Floating fan air gap response statistical property forecasting method based on neural network

Abstract

The invention provides a floating fan air gap response statistical characteristic forecasting method based on a neural network. The floating fan air gap response statistical characteristic forecasting method comprises the steps of generating full-working-condition wave time history data; dangerous wave segments are identified, and CFD local simulation is carried out; preprocessing CFD data to construct a training data set; training an air gap response time sequence forecast agent model; forecasting an air gap response time history value; all minimum values in the forecast values are taken out, and data screening is carried out; a proper distribution function is selected for fitting; and analyzing air gap response key statistical characteristics. According to the method, quick forecasting of the air gap response of the floating fan is achieved, an intelligent forecasting scheme suitable for the design of the floating fan is provided, and key statistical parameters concerned by engineering can be directly output.

Description

Technical Field

[0001] This invention relates to the field of marine floating wind turbine technology, and more specifically, to a method for predicting the statistical characteristics of air gap response of floating wind turbines based on neural networks. Background Technology

[0002] Floating offshore wind power, as a core technology for developing deep-sea wind energy resources, has entered a new stage of development. When a floating wind turbine platform operates in a real marine environment, the minimum distance between its deck and waves (called the air gap) directly affects the safety of the platform structure. However, accurately predicting this distance is extremely difficult, mainly due to two problems: First, while the potential flow method commonly used in engineering is relatively fast, it cannot assess nonlinear effects such as wave breaking and wave intrusion in extreme sea states, leading to a systematic underestimation of the predicted value under extreme sea states, resulting in safety hazards. Second, while CFD-based viscous flow simulation can simulate nonlinear processes well, it requires extremely long computation time and computational resources. Completing a single 300-second simulation requires more than 200 hours of computation time, while actual engineering designs require the assessment of multiple sea state combinations, making this speed simply unacceptable.

[0003] With the increasing size of floating wind turbine units, damage to floating foundations will lead to significant economic losses. Conservative design and excessive structural reinforcement significantly increase power costs, posing a huge challenge to the accurate control of air gap safety. How to achieve rapid assessment of the statistical characteristics of air gap extreme response while ensuring forecast accuracy and engineering efficiency has become an important problem for the safe deployment of floating wind turbines. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method for predicting the statistical characteristics of air gap response of floating wind turbines based on neural networks, thereby solving the problem that traditional CFD time-domain simulation methods cannot quickly obtain the statistical parameters of engineering interest due to long computation time and high resource consumption.

[0005] To solve the above problems, the technical solution of the present invention is as follows:

[0006] A method for predicting the statistical characteristics of air gap response of floating wind turbines based on neural networks includes the following steps:

[0007] Generate wave time history data for all operating conditions;

[0008] Identify dangerous wave segments and perform local CFD simulation;

[0009] Preprocess CFD data to build the training dataset;

[0010] Training a surrogate model for air gap response time-series prediction;

[0011] Predict the air gap response time history;

[0012] Extract all the minimum values ​​from the forecast values ​​and perform data filtering;

[0013] Choose a suitable distribution function for fitting;

[0014] Analyze the key statistical characteristics of the air gap response.

[0015] Preferably, the step of generating full-condition wave time history data specifically includes: using the potential flow method to complete a 3-hour irregular wave simulation, inputting JONSWAP spectral parameters, and outputting the full-time domain wave height time history curve.

[0016] Prior to this, the step of identifying dangerous wave segments and performing CFD local simulation specifically includes: extracting the continuous dangerous wave input to the viscous flow solver, solving to obtain the motion response and free surface elevation time history of the floating wind turbine platform, and obtaining the air gap response of the floating wind turbine after post-processing.

[0017] Preferably, the air gap response of the floating wind turbine is determined by the following formula:

[0018]

[0019] in: It is the air gap value at any spatial point near the wind turbine foundation platform, based on the highest point of the floating structure column. Location coordinates, time history, and wavefront height at that location The air gap time history at any point near the wind turbine foundation platform can be calculated. .

[0020] Preferably, the step of constructing a training dataset from preprocessed CFD data specifically includes: using wave height history data from time T-20 to T-1 in the CFD data as a set of input features for the air gap response at time T.

[0021] Prior to this, the step of training the air gap response time series prediction proxy model specifically includes: configuring an input layer, a hidden layer and an output layer based on a multilayer perceptron neural network, selecting an appropriate loss function and optimizer for training, and establishing a nonlinear mapping proxy model from input features to the air gap response time series.

[0022] Preferred, the step of predicting the air gap response time history specifically includes: selecting wave history data after the largest wave in the potential flow wave, and using the wave height history data from time T-20 to time T-1 as the input feature for predicting the air gap response at time T.

[0023] Prior to this, the step of extracting all local minima from the forecast values ​​and performing data filtering specifically includes: extracting local minima from all forecast time-history points, filtering effective extreme value samples at wave cycles, eliminating instantaneous noise interference, and generating an air gap minimum value sequence dataset; the step of extracting the minimum values ​​is as follows: calculating the arithmetic mean of the air gap response, subtracting the time-history data to obtain a zero-mean sequence, identifying each upper and lower zero-crossing point in the sequence, locating the minimum value point between adjacent upper and lower zero-crossing points, and the original air gap value at this point is the effective minimum value sample.

[0024] Prior to this, the step of selecting a suitable distribution function for fitting specifically includes: fitting the selected air gap minimum value sequence with the Weibull distribution function, calculating the shape parameter k and scale parameter λ using the least squares method, and establishing a cumulative distribution function model.

[0025] Preferably, the Weibull distribution function is:

[0026]

[0027] Where k is the shape parameter, λ is the scaling parameter, and the cumulative distribution function is:

[0028] .

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. This invention extracts the most dangerous 300-second waveband (accounting for 5% of the total) from a 3-hour wave and performs high-precision CFD simulation to accurately capture nonlinear physical effects such as wave breaking and viscous damping. Compared with the traditional potential flow method, which ignores nonlinear effects, it has higher accuracy under the same extreme working conditions, thus reducing the prediction error of air gap extreme values.

[0031] 2. This invention generalizes the remaining non-dangerous wave segments using a multilayer perceptron neural network surrogate model, and combines it with local CFD calibration of the 5% dangerous segment to ensure calculation accuracy. At the same time, the time taken to predict the remaining air gap response under a single working condition is less than 10 minutes, which greatly reduces the calculation time compared to the traditional full CFD calculation which takes more than 200 hours in 300 seconds, and is more than 100 times faster than the traditional method. Attached Figure Description

[0032] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0033] Figure 1 This is a flowchart of the method for predicting the statistical characteristics of air gap response of floating wind turbines based on neural networks, as described in this invention.

[0034] Figure 2This is a 3-hour potential flow wave time-history curve.

[0035] Figure 3 This is a 300s wave height duration curve;

[0036] Figure 4 This is a schematic diagram of a multilayer perceptron neural network model;

[0037] Figure 5 The graph shows the test results of the neural network on the CFD dataset.

[0038] Figure 6 The time-history curve of the air gap response predicted by the neural network model;

[0039] Figure 7 This is a distribution diagram of the minimum points of the air gap response;

[0040] Figure 8 This is a frequency histogram of the minimum points of the air gap response;

[0041] Figure 9 A forecast map of the minimum point with a 90% probability of occurrence fitted to the Weibull distribution. Detailed Implementation

[0042] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0043] Specifically, this invention provides a method for predicting the statistical characteristics of the air gap response of a floating wind turbine based on a neural network, such as... Figure 1 As shown, the method includes the following steps:

[0044] S1: Generate wave time history data for all operating conditions;

[0045] Specifically, a 3-hour irregular wave simulation was performed using the potential flow method. The meaningful wave height of the irregular wave was selected as 11.71 m, the peak period as 14.58 s, and the wave spectrum was chosen as the JONSWAP spectrum. The JONSWAP spectrum parameters were input, and the full-time domain wave height-time history curve was output. Figure 2 The 3-hour wave height time-history curve calculated by potential flow theory is given.

[0046] S2: Identify dangerous wave segments and perform local CFD simulation;

[0047] The continuous dangerous wave is extracted and input into the viscous flow solver to obtain the motion response and free surface elevation time history of the floating wind turbine platform. After post-processing, the air gap response of the floating wind turbine is obtained.

[0048] Specifically, based on the wave height duration curve obtained from the potential flow method, the continuous dangerous wave within the 300-second wave height duration near the maximum wave height (i.e., calculation time 4150s) is input into the viscous flow solver. The 300-second wave height duration curve is as follows: Figure 3 As shown, the motion response and free surface elevation time history of the floating wind turbine platform are obtained by solving the equation, and the air gap response of the floating wind turbine is obtained after post-processing. The air gap response of the floating wind turbine is determined by the following equation:

[0049]

[0050] in: It is the air gap value at any spatial point near the wind turbine foundation platform, based on the highest point of the floating structure column. Location coordinates, time history, and wavefront height at that location The air gap time history at any point near the wind turbine foundation platform can be calculated. .

[0051] Based on the initial center of gravity G of the floating wind turbine structure, the initial position C of the blade disk, and the highest point A of the floating structure column, the coordinates of point A need to be updated in real time at every moment. The formula for updating the coordinates of point A is:

[0052]

[0053] Where d and θ represent the translation and rotation of the floating wind turbine, respectively, ϕ represents the influence of the tilt angle of the wind turbine blade disk at the initial moment on the calculation of the position coordinates of point A, and r represents the radius of the wind turbine blade.

[0054] S3: Preprocess CFD data to build the training dataset;

[0055] Specifically, the irregular spectral peak period in the selected operating condition is 11.71s, and the time step of the CFD data results is 1s. Therefore, the data sliding window length is selected as 20. The wave height time history data from time T-20 to T-1 in the input CFD wave height data is used as a set of input features for the air gap response at time T. Similarly, the wave height time history data from time T-20 to T-1 in step S2 is used as a set of input features for the air gap response at time T. The selection of the number of input features is related to the time step of the data and the wave period. When the wave height feature duration is greater than the dominant wave period, the extreme value evolution characteristics can be fully captured.

[0056] S4: Training the air gap response time-series prediction surrogate model;

[0057] Specifically, based on the multi-layer perceptron (MLP) artificial neural network, an input layer, hidden layer, and output layer are configured. An appropriate loss function and optimizer are selected for training to establish a nonlinear mapping proxy model from input features to the air gap response time history. A schematic diagram of the multi-layer perceptron neural network is shown below. Figure 4 As shown, the input layer has 20 features, the hidden layers have 12 features, the output layer is the air gap response value, the loss function is MSE, the optimizer is Adamx, and a nonlinear mapping surrogate model is trained. The trained model's performance on CFD data is shown below. Figure 5 As shown.

[0058] S5: Predicted air gap response time history value;

[0059] Specifically, wave history data following the largest wave in the potential-current wave pattern are selected, and wave height history data from time T-20 to T-1 are used as input features for predicting the air gap response at time T. Using the trained neural network model, the air gap response corresponding to the wave history is predicted, and the air gap response history curve is shown below. Figure 6 As shown. Furthermore, the selected input wave history data needs to be greater than 100 wave cycles. When the duration of the input data exceeds 100 times the characteristic cycle, it can be ensured that the statistical characteristics of the air gap response converge to a stationary random process.

[0060] S6: Extract all the minimum values ​​in the forecast values ​​and perform data filtering;

[0061] Specifically, local minima are extracted from all forecast time-history points, effective extreme value samples are screened at wave cycles, instantaneous noise interference is removed, and an air gap minimum value sequence dataset is generated.

[0062] The steps for extracting the minimum value are as follows: Calculate the arithmetic mean of the air gap response, subtract the time-history data to obtain a zero-mean sequence, identify each upper and lower zero-crossing point in the sequence, locate the minimum point between adjacent upper and lower zero-crossing points, and the original air gap value at this point is the effective minimum value sample. Valid extreme value samples are filtered at wave period intervals of 11.71s to remove instantaneous noise interference, generating an air gap minimum value sequence dataset. The distribution of all obtained minimum value points is selected as follows: Figure 7 As shown.

[0063] S7: Select a suitable distribution function for fitting;

[0064] Specifically, count the number of times the minimum value occurs and draw a frequency histogram, such as... Figure 8 As shown, the selected air gap minimum value sequence is fitted with the Weibull distribution function, and the shape parameter k and scale parameter λ are calculated by the least squares method to establish the cumulative distribution function model.

[0065] The Weibull distribution function is:

[0066]

[0067] Where k is the shape parameter and λ is the scaling parameter. The cumulative distribution function is:

[0068]

[0069] S8: Analyze the key statistical characteristics of the air gap response.

[0070] Specifically, based on the minimum probability of 90% occurrence of the air gap response predicted by the Weibull distribution, the predicted value is as follows: Figure 9 As shown, this matches the obtained statistical values.

[0071] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for predicting the statistical characteristics of the air gap response of a floating wind turbine based on a neural network, characterized in that, The method includes the following steps: Generate wave time history data for all operating conditions; Identify dangerous wave segments and perform local CFD simulation; Preprocess CFD data to build the training dataset; Training a surrogate model for air gap response time-series prediction; Predict the air gap response time history; Extract all the minimum values ​​from the forecast values ​​and perform data filtering; Choose a suitable distribution function for fitting; Analyze the key statistical characteristics of the air gap response.

2. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 1, characterized in that, The steps for generating full-condition wave time history data specifically include: using the potential flow method to complete a 3-hour irregular wave simulation, inputting JONSWAP spectral parameters, and outputting the full-time domain wave height time history curve.

3. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 1, characterized in that, The steps of identifying dangerous wave segments and performing CFD local simulation specifically include: extracting continuous dangerous waves into the viscous flow solver, solving to obtain the motion response and free surface elevation time history of the floating wind turbine platform, and obtaining the air gap response of the floating wind turbine after post-processing.

4. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 3, characterized in that, The air gap response of the floating wind turbine is determined by the following formula: in: It is the air gap value at any spatial point near the wind turbine foundation platform, based on the highest point of the floating structure column. Location coordinates, time history, and wavefront height at that location. The air gap time history at any point near the wind turbine foundation platform can be calculated. .

5. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 1, characterized in that, The step of constructing a training dataset from preprocessed CFD data specifically includes: using wave height history data from time T-20 to T-1 in the CFD data as a set of input features for the air gap response at time T.

6. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 1, characterized in that, The steps for training the air gap response time series prediction proxy model specifically include: configuring an input layer, hidden layer and output layer based on a multilayer perceptron neural network, selecting an appropriate loss function and optimizer for training, and establishing a nonlinear mapping proxy model from input features to air gap response time series.

7. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 1, characterized in that, The steps for predicting the air gap response time history include: selecting wave history data from the largest wave in the potential current wave, and using the wave height history data from time T-20 to time T-1 as the input feature for predicting the air gap response at time T.

8. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 1, characterized in that, The step of extracting all local minima from the forecast values ​​and performing data filtering specifically includes: extracting local minima from all forecast time-history points, filtering effective extreme value samples at wave cycles, eliminating instantaneous noise interference, and generating an air gap minimum value sequence dataset; the step of extracting minimum values ​​is as follows: calculating the arithmetic mean of the air gap response, subtracting the time-history data to obtain a zero-mean sequence, identifying each upper and lower zero-crossing point in the sequence, locating the minimum value point between adjacent upper and lower zero-crossing points, and the original air gap value at this point is the effective minimum value sample.

9. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 1, characterized in that, The step of selecting a suitable distribution function for fitting specifically includes: fitting the selected air gap minimum value sequence with the Weibull distribution function, calculating the shape parameter k and scale parameter λ using the least squares method, and establishing a cumulative distribution function model.

10. The method for predicting the statistical characteristics of air gap response of a floating wind turbine based on a neural network according to claim 9, characterized in that, The Weibull distribution function is: Where k is the shape parameter, λ is the scaling parameter, and the cumulative distribution function is: 。

Images