Method and system for filling missing values ​​in wave field based on high-frequency ground wave radar observation

Abstract

The present invention discloses a method and system for filling missing values ​​of a wave field based on high-frequency ground wave radar observation, the method comprising: obtaining wave field information of the high-frequency ground wave radar; performing statistics of missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements; establishing an offshore ocean model according to the missing values ​​of wave elements; obtaining model simulation values ​​according to the offshore ocean model, and filling long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar according to the model simulation values. The present invention improves accuracy and can be widely used in the field of data processing technology.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for filling missing values ​​of a wave field based on high-frequency ground wave radar observation. Background Art

[0002] High-Frequency Surface Wave Radar (HFSWR) is a high-tech means of ocean remote sensing that can detect sea conditions in a large area and monitor and locate moving targets at sea beyond visual range.

[0003] In recent years, as coastal countries have vigorously built shore-based HFSWR networks in coastal areas for monitoring, the use of HFSWR to invert wave elements has been proven to have good reliability, gradually becoming mature and developing towards business development. Since the working principle of HFSWR is to use the characteristics of vertically polarized high-frequency (3-30MHz) electromagnetic waves with low diffraction propagation attenuation on the conductive ocean surface and the scattering mechanism of radio waves on the ocean surface, it can realize the detection of sea conditions in a large range and over-the-horizon monitoring and positioning of moving targets at sea, and has become one of the main observations of offshore wave elements. However, in actual observations, high-frequency ground wave radars are often interfered by external factors, causing the loss of radar echoes, resulting in spatial discontinuity in the detected wave information.

[0004] Waves are one of the most common forms of seawater movement, an important research topic in physical oceanography and an important parameter in the field of marine engineering. Accurate and real-time wave data is helpful for ship navigation, fishery production, offshore oil development, and has guiding significance for port hydraulic construction and beach evolution. The existing wave observation in offshore waters is difficult and has little measured data. Its technical shortcomings can be summarized as follows:

[0005] 1. Traditional wave observation instruments mainly include wave measuring rods, acoustic wave meters, buoys, laser wave meters, etc. Wave measuring rods, acoustic wave meters, laser wave meters, etc. have the advantages of easy installation, high observation accuracy, and low cost, but they require fixed platforms, regular maintenance, are easily affected by extreme weather, and cannot measure large areas.

[0006] 2. Wave remote sensing observation mainly uses high-frequency ground wave radar and satellite altimeter. The wave parameters observed by satellite altimeter have the disadvantages of large spatial resolution and long revisit period, and it is difficult to conduct real-time monitoring of the target sea area. Compared with satellite altimeter, the data volume obtained by inverting the sea condition information of the target sea area using HFSWR has lower cost, higher time frequency and better observation accuracy.

[0007] However, due to the complexity of the actual sea conditions, affected by objective factors such as ionospheric interference, transient interference, equipment failure, power failure, etc., the wave information (wave height, wave direction, wave period) inverted by the high-frequency ground wave mine receiving radar echo signal will be discontinuous in time and space. At the same time, in the process of data collection, in addition to the technical inability to obtain data, there are also high data acquisition costs and no response from units due to confidentiality or privacy issues, which will affect the integrity of the data. After the sample collection is completed, the missing data due to human factors, or the data is incorrect or unavailable, will indirectly cause incomplete or intermittent data.

[0008] The existing technology has not yet solved the problem of temporal and spatial discontinuity of HFSWR wave field observation elements, and the technology for filling missing values ​​of long-term HFSWR needs to be improved. Summary of the invention

[0009] In view of this, an embodiment of the present invention provides a method and system for filling missing values ​​of a wave field based on high-frequency ground wave radar observation with high accuracy.

[0010] One aspect of the present invention provides a method for filling missing values ​​of a wave field based on high-frequency ground wave radar observation, comprising:

[0011] Obtain wave field information from high-frequency ground wave radar;

[0012] Performing statistics on missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements;

[0013] Establishing a nearshore ocean model according to the missing values ​​of the wave elements;

[0014] Model simulation values ​​are obtained according to the offshore ocean model, and long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar are filled according to the model simulation values.

[0015] Optionally, the acquiring of wave field information of a high-frequency ground wave radar includes:

[0016] The shore antenna array of the shore-based high-frequency ground wave radar transmits vertically polarized shortwave to the sea area and obtains the radar echo signal;

[0017] The radar echo signal is processed by using full integral inversion and fitting wave spectrum model to obtain wave field information from the second-order radar spectrum;

[0018] The wave field information includes but is not limited to radar coverage information, wave height information obtained by inversion, wave period information and wave direction information.

[0019] Optionally, performing statistics of missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements includes:

[0020] According to the wave field information, first missing value records of different wave elements in the radial direction of the shore-based high-frequency ground wave radar are counted;

[0021] According to the wave field information, counting the second missing value records at different observation locations;

[0022] According to the wave field information, counting the third missing value records at different measurement times;

[0023] Determine the missing values ​​of wave elements based on the first missing value record, the second missing value record and the third missing value record.

[0024] Optionally, the method further comprises the step of performing block processing on the target area, the step comprising:

[0025] Obtain the nautical chart data of the target sea area and extract the nearshore coastline data and water depth data within the coverage area of ​​the high-frequency ground wave radar;

[0026] Based on the nearshore coastline data and water depth data, delineate shallow water areas and deep water areas;

[0027] Obtaining meteorological data of the target sea area and extracting wind speed data within the coverage area of ​​the high-frequency ground wave radar;

[0028] Based on the wind speed data, normal wind speed periods and extreme wind speed periods are delineated.

[0029] Optionally, establishing a nearshore ocean model according to the missing values ​​of the wave elements includes:

[0030] The shallow water area is used as a research area of ​​the offshore wave model SWAN model, and the deep water area is used as a research area of ​​the ocean wave model WW3 model, and the grid area is obtained by dividing the area;

[0031] Obtaining background sea conditions and meteorological data, and processing the background sea conditions and meteorological data into corresponding network resolution accuracy; wherein the background sea conditions and meteorological data include historical wind field data, historical flow field data, and initial wave field data at the time of model operation;

[0032] According to the divided grid areas, the SWAN model and the WW3 model are run respectively;

[0033] The parameters of the SWAN model and the WW3 model are adjusted, and the constructed models are quantitatively evaluated to obtain a nearshore ocean model that meets the accuracy requirements.

[0034] Optionally, acquiring model simulation values ​​according to the offshore ocean model, and filling long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar according to the model simulation values, comprises:

[0035] Combined with the model simulation values, the long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar are filled; among them, the short-term missing periods of different points within the research range of the high-frequency ground wave radar are interpolated, and the long-term missing periods are interpolated by model values;

[0036] The interpolation process for the short-term missing periods of different points within the research range of the high-frequency ground wave radar includes:

[0037] The target point missing period is recorded as a blank value, and the seasonal time series wave element information model is established;

[0038] Perform sliding average processing on the data of the seasonal time series wave element information model to obtain the baseline of the wave element time series of the target point;

[0039] The data is interpolated using the baseline of the time series.

[0040] Optionally, the method further comprises the step of constructing a data classification model based on information entropy, the step comprising:

[0041] Acquire initial data, and perform data preprocessing on the initial data to obtain target data;

[0042] extracting target features from the target data;

[0043] According to the target features, a learning algorithm constructed by a feedforward neural network is used for training to construct a data classification model based on information entropy.

[0044] Optionally, the method further comprises the step of constructing a wave forecast model, the step comprising:

[0045] Acquire sample data of the effective wave height credible value according to the data classification model based on information entropy, and construct a training sample set and a verification set;

[0046] The training sample set is trained by a multi-layer perceptron model and a random forest model, and the significant wave height data is output as the first layer model learning training result;

[0047] Constructing a new training set, using an extreme tree model to train the new training set, and outputting effective wave height data as a second-layer model learning and training result;

[0048] A wave forecasting model is constructed according to the verification results of the first layer model learning and training results and the second layer model learning and training results.

[0049] Another aspect of the embodiment of the present invention further provides a system for filling missing values ​​of a wave field based on high-frequency ground wave radar observation, comprising:

[0050] The first module is used to obtain the wave field information of the high-frequency ground wave radar;

[0051] The second module is used to perform statistics on missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements;

[0052] The third module is used to establish a near-shore ocean model according to the missing values ​​of the wave elements;

[0053] The fourth module is used to obtain model simulation values ​​according to the offshore ocean model, and fill in the long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar according to the model simulation values.

[0054] Another aspect of an embodiment of the present invention further provides an electronic device, including a processor and a memory;

[0055] The memory is used to store programs;

[0056] The processor executes the program to implement the method described above.

[0057] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0058] The embodiment of the present invention obtains the wave field information of the high frequency ground wave radar; performs statistics of the missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements; establishes an offshore ocean model according to the missing values ​​of wave elements; obtains model simulation values ​​according to the offshore ocean model, and fills the long-term missing values ​​and short-term missing values ​​of the high frequency ground wave radar according to the model simulation values. The present invention improves accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0060] Figure 1 An overall step flow chart provided for an embodiment of the present invention;

[0061] Figure 2 A flowchart for obtaining HFSWR wave field information provided by an embodiment of the present invention;

[0062] Figure 3 A flow chart of the statistics of missing values ​​of wave elements provided by an embodiment of the present invention;

[0063] Figure 4 A flowchart of a target area segmentation process provided by an embodiment of the present invention;

[0064] Figure 5 A flow chart of establishing a near-shore ocean model provided by an embodiment of the present invention;

[0065] Figure 6 A flow chart for filling missing values ​​of HFSWR wave fields provided by an embodiment of the present invention;

[0066] Figure 7 A flowchart of establishing a data classification model based on information entropy provided by an embodiment of the present invention;

[0067] Figure 8 A flow chart of a wave forecasting model integrating machine learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0069] In view of the problems existing in the prior art, one aspect of the present invention provides a method for filling missing values ​​of a wave field based on high-frequency ground wave radar observation, comprising:

[0070] Obtain wave field information from high-frequency ground wave radar;

[0071] Performing statistics on missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements;

[0072] Establishing a nearshore ocean model according to the missing values ​​of the wave elements;

[0073] Model simulation values ​​are obtained according to the offshore ocean model, and long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar are filled according to the model simulation values.

[0074] Optionally, the acquiring of wave field information of a high-frequency ground wave radar includes:

[0075] The shore antenna array of the shore-based high-frequency ground wave radar transmits vertically polarized shortwave to the sea area and obtains the radar echo signal;

[0076] The radar echo signal is processed by using full integral inversion and fitting wave spectrum model to obtain wave field information from the second-order radar spectrum;

[0077] The wave field information includes but is not limited to radar coverage information, wave height information obtained by inversion, wave period information and wave direction information.

[0078] Optionally, performing statistics of missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements includes:

[0079] According to the wave field information, first missing value records of different wave elements in the radial direction of the shore-based high-frequency ground wave radar are counted;

[0080] According to the wave field information, counting the second missing value records at different observation locations;

[0081] According to the wave field information, counting the third missing value records at different measurement times;

[0082] Determine the missing values ​​of wave elements based on the first missing value record, the second missing value record and the third missing value record.

[0083] Optionally, the method further comprises the step of performing block processing on the target area, the step comprising:

[0084] Obtain the nautical chart data of the target sea area and extract the nearshore coastline data and water depth data within the coverage area of ​​the high-frequency ground wave radar;

[0085] Based on the nearshore coastline data and water depth data, delineate shallow water areas and deep water areas;

[0086] Obtaining meteorological data of the target sea area and extracting wind speed data within the coverage area of ​​the high-frequency ground wave radar;

[0087] Based on the wind speed data, normal wind speed periods and extreme wind speed periods are delineated.

[0088] Optionally, establishing a nearshore ocean model according to the missing values ​​of the wave elements includes:

[0089] The shallow water area is used as a research area of ​​the offshore wave model SWAN model, and the deep water area is used as a research area of ​​the ocean wave model WW3 model, and the grid area is obtained by dividing the area;

[0090] Obtaining background sea conditions and meteorological data, and processing the background sea conditions and meteorological data into corresponding network resolution accuracy; wherein the background sea conditions and meteorological data include historical wind field data, historical flow field data, and initial wave field data at the time of model operation;

[0091] According to the divided grid areas, the SWAN model and the WW3 model are run respectively;

[0092] The parameters of the SWAN model and the WW3 model are adjusted, and the constructed models are quantitatively evaluated to obtain a nearshore ocean model that meets the accuracy requirements.

[0093] Optionally, acquiring model simulation values ​​according to the offshore ocean model, and filling long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar according to the model simulation values, comprises:

[0094] Combined with the model simulation values, the long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar are filled; among them, the short-term missing periods of different points within the research range of the high-frequency ground wave radar are interpolated, and the long-term missing periods are interpolated by model values;

[0095] The interpolation process for the short-term missing periods of different points within the research range of the high-frequency ground wave radar includes:

[0096] The target point missing period is recorded as a blank value, and the seasonal time series wave element information model is established;

[0097] Perform sliding average processing on the data of the seasonal time series wave element information model to obtain the baseline of the wave element time series of the target point;

[0098] The data is interpolated using the baseline of the time series.

[0099] Optionally, the method further comprises the step of constructing a data classification model based on information entropy, the step comprising:

[0100] Acquire initial data, and perform data preprocessing on the initial data to obtain target data;

[0101] extracting target features from the target data;

[0102] According to the target features, a learning algorithm constructed by a feedforward neural network is used for training to construct a data classification model based on information entropy.

[0103] Optionally, the method further comprises the step of constructing a wave forecast model, the step comprising:

[0104] Acquire sample data of the effective wave height credible value according to the data classification model based on information entropy, and construct a training sample set and a verification set;

[0105] The training sample set is trained by a multi-layer perceptron model and a random forest model, and the significant wave height data is output as the first layer model learning training result;

[0106] Constructing a new training set, using an extreme tree model to train the new training set, and outputting effective wave height data as a second-layer model learning and training result;

[0107] A wave forecasting model is constructed according to the verification results of the first layer model learning and training results and the second layer model learning and training results.

[0108] Another aspect of the embodiment of the present invention further provides a system for filling missing values ​​of a wave field based on high-frequency ground wave radar observation, comprising:

[0109] The first module is used to obtain the wave field information of the high-frequency ground wave radar;

[0110] The second module is used to perform statistics on missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements;

[0111] The third module is used to establish a near-shore ocean model according to the missing values ​​of the wave elements;

[0112] The fourth module is used to obtain model simulation values ​​according to the offshore ocean model, and fill in the long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar according to the model simulation values.

[0113] Another aspect of an embodiment of the present invention further provides an electronic device, including a processor and a memory;

[0114] The memory is used to store programs;

[0115] The processor executes the program to implement the method described above.

[0116] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0117] The specific implementation process of the present invention is described in detail below in conjunction with the accompanying drawings:

[0118] Existing technologies have not yet solved the temporal and spatial discontinuities of HFSWR wave field observation elements, and the technology for filling missing values ​​of HFSWR over long periods of time needs to be improved. This application intends to use the SVM algorithm, combined with the wave model, to fill in the missing values ​​of HFSWR, and to classify the data credibility based on the data classification model of information entropy, and to establish a wind-water depth-current-wave forecast model through the wave forecast model integrated with machine learning, and complete the system of complete wave field reconstruction.

[0119] The technical route steps of the present invention are shown in Figure 1 , the detailed process steps are as follows:

[0120] The first step is to obtain HFSWR wave field information. The technical process is shown in Figure 2 .

[0121] The shore antenna array of the shore-based HFSWR transmits vertically polarized shortwaves with a frequency range of (3MHz to 30MHz) to the sea area, and obtains radar echo signals based on the Bragg back-reflection principle of the sea surface. At present, the signal processing of radar echoes mainly uses full integral inversion and fitting of the wave spectrum model to obtain wave element information from the second-order radar spectrum. The wave field information obtained includes radar coverage and inverted wave height, wave period, wave direction and other information.

[0122] The second step is to count the missing values ​​of wave elements. See the technical process for details. Figure 3 .

[0123] First, based on the historical HFSWR inversion wave element information of the target sea area, the missing values ​​of different wave elements in the radial direction of the shore-based HFSWR, at each observation point, and at different measurement times are recorded and counted. Secondly, the data is screened and processed to eliminate unreasonable data and calibrate them as missing values. Finally, for the missing values ​​of each HFR point in all time periods, for each point data, the missing data is defined as a short-term missing value within a time step of one hour, and the missing data of the point exceeding the time step of one hour is defined as a long-term missing value. Note that since the radar echo signal is easily affected by objective factors, resulting in weak or no echo signal at a certain point at a certain time, an observation point can have multiple long and short-term missing periods.

[0124] The third step is to process the target area in blocks. See the technical flow field for details. Figure 4 .

[0125] The inversion of waves by HFSWR, in areas with smaller water depths, no longer conforms to the deep-water wave dispersion relationship derived from the micro-wave theory used in remote sensing observation technology, and the inversion of wave parameters in intertidal shoals also has large precision errors. Therefore, by obtaining the nautical chart data of the target sea area, the water depth data of the HFSWR coverage area is extracted. The location of the target sea area is determined according to the HFR measurement range, and the nearshore coastline data, water depth and other data of the corresponding range are obtained. With the water depth of 20m as the dividing line, the water depth of less than 20m in the HFSWR covered sea area is defined as the nearshore shallow water area, and the water depth greater than 20m is defined as the deep water area.

[0126] At the same time, another limitation of the HFSWR extraction and calculation wave theory is related to the maximum wave height. The maximum effective wave height inverted by radars with different transmission frequencies has a theoretical cutoff value. For example, the theoretical cutoff value of the effective wave height of the 13.5MHz frequency SeaSondeHF radar deployed in the Malta Strait in Europe is about 7.5m. When the wave height of large waves exceeds the limit defined by the radar transmission frequency, the radar spectrum is saturated, so the actual wave height value is underestimated. The wind in the climate element is the most important factor affecting the effective wave height. Therefore, by obtaining the meteorological data of the target sea area, high-precision wind speed data of the coverage area of ​​HFSWR is extracted. The wind speed of 20.7m / s at 10m above the ground is defined as the normal wind period (the maximum wave height on the sea surface is 7.5m at this time), and the wind speed in the study area is judged. The wind speed in the sea area covered by HFSWR is less than 20.7m / s as the normal wind period, and greater than 20.7m / s as the extreme wind speed period.

[0127] The fourth step is to establish a nearshore ocean model (SWAN, WW3). The technical flow chart is as follows: Figure 5 , the specific steps are as follows:

[0128] 1. Divide the grid area: First, the shallow water area divided in the third step above is used as the research area of ​​the offshore wave model SWAN model, and the deep water area is used as the ocean wave model WW3 model area. In the shallow water area, a small grid with high resolution is selected to model the shallow water area. Among them, unstructured grids or structured grids with the same or higher resolution as the offshore resolution of HFSWR can be selected. In the deep water area, a larger grid with higher resolution is selected to model the deep water area. The WW3 model only accepts structured grids. Among them, the selected grid must be consistent with or higher than the resolution of HFSWR.

[0129] 2. Obtain background sea conditions and meteorological data: The background sea conditions and meteorological data obtained include historical wind field data, historical flow field data, and initial wave field data at the time of model operation, and are processed into corresponding grid resolution accuracy.

[0130] 3. Model nesting: Run the SWAN model with a higher resolution offshore grid, run the WW3 ocean model with a lower resolution ocean grid, and nest the SWAN model into the WW3 model.

[0131] 4. Model parameter adjustment: Adjust the model parameters based on the existing research data of the target sea area, such as published papers, previous experience, etc.

[0132] 5. Model evaluation: Evaluate the model based on available measured data, such as buoys, public or available satellite data, and HFSWR observed wave data within the evaluated target sea area. This application uses the correlation coefficient r and root mean square error RMSE as the main indicators to conduct a quantitative comprehensive evaluation of the model. The formula is as follows:

[0133] Correlation coefficient r:

[0134]

[0135] Root mean square error RMSE:

[0136]

[0137] In the formula, x and y are the series of data obtained from different observation channels for the same wave element. The larger the correlation coefficient r, the smaller the root mean square error RMSE, which means that the model simulation value is closer to the measured data and the higher the model simulation accuracy.

[0138] The fifth step is to fill in the missing values ​​of the HFSWR wave field. The technical flow chart is as follows: Figure 6 .

[0139] Combined with the model simulation values, the long-term missing values ​​and short-term missing values ​​of HFSWR are filled. Among them, the short-term missing periods of different points within the HFSWR research range are interpolated, and the long-term missing periods are interpolated with model values.

[0140] Due to the lack of high-frequency ground wave radar echo signals, the inverted information of multiple wave elements is simultaneously missing, and it is impossible to find the association between different attributes for the variable time series. For the interpolation filling of missing values ​​in time series, there are mainly mean, mode, random sample interpolation, linear interpolation, etc. Based on the data characteristics of wave elements with trends and seasonality, this application uses an interpolation filling method based on sliding average to interpolate short-term missing periods. The steps are as follows: 1. The missing period of the target point is recorded as a blank value, and a seasonal time series wave element information model is established. 2. Perform sliding average processing on the model data to obtain the baseline of the target point wave element time series. 2. Interpolate the data using the time series baseline.

[0141] However, actual ocean waves are random, and are affected by the ever-changing wind speed and direction on the sea surface. Their wave height, wavelength, and period are all random quantities. For missing values ​​within the HFSWR research scope, when the proportion of missing data increases, the errors of all filling methods will increase, and they cannot be compensated by algorithms. Due to data confidentiality or privacy issues, the wave elements inverted by high-frequency ground wave radar may have long-term missing values. Therefore, model value interpolation is performed for long-term missing periods within the HFSWR research scope.

[0142] The sixth step is to establish a data classification model based on information entropy. The technical flow chart is shown below. Figure 7 , the steps are as follows:

[0143] Since the inversion accuracy of wave elements inverted by high-frequency ground wave radar is large in shallow water areas (not in line with the dispersion relationship of deep water waves, poor inversion accuracy), long-distance areas (there are distance restrictions, poor inversion accuracy), in the straight line direction of two adjacent radar base stations, and under extreme wind speed conditions (due to the theoretical maximum cutoff value of effective wave height caused by radar transmission frequency, there is an underestimation of wave elements), there are still some good improvement methods. Therefore, in order to solve the problem of inaccurate HFSWR sampling data and filling data, this paper proposes a machine learning network model using information entropy to classify the data into shallow water outliers, long-distance outliers, extreme wind speed outliers, and credible values, so as to achieve information capture and make data classification processing in advance for the next step of machine learning to explore the mapping relationship between meteorological elements and waves.

[0144] 1. Data preprocessing. Data preprocessing includes data cleaning and data normalization. Data cleaning is to remove abnormal feature values ​​and delete duplicate information. The data normalization method of this application adopts the maximum-minimum normalization method. Let minA and minA be the minimum and maximum values ​​of attribute A respectively. Map an original data x of A to x′ in the interval [0,1] through maximum-minimum normalization. The formula is:

[0145]

[0146] 2. Feature extraction. Information entropy actually reflects the uncertainty of information. In a random event, the greater the uncertainty of an event, the greater the entropy, and the more information we need to figure out. Since information entropy is good at describing unknown information, this application performs redundancy removal on the preprocessed data based on information entropy and uses the mutual information matrix to extract the main component features.

[0147] Assume that the preprocessed data set is D and the feature distribution space is R a×b Assume that any data set D in D i =(d i1 ,d i2 ,...,d ib ), d ij It is the jth feature quantity corresponding to the description data i. The information entropy is introduced and the information entropy formula of each feature attribute is obtained as follows:

[0148]

[0149] in, It is the probability relationship between the j feature quantity corresponding to data i and the j feature quantities of all data.

[0150] Introducing mutual information R(D i ,D j )=H(D i )+H(D j)-H(D i ,D j ), where H(D i ,D j ) by d i and d j The joint probability is obtained; R(D i ,D j ) is a quantitative determination of entropy characteristic attributes, which can be applied to nonlinear relationships. Solving the mutual information of all data features, the matrix is ​​as follows:

[0151]

[0152] In the matrix R, R 11 , R 22 ...,R bb is self-information, and the rest is mutual information, that is, R ij =R ji . Assume that the feature order set of mutual information R is r1, r2, ..., r n , the corresponding eigenvectors are a1, a2, ..., a n The formula for calculating the principal component dimension based on the eigenvalue of the symmetric matrix R is:

[0153]

[0154] Solve the dimension and value of the top-ranked eigenvalues sum a Compared with the threshold, when sum a When the judgment threshold is exceeded, A is determined a =(a1,a2,...,a a ) is the principal component matrix, and then A′ a D can complete the principal component feature extraction. Since this feature extraction method has good nonlinearity, it can effectively process non-uniformly distributed cross-coverage information.

[0155] 3. Information classification based on machine learning. Taking into account the timeliness and accuracy of the model for data classification, the Single-hidden Layer Feed forward Network (SLFN) is selected to construct the learning algorithm, which is specifically divided into three levels: input, output, and hidden. The SLNF algorithm will randomly generate the connection weights between the input layer and the hidden layer and the thresholds of the hidden layer neurons, and no adjustments are required during the training process. Only the number of neurons in the hidden layer needs to be set to obtain the only optimal solution. Compared with traditional training methods, this method has the advantages of fast learning rate and good generalization performance.

[0156] Assume that the input data volume is q, and the input sample is {(ei ,u i ),i∈[1,q]}, the input vector is e i =[e i1 ,e i2 ,...,e iβ ] T , define e i The ideal output is u i =[u i1 ,u i2 ,...,u iβ ] T . Then the number of input and output neurons can be determined as β and ε. Then the SLFN network expression is:

[0157]

[0158] Where N represents the number of neurons in the hidden layer; ζ i =[ζ i1 ,ζ i2 ,...,ζ iβ ] T , i =[ξ i1 ,ζ i2 ,...,ζ iβ ] T They are used to describe the weighted values ​​between the neurons of hidden layer i and the output and input respectively; J i (x) is the activation function; δ i is the error. According to the relationship between each layer, the hidden layer output can be obtained as:

[0159]

[0160] Among them, G is the hidden layer output; O is the ideal output. Introducing regularization processing, the regular model of the Lagrange function is constructed as follows:

[0161]

[0162] Where η=[η1,η2,...,η β ] T is the Lagrange multiplier, ‖ζ i ‖ 2 represents the network complexity, ‖Δ i ‖ 2 Represents the learning bias. The network output weights are calculated as follows:

[0163]

[0164] If the amount of learning data in the network exceeds N, U represents an N×N matrix; otherwise, U represents an M×M matrix.

[0165] 4. Evaluation indicators. This application selects three indicators, Acc (Accuracy), F-Measure and G-Mean, to comprehensively test the performance of the cross-coverage information capture algorithm.

[0166] Acc is the overall classification accuracy, which is used to describe the capture accuracy of the model. The formula is:

[0167]

[0168] Among them, TP represents the correctly predicted positive classification result, FP represents the incorrectly predicted positive classification result, and FN represents the incorrectly predicted negative classification result.

[0169] F-Measure is an indicator commonly used to evaluate the quality of classification models. The formula is:

[0170]

[0171] Among them, β is the weighted harmonic function, P = TP / (TP+FP) represents the classification accuracy, and R = TP / (TP+TN) represents the recall rate.

[0172] G-Mean is a comprehensive indicator used to describe the recall rate of positive and negative classes. G-Mean increases only when the R index of both positive and negative classes increases. It has good robustness to unbalanced data. The formula is:

[0173]

[0174] Step 7. Integrate machine learning wave forecast model: The integration method combines concentrated machine learning techniques into a meta-algorithm of a prediction model to achieve the effect of reducing variance (bagging), bias (boosting) or improving prediction (stacking). Since the amount of filled HFSWR radar data obtained is large, there are many variable factors, and there may be multiple mapping relationships, this application establishes a wave forecast model by selecting the Stacking method. Using the data of the improved HFR wave field as a sample, a wind-water depth-current-wave forecast model is established through integrated learning. For ease of understanding, the following steps establish a wave forecast model based on the relationship between wind speed and effective wave height. See below for the technical flow chart Figure 8 , the specific steps are as follows:

[0175] 1. Establish a training sample set. First, select the effective wave height credible value classified by information entropy in step 6 as the total sample, and divide it into 10 sample sets in proportion to the wave height level, of which 7 samples are used as training sets and 3 sample sets are used as validation sets. The training sample is defined as in, Refers to the wind speed data corresponding to the credible value of effective wave height. Refers to the reliable value data of significant wave height. For different sea conditions, it is divided into 10 sample sets according to the significant wave height level / wind speed level, of which 7 samples are used as training sets and 3 sample sets are used as verification sets.

[0176] 2. First-layer model learning and training. The first-layer model is divided into two individual learners: a multi-layer perceptron model and a random forest model. Through the training of the sample set by different learners, the significant wave height data predicted by each individual learner is output.

[0177] 3. Create a new training set. To avoid the situation where the model cannot fit the same data set due to its high adaptability to the current training set, a new training set D′ is regenerated as the training sample for the second-layer model. t represents the number of the individual learner in the first layer model, and j represents the sample number.

[0178] 4. Second-layer model learning and training: The individual learner used in the second-layer model is the extreme tree model.

[0179] 5. Verification of the wind speed and significant wave height single-unit mapping relationship forecast model. The wind speed and significant wave height forecast model composed of the above two layers of models was put into the validation set for verification. This application uses the goodness of fit R 2 Perform a quantitative evaluation of the model.

[0180]

[0181] in, represents the prediction result of the algorithm, y represents the effective wave height of the validation set, Represents the average value of the effective wave height in the validation set. R 2 The value range of R is [0,1]. 2 The larger it is, the stronger the correlation between wind speed and significant wave height established by the model.

[0182] 6. Wave multi-factor mapping relationship forecast model. By gradually adding influencing factors such as current and water depth as the first-layer model input training set, a wind-water depth-current-wave influence mechanism is established, and a multi-factor wave forecast model for the target sea area is established.

[0183] 7. Wave field data reconstruction and prediction. The wind speed conditions corresponding to the effective wave heights classified as shallow water anomalies, long-distance anomalies, and extreme wind speed anomalies are put into the wave forecasting algorithm of stacking integrated machine learning. The obtained forecast results are compared with the wave elements (such as effective wave height) classified as shallow water anomalies, long-distance anomalies, and extreme wind speed anomalies. The data is overwritten and the model is re-run to improve the accuracy and stability of the model response mechanism results.

[0184] In summary, the present invention aims at the technical limitations of the existing high-frequency ground wave radar on the missing wave data and the lack of discussion on the wave dynamic mechanism of the target sea area, and proposes a filling model based on the wave numerical model and the SVM algorithm, and establishes a wind-water depth-current-wave forecast model based on the information entropy classification model and integrated machine learning to complete the complete wave field reconstruction system. The specific advantages are as follows:

[0185] 1. Multi-source data complementarity: synchronously collect data on wind, current, water depth and available wave buoys in the target sea area, establish the offshore numerical model SWAN and the ocean numerical model WW3, and fill in and compare the long-term missing values ​​of HFSWR.

[0186] 2. Considering long and short missing values ​​HFSWR data filling: According to the time continuity of the missing data of the target point, it is divided into long / short missing values. For long missing values, the wave model data is used to fill, and for short missing values, the SVM algorithm model is used to fill.

[0187] 3. Reasonableness of data processing: Using the data classification model based on information entropy, HFSWR data is classified to improve the reliability of the input sample set of wave forecasting for integrated machine learning. At the same time, using the abnormal data sample set as the control data for integrated machine learning can further explore the scope and limitations of HFSWR inversion wave elements.

[0188] 4. Wave forecasting with integrated machine learning: Integrated machine learning is divided into the first layer model and the second layer model, and the training set of the second layer model is output through the first layer model. Compared with a single machine learning model, it can explore the response mechanism of wind, water depth, current, etc. to waves in more depth, greatly improving the accuracy and stability of wave element forecasting. (When exploring the response mechanism of wind to waves alone, the reliable wave Yang element sample set corresponding to the wind field element sample set is used as the input sample set of the first layer model, and the output result of the first layer model is used as the input sample set of the second layer model, thereby establishing the wind-wave response mechanism and forecasting model).

[0189] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0190] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features described may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0191] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0192] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0193] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0194] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0195] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0196] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0197] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A method for filling missing values ​​in wave fields based on high-frequency ground wave radar observations, characterized in that: include: Obtain wave field information from high-frequency ground wave radar; Performing statistics on missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements; Establishing a nearshore ocean model according to the missing values ​​of the wave elements; Acquire model simulation values ​​according to the offshore ocean model, and fill in long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar according to the model simulation values; The step of obtaining the wave field information of the high frequency ground wave radar comprises: The shore antenna array of the shore-based high-frequency ground wave radar transmits vertically polarized shortwave to the sea area and obtains the radar echo signal; The radar echo signal is processed by using full integral inversion and fitting wave spectrum model to obtain wave field information from the second-order radar spectrum; The wave field information includes but is not limited to radar coverage information, wave height information obtained by inversion, wave period information and wave direction information; The method of establishing a nearshore ocean model according to the missing values ​​of the wave elements comprises: The shallow water area is used as the research area of ​​the offshore wave model SWAN model, and the deep water area is used as the research area of ​​the ocean wave model WW3 model, and the grid area is obtained; Obtaining background sea conditions and meteorological data, and processing the background sea conditions and meteorological data into corresponding network resolution accuracy; wherein the background sea conditions and meteorological data include historical wind field data, historical flow field data, and initial wave field data at the time of model operation; According to the divided grid areas, the SWAN model and the WW3 model are run respectively; Adjusting parameters of the SWAN model and the WW3 model, and quantitatively evaluating the constructed models to obtain a nearshore ocean model that meets accuracy requirements; The method of obtaining model simulation values ​​according to the offshore ocean model and filling in long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar according to the model simulation values ​​includes: Combined with the model simulation values, the long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar are filled; among them, the short-term missing periods of different points within the research range of the high-frequency ground wave radar are interpolated, and the long-term missing periods are interpolated by model values; The interpolation process for the short-term missing periods of different points within the research range of the high-frequency ground wave radar includes: The target point missing period is recorded as a blank value, and the seasonal time series wave element information model is established; Perform sliding average processing on the data of the seasonal time series wave element information model to obtain the baseline of the wave element time series of the target point; The data is interpolated using the baseline of the time series.

2. The method for filling missing values ​​of wave fields based on high-frequency ground wave radar observation according to claim 1 is characterized in that: The performing statistics of missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements comprises: According to the wave field information, first missing value records of different wave elements in the radial direction of the shore-based high-frequency ground wave radar are counted; According to the wave field information, counting the second missing value records at different observation locations; According to the wave field information, counting the third missing value records at different measurement times; Determine the missing values ​​of wave elements based on the first missing value record, the second missing value record and the third missing value record.

3. The method for filling missing values ​​of wave fields based on high-frequency ground wave radar observation according to claim 1 is characterized in that: The method further comprises the step of performing block processing on the target area, the step comprising: Obtain the nautical chart data of the target sea area and extract the nearshore coastline data and water depth data within the coverage area of ​​the high-frequency ground wave radar; Based on the nearshore coastline data and water depth data, delineate shallow water areas and deep water areas; Obtaining meteorological data of the target sea area and extracting wind speed data within the coverage area of ​​the high-frequency ground wave radar; Based on the wind speed data, normal wind speed periods and extreme wind speed periods are delineated.

4. The method for filling missing values ​​of wave fields based on high-frequency ground wave radar observation according to claim 1 is characterized in that: The method further comprises the step of constructing a data classification model based on information entropy, the step comprising: Acquire initial data, and perform data preprocessing on the initial data to obtain target data; extracting target features from the target data; According to the target features, a learning algorithm constructed by a feedforward neural network is used for training to construct a data classification model based on information entropy.

5. The method for filling missing values ​​of wave fields based on high-frequency ground wave radar observation according to claim 4 is characterized in that: The method further comprises the step of constructing a wave forecast model, the step comprising: Acquire sample data of the effective wave height credible value according to the data classification model based on information entropy, and construct a training sample set and a verification set; The training sample set is trained by a multi-layer perceptron model and a random forest model, and the significant wave height data is output as the first layer model learning training result; Constructing a new training set, using an extreme tree model to train the new training set, and outputting effective wave height data as a second-layer model learning and training result; A wave forecasting model is constructed according to the verification results of the first layer model learning and training results and the second layer model learning and training results.

6. A system using the method for filling missing values ​​of a wave field based on high-frequency ground wave radar observation as described in any one of claims 1 to 5, characterized in that: include: The first module is used to obtain the wave field information of the high-frequency ground wave radar; The second module is used to perform statistics on missing values ​​of wave elements according to the wave field information to determine the missing values ​​of wave elements; The third module is used to establish a near-shore ocean model according to the missing values ​​of the wave elements; The fourth module is used to obtain model simulation values ​​according to the offshore ocean model, and fill in the long-term missing values ​​and short-term missing values ​​of the high-frequency ground wave radar according to the model simulation values.

7. An electronic device, characterized in that: including a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 5.

Images