example, in this embodiment, the loss function of the physical constraint loss part is:

[0070] ;

[0071] in, Where is the propeller radius. The propeller hydrodynamic pitch angle, The lift-to-drag ratio of the propeller. This refers to the axial induced velocity during propeller operation. This refers to the circumferential induced velocity during propeller operation. The circulation of fluid around the propeller. The density of the fluid around the propeller. For propeller mass.

[0072] The loss function for the data forecasting part is:

[0073] ;

[0074] in, This represents the actual efficiency of the propeller. To improve the prediction efficiency of neural networks

[0075] The root cause of overfitting during network training is that the model, in its attempt to fit local details in the training samples, leads to excessively large values ​​for network parameters. This results in model complexity far exceeding the requirements for representing the core principles of propeller efficiency, thus losing the ability to generalize to new operating conditions and propeller models. Regularization, by imposing numerical constraints such as parameter sum-of-squares penalties and absolute value penalties on the model's network parameters in the total loss function, forces the parameters to converge towards smaller values, reducing redundant dependencies between parameters and suppressing overlearning of non-common features of the training samples. This embodiment uses parameter mean penalty to focus the model on the essential physical relationship between core parameters such as advance rate, radius, rotational speed, and mass and propeller efficiency. This effectively reduces model complexity while ensuring prediction accuracy and physical interpretability, fundamentally alleviating the overfitting problem and improving the model's adaptability to different application scenarios. The specific expression is:

[0076] ;

[0077] in, The predicted value for propeller efficiency is given by the neural network. The average propeller efficiency in the training set

[0078] This leads to the final loss function:

[0079] ;

[0080] in, The weights of the two loss terms and the regularization term are respectively.

[0081] S4, Training the Physical Information Neural Network Model using the Adam optimizer.

[0082] Using the Adam optimizer for training neural network models ensures both training efficiency and model accuracy. The Adam optimizer helps neural network models approach the global optimum in a shorter time and can implement functions such as gradient moving average and bias correction. This strategy ensures the stability and accuracy of the model.

[0083] For example, training a neural network using the Adam optimizer involves: Adam can simultaneously calculate the exponential moving averages of the first and second moments of the gradient and correct for biases, ensuring that the gradient estimates are not biased towards zero in the early stages of training. The Adam optimizer calculates the exponential moving averages of the first and second moments of the gradient separately, using a decay coefficient to smooth the historical gradient direction and fluctuation amplitude; since the initial moment estimates are biased towards zero, bias correction is applied to both moments to correct the initial error; subsequently, the corrected first moment provides momentum, and the second moment is used for adaptive learning rate scaling, combined with the base learning rate to stably update parameters, which not only accelerates the initial convergence speed and reduces oscillations, but also improves the optimization accuracy in the later stages, more robustly approaching the global optimum.

[0084] The specific workflow of the Adam optimizer is as follows:

[0085] 1. Initialize the optimizer by estimating the first and second moments as 0, and setting the learning rate, the decay rate of the first moment, the decay rate of the second moment, and a small constant to prevent the denominator from being zero.

[0086] 2. Calculate the gradient and update the estimates of the first and second moments;

[0087] 3. Bias Correction: Since the first and second moments are initialized to 0 in the early stages of training, their estimated values ​​will be biased towards 0. To correct this bias, bias correction is performed.

[0088] 4. Parameter updates are performed using bias-corrected first and second moment estimates. By adjusting the learning rate, the parameter update step size is adaptively adjusted based on historical gradient information, thus achieving more efficient parameter updates.

[0089] For example, the gradient estimation in the initial stage of a traditional optimizer may be too low. Bias correction can be used to accelerate the initial learning rate, thereby quickly approaching the global optimum.

[0090] S5 uses a trained neural network to predict propeller efficiency.

[0091] This study utilizes a trained physical information neural network model to make predictions on a new dataset, focusing primarily on the model's performance in predicting ship efficiency. By comparing the predictions with experimental data, the effectiveness and superiority of the physical information neural network in predicting propeller efficiency are further enhanced.

[0092] For example, using a trained physical information neural network model to predict a new dataset includes: taking the sailing time and current ship speed as input parameters, and using a partial differential physical information neural network to predict the propeller efficiency at the current speed. Specifically, the partial differential equation is removed and replaced with a physical information neural network. The calculation process is described as follows: different main engine efficiencies can be calculated, the network-predicted main engine efficiency is compared with a given main engine efficiency, and the neural network is trained by continuously reducing the error value until the error is less than the allowable range, at which point the training is successful.

[0093] S5, Iterative optimization and application of the trained physical information neural network model.

[0094] Based on the prediction and evaluation results, the trained physical information neural network model is iteratively optimized. Improvements are mainly made by adjusting the network structure, optimizing the loss function, and selecting the optimal algorithm, thereby enhancing the model's prediction accuracy and generalization ability. Finally, the trained model is applied to the rapid prediction of ship propeller x-efficiency.

[0095] For example, by introducing an attention mechanism, training costs can be saved, thereby making the neural network's predictions more efficient.

[0096] Example 2, as Figure 2 As shown, the ship propeller efficiency prediction system provided in this embodiment of the invention includes:

[0097] Data preprocessing module 1 is used to initialize the network and perform data preprocessing;

[0098] Loss function construction module 2 is used to construct the loss function of the physical information neural network using the energy conservation physical equation related to propeller efficiency;

[0099] Model training module 3 is used to train physical information neural network models using the Adam optimizer;

[0100] Propeller efficiency prediction module 4 is used to predict propeller efficiency using a trained physical information neural network model.

[0101] Iterative optimization module 5 is used for iterative optimization and application of the trained physical information neural network model.

[0102] To further illustrate the effects of the embodiments of the present invention, the following experiments were conducted.

[0103] Network framework and experimental setup:

[0104] The network framework of the PINNs method of this invention is detailed in the appendix. Figure 3This framework details the key information required by the network, such as the input variables, network structure, loss function construction, optimizer selection, and iteration cutoff conditions, providing clear technical features for understanding and implementing the PINN method.

[0105] Through the PINN method described above and its experimental verification, this invention provides an effective solution for real-time forecasting of ship efficiency, combining high-precision prediction with a strong physical foundation, and providing a new technical path and research basis for related research fields.

[0106] Step 1: To accurately test the propeller's hydrodynamic performance and open-water performance, an independent simulation of the propeller is first performed, requiring the establishment of a propeller computational domain model. A Multiple Reference Frame (MRF) model is used to build the model. During modeling, the computational domain is divided into two cylinders, with the outer computational domain being the stationary domain. A velocity inlet is set on the surface of the stationary domain closest to the propeller, and a pressure outlet is set on the surface farther from the propeller. Other surfaces of the stationary domain are set as symmetrical planes, and the propeller model surface is set as a wall. To enable data exchange between the stationary and rotating domains, the interface between the two computational domains is set as an interface. The diameter of the stationary domain is 4. Length is 18 Speed ​​input data propeller 5 The diameter of the rotation domain is 1.2. The length is 0.8 To improve accuracy, the mesh around the propeller was refined. To meet practical requirements, an open-water propeller simulation was conducted. The volume force propeller method was adopted without incurring the additional costs of using a fine mesh and a smaller time scale for propeller geometry modeling. The main models include: the fluid domain volume (VOF) of water and air, and dynamic fluid-solid interaction (DFBI), used to calculate the ship's motion under fluid influence, simulating forward and backward movement, free undulation, and turbulence. Table 1 shows the propeller-related geometric parameters used in this invention.

[0107] Table 1 Propeller Parameters

[0108]

[0109] Step 2: Using the data collected in Step 1, the autoencoder was trained and tested. The model was driven at propeller speed. propeller radius propeller speed propeller mass As input, the desired training effect is that after encoding by the encoder and decoding by the decoder, the input parameters can be reproduced, ensuring that the encoder can extract key feature parameters.

[0110] Step 3, the specific steps for constructing the neural network loss function module are as follows:

[0111] Once the specific model and current advance speed of the ship's propeller are determined, the propeller efficiency is then determined. The prediction results of the physical information neural network are compared with the propeller efficiency to obtain the loss function:

[0112] ;

[0113] ;

[0114] ;

[0115] The total loss function combines the two losses mentioned above, where, These are the weighting coefficients for the two loss terms. When predicting propeller efficiency, these two weighting parameters are uniformly set to 1 to balance the model's data fitting ability and adherence to physical laws.

[0116]

[0117] In step 3, the Adam optimizer is applied to train the network. If the loss value does not decrease within 10 iterations, the learning rate is adjusted by increasing or decreasing it by 20%. The entire training process comprises 200,000 iterations. At the beginning of training, the loss decreases slowly, but rapidly decreases during training intervals of 30,000 to 80,000 steps. Finally, the trained network is used to predict the current efficiency of the ship's propeller, with a prediction error of approximately 12%.

[0118] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention and within the spirit and principles of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting ship propeller efficiency, characterized in that, The method includes the following steps: S1, construct an autoencoder and pre-train the data; S2, using the encoder part of the trained autoencoder as input, constructs a physical information neural network; S3, Introduce a regularization term to determine the final loss function; S4, using the Adam optimizer to train the physical information neural network model; S5 uses a trained neural network to predict propeller efficiency. S6, Model Iterative Optimization and Application; In step S2, the output of the physical information neural network model is the propeller lift-to-drag ratio, circulation, circumferential induced velocity, and axial induced velocity. The loss function of the physical information neural network includes constraints on energy conservation, constraints on the real-time propeller propulsion efficiency output parameters, and regularization terms, given a known and determined ship type. In step S2, the method for constructing the loss function of the physical information neural network specifically includes: For cases where the ship's propulsion efficiency is determined, the loss between the predicted efficiency and the determined propulsion efficiency of the physical information neural network is used as a function to impose physical constraints on the network, constructing the following loss function: ； In the formula, This is a predicted value for the current propeller efficiency. The difference between the prediction efficiency and the deterministic efficiency of a neural network for physical information is represented by a regularization constraint. These are the weights of the two loss terms and the regularization term, respectively. This is the total loss function; The gap between the prediction efficiency and deterministic efficiency of physical information neural networks The expression is: ； In the formula, Where is the propeller radius. is the propeller hydrodynamic pitch angle, and is the propeller lift-to-drag ratio. This refers to the axial induced velocity during propeller operation. Let be the circumferential induced velocity during propeller operation, be the fluid circulation around the propeller, and be the fluid density around the propeller. For propeller mass; Current propeller efficiency prediction The expression is: ； In the formula, This represents the actual efficiency of the propeller. To improve the prediction efficiency of neural networks; Regularization constraints The expression is: ； In the formula, The predicted value for propeller efficiency is given by the neural network. The average propeller efficiency in the training set.

2. The method for predicting ship propeller efficiency according to claim 1, characterized in that, In steps S1 and S2, when extracting key parameters from the data, the pre-trained autoencoder is used as the input to the subsequent network.

3. The method for predicting ship propeller efficiency according to claim 1, characterized in that, In step S4, training the physical information neural network model using the Adam optimizer includes: Adam simultaneously calculates the exponential moving average of the first and second moments of the gradient, and performs bias correction to accelerate the initial learning rate and approach the global optimum.

4. The method for predicting ship propeller efficiency according to claim 1, characterized in that, In step S5, the trained physical information neural network model is used to predict the propeller circulation. Propeller lift-to-drag ratio, axial induced velocity Circumferential induced velocity The propeller propulsion efficiency, further calculated based on the predicted values, includes: The propeller's advance rate propeller radius, propeller speed, propeller mass As input parameters, key features are extracted using an autoencoder, and then the axial velocity of the ship's propeller at the current speed is predicted using a physical information neural network. Circumferential speed Rise-to-drag ratio Circulation; the expression is: ； In the formula, Where is the propeller radius. The propeller hydrodynamic pitch angle; The propeller efficiency is calculated and expressed as follows: ； ； ； ； In the formula, is the propeller thrust, and is the propeller torque. For propeller advance speed, is the propeller angular velocity, and is the propeller efficiency.

5. A ship propeller efficiency prediction system, characterized in that, The system implements the ship propeller efficiency prediction method as described in any one of claims 1-4, and the system includes: The data preprocessing module (1) constructs an autoencoder and pre-trains the data; The loss function construction module (2) takes the encoder part of the trained autoencoder as input to construct a physical information neural network; introduces a regularization term to determine the final loss function; The model training module (3) is used to train the physical information neural network model using the Adam optimizer; Efficiency prediction module (4) is used to predict propeller efficiency using a trained neural network; The iterative optimization module (5) is used for model iterative optimization and application.

6. The ship propeller efficiency prediction system according to claim 5, characterized in that, The aforementioned ship propeller efficiency forecasting system is mounted on a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can realize the functions of the aforementioned ship propeller efficiency forecasting system.

7. The ship propeller efficiency prediction system according to claim 5, characterized in that, The aforementioned ship propeller efficiency forecasting system is mounted on an information data processing terminal, which provides a user input interface to implement the functions of the ship propeller efficiency forecasting system as described above when executed on an electronic device.

Images