A method for improving the accuracy of a wave numerical model

Abstract

The application provides a method for improving the precision of a wave numerical model, comprising the following steps: step A, establishing an improved ICAST model; step A1, trend similarity identification; step A2, value similarity identification; step B, wave period correction based on the ICAST model; and step C, wave height data correction based on the ICAST model. The application provides a brand-new similarity criterion algorithm, which simultaneously considers the numerical value and the trend information of a vector element, that is, the new similarity algorithm is a two-dimensional algorithm which takes into account the numerical value and the trend information. Compared with the original one-dimensional algorithm, the identification precision and the correction effect of the ICAST model are effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of natural environment simulation technology, specifically relating to a method for improving the accuracy of wave numerical models. Background Technology

[0002] In sea areas lacking long-term measured data, wave mathematical models are typically used to simulate multi-year wave data, which is then used as a sample for relevant research. Over the past thirty years, researchers have proposed various parameterization schemes to improve the accuracy of wave numerical model simulations. Currently, in the fields of wave forecasting and forecasting, research institutions widely use third-generation wave numerical models to calculate wave data; however, practical applications have revealed significant errors between the calculated results and actual values.

[0003] In 2005, Caires and Stell established a wave height data error correction model (CAST model) using a nonparametric method. This model is based on the assumption that "ocean wave height data has the characteristics of a conditionally stationary stochastic process" and believes that under similar "conditional states", wave height data have approximately consistent error characteristics.

[0004] The CAST model identifies whether two wave processes share a consistent "conditional state" by using a similarity criterion. If the wave data to be corrected and a certain wave process in the training set satisfy the similarity criterion, the error information corresponding to the training set can be used to modify the data to be corrected. The core of the CAST model is the "similarity criterion" algorithm. The main steps in establishing the CAST model are as follows:

[0005] (1) Constructing a training dataset: The training dataset is a crucial foundation for machine algorithms, determining the computational accuracy of the model. Generally, 2-3 years of data are sufficient. The training dataset for the CAST model consists of model data and experimental data, and can be written as follows: (in For model data, (For actual test data), the model data in the training set and the actual test data must maintain consistency in location and time. Extract the target vector sequentially from the training set. Simultaneously, synchronized measured data can be obtained. Where Δt is the data sampling interval, and the target vector consists of 3 elements.

[0006] (2) Construct a modified number set and extract a condition vector from it. The condition vector is composed of the modified number set in chronological order and consists of 3 elements. The extracted condition vector is: The sampling interval Δt must be consistent with the sampling interval of the target vector.

[0007] (3) Based on the similarity criterion algorithm, determine whether the condition vector and the target vector are similar. If the condition vector and the target vector are similar, the error of the last element of the condition vector can be obtained from the error of the last element of the target vector. The calculation yielded the following result. The CAST model establishes a similarity criterion based on the comparison of the magnitudes of the target vector and the condition vector. If the difference between the three elements in the two vectors is less than the smoothing parameter h... n This allows us to determine if two vectors are similar (see Equation 1). After traversing the entire target vector, the model completes the collection of error information and result correction according to Equation (2). These are elements of the target vector. These are elements of the condition vector. This is the final corrected wave height data. In equation (2), p indicates that there are p target vectors and condition vectors similar in the training set, x i Let x represent the geographical location of the data collection point (in latitude and longitude), and t represent the location of the data collection point. i t represents the data acquisition time.

[0008]

[0009]

[0010] The smoothness parameter h in the formula n The result is obtained according to formula (3), where c = 0.3, n = 700, and m = 3, indicating that the element lengths of both the target vector and the condition vector are 3. That is, the model uses the relevant information of 3 consecutive elements to determine whether the target vector and the condition vector are similar.

[0011] h n =(cn) α-1 logn 2 ) 1 / m (3)

[0012] Equations (1) to (3) are the main equations of the CAST model. The similarity algorithm of the CAST model determines whether two vectors are similar by comparing the magnitudes of three consecutive vector elements. If the difference between the three elements in the two vectors is less than the smoothing parameter, the two vectors are considered similar. After traversing the entire target vector, the model completes the collection of error information and result correction. The CAST model determines whether two vectors are similar by comparing the magnitudes of three consecutive vector elements. Since its similarity criterion is based on the magnitudes of the sample data, it is a one-dimensional similarity determination. In practical applications, it has been found that the one-dimensional similarity determination of the CAST model usually cannot correctly determine whether two processes are similar. Because it cannot properly identify whether the target vector and the condition vector are similar, the CAST model often exhibits over-correction, which further widens the gap between the modified wave data and the actual value. When used in engineering design, it is easy to over-design or under-design, which leads to a decrease in the economy and safety of the actual project. Summary of the Invention

[0013] To improve the accuracy of wave numerical models and reduce the uncertainty of model data, this invention proposes a method for improving the accuracy of wave numerical models, which adopts the following technical solution: A method for improving the accuracy of wave numerical models, comprising:

[0014] Step A: Establish an improved ICAST model;

[0015] Step A1: Trend similarity identification;

[0016] After obtaining the trend information of the target vector and the conditional vector respectively, according to the formula: To determine whether two vectors have the same trend, k is an adjustable parameter, dh m L The slope of the condition vector; The slope of the target vector;

[0017] Step A2: Value Similarity Identification

[0018] The values ​​of the condition vector and the target vector are considered to be consistent when the following formula is satisfied;

[0019] in For the target vector, It is a conditional vector;

[0020] Step B: Wave cycle correction based on the ICAST model;

[0021] Step C: Wave height data correction based on the ICAST model.

[0022] Furthermore, step A also includes:

[0023] Step A3, according to Obtain the absolute error value of the last element of the vector. If this error satisfies... If the error is determined to be an erroneous value, the ICAST model will not collect this error information. The measured value of the target vector. This represents the target vector model value.

[0024] Furthermore, the trend information of the conditional vector is determined according to the following formula:

[0025]

[0026] Target vector The trend information is determined by the following formula, where the subscript i indicates that the vector is the target vector extracted from the training set.

[0027]

[0028] Further, step C includes:

[0029] Step C1: Construct the training dataset;

[0030] Step C2: Obtain the correction coefficients for different points in the statistical sea area at various times throughout the year;

[0031] Step C3: Interpolate the correction coefficients over time and identify the validity of the model error;

[0032] Step C4: Correct the model wave height data sequence using correction coefficients with time process information generated by interpolation.

[0033] Compared with existing technologies, the advantages and positive effects of this invention are as follows: This invention proposes a novel similarity criterion algorithm that simultaneously considers the numerical magnitude of vector elements and the trend information of the vector process; that is, the new similarity algorithm is a two-dimensional algorithm that takes into account both numerical magnitude and process trend. Compared with the original one-dimensional algorithm, the recognition accuracy and correction effect of the ICAST model are effectively improved. Attached Figure Description

[0034] Figure 1 This is a statistical comparison chart of the correction results of the CAST model and the ICAST model for two buoy period data points;

[0035] Figure 2 This is a partial comparison chart of the correction effects of the CAST model and the ICAST model on data from two buoy cycles.

[0036] Figure 3This is a comparison chart (partial) of the wave height results of a certain buoy model after correction using the WHCAE method;

[0037] Figure 4 This is a comparison chart (partial) of the wave height results of a certain buoy model after correction using the WHCAE method;

[0038] Figure 5 This is a comparison chart (partial) of the wave height results of a certain buoy model after correction using the WHCAE method; Detailed Implementation

[0039] The applicant found that in the northern South China Sea, wave models driven by wind fields tend to underestimate typhoon waves, while in the southern South China Sea, the models tend to overestimate non-typhoon waves. Data obtained from wave mathematical models are not suitable for direct use in formulating engineering design standards, but the academic community currently lacks practical and reliable error correction techniques to effectively correct the model results.

[0040] Based on the above, the present invention proposes a method to improve the accuracy of wave numerical models. The present invention will be described in detail below with reference to specific embodiments.

[0041] I. Improved Model—ICAST Model

[0042] This invention proposes a novel two-dimensional similarity algorithm for the CAST model, which primarily uses magnitude as the criterion for comparison. This new algorithm considers both the development trend of the process and the similarity of the magnitude; that is, the similarity algorithm can only determine the similarity between the condition vector and the target vector when both the process trend and magnitude of the magnitudes are simultaneously satisfied. The data set structure of the ICAST model is consistent with that of the CAST model.

[0043] (1) Trend similarity identification

[0044] Assume there is a conditional vector The time interval of the sample data is Δt, and the vector length is m = 3. The trend information of the conditional vector is determined according to equation (4).

[0045]

[0046] Similarly, the target vector The trend information is determined by equation (5), where the subscript i indicates that the vector is the target vector extracted from the training set.

[0047]

[0048] After obtaining the trend information of the target vector and the conditional vector respectively, it is determined whether the trends of the two vectors are consistent according to equation (6), where dh m(m = 1, 2, 3) represents the difference between the slopes of the target vector and the conditional vector (which can be seen as the trend of vector change), dh m L and Calculated from equation (5), k is an adjustable parameter. The smaller the value, the stricter the similarity conditions. Multiple debugging sessions have shown that the ICAST model can achieve better results when k = 0.01. That is, only when all three elements in the vector simultaneously satisfy equation (6) are the trends of the condition vector and the target vector considered to be consistent.

[0049]

[0050] (2) Quantitative similarity identification

[0051] The CAST model uses the absolute difference in wave height values ​​of vector elements as the criterion for determining whether two vectors are similar in magnitude. The ICAST model uses the relative error rate of wave height (see Equation 7) as the criterion for determining whether vectors are similar in magnitude. That is, when the relative error rate of all elements of a vector is less than (r×100)%, the two vectors are considered similar. Multiple trials have shown that r = 0.01 is sufficient for general use. In the ICAST model proposed in this invention, when the difference between two element values ​​is within 1%, they are considered consistent. Only when all three conditions in Equation 7 are met can the magnitudes of the condition vector and the target vector be considered consistent. Equation 8 is the correspondence between the condition vector elements and the target vector elements derived from Equation 7.

[0052]

[0053]

[0054] (3) Error constraint parameters

[0055] After trend similarity identification (Equation 6) and magnitude similarity identification (Equation 7), the probability of model over-correction is significantly reduced, greatly improving the accuracy of vector similarity identification. To enable the ICAST model to distinguish between normal errors and model calculation errors (such as errors caused by missing wind fields), and to prevent the model from treating obvious errors as normal errors, an error constraint coefficient e is introduced into the ICAST model. s In other words, when the relative error between the wave model value and the measured value exceeds the error constraint coefficient, the ICAST model considers this error to be an erroneous value and does not collect this error information. By introducing error constraint parameters, the reliability of the model is enhanced.

[0056] Assuming that a target vector is identified as similar to a condition vector based on equations 6 and 7, then according to... Obtain the absolute error value of the last element of the vector. If this error satisfies... Then the error can be identified as an erroneous value, and Equation 2 will not collect the error (the ICAST model uses Equation 2 of the CAST model).

[0057] Experiments show that the error constraint coefficient e can be ignored when correcting the buoy wave period. s When correcting buoy wave height data, e s The correction result has a significant impact, and its value is closely related to the length of the training set. It can be calculated according to e. s =19.29×(n / 10) 4 ) -2.45 calculate.

[0058] The following section will correct the wave model period data and wave height data based on the ICAST model. Assume that a wave buoy B is located at P(x,y), and wave data of buoy B for many consecutive years has been obtained through the wave numerical model, including wave period and wave height data. The correction methods for wave period and wave height data will be introduced below.

[0059] II. Wave Period Correction Method Based on ICAST Model

[0060] (I) Constructing the dataset

[0061] The training dataset consists of 2-3 years of measured periodic data from buoy B and synchronized model data from the same location. It is necessary to ensure that the two sets of data are synchronized in time.

[0062] The correction set is the computational data of the model to be corrected. The length of this data is not limited by time, but it does not include the model data in the training set. The model data in the correction set must be computed by the same model in the same driving environment as the model data in the training set.

[0063] (II) ICAST Model Correction

[0064] The above equations (4)-(8) are used to make similarity judgments, collect error information, and correct the wave cycle of the model data.

[0065] III. Wave Height Data Correction Method Based on ICAST Model (WHCAE Method)

[0066] Practical applications show that using the ICAST model alone cannot effectively correct model wave height data. The WHCAE method is a wave height data correction method built on the ICAST model. This method performs time-domain interpolation on the wave height error of the satellite remote sensing synchronous model, and corrects the hourly wave height data of the model based on the error sequence formed after interpolation. This method can make full use of shared data distributed online to correct model wave height data for any sea area in the world.

[0067] (i) Extract 2-3 years of wave height data from the sea area surrounding buoy B with P(x,y) as the center within a certain latitude and longitude range (not more than 5.5 degrees) to construct a training set. Use global satellite remote sensing wave height data as the measured wave height. Extract synchronous calculation data from the wave model to construct a training set. Model data not included in the training set constitutes a correction set to be corrected.

[0068] (ii) The ICAST model was used to correct the model data, and the correction coefficients for different points in the statistical sea area at various times throughout the year were obtained based on the model data before and after correction. Based on the assumption that "the wave height error characteristics of the sea area within a certain range around the buoy are approximately consistent with the buoy data error characteristics", the correction coefficients at different locations in the statistical sea area can be regarded as the correction coefficients of buoy B at different times.

[0069] (III) Interpolate the correction coefficients over time and identify the validity of the model error. Assume that multiple correction coefficients are calculated at multiple times over several consecutive days (d-1, d, ..., d+n) in the sea area. Then in t1—t m The correction coefficient at any time T between the two points can be obtained by Equation 9, where F is the interpolation function (it is recommended to use the smooth averaging method). By appropriately increasing the sea area range for extracting model data (but not exceeding 5.5 degrees of latitude and longitude), it can be basically guaranteed that there are 1-2 different time periods for correction coefficients every day. Using the smooth averaging method to process the above coefficients can effectively avoid the impact of abnormal correction coefficients at individual times. According to the trial calculation, a smoothing period of 12 hours can achieve a better correction effect.

[0070]

[0071] To prevent the model from over-correcting the data, it is necessary to identify whether the data to be corrected truly needs correction. Let h be the model wave height at time T in the sea area P(x,y) where buoy B is located. s (T), the globally measured wave height at a certain location in the surrounding sea area is h. s alt (T), the model synchronizes global wave data as h s aw3 (T), calculate the wave height difference between them according to formula 10, when d a ×d b When >0, the model will target h. s (T) is corrected; otherwise, no correction is made, and the correction factor is adjusted to 1.

[0072]

[0073] (iv) Correct the model wave height data sequence at buoy B using correction coefficients with time process information generated by interpolation.

[0074] (V) Verification of Error Correction Effectiveness

[0075] Figures 1-5 A comparison chart of the periodic data and the corrected wave height data using the ICAST model is presented, including a comparison with the results of the CAST model. (By...) Figure 1 , Figure 2 It can be seen that the accuracy of the corrected model period is significantly improved, and the trend of the model period being too small is significantly improved. Compared with the uncorrected data, the accuracy of the corrected period data is improved by more than 50% overall. Figures 3-5 The results of correcting the buoy model wave height data are presented. Analysis of the wave height time-series curves shows that the corrected wave height curves agree more closely with the measured curves, improving wave height accuracy by approximately 10%. The ICAST model has the following advantages:

[0076] (1) High recognition accuracy

[0077] The original CAST model is relatively unstable and prone to over-correction. The ICAST model established in this invention has a significantly better overall correction effect than the original CAST model, with excellent correction results. The two-dimensional algorithm's recognition and correction capabilities are superior to the original model's one-dimensional recognition algorithm.

[0078] (2) The longer the training dataset duration, the better the correction effect of the ICAST model.

[0079] The ICAST model trained on a longer training set exhibits more stable correction results. When the training set is 15 years long, the corrected wave height curve is smoother than when the training set is 2 years long, and extreme wave heights are also corrected to varying degrees. Under suitable conditions, the length of the training set can be appropriately increased to achieve better correction results, but increasing the length of the training set will increase computation time.

[0080] (3) Strong spatiotemporal compatibility

[0081] The ICAST model has strong compatibility with training datasets from different periods, and the model can be trained and corrected using training datasets from different periods. The buoy synchronization model data can be corrected using buoy error information from nearby sea areas. Due to its strong spatiotemporal compatibility, the model has broad application prospects.

[0082] (4) It has strong practical application value

[0083] This invention can effectively correct the results of third-generation wave numerical models, improving the accuracy of wave period and wave height calculations. Wave data is a key input element for the design and construction of marine engineering projects; higher-precision wave data can enhance the economy and safety of various marine engineering projects. Furthermore, this invention also has significant application value in wave forecasting, significantly improving the forecast accuracy of wave period data.

[0084] The embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for improving the accuracy of wave numerical models, characterized in that... include: Step A: Establish an improved ICAST model; Step A1: Trend similarity identification; After obtaining the trend information of the target vector and the conditional vector respectively, according to the formula: To determine whether two vectors have the same trend, k is an adjustable parameter. The slope of the condition vector; The slope of the target vector; The trend information of the conditional vector is determined according to the following formula: Target vector The trend information is determined by the following formula, where the subscript i indicates that the vector is the target vector extracted from the training set. ； Step A2: Value Similarity Identification The values ​​of the condition vector and the target vector are considered to be consistent when the following formula is satisfied; ,in For the target vector, It is a conditional vector; Step B: Wave cycle correction based on the ICAST model; Step C: Wave height data correction based on the ICAST model.

2. The method for improving the accuracy of wave numerical models according to claim 1, characterized in that, Step A further includes: Step A3, according to Obtain the absolute error value of the last element of the vector. If this error satisfies... If the error is determined to be an erroneous value, the ICAST model will not collect this error information. and These are the elements of the target vector.

3. The method for improving the accuracy of wave numerical models according to claim 1, characterized in that, Step C includes: Step C1: Construct the training dataset; Step C2: Obtain the correction coefficients for different points in the statistical sea area at various times throughout the year; Step C3: Interpolate the correction coefficients over time and identify the validity of the model error; Step C4: Correct the model wave height data sequence using correction coefficients with time process information generated by interpolation.

Images