A severe convective weather similarity analysis method

Abstract

The application discloses a strong convective weather similarity analysis method, which comprises the following steps: obtaining current echo intensity data and current waveform data in current large atmospheric echo data, and obtaining historical echo intensity data of each strong convective weather; preprocessing the current echo intensity data and the current waveform data to obtain preprocessed echo intensity data and preprocessed waveform data; calculating the Hamming distance and the cosine similarity corresponding to each strong convective weather according to the preprocessed echo intensity data and the historical echo intensity data of each strong convective weather; determining the waveform similarity corresponding to each strong convective weather according to the preprocessed waveform data; and determining the target strong convective weather similarity based on the Hamming distance, the cosine similarity and the waveform similarity. The application can improve the accuracy of judging strong convective weather, thereby reducing the misjudgment and missed judgment of strong convective weather.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of weather monitoring, in particular to a strong convective weather similarity analysis method. BACKGROUND

[0002] Strong convective weather, as a weather phenomenon with great destructive power in nature, is known for its suddenness, intensity, and wide impact. Such weather, such as thunderstorm, hail, short-term heavy rain, and even tornado, often brings great impact to the environment and human society in a short time. In the monitoring and analysis of strong convective weather, radar technology plays a crucial role. Radar can obtain dynamic information of weather systems in real time by emitting electromagnetic waves and receiving signals reflected from weather targets. Among them, echo intensity and waveform are two key characteristic parameters.

[0003] The existing technology has several defects in strong convective similarity analysis, mainly reflected in the insufficient quantitative evaluation of echo characteristics and the limitation of waveform analysis method. Traditional methods usually only rely on simple intensity indicators, and fail to deeply mine the timing changes of echoes, so the overall understanding of complex weather patterns and the warning ability are limited. In addition, the existing waveform analysis mostly uses linear models, which lack the capture of nonlinear features, leading to misjudgment and missed judgment of strong convective weather in actual application. SUMMARY

[0004] The present application aims to provide a strong convective weather similarity analysis method, which can improve the accuracy of judging strong convective weather, thereby reducing the misjudgment and missed judgment of strong convective weather.

[0005] The present application provides a strong convective weather similarity analysis method, which comprises:

[0006] Obtaining current echo intensity data and current waveform data in current large-scale echo data, and obtaining historical echo intensity data of each strong convective weather;

[0007] Preprocessing the current echo intensity data and the current waveform data to obtain preprocessed echo intensity data and preprocessed waveform data;

[0008] According to the preprocessed echo intensity data and the historical echo intensity data of each strong convective weather, the corresponding Hamming distance and cosine similarity of each strong convective weather are calculated;

[0009] According to the preprocessed waveform data, the waveform similarity corresponding to each strong convective weather is determined;

[0010] Based on the Hamming distance, the cosine similarity and the waveform similarity, the target strong convective weather similarity is determined.

[0011] Compared with the prior art, the present application has the following beneficial effects:

[0012] The method can more comprehensively reflect the change characteristics of the echo intensity by calculating the Hamming distance and the cosine similarity corresponding to each severe convective weather according to the preprocessed echo intensity data and the historical echo intensity data of each severe convective weather, and using the combination of the Hamming distance and the cosine similarity to evaluate the similarity of the echo intensity. Then, the waveform similarity corresponding to each severe convective weather is determined according to the preprocessed waveform data, and the target severe convective weather similarity is determined based on the Hamming distance, the cosine similarity and the waveform similarity. In this way, by comprehensively considering the Hamming distance and the cosine similarity calculated based on the echo intensity data and considering the waveform similarity calculated based on the waveform data, the target severe convective weather similarity is determined together, which can improve the accuracy of the determined target severe convective weather similarity, so that the current echo intensity data and the current waveform data correspond to which kind of severe convective weather can be judged according to the accurate target severe convective weather similarity, which can improve the accuracy of judging the severe convective weather, thereby reducing the misjudgment and missed judgment of the severe convective weather.

[0013] In some embodiments, the calculating the Hamming distance and the cosine similarity corresponding to each severe convective weather according to the preprocessed echo intensity data and the historical echo intensity data of each severe convective weather comprises:

[0014] The preprocessed echo intensity data is represented as a first echo intensity binary sequence, and the historical echo intensity data of each severe convective weather is represented as a second echo intensity binary sequence.

[0015] The Hamming distance corresponding to each severe convective weather is calculated according to the first echo intensity binary sequence and the second echo intensity binary sequence.

[0016] In some embodiments, the Hamming distance corresponding to each severe convective weather is calculated according to the first echo intensity binary sequence and the second echo intensity binary sequence in the following manner:

[0017]

[0018] Wherein, A represents the first echo intensity binary sequence, B represents the second echo intensity binary sequence, a i represents the i th element in the first echo intensity binary sequence, b i represents the i th element in the second echo intensity binary sequence, and n represents the total number of elements in the echo intensity binary sequence.

[0019] In some embodiments, the Hamming distance and the cosine similarity corresponding to each severe convective weather are calculated according to the preprocessed echo intensity data and the historical echo intensity data of each severe convective weather, including:

[0020] The preprocessed echo intensity data is represented as a first echo intensity vector, and the historical echo intensity data of each severe convective weather is represented as a second echo intensity vector;

[0021] The cosine similarity corresponding to each severe convective weather is calculated according to the first echo intensity vector and the second echo intensity vector.

[0022] In some embodiments, the cosine similarity corresponding to each severe convective weather is calculated according to the first echo intensity vector and the second echo intensity vector in the following manner:

[0023]

[0024] wherein X represents the first echo intensity vector, Y represents the second echo intensity vector, x i represents the i-th component in the first echo intensity vector, y i represents the i-th component in the second echo intensity vector, and k represents the total number of components in the echo intensity vector.

[0025] In some embodiments, the waveform similarity corresponding to each severe convective weather is determined according to the preprocessed waveform data, including:

[0026] The preprocessed waveform data is input into a trained BP neural network model to obtain the waveform similarity corresponding to each severe convective weather output by the trained BP neural network model; wherein the trained BP neural network model is trained using waveform data of different single severe convective weather.

[0027] In some embodiments, the target severe convective weather similarity is determined based on the Hamming distance, the cosine similarity and the waveform similarity, including:

[0028] The Hamming distance, the cosine similarity and the waveform similarity corresponding to each severe convective weather are obtained;

[0029] The severe convective weather similarity corresponding to each severe convective weather is calculated based on the waveform similarity, the Hamming distance and the cosine similarity corresponding to each severe convective weather;

[0030] The severe convective weather similarities corresponding to each severe convective weather are sorted, and the maximum severe convective weather similarity is selected as the target severe convective weather similarity.

[0031] In some embodiments, the strong convective weather similarity corresponding to each type of strong convective weather is calculated based on the waveform similarity, the Hamming distance and the cosine similarity corresponding to each type of strong convective weather in the following manner:

[0032] S = aS H + bS C + gS BP

[0033] wherein S represents the strong convective weather similarity, a, b and g represent weight coefficients, S H represents the Hamming distance, S C represents the cosine similarity, and S BP represents the waveform similarity.

[0034] In some embodiments, the preprocessing of the current echo intensity data and the current waveform data to obtain preprocessed echo intensity data and preprocessed waveform data comprises:

[0035] denoising the current echo intensity data and the current waveform data to obtain denoised echo intensity data and denoised waveform data;

[0036] normalizing the denoised echo intensity data and the denoised waveform data to obtain normalized echo intensity data and normalized waveform data;

[0037] extracting features from the normalized echo intensity data and the normalized waveform data to obtain preprocessed echo intensity data and preprocessed waveform data.

[0038] In some embodiments, the denoised echo intensity data and the denoised waveform data are normalized in the following manner:

[0039]

[0040] wherein x1’ represents the normalized echo intensity data, x1 represents the denoised echo intensity data, x1 min represents the minimum value in the denoised echo intensity data, x1 max represents the maximum value in the denoised echo intensity data, x2’ represents the normalized waveform data, x2 represents the denoised waveform data, x2 min represents the minimum value in the denoised waveform data, x2 max represents the maximum value in the denoised waveform data. BRIEF DESCRIPTION OF DRAWINGS

[0041] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the following drawings in which:

[0042] Figure 1 is a flowchart of an embodiment of a strong convective weather similarity analysis method provided by the present application;

[0043] Figure 2 is a flowchart of an embodiment of a strong convective weather similarity analysis method provided by the present application;

[0044] Figure 3 is a radar echo map in an embodiment of a strong convective weather similarity analysis method provided by the present application. DETAILED DESCRIPTION

[0045] Embodiments of the present application are described in detail below with reference to the attached drawing figures, wherein the same or like reference numerals and characters are used throughout the drawing figures to designate the same components. The embodiments described below are examples only, and are not intended to limit the present application, unless otherwise explicitly indicated herein.

[0046] In the description of the present application, if there is a description to the first, second, etc., it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the order of the indicated technical features.

[0047] In the description of the present application, it should be understood that, in relation to the orientation description, for example, the orientation or position relationship indicated by up, down, etc. is based on the orientation or position relationship shown in the drawings, and is only for the purpose of facilitating the description of the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the device or element indicated must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application.

[0048] In the description of the present application, it should be noted that, unless otherwise explicitly limited, the words such as setting, installing, connecting, etc. should be broadly understood, and the person skilled in the art can reasonably determine the specific meaning of the above words in the present application in combination with the specific content of the technical solution.

[0049] Severe convective weather, as a kind of weather phenomenon with great destructive power in nature, is known for its suddenness, intensity and wide influence. Such weather, such as thunderstorm, hail, short-time heavy rain and even tornado, often brings great impact on the environment and human society in a short time. In the monitoring and analysis of severe convective weather, radar technology plays a crucial role. Radar can obtain dynamic information of weather system in real time by transmitting electromagnetic waves and receiving signals reflected from weather targets. Among them, echo intensity and waveform are two key characteristic parameters.

[0050] The existing technology has several defects in the analysis of severe convective similarity, mainly reflected in the insufficient quantitative evaluation of echo characteristics and the limitation of waveform analysis method. The traditional method usually only relies on simple intensity index and fails to deeply mine the timing changes of echo, so the overall understanding of complex weather patterns and the warning ability are limited. In addition, the existing waveform analysis mostly uses linear model, which lacks the capture of nonlinear features, leading to misjudgment and missed judgment of severe convective weather in actual application.

[0051] To solve the above-mentioned problem that the linear model lacks the capture of nonlinear features, leading to misjudgment and missed judgment of severe convective weather in actual application, the present application provides a severe convective weather similarity analysis method.

[0052] With reference to Figure 1 The embodiment of the present application provides a severe convective weather similarity analysis method, which comprises the following steps:

[0053] Step S100, obtaining current echo intensity data and current waveform data in current large echo data, and obtaining historical echo intensity data of each severe convective weather;

[0054] Step S200, preprocessing the current echo intensity data and the current waveform data to obtain preprocessed echo intensity data and preprocessed waveform data;

[0055] Step S300, calculating the Hamming distance and cosine similarity corresponding to each severe convective weather according to the preprocessed echo intensity data and the historical echo intensity data of each severe convective weather;

[0056] Step S400, determining the waveform similarity corresponding to each severe convective weather according to the preprocessed waveform data;

[0057] Step S500, determining the target severe convective weather similarity based on the Hamming distance, cosine similarity and waveform similarity.

[0058] In this embodiment, the Hamming distance and cosine similarity corresponding to each severe convective weather event are calculated based on the preprocessed echo intensity data and the historical echo intensity data for each severe convective weather event. A method combining the Hamming distance and cosine similarity is used to evaluate echo intensity similarity, which can more comprehensively reflect the changing characteristics of echo intensity. The waveform similarity corresponding to each severe convective weather event is then determined based on the preprocessed waveform data. The target severe convective weather event similarity is determined based on the Hamming distance, cosine similarity, and waveform similarity. Thus, by comprehensively considering the Hamming distance and cosine similarity calculated based on the echo intensity data, as well as the waveform similarity calculated based on the waveform data, the target severe convective weather event similarity is determined. This improves the accuracy of the determined target severe convective weather event similarity. Consequently, the current echo intensity data and the current waveform data can be determined based on the accurate target severe convective weather similarity, thereby improving the accuracy of severe convective weather event determination and reducing the incidence of misjudgments and missed detections of severe convective weather events.

[0059] The above-mentioned acquisition of the current echo intensity data and the current waveform data in the current atmospheric echo data, as well as the acquisition of the historical echo intensity data of each severe convective weather, can be achieved by acquiring the current echo intensity data and the current waveform data in the current atmospheric echo data through radar, as well as the acquisition of the historical echo intensity data of each severe convective weather.

[0060] The above-mentioned preprocessing of the current echo intensity data and the current waveform data may be performed by using a denoising, normalization or other method to preprocess the current echo intensity data and the current waveform data.

[0061] The waveform similarity corresponding to each severe convective weather event can be determined based on the preprocessed waveform data. The preprocessed waveform data and historical waveform data for each severe convective weather event can be vectorized, and then the waveform similarity corresponding to each severe convective weather event can be calculated using cosine similarity based on the vectorized waveform data and the historical waveform data for each severe convective weather event. A neural network model can also be used to predict the preprocessed waveform data to determine the waveform similarity corresponding to each severe convective weather event. The neural network model can be a prior art neural network model known to those skilled in the art, such as a BP neural network model.

[0062] In some embodiments, the Hamming distance and cosine similarity corresponding to each severe convective weather event are calculated based on the preprocessed echo intensity data and the historical echo intensity data of each severe convective weather event, including:

[0063] The pre-processed echo intensity data is represented as a first echo intensity binary sequence, and the historical echo intensity data of each severe convective weather is represented as a second echo intensity binary sequence;

[0064] According to the first echo intensity binary sequence and the second echo intensity binary sequence, a Hamming distance corresponding to each severe convective weather is calculated.

[0065] In this embodiment, the Hamming distance is used to measure the number of different bits between two binary strings, and can intuitively reflect the difference between the echo intensity data. The Hamming distance can measure the difference between two groups of echo data from a discrete perspective, focuses on the presence or absence of echo data, and is suitable for processing discrete features.

[0066] In some embodiments, according to the first echo intensity binary sequence and the second echo intensity binary sequence, a Hamming distance corresponding to each severe convective weather is calculated in the following manner:

[0067]

[0068] Wherein, A represents the first echo intensity binary sequence, B represents the second echo intensity binary sequence, a i represents the i-th element in the first echo intensity binary sequence, b i represents the i-th element in the second echo intensity binary sequence, and n represents the total number of elements in the echo intensity binary sequence.

[0069] In some embodiments, according to the preprocessed echo intensity data and the historical echo intensity data of each severe convective weather, a Hamming distance and a cosine similarity corresponding to each severe convective weather are calculated, including:

[0070] The preprocessed echo intensity data is represented as a first echo intensity vector, and the historical echo intensity data of each severe convective weather is represented as a second echo intensity vector;

[0071] According to the first echo intensity vector and the second echo intensity vector, a cosine similarity corresponding to each severe convective weather is calculated.

[0072] In this embodiment, the cosine similarity evaluates the similarity between two vectors by calculating the cosine of the included angle, is not sensitive to the magnitude of the data, pays more attention to the difference in direction, and considers the directionality and overall trend of the data.

[0073] In some embodiments, according to the first echo intensity vector and the second echo intensity vector, a cosine similarity corresponding to each severe convective weather is calculated in the following manner:

[0074]

[0075] Wherein, X represents the first echo intensity vector, Y represents the second echo intensity vector, x i represents the i-th component in the first echo intensity vector, and yi represents the i-th component in the second echo intensity vector, k represents the total number of components in the echo intensity vector.

[0076] In some embodiments, according to the preprocessed waveform data, the waveform similarity corresponding to each severe convective weather is determined, including:

[0077] The preprocessed waveform data is input into the trained BP neural network model to obtain the waveform similarity corresponding to each severe convective weather output by the trained BP neural network model; wherein the trained BP neural network model is trained by using waveform data of different single severe convective weather.

[0078] In this embodiment, the BP neural network model is a powerful pattern recognition tool, which can extract deep features from complex waveform data and analyze the waveform features in depth through its nonlinear mapping capability. Through BP neural network training, the trained BP neural network model can accurately identify the waveform features under different weather conditions, thereby improving the accuracy of waveform similarity calculation.

[0079] In some embodiments, based on the Hamming distance, the cosine similarity and the waveform similarity, the target severe convective weather similarity is determined, including:

[0080] The Hamming distance, the cosine similarity and the waveform similarity corresponding to each severe convective weather are obtained;

[0081] Based on the waveform similarity, the Hamming distance and the cosine similarity corresponding to each severe convective weather, the severe convective weather similarity corresponding to each severe convective weather is calculated;

[0082] The severe convective weather similarity corresponding to each severe convective weather is sorted, and the maximum value of the severe convective weather similarity is selected as the target severe convective weather similarity.

[0083] In this embodiment, by comprehensively considering the Hamming distance and the cosine similarity calculated based on the echo intensity data, and considering the waveform similarity calculated based on the waveform data, the target severe convective weather similarity is determined, which can improve the accuracy of the determined target severe convective weather similarity. According to the accurate target severe convective weather similarity, it can be judged which kind of severe convective weather the current echo intensity data and the current waveform data correspond to, which can improve the accuracy of judging the severe convective weather, thereby reducing the misjudgment and omission of the severe convective weather.

[0084] In some embodiments, the severe convective weather similarity corresponding to each severe convective weather is calculated based on the waveform similarity, the Hamming distance and the cosine similarity corresponding to each severe convective weather in the following manner:

[0085] S = aS + bD + cCH + βS C + γS BP

[0086] wherein S represents the strong convective weather similarity, α, β and γ represent the weight coefficients, S H represents the Hamming distance, S C represents the cosine similarity, S BP represents the waveform similarity.

[0087] In some embodiments, the current echo intensity data and the current waveform data are preprocessed to obtain preprocessed echo intensity data and preprocessed waveform data, including:

[0088] The current echo intensity data and the current waveform data are denoised to obtain denoised echo intensity data and denoised waveform data;

[0089] The denoised echo intensity data and the denoised waveform data are normalized to obtain normalized echo intensity data and normalized waveform data;

[0090] The normalized echo intensity data and the normalized waveform data are feature extracted to obtain the preprocessed echo intensity data and the preprocessed waveform data.

[0091] In this embodiment, by preprocessing the current echo intensity data and the current waveform data, a good data foundation is laid for later Hamming distance and cosine similarity calculation, waveform similarity calculation and determination of the target strong convective weather similarity, ensuring the accuracy of the echo intensity data and the waveform data.

[0092] In some embodiments, the denoised echo intensity data and the denoised waveform data are normalized by the following method:

[0093]

[0094] wherein x1' represents the normalized echo intensity data, x1 represents the denoised echo intensity data, x1 min represents the minimum value in the denoised echo intensity data, x1 max represents the maximum value in the denoised echo intensity data, x2' represents the normalized waveform data, x2 represents the denoised waveform data, x2 min represents the minimum value in the denoised waveform data, x2 max represents the maximum value in the denoised waveform data.

[0095] For the convenience of those skilled in the art, a set of best embodiments is provided as follows:

[0096] The embodiment effectively evaluates the change of echo intensity by combining Hamming distance and cosine similarity, deeply analyzes waveform features by using a BP neural network model, and uses the powerful nonlinear mapping and pattern recognition capability of the BP neural network model to realize more accurate analysis of severe convective weather. Through the self-learning and optimization process, the BP neural network model can automatically extract the key features in the waveform and establish the mapping relationship between the echo waveform and the severe convective weather type, thereby improving the accuracy and efficiency of weather forecasting.

[0097] In summary, with reference to Figure 2 , the purpose of the embodiment is to overcome the shortcomings of the prior art, realize more accurate and comprehensive monitoring and analysis of severe convective weather by introducing new analysis methods and algorithms, and thereby enhance the response capability to the influence of severe convective weather.

[0098] 1. Data preprocessing.

[0099] (1) Data acquisition and extraction.

[0100] Radar is used to obtain echo data in the atmosphere, mainly including echo intensity data and waveform information (i.e. waveform data). Echo intensity reflects the reflection ability of the target object, while waveform provides more detailed features of the target object. The original data of echo intensity is a radar echo map, with reference to Figure 3 .

[0101] (2) Data denoising.

[0102] Filtering techniques (e.g. Gaussian filtering) are used to denoise the original data (i.e. echo intensity data and waveform data obtained by radar) to remove noise caused by environmental factors or equipment errors, ensuring data point accuracy. Gaussian filtering uses a Gaussian function to perform weighted averaging on the data, making the data smoother.

[0103] (3) Data normalization.

[0104] The denoised echo intensity data and waveform data are normalized to standardize the data to the same scale range, usually 0, 1, to facilitate subsequent similarity calculation and model training. The normalization formula is:

[0105]

[0106] where x1' represents the normalized echo intensity data, x1 represents the denoised echo intensity data, x1 min represents the minimum value in the denoised echo intensity data, x1 max represents the maximum value in the denoised echo intensity data, x2' represents the normalized waveform data, x2 represents the denoised waveform data, x2 minIndicates the minimum value in the denoised waveform data, x2 max Indicates the maximum value in the denoised waveform data.

[0107] (4) Feature extraction.

[0108] Feature values ​​are extracted from the normalized data, including statistical features such as peak value, mean value, and variance extracted from the echo intensity data, and features such as frequency component extracted from the waveform data. The waveform is converted from the time domain to the frequency domain through Fourier transform to obtain the frequency component.

[0109] 2. Calculation of echo intensity similarity.

[0110] The echo intensity similarity calculation adopts a method combining Hamming distance and cosine similarity to ensure accurate comparison of echo intensities of different severe convective weather.

[0111] (1) Hamming distance calculation.

[0112] The currently acquired echo intensity data is expressed as a binary sequence (i.e., the first echo intensity binary sequence) a1, a2, ..., a n , the historical echo intensity data is represented as a binary sequence (i.e., the second echo intensity binary sequence) b1, b2, ..., b n The Hamming distance is used to measure the degree of difference between two binary sequences.

[0113]

[0114] Among them, A represents the first echo intensity binary sequence, B represents the second echo intensity binary sequence, a i Represents the i-th element in the binary sequence of the first echo intensity, b i represents the i-th element in the second echo intensity binary sequence, n represents the total number of elements in the echo intensity binary sequence, d(A, B) is the obtained Hamming distance, and the closer the d(A, B) value is to 0, the better the similarity.

[0115] (2) Cosine similarity calculation.

[0116] The currently acquired echo intensity data and the historical echo intensity data are represented as vector X (ie, the first echo intensity vector) and vector Y (ie, the second echo intensity vector), and the cosine similarity is used to measure the similarity between the two vectors.

[0117]

[0118] Among them, x j and y irespectively represent the i-th component of vector X and vector Y, and cos(x, y) represents the cosine of the angle between variables x and y. If the correlation between variables is close, the angle approaches 0 and the cosine approaches 1; otherwise, it approaches 0.

[0119] 3. Waveform similarity analysis.

[0120] Waveform similarity analysis uses a BP neural network model to identify the similarity of waveforms by training the model.

[0121] (1) Construction of BP neural network model.

[0122] 1) Network structure.

[0123] The construction of a BP neural network (Back propagation Neural Network) refers to building a multi-layer perceptron model and training it through a backpropagation algorithm to solve non-linear mapping problems. This process includes defining the number of network layers, nodes, activation functions, weight initialization, error calculation, and training methods.

[0124] Input layer: The number of nodes is n, representing the dimension of the input waveform data.

[0125] Hidden layer: One or more hidden layers, with the number of nodes in each layer set according to actual needs, generally m.

[0126] Output layer: The number of nodes is 1, representing the waveform similarity.

[0127] 2) Activation function.

[0128] The activation function is used for non-linear transformation between the hidden layer and the output layer. Common activation functions include the Sigmoid function and the ReLU function, where the Sigmoid function is suitable for scenarios where output values are between 0 and 1, and the ReLU function is suitable for handling high-dimensional data. By introducing the activation function, the BP neural network can capture the non-linear patterns in strong convective weather data, especially the complex relationship between echo intensity over time and space.

[0129] (2) BP neural network training.

[0130] 1) Forward propagation.

[0131] Forward propagation is the process of calculating the output of the model based on the input data. The output of each layer of nodes is added through weights and biases, and then processed through an activation function, passing layer by layer to the output layer. The output of each node is calculated by the formula:

[0132]

[0133] where o jThe output of the jth hidden layer node is w ij The weight from the input layer to the hidden layer is x i The input is b j The bias is f, and the activation function is f.

[0134] 2) Error calculation.

[0135] The error of the output layer refers to the difference between the model output and the actual target, and the mean square error (MSE) is often used to calculate it. To calculate the error of the output layer, the formula is:

[0136]

[0137] where E is the error function, y k is the expected output, o k is the actual output.

[0138] 3) Error backpropagation.

[0139] Error backpropagation is a process of adjusting weights and biases layer by layer based on output errors. The BP algorithm calculates the gradient of the error and updates the parameters of each layer in reverse, so that the network gradually approaches the target output. According to the error backpropagation algorithm, the gradient of the weight and bias is calculated, and the weight and bias are updated, the formula is:

[0140]

[0141] where η is the learning rate, indicates the bias, and the error between the output and the true value is calculated to update the weight and bias in reverse.

[0142] 4. Similarity comprehensive analysis.

[0143] The echo intensity similarity (Hammig distance and cosine similarity) and waveform similarity (BP neural network output) are combined to calculate the comprehensive similarity, so as to realize the accurate identification of severe convective weather.

[0144] S = αS H + βS C + γS BP

[0145] where S H represents the Hammig distance similarity, and represents the binary difference of echo intensity; S C represents the cosine similarity, and represents the vector similarity of echo intensity; S BP represents the waveform similarity of the BP neural network output; α, β and γ are weight coefficients, and satisfy α + β + γ = 1. According to the actual demand and historical data, the weight coefficients are determined through experiments, and the machine learning algorithm is used to optimize the weight to improve the accuracy and robustness of the similarity calculation.

[0146] In this embodiment, echo intensity is an important indicator for identifying severe convective weather. By combining Hamming distance and cosine similarity, the similarity of echo intensity can be more comprehensively evaluated. Hamming distance can measure the difference between two sets of echo data from a discrete perspective, focusing on the presence or absence of echo data, and is suitable for processing discrete features. Cosine similarity, on the other hand, evaluates the vector angle of echo data from an angle, mainly considering the directionality and overall trend of the data. The combination of the two makes the echo intensity analysis focus on both local differences and overall similarity, forming a multi-dimensional and three-dimensional basis for severe convective weather identification.

[0147] However, echo intensity is only a part of the characteristics of severe convective weather. In order to further improve the identification accuracy, it is particularly important to add BP neural network for waveform similarity analysis. As a powerful pattern recognition tool, BP neural network can extract deep features from complex waveform data and perform in-depth analysis of waveform features through its nonlinear mapping capability. Through training, BP neural network can accurately identify waveform features under different weather conditions, thereby improving the accuracy of waveform similarity calculation.

[0148] This comprehensive similarity calculation method that combines echo intensity similarity and waveform similarity reflects the multi-level data fusion in severe convective weather analysis. By effectively integrating information at different levels, comprehensive similarity calculation not only improves the dependence on a single data source, but also enhances the accuracy and reliability of the analysis through cross-validation and complementary of multi-source information. Especially in the case of severe convective weather, a complex meteorological phenomenon, traditional single similarity calculation methods often fail to cope with its high complexity and nonlinear changes, while this comprehensive similarity method can comprehensively analyze the weather evolution process from multiple angles, greatly improving the precision of severe convective weather identification.

[0149] Combining traditional Hamming distance and cosine similarity with BP neural network breaks the limitations of single similarity calculation and builds a new identification framework integrating statistics, geometry and deep learning. This method not only has uniqueness in the accurate analysis of severe convective weather, but also provides an expandable and innovative direction for future applications in other complex meteorological systems. By combining BP neural network with Hamming distance and cosine similarity, not only can the changes in echo intensity be preliminarily evaluated, but also the complex nonlinear relationships of waveform features can be further captured through deep learning. This multi-level and multi-dimensional similarity analysis method provides an innovative solution for the accurate prediction and evaluation of severe convective weather, effectively improving the scientificity and practicality of severe convective weather similarity analysis.

[0150] Compared with the prior art, the embodiment has the following beneficial effects:

[0151] 1) Comprehensive evaluation of echo intensity.

[0152] The embodiment of the present application adopts a method combining Hamming distance and cosine similarity to evaluate the similarity of echo intensity. Hamming distance is used to measure the number of different bits between two binary strings, which can intuitively reflect the differences between echo intensity data; while cosine similarity evaluates the similarity between two vectors by calculating the cosine value of the included angle, which is not sensitive to the magnitude of data and pays more attention to the difference in direction. The combination of the two can more comprehensively reflect the characteristics of the change of echo intensity.

[0153] 2) Waveform feature extraction and classification.

[0154] The embodiment of the present application uses BP neural network (Back Propagation Neural Network) to extract and classify the waveform features of radar echoes. BP neural network has strong non-linear mapping ability and pattern recognition ability, which can automatically extract key features from waveform data and establish the mapping relationship between waveform and severe convective weather type. This method can effectively deal with complex and variable waveform patterns, and improve the accuracy and efficiency of classification.

[0155] 3) Similarity analysis method.

[0156] Based on the above analysis of echo intensity and waveform, the embodiment of the present application proposes a new similarity analysis method for evaluating the similarity between different severe convective weather processes. This method not only considers the change of echo intensity, but also integrates the information of waveform features, which can more comprehensively reflect the evolution law and characteristics of weather systems.

[0157] The above describes the embodiments of the present application in combination with the drawings, but the present application is not limited to the above embodiments, and various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the purpose of the present application.

Claims

1. A method for similarity analysis of severe convective weather, characterized in that: The method comprises: obtaining current echo intensity data and current waveform data in current large area echo data, and obtaining historical echo intensity data of each severe convective weather; preprocessing the current echo intensity data and the current waveform data to obtain preprocessed echo intensity data and preprocessed waveform data; According to the preprocessed echo intensity data and the historical echo intensity data of each severe convective weather, the Hamming distance and the cosine similarity corresponding to each severe convective weather are calculated, including: The preprocessed echo intensity data is represented as a first echo intensity binary sequence and a first echo intensity vector, and the historical echo intensity data of each severe convective weather is represented as a second echo intensity binary sequence and a second echo intensity vector; According to the first echo intensity binary sequence and the second echo intensity binary sequence, the Hamming distance corresponding to each severe convective weather is calculated, specifically: in, represents the first echo intensity binary sequence, represents the second echo intensity binary sequence, Indicates the first echo intensity binary sequence elements, Indicates the first echo intensity in the second echo intensity binary sequence elements, Indicates the total number of elements in the binary sequence of echo intensity; According to the first echo intensity vector and the second echo intensity vector, the cosine similarity corresponding to each severe convective weather is calculated; According to the preprocessed waveform data, the waveform similarity corresponding to each severe convective weather is determined; Based on the Hamming distance, the cosine similarity and the waveform similarity, the target severe convective weather similarity is determined.

2. The severe weather similarity analysis method of claim 1, wherein, The cosine similarity corresponding to each severe convective weather is calculated according to the first echo intensity vector and the second echo intensity vector in the following way: wherein, denotes the first echo intensity vector, denotes the second echo intensity vector, denotes the i-th component of the first echo intensity vector, denotes the i-th component of the second echo intensity vector, denotes the i-th component of the first echo intensity vector, denotes the i-th component of the second echo intensity vector, denotes the total number of components in the echo intensity vector.

3. The severe weather similarity analysis method of claim 1, wherein, The determination of the waveform similarity corresponding to each severe convective weather according to the preprocessed waveform data comprises: inputting the preprocessed waveform data into a trained BP neural network model to obtain the waveform similarity corresponding to each severe convective weather output by the trained BP neural network model; wherein the trained BP neural network model is trained by using waveform data of different single severe convective weather.

4. The severe weather similarity analysis method of claim 1, wherein, The determination of the target severe convective weather similarity based on the Hamming distance, the cosine similarity and the waveform similarity comprises: obtaining the Hamming distance, the cosine similarity and the waveform similarity corresponding to each severe convective weather; Based on the waveform similarity, the Hamming distance and the cosine similarity corresponding to each severe convective weather, the severe convective weather similarity corresponding to each severe convective weather is calculated; The severe convective weather similarity corresponding to each severe convective weather is sorted, and the maximum value of the severe convective weather similarity is selected as the target severe convective weather similarity.

5. The severe weather similarity analysis method of claim 4, wherein, The severe convective weather similarity corresponding to each severe convective weather is calculated based on the waveform similarity, the Hamming distance and the cosine similarity corresponding to each severe convective weather in the following way: wherein, represents the strong convection weather similarity, , and represents a weight coefficient, represents the Hamming distance, represents the cosine similarity, represents the waveform similarity.

6. The severe weather similarity analysis method of claim 1, wherein, The preprocessing of the current echo intensity data and the current waveform data to obtain preprocessed echo intensity data and preprocessed waveform data comprises: de-noising the current echo intensity data and the current waveform data to obtain de-noised echo intensity data and de-noised waveform data; normalizing the denoised echo intensity data and the denoised waveform data to obtain normalized echo intensity data and normalized waveform data; extracting features from the normalized echo intensity data and the normalized waveform data to obtain preprocessed echo intensity data and preprocessed waveform data.

7. The severe-weather similarity analysis method of claim 6, wherein, The denoised echo intensity data and the denoised waveform data are normalized in the following manner: wherein, denotes the normalized echo intensity data, denotes the de-noised echo intensity data, denotes the minimum value in the de-noised echo intensity data, denotes the maximum value in the de-noised echo intensity data, denotes the normalized waveform data, denotes the de-noised waveform data, denotes the minimum value in the de-noised waveform data, denotes the maximum value in the de-noised waveform data.

Images