Dimension reduction training method for ship propeller open water performance intelligent forecasting model
By standardizing and dimensionality reduction processing of the geometric shape and open water performance data of ship propellers, a suitable data set is built for intelligent forecasting model training, which solves the problem of excessive loss of feature information and lack of obvious dimensionality reduction effects in the existing technology, and achieves efficient open water performance forecasting.
Patent Information
- Application Number
- CN202510201751.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-24
- Publication Date
- 2025-05-13
AI Technical Summary
The existing technology has problems such as excessive loss of original feature information or lack of dimensionality reduction in the intelligent forecast model for open water performance of ship propellers, resulting in high training difficulty and low forecast accuracy.
By normalizing the geometric shape data and open water performance data of the propeller sample, simplified geometric shape data is expressed using B-spline curves, quadratic polynomial fit simplified open water performance data, building the propeller thrust sample data set and torque sample data set, and intelligent forecast model training is carried out.
It realizes efficient dimensionality reduction of propeller feature parameters, improves the training efficiency and calculation accuracy of the intelligent forecast model of open water performance, reduces the training data demand, and shortens the model training time.
Smart Images
Figure CN119988980A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of open water performance test of ship propellers, and in particular to a dimensionality reduction training method for an intelligent prediction model of open water performance of ship propellers. Background Art
[0002] The propeller has a simple structure, is easy to use and has high efficiency. It is currently the most widely used ship propulsion device. Open water performance characterizes the comprehensive hydrodynamic performance of the propeller under any working conditions. It is the first to be determined and paid attention to in propeller design, and directly affects the matching performance of the ship-engine-propeller and the propulsion performance of the ship. Designers often use small-scale model tests, numerical simulation calculations and other methods to predict the open water performance of ship propellers, but the manpower, material resources and time costs are relatively large. With the continuous development of artificial intelligence technology, the intelligent prediction model of ship propeller open water performance based on machine learning algorithms has gradually been applied, which can greatly shorten the prediction time of ship propeller open water performance and improve the overall design efficiency.
[0003] The intelligent prediction model for open water performance of ship propellers is essentially a kind of proxy model. It is driven by data and requires a large amount of data as support during the training process. However, the open water performance data of ship propellers is still difficult to reach this level. In order to reduce the demand for data in the process of establishing the intelligent prediction model, it is necessary to extract features and reduce the dimension of ship propeller data. Currently, the commonly used dimensionality reduction methods include feature decomposition, popular learning and neural networks, but in the training of the intelligent prediction model for open water performance of ship propellers, there are problems such as excessive loss of original feature information or unclear dimensionality reduction effect. Therefore, it is urgent to establish a dimensionality reduction method suitable for the training of the intelligent prediction model for open water performance of ship propellers, so as to reduce the difficulty of training the intelligent prediction model for open water performance of ship propellers, support the construction of a high-precision intelligent prediction model for open water performance of ship propellers, and meet the engineering design application requirements of rapid prediction and efficient iteration of ship propellers. Summary of the invention
[0004] In view of the above problems and technical requirements, the inventors have proposed a dimensionality reduction training method for a ship propeller open water performance intelligent prediction model, which can achieve efficient dimensionality reduction of propeller characteristic parameters, further improve the training efficiency of the open water performance intelligent prediction model, and significantly improve the calculation accuracy of the prediction model while reducing the demand for training data. The technical solution of the present invention is as follows:
[0005] A dimensionality reduction training method for an intelligent prediction model of ship propeller open water performance comprises the following steps:
[0006] Normalize the geometry data and open water performance data of each propeller sample into a standard format;
[0007] The normalized geometric shape data is processed for dimensionality reduction to obtain a small number of radial and chord-wise control points for each propeller sample;
[0008] The normalized open water performance data is subjected to dimensionality reduction processing to obtain the coefficients of the thrust coefficient curve and torque coefficient curve of each propeller sample;
[0009] Construct propeller thrust sample data sets and propeller torque sample data sets, where each set of sample data sets takes a small number of radial and chord-wise control points, as well as the number of propeller blades and the hub-diameter ratio as inputs, and the two sets of sample data sets take the coefficients of the propeller thrust coefficient curve and torque coefficient curve as corresponding outputs respectively;
[0010] The propeller thrust sample data set and the propeller torque sample data set are used to train the intelligent prediction models of propeller thrust coefficient and propeller torque coefficient respectively.
[0011] A further technical solution is to normalize the geometric shape data and open water performance data of each propeller sample in a standard format, including:
[0012] The geometric shape data of each propeller sample is expressed by the dimensionless principal parameters corresponding to the propeller radial position and the dimensionless section parameters corresponding to the propeller chordwise position;
[0013] The dimensionless main parameters include the profile radial pitch ratio P / D, the profile radial maximum thickness ratio tmax / D, the radial profile chord length ratio C / D, the profile radial side skew angle skew, the profile radial longitudinal inclination Zr / D and the profile radial maximum camber ratio fmax / C, and they are dimensionless.
[0014] The dimensionless profile parameters include the profile chord-wise thickness distribution t / tmax and the profile chord-wise camber distribution f / fmax, and they are dimensionless.
[0015] The open water performance data of each propeller sample is expressed by an open water characteristic curve, including the propeller thrust coefficient K T , Torque coefficient K Q and the distribution curve of open water efficiency η as the advance rate coefficient J changes;
[0016] Among them, P is the propeller profile pitch, D is the propeller diameter, tmax is the maximum profile thickness, C is the profile chord length, Zr is the profile longitudinal inclination, fmax is the maximum profile arch, t is the profile chord thickness corresponding to the current radial position, and f is the profile chord arch corresponding to the current radial position.
[0017] A further technical solution is to perform dimensionality reduction processing on the normalized geometric shape data, including:
[0018] Select m data corresponding to each parameter type in the dimensionless main parameters of each propeller sample as radial control points, and select m data corresponding to each parameter type in the dimensionless profile parameters of each propeller sample as chord-wise control points;
[0019] Among them, the selected control points should satisfy the smooth geometric parameter distribution B-spline curve along the radial and chord directions obtained through the control points, and basically coincide with the prototype geometric parameter curve.
[0020] A further technical solution is that, for each parameter type, the selected control points at least include data corresponding to the first and last radial positions or the first and last chord positions.
[0021] A further technical solution is that the number of control points is 5≤m≤m0;
[0022] Where m0 is the number of propeller radial stations, and m0≥11; n0 is the number of propeller chord-wise stations, and n0≥16.
[0023] A further technical solution is to perform dimensionality reduction processing on the normalized open water performance data, including:
[0024] Taking the advance coefficient J as the variable, high-order polynomials are used to fit the thrust coefficient curve and torque coefficient curve of each propeller sample, and the coefficient of each term in the high-order polynomial is used as the coefficient of the thrust coefficient curve and the torque coefficient curve.
[0025] A further technical solution is to use a second-order polynomial to fit the thrust coefficient curve K of each propeller sample. T =a0×J 2 +b0×J+c0 and torque coefficient curve K Q =a1×J 2 +b1×J+c1, thus changing the original training output target from a series of thrust coefficients K within a certain speed coefficient range T , Torque coefficient K Q The six quadratic polynomial coefficients converted into thrust coefficient curve and torque coefficient curve are a0, b0, c0 and a1, b1, c1 respectively.
[0026] A further technical solution is to use a propeller thrust sample data set to train a propeller thrust coefficient intelligent prediction model, which is the same as the method of using a propeller torque sample data set to train a propeller torque coefficient intelligent prediction model, wherein the training process of the propeller thrust coefficient intelligent prediction model includes:
[0027] The propeller thrust sample data set is divided into a training set and a test set according to a certain ratio;
[0028] Based on the training set, different machine learning algorithms were selected to train and test the propeller thrust coefficient intelligent prediction model. The model hyperparameters were optimized through cross-validation and grid search methods. The thrust coefficient intelligent prediction model with the best prediction effect was selected based on the mean square error (MSE) and saved in a file in the specified format.
[0029] Its further technical solution is that the method further comprises:
[0030] Based on the trained intelligent prediction model of propeller optimal thrust coefficient and optimal torque coefficient, a complete open water characteristic curve including propeller open water efficiency is obtained.
[0031] A further technical solution is to obtain a complete open water characteristic curve including the propeller open water efficiency based on the trained propeller optimal thrust coefficient and optimal torque coefficient intelligent prediction model, including:
[0032] For a specific ship propeller, the data corresponding to the radial and chord control points are selected from the data corresponding to each parameter type in its geometric shape data, and are input into the optimal thrust coefficient and optimal torque coefficient intelligent prediction model respectively to calculate the thrust coefficient K T and torque coefficient K Q Curve fitting formula that changes with the advance speed coefficient J;
[0033] Substituting the curve fitting formula into the following formula, the open water efficiency η curve fitting formula is obtained, which together with the thrust coefficient and torque coefficient curves constitute the complete open water characteristic curve of the ship propeller;
[0034]
[0035] The beneficial technical effects of the present invention are:
[0036] This method simplifies the large-scale geometric shape data of the propeller into a small number of radial and chord control points based on B-spline curve expression, and uses quadratic polynomials to simplify a large range of open water performance data into six quadratic polynomial coefficients of the thrust coefficient curve and the torque coefficient curve, achieving parameter dimension reduction of more than 50% during the training process of the propeller open water performance intelligent prediction model, which can greatly reduce the data demand for model training, shorten the training time of the open water performance intelligent prediction model, and further improve the calculation accuracy of the open water performance intelligent prediction model. Overall, this method has a clear process and simple operation, which helps to support the construction of a high-precision ship propeller open water performance intelligent prediction model, and meet the engineering design application requirements of rapid prediction and efficient iteration of ship propellers. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1It is a schematic diagram of the normalized processing of propeller training data provided in the present application, wherein (1) is a schematic diagram of the normalized processing of geometric shape data, and (2) is a schematic diagram of the normalized processing of open water performance data.
[0038] Figure 2 Schematic diagram of B-spline dimensionality reduction expression of typical geometric characteristic parameters of propellers provided in the present application, wherein (1) is a schematic diagram of B-spline dimensionality reduction expression of pitch ratio, (2) is a schematic diagram of B-spline dimensionality reduction expression of skew angle, (3) is a schematic diagram of B-spline dimensionality reduction expression of blade section thickness distribution, and (4) is a schematic diagram of B-spline dimensionality reduction expression of blade section camber distribution.
[0039] Figure 3 It is a schematic diagram of the quadratic polynomial dimensionality reduction expression of propeller open water performance data provided in this application.
[0040] Figure 4 It is a schematic diagram of a propeller thrust sample data set and a propeller torque sample data set provided in this application.
[0041] Figure 5 This is a flow chart of the dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance provided by this application.
[0042] Figure 6 This is a schematic diagram comparing the parameter dimensions before and after dimensionality reduction and the accuracy of the intelligent prediction model for propeller open water performance provided by this application. DETAILED DESCRIPTION
[0043] The specific implementation of the present invention will be further described below in conjunction with the accompanying drawings.
[0044] An embodiment of the present application provides a dimensionality reduction training method for an intelligent prediction model of ship propeller open water performance. The overall training method refers to Figure 5 As shown, the specific steps include:
[0045] Step 1: Normalize the propeller training data. Figure 1 As shown in Figure 2, the geometric shape data and open water performance data of each propeller sample are normalized in a standard format.
[0046] Specifically, the geometric shape data of each propeller sample is expressed by the dimensionless main parameters corresponding to the propeller radial position r / R and the dimensionless section parameters corresponding to the propeller chord position x / C. Among them, r is the radial position radius, R is the propeller radius, x is the chord position, and C is the section chord length. Let the number of radial positions and the number of chord positions be m0 and n0 respectively. In order to accurately express the blade geometry, it is recommended that m0 be ≥11 and n0 be ≥16. The dimensionless main parameters include the section radial pitch ratio P / D, the section radial maximum thickness ratio tmax / D, the radial section chord length ratio C / D, the section radial side skew angle skew (°), the section radial pitch Zr / D and the section radial maximum camber ratio fmax / C, and they are dimensionless. The dimensionless section parameters include the section chord thickness distribution t / tmax and the section chord camber distribution f / fmax, and they are dimensionless. Among them, P is the pitch of the propeller profile, D is the propeller diameter, tmax is the maximum thickness of the profile, Zr is the profile trim, fmax is the maximum camber of the profile, t is the chord-wise thickness of the profile corresponding to the current radial position, and f is the chord-wise camber of the profile corresponding to the current radial position. Considering the number of propeller blades Z and the hub-diameter ratio Hr, assuming that the profiles of each radial position of the propeller are consistent, the geometric characteristic parameters of the propeller are 11*6+16*2+2=100, Figure 1 -(1) The gray background area is the propeller geometric characteristic parameters.
[0047] The open water performance data of each propeller sample is expressed by an open water characteristic curve, including the propeller thrust coefficient K T , Torque coefficient K Q And the distribution curve of open water efficiency η with the change of advance coefficient J. In this embodiment, the advance coefficient interval is usually ΔJ=0.05, and the maximum advance coefficient range of the propeller is 1.0, then the thrust coefficient K T , Torque coefficient K Q The data parameter dimension is 1.0 / 0.05*2=40.
[0048] Step 2: Perform dimensionality reduction on the normalized geometric data to obtain a small number of radial and chord-wise control points for each propeller sample.
[0049] Specifically, from the data corresponding to each parameter type in the dimensionless main parameters of each propeller sample (i.e., P / D, tmax / D, C / D, skew, Zr / D, fmax / C series data), m are selected as radial control points P0, P1, ..., P4, and from the data corresponding to each parameter type in the dimensionless profile parameters of each propeller sample (i.e., t / tmax, f / fmax series data), m are also selected as chord control points P0, P1, ..., P4. The selected control points should satisfy the smooth geometric parameter distribution B-spline curve along the radial direction and the chord direction obtained by fewer control points, and basically coincide with the prototype geometric parameter curve. It is recommended that the number of control points is 5≤m≤m0. In this embodiment, m=5 is preferred, and the data corresponding to the first and last radial positions or the first and last chord positions and the middle three-point positions of the original parameter distribution curve are selected as control points. Then, the B-spline curve fitted based on the control points can re-express the dimensionless main parameters and profile parameters of each propeller.
[0050] Among them, the p-order B-spline curve is defined as: Where N i,p (u) is the i-th p-order basis function, and the basis function expression is:
[0051]
[0052]
[0053] Figure 2 -(1)~(4) give schematic diagrams of the B-spline dimensionality reduction expression of typical geometric feature parameters. In the case of 5 control points, the geometric parameter distribution B-spline curve basically coincides with the prototype geometric parameter curve, and the two are in good agreement.
[0054] Step 3: Perform dimensionality reduction processing on the normalized open water performance data to obtain the coefficients of the thrust coefficient curve and torque coefficient curve of each propeller sample.
[0055] Specifically, taking the advance coefficient J as the variable, a high-order polynomial is used to fit the thrust coefficient curve and torque coefficient curve of each propeller sample, and the coefficient of each term in the high-order polynomial is used as the coefficient of the thrust coefficient curve and the torque coefficient curve. The open water efficiency η is the thrust coefficient K T and torque coefficient K Q The derived quantity of η is thus unnecessary to process. The derived formula of η is:
[0056]
[0057] In this embodiment, a second-order polynomial is used to fit the thrust coefficient curve K of each propeller sample. T =a0×J 2+b0×J+c0 and torque coefficient curve K Q =a1×J 2 +b1×J+c1, such as Figure 3 As shown in the figure, the original training output target is changed from a series of thrust coefficients K within a certain speed coefficient range. T , Torque coefficient K Q The six quadratic polynomial coefficients converted into thrust coefficient curve and torque coefficient curve are a0, b0, c0 and a1, b1, c1 respectively.
[0058] Step 4: Construct a propeller thrust sample data set and a propeller torque sample data set containing x samples, where each set of sample data sets uses the radial and chord control points P0, P1, ..., P4 corresponding to each parameter type selected in step 2, as well as the number of propeller blades Z and the hub diameter ratio Hr as inputs, and the two sets of sample data sets use the coefficients a0, b0, c0 of the propeller thrust coefficient curve and the coefficients a1, b1, c1 of the torque coefficient curve as corresponding outputs, respectively. The input feature parameter dimension is 42, and the output parameter dimension of the two sets of data sets is 3, such as Figure 4 The recommended data set sample number x in this embodiment is ≥ 300, and one sample includes geometric shape data and open water performance data of one propeller.
[0059] Step 5: Use the propeller thrust sample data set and the propeller torque sample data set to train the intelligent prediction models of the propeller thrust coefficient and the propeller torque coefficient respectively.
[0060] like Figure 5 As shown, first, the propeller thrust sample data set constructed in step 4 is divided into a training set and a test set according to a certain ratio. Then, based on the training set, nonlinear machine learning regression algorithms such as random forest, feedforward neural network, and support vector machine are selected to train and test the propeller thrust coefficient intelligent prediction model, and the model hyperparameters are optimized through cross-validation and grid search methods. The thrust coefficient intelligent prediction model with the best prediction effect is selected based on the mean square error MSE as the criterion and saved as a .pickle format file. Training method of propeller torque coefficient intelligent prediction model and thrust coefficient K T The intelligent prediction models are the same, except that the output target quantities of the model training of the propeller torque sample data set are changed from a0, b0, c0 to a1, b1, c1.
[0061] Step 6: Training and application of the open water performance intelligent prediction model. Based on the trained propeller optimal thrust coefficient and optimal torque coefficient intelligent prediction model, a complete open water characteristic curve including the propeller open water efficiency is obtained.
[0062] Specifically, load the optimal thrust coefficient K T Intelligent prediction model and torque coefficient KQ The .pickle file of the intelligent prediction model selects the data corresponding to the radial and chord control points from the data corresponding to each parameter type in the geometric shape data for a specific ship propeller, and inputs them into the optimal thrust coefficient K T and the optimum torque coefficient K Q Intelligent prediction model, calculate the thrust coefficient K T and torque coefficient K Q The curve fitting formula that changes with the advance coefficient J. Substituting the curve fitting formula into formula (1) to calculate the open water efficiency η curve fitting formula, together with the thrust coefficient and torque coefficient curves, constitutes the complete open water characteristic curve of the ship propeller.
[0063] Figure 6 The parameter dimension before and after dimensionality reduction and the accuracy of the intelligent prediction model for propeller open water performance are compared. The propeller sample data set of a surface ship was verified. After using the above dimensionality reduction training method, the parameter dimension decreased by more than 50%, and the prediction accuracy of the obtained model was improved by more than 30%, which fully verified the effectiveness of the above dimensionality reduction training method.
[0064] The above is only a preferred embodiment of the present application, and the present invention is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present invention should be considered to be included in the protection scope of the present invention.
Claims
1. A dimensionality reduction training method for an intelligent prediction model of ship propeller open water performance, characterized in that: The method comprises: Normalize the geometry data and open water performance data of each propeller sample into a standard format; The normalized geometric shape data is processed for dimensionality reduction to obtain a small number of radial and chord-wise control points for each propeller sample; The normalized open water performance data is subjected to dimensionality reduction processing to obtain the coefficients of the thrust coefficient curve and torque coefficient curve of each propeller sample; Construct propeller thrust sample data sets and propeller torque sample data sets, where each set of sample data sets takes a small number of radial and chord-wise control points, as well as the number of propeller blades and the hub-diameter ratio as inputs, and the two sets of sample data sets take the coefficients of the propeller thrust coefficient curve and torque coefficient curve as corresponding outputs respectively; The propeller thrust sample data set and the propeller torque sample data set are used to respectively train the intelligent prediction models of the propeller thrust coefficient and the propeller torque coefficient.
2. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to claim 1 is characterized in that: The geometric shape data and open water performance data of each propeller sample are normalized in a standard format, including: The geometric shape data of each propeller sample is expressed by the dimensionless principal parameters corresponding to the propeller radial position and the dimensionless section parameters corresponding to the propeller chordwise position; The dimensionless main parameters include the profile radial pitch ratio P / D, the profile radial maximum thickness ratio tmax / D, the radial profile chord length ratio C / D, the profile radial side skew angle skew, the profile radial longitudinal inclination Zr / D and the profile radial maximum camber ratio fmax / C, and are dimensionally processed; The dimensionless profile parameters include the profile chord-wise thickness distribution t / tmax and the profile chord-wise camber distribution f / fmax, and are dimensionally processed; The open water performance data of each propeller sample is expressed by an open water characteristic curve, including the propeller thrust coefficient K T , Torque coefficient K Q and the distribution curve of open water efficiency η as the advance rate coefficient J changes; Among them, P is the propeller profile pitch, D is the propeller diameter, tmax is the maximum profile thickness, C is the profile chord length, Zr is the profile longitudinal inclination, fmax is the maximum profile arch, t is the profile chord thickness corresponding to the current radial position, and f is the profile chord arch corresponding to the current radial position.
3. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to claim 2 is characterized in that: The step of performing dimensionality reduction processing on the normalized geometric shape data comprises: Select m data corresponding to each parameter type in the dimensionless main parameters of each propeller sample as radial control points, and select m data corresponding to each parameter type in the dimensionless profile parameters of each propeller sample as chord-wise control points; The selected control points should satisfy the smoothness of the geometric parameter distribution B-spline curve along the radial direction and the chord direction obtained through the control points, and basically coincide with the prototype geometric parameter curve.
4. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to claim 3 is characterized in that: For each parameter type, the selected control points at least include data corresponding to the first and last radial stations or the first and last chordal stations.
5. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to claim 3 is characterized in that: The number of control points is 5≤m≤m0; Where m0 is the number of propeller radial stations, and m0≥11; n0 is the number of propeller chord-wise stations, and n0≥16.
6. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to claim 2 is characterized in that: The dimension reduction processing of the normalized open water performance data includes: The advance coefficient J is used as a variable, and a high-order polynomial is used to fit the thrust coefficient curve and the torque coefficient curve of each propeller sample respectively, and the coefficient of each item in the high-order polynomial is used as the coefficient of the thrust coefficient curve and the torque coefficient curve.
7. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to claim 6 is characterized in that: The thrust coefficient curve K of each propeller sample is fitted using a second-order polynomial T =a0×J 2 +b0×J+c0 and torque coefficient curve K Q =a1×J 2 +b1×J+c1, thus changing the original training output target from a series of thrust coefficients K within a certain speed coefficient range T , Torque coefficient K Q The six quadratic polynomial coefficients converted into thrust coefficient curve and torque coefficient curve are a0, b0, c0 and a1, b1, c1 respectively.
8. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to claim 1 is characterized in that: The method of using the propeller thrust sample data set to train the propeller thrust coefficient intelligent prediction model is the same as the method of using the propeller torque sample data set to train the propeller torque coefficient intelligent prediction model, wherein the training process of the propeller thrust coefficient intelligent prediction model includes: Dividing the propeller thrust sample data set into a training set and a test set according to a certain ratio; Based on the training set, different machine learning algorithms are selected to train and test the propeller thrust coefficient intelligent prediction model, and the model hyperparameters are optimized through cross-validation and grid search methods. The thrust coefficient intelligent prediction model with the best prediction effect is selected based on the mean square error (MSE) and saved as a file in a specified format.
9. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to any one of claims 1 to 8, characterized in that: The method further comprises: Based on the trained intelligent prediction model of propeller optimal thrust coefficient and optimal torque coefficient, a complete open water characteristic curve including propeller open water efficiency is obtained.
10. The dimensionality reduction training method for the intelligent prediction model of ship propeller open water performance according to claim 9 is characterized in that: The intelligent prediction model of the propeller optimal thrust coefficient and optimal torque coefficient obtained based on the training obtains a complete open water characteristic curve including the propeller open water efficiency, including: For a specific ship propeller, the data corresponding to the radial and chord control points are selected from the data corresponding to each parameter type in the geometric shape data, and are respectively input into the optimal thrust coefficient and optimal torque coefficient intelligent prediction model to calculate the thrust coefficient K T and torque coefficient K Q Curve fitting formula that changes with the advance speed coefficient J; Substituting the curve fitting formula into the following formula, the open water efficiency η curve fitting formula is obtained, which together with the thrust coefficient and torque coefficient curves constitutes a complete open water characteristic curve of the ship propeller;