A non-stationary platform wind speed data cleaning and quality grading method and system

Abstract

The application discloses a non-stationary platform wind speed data cleaning and quality grading method and system, and the method comprises the following steps: obtaining wind speed observation values, platform attitude characteristic parameters and environmental dynamic characteristic parameters within a preset time window; calculating a comprehensive tilt angle based on the attitude parameters through vector composition operation; combining the environmental dynamic parameters to calculate a motion influence index according to a preset motion influence index model to quantize the motion interference degree; comparing the motion influence index with a quality grading threshold to determine the data quality grade, wherein the grade at least comprises a high-risk grade; and cleaning the data according to the quality grade, and executing elimination or marking on the high-risk data. According to the application, only low-frequency statistical characteristics are utilized, and high-frequency original motion sequences are not needed, so that the quality evaluation and grading cleaning of the non-stationary platform wind speed data can be realized, the hidden system deviation caused by platform motion can be effectively identified and processed, the calculation complexity is low, and the application is suitable for automatic offline batch processing of large-scale historical data.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for cleaning and quality grading wind speed data on non-stationary platforms. Background Technology

[0002] In the field of meteorological monitoring and data analysis, the accuracy of wind speed observation data has a decisive impact on key applications such as numerical weather prediction, extreme weather event analysis, and wind energy resource assessment. However, in actual observations, wind speed sensors are often deployed on non-stationary platforms, such as ocean buoys, vehicle-mounted or shipborne mobile observation stations, etc. These platforms experience continuous and complex attitude fluctuations under the influence of environmental dynamics such as waves and vibrations, leading to a systematic negative bias in the measurement results of mechanical wind speed sensors due to geometric projection effects and additional motion disturbances. The insidious nature of this error lies in the fact that its numerical value is often still within a reasonable physical range, but the physical authenticity of the data has been compromised, constituting a difficult-to-detect "hidden quality problem" that seriously threatens the reliability of subsequent data applications.

[0003] Currently, data quality control technologies for addressing such issues face three main limitations. First, there is a contradiction between algorithm dependence and data availability: effective real-time motion compensation methods typically rely on continuous, high-frequency raw attitude or acceleration sequences (such as data from IMU / INS). However, in many operational observation systems, limited by storage capacity or data transmission bandwidth, platforms often only store or report low-frequency statistical products (such as 10-minute or 30-minute average wind speed, average attitude angle, significant wave height, etc.), resulting in compensation algorithms based on high-frequency raw sequences being unable to be applied to offline processing of existing historical data and routine operational data due to a lack of necessary input. Second, the quality assessment models are too simplistic: existing methods are mostly limited to threshold checks based on the wind speed data itself (such as value range, abrupt change detection, etc.), primarily used to identify explicit numerical anomalies, and are insufficient to effectively diagnose and identify implicit system biases caused by platform attitude fluctuations. Finally, there are shortcomings in feature utilization: existing methods usually make simple judgments based on a single attitude index, failing to deeply integrate and collaboratively model environmental dynamic intensity features with platform attitude features, thus making it difficult to quantify the comprehensive reliability of observation data under complex and dynamic environmental conditions.

[0004] Therefore, there is an urgent need for a wind speed data quality assessment and cleaning method that can break free from dependence on high-frequency original motion sequences and can be implemented solely based on low-frequency statistical characteristics commonly found in business systems, in order to solve the problem of effectively identifying and processing hidden quality issues in observation data from non-stationary platforms, especially historical stock data. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for cleaning and grading the quality of wind speed data from non-stationary platforms. By integrating the low-frequency statistical characteristics of platform attitude and environmental dynamics, it achieves quantitative evaluation and grading of wind speed data quality from non-stationary platforms without relying on high-frequency original motion sequences, effectively solving the problem of identifying implicit system deviations caused by platform motion.

[0006] To address the aforementioned technical problems, a first aspect of this invention provides a method for cleaning and quality grading wind speed data from a non-stationary platform, comprising the following steps:

[0007] Acquire multi-source observation data within a preset time window, including wind speed observations, attitude characteristic parameters of the non-stationary platform, and environmental dynamic characteristic parameters;

[0008] Based on the attitude feature parameters, a comprehensive tilt angle is calculated through vector synthesis operation to characterize the overall attitude state of the non-stationary platform.

[0009] Based on the comprehensive tilt angle and the environmental dynamic characteristic parameters, the motion impact index is calculated according to the preset motion impact index model. The motion impact index is used to quantify the degree of interference of the non-stationary platform motion on the wind speed observation value.

[0010] The motion impact index is compared with a preset quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window. The quality level includes at least a high-risk level. The quality grading threshold is dynamically calculated based on the statistical quantile of the historical motion impact index set.

[0011] The wind speed observations are cleaned according to the quality level, and the cleaning process includes at least removing or marking the wind speed observations that belong to the high-risk level.

[0012] Furthermore, the calculation of the motion impact index based on the comprehensive tilt angle and the environmental dynamic characteristic parameters according to the preset motion impact index model includes:

[0013] The environmental dynamic characteristic parameters are obtained, including the significant wave height;

[0014] Based on the effective wave height and the comprehensive tilt angle, combined with the response sensitivity coefficient and normalized reference factor of the non-stationary platform, the motion influence index is calculated.

[0015] The response sensitivity coefficient and the normalized reference factor are preset values ​​or obtained based on historical data.

[0016] Furthermore, the formula for calculating the motion influence index is as follows:

[0017] ;

[0018] in, To take into account the tilt angle, These are environmental dynamic characteristic parameters. The response sensitivity coefficient is... is the normalized reference factor.

[0019] Furthermore, before calculating the motion influence index, the process further includes:

[0020] Obtain data from multiple historical time windows of the non-stationary platform within a historical time period;

[0021] For each historical time window, calculate the combined tilt angle and significant wave height within the historical time window;

[0022] A linear regression model is constructed using the effective wave height of each historical time window as the independent variable and the corresponding comprehensive tilt angle as the dependent variable.

[0023] The regression coefficients of the linear regression model are determined as the response sensitivity coefficients.

[0024] Further, the step of comparing the motion impact index with a preset quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window includes:

[0025] Obtain the set of motion influence indices of the non-stationary platform over a preset number of consecutive historical periods prior to the current time window;

[0026] Calculate the first quantile, second quantile, and third quantile of the motion influence index set, and set the first quantile, second quantile, and third quantile as the first mass threshold, second mass threshold, and third mass threshold, respectively.

[0027] The motion impact index of the current time window is compared sequentially with the first quality threshold, the second quality threshold, and the third quality threshold to determine the quality level to which the motion impact index of the current time window belongs.

[0028] Further, obtaining the set of motion influence indices of the non-stationary platform over a predetermined number of consecutive historical periods prior to the current time window includes:

[0029] Set the maximum and minimum backtracking time lengths;

[0030] Within the range defined by the minimum backtracking time length and the maximum backtracking time length, multiple candidate historical periods of different time lengths are generated by increasing the preset time length step.

[0031] The stability metric value of the motion influence index sequence in each of the candidate historical periods is calculated. The method for calculating the stability metric value is as follows: the coefficient of variation of the motion influence index sequence is calculated. The coefficient of variation is the ratio of the absolute value of the standard deviation of the motion influence index sequence to the absolute value of the mean. When the absolute value of the mean of the motion influence index sequence is less than a first preset threshold, the stability metric value is set to a preset maximum value.

[0032] Compare the stability metric values ​​of all the candidate historical periods, and determine the candidate historical period with the smallest stability metric value as the consecutive preset number of historical periods;

[0033] From the aforementioned consecutive preset number of historical periods, all motion impact indices are obtained to obtain the motion impact index set.

[0034] Further, the step of comparing the stability metric values ​​of all the candidate historical periods and determining the candidate historical period with the smallest stability metric value as the consecutive preset number of historical periods includes:

[0035] From all the candidate historical periods, the top N candidate historical periods with the smallest stability metric values ​​are selected to form the optimal candidate subset, where N is an integer greater than 1;

[0036] Calculate the absolute value of the first-order autocorrelation coefficient of the motion influence index sequence within each of the candidate historical periods in the optimal candidate subset;

[0037] From the optimal candidate subset, the candidate historical period with the smallest absolute value of the first-order autocorrelation coefficient is selected and determined as the consecutive preset number of historical periods.

[0038] Further, the calculation of the first quantile, second quantile, and third quantile values ​​of the motion influence index set includes:

[0039] The time decay weight value corresponding to each motion influence index in the set of motion influence indices is determined. The time decay weight value is calculated by a preset decay function model based on the time interval between its corresponding time window and the current time window. The smaller the time interval, the larger the time decay weight value.

[0040] The time decay weight value of each motion influence index is rounded down to obtain the corresponding representative frequency value;

[0041] Based on the representative frequency value, each motion influence index is repeated a corresponding number of times and combined to obtain an amplified motion influence index sequence;

[0042] The amplified motion influence index sequence is numerically sorted to obtain a sorted sequence.

[0043] Based on the total length of the sorted sequence, the first ordinal position, the second ordinal position, and the third ordinal position are calculated respectively.

[0044] The values ​​located at the first ordinal position, the second ordinal position, and the third ordinal position in the sorted sequence are respectively assigned the first quantile value, the second quantile value, and the third quantile value.

[0045] Further, obtaining the set of motion influence indices of the non-stationary platform over a predetermined number of consecutive historical periods prior to the current time window includes:

[0046] Obtain the environmental dynamic feature parameters and attitude feature parameters of the current time window as the current feature vector;

[0047] From the aforementioned consecutive preset number of historical periods, obtain the feature vectors of all historical time windows, the feature vectors including the environmental dynamic feature parameters and attitude feature parameters of the corresponding window;

[0048] The current feature vector and the feature vectors of all the historical time windows are normalized respectively. The normalization process includes standardizing each feature according to its mean and standard deviation in the historical period. Based on the normalized feature vector, the Euclidean distance between the feature vector of each historical time window and the current feature vector is calculated.

[0049] Based on the Euclidean distance, select the K closest historical time windows, where K is a positive integer;

[0050] Obtain the motion impact index corresponding to K of the historical time windows, and form the motion impact index set.

[0051] Accordingly, a second aspect of the present invention provides a system for cleaning and classifying wind speed data on non-stationary platforms, which performs wind speed data cleaning and quality classification based on the above-described method for cleaning and classifying wind speed data on non-stationary platforms, including:

[0052] The data acquisition module is used to acquire multi-source observation data within a preset time window. The multi-source observation data includes wind speed observations, attitude characteristic parameters of the non-stationary platform, and environmental dynamic characteristic parameters.

[0053] An angle calculation module is used to calculate, based on the attitude feature parameters, a comprehensive tilt angle that characterizes the overall attitude state of the non-stationary platform through vector synthesis operations.

[0054] The index calculation module is used to calculate the motion impact index based on the comprehensive tilt angle and the environmental dynamic characteristic parameters, according to a preset motion impact index model. The motion impact index is used to quantify the degree of interference of the non-stationary platform motion on the wind speed observation value.

[0055] The quality grading module is used to compare the motion impact index with a dynamically generated preset quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window. The quality level includes at least a high-risk level. The quality grading threshold is dynamically calculated based on the statistical quantile of the historical motion impact index set.

[0056] The data cleaning module is used to clean the wind speed observations according to the quality level. The cleaning process includes at least removing or marking the wind speed observations that belong to the high-risk level.

[0057] Accordingly, a third aspect of the present invention provides an electronic device, comprising: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the above-described non-stationary platform wind speed data cleaning and quality classification method.

[0058] Accordingly, a fourth aspect of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the above-described method for cleaning and quality grading wind speed data on non-stationary platforms.

[0059] The above-described technical solutions of the embodiments of the present invention have the following beneficial technical effects:

[0060] 1. By statistically fusing the platform's attitude characteristic parameters with environmental dynamic characteristic parameters and constructing the Motion Influence Index (MII), a quantitative evaluation model that does not rely on high-frequency original motion sequences was established. This model can effectively perceive and quantify the data quality degradation caused by the combined effects of complex platform motion and environmental disturbances, thereby significantly improving the detection capability of "hidden systematic biases" that are difficult to identify by traditional thresholding methods, and solving the long-standing problem of the difficulty in evaluating systematic errors in wind speed observation of non-stationary platforms.

[0061] 2. By introducing a dynamic quality threshold generation mechanism based on historical data statistical quantiles and combining it with an adaptive historical data selection strategy (such as based on stability or feature similarity), the quality grading standard is autonomously optimized and dynamically adjusted. The threshold is generated through a historical stability screening mechanism to avoid threshold distortion caused by short-term drastic fluctuations. This enables data quality assessment to adapt to different platform characteristics, diverse sea conditions, and long-term performance changes, outputting more refined, objective, and reliable quality grades (such as A, B, C, and D levels). This significantly improves the accuracy and practicality of the grading results and solves the problem of poor adaptability of fixed thresholds or single assessment models in complex environments.

[0062] 3. The overall technical solution is based entirely on low-frequency statistical features for calculation. The algorithm structure is clear, the computational complexity is low, and each module (such as feature extraction, index calculation, threshold determination, and hierarchical cleaning) forms a complete automated processing closed loop. It can efficiently perform offline batch processing on large-scale historical data or real-time business data without manual intervention, which greatly reduces the implementation threshold and computational resource consumption of data cleaning and solves the engineering practice bottleneck that existing compensation algorithms that rely on high-frequency data cannot be applied to the cleaning of business historical data.

[0063] 4. At the computer implementation level, the technical solution of the present invention has significant advantages: the processing of a single time window involves only a limited number of floating-point operations and array sorting operations, which has low requirements for processor computing power and memory capacity; the various links from data reading from memory, exponential calculation, threshold generation, quality grading to cleaning output form a pipeline-style automated processing link, which supports batch traversal processing of large-scale historical data in memory, and significantly improves the computational throughput efficiency and automation level of data cleaning. Attached Figure Description

[0064] Figure 1 This is a flowchart of the non-stationary platform wind speed data cleaning and quality classification method provided in the embodiments of the present invention;

[0065] Figure 2 This is a schematic diagram of the non-stationary platform wind speed data processing logic provided in an embodiment of the present invention;

[0066] Figure 3 This is a schematic diagram of the MII index calculation logic provided in an embodiment of the present invention;

[0067] Figure 4 This is a box plot showing the statistical distribution of the overall tilt angle under different quality levels provided in this embodiment of the invention;

[0068] Figure 5 This is a box plot showing the statistical distribution of theoretical wind speed deviations under different quality levels provided in this embodiment of the invention;

[0069] Figure 6This is a block diagram of a non-stationary platform wind speed data cleaning and quality grading system provided in an embodiment of the present invention.

[0070] Figure label:

[0071] 1. Data acquisition module; 2. Angle calculation module; 3. Index calculation module; 4. Quality grading module; 5. Data cleaning module. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0073] The method described in this embodiment of the invention can run on electronic devices equipped with a general-purpose processor (such as an x86 or ARM architecture CPU) and at least 4GB of memory. The operating system can be a general-purpose computing platform such as Linux or Windows, and the programming implementation can be done using languages ​​such as Python, MATLAB, or C++. Input data is read from local storage or a network database in standard formats such as CSV or NetCDF, and the output quality grading results are written back to storage in the same or compatible format as the cleaned data.

[0074] In the fields of meteorology, oceanography, and wind energy engineering, obtaining high-precision wind speed observation data is crucial for applications such as numerical weather prediction, extreme weather warning, climate research, and wind energy resource assessment. To achieve wide-area, continuous wind speed monitoring, wind speed sensors are often deployed on various non-stationary platforms, such as ocean buoys, shipboard observation systems, and mobile vehicle-mounted observation stations. These platforms, under the dynamic influence of the natural environment—such as wave undulations, ship swaying, or vehicle jolting—experience continuous and complex six-degree-of-freedom motion, causing dynamic changes in the measurement reference plane of the mechanical wind speed sensors mounted on them. Under these conditions, a geometric projection error occurs between the wind speed sensor's actual perceived wind direction and the combined direction of the platform's motion. Simultaneously, the sensor itself may be subject to additional acceleration interference, introducing a systematic negative bias into the observation data. This error introduced by platform motion is particularly insidious, as it often does not manifest as obvious out-of-range or abrupt anomalies. The data values ​​may still be within a reasonable physical range, but their physical accuracy has been compromised, forming a "hidden quality problem" that is difficult to effectively identify using conventional quality control methods. Furthermore, in many operational historical and real-time observation systems, limited by data storage costs and transmission bandwidth, only low-frequency statistical products (such as 10-minute or 30-minute averages) are typically stored or transmitted long-term. High-frequency raw attitude and acceleration sequences used for precise motion compensation are often not recorded, further limiting the possibility of effective quality backtracking and correction of existing data. Therefore, how to effectively assess and clean such hidden quality problems in non-stationary platform wind speed data, especially historical observation data, using limited low-frequency statistical features has become a prominent technical challenge for improving the usability of related data products and the reliability of research. To address these issues, this invention proposes a method and system for cleaning and quality grading non-stationary platform wind speed data.

[0075] Please refer to Figure 1 and Figure 2 The first aspect of this invention provides a method for cleaning and quality grading wind speed data from a non-stationary platform, comprising the following steps:

[0076] Step S100: Obtain multi-source observation data within a preset time window. The multi-source observation data includes wind speed observations, attitude characteristic parameters of the non-stationary platform, and environmental dynamic characteristic parameters.

[0077] This invention primarily targets time-series data collected by non-stationary observation platforms such as ocean buoys, research vessels, or mobile observation stations. In practical operational systems, data is typically stored and processed in fixed time windows (e.g., 30 minutes). Multi-source observation data records within a preset time window are retrieved from the storage system or data stream. These records include at least the observed wind speed values ​​(e.g., average or maximum wind speed), attitude characteristic parameters representing the platform's spatial orientation (usually statistical values ​​of pitch and roll angles, such as average values), and environmental dynamic characteristic parameters representing the intensity of external environmental disturbances (usually significant wave height in marine scenarios). These low-frequency statistical features are common and long-term available data formats in operational observation systems, forming the foundational data source for subsequent offline quality assessments.

[0078] Step S200: Based on the attitude feature parameters, the comprehensive tilt angle used to characterize the overall attitude state of the non-stationary platform is calculated through vector synthesis operation.

[0079] The platform's two-dimensional attitude information is synthesized into a scalar index that comprehensively reflects its overall deviation from the horizontal plane. Specifically, the processor reads the average pitch and roll angle values ​​within the current time window and obtains the composite tilt angle through vector synthesis operations (e.g., calculating the square root of the sum of the squares of the two angles). This angle has a clear physical meaning, representing the average comprehensive deviation of the platform's reference plane from the horizontal plane within that time window. It more comprehensively characterizes the platform's geometric attitude than using pitch or roll angles alone, providing a unified and measurable input variable for subsequent quantification of the potential impact of attitude on wind speed measurements.

[0080] Step S300: Based on the comprehensive tilt angle and environmental dynamic characteristic parameters, the motion impact index is calculated. The motion impact index is used to quantify the degree of interference of the non-stationary platform motion on the wind speed observation.

[0081] The purpose of calculating the Motion Influence Index (MII) is to establish a composite model that integrates the platform's static attitude and environmental dynamic disturbances. This model quantifies the interference level caused by the combined effect of these two factors on wind speed observations. The calculation is based on the composite tilt angle obtained in step S200 and the environmental dynamic characteristic parameters (such as significant wave height) acquired in step S100, combined with a response sensitivity coefficient and a normalized reference factor pre-calibrated for a specific platform type. A typical implementation uses the formula MII = Composite Tilt Angle × (1 + Response Sensitivity Coefficient × Significant Wave Height / Normalized Reference Factor). The basic principle of this model is that the platform's composite tilt angle constitutes the baseline component of the disturbance, while the environmental dynamic parameters (significant wave height) modulate the baseline disturbance through an amplification factor (determined by the response sensitivity coefficient and the normalized reference factor), thereby simulating the physical phenomenon that the same attitude angle may cause more severe data distortion under adverse sea conditions.

[0082] Step S400: Compare the motion impact index with the preset quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window. The quality level includes at least the high-risk level. The quality grading threshold is dynamically calculated based on the statistical quantile of the historical motion impact index set.

[0083] The current wind speed observation data is classified into quality levels based on the calculated motion impact index. The preset quality classification thresholds are not fixed values ​​but are dynamically generated based on the statistical distribution (e.g., quantiles) of historical motion impact index data, thus forming a quality judgment standard that adapts to the current data characteristics. In practice, the motion impact index of the current time window is compared with a set of dynamic thresholds (e.g., first, second, and third quality thresholds). The wind speed data within that window is determined to belong to a specific quality level based on the range in which it falls. The classification typically includes high confidence, good, marginally usable, and high-risk levels, forming a quality spectrum from best to worst, achieving a refined distinction in data reliability.

[0084] Step S500: Clean the wind speed observations according to the quality level. The cleaning process includes at least removing or marking the wind speed observations that belong to the high-risk level.

[0085] For data identified as high-risk, the system will execute a pre-defined cleaning strategy, such as directly removing wind speed observations for that time window from the valid data sequence or adding a prominent quality warning mark to its data record. Data of other risk levels will be retained and processed or recorded differently according to their risk level. Ultimately, the system outputs a cleaned and labeled wind speed data sequence with controllable quality. This sequence can be directly used for subsequent scientific analysis or operational applications such as wind energy resource assessment and numerical weather prediction assimilation, effectively improving the overall reliability and application value of the dataset.

[0086] This invention innovatively integrates low-frequency attitude statistics of non-stationary platforms with environmental dynamic parameters to construct a motion impact index model, achieving quantitative assessment and adaptive grading of wind speed observation data quality without relying on high-frequency raw motion data. This method effectively identifies implicit systematic biases caused by complex platform motions that are difficult to detect using traditional thresholding methods, and adapts to different platform and sea state conditions through a dynamic thresholding mechanism. The entire scheme boasts high computational efficiency and automated processes, making it particularly suitable for batch, offline quality cleaning and control of large-scale historical archive data and routine operational observation data, significantly improving the usability and reliability of wind speed data products from non-stationary platforms.

[0087] Specifically, in step S300, the motion impact index is calculated based on the comprehensive tilt angle and environmental dynamic characteristic parameters according to the preset motion impact index model, including:

[0088] Step S310: Obtain environmental dynamic characteristic parameters, including significant wave height.

[0089] In non-stationary platform applications in oceans, lakes, or other water bodies, the environmental dynamic characteristic parameter is typically the significant wave height (SWH). SWH is a commonly used physical quantity in oceanography and engineering to describe the statistical characteristics of random wave heights. It reflects the average energy state of sea surface waves and is the primary environmental driving force causing complex motions such as heave and roll of buoys and ships. In data processing practice, this value is usually measured and calculated by wave sensors on the platform, directly obtained as a low-frequency statistical product (e.g., the average significant wave height every 30 minutes) recorded synchronously with wind speed observations. Choosing SWHH as the environmental dynamic characteristic parameter allows the complex, time-varying intensity of environmental disturbances to be introduced into subsequent quality assessment models in a stable and measurable scalar form, thereby establishing a correlation between environmental conditions and potential risks to data quality.

[0090] Step S320: Based on the significant wave height and the overall tilt angle, combined with the response sensitivity coefficient and normalized reference factor of the non-stationary platform, the motion influence index is calculated. The response sensitivity coefficient and normalized reference factor are preset values ​​or obtained based on historical data.

[0091] The core of the calculation process for the Motion Influence Index (MII) lies in establishing a system that integrates the platform's static attitude deviation (derived from the combined tilt angle). Characterization) and environmental dynamic disturbances (based on effective wave height H) s A quantitative model for characterization is proposed; this model refines the fusion relationship by introducing two key parameters: the response sensitivity coefficient k and the normalized reference factor. The response sensitivity coefficient k is used to quantify the intensity of the attitude response of a specific type of platform (such as a certain model of buoy) to changes in environmental dynamics. A larger value indicates that the same environmental dynamic change (such as an increase in wave height) may lead to more drastic changes in platform attitude or a greater amplification effect on wind speed measurements. Normalized reference factor. It is mainly used to perform dimensionless processing of significant wave height. Its physical meaning can be understood as the typical wave height background value or design tolerance threshold of the sea area where the platform is located, and it is used to calibrate the calculation scale under different sea areas or different wave height levels.

[0092] Parameter k and This can be achieved by presetting platform design parameters, or more commonly by performing offline statistical analysis on the platform's historical observation data (e.g., calibrating k through regression analysis using historical wave height and attitude angle data, and determining k based on long-term wave height distribution). This is obtained by using [methods], thus enabling the model to adapt to specific platforms and deployment locations.

[0093] like Figure 3 As shown, in a specific embodiment of the present invention, the calculation formula for the above-mentioned motion influence index is as follows:

[0094] .

[0095] in, To take into account the tilt angle, These are environmental dynamic characteristic parameters. For the response sensitivity coefficient, This is the normalization reference factor.

[0096] The Motion Impact Index (MII) calculation formula integrates the combined effects of static attitude deviation and dynamic environmental disturbances on wind speed measurement quality through a concise mathematical form. The first term in the formula is θ. tiltThe overall tilt angle represents the degree to which the platform's average spatial attitude deviates from its horizontal state within the observation window, constituting the "fundamental component" or "static component" of motion disturbance. Its physical meaning is that even in calm sea conditions (weak environmental dynamics), the platform's inherent tilt attitude will directly cause a deviation between the apparent wind speed measured by the wind speed sensor and the actual wind speed due to geometric projection effects. (The part within the parentheses in the formula...) This then acts as an "environmental dynamic modulation factor".

[0097] The entire formula uses a product form, reflecting the coupling relationship between "basic attitude disturbance" and "environmental dynamic amplification effect." When the environmental dynamics are weak ( When the modulation factor approaches 0, the modulation factor approaches 1, at which point the MII is mainly determined by the overall tilt angle. Decision. With the enhancement of environmental dynamics ( As the modulation factor increases, it becomes greater than 1. Under the same platform attitude angle, severe sea conditions will cause the calculated MII value to increase significantly, reflecting the physical reality that "the risk is different under the same attitude but different sea conditions".

[0098] In one specific embodiment of the present invention, before calculating the motion influence index in step S320, a data-driven calibration process for the response sensitivity coefficient k is also included:

[0099] Step S321: Obtain data from multiple historical time windows of the non-stationary platform within a historical time period.

[0100] In practice, a representative continuous historical period needs to be selected from the long-term observation archives of non-stationary platforms. This period should cover various typical environmental conditions that the platform may experience (such as sea states ranging from calm to severe) to ensure the robustness of the calibration results. The acquired data consists of records from multiple historical time windows arranged chronologically. Each record should at least include the platform's average attitude angles (pitch and roll angles) and the average significant wave height observed simultaneously within that time window.

[0101] Step S322: For each historical time window, calculate the combined tilt angle and effective wave height within the historical time window.

[0102] First, based on the average pitch and roll angles of each window, the historical composite tilt angle for that window is calculated using the same vector synthesis method as in online processing. Simultaneously, the average significant wave height recorded for that window is directly read. Through this step, a pair of corresponding observations is generated for each historical time window: the significant wave height as a potential causal variable. , and the overall tilt angle as the outcome variable .

[0103] Step S323: Construct a linear regression model with the effective wave height of each historical time window as the independent variable and the corresponding comprehensive tilt angle as the dependent variable.

[0104] All historical data pairs obtained in step S322 ( , Using the effective wave height as a sample Let X be the independent variable, and let the overall tilt angle be the inclination angle. For the dependent variable (Y), construct a univariate linear regression model. This model aims to fit a regression model of the form... The optimal straight line is given by , where a is the intercept and k is the desired regression coefficient (slope). Its statistical meaning is that, within the observed data range, for every unit increase in significant wave height, the average expected change in the platform's overall tilt angle is k units. This directly characterizes the average response intensity of the platform's attitude to changes in environmental dynamics.

[0105] Step S324: Determine the regression coefficients of the linear regression model as the response sensitivity coefficients.

[0106] After completing the linear regression, the regression coefficients (i.e., the slope k) of the established model are extracted. These statistically significant regression coefficients k are directly adopted as the response sensitivity coefficient k required in the Motion Impact Index (MII) calculation formula. The k value determined in this way is not an arbitrary preset value, but rather an empirical parameter reflecting the specific response characteristics of the platform, mined from its historical behavioral data. This data-based calibration method allows the Motion Impact Index model to adapt to the dynamic characteristics of a specific platform. For example, a smaller buoy, more sensitive to waves, might be calibrated with a larger k value, while a more stable platform would be calibrated with a smaller k value, thus achieving personalization and optimization of the model parameters.

[0107] The dynamic calculation method for the response sensitivity coefficient k transforms k from a static value that needs to be pre-set based on experience or depends on platform design parameters into a dynamic calibration value learned from the platform's own historical behavior through a data-driven approach. First, the dynamic calculation method significantly improves the adaptability and accuracy of the motion impact index model. Through linear regression analysis based on historical platform data, the obtained k value truly reflects the average response characteristics of the specific platform in the real marine environment, enabling the final calculated motion impact index to more accurately characterize the platform's disturbance risk in the current environment, achieving a leap from "general approximation" to "individual adaptation" in the model. Second, this method perfectly suits the application scenarios of offline processing and historical data cleaning. The entire calibration process relies entirely on low-frequency statistical characteristics (historical average wave height and attitude angle) commonly stored in the business system, without requiring any additional high-frequency motion sensor raw data. This allows advanced, parameter-personalized quality assessment models to be applied to the batch processing of existing historical data, solving the technical bottleneck of not being able to use refined models due to data source limitations. Finally, this method lowers the technical threshold for model application and enhances the objectivity of the results; it avoids the subjectivity and uncertainty of manually preset parameters, and automatically generates key parameters through objective statistical calculations, making the entire quality assessment scheme easier to standardize and verify on non-stationary platforms of different types and sea areas, thereby improving the repeatability and engineering practical value of the method.

[0108] Specifically, step S400 compares the motion impact index with a preset quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window, including:

[0109] Step S410: Obtain the set of motion impact indices of the non-stationary platform over a preset number of historical periods prior to the current time window.

[0110] A reasonable and continuous historical review period is determined to collect the "motion impact index set" that forms the basis for threshold calculation. Here, "a predetermined number of continuous historical periods" typically refers to a closely adjacent and continuous time span preceding the current evaluation window, such as the MII values ​​for all time windows in the past 24 hours or the past 7 days. This design ensures that the data used for statistics has temporal proximity to the current state and potential correlation with the environmental conditions. This allows the generated threshold to promptly reflect the typical environmental disturbance levels and data quality distribution characteristics of the platform in recent times, avoiding the inappropriateness problems that may arise from using fixed thresholds or excessively long historical data.

[0111] Step S420: Calculate the first quantile, second quantile, and third quantile values ​​of the motion influence index set, and set the first quantile, second quantile, and third quantile values ​​as the first mass threshold, second mass threshold, and third mass threshold, respectively.

[0112] Specifically, statistical quantiles are calculated for the set of motion impact indices obtained in step S410. Typically, the first quantile (e.g., the 25th quantile), the second quantile (e.g., the 50th quantile, i.e., the median), and the third quantile (e.g., the 75th quantile) are calculated separately. These three quantiles are then set as the first, second, and third quality thresholds, respectively. The principle behind this approach is that quantiles themselves describe the internal structure of historical data distribution: for example, MII values ​​below the 25th quantile represent "excellent" data with the lowest level of interference in history, while values ​​above the 75th quantile represent "poor" data with a higher level of interference. Converting these statistics into dynamic thresholds means that the quality level classification criteria are not fixed but automatically adjust as the historical data distribution changes (e.g., different seasons, different sea state periods), thus achieving an adaptive classification standard to the characteristics of the data itself.

[0113] Step S430: Compare the motion impact index of the current time window with the first quality threshold, the second quality threshold, and the third quality threshold in sequence to determine the quality level to which the motion impact index of the current time window belongs.

[0114] The motion impact index calculated for the current time window is compared sequentially with three quality thresholds dynamically generated in step S420. Based on the numerical range it falls into (e.g., less than or equal to the first quality threshold, between the first and second thresholds, between the second and third thresholds, or greater than the third quality threshold), it is mapped to a preset discrete quality level. Typically, these levels are defined as high confidence level, good level, marginally usable level, and high-risk level. This determination process is a clear decision logic that transforms continuous MII indicators into category labels with explicit quality semantics. For example, data classified as "high-risk" means that its degree of motion interference exceeds the majority of historical cases (e.g., 75%), its reliability is questionable, and it needs to be handled more carefully in subsequent cleaning.

[0115] In one embodiment of the present invention, step S410, obtaining the set of motion influence indices of the non-stationary platform over a predetermined number of consecutive historical periods prior to the current time window, includes:

[0116] Step S411a: Set the maximum backtracking time length and the minimum backtracking time length.

[0117] In practice, two time length parameters need to be predefined: the maximum backtracking time length and the minimum backtracking time length. The maximum backtracking time length specifies the upper limit of the range of historical data that the algorithm can search, to prevent the use of outdated data that may not reflect the current environmental state; the minimum backtracking time length sets a lower limit to ensure that the amount of data used for statistics has a basic sample size, avoiding unreliable statistical results due to insufficient data. These two parameters together define a reasonable selection range. For example, the minimum backtracking time length can be set to 6 hours, and the maximum backtracking time length can be set to 72 hours. The algorithm will search for the optimal historical review duration within this range.

[0118] Step S412a: Within the range defined by the minimum backtracking time length and the maximum backtracking time length, multiple candidate historical periods with different time lengths are generated by increasing the preset time length step.

[0119] Based on a preset time step (e.g., 6 hours), starting from the minimum backtracking time, the duration is gradually increased until the maximum backtracking time is reached, thus generating a series of candidate historical periods with different durations. For example, if the minimum is 6 hours, the maximum is 72 hours, and the step size is 6 hours, then multiple candidate historical periods with durations of 6, 12, 18, ..., 72 hours will be generated. Each candidate historical period refers to a continuous historical time segment that ends before the current processing time window and backtracks by the corresponding duration.

[0120] Step S413a: Calculate the stability metric of the motion influence index sequence for each candidate historical period. The stability metric is calculated by calculating the coefficient of variation of the motion influence index sequence. The coefficient of variation is the ratio of the absolute value of the standard deviation of the motion influence index sequence to the absolute value of the mean. When the absolute value of the mean of the motion influence index sequence is less than a first preset threshold, the stability metric is set to a preset maximum value.

[0121] For each candidate historical period, all Motion Influence Indices (MIIs) contained within it are extracted to form a time series. The coefficient of variation (COP) of this series is calculated as its stability measure. The COP is defined as the absolute value of the series' standard deviation divided by its mean. It is a dimensionless statistic that effectively measures the dispersion (fluctuation) of data relative to its average level. The smaller the COP, the more stable the MII values ​​are within that period, and the less volatile they are. A special processing rule is set: when the absolute value of the series mean is less than a very small first preset threshold (close to zero), it means that the series mean is almost zero. In this case, calculating the COP may result in a maximum value or instability due to the small denominator. To standardize the processing, in this case, a preset maximum value (e.g., 10^6) is directly assigned to the stability measure of the candidate period, thereby automatically excluding it from the selection of the "most stable" period in subsequent comparisons.

[0122] Step S414a: Compare the stability metric values ​​of all candidate historical periods, and determine the candidate historical period with the smallest stability metric value as a consecutive preset number of historical periods.

[0123] After calculating the stability metric values ​​for all candidate historical periods, these values ​​are compared, and the candidate historical period with the smallest stability metric value is selected. According to the definition in step S413a, this means selecting the historical period among all candidate periods whose internal motion influence index sequence has the smallest relative fluctuation and is the most stable. This period is ultimately determined as the "continuous preset number of historical periods" used to calculate the quality threshold. The rationale behind this selection is that a historical period with relatively small fluctuations in internal disturbance levels has a relatively concentrated and stable data quality distribution, and the statistical quantiles (quality thresholds) calculated based on the data from this period are likely to be more representative and reliable.

[0124] Step S415a: Obtain all motion impact indices from a preset number of consecutive historical periods to obtain a set of motion impact indices.

[0125] After determining the optimal historical period in step S414a, the Motion Influence Index (MII) for each window is extracted from all time window records corresponding to that period. These MII values ​​are then aggregated to form the "Motion Influence Index Set" used in the subsequent step S420 to calculate the dynamic quality threshold.

[0126] Furthermore, step S414a, which compares the stability metric values ​​of all candidate historical periods and determines the candidate historical period with the smallest stability metric value as a consecutive preset number of historical periods, includes:

[0127] Step S414a1: Select the top N candidate historical periods with the smallest stability metric values ​​from all candidate historical periods to form the optimal candidate subset, where N is an integer greater than 1.

[0128] In practice, all candidate historical periods are first sorted according to their stability metrics (i.e., coefficient of variation). Then, the top N periods with the smallest values ​​(e.g., the top 3 or 5) are selected to form the "optimal candidate subset." The parameter N is an integer greater than 1, which ensures that the subset contains multiple candidates that perform best and at similar levels on the primary stability metric. In real-world data, multiple periods may simultaneously exhibit low and minimal data volatility. Relying solely on a single stability metric may not be sufficient to determine a unique optimal candidate, thus providing a reasonable input for subsequent, more refined secondary screening.

[0129] Step S414a2: Calculate the absolute value of the first-order autocorrelation coefficient of the motion influence index sequence in each candidate historical period within the optimal candidate subset.

[0130] For each candidate historical period in the "optimal candidate subset," its motion influence index sequence is extracted, and the absolute value of the first-order autocorrelation coefficient of this sequence is calculated. The first-order autocorrelation coefficient measures the degree of linear correlation between adjacent data points in a time series; the closer its absolute value is to 0, the weaker the numerical correlation between consecutive moments in the sequence, i.e., the lower the short-term memory of the sequence, the smaller its inherent temporal dependence, and its fluctuations are closer to random noise. Calculating the absolute value is to focus on the strength of the correlation while ignoring its positive or negative direction. Statistically, this identifies historical periods that not only have small overall fluctuations but also have a "purer" internal data structure and are less affected by potential short-term trends or periodic patterns.

[0131] Step S414a3: Select the candidate historical period with the smallest absolute value of the first-order autocorrelation coefficient from the optimal candidate subset and determine it as a continuous preset number of historical periods.

[0132] After calculating the absolute value of the first-order autocorrelation coefficient for each period in the "optimal candidate subset," these absolute values ​​are compared, and the candidate historical period with the smallest absolute value of the first-order autocorrelation coefficient is selected. The finally selected historical period is the one in the group with the best primary stability (low coefficient of variation) whose data sequence has the weakest intrinsic temporal correlation and is closest to random independence. This period is ultimately determined as the "continuous preset number of historical periods" used for subsequent threshold calculation. The rationale for this selection is that a sequence with small internal fluctuations and weak temporal dependence may have more robust and general statistical distribution characteristics (such as quantiles), and a lower risk of being "hijacked" by specific historical fluctuation patterns. Therefore, the dynamic quality threshold calculated based on it may be more universal and have more reference value.

[0133] In another embodiment of the present invention, step S410, obtaining the set of motion influence indices of the non-stationary platform over a predetermined number of consecutive historical periods prior to the current time window, includes:

[0134] Step S411b: Obtain the environmental dynamic feature parameters and attitude feature parameters of the current time window as the current feature vector.

[0135] In practice, the environmental dynamic characteristic parameters (such as significant wave height) and attitude characteristic parameters (such as pitch angle and roll angle) of the current time window to be evaluated are extracted, and these two parameters are combined to form a two-dimensional or three-dimensional current feature vector. This vector is a point in mathematical space, uniquely representing the combined state of "environmental conditions-platform attitude" corresponding to the current time window. For example, a feature vector may be represented as [significant wave height = 2.5 meters, combined tilt angle = 8 degrees].

[0136] Step S412b: Obtain feature vectors for all historical time windows within a consecutive preset number of historical periods. The feature vectors include environmental dynamic feature parameters and attitude feature parameters for the corresponding window.

[0137] According to the overall requirements of step S410, a "continuous preset number of historical periods" (e.g., 24 hours before the current time window) is first determined. Then, within this historical period, for each historical time window (e.g., each 30-minute window), the same operation as in step S411b is performed: the environmental dynamic feature parameters and attitude feature parameters corresponding to the historical window are extracted, and its historical feature vector is constructed. Finally, a set consisting of the feature vectors corresponding to all time windows within this historical period is obtained.

[0138] Step S413b: Normalize the current feature vector and the feature vectors of all historical time windows respectively.

[0139] Normalization involves standardizing each feature dimension according to its mean and standard deviation over the stated historical period. Normalization addresses the imbalance in weighting of features with different physical dimensions (e.g., significant wave height in meters, attitude angle in degrees) in distance calculations, ensuring that each feature dimension has equal contribution weight in subsequent distance measurements.

[0140] Step S414b: Based on the normalized feature vector, calculate the Euclidean distance between the feature vector of each historical time window and the current feature vector.

[0141] For each historical time window feature vector obtained in step S412b, the Euclidean distance between it and the current feature vector constructed in step S411b is calculated. Euclidean distance is a classic method for measuring the straight-line distance between two points in multidimensional space; the smaller the value, the more similar the "environment-pose" states represented by the two vectors. Through this calculation, each historical window within a historical period is assigned a numerical value, which quantitatively expresses the degree of similarity between the state of that historical window and the state of the current window.

[0142] Step S415b: Based on the Euclidean distance, select the K closest historical time windows, where K is a positive integer.

[0143] The Euclidean distances of all historical time windows calculated in step S414b are sorted, and then the K historical time windows with the smallest Euclidean distances are selected. The parameter K is a preset positive integer (e.g., K=20), and its size determines the number of historical samples included in subsequent calculations. The logic behind this operation is that the historical data moments most similar to the current observation moment in terms of environmental conditions and platform attitude are likely to have the most similar motion disturbance characteristics. Therefore, the empirical distribution formed by data from these "most similar states" may have the highest reference value for assessing the quality level of the current data.

[0144] Step S416b: Obtain the motion impact index corresponding to K historical time windows to form a motion impact index set.

[0145] After determining the K most similar historical time windows in step S415b, the pre-calculated motion impact indices (MIIs) for each of these K windows are retrieved from the data storage. These K MII values ​​are then aggregated to form a "motion impact index set" used in step S420 to calculate the dynamic quality threshold. This set does not originate from a continuous period of time, but rather from multiple discrete moments in history that are most similar to the current state.

[0146] Further, the calculation of the first quantile, second quantile, and third quantile values ​​of the motion influence index set in step S420 includes:

[0147] Step S421: Determine the time decay weight value corresponding to each motion influence index in the motion influence index set. The time decay weight value is calculated by using a preset decay function model based on the time interval between its corresponding time window and the current time window. The smaller the time interval, the larger the time decay weight value.

[0148] In practice, for each motion influence index in the set, the time interval (e.g., the number of hours or days) between its corresponding original observation time window and the current time window to be evaluated is first determined. Then, a time decay weight value is calculated based on this time interval using a pre-defined decay function model. The design of this function model follows a core principle: the smaller the time interval, i.e., the closer the historical data point is to the current moment, the larger the calculated weight value. A common implementation uses an exponential decay function, where the weight value is proportional to a negative exponential function of the time interval, thus making the weight of recent data significantly higher than that of distant data. By introducing the concept of "data timeliness," it is reflected that in non-stationary dynamic environments, recent observational experience usually has higher reference value for assessing the current state.

[0149] Step S422: Round the time decay weight value of each motion influence index to obtain the corresponding representative frequency value.

[0150] The time decay weight value (usually a floating-point number) for each data point calculated in step S421 is rounded. Common rounding methods include rounding to the nearest integer and rounding up. The rounded value yields the representative frequency value for each data point, which is a non-negative integer (usually a positive integer). For example, a data point with a weight value of 2.3, after rounding, has a representative frequency value of 2. This value means that in the subsequent construction of the statistical distribution, this original data point will be considered to have occurred "times".

[0151] Step S423: Based on the representative frequency value, each motion influence index is repeated a corresponding number of times and merged to obtain an amplified motion influence index sequence.

[0152] Each original data point in the Motion Influence Index (MII) set is repeated based on its representative frequency value. For example, a data point with an MII of 5.2 and a representative frequency of 3 will appear three times consecutively in the new sequence. After repeating all original data points according to this rule, all the repeated values ​​are merged sequentially to form a new, longer, amplified MII sequence. This new sequence is characterized by more recent data (with higher representative frequencies due to greater time decay weights) occupying more "seats," while older data occupy fewer "seats," thus achieving a time-weighted effect in the data structure.

[0153] Step S424: Numerically sort the amplified motion influence index sequence to obtain the sorted sequence.

[0154] The amplified motion influence index sequence obtained in step S423 is numerically sorted, usually in ascending order, resulting in a sorted sequence arranged from minimum to maximum value. This sorted sequence forms the basis for calculating any ordinal-based statistics, such as quantiles.

[0155] Step S425: Based on the total length of the sorted sequence, calculate the first ordinal position, the second ordinal position, and the third ordinal position respectively.

[0156] First, obtain the total length L of the sorted sequence (i.e., the number of all values ​​in the sequence). Then, calculate the corresponding ordinal positions according to the required quantile definition. For example, to calculate the first quantile (25th quantile), its ordinal position index is typically 0.25 * L (rounding or interpolation may be required depending on the specific quantile definition). Similarly, calculate the ordinal positions of the second quantile (median, 50th quantile) and the third quantile (75th quantile). These ordinal positions indicate which (or which) positions in the sorted sequence will be used as quantile values.

[0157] Step S426: Assign the values ​​located at the first ordinal position, the second ordinal position, and the third ordinal position in the sorted sequence to the first quantile value, the second quantile value, and the third quantile value, respectively.

[0158] Based on the first, second, and third ordinal positions calculated in step S425, the specific values ​​located at these positions are found in the sorted sequence obtained in step S424. These values ​​are then extracted and assigned the values ​​as the first, second, and third quantiles, respectively. Since the sorted sequence is time-weighted, the quantiles calculated based on it naturally give greater influence to recent historical data, allowing the final quality threshold to more sensitively reflect the recent data quality distribution characteristics.

[0159] like Figure 4 The figure shows a box plot of the statistical distribution of the comprehensive tilt angle under different quality levels provided in this embodiment of the invention. The horizontal axis represents the quality level, corresponding to four levels: A (Excellent), B (Good), C (Questionable), and D (High Risk); the vertical axis represents the comprehensive tilt angle (unit: degrees). As can be seen from the figure, the median comprehensive tilt angle for level A data is approximately 5°, with the overall distribution concentrated at a low level; the median comprehensive tilt angle for level B data is approximately 10°, with a wider distribution range; the median comprehensive tilt angle for level C data is approximately 16°, with a significantly increased interquartile range; and the median comprehensive tilt angle for level D (High Risk) data is approximately 22°, with the widest distribution range, and the upper edge approaching 35°. These distribution results demonstrate that the quality grading method proposed in this invention can effectively distinguish data under different levels of attitude interference. The comprehensive tilt angle shows a progressively increasing trend from level A to level D, verifying that the motion influence index and its grading threshold have good distinguishing ability for the degree of platform attitude deviation.

[0160] like Figure 5The figure shows a box plot of the statistical distribution of theoretical wind speed deviations under different quality levels provided in this embodiment of the invention. The horizontal axis represents the quality level, corresponding to four levels: A (Excellent), B (Good), C (Questionable), and D (High Risk); the vertical axis represents the theoretical wind speed deviation. As can be seen from the figure, the median theoretical wind speed deviation for level A data is close to 0.5, with a highly concentrated distribution, indicating that the platform motion has a very limited impact on wind speed measurement at this level. The median theoretical wind speed deviation for level B data is approximately 1.5, with a slightly wider distribution range. The median theoretical wind speed deviation for level C data is approximately 3.5, with a significantly increased interquartile range, indicating a significant decrease in data reliability. The median theoretical wind speed deviation for level D (High Risk) data is approximately 7, with the widest distribution range, and the upper edge is close to 10 or higher. These results further verify the effectiveness of the quality grading method proposed in this invention: data judged as high-risk does indeed correspond to a larger theoretical wind speed deviation, i.e., more severe motion-induced measurement errors, indicating that performing removal or labeling operations on such data has sufficient physical basis and engineering rationale.

[0161] Accordingly, please refer to Figure 6 The second aspect of this invention provides a system for cleaning and classifying wind speed data on non-stationary platforms, which performs wind speed data cleaning and quality classification based on the aforementioned method for cleaning and classifying wind speed data on non-stationary platforms, including:

[0162] Data acquisition module 1 is used to acquire multi-source observation data within a preset time window. The multi-source observation data includes wind speed observations, attitude characteristic parameters of the non-stationary platform, and environmental dynamic characteristic parameters.

[0163] Angle calculation module 2 is used to calculate the comprehensive tilt angle, which characterizes the overall attitude state of the non-stationary platform, based on attitude feature parameters.

[0164] The index calculation module 3 is used to calculate the motion impact index based on the comprehensive tilt angle and environmental dynamic characteristic parameters. The motion impact index is used to quantify the degree of interference of the non-stationary platform motion on the wind speed observation value.

[0165] Quality grading module 4 is used to compare the motion impact index with the preset quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window. The quality level includes at least the high-risk level.

[0166] The data cleaning module 5 is used to clean the wind speed observations according to the quality level. The cleaning process includes at least removing or marking the wind speed observations that belong to the high-risk level.

[0167] Accordingly, a third aspect of the present invention provides an electronic device, comprising: at least one processor and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the aforementioned non-stationary platform wind speed data cleaning and quality classification method.

[0168] Accordingly, a fourth aspect of the present invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for cleaning and quality grading wind speed data on non-stationary platforms. The computer-readable storage medium includes, but is not limited to, media capable of storing computer program code such as read-only memory (ROM), random access memory (RAM), disk storage media, optical storage media, and flash memory media.

[0169] The embodiments of this invention aim to protect a method and system for cleaning and quality grading wind speed data on non-stationary platforms, which has the following effects:

[0170] 1. By statistically fusing the platform's attitude characteristic parameters with environmental dynamic characteristic parameters and constructing the Motion Influence Index (MII), a quantitative evaluation model that does not rely on high-frequency original motion sequences was established. This model can effectively perceive and quantify the data quality degradation caused by the combined effects of complex platform motion and environmental disturbances, thereby significantly improving the detection capability of "hidden systematic biases" that are difficult to identify by traditional thresholding methods, and solving the long-standing problem of the difficulty in evaluating systematic errors in wind speed observation of non-stationary platforms.

[0171] 2. By introducing a dynamic quality threshold generation mechanism based on historical data statistical quantiles and combining it with an adaptive historical data selection strategy (such as based on stability or feature similarity), the quality grading standard is autonomously optimized and dynamically adjusted. The threshold is generated through a historical stability screening mechanism to avoid threshold distortion caused by short-term drastic fluctuations. This enables data quality assessment to adapt to different platform characteristics, diverse sea conditions, and long-term performance changes, outputting more refined, objective, and reliable quality grades (such as A, B, C, and D levels). This significantly improves the accuracy and practicality of the grading results and solves the problem of poor adaptability of fixed thresholds or single assessment models in complex environments.

[0172] 3. The overall technical solution is based entirely on low-frequency statistical features for calculation. The algorithm structure is clear, the computational complexity is low, and each module (such as feature extraction, index calculation, threshold determination, and hierarchical cleaning) forms a complete automated processing closed loop. It can efficiently perform offline batch processing on large-scale historical data or real-time business data without manual intervention, which greatly reduces the implementation threshold and computational resource consumption of data cleaning and solves the engineering practice bottleneck that existing compensation algorithms that rely on high-frequency data cannot be applied to the cleaning of business historical data.

[0173] 4. At the computer implementation level, the technical solution of the present invention has significant advantages: the processing of a single time window involves only a limited number of floating-point operations and array sorting operations, which has low requirements for processor computing power and memory capacity; the various links from data reading from memory, exponential calculation, threshold generation, quality grading to cleaning output form a pipeline-style automated processing link, which supports batch traversal processing of large-scale historical data in memory, and significantly improves the computational throughput efficiency and automation level of data cleaning.

[0174] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0175] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0176] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The function specified in one or more boxes.

[0177] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0178] 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 it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for cleaning and quality grading wind speed data from a non-stationary platform, characterized in that, Includes the following steps: Acquire multi-source observation data within a preset time window, including wind speed observations, attitude characteristic parameters of the non-stationary platform, and environmental dynamic characteristic parameters; Based on the attitude feature parameters, a comprehensive tilt angle is calculated through vector synthesis operation to characterize the overall attitude state of the non-stationary platform. Based on the comprehensive tilt angle and the environmental dynamic characteristic parameters, the motion impact index is calculated according to the preset motion impact index model. The motion impact index is used to quantify the degree of interference of the non-stationary platform motion on the wind speed observation value. The motion impact index is compared with the dynamically generated quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window. The quality level includes at least the high-risk level. The quality grading threshold is dynamically calculated based on the statistical quantile of the historical motion impact index set. The wind speed observations are cleaned according to the quality level, and the cleaning process includes at least removing or marking the wind speed observations that belong to the high-risk level. The motion impact index is calculated based on the comprehensive tilt angle and the environmental dynamic characteristic parameters according to a preset motion impact index model, including: The environmental dynamic characteristic parameters are obtained, including the significant wave height; Based on the effective wave height and the comprehensive tilt angle, combined with the response sensitivity coefficient and normalized reference factor of the non-stationary platform, the motion influence index is calculated. The response sensitivity coefficient and the normalized reference factor are preset values ​​or obtained based on historical data.

2. The method for cleaning and quality grading wind speed data on non-stationary platforms according to claim 1, characterized in that, The formula for calculating the motion impact index is as follows: ； in, To take into account the tilt angle, These are environmental dynamic characteristic parameters. The response sensitivity coefficient is... is the normalized reference factor.

3. The method for cleaning and quality grading wind speed data on non-stationary platforms according to claim 2, characterized in that, Before calculating the motion impact index, the method further includes: Obtain data from multiple historical time windows of the non-stationary platform within a historical time period; For each historical time window, calculate the combined tilt angle and significant wave height within the historical time window; A linear regression model is constructed using the effective wave height of each historical time window as the independent variable and the corresponding comprehensive tilt angle as the dependent variable. The regression coefficients of the linear regression model are determined as the response sensitivity coefficients.

4. The method for cleaning and quality grading of wind speed data on non-stationary platforms according to any one of claims 1-3, characterized in that, The step of comparing the motion impact index with a preset quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window includes: Obtain the set of motion influence indices of the non-stationary platform over a preset number of consecutive historical periods prior to the current time window; Calculate the first quantile, second quantile, and third quantile of the motion influence index set, and set the first quantile, second quantile, and third quantile as the first mass threshold, second mass threshold, and third mass threshold, respectively. The motion impact index of the current time window is compared sequentially with the first quality threshold, the second quality threshold, and the third quality threshold to determine the quality level to which the motion impact index of the current time window belongs.

5. The method for cleaning and quality grading wind speed data on non-stationary platforms according to claim 4, characterized in that, The step of obtaining the set of motion impact indices of the non-stationary platform over a predetermined number of consecutive historical periods prior to the current time window includes: Set the maximum and minimum backtracking time lengths; Within the range defined by the minimum backtracking time length and the maximum backtracking time length, multiple candidate historical periods of different time lengths are generated by increasing the preset time length step. The stability metric of the motion influence index sequence in each candidate historical period is calculated. The method for calculating the stability metric is as follows: the coefficient of variation of the motion influence index sequence is calculated as the stability metric. The coefficient of variation is the ratio of the absolute value of the standard deviation of the motion influence index sequence to the absolute value of the mean. When the absolute value of the mean of the motion influence index sequence is less than a first preset threshold, the stability metric is set to a preset maximum value. Compare the stability metric values ​​of all the candidate historical periods, and determine the candidate historical period with the smallest stability metric value as the consecutive preset number of historical periods; From the aforementioned consecutive preset number of historical periods, all motion impact indices are obtained to obtain the motion impact index set.

6. The method for cleaning and quality grading wind speed data on non-stationary platforms according to claim 5, characterized in that, The step of comparing the stability metric values ​​of all the candidate historical periods and determining the candidate historical period with the smallest stability metric value as the consecutive preset number of historical periods includes: From all the candidate historical periods, the top N candidate historical periods with the smallest stability metric values ​​are selected to form the optimal candidate subset, where N is an integer greater than 1; Calculate the absolute value of the first-order autocorrelation coefficient of the motion influence index sequence within each of the candidate historical periods in the optimal candidate subset; From the optimal candidate subset, the candidate historical period with the smallest absolute value of the first-order autocorrelation coefficient is selected and determined as the consecutive preset number of historical periods.

7. The method for cleaning and quality grading wind speed data on non-stationary platforms according to claim 4, characterized in that, The calculation of the first quantile, second quantile, and third quantile values ​​of the motion influence index set includes: The time decay weight value corresponding to each motion influence index in the set of motion influence indices is determined. The time decay weight value is calculated by a preset decay function model based on the time interval between its corresponding time window and the current time window. The smaller the time interval, the larger the time decay weight value. The time decay weight value of each motion influence index is rounded down to obtain the corresponding representative frequency value; Based on the representative frequency value, each motion influence index is repeated a corresponding number of times and combined to obtain an amplified motion influence index sequence; The amplified motion influence index sequence is numerically sorted to obtain a sorted sequence. Based on the total length of the sorted sequence, the first ordinal position, the second ordinal position, and the third ordinal position are calculated respectively. The values ​​located at the first ordinal position, the second ordinal position, and the third ordinal position in the sorted sequence are respectively assigned the first quantile value, the second quantile value, and the third quantile value.

8. The method for cleaning and quality grading wind speed data on non-stationary platforms according to claim 4, characterized in that, The step of obtaining the set of motion impact indices of the non-stationary platform over a predetermined number of consecutive historical periods prior to the current time window includes: Obtain the environmental dynamic feature parameters and attitude feature parameters of the current time window as the current feature vector; From the aforementioned consecutive preset number of historical periods, obtain the feature vectors of all historical time windows, the feature vectors including the environmental dynamic feature parameters and attitude feature parameters of the corresponding window; The current feature vector and the feature vectors of all the historical time windows are normalized respectively. The normalization process includes standardizing each feature according to its mean and standard deviation in the historical period. Based on the normalized feature vector, calculate the Euclidean distance between the normalized feature vector of each historical time window and the normalized current feature vector. Based on the Euclidean distance, select the K closest historical time windows, where K is a positive integer; Obtain the motion impact index corresponding to K of the historical time windows, and form the motion impact index set.

9. A system for cleaning and grading wind speed data on a non-stationary platform, characterized in that, Wind speed data cleaning and quality classification processing is performed based on the non-stationary platform wind speed data cleaning and quality classification method described in any one of claims 1-8, including: The data acquisition module is used to acquire multi-source observation data within a preset time window. The multi-source observation data includes wind speed observations, attitude characteristic parameters of the non-stationary platform, and environmental dynamic characteristic parameters. An angle calculation module is used to calculate, based on the attitude feature parameters, a comprehensive tilt angle that characterizes the overall attitude state of the non-stationary platform through vector synthesis operations. The index calculation module is used to calculate the motion impact index based on the comprehensive tilt angle and the environmental dynamic characteristic parameters, according to a preset motion impact index model. The motion impact index is used to quantify the degree of interference of the non-stationary platform motion on the wind speed observation value. The quality grading module is used to compare the motion impact index with the dynamically generated quality grading threshold to determine the quality level corresponding to the wind speed observation value within the time window. The quality level includes at least a high-risk level. The quality grading threshold is dynamically calculated based on the statistical quantile of the historical motion impact index set. The data cleaning module is used to clean the wind speed observations according to the quality level. The cleaning process includes at least removing or marking the wind speed observations that belong to the high-risk level.

10. An electronic device, characterized in that, include: At least one processor; A memory connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the non-stationary platform wind speed data cleaning and quality classification method as described in any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the non-stationary platform wind speed data cleaning and quality classification method as described in any one of claims 1-8.

Images