Sea condition intelligent compensation method based on machine learning

Abstract

The invention relates to the technical field of ocean wave monitoring, in particular to an intelligent sea condition compensation method based on machine learning, which corrects an error generated by a buoy motion condition under a high sea condition, which cannot be simulated by a rotary wave simulation test board adopting sine waves as an excitation source, through a large-scale wave flume extreme working condition test. A buoy wave measurement data compensation method in a severe environment is researched, and the buoy measurement data precision is improved.

Description

Technical Field

[0001] This invention relates to the field of ocean wave monitoring technology, and in particular to a machine learning-based intelligent sea condition compensation method. Background Technology

[0002] Currently, wave buoys are crucial equipment for acquiring ocean wave parameters in the field of ocean wave monitoring technology. Existing technologies typically employ wave measurement methods based on accelerometers. The basic process involves acquiring raw acceleration data using accelerometers on the buoy, and then calculating wave parameters using numerical integration methods. Traditional wave buoy data processing methods primarily rely on classical numerical integration techniques, such as rectangular integration and trapezoidal integration.

[0003] In existing technologies, wave buoy data processing typically employs the following traditional methods: first, the raw data collected by the accelerometer is simply filtered; then, velocity and displacement are calculated using a fixed numerical integration formula; and finally, wave parameters are obtained based on statistical methods. While this method can meet basic measurement needs under normal sea conditions, it has significant technical limitations. Most existing technologies use a single integration algorithm, lacking adaptability to complex sea conditions and failing to effectively handle measurement errors under extreme wave conditions.

[0004] The existing technical problems include:

[0005] 1. Traditional numerical integration methods produce significant cumulative errors in long-term data processing. This is especially true in wave buoy measurements, where the error is amplified by the double integration of acceleration data, leading to serious deviations in the final wave parameter measurements.

[0006] 2. Existing technologies suffer a sharp decline in measurement accuracy when processing wave data under extreme conditions such as high sea states and steep waves. Traditional methods cannot effectively handle nonlinear wave characteristics, resulting in insufficient reliability of measurement data under severe sea conditions.

[0007] 3. Most existing technologies use fixed algorithm parameters, which cannot be adaptively adjusted according to actual sea conditions. The lack of intelligent identification and compensation capabilities for measurement errors results in poor robustness of the measurement system.

[0008] 4. Traditional methods typically employ linear data processing flows, lacking the ability to integrate and utilize multi-source information. This prevents the full utilization of wave physical characteristics for data verification and compensation, thus limiting further improvements in measurement accuracy. Summary of the Invention

[0009] The technical solution adopted by this invention to solve the above-mentioned technical problems is: a sea state intelligent compensation method based on machine learning, comprising the following steps:

[0010] S1. Physical model tests were conducted in a water tank using wave buoy models at scales of 1:2, 1:3, and 1:5. Acceleration sensors were attached to the buoys, and raw acceleration time history data were collected through the acceleration sensors on the wave buoys.

[0011] S2. Perform a Fourier transform on the original acceleration time history data to extract the acceleration components within the effective wave frequency range;

[0012] S3. The filtered acceleration data is integrated using the trapezoidal integral method to obtain the buoy motion velocity time history data.

[0013] S4. Intelligent compensation for accumulated errors generated during the integration process based on machine learning models;

[0014] S5. Perform a second integration on the compensated velocity data to obtain the buoy's vertical displacement time history data;

[0015] S6. Calculate wave parameters based on displacement time history data and output the compensated measurement results.

[0016] Furthermore, the step S2, which involves performing a Fourier transform on the original acceleration time history data to extract the acceleration components within the effective wave frequency range, specifically includes the following steps:

[0017] S2-1. Perform Fourier transform on the acceleration time history data a(t);

[0018] S2-2, Set the effective frequency range for first-order and second-order waves;

[0019] S2-3. Set the acceleration amplitude outside the effective frequency range to zero, and retain the data within the effective frequency range;

[0020] S2-4. Perform an inverse Fourier transform on the processed frequency domain data.

[0021] Furthermore, the step S3, which uses the trapezoidal integral method to integrate the filtered acceleration data, specifically includes the following steps:

[0022] S3-1, Set the sampling time interval Δt;

[0023] S3-2, According to the formula The velocity is calculated, where v(t) is the velocity value at the current moment, v(t-Δt) is the velocity value at the previous moment, Δt is the sampling time interval, ã(t) is the filtered acceleration at the current moment, and ã(t-Δt) is the filtered acceleration at the previous moment.

[0024] 3-3. Repeat the above calculations for each sampling point to obtain complete velocity time history data.

[0025] Furthermore, the intelligent compensation for the accumulated error generated during the integration process based on the machine learning model in S4 specifically includes the following steps:

[0026] S4-1. Construct a feature vector containing wave buoy geometric scale, wave steepness parameter, and wave period;

[0027] S4-2. Input the feature vectors into the pre-trained deep learning neural network model;

[0028] S4-3. Obtain the error compensation coefficients output by the deep learning neural network model;

[0029] S4-4. Correct the speed time history data using the compensation coefficient.

[0030] Furthermore, the second integration of the compensated velocity data in step S5 specifically includes the following steps:

[0031] S5-1. Displacement calculation is performed using the Gaussian integral method.

[0032] S5-2. Map the integration interval to the [-1,1] interval through variable transformation;

[0033] S5-3, Calculate using the five-point Gauss-Legend integral formula;

[0034] S5-4. Obtain the time history data of the buoy's vertical displacement based on the integration results.

[0035] Furthermore, when using the five-point Gauss-Legend integral formula in S5-3 for calculation, the integration nodes are:

[0036] T1=-0.9061798459,t2=-0.5384693101,t3=0,

[0037] T4=0.5384693101,t5=0.9061798459;

[0038] The corresponding quadrature coefficients are:

[0039] λ1=0.2369268851, λ2=0.4786286705, λ3=0.5688888889,

[0040] λ4=0.4786286705, λ5=0.2369268851.

[0041] Furthermore, the wave parameter calculation based on displacement time history data in step S6 specifically includes the following steps:

[0042] S6-1. Extract wave height sequence from displacement time history data;

[0043] S6-2, Calculate the maximum wave height H max Significant wave height H 1 / 3 , average wave height H m ;

[0044] S6-3, Calculate the effective wave height period T H1 / 3 Average period T m ;

[0045] S6-4, Regarding the maximum wave height H max Significant wave height H 1 / 3 , average wave height H m Effective wave height period T H1 / 3 Average period T m The parameters are subjected to quality inspection and error assessment.

[0046] Furthermore, step S4 also includes a model optimization step:

[0047] S4-5. Establish an error evaluation function to calculate the deviation between the measurement results before and after compensation and the true value;

[0048] S4-6. Use optimization algorithms to adjust the parameters of the machine learning model;

[0049] S4-7. Adjust the compensation strategy dynamically based on the optimization results.

[0050] Furthermore, it also includes quality control steps:

[0051] S7. Real-time monitoring of the signal-to-noise ratio of acceleration data;

[0052] S8. When the signal-to-noise ratio is lower than the preset threshold, automatically increase the filtering strength;

[0053] S9. Mark and remove abnormal fluctuation data;

[0054] S10. Output measurement results and provide a data quality assessment report.

[0055] The advantages of this invention are as follows: This invention provides a sea state intelligent compensation method based on machine learning, which has the following advantages:

[0056] 1. This invention uses Fourier transform to convert acceleration data to the frequency domain for precise filtering, combined with an optimized trapezoidal integral formula. This significantly reduces error accumulation during numerical integration. For example... Figure 23 As shown, by comparing the error distribution of the traditional method and the method of the present invention, this technology controls the average wave height measurement error to within 5%, the effective wave height error to less than 8%, and the maximum wave height error to be reduced by more than 60%, thus completely solving the problem of amplified integral error mentioned in the background technology.

[0057] 2. This invention introduces a deep learning-based intelligent compensation model. By constructing an adaptive compensation system that includes feature vectors such as wave buoy geometric scale, wave steepness parameters, and wave period, it effectively copes with extreme conditions such as high sea states and steep waves. For example... Figure 26 and Figure 27 The comparison results show that even in severe sea conditions with wave steepness greater than 0.05, this method can still maintain stable measurement accuracy, solving the technical problem of the sharp decline in accuracy of traditional methods in nonlinear wave characteristic measurement.

[0058] 3. This invention establishes a complete model optimization system, including error evaluation functions, gradient descent algorithm parameter optimization, and dynamic compensation strategy adjustment. This technology can automatically adjust algorithm parameters according to real-time sea conditions, solving the problems of fixed parameters and lack of adaptability in traditional methods. Through the model optimization steps, the system maintains optimal performance under different sea conditions, significantly improving the robustness of the measurement system.

[0059] 4. This invention employs multi-parameter statistical calculations and a rigorous quality control process, fully utilizing the physical characteristics of waves for data verification. Through quality inspection and error assessment steps, combined with wave height sequence analysis, parameter statistics, and quality monitoring, comprehensive verification of measurement results is achieved. This technology breaks through the limitations of traditional linear processing procedures, elevating measurement accuracy to a new level through multi-source information fusion. Attached Figure Description

[0060] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram of the Gaussian integral method used in this invention. Figure 2 This is a schematic diagram of the rectangular integration method that can be used in this invention. Figure 3 This is a schematic diagram of the trapezoidal integral method that can be used in this invention. Figure 4 This is a schematic diagram of the wavefront measurement results from the wave height sensor (operating condition 1 - small wave height, small period). Figure 5 This is a schematic diagram of the wavefront measurement results from a wave measuring radar (operating condition 1 - wavelet height and short period). Figure 6 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 1, 1:2 scale ratio results). Figure 7 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 1, 1:3 scale ratio results). Figure 8This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 1, 1:5 scale ratio results). Figure 9 Schematic diagram of wavefront measurement results from different instruments (Working condition 2, wave height sensor). Figure 10 Schematic diagram of wavefront measurement results from different instruments (Working condition 2, wave measuring radar). Figure 11 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 2, 1:2 scale ratio results). Figure 12 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 2, 1:3 scale ratio results). Figure 13 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 2, 1:5 scale ratio results). Figure 14 A schematic diagram of wavefront measurement results from different instruments (condition 7, wave height sensor). Figure 15 Schematic diagram of wavefront measurement results from different instruments (Operating Condition 7, wave measuring radar). Figure 16 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 7, 1:2 scale ratio results). Figure 17 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 7, 1:3 scale ratio results). Figure 18 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 7, 1:5 scale ratio results). Figure 19 Schematic diagram of wavefront measurement results from different instruments (operating condition 12) Figure 19 a is a wave height sensor. Figure 19 b represents wave-measuring radar. Figure 20 A schematic diagram showing the measurement and calculation results of buoys with different scales (operating condition 12). Figure 20 a represents a 1:2 scale result. Figure 20 b represents the 1:3 scale result. Figure 20 c represents a 1:5 scale result. Figure 21 Schematic diagram of wavefront measurement results from different instruments (operating condition 19) Figure 21 a is a wave height sensor. Figure 21 b represents wave-measuring radar. Figure 22 A schematic diagram showing the measurement and calculation results of buoys with different scales (operating condition 19). Figure 22 a represents a 1:2 scale result; Figure 22 b represents the 1:3 scale result. Figure 22 c represents a 1:5 scale result. Figure 23 A diagram showing the comparison of statistical errors for different wave heights in each group ( Figure 23 'a' represents the average wave height. Figure 23 b represents the significant wave height. Figure 23 c represents the maximum wave height. Figure 24 A schematic diagram comparing the statistical errors of different wave heights based on different periods. Figure 1 ( Figure 24 'a' represents the average wave height. Figure 24 b represents the significant wave height. Figure 24 c represents the maximum wave height. Figure 25 A schematic diagram comparing the statistical errors of different wave heights based on different periods. Figure 2 ( Figure 25 'a' represents the average wave height. Figure 25 b represents the significant wave height. Figure 25 c represents the maximum wave height. Figure 26 This is a diagram comparing the statistical errors of two wave height calculation methods based on different periods. Figure 26 'a' represents the average wave height. Figure 26 b represents the significant wave height. Figure 26 c represents the maximum wave height. Figure 27 This is a schematic diagram comparing the statistical errors of two wave height calculation methods based on different wave steepness. Figure 27 'a' represents the average wave height. Figure 27 b represents the significant wave height. Figure 27 c represents the maximum wave height. Detailed Implementation

[0062] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0063] Example 1:

[0064] This invention employs 1:2, 1:3, and 1:5 scale wave buoy models for physical model testing. The buoy models are equipped with anchorage structures of corresponding scales and tested in a large water tank. Accelerometers are attached to the surface of the model buoys, and the obtained acceleration sensor data reflects the wave data measured by the buoys. Simultaneously, capacitive wave height sensors and radar wave meters are placed near the buoys, and the wave data measured by these represent the true wave values. The relationship between the two is compared to study the compensation method for the buoy wave measurement data. The models are designed according to the gravity similarity criterion, and the buoy dimensions satisfy geometric similarity. The scale relationships are as follows: , , , , ; In the formula Represents the model length scale; Represents the length of the prototype; Represents the model length; Represents a time scale; Represents the unit width flow ratio; Represents wave height ratio, Represents the wave period scale, The geometric scales represent force ratios. Three types of buoy models were used in this experiment, with geometric scales of 1:2, 1:3, and 1:5. The parameter scales of the 1:2 buoy model are as follows: dimensional scale: λ = 1:2, gravity scale: λ F =1:8, wave height scale: λ=1:2, wave period scale: λ t =1:2 0.5, 1:3. The scale ratios of the buoy model parameters are as follows: Dimension scale: λ=1:3, Gravity scale: λ F =1:27, wave height scale: λ=1:3, wave period scale: λ t =1:3 0.5. The scales of the parameters of the 1:5 buoy model are as follows: Dimension scale: λ=1:5, Gravity scale: λ F =1:125, wave height scale: λ=1:5, wave period scale: λ t =1:50.5. While ensuring similarity in gravity and dimensions, similarity in the center of gravity also needs to be guaranteed. Inertia testing is required. Irregular waves will be selected for the experiment to closely reflect actual conditions. JONSWAP spectrum simulation will be used, and its analytical expression is: (3-1)

[0065] In the formula: ; ;

[0066] Where: γ is the spectral peak factor, taken as 3.3, fp is the peak frequency, which is the reciprocal of the spectral peak frequency period TP, S(f) is the spectral density, and H... 1 / 3 The significant wave height is f, and the frequency is f. The average period is used. Given the effective wave height and period, the data is fed into a computer for wave spectrum simulation. After correction, the spectral density, peak frequency, spectral energy, and effective wave height near the peak frequency meet the requirements of the "Technical Specification for Simulation Tests of Water Transport Engineering" (JTS / T231-2021). The wave train for each wave element maintains a wave count of over 150, ensuring that both the wave buoy and the benchmark wave sensor can collect over 100 waves to guarantee statistical significance. Model tests are conducted in a large-scale wave flume (456 m long, 5.0 m wide, 8-12 m deep, capable of generating waves up to 3.5 meters high). The flume is divided into wave-generating, wave-producing, test, and wave-dissipating sections. The platform model system is positioned in the stable wave region of the flume to minimize sidewall effects. The wave measurement system used in this invention is a large-scale dynamic capacitive wave height measurement system with a range of 1.8 m. The system performs sensor...

[0067] Temperature correction was applied, and a radar wave meter was used as the wave measurement system. Wave height and period parameters were considered in the test setup. Data from the buoy acceleration, capacitive wave height sensor, and radar wave meter were recorded during the experiment. A total of 20 large test groups were set up, subdivided into 60 smaller groups. Each smaller group was tested three times.

[0068]

[0069] like Figure 1 As shown, this invention uses the Gaussian integration method to calculate the acceleration of wave buoys. The Gaussian integration method is a numerical integration method based on orthogonal polynomials. It selects appropriate nodes (Gaussian points) and corresponding weights within the integration interval to ensure that the integration result is accurate for a given degree polynomial. For example, the Gauss-Legendary integration method is used to calculate the integral over the interval [-1, 1]. For n Gaussian points, a polynomial of degree 2n-1 can be integrated accurately. Its basic idea is to select nodes and weights to ensure that the approximate integration formula has the highest algebraic accuracy in polynomial space. The Gaussian integration method has high accuracy, especially for smooth functions, achieving high accuracy with a relatively small number of nodes. Compared to the previous equidistant node integration method, it uses non-equidistant nodes, and the node positions and weights are determined based on the properties of orthogonal polynomials, fully utilizing the information of the function and improving integration efficiency.

[0070]

[0071] First, perform the transformation.

[0072]

[0073] The original integral is transformed into an integral over the interval [-1, 1].

[0074]

[0075] According to the interpolation formula

[0076]

[0077] If the n interpolation nodes take the Legendre polynomial

[0078]

[0079] For n zeros in the interval [-1, 1], the algebraic accuracy of the interpolation quadrature formula is 2n-1.

[0080] In this example, n=5, and a variable step size integration method is used. The interpolation nodes are then:

[0081]

[0082] The product coefficient is

[0083]

[0084] This invention can also use the Chebyshev quadrature method to calculate the acceleration of wave buoys:

[0085] First, perform the transformation.

[0086]

[0087] The original integral is transformed into an integral over the interval [-1, 1].

[0088]

[0089] Chebyshev's quadrature formula is

[0090]

[0091] In this example, n=5, and a variable step size integration method is used. The interpolation nodes are then:

[0092]

[0093] To address this invention, a trapezoidal integral is used to analyze the measured acceleration results, and a certain numerical method is employed to avoid errors caused by the initial acceleration. Assuming the measured acceleration of the buoy in the experiment is a(t), a bandpass filter is first used to remove the cumulative errors in velocity and displacement caused by the initial acceleration.

[0094]

[0095] In the formula, ℱ represents the Fourier transform, N represents the number of data points in the Fourier transform, and ω represents the angular frequency of the Fourier transform; a(ω) represents the frequency domain result of the acceleration time history curve, which is the energy distribution of the time series at different frequencies. The Fourier transform completes the conversion from the time domain to the frequency domain, obtaining the amplitude spectrum curve a(ω) of the floating body displacement in the i-direction over time. Analyzing the amplitude spectrum curve, the effective frequency ranges of the first and second-order waves are selected, and the amplitudes of other invalid frequency ranges are assigned 0, retaining the data within the effective frequency range to obtain the processed amplitude spectrum curve; then, an inverse Fourier transform is performed on the processed amplitude spectrum curve.

[0096] In the formula, ℱ -1 This represents the inverse Fourier transform, thus obtaining the acceleration time history curves of the buoy at the first and second wave frequencies. The vertical velocity of the buoy is then derived from the corrected acceleration time history curves to obtain the velocity time history curves.

[0097] To obtain the buoy's displacement more accurately through integration, the velocity time history curve also needs to be processed, using the same method as for acceleration. The expression is as follows:

[0098]

[0099]

[0100] Based on the specific circumstances, the corrected speed was then adjusted. Perform compensation calculations.

[0101]

[0102] Obtain the buoy velocity time history curve after compensation calculation Therefore, the vertical displacement of the buoy is obtained based on the corrected and compensated velocity time history curve, resulting in the displacement time history curve.

[0103]

[0104] Similarly, displacement compensation calculations are performed based on specific circumstances.

[0105] The buoy displacement s obtained based on compensation calculations c(t), the effective value of displacement is statistically obtained and compared with the effective value obtained by the wave height sensor, and the maximum value of displacement is compared with the maximum value obtained by the wave height sensor.

[0106] like Figure 3 As shown, this invention can also use the trapezoidal integral method to calculate the acceleration of wave buoys. The trapezoidal integral method divides the integration interval into n smaller intervals, and then approximates the area between the curve and the x-axis using a trapezoid in each smaller interval. For each smaller interval [xi, xi+1], the upper base of the trapezoid is f(xi), the lower base is f(xi+1), and the height is ∆x. The area of ​​the trapezoid is (f(xi) + f(xi+1)) / 2 × ∆x. Adding the areas of all trapezoids gives an approximate value of the integral. This invention can also use the rectangular integral method to calculate the acceleration of wave buoys, specifically, as shown... Figure 2 As shown, the integration interval is divided into several smaller intervals, each approximated as a rectangular region for area calculation. For the integral of the function f(x) on the interval [a, b], the interval is divided into n smaller intervals, with the width of each smaller interval being ∆x = (ba) / n. In each smaller interval [xi, xi+1], a point is taken within that smaller interval (usually the left endpoint, right endpoint, or midpoint), and the area of ​​the rectangle is used to approximate the area between the curve and the x-axis within that smaller interval. For example, if the left endpoint is taken, the area of ​​the i-th rectangle is f(xi) × ∆x. Then, the areas of all rectangles are added together to obtain an approximate value of the integral. The rectangle method is a simple numerical integration method that directly approximates the area under the curve using a rectangle. Its accuracy is relatively low, especially for functions with drastic changes, where the error is relatively large. Different point selection methods (left endpoint, right endpoint, midpoint) will bring different error characteristics. Generally speaking, the midpoint rectangle method is more accurate than the left endpoint or right endpoint rectangle method because it balances the changes of the function within the smaller intervals to some extent. The trapezoidal integration method described above is more accurate than the rectangle integration method. Because it considers the function values ​​at the two endpoints of the interval and uses a trapezoid to approximate the curve, it can better adapt to the changing trend of the function. However, when the second derivative of the function is large, the trapezoidal integral method will also have some error. This invention can also use Simpson's integral method to calculate the acceleration of wave buoys. Simpson's integral method approximates the curve with a parabola to calculate the integral. It requires the integration interval to be divided into an even number of intervals, with each pair of intervals forming a parabolic segment. For each parabolic segment, a quadratic function is used to fit the three points (two dividing points and the middle point), and then the area between the parabolic segment and the x-axis is calculated. Assuming the width between each interval is h, for three points x... i , x i+1 ,x i+2 The corresponding function values ​​are f(x) i ), f(x i+1 ), f(xi+2 The area calculation formula is (h / 3)×[f(x)]. i )+4f(x i+1 )+f(x i+2 The approximate value of the integral is obtained by summing the areas of all parabolic segments. Simpson's integral method is more accurate than the trapezoidal and rectangular integral methods. It uses a quadratic function to approximate the curve, which can better adapt to changes in the curve, especially for functions with continuous second derivatives, where Simpson's integral method has smaller errors. However, it requires an odd number of division points in the integration interval, which may be inconvenient in some cases. This invention can also use an automatic integration method to calculate the acceleration of wave buoys. The automatic integration method is an adaptive integration method. It automatically adjusts the division of the integration interval by analyzing the local properties of the integrand. In regions where the function changes drastically, the interval is divided more finely; in regions where the function changes gently, the interval is divided more coarsely. Its basic idea is to use a basic integration rule (such as Simpson's rule or trapezoidal rule) to calculate the integral and estimate the error by comparing the integration results at different accuracies. If the error exceeds a predetermined tolerance, the interval is subdivided until the accuracy requirement is met.

[0107] Automatic integration methods can automatically adjust the integration strategy according to the characteristics of the function, making them suitable for situations where the integration function varies significantly. This effectively improves the accuracy and efficiency of integration. Compared to the previous fixed-interval integration methods, it is more flexible and intelligent, reducing computational load while maintaining accuracy.

[0108] This invention uses accelerometer data to reflect wave data measured by a buoy, and compares and analyzes the wave buoy data with that of a wave height meter and a radar wave meter. Specifically, a total of 20 experimental groups were conducted, including physical model experiments at different scales. Each experimental group is described below. Wave height sensors and wave-measuring radars were installed at different locations in the large water tank to monitor wave height and statistically analyze the wave height and period information for each operating condition. Several examples are used to illustrate the wave height measurement results and buoy wave buoy data. First, a comparative analysis is performed on the small wave height and small period results for an effective wave height of approximately 20cm, such as... Figure 4 and Figure 5 The diagram shows the wavefront measurement results of the wave height sensor (operating condition 1 - small wave height, small period). Figure 5 The schematic diagram of the wavefront measurement results of the wave measuring radar (operating condition 1 - wavelet height, short period) is shown in the table below. Table 4-1 shows the significant wave height (H). 1 / 3 ), maximum wave height (H max ), average wave height (H m ), effective wave height period (T) H1 / 3 ) and average period (T) m ) Statistical results.

[0109] Table 4-1: Statistical Results of Incident Wave Height and Period (Condition 1)

[0110]

[0111] The model values ​​of the measurement results of each buoy are as follows: Figure 6 , 7 As shown in Figures 8 and 9, the statistical results of different scales converted back to the original model are shown in Tables 4-2 to 4-4. Through comparative analysis, under the condition of small wave height and small period, the buoy measurement results of each scale are very accurate. The statistical results of significant wave height, average wave height and period are all within 5%, and the maximum wave height is slightly different.

[0112] Table 4-2: Comparison of results for 1:2 scale buoys (Operating condition 1 - small wave height, small period)

[0113]

[0114] Table 4-3: Comparison of results for 1:3 scale buoys (Operating Condition 1 - Small wave height, small period)

[0115]

[0116] Table 4-4: Comparison of results for 1:5 scale buoys (Operating condition 1 - small wave height, small period)

[0117]

[0118] Subsequently, with similar wave heights, the wave period was gradually increased to test the buoy's performance and measurement analysis methods. The average wave period was increased to 2.82 s, designated as Condition 2, and physical model tests were conducted. Figure 9 and Figure 10 As shown, Figure 9 Schematic diagram of wavefront measurement results from different instruments (Working condition 2, wave height sensor). Figure 10 Schematic diagram of wavefront measurement results from different instruments (Working condition 2, wave measuring radar).

[0119] Table 4-5: Statistical Results of Incident Wave Height and Period (Operating Condition 2)

[0120]

[0121] Subsequently, acceleration time history curves for buoys of different scales were obtained, and displacement time history curves were calculated, as follows: Figure 11 , 12 As shown in Figure 13, Figure 11 This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 2, 1:2 scale ratio results). Figure 12This is a schematic diagram of the measurement and calculation results for buoys with different scale ratios (condition 2, 1:3 scale ratio results). Figure 13 The diagram illustrates the measurement and calculation results for buoys of different scales (condition 2, 1:5 scale results). Similarly, the statistical results of different scales converted back to the original model are shown in Tables 4-6 to 4-8. Comparative analysis also reveals that under this condition, the buoy measurement results for each scale are highly accurate, with the statistical results for significant wave height, average wave height, and period all within 5%, and only slight differences in maximum wave height.

[0122] Table 4-6: Comparison of results for 1:2 scale buoys (operating condition 2)

[0123]

[0124] Table 4-7: Comparison of results for 1:3 scale buoys (operating condition 2)

[0125]

[0126] Table 4-8: Comparison of results for 1:5 scale buoys (operating condition 2)

[0127]

[0128] Further increasing the wave height and wave period, the buoy's performance and measurement analysis methods were tested. The effective wave height was increased to approximately 0.50 m, and the average wave period was approximately 3.40 s, designated as Condition 7. Physical model tests were then conducted. Figure 14 , 15 As shown, Figure 14 Schematic diagram of wavefront measurement results from different instruments (Operating Condition 7, wave height sensor). Figure 15 The diagram shows the wavefront measurement results of different instruments (operating condition 7, wave measuring radar). Table 4-9 shows the statistical results of different wave heights and periods.

[0129] Table 4-9: Statistical Results of Incident Wave Height and Period (Operating Condition 7)

[0130]

[0131] Subsequently, displacement time history curves were calculated based on the acceleration time history curves of buoys with different scales, such as... Figure 16 , 17 As shown in Figure 18, Figure 16 This is a schematic diagram showing the measurement and calculation results of buoys with different scale ratios (operating condition 7, 1:2 scale ratio results). Figure 17 This diagram illustrates the measurement and calculation results for buoys with different scale ratios (condition 7, 1:3 scale ratio results). Figure 18The diagram illustrates the measurement and calculation results for buoys with different scales (condition 7, 1:5 scale results). Similarly, the statistical results of different scales converted back to their original form are shown in Tables 4-10 to 4-12. Comparative analysis also reveals that under this condition, the measurement results for buoys with a scale of 1:2 are highly accurate, with statistical results for significant wave height, average wave height, and period all within 5%. However, buoys with scales of 1:3 and 1:5 exhibit errors in measuring the maximum wave height, with the error increasing as the wave height increases.

[0132] Table 4-10: Comparison of results for 1:2 scale buoys (operating condition 7)

[0133]

[0134] Table 4-11: Comparison of results for 1:3 scale buoys (operating condition 7)

[0135]

[0136] Table 4-12: Comparison of results for 1:5 scale buoys (operating condition 7)

[0137]

[0138] Further increasing the wave height and wave period, the buoy's performance and measurement analysis methods were tested. The effective wave height was increased to approximately 0.65 m, and the average wave period was approximately 4.10 s, designated as Condition 12, and physical model tests were conducted. Figure 19 As shown, Figure 19 Schematic diagram of wavefront measurement results from different instruments (operating condition 12) Figure 19 a is a wave height sensor. Figure 19 (b represents the wave-measuring radar), Table 4-13 shows the statistical results for different wave heights and periods.

[0139] Table 4-13: Statistical Results of Incident Wave Height and Period (Operating Condition 12)

[0140]

[0141] Subsequently, displacement time history curves were calculated based on the acceleration time history curves of buoys with different scales, such as... Figure 20 As shown, Figure 20 A schematic diagram showing the measurement and calculation results of buoys with different scales (operating condition 12). Figure 20 a represents a 1:2 scale result. Figure 20 b represents the 1:3 scale result. Figure 20(c represents the result of a 1:5 scale). Similarly, the statistical results of different scales converted back to the original model are shown in Tables 4-14 to 4-16. Through comparative analysis, it was also found that under this working condition, the buoy measurement results of a 1:2 scale were very accurate, with the statistical results of significant wave height, average wave height, and period all within 5%. However, the 1:3 and 1:5 scales showed errors in the measurement of the maximum wave height, with the error increasing as the wave height increased.

[0142] Table 4-14: Comparison of results for 1:2 scale buoys (operating condition 12)

[0143]

[0144] Table 4-15: Comparison of results for 1:3 scale buoys (operating condition 12)

[0145]

[0146] Table 4-16: Comparison of results for 1:5 scale buoys (operating condition 12)

[0147]

[0148] To address extreme wave conditions, the wave height and period were increased to test the buoy's performance and measurement analysis methods. The effective wave height was increased to approximately 1.043 m, and the average wave period was approximately 5.95 s, designated as Condition 19, and physical model tests were conducted. Figure 21 As shown, Figure 21 Schematic diagram of wavefront measurement results from different instruments (operating condition 19) Figure 21 a is a wave height sensor. Figure 21 (b represents wave measuring radar). Table 4-17 shows the statistical results of different wave heights and periods for wave height sensors and wave measuring radars.

[0149] Table 4-17: Statistical Results of Incident Wave Height Period (Operating Condition 19)

[0150]

[0151] Subsequently, displacement time history curves were calculated based on the acceleration time history curves of buoys with different scales, such as... Figure 22 As shown, Figure 22 A schematic diagram showing the measurement and calculation results of buoys with different scales (operating condition 19). Figure 22 a represents a 1:2 scale result; Figure 22 b represents the 1:3 scale result. Figure 22(c represents the result of a 1:5 scale). Similarly, the statistical results of different scales converted back to the original model are shown in Tables 4-18 to 4-20. Through comparative analysis, it was also found that under this working condition, the buoy measurement results of a 1:2 scale were very accurate, with the statistical results of significant wave height, average wave height, and period all within 5%. However, the 1:3 and 1:5 scales showed errors in the measurement of the maximum wave height, with the error increasing as the wave height increased.

[0152] Table 4-18: Comparison of results for 1:2 scale buoys (operating condition 19)

[0153]

[0154] Table 4-19: Comparison of results for 1:3 scale buoys (operating condition 19)

[0155]

[0156] Table 4-20: Comparison of results for 1:5 scale buoys (operating condition 19)

[0157]

[0158] For other operating conditions, time history curves will no longer be displayed. Only the comparison results of statistical wave heights (maximum wave height, significant wave height, average wave height) and periods (significant wave height period, average period) will be shown. The specific comparison results are shown in Tables 4-20 to 4-35.

[0159] Table 4-21: Comparison Results of Measurements at Different Scales under Working Condition 3

[0160]

[0161] Table 4-22: Comparison Results of Measurements at Different Scales under Working Condition 4

[0162]

[0163] Table 4-23: Comparison Results of Measurements at Different Scales under Working Condition 5

[0164]

[0165] Table 4-24: Comparison Results of Measurements at Different Scales under Working Condition 6

[0166]

[0167] Table 4-25: Comparison Results of Measurements at Different Scales under Working Condition 8

[0168]

[0169] Table 4-26: Comparison Results of Measurements at Different Scales under Working Condition 9

[0170]

[0171] Table 4-27: Comparison Results of Measurements at Different Scales under Working Condition 10

[0172]

[0173] Table 4-28: Comparison Results of Measurements at Different Scales under Working Condition 11

[0174]

[0175] Table 4-29: Comparison Results of Measurements at Different Scales under Working Condition 13

[0176]

[0177] Table 4-30: Comparison Results of Measurements at Different Scales under Working Condition 14

[0178]

[0179] Table 4-31: Comparison Results of Measurements at Different Scales under Working Condition 15

[0180]

[0181] Table 4-32: Comparison Results of Measurements at Different Scales under Working Condition 16

[0182]

[0183] Table 4-33: Comparison Results of Measurements at Different Scales under Working Condition 17

[0184]

[0185] Table 4-34: Comparison Results of Measurements at Different Scales under Working Condition 18

[0186]

[0187] Table 4-35: Comparison Results of Measurements at Different Scales under Working Condition 22

[0188]

[0189] This invention analyzes the error offset pattern and further compares the errors of different groups, dividing the comparison into the following parts: first, comparing the errors of each scale of mean wave height, significant wave height, and maximum wave height, such as... Figure 23 As shown, Figure 23 A diagram showing the comparison of statistical errors for different wave heights in each group ( Figure 23 'a' represents the average wave height. Figure 23 b represents the significant wave height. Figure 23(c represents the maximum wave height). Comparative analysis shows that the error in the average wave height is relatively small, generally within 5%. The error in the effective wave height has increased slightly, but remains less than 8%. The error in the maximum wave height is larger, with some groups potentially exceeding 20%. This is due to two main reasons: First, limited data and experimental time constraints mean that each group of tests only had 100-200 waves, making errors more likely in the maximum value. Increasing the number of waves would reduce the error. Second, the methodological principle is flawed; integral calculations at the maximum value are prone to over-prediction. Numerical methods can be used to optimize the maximum value. Furthermore, it was found that smaller scale ratios lead to larger errors. Overall, scale ratios of 1:2 and 1:3 show relatively small errors, while scale ratios of 1:5 show larger errors in some groups. This indicates a scale effect in buoy performance comparison. Where possible, large-scale physical model tests should be used to better measure accuracy. Error analysis was then conducted on each group after converting it back to the prototype for different wave periods and wave steepness conditions. Figure 24 A schematic diagram comparing the statistical errors of different wave heights based on different periods. Figure 1 ( Figure 24 'a' represents the average wave height. Figure 24 b represents the significant wave height. Figure 24 (c is the maximum wave height) Figure 25 A schematic diagram comparing the statistical errors of different wave heights based on different periods. Figure 2 ( Figure 25 'a' represents the average wave height. Figure 25 b represents the significant wave height. Figure 25 (c is the maximum wave height), from Figure 24 and Figure 25 The analysis results show that the error distribution is relatively uniform for the average wave height and significant wave height. There are certain degrees of error for each period and each wave steepness, but the overall error is controllable. However, for the error of the maximum wave height, the larger the wave steepness, the greater the error, which means the stronger the wave nonlinearity, the greater the error of the buoy.

[0190] Furthermore, this invention requires compensation analysis for cases with small scale and steep waves to ensure the accuracy of buoy wave height calculation. After compensation analysis, the error results are as follows: Figure 26 and Figure 27 As shown in the figure. The results analysis shows that for the average wave height and significant wave height, the compensation effect is better because the error is not large. For the maximum wave height, the compensation algorithm is very effective, reducing the error by up to 60%, which effectively ensures the accuracy of the buoy's measurement in extreme wave environments.

[0191] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine learning-based intelligent sea state compensation method, characterized in that, Includes the following steps: S1. Physical model tests were conducted in a water tank using wave buoy models at scales of 1:2, 1:3, and 1:

5. Acceleration sensors were attached to the buoys, and raw acceleration time history data were collected through the acceleration sensors on the wave buoys. S2. Perform a Fourier transform on the original acceleration time history data to extract the acceleration components within the effective wave frequency range; S3. The filtered acceleration data is integrated using the trapezoidal integral method to obtain the buoy motion velocity time history data. S4. Intelligent compensation for accumulated errors generated during the integration process based on machine learning models; S5. Perform a second integration on the compensated velocity data to obtain the buoy's vertical displacement time history data; S6. Calculate wave parameters based on displacement time history data and output the compensated measurement results.

2. The intelligent sea state compensation method based on machine learning according to claim 1, characterized in that, The step S2, which involves performing a Fourier transform on the original acceleration time history data to extract acceleration components within the effective wave frequency range, specifically includes the following steps: S2-1. Perform Fourier transform on the acceleration time history data a(t); S2-2, Set the effective frequency range for first-order and second-order waves; S2-3. Set the acceleration amplitude outside the effective frequency range to zero, and retain the data within the effective frequency range; S2-4. Perform an inverse Fourier transform on the processed frequency domain data.

3. The intelligent sea state compensation method based on machine learning according to claim 1, characterized in that, The step S3, which uses the trapezoidal integral method to integrate the filtered acceleration data, specifically includes the following steps: S3-1, Set the sampling time interval Δt; S3-2, According to the formula The velocity is calculated, where v(t) is the velocity value at the current moment, v(t-Δt) is the velocity value at the previous moment, Δt is the sampling time interval, ã(t) is the filtered acceleration at the current moment, and ã(t-Δt) is the filtered acceleration at the previous moment. 3-3. Repeat the above calculations for each sampling point to obtain complete velocity time history data.

4. The intelligent sea state compensation method based on machine learning according to claim 1, characterized in that, The intelligent compensation for the accumulated error generated during the integration process based on the machine learning model in S4 specifically includes the following steps: S4-1. Construct a feature vector containing wave buoy geometric scale, wave steepness parameter, and wave period; S4-2. Input the feature vectors into the pre-trained deep learning neural network model; S4-3. Obtain the error compensation coefficients output by the deep learning neural network model; S4-4. Correct the speed time history data using the compensation coefficient.

5. The intelligent sea state compensation method based on machine learning according to claim 1, characterized in that, The second integration of the compensated velocity data in step S5 specifically includes the following steps: S5-1. Displacement calculation is performed using the Gaussian integral method. S5-2. Map the integration interval to the [-1,1] interval through variable transformation; S5-3, Calculate using the five-point Gauss-Legend integral formula; S5-4. Obtain the time history data of the buoy's vertical displacement based on the integration results.

6. The intelligent sea state compensation method based on machine learning according to claim 5, characterized in that, When using the five-point Gauss-Legend integral formula in S5-3, the integration nodes are: T1=-0.9061798459,t2=-0.5384693101,t3=0, T4=0.5384693101,t5=0.9061798459; The corresponding quadrature coefficients are: λ1=0.2369268851, λ2=0.4786286705, λ3=0.5688888889, λ4=0.4786286705, λ5=0.2369268851.

7. The intelligent sea state compensation method based on machine learning according to claim 1, characterized in that, The wave parameters calculated based on displacement time history data in S6 specifically include the following steps: S6-1. Extract wave height sequence from displacement time history data; S6-2, Calculate the maximum wave height H max Significant wave height H 1 / 3 , average wave height H m ; S6-3, Calculate the effective wave height period T H1 / 3 Average period T m ; S6-4, Regarding the maximum wave height H max Significant wave height H 1 / 3 , average wave height H m Effective wave height period T H1 / 3 Average period T m The parameters are subjected to quality inspection and error assessment.

8. The intelligent sea state compensation method based on machine learning according to claim 1, characterized in that, S4 also includes a model optimization step: S4-5. Establish an error evaluation function to calculate the deviation between the measurement results before and after compensation and the true value; S4-6. Use optimization algorithms to adjust the parameters of the machine learning model; S4-7. Adjust the compensation strategy dynamically based on the optimization results.

9. The intelligent sea state compensation method based on machine learning according to claim 1, characterized in that, It also includes quality control steps: S7. Real-time monitoring of the signal-to-noise ratio of acceleration data; S8. When the signal-to-noise ratio is lower than the preset threshold, automatically increase the filtering strength; S9. Mark and remove abnormal fluctuation data; S10. Output measurement results and provide a data quality assessment report.

Images