A Wave Dynamic Weather Classification Method Based on Environmental Field Feature Extraction

Abstract

The present invention discloses a wave dynamic weather classification method based on environmental field feature extraction, which belongs to the field of coastal and marine engineering technology. The method comprises the following steps: obtaining n-year time series historical data of wave elements of a certain station, and obtaining n-year time series historical data of wind field and sea level pressure field near the station; identifying independent wave events from the time series wave height sequence based on a 5-day time window, and screening stronger independent wave events as events to be classified according to the threshold method; exploring the regional correlation between wind speed and independent wave process peak elements under different time lag conditions to determine the time domain and space domain of wind field and sea level pressure field for classification; inputting the determined wind field and sea level pressure field into the SOM+K-means clustering algorithm to obtain the weather classification results of independent wave events. This method improves the blindness of the initial clustering center selection of the K-means clustering algorithm, and has the advantages of short convergence time and high result stability.

Description

Technical Field

[0001] The invention belongs to the technical field of coastal and marine engineering, and specifically relates to a wave dynamic weather classification method based on environmental field feature extraction. Background Art

[0002] In recent years, while the development of marine engineering has faced new opportunities, it has also faced severe challenges. Atmospheric circulation is a key factor driving wave climate. Their occurrence and persistence control the development of waves, including extreme wave events that may pose a serious threat to marine engineering. Therefore, it is necessary to understand the connection between ocean waves and atmospheric circulation. However, the physical connection between the two is complex and nonlinear because it involves processes at multiple spatial and temporal scales. Therefore, attempts to model wave characteristics from atmospheric circulation patterns can be difficult and time-consuming. In recent years, thanks to the development of weather climatology, more attempts have been made to obtain the statistical relationship between atmospheric circulation patterns and regional wave climate observations in order to statistically model regional wave climate using atmospheric circulation patterns.

[0003] Establishing the connection between atmospheric circulation patterns and regional wave climate through statistical techniques has many advantages. Atmospheric circulation patterns are closely related to the distribution of wave parameters (such as wave height, wave direction, etc.). Wave parameters under the influence of similar circulation patterns show statistical similarities. Based on this, when a specific type of circulation pattern appears, the relevant wave parameters can be predicted by statistical methods. In addition, wave classification based on atmospheric circulation patterns helps to reveal the causal mechanism and unique spatial and temporal distribution laws of different weather-type waves. This research can contribute to the green and low-carbon development of the marine economy, and also provide an important reference for the scientific advancement of disaster prevention and mitigation and engineering construction.

[0004] At present, the weather type classification methods of waves mainly include subjective weather classification and objective weather classification. They are based on factors such as sea level pressure and wind fields at different levels to identify weather backgrounds with different characteristics and then classify the wave process. Compared with subjective weather classification, objective weather classification has advantages in processing large-scale and long-time series data. The K-means clustering algorithm is often used for weather type classification of waves. This method classifies wave processes into K categories through similarity measurement, so that the data in each category are as similar as possible and the data in different categories are as different as possible. However, this method needs to set the number of clusters based on prior knowledge. Inconsistent discrimination criteria may lead to individual differences, making it difficult to obtain unified clustering results. In addition, this method is easily affected by the initial cluster center, which may reduce the convergence speed of the algorithm and may also cause the algorithm to fall into a local minimum, affecting the reliability of the clustering results. Summary of the invention

[0005] In view of this, the present invention discloses a wave dynamic weather classification method based on environmental field feature extraction. The method mainly solves the problem that the traditional weather classification method is easily affected by the initial cluster center and easily falls into the local minimum, thereby improving the stability of the classification results.

[0006] The object of the present invention is achieved by the following technical solutions:

[0007] A wave dynamic weather classification method based on environmental field feature extraction comprises the following steps:

[0008] S1. Get the wave elements of a certain site Yearly time series historical data to obtain the wind field and sea level pressure field near the site Yearly time series historical data, where n is greater than 20 years;

[0009] S2, identifying independent wave events from the time series wave height sequence based on a 5-day time window, and selecting stronger independent wave events as events to be classified according to the threshold method;

[0010] S3. Explore the regional correlation between wind speed and independent wave process peak elements under different time lag conditions to determine the time and space domains of wind field and sea level pressure field for classification;

[0011] S4. Input the determined wind field and sea level pressure field into the SOM+K-means clustering algorithm to obtain the weather classification results of independent wave events.

[0012] Preferably, the wave elements in step S1 include wave height and wave direction, and the wind field includes wind speed.

[0013] Preferably, considering the difference in variable attributes of wind speed and wave peak value elements (the former is a vector with directionality, while the latter is a scalar), step S3 includes the following steps:

[0014] S31, defined when the time lag is In the case of The latitudinal and meridional components of wind speed ;

[0015] S32, project the wind speed sequence in all possible directions, perform correlation analysis on the projection results and the wave height peak element sequence, and select the result in the direction with the maximum correlation as the latitude and longitude point The correlation measure of the two factors is calculated as follows:

[0016] ,

[0017] in, is the latitude and longitude Time lag Conditional wave peak element sequence The correlation coefficient with wind speed, Expressed as a linear correlation function, Along the direction The projected wind speed sequence is calculated according to the following formula:

[0018] ,

[0019] S33. Calculate the correlation between wind speed and wave height sequence under different time lag conditions at each longitude and latitude point in the area near the site to be classified, and determine the time domain and space domain of the wind field and sea level pressure field used for clustering input based on the correlation distribution diagram.

[0020] Preferably, step S4 comprises the following steps:

[0021] S41, normalizing the wind field and sea level pressure field input by clustering, and inputting them into the SOM neural network for training;

[0022] S42. After training, construct a heterogeneity matrix representing the characteristics of each neuron in the SOM network , as shown below:

[0023] ,

[0024] in, is the weight of each neuron in the SOM network after training, The weight for each input data;

[0025] S43. Input the heterogeneity matrix into the K-means algorithm for clustering, calculate the Calinski-Harabasz index for different cluster numbers, and determine the optimal cluster number.

[0026] Beneficial Effects

[0027] The present invention provides a wave dynamic weather typing method based on environmental field feature extraction. The method combines the feature extraction of the SOM neural network and the fast processing of the K-means clustering algorithm, optimizes the initial clustering center of the K-means with the SOM clustering result, improves the blindness of the initial clustering center selection of the K-means clustering algorithm, and has the advantages of short convergence time and high result stability. Based on the dynamic clustering method, the driving weather characteristics of the wave process at a certain site can be revealed to further clarify the physical mechanism of extreme wave generation and its time and direction characteristics, providing an important reference for the design and construction of marine and coastal engineering. In addition, the dynamic typing method is applied to a certain sea area to reveal the spatial distribution pattern of extreme waves driven by different weather conditions, thereby ensuring the safety and sustainable development of coastal areas and marine activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0029] Figure 1 It is a flow chart of the wave dynamic weather classification method based on environmental field feature extraction disclosed in the present invention;

[0030] Figure 2 is the correlation distribution between the wind speed and the peak element sequence at the research point under different time lag conditions in the embodiment;

[0031] Figure 3 The distribution of grid points in the SOM competition layer after training in the embodiment;

[0032] Figure 4 The curve of the change of Calinski-Harabasz index with the number of aggregations in the embodiment;

[0033] Figure 5 The wave height and wave direction distribution of different cluster wave processes in the embodiment;

[0034] Figure 6 Monthly distribution of different cluster wave processes in the embodiment;

[0035] Figure 7 ] are the average sea level pressure field and wind field corresponding to the peak elements of the two types of wave processes in the embodiment. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Example 1

[0038] The embodiment of the present invention discloses a wave dynamic weather classification method based on environmental field feature extraction. Figure 1 This is a flow chart of this embodiment. The research point is selected in the middle of the Yellow Sea, with longitude and latitude of 124°E, 37°N.

[0039] This embodiment includes the following steps:

[0040] S1. Obtain 40 years of time series historical data of wave elements at the station in the central Yellow Sea, and obtain 40 years of time series historical data of wind field and sea level pressure field near the station. In this example, the wave elements (wave height, wave direction) at the study point are simulated by the SWAN model, with a time range of 1979-2018 and a time step of 1 hour. The wind field (wind speed) and sea level pressure field near the study point are provided by the European Center for Medium-Range Weather Forecasts ERA5, with a time span of 40 years (1979-2018), a time step of 1 hour, and a spatial resolution of 0.25°.

[0041] S2. Identify independent wave events from the time series of wave heights based on a 5-day time window, and select stronger independent wave events as events to be classified based on the threshold method. In this example, the 5-day time window was used as the benchmark for the central Yellow Sea station to identify independent wave events in the 40-year time series, and 2.84m was initially selected as the threshold to screen out 470 stronger independent wave processes. Considering that some of these wave processes are driven by tropical cyclones, and the historical records of such weather are clear and sufficient, they can be identified by tracking tropical cyclone trajectories and matching record times. In this example, 70 independent wave processes driven by tropical cyclones were identified from 470 independent wave processes (screened out), and the remaining 400 wave processes were further classified.

[0042] S3. Explore the regional correlation between wind speed and independent wave process peak elements under different time lag conditions to determine the time and space domains of wind field and sea level pressure field for classification;

[0043] The steps include:

[0044] S31, defined when the time lag is In the case of The latitudinal and meridional components of wind speed ;

[0045] S32, project the wind speed sequence in all possible directions, perform correlation analysis on the projection results and the wave height peak element sequence, and select the result in the direction with the maximum correlation as the latitude and longitude point The correlation measure of the two factors is calculated as follows:

[0046] ,

[0047] in, is the latitude and longitude Time lag Conditional wave peak element sequence The correlation coefficient with wind speed, Expressed as a linear correlation function, Along the direction The projected wind speed sequence is calculated according to the following formula:

[0048] ,

[0049] S33. Calculate the correlation between wind speed and wave height sequence under different time lag conditions at each longitude and latitude point in the area near the site to be classified, and determine the time domain and space domain of the wind field and sea level pressure field used for clustering input based on the correlation distribution diagram.

[0050] In this example, the correlation between wind speed and wave peak element sequence under the conditions of 0 h, 12 h, 24 h and 48 h with a 21°×21° center range at the research point is explored. Figure 2 As shown. It can be seen that in terms of time lag, the overall correlation decreases rapidly after more than 12 hours. Spatially, it was found that areas with similar correlations were not regularly distributed around the study points. In addition, no longitude and latitude points with high correlation were found on the boundary, so this range is sufficient to capture the weather characteristics that drive the waves. In summary, the time lag conditions of 0 and 12 h were used as the wind field and time domain of the clustering input, and the 21°×21° center range of the study point was used as the spatial domain of the wind field of the clustering input. Similar to the time domain and space domain of the wind field, the sea level pressure field uses consistent time domain and space domain settings.

[0051] S4. Input the determined wind field and sea level pressure field into the SOM+K-means clustering algorithm to obtain the weather classification results of independent wave events.

[0052] The steps include:

[0053] S41. Normalize the clustered input wind field and sea level pressure field, and input them into the SOM neural network for training. In this example, a two-dimensional hexagonal grid is used as the competitive layer network structure of the SOM neural network, which can more effectively preserve the topological structure of the input space than a rectangular grid. By trying a variety of grid array configurations, a 10×10 grid array is selected, which enables the input data to be adaptively adjusted and can achieve relatively fast processing time. In addition, the number of training iterations is set to 1000 times, which has been verified to be sufficient to enable the SOM neural network to gradually align with the input data during the iteration process, form clusters, and maintain the topological relationship between the input data throughout the iteration process. Figure 3 The figure shows the distribution of grid points in the SOM competition layer after training, where each white hexagon represents a grid point and the numbers represent the input data classified into each grid point.

[0054] S42. After training, construct a heterogeneity matrix representing the characteristics of each neuron in the SOM network , as shown below:

[0055] ,

[0056] in, is the weight of each neuron in the SOM network after training, is the weight for each input data; in this example, the heterogeneity matrix is ​​as follows:

[0057] ,

[0058] S43, input the heterogeneity matrix into the K-means algorithm for clustering, calculate the Calinski-Harabasz index for different cluster numbers, and determine the optimal cluster number. In this example, the curve of the change of the Calinski-Harabasz index with the number of clusters is as follows: Figure 4 As shown in the figure, it can be seen that the optimal number of clusters is 2, and the Calinski-Harabasz value is the largest, about 165.31. Therefore, at this research point, 400 independent wave processes can be divided into two weather types.

[0059] Example 2

[0060] In order to verify the rationality of weather classification, the embodiment of the present invention uses MATLAB and ORIGIN software to perform statistical analysis on the time characteristics and direction characteristics of the peak elements of two types of wave processes.

[0061] Figure 5 The wave direction distribution corresponding to the two types of wave process peak elements at the research site is provided. The directional characteristics of the two types of wave process peak elements are significant. The first type is concentrated in the northwest direction (standard deviation is 15.26°), while the second type is more dispersed (standard deviation is 90.41°), mainly concentrated in the southeast direction. In addition, the two types of wave process peak elements also show significant differences in time distribution, and the seasonal differences are obvious. Figure 6 The monthly distribution of the two types of wave process peak elements at the research site is provided. The first type of wave process peak elements mainly occur from October to March, with typical seasons being winter and autumn; while the second type of wave process peak elements mainly occur from April to August, with typical seasons being spring and summer. Based on the above analysis, although the wave process peak elements are not directly involved in the weather type clustering, the time and direction characteristics of the wave process peak elements in the same cluster in the clustering results have obvious homogeneity, and the differences between different clusters are significant. These conclusions fully confirm the rationality of the wave process weather type classification, and the two types of wave processes may be driven by two weather systems.

[0062] To confirm the above speculation, the MATLAB software was used in this embodiment to further calculate the average sea level pressure field and wind field corresponding to the peak elements of the two types of wave processes when the time lag At is 0. The results are as follows: Figure 7 As shown. It can be seen that the first type of wave process may be driven by a cold wave. The northwest of the study point is controlled by a strong high-pressure system, and the southeast is a low-pressure system. This type of pressure system produces strong northwest winds near the study point. Therefore, the direction distribution of the first type of wave process is mainly northwest. The second type of wave process may be driven by a temperate cyclone. The study point is controlled by the low-pressure center of the temperate cyclone, and the periphery of the temperate cyclone is a high-pressure system. The vicinity of this point is mainly affected by the southeast strong wind in the east of the temperate cyclone. Therefore, the direction distribution of the second type of wave process is mainly southeast. It was also found that the western part of some temperate cyclones is a strong wind area. Therefore, the wave processes in the northwest direction are not all driven by cold waves. A few are driven by temperate cyclones. This conclusion is consistent with Figure 5 The directional distribution characteristics of the second type of wave process are highly consistent. In addition, the clustering results of this method were compared with those of the traditional clustering method. After 1000 tests, it was found that the classification stability can be improved by about 25%. In summary, the wave dynamic weather classification method based on environmental field feature extraction can classify a variety of weather-type wave processes with different weather situation characteristics from independent wave processes after screening out tropical cyclone-type wave processes, providing an important reference for the design, construction and protection of marine and coastal engineering.

[0063] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0064] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A wave dynamic weather classification method based on environmental field feature extraction, characterized in that: The following steps are involved: S1. Obtain n-year time series historical data of wave elements at a certain station, and obtain n-year time series historical data of wind field and sea level pressure field near the station, where n is greater than 20 years; S2, identifying independent wave events from the time series wave height sequence based on a 5-day time window, and selecting stronger independent wave events as events to be classified according to the threshold method; S3. Explore the regional correlation between wind speed and independent wave process peak elements under different time lag conditions to determine the time and space domains of wind field and sea level pressure field for classification; The steps include: S31, defined when the time lag is In the case of The latitudinal and meridional components of wind speed ; S32, project the wind speed sequence in all possible directions, perform correlation analysis on the projection results and the wave height peak element sequence, and select the result in the direction with the maximum correlation as the latitude and longitude point The correlation measure of the two factors is calculated as follows: ， in, is the latitude and longitude Time lag Conditional wave peak element sequence The correlation coefficient with wind speed, Expressed as a linear correlation function, Along the direction The projected wind speed sequence is calculated according to the following formula: ， S33, calculating the correlation between wind speed and wave height sequence under different time lag conditions at each longitude and latitude point in the area near the site to be classified, and determining the time domain and space domain of the wind field and sea level pressure field for clustering input according to the correlation distribution diagram; S4. Input the determined wind field and sea level pressure field into the SOM+K-means clustering algorithm to obtain the weather classification results of independent wave events.

2. The wave dynamic weather classification method based on environmental field feature extraction according to claim 1 is characterized in that: The wave elements in step S1 include wave height and wave direction, and the wind field includes wind speed.

3. The wave dynamic weather classification method based on environmental field feature extraction according to any one of claims 1 or 2 is characterized in that: The step S4 includes the following steps S41 normalizes the clustered input wind field and sea level pressure field and inputs them into the SOM neural network for training; S42. After training, construct a heterogeneity matrix representing the characteristics of each neuron in the SOM network , as shown below: ， in, is the weight of each neuron in the SOM network after training, The weight for each input data; S43. Input the heterogeneity matrix into the K-means algorithm for clustering, calculate the Calinski-Harabasz index for different cluster numbers, and determine the optimal cluster number.

Images