Deep Learning-Based Meteorological Report Generation Method and System

Abstract

The present application provides a method and system for generating a meteorological report based on deep learning. By obtaining target meteorological acquisition data; loading the target meteorological acquisition data into a pre-debugged extreme climate pre-recognition neural network for processing to obtain an extreme climate pre-recognition result; generating a meteorological report based on the predicted extreme climate pre-recognition result. During network debugging, augmentation of the extreme climate data representation vector is completed in the dimension of data representation. The extreme climate data representation vector is represented as a set of multiple representation vector groups in the vector domain. An enhanced processing is performed on the centroid of the group set to construct a second extreme climate data representation vector, ensuring that the second extreme climate data representation vector is more similar to the existing extreme climate data representation vectors, so as to increase the accuracy of the constructed extreme climate data representation vector and improve the augmentation quality of the extreme climate data representation vector, thereby making the debugging result of the extreme climate pre-recognition neural network more accurate.

Description

Technical Field

[0001] This application relates to, but is not limited to, the field of data processing technologies, and in particular, to a method and system for generating meteorological reports based on deep learning. Background Art

[0002] In the field of meteorological data processing and analysis, the problem of data imbalance has always been a major challenge faced by researchers. Especially in the prediction and identification of extreme climate events, the scarcity of extreme climate training data compared to the abundance of non-extreme climate data has brought great difficulties to the training and prediction of models. Traditional data augmentation methods, such as data synthesis based on meteorological data, although alleviating the problem of data imbalance to a certain extent, these methods are often cumbersome and the generated extreme climate training data has a large difference from real data, affecting the accuracy of model prediction. Therefore, it is necessary to propose a new deep learning model training scheme to help debug an accurate deep learning model to help generate high-value meteorological reports. Summary of the Invention

[0003] In view of this, this application provides at least a method and system for generating meteorological reports based on deep learning. The technical solution of this application is implemented as follows:

[0004] On the one hand, an embodiment of this application provides a method for generating a meteorological report based on deep learning. The method includes: obtaining target meteorological acquisition data; loading the target meteorological acquisition data into a pre-debugged extreme climate pre-identification neural network for processing to obtain an extreme climate pre-identification result; generating a meteorological report based on the predicted extreme climate pre-identification result; where the debugging process of the extreme climate pre-identification neural network includes the following steps: extracting extreme climate data characterization vectors corresponding to multiple extreme climate training data respectively; where the extreme climate training data corresponds to a target acquisition area, and when performing extreme climate pre-identification on the target acquisition area according to the extreme climate training data, the obtained extreme climate pre-identification result is used to indicate that the target acquisition area has extreme climate conditions; performing a grouping and clustering operation on the multiple extreme climate data characterization vectors to obtain one or more sets of characterization vector groupings; performing enhancement processing on the first extreme climate data characterization vector corresponding to the centroid of each set of characterization vector groupings to obtain a second extreme climate data characterization vector corresponding to each first extreme climate data characterization vector; iteratively debugging the extreme climate pre-identification neural network based on the second extreme climate data characterization vector, and stopping debugging when reaching a preset convergence condition to obtain the debugged extreme climate pre-identification neural network, and the extreme climate pre-identification neural network is used to perform extreme climate pre-identification on target meteorological acquisition data.

[0005] On the other hand, the present application provides a computer system, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the program, the steps in the above-mentioned method are implemented.

[0006] The beneficial effects of the present application at least include: The method and system for generating a weather report based on deep learning provided by the present application obtain target meteorological acquisition data; load the target meteorological acquisition data into a pre-debugged extreme climate pre-identification neural network for processing to obtain an extreme climate pre-identification result; based on the predicted extreme climate pre-identification result, generate a weather report. During the debugging process of the neural network, first extract the extreme climate data characterization vectors corresponding to multiple extreme climate training data. The extreme climate training data corresponds to the target acquisition area. When performing extreme climate pre-identification on the target acquisition area according to the extreme climate training data, the obtained extreme climate pre-identification result is used to indicate that the target acquisition area has extreme climate conditions; then perform a grouping and clustering operation on the multiple extreme climate data characterization vectors to obtain one or more sets of characterization vector groups; enhance the first extreme climate data characterization vector corresponding to the centroid of each set of characterization vector groups to obtain the second extreme climate data characterization vector corresponding to each first extreme climate data characterization vector. The second extreme climate data characterization vector is an augmented feature of the extreme climate data characterization vector, which is convenient for calibrating the extreme climate pre-identification neural network based on the second extreme climate data characterization vector. The present application augments the extreme climate data characterization vector in the dimension of data representation. The extreme climate data characterization vector is represented as multiple sets of characterization vector groups in the vector domain. By enhancing the centroid of the grouping set to construct the second extreme climate data characterization vector, it is ensured that the constructed second extreme climate data characterization vector is more similar to the existing extreme climate data characterization vector, so as to increase the accuracy of the constructed extreme climate data characterization vector and the quality of the augmentation of the extreme climate data characterization vector, thereby making the debugging result of the extreme climate pre-identification neural network more accurate, facilitating the obtaining of an accurate and reliable extreme climate pre-identification result, and generating an accurate weather report. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is a schematic flowchart of the implementation of a method for generating a weather report based on deep learning provided by an embodiment of the present application.

[0008] Figure 2 It is a schematic diagram of the hardware entity of a computer system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0009] An embodiment of the present application provides a method for generating a weather report based on deep learning, which can be executed by a processor of a computer system. Among them, the computer system may refer to devices with data processing capabilities such as servers, laptops, tablets, desktop computers, etc.

[0010] In a weather data processing model, the problem of data imbalance may be faced. Data imbalance is specifically manifested as the number of extreme climate training data being much less than the number of non-extreme climate training data. Therefore, it is a common practice to generate some forged extreme climate training data during training to augment the number of extreme climate training data. Usually, most of the methods for generating extreme climate training data are implemented in the actual data dimension. For example, extreme climate training data is synthesized based on supplementary training data and random non-extreme climate training data. However, such a generation method usually has cumbersome steps, and the generated extreme climate training data has a large difference from the real extreme climate training data. In this case, the inventor found that compared with constructing extreme climate training data in the actual data space, it is more effective to construct an extreme climate data representation vector in the feature space, and it is closer to the extreme climate data representation vector of the real extreme climate training data. Therefore, in the embodiment of the present application, when debugging a neural network, a method for generating an extreme climate data representation vector is provided, which can use the existing extreme climate data representation vector to construct an extreme climate data representation vector that is closer to the actual extreme climate data representation vector.

[0011] The following is the process of the method for generating a weather report based on deep learning provided by the present application. Figure 1 It is a schematic diagram of the implementation process of a method for generating a weather report based on deep learning provided by an embodiment of the present application, as Figure 1 shown, the method includes:

[0012] Step S100: Obtain target meteorological acquisition data.

[0013] Specifically, the computer system can be connected to multiple meteorological data collection points, which may be surface meteorological stations, ocean buoys, meteorological satellites, radar systems, or unmanned aerial vehicles, etc. Each collection point is equipped with sensors capable of measuring various meteorological parameters, such as temperature, humidity, wind speed, wind direction, air pressure, precipitation, etc. These sensors will send the collected data to the computer system in real time or at regular intervals. Taking a surface meteorological station as an example, it usually includes devices such as thermometers, hygrometers, anemometers, and barometers, which continuously monitor and record the local meteorological parameters. At regular time intervals (such as every hour or every minute), the meteorological station transmits the collected data to the central computer system through the network. The computer system stores this data in a database for subsequent analysis and processing. In addition to real-time data, the computer system can also obtain historical meteorological data. This data may come from the archives of meteorological bureaus, databases of scientific research, or public online resources. Historical data is very important for meteorological forecasting and climate analysis because it can provide information such as past climate patterns, records of extreme events, and long-term trends. When obtaining the target meteorological collection data, the computer system also needs to clean and preprocess the data. This includes checking the integrity of the data, correcting incorrect data points, handling missing values, and converting the data into a unified format and unit. The target meteorological collection data can exist in the form of sequence data or matrix data.

[0014] Step S200: Load the target meteorological collection data into the pre-debugged extreme climate pre-identification neural network for processing to obtain the extreme climate pre-identification result.

[0015] In the embodiment of the present application, step S200 loads the target meteorological collection data into the pre-debugged extreme climate pre-identification neural network to process and obtain the pre-identification result of extreme climate.

[0016] First, the computer system has pre-constructed and debugged an extreme climate pre-identification neural network in advance. This neural network may have adopted deep learning techniques, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), combined with appropriate fully connected layers to capture complex patterns and features in meteorological data. This network has learned during the training phase how to identify the characteristics of extreme climate events from meteorological data. When the computer system obtains the target meteorological acquisition data, it will load this data into the extreme climate pre-identification neural network. This usually involves data preprocessing steps, such as data standardization, normalization, or encoding, to ensure that the data format matches the requirements of the neural network input. After loading the data, the neural network starts to process this data, and through a series of network layers (such as convolutional layers, pooling layers, fully connected layers, etc.), extracts the key features in the meteorological data. These features may include data change patterns in multiple dimensions such as temperature, humidity, wind speed, and air pressure. During the processing, the neural network will compare these features with the extreme climate features it learned during the training phase. By applying activation functions and loss functions, the neural network can calculate the similarity or matching degree between the target meteorological acquisition data and the extreme climate features.

[0017] Finally, the neural network outputs an extreme climate pre-identification result. This result may be a probability value indicating the likelihood of an extreme climate event occurring in the target acquisition area; it may also be a classification label directly indicating whether an extreme climate event is predicted. For example, if the probability value exceeds a certain threshold (such as 0.8), the computer system may determine that an extreme climate event is about to occur in this area and generate corresponding warning information.

[0018] In this way, the computer system can quickly and accurately predict the likelihood of extreme climate events based on the target meteorological acquisition data using the extreme climate pre-identification neural network, providing an important basis for meteorological warnings, disaster prevention, and decision-making support.

[0019] Step S300: Generate a meteorological report based on the predicted extreme climate pre-identification result.

[0020] When the computer system processes the target meteorological acquisition data through the extreme climate pre-identification neural network and predicts the likelihood of extreme climate, a detailed meteorological report is generated based on this prediction result. The computer system determines whether there is a possibility of an extreme climate event occurring in the target acquisition area according to the extreme climate pre-identification result output by the neural network. This prediction result is usually a probability value or a classification label, which directly reflects the likelihood of an extreme climate event occurring. Then, the computer system constructs the content of the meteorological report based on this prediction result. The meteorological report may include, for example, the following key parts: event overview, prediction details, impact assessment, recommended measures, data support, etc.

[0021] The debugging process of the extreme climate pre-identification neural network may include the following steps:

[0022] Step S10: Extract the extreme climate data characterization vectors corresponding to multiple extreme climate training data; wherein, the extreme climate training data corresponds to the target acquisition area, and when the extreme climate pre-identification is performed on the target acquisition area according to the extreme climate training data, the obtained extreme climate pre-identification result is used to indicate that the target acquisition area has extreme climate conditions.

[0023] In the embodiment of the present application, in step S10, the extreme climate data characterization vectors corresponding to multiple extreme climate training data are extracted, and these vectors will serve as the basis for neural network training. First, the computer system needs to screen out the extreme climate training data corresponding to the target acquisition area from the existing meteorological data sets. These extreme climate training data usually contain the relevant records of extreme climate events that occurred in this area in history, such as heavy rain, heavy snow, extreme high temperature or low temperature, etc. These records contain data of various meteorological parameters, such as temperature, humidity, wind speed, air pressure, etc., which together constitute a complete description of the extreme climate event.

[0024] Next, the computer system performs feature extraction on these extreme climate training data to obtain the corresponding extreme climate data characterization vectors. Feature extraction is an important step in machine learning. It captures the key information in the data by converting the original data into a form that is easier to analyze and process. In the embodiment of the present application, feature extraction can be implemented through various algorithms, such as principal component analysis (PCA), independent component analysis (ICA), or autoencoder, etc. Taking the autoencoder as an example, it is an unsupervised neural network model that can extract key features by learning the low-dimensional representation of the input data. In this scenario, the computer system can use the extreme climate training data as the input of the autoencoder and train the autoencoder to learn the low-dimensional representation of these data through training. These low-dimensional representations are the extreme climate data characterization vectors, which capture the key features of extreme climate events in multiple dimensions.

[0025] For example, assume a training dataset containing multiple extreme rainstorm events. For each rainstorm event, the dataset records multiple meteorological parameters such as temperature, humidity, wind speed, and air pressure at the time of the event. After feature extraction by an autoencoder, an extreme climate data representation vector corresponding to each rainstorm event is obtained. These vectors contain key features describing the rainstorm event, such as rainfall intensity, duration, and affected area. Finally, these extreme climate data representation vectors will be used in the subsequent neural network training process. By training a neural network model capable of recognizing these representation vectors, the computer system will be able to learn how to identify potential extreme climate events from new meteorological data. This will provide strong technical support for meteorological early warning, disaster prevention, and decision-making support.

[0026] Step S20: Perform a grouping and clustering operation on multiple extreme climate data representation vectors to obtain one or more sets of grouped representation vectors.

[0027] In the embodiments of the present application, the computer system will perform a grouping and clustering operation on multiple extreme climate data representation vectors, that is, cluster analysis, to obtain one or more sets of grouped representation vectors. These grouped sets will be used in subsequent steps to generate enhanced extreme climate data representation vectors, thereby enhancing the training effect of the extreme climate pre-identification neural network. First, the extreme climate data representation vectors are obtained from extreme climate training data through feature extraction techniques. These vectors contain key features describing extreme climate events, such as rainfall intensity, duration, and wind speed. These vectors usually have high-dimensional characteristics, containing a large amount of information, but at the same time increasing the complexity of analysis. To simplify the analysis and improve the efficiency of data processing, the computer system performs cluster analysis (i.e., clustering) on these extreme climate data representation vectors. Clustering is an unsupervised learning technique that can divide similar data points into the same group, while data points in different groups have significant differences. In the embodiments of the present application, clustering can help the computer system identify extreme climate events with similar features, thus providing a basis for subsequent data enhancement and neural network training.

[0028] Specifically, the computer system can adopt various clustering algorithms to implement the grouping and clustering operation of the extreme climate data representation vectors. For example, the K-means algorithm is a commonly used clustering algorithm that can divide data points into K clusters through iteration, making the data points within the same cluster as similar as possible, while the data points between different clusters are as different as possible. In this scenario, the computer system can preset a suitable number of clusters K in advance and then apply the K-means algorithm to cluster the extreme climate data representation vectors. Through cluster analysis, the computer system can obtain one or more sets of grouped representation vectors. Each grouped set represents a type of extreme climate event with similar characteristics. For example, one grouped set may contain all the extreme climate data representation vectors describing rainstorm events, while another grouped set may contain the vectors describing extreme high temperature events. These grouped sets will provide a basis for subsequent data enhancement steps to help the computer system generate more diverse and more realistic enhanced data representation vectors for extreme climate events. By performing cluster analysis on the extreme climate data representation vectors and dividing similar extreme climate events into the same group, it provides strong support for subsequent data enhancement and neural network training. This clustering method not only simplifies the complexity of data analysis but also improves the efficiency and accuracy of data processing.

[0029] Step S30: Perform enhancement processing on the first extreme climate data representation vector corresponding to the centroid of each set of grouped representation vectors to obtain the second extreme climate data representation vector corresponding to each first extreme climate data representation vector.

[0030] In the embodiment of the present application, step S30 generates more diverse training data by performing specific enhancement processing on the extreme climate data representation vectors, thereby improving the generalization ability of the extreme climate pre-recognition neural network.

[0031] In step S20, the computer system has performed cluster analysis on multiple extreme climate data representation vectors, obtaining one or more sets of representation vector groupings. Each grouping set represents a type of extreme climate event with similar characteristics. The set centroid is a certain center or representative vector of all vectors in the grouping set, which can usually reflect the main characteristics of the grouping set. In step S30, the computer system will perform enhancement processing on the first extreme climate data representation vector corresponding to the set centroid of each representation vector grouping set. Here, the enhancement processing specifically refers to adding perturbation information, or called adding noise. This noise addition operation is to add a certain amount of random noise or perturbation to the original data representation vector to simulate the uncertainty and data diversity in the real world. Specifically, the computer system can adopt various methods to implement this enhancement processing process. For example, for the first extreme climate data representation vector corresponding to each set centroid, a random noise value within a certain range can be added to each of its feature dimensions. This random noise value can be generated based on a certain probability distribution (such as Gaussian distribution) to ensure that the enhanced data representation vector has a certain degree of diversity while maintaining the original features.

[0032] For example, suppose there is a set of extreme climate data representation vectors for grouping rainstorm events, and the first extreme climate data representation vector corresponding to its set centroid can be represented as a multi-dimensional array, such as [temperature, humidity, wind speed, air pressure] = [25, 90, 30, 980]. During the enhancement processing, the computer system can add random noise to each dimension of this vector. For example, generate a small random perturbation value, such as adding a random number between -0.5 and 0.5 to the temperature, and adding a random number between -2 and 2 to the humidity, and so on. In this way, the original first extreme climate data representation vector may become [24.8, 89.5, 30.3, 980.2] or similar values, thus generating a new second extreme climate data representation vector.

[0033] Through this method, the computer system can generate multiple enhanced second extreme climate data representation vectors for each representation vector grouping set. While maintaining the main characteristics of the original extreme climate events, these enhanced vectors increase the data diversity and uncertainty, which helps to train a more robust and stronger generalization ability extreme climate pre-identification neural network.

[0034] Step S40: Iteratively debug the extreme climate pre-identification neural network based on the second extreme climate data representation vectors, and stop the debugging when the preset convergence condition is reached, obtaining a debugged extreme climate pre-identification neural network, which is used to perform extreme climate pre-identification on target meteorological acquisition data.

[0035] In step S40 of the embodiment of the present application, the computer system will iteratively debug the extreme climate pre-identification neural network based on the second extreme climate data characterization vector generated in step S30, and stop the debugging when the preset convergence condition is met, so as to obtain a neural network model with completed debugging.

[0036] First of all, the extreme climate pre-identification neural network is a deep learning model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN), which can learn and identify extreme climate patterns in meteorological data. During the training process, the neural network continuously adjusts its internal parameters to minimize the difference between the predicted result and the true result. In step S40, the second extreme climate data characterization vector is used as the input data of the neural network. These vectors are generated by adding perturbation information to the original extreme climate data characterization vectors, and they simulate the uncertainty in the real world, thus enhancing the generalization ability of the neural network. Next, the computer system compares the output of the neural network with the true extreme climate labels and calculates the prediction error. This error is usually measured by a loss function, such as cross-entropy loss or mean squared error loss. The loss function can quantify the difference between the predicted result and the true result and provide guidance for the parameter adjustment of the neural network. Then, the computer system uses the backpropagation algorithm to update the parameters of the neural network. This process is continuously repeated until the preset convergence condition is met. The convergence condition usually includes indicators such as the number of iterations, the change range of the loss function value, or the performance on the validation set. When the performance of the neural network on the validation set reaches stability or no longer improves significantly, the computer system stops the iterative debugging and considers that the neural network has been trained.

[0037] For example, assume that an extreme climate pre-identification neural network uses a convolutional neural network (CNN) as the basic architecture. In step S40, the second extreme climate data characterization vector is used as the input data, and the output of the neural network is calculated through forward propagation. Then, the output is compared with the true extreme climate labels, and the cross-entropy loss is calculated. Next, the backpropagation algorithm is used to calculate the gradient of the loss function with respect to the neural network parameters and update the parameters to reduce the loss value. This process will be continuously repeated until the accuracy on the validation set no longer improves significantly or reaches the preset maximum number of iterations. Finally, an extreme climate pre-identification neural network with completed debugging will be obtained, which can accurately identify extreme climate events in the target meteorological acquisition data.

[0038] As an implementation manner, before step S30, which performs enhancement processing on the first extreme climate data characterization vector corresponding to the centroid of each set of characterization vector groups to obtain the second extreme climate data characterization vector corresponding to each first extreme climate data characterization vector, the method may further include:

[0039] Step S301: Construct one or more first enhancement information, where each first enhancement information is different from each other.

[0040] In the embodiment of the present application, step S301 aims to provide a basis for subsequent enhanced processing of extreme climate data representation vectors. In this step, the computer system needs to construct one or more first enhancement information, and each enhancement information is different from each other to ensure that the enhanced data has sufficient diversity and richness.

[0041] In the embodiment of the present application, the first enhancement information can be regarded as a kind of perturbation or change to the original extreme climate data representation vector. This kind of perturbation can simulate the uncertainty of meteorological data in the real world, thereby helping the extreme climate pre-identification neural network to better learn and generalize.

[0042] In step S301, the computer system can adopt various methods to construct the first enhancement information. The following is a possible method: The computer system analyzes historical meteorological data, especially the data related to the target acquisition area and extreme climate events. Through statistical analysis and machine learning techniques, the system can identify the main change patterns and feature distributions in the meteorological data. For example, the system can find that the rainfall usually presents a normal distribution in extreme rainstorm events, while the change of wind speed may be more in line with a certain skewed distribution. Based on these statistical information and distribution characteristics, the computer system can generate multiple first enhancement information. Each enhancement information is a vector or array containing random perturbations, and its dimension matches the original extreme climate data representation vector. These random perturbations can be generated based on a certain probability distribution to ensure that the enhanced data not only conforms to the statistical laws of the real world but also has a certain degree of diversity.

[0043] Specifically, assume that the focus is on the representation vector of extreme rainstorm events, which may include features such as rainfall amount, rainfall duration, and rainfall intensity. In step S301, the computer system can generate one or more enhancement information for each feature dimension. For the rainfall amount, the system can generate a series of random rainfall perturbation values based on the normal distribution; for the rainfall duration, the system can generate random increments or decrements based on the uniform distribution or exponential distribution; for the rainfall intensity, the system can generate different levels of intensity changes based on the ordinal classification or regression model. For example, assume that the representation vector of the original extreme rainstorm event is [rainfall amount: 100mm, rainfall duration: 12 hours, rainfall intensity: moderate]. In step S301, the computer system can generate the following two first enhancement information: Enhancement information 1: [rainfall perturbation: +5mm, rainfall duration perturbation: -0.5 hours, rainfall intensity perturbation: no change], corresponding to the enhanced representation vector [rainfall amount: 105mm, rainfall duration: 11.5 hours, rainfall intensity: moderate]. Enhancement information 2: [rainfall perturbation: -10mm, rainfall duration perturbation: +1 hour, rainfall intensity perturbation: becomes strong], corresponding to the enhanced representation vector [rainfall amount: 90mm, rainfall duration: 13 hours, rainfall intensity: strong].

[0044] In step S30, perform enhancement processing on the first extreme climate data representation vectors corresponding to the set centroids of each set of representation vectors to obtain the second extreme climate data representation vectors corresponding to each first extreme climate data representation vector. Specifically, it may include:

[0045] Perform the following operations on each set of representation vectors respectively:

[0046] Step S31: Incorporate the first enhancement information into the first extreme climate data representation vectors corresponding to the set centroids of the set of representation vectors respectively to obtain one or more second extreme climate data representation vectors corresponding to the first extreme climate data representation vectors.

[0047] In the embodiment of the present application, step S31 is a key step in the data enhancement process. It generates one or more second extreme climate data representation vectors by incorporating the first enhancement information for the first extreme climate data representation vectors corresponding to the set centroids of each set of representation vectors. The following is a detailed explanation of this step, and examples are given in combination with specific application scenarios for auxiliary illustration.

[0048] In step S31, the computer system will perform enhancement processing on the first extreme climate data representation vectors corresponding to the set centroids of each set of representation vector groups. The enhancement processing here is achieved by incorporating the first enhancement information, which is constructed in step S301. Each piece of enhancement information is different from each other and is used to simulate the changes in meteorological data in the real world.

[0049] Specifically, for each set of representation vector groups, the computer system performs the following operations:

[0050] Select the set centroid vector: First, select the first extreme climate data representation vector corresponding to the set centroid from the current group set. This vector represents the main characteristics of the group set and is the starting point for the enhancement processing.

[0051] Incorporate the first enhancement information: Select one or more pieces of enhancement information from the first enhancement information generated in step S301 and incorporate them into the selected first extreme climate data representation vector respectively. The incorporation method can be to directly add the corresponding enhancement factor to each feature dimension of the representation vector, or to incorporate the enhancement information into the original vector through a more complex transformation function. The purpose of doing this is to increase the diversity and uncertainty of the data while maintaining the main characteristics of the original data.

[0052] Generate the second extreme climate data representation vectors: By incorporating different first enhancement information, the computer system can generate one or more second extreme climate data representation vectors for each first extreme climate data representation vector. These new representation vectors not only retain the key characteristics of the original data but also simulate the changes in meteorological data in the real world, thus helping to train a more robust and generalization-capable extreme climate pre-identification neural network.

[0053] Taking the characterization vector of a rainstorm event as an example, assume that the first extreme climate data characterization vector corresponding to the centroid of a set of characterization vector groups is [rainfall: 100 mm, rainfall duration: 12 hours, rainfall intensity: moderate]. In step S31, the computer system can select two first enhancement information for enhancement processing: Enhancement information 1: [rainfall perturbation: +5 mm, rainfall duration perturbation: -0.5 hours, rainfall intensity perturbation: no change]. Enhancement information 2: [rainfall perturbation: -10 mm, rainfall duration perturbation: +1 hour, rainfall intensity perturbation: becomes strong]. After integrating these two enhancement information into the original characterization vector respectively, two second extreme climate data characterization vectors can be obtained: Second extreme climate data characterization vector 1: [rainfall: 105 mm, rainfall duration: 11.5 hours, rainfall intensity: moderate]. Second extreme climate data characterization vector 2: [rainfall: 90 mm, rainfall duration: 13 hours, rainfall intensity: strong]. Through such step S31, the computer system can generate multiple second extreme climate data characterization vectors for each set of characterization vector groups. These vectors not only retain the key features of the original data, but also increase the diversity and uncertainty of the data, which helps to improve the performance of the extreme climate pre-identification neural network.

[0054] As an implementation manner, step S301, constructing one or more first enhancement information, may specifically include:

[0055] Step S3011A: performing one or more sparse extractions on the data that meets the first data dispersion condition to obtain one or more sparse extraction results, and using each sparse extraction result as each first enhancement information, and the sparse extraction results are different from each other.

[0056] In the embodiment of the present application, step S3011A performs sparse extraction based on a data set that meets a specific data dispersion condition to simulate the changes in meteorological data in the real world.

[0057] First of all, step S3011A focuses on the data that meets the first data dispersion condition. In the embodiment of the present application, the data distribution of some meteorological parameters (such as rainfall, wind speed, temperature, etc.) may conform to a specific probability distribution, such as normal distribution, skewed distribution, etc. These distributions reflect the change rules of meteorological parameters under different conditions. In order to construct the first enhancement information that conforms to these distributions, the computer system will first determine the meteorological parameters to be simulated and their corresponding data dispersion conditions. Taking rainfall as an example, assume that historical data shows that the rainfall in a certain area approximately follows a normal distribution with a mean of 100 mm and a standard deviation of 20 mm. This normal distribution is the first data dispersion condition.

[0058] Next, the computer system generates a set of random data according to this normal distribution, or filters out a subset of data that conforms to this distribution from historical meteorological data. This set of data will be used as the original data set for subsequent sparse extraction. Then, the computer system performs sparse extraction on this set of data, that is, selects a part of the data as sparse samples according to a certain rule. The purpose of sparse extraction is to select representative data points from the original data set to simulate the changes in meteorological data in the real world. The extraction method can be random sampling or sampling based on a certain strategy (such as importance sampling).

[0059] During the sparse extraction process, each extraction will obtain a sparse extraction result, that is, one or more data points. Since the extraction is random or strategic, the results obtained from each extraction will be different, thus ensuring the diversity of the enhanced information. Finally, the computer system takes each sparse extraction result as an independent first enhanced information. These enhanced information will be used in the subsequent extreme climate data characterization vector enhancement processing step to simulate the changes in meteorological data in the real world, thereby enhancing the generalization ability and robustness of the neural network. Taking rainfall as an example, assume that the computer system randomly sparsely extracts 5 data points from a rainfall data set that conforms to a normal distribution (mean 100mm, standard deviation 20mm) as the first enhanced information. These enhanced information may be different rainfall values such as 95mm, 112mm, 88mm, 103mm, and 125mm. These enhanced information will be used in the subsequent steps to fuse with the original extreme climate data characterization vector to generate more diverse training data to improve the performance of the extreme climate pre-identification neural network.

[0060] Alternatively, in another embodiment, constructing one or more first enhanced information may include step S3011B: obtaining data that satisfies one or more second data dispersion conditions, and for each second data dispersion condition, performing sparse extraction on the data that satisfies the second data dispersion condition to obtain the first enhanced information, and each second data dispersion condition is different from each other.

[0061] In the embodiments of the present application, step S3011B provides a flexible and precise method to construct the first enhancement information to adapt to the data dispersion characteristics under different meteorological conditions. The core idea of this step is to extract sparse samples from the corresponding data dispersion as enhancement information for each specific meteorological condition or season. The computer system identifies and obtains data that meet one or more second data dispersion conditions. In the embodiments of the present application, these data dispersion conditions may represent the meteorological data distribution characteristics in different seasons, different geographical locations, or different climate types. For example, rainfall may follow different normal distributions in summer and winter, and wind speed may show different skewed distributions in different regions. Once these second data dispersion conditions are determined, the computer system will perform a sparse extraction operation on each corresponding dataset. Sparse extraction is a method of selecting representative samples from the original dataset, which can be achieved by random sampling or sampling based on specific strategies. In this scenario, the purpose of sparse extraction is to select samples that can reflect the characteristics from each data dispersion condition as the first enhancement information. Taking rainfall as an example, assume that it is known that rainfall follows different normal distributions in summer and winter. The rainfall in summer may be concentrated between 80mm and 120mm, while the rainfall in winter may be more concentrated between 40mm and 80mm. The computer system will first screen out the data subsets that meet these two distributions from the historical meteorological data, and then perform sparse extraction on each subset respectively.

[0062] In the summer rainfall data subset, the computer system can randomly or based on a specific strategy extract several rainfall values such as 95mm, 108mm, 115mm, etc. as the sparse extraction results. Similarly, in the winter rainfall data subset, it can extract rainfall values such as 55mm, 68mm, 72mm, etc. These sparse extraction results represent the typical distribution characteristics of summer and winter rainfall, so they are used as the first enhancement information for the subsequent extreme climate data characterization vector enhancement processing step.

[0063] In this way, step S3011B not only considers the data dispersion characteristics under different meteorological conditions, but also obtains representative samples that can reflect these characteristics as enhancement information through the method of sparse extraction. This makes the subsequent generated enhanced data closer to the real-world meteorological data changes, and helps to improve the generalization ability and accuracy of the extreme climate pre-identification neural network.

[0064] As an implementation, before step S30, which enhances the first extreme climate data characterization vector corresponding to the centroid of each set of characterization vector groups to obtain the second extreme climate data characterization vector corresponding to each first extreme climate data characterization vector, the method may further include:

[0065] Step S302: Construct the second enhancement information corresponding to each set of representation vector groups.

[0066] In the embodiments of the present application, step S302 focuses on constructing specific second enhancement information for each set of representation vector groups. These enhancement information will be used in subsequent steps to enhance the first extreme climate data representation vectors corresponding to the set centroids of the representation vector group sets, so as to generate more diverse second extreme climate data representation vectors. In the previous steps, the computer system has divided a plurality of extreme climate data representation vectors into several sets of representation vector groups through cluster analysis, and each set represents a class of extreme climate events with similar characteristics. These sets not only contain the main characteristics of extreme climate events, but also reflect the distribution law and change trend of meteorological data under specific conditions.

[0067] In step S302, the goal of the computer system is to construct the second enhancement information for each set of representation vector groups. These enhancement information are generated based on the unique characteristics and data dispersion of each set, aiming to simulate the possible change patterns and data fluctuations of extreme climate events within the set.

[0068] Specifically, the computer system analyzes the data characteristics of each set of representation vector groups. These characteristics may include the type of extreme climate events (such as heavy rain, heavy snow, extreme high temperature, etc.), the occurrence time, geographical location, season, etc. By comprehensively analyzing these characteristics, the main characteristics and data distribution law of extreme climate events within the set can be understood. Next, the computer system constructs the second enhancement information according to the data characteristics and data dispersion of each set. This process can be realized based on a variety of statistical methods and machine learning techniques. For example, the system can calculate statistical quantities such as the mean, standard deviation, skewness, kurtosis, etc. of the data within the set to understand the central tendency and dispersion of the data. Then, based on these statistical quantities, the system can generate random perturbation values or change patterns that conform to the data dispersion of the set as the second enhancement information.

[0069] In addition, the computer system can also use machine learning models (such as Generative Adversarial Networks (GAN), Variational Autoencoders (VAE), etc.) to construct the second enhanced information. These models can learn the distribution laws and potential features of the data from the original data and generate new data similar to the original data. In the embodiments of the present application, these models can help the computer system generate more diverse extreme climate data representation vectors to enrich the training dataset and improve the generalization ability of the neural network. Taking the grouped set of representation vectors of rainstorm events as an example, assume that this set mainly reflects the data characteristics of rainstorm events in coastal areas in summer. In step S302, the computer system will first analyze the data characteristics of this set, such as the distribution laws and changing trends of parameters such as rainfall amount, rainfall duration, and wind speed. Then, based on these characteristics and data dispersion, the system can generate random perturbation values or change patterns that conform to the data characteristics of this set, such as a small fluctuation range of rainfall amount, an increase or decrease in rainfall duration, etc. These random perturbation values or change patterns will be used as the second enhanced information in the subsequent enhancement processing steps of the first extreme climate data representation vectors.

[0070] In this way, step S302 constructs specific second enhanced information for each grouped set of representation vectors, providing strong support for the subsequent enhancement processing steps. These enhanced information not only simulate the changes in meteorological data in the real world but also increase the diversity and uncertainty of the data, helping to improve the performance of the extreme climate pre-identification neural network.

[0071] Based on this, in step S30, the first extreme climate data representation vectors corresponding to the set centroids of each grouped set of representation vectors are enhanced to obtain the second extreme climate data representation vectors corresponding to each first extreme climate data representation vector. Specifically, it may include:

[0072] For each grouped set of representation vectors, the following operations are performed respectively:

[0073] Step S32: Incorporate the second enhanced information corresponding to the grouped set of representation vectors into the first extreme climate data representation vector corresponding to the set centroid to obtain the second extreme climate data representation vector corresponding to the first extreme climate data representation vector.

[0074] In the embodiments of the present application, step S32 involves incorporating the second enhanced information corresponding to the grouped set of representation vectors into the first extreme climate data representation vector corresponding to the set centroid to generate the second extreme climate data representation vector. The purpose of this step is to simulate the changes in meteorological data in the real world, thereby enriching the training dataset and improving the performance of the extreme climate pre-identification neural network.

[0075] In step S32, the computer system will perform the following operations on each grouped set of representation vectors:

[0076] Select the first extreme climate data representation vector corresponding to the centroid of the set: For each set of representation vector groups, the system first selects the first extreme climate data representation vector corresponding to its centroid. This vector represents the main characteristics of extreme climate events within the set and is the starting point for the enhancement process.

[0077] Integrate the second enhancement information: Next, the system integrates the second enhancement information corresponding to the set of representation vector groups into the selected first extreme climate data representation vector. The integration method can be to directly add the corresponding enhancement factor to each feature dimension of the representation vector, or to integrate the enhancement information into the original vector through a more complex transformation function. The purpose of this step is to increase the diversity and uncertainty of the data while maintaining the main characteristics of the original data.

[0078] Generate the second extreme climate data representation vector: By integrating the second enhancement information, the system can generate one or more second extreme climate data representation vectors for each first extreme climate data representation vector. These new representation vectors not only retain the key characteristics of the original data but also simulate the changes in meteorological data in the real world, thereby increasing the richness and generalization ability of the data.

[0079] Taking the set of representation vectors for heavy rain events as an example, assume that this set mainly reflects the data characteristics of heavy rain events in coastal areas during summer. In step S32, the computer system first selects the first extreme climate data representation vector corresponding to the centroid of this set, such as [rainfall: 150 mm, rainfall duration: 10 hours, wind speed: 20 m / s]. Then, the system integrates the second enhancement information representing the data dispersion characteristics of this set into this vector. Assume that the second enhancement information includes a small fluctuation range of rainfall (+10 mm to -5 mm) and a small increase or decrease in rainfall duration (+0.5 hours to -0.5 hours). A random enhancement factor can be selected, such as rainfall +5 mm and rainfall duration -0.2 hours, and then these enhancement factors are added to the original vector to generate a new second extreme climate data representation vector [rainfall: 155 mm, rainfall duration: 9.8 hours, wind speed: 20 m / s]. Through such step S32, the computer system can generate multiple second extreme climate data representation vectors for each set of representation vector groups. These vectors not only retain the key characteristics of the original data but also simulate the changes in meteorological data in the real world. These enhanced data will be used in the subsequent training process to improve the performance and generalization ability of the extreme climate pre-identification neural network.

[0080] As an implementation manner, step S302, constructing the second enhancement information corresponding to each set of representation vector groups, may include: performing the following operations for each set of representation vector groups respectively:

[0081] Step S3021: Determine the vector probability density distribution representing the vector grouping set.

[0082] In the embodiment of the present application, the purpose of step S3021 is to understand and describe the distribution of the extreme climate data representation vectors in each grouping set, so as to provide a basis for subsequent data augmentation and neural network training. The previous steps have divided multiple extreme climate data representation vectors into several representation vector grouping sets through cluster analysis. The vectors within each set represent extreme climate events with similar characteristics. The purpose of step S3021 is to determine the probability density distribution of the vectors in these sets. In step S3021, the computer system first performs statistical analysis on the vectors in each representation vector grouping set. This includes calculating statistics such as the mean, variance, skewness, and kurtosis of each feature dimension to understand the basic distribution of the data. Then, one or more probability density estimation methods are used to determine the vector probability density distribution. A commonly used probability density estimation method is Kernel Density Estimation (KDE). KDE is a non-parametric estimation method that estimates the probability density distribution of the entire data by using a smooth kernel function for each data point. In the embodiment of the present application, each extreme climate data representation vector can be regarded as a point in a multi-dimensional space, and then KDE is used to estimate the probability density distribution of these points in the multi-dimensional space. Through step S3021, the computer system can accurately determine the vector probability density distribution of each representation vector grouping set. This will provide important reference information for the subsequent data augmentation steps, help generate data representation vectors that are more close to real-world extreme climate events, and thus improve the performance of the extreme climate pre-identification neural network.

[0083] Step S3022: Construct data that satisfies the vector probability density distribution representing the vector grouping set.

[0084] In the embodiments of the present application, the main objective of step S3022 is to construct new data points based on the vector probability density distribution of the determined set of characterization vector groups, and these data points will conform to the distribution characteristics of the original data set. In step S3022, the computer system refers to the vector probability density distribution determined in step S3021. This probability density distribution describes the distribution of data points in the multi-dimensional space within the set of characterization vector groups, reflecting the statistical characteristics of extreme climate events in different characteristic dimensions. Next, the computer system uses one or more methods to construct new data points that satisfy this probability density distribution. A commonly used method is Monte Carlo Simulation. Monte Carlo Simulation is a statistical method based on random sampling, which estimates the solution of complex problems through a large number of random trials. Here, the computer system can randomly extract sample points from the known vector probability density distribution, and these sample points are the newly constructed data points. Taking the set of characterization vector groups of rainstorm events as an example, assume that the vector probability density distribution of this set describes the joint probability distribution of three characteristic dimensions: rainfall, rainfall duration, and wind speed. In step S3022, the computer system can use the Monte Carlo simulation method to randomly extract a large number of sample points from this joint probability distribution. Each sample point represents a possible rainstorm event, containing specific values of rainfall, rainfall duration, and wind speed. These newly constructed data points not only conform to the distribution characteristics of the original data set, but also increase the diversity and richness of the data. They will be used as enhanced information in subsequent steps, and fused with the original extreme climate data characterization vectors to generate more diverse training data. Through step S3022, the computer system can construct new data points based on the known vector probability density distribution, and these data points will be used to simulate extreme climate events in the real world. This will help improve the generalization ability and robustness of the extreme climate pre-identification neural network, enabling it to more accurately predict possible future extreme climate events.

[0085] Step S3023: Sparsely extract from the data that satisfies the vector probability density distribution of the set of characterization vector groups to obtain the second enhanced information corresponding to the set of characterization vector groups.

[0086] In the embodiment of the present application, step S3023 performs sparse extraction from the data that satisfies the vector probability density distribution of the specific representation vector grouping set. The purpose of this step is to select representative samples from the large amount of generated data as enhanced information for subsequent enhanced processing of the extreme climate data representation vector. In step S3022, the computer system has constructed a large number of new data points based on the vector probability density distribution of the representation vector grouping set. These data points not only conform to the distribution characteristics of the original data set, but also increase the diversity and richness of the data. However, directly using all the generated data points for enhancement processing may not be efficient because it may contain redundant or highly similar samples. Therefore, step S3023 selects representative samples from these data points as the second enhanced information through sparse extraction.

[0087] Sparse extraction is a method of selecting a part of representative samples from a large amount of data. In a computer system, this can be achieved through various strategies, such as random sampling, importance sampling, or distance-based sampling, etc. Random sampling is a simple and effective method that can randomly select a part of the generated data points as enhanced information. Importance sampling samples according to the importance of the data points (such as probability density values) to ensure that the selected samples can cover the main feature areas of the original data. Distance-based sampling considers the spatial distribution between data points to ensure that the selected samples have a certain degree of dispersion in space. Specifically, each enhanced information is a multi-dimensional vector containing the specific numerical values of features such as rainfall, rainfall duration, and wind speed. For example, an enhanced information may be [rainfall: 120mm, rainfall duration: 8 hours, wind speed: 25m / s]. These enhanced information will simulate the possible change patterns of rainstorm events in the real world, helping the extreme climate pre-identification neural network to better learn and generalize.

[0088] Through the sparse extraction operation of step S3023, the computer system can select representative samples from the large amount of generated data as the second enhanced information. These enhanced information not only retain the main features of the original data, but also increase the diversity and uncertainty of the data, which helps to improve the performance of the extreme climate pre-identification neural network.

[0089] As an implementation manner, after step S30, which performs enhancement processing on the first extreme climate data representation vector corresponding to the set centroid of each representation vector grouping set to obtain the second extreme climate data representation vector corresponding to each first extreme climate data representation vector, the method may further include:

[0090] Step S30A: Obtain supplementary training data, and the supplementary training data does not originate from the target acquisition area.

[0091] In the embodiments of the present application, step S30A aims to enhance the generalization ability of the extreme climate pre-recognition neural network by obtaining supplementary training data. These supplementary training data do not originate from the target acquisition area, but come from other geographical areas with similar or different climate conditions.

[0092] The target acquisition area is the main source of meteorological data of concern and analysis, which may be a specific city, region or country. However, due to the complexity and diversity of meteorological data, relying solely on the data of the target acquisition area for neural network training may lead to insufficient generalization ability of the model and inability to accurately predict extreme climate events outside the target area. Therefore, in step S30A, the computer system needs to obtain supplementary training data. These supplementary data do not directly come from the target acquisition area, but they contain information on climate conditions and extreme climate events similar to or different from those of the target area. The sources of these supplementary data can be extensive and diverse, including but not limited to historical meteorological records, remote sensing data, simulation data, etc. of other cities, regions, countries.

[0093] For example, assume that the target acquisition area is a coastal city in the tropical region, which is often affected by typhoons and heavy rains. To enhance the generalization ability of the extreme climate pre-recognition neural network, the computer system needs to obtain supplementary training data from other regions. These data may include:

[0094] Data under similar climate conditions: For example, historical meteorological records from other tropical coastal cities that are also often affected by typhoons and heavy rains. These data will provide information on climate conditions and extreme climate events similar to those of the target area, helping the neural network learn the common characteristics of these events.

[0095] Data under different climate conditions: Historical meteorological data from inland cities or temperate regions, where the climate conditions are significantly different from those of the target area. Although these regions may not be directly affected by typhoons and heavy rains frequently, their climate data still contain useful information, such as the variation laws of parameters such as temperature, humidity, and air pressure.

[0096] Simulation data: Data generated by simulating climate models is also an effective source of supplementary training data. These simulation data can simulate different climate conditions and extreme climate events, providing a rich variety of training samples for the neural network.

[0097] After obtaining the supplementary training data, the computer system will preprocess and extract features from these data to generate corresponding supplementary training data representation vectors. These vectors will be used in subsequent steps to fuse or integrate with the original extreme climate data representation vectors to generate more abundant and diverse training data samples. In this way, the neural network will be able to learn more diverse features of extreme climate events, thereby improving its prediction accuracy and generalization ability.

[0098] Step S30B: Extract the supplementary training data representation vectors of the supplementary training data.

[0099] In the embodiment of the present application, step S30B extracts the supplementary training data representation vectors from the supplementary training data. These representation vectors are the basis for subsequent enhancement processing and neural network training, and are crucial for improving the performance of the extreme climate pre-recognition neural network.

[0100] In step S30A, the computer system has obtained the supplementary training data. These supplementary data do not originate from the target acquisition area, but contain climate condition information similar to or different from that of the target area. In step S30B, the computer system needs to process these supplementary data to extract the representation vectors that can represent the key features of extreme climate events. In the embodiment of the present application, the representation vectors are usually multi-dimensional arrays or vectors composed of multiple meteorological parameters. These parameters may include rainfall, rainfall duration, wind speed, wind direction, temperature, humidity, air pressure, etc., which jointly describe the main features of extreme climate events. To extract these representation vectors from the supplementary training data, the computer system can adopt various methods, such as feature extraction algorithms or machine learning models.

[0101] A commonly used feature extraction method is the method based on statistical characteristics. The computer system can calculate the statistics of each meteorological parameter in the supplementary training data, such as mean, standard deviation, skewness, kurtosis, etc., to capture the distribution law and change characteristics of the data. These statistics can be used as part of the representation vectors to describe the overall characteristics of extreme climate events in specific parameters. However, relying solely on statistical characteristics may not be able to fully capture the complexity and diversity of extreme climate events. Therefore, the computer system can also adopt more complex machine learning models to extract the representation vectors. For example, unsupervised learning models such as Autoencoder or Variational Autoencoder can be used. These models can extract key features by learning the low-dimensional representation of the input data and represent these features as representation vectors. In the embodiment of the present application, the supplementary training data can be used as the input of the model, and the model can be trained to learn the representation vectors that can represent the main features of extreme climate events.

[0102] Specifically, taking the supplementary training data of rainstorm events as an example. Suppose there is a set of historical data of rainstorm events from different regions, and this data includes multiple meteorological parameters such as rainfall, rainfall duration, wind speed, etc. In step S30B, the computer system will first preprocess this data, such as data cleaning, normalization, etc., to ensure the quality and consistency of the data. Then, the system can use a pre-trained autoencoder model to extract the feature vectors. This model learns the low-dimensional representations of the input data through a multi-layer neural network structure and represents these representations as multi-dimensional vectors. By inputting the supplementary training data into the autoencoder, the system can obtain the feature vectors of each rainstorm event, and these vectors contain the key information that can represent the main features of the rainstorm event. Finally, these extracted supplementary training data feature vectors will be used in subsequent steps, such as fusing or integrating with the original extreme climate data feature vectors to generate more abundant and diverse training data samples. In this way, the computer system can make full use of the supplementary training data from different regions to improve the generalization ability and accuracy of the extreme climate pre-identification neural network.

[0103] Step S30C: Integrate the second extreme climate data feature vector and the supplementary training data feature vector to obtain the third extreme climate data feature vector; and / or; Step S30D: Use the supplementary training data feature vector as the third extreme climate data feature vector.

[0104] In the embodiment of the present application, step S30C integrates the second extreme climate data feature vector and the supplementary training data feature vector to generate the third extreme climate data feature vector. This step aims to combine data from different sources to enrich the training data set and improve the generalization ability and accuracy of the extreme climate pre-identification neural network. In step S30, the computer system has enhanced the first extreme climate data feature vector corresponding to the set centroid of the feature vector grouping set to generate the second extreme climate data feature vector. These vectors simulate the possible change patterns of extreme climate events in the target acquisition area. At the same time, in steps S30A and S30B, the system obtains the supplementary training data and extracts the supplementary training data feature vectors from it. These vectors represent the characteristics of extreme climate events under climate conditions similar to or different from the target acquisition area.

[0105] In step S30C, the goal of the computer system is to integrate these two feature vectors to generate the third extreme climate data feature vector. Different strategies can be adopted for the integration process according to specific requirements. The following are examples of two possible integration strategies and their application scenarios:

[0106] Strategy 1: Direct concatenation. A simple integration strategy is to directly concatenate the second extreme climate data representation vector and the supplementary training data representation vector. Suppose the second extreme climate data representation vector is a three-dimensional vector containing rainfall, wind speed, and air pressure, such as [rainfall: 150mm, wind speed: 30m / s, air pressure: 980hPa]. At the same time, the supplementary training data representation vector is also a three-dimensional vector, but it may contain different features, such as temperature, humidity, and wind direction, such as [temperature: 25°C, humidity: 80%, wind direction: northeast wind]. By directly concatenating these two vectors, a six-dimensional third extreme climate data representation vector with more features can be obtained, such as [rainfall: 150mm, wind speed: 30m / s, air pressure: 980hPa, temperature: 25°C, humidity: 80%, wind direction: northeast wind]. This integration strategy is suitable when the two representation vectors are complementary, that is, they contain different key features. By concatenation, a more comprehensive representation vector for describing extreme climate events can be obtained, providing more training information for the neural network.

[0107] Strategy 2: Feature fusion network. Another more complex integration strategy is to use a Feature Fusion Network. This network can accept representation vectors from different sources as inputs and perform feature fusion and transformation through a series of network layers (such as fully connected layers, convolutional layers, etc.), and finally output a fused third extreme climate data representation vector.

[0108] In the feature fusion network, representation vectors from different sources first pass through their respective preprocessing layers to extract key features and adjust the data dimensions. Then, these feature vectors are sent to the fusion layer for fusion. The fusion layer can use various methods to achieve feature fusion, such as element-wise addition, element-wise multiplication, transformation through a fully connected layer after concatenation, etc. Finally, the fused feature vectors pass through the output layer for transformation and adjustment to obtain the final third extreme climate data representation vector. This integration strategy is applicable when there is a certain correlation or similarity between the two representation vectors. Through the learning and optimization of the feature fusion network, the system can automatically discover the associations and interactions between different features, thereby generating a more accurate and meaningful third extreme climate data representation vector.

[0109] Regardless of which integration strategy is adopted, the goal of step S30C is to combine data from different sources to generate a richer and more diverse third extreme climate data representation vector. These vectors will be used as part of the training data for the training and optimization of the extreme climate pre-identification neural network, thereby improving the generalization ability and accuracy of the model.

[0110] In steps S30A and S30B, the computer system has obtained supplementary training data and extracted supplementary training data representation vectors from it. These vectors represent the characteristics of extreme climate events under climate conditions similar to or different from the target acquisition area. The proposal of step S30D is based on an assumption that in some cases, these supplementary training data representation vectors are rich and diverse enough to be directly used for the training of the neural network without complex integration or fusion with other representation vectors. In step S30D, the operation of the computer system is relatively simple and straightforward. It no longer performs additional integration or fusion steps, but directly uses the supplementary training data representation vectors as the third extreme climate data representation vectors. This means that these vectors will be directly input into the extreme climate pre-recognition neural network as part of the training data for training.

[0111] The advantage of this method lies in its simplicity and efficiency. By directly using the supplementary training data representation vectors, the computer system can quickly expand the training data set without performing complex feature fusion or transformation operations. In addition, since these vectors already contain rich characteristics of extreme climate events, they can provide valuable training information for the neural network, helping to improve the generalization ability and accuracy of the model.

[0112] In this way, the computer system can quickly increase the diversity and richness of the training data set and improve the performance of the extreme climate pre-recognition neural network. At the same time, because this method is relatively simple and straightforward, it has high feasibility and operability in practical applications.

[0113] As an implementation, after step S30, which enhances the first extreme climate data representation vectors corresponding to the set centroids of each group of representation vectors to obtain the second extreme climate data representation vectors corresponding to each first extreme climate data representation vector, the method may further include:

[0114] Step S30I: Extracting the non-extreme climate data representation vectors of non-extreme climate training data and constructing the third enhancement information; where the non-extreme climate training data corresponds to the target acquisition area, and when the extreme climate pre-recognition is performed on the target acquisition area according to the non-extreme climate training data, the obtained extreme climate pre-recognition result is used to indicate that there is no extreme climate situation in the target acquisition area.

[0115] In the embodiments of the present application, step S30I involves the processing of non-extreme climate training data, aiming to simulate and expand extreme climate data by extracting non-extreme climate data characterization vectors and constructing third enhancement information, thereby improving the generalization ability of the extreme climate pre-identification neural network. Non-extreme climate training data refers to meteorological data that corresponds to the target acquisition area but is judged to have no extreme climate conditions during extreme climate pre-identification. These data contain the daily or common meteorological conditions within the target area, such as normal rainfall, temperature, wind speed, etc. In the embodiments of the present application, these data are also of great value because they provide the basic background and reference information for meteorological changes in the target area.

[0116] In step S30I, the primary task of the computer system is to extract the non-extreme climate data characterization vectors of the non-extreme climate training data. These characterization vectors are highly generalized and abstracted features of non-extreme climate data, usually containing multiple meteorological parameters such as rainfall, temperature, humidity, wind speed, etc. Through feature extraction algorithms or machine learning models, the computer system can extract these characterization vectors from the non-extreme climate training data as the basis for subsequent analysis and processing. At the same time, in order to simulate the transition or change from non-extreme climate to extreme climate, the computer system needs to construct third enhancement information. These enhancement information are used to simulate the extreme climate characteristics that non-extreme climate data may exhibit under certain changes or perturbations. The construction of enhancement information can be based on various methods, such as the variation law of historical meteorological data, expert knowledge, or machine learning models. For example, deep learning models such as generative adversarial networks (GANs) or variational autoencoders (VAEs) can be used to simulate the potential change paths from non-extreme climate data to extreme climate data, thereby generating appropriate enhancement information.

[0117] Specifically, assume that the target acquisition area is a coastal city that experiences normal rainfall and temperature conditions most of the time. However, under certain specific conditions, this city may also be affected by extreme climate events such as typhoons or heavy rains. In step S30I, the computer system will first extract non-extreme climate training data from the historical meteorological data of this city and extract the corresponding non-extreme climate data characterization vectors. These vectors may contain information such as normal rainfall (e.g., 50 mm / day), temperature (e.g., 25 °C), and wind speed (e.g., 10 m / s). Then, in order to simulate the transition from these non-extreme climate conditions to extreme climate conditions, the computer system needs to construct third enhancement information. These information may include a significant increase in rainfall (e.g., from 50 mm / day to 200 mm / day), a sharp increase in wind speed (e.g., from 10 m / s to 50 m / s), and a sharp drop in air pressure, etc. These enhancement information can be generated by analyzing the variation law of historical extreme climate events or using machine learning models.

[0118] By extracting the non-extreme climate data representation vectors and constructing the third enhanced information, the computer system can provide a rich data basis for subsequent steps, which helps to simulate and expand the extreme climate data, and improve the generalization ability and accuracy of the extreme climate pre-identification neural network.

[0119] Step S30II: Based on the non-extreme climate data representation vectors and the third enhanced information, generate the fourth extreme climate data representation vector through the extreme representation vector construction network.

[0120] In the embodiment of the present application, step S30II is based on the non-extreme climate data representation vectors and the third enhanced information, and generates the fourth extreme climate data representation vector through the extreme representation vector construction network. The goal of this step is to further enrich the extreme climate training data set by simulating the transition from non-extreme climate to extreme climate, thereby improving the generalization ability of the extreme climate pre-identification neural network. In step S30II, the computer system will use these non-extreme climate data representation vectors and the third enhanced information as inputs, and generate the fourth extreme climate data representation vector through the extreme representation vector construction network. This extreme representation vector construction network can be a deep learning model, such as a recurrent neural network (RNN), a long short-term memory network (LSTM), or a generative adversarial network (GAN). Taking the generative adversarial network (GAN) as an example, the GAN consists of two parts: a generator and a discriminator. In this scenario, the generator will receive the non-extreme climate data representation vectors and the third enhanced information as inputs, and learn to generate representation vectors with extreme climate characteristics. The task of the discriminator is to distinguish the generated extreme climate data representation vectors from the real extreme climate data representation vectors. Through continuous adversarial training, the representation vectors generated by the generator will become closer and closer to the real extreme climate data.

[0121] Through continuous iterative training, the extreme representation vector construction network (i.e., the GAN model) will be able to generate more and more realistic fourth extreme climate data representation vectors. These generated representation vectors, together with the original second extreme climate data representation vectors, will be used as training data for the training and optimization of the extreme climate pre-identification neural network, thereby improving the prediction accuracy and generalization ability of the model.

[0122] Step S30III: Integrate the second extreme climate data representation vector and the fourth extreme climate data representation vector to obtain the fifth extreme climate data representation vector.

[0123] In the embodiments of the present application, in step S30III, the second extreme climate data representation vector (obtained by enhancing the original extreme climate data) is integrated with the fourth extreme climate data representation vector (generated based on non-extreme climate data and enhancement information) to generate a fifth extreme climate data representation vector. The purpose of this step is to enrich the extreme climate training data set and improve the generalization ability and accuracy of the extreme climate pre-recognition neural network. In step S30, the computer system has enhanced the first extreme climate data representation vector corresponding to the set centroid of the representation vector grouping set to generate the second extreme climate data representation vector. These vectors simulate the possible change patterns of extreme climate events in the target acquisition area. At the same time, in step S30II, based on the non-extreme climate data representation vector and the third enhancement information, a fourth extreme climate data representation vector is generated through an extreme representation vector construction network (such as a generative adversarial network GAN). These vectors simulate the transition from non-extreme climate to extreme climate.

[0124] In step S30III, the task of the computer system is to integrate these two representation vectors to generate a fifth extreme climate data representation vector. Different strategies can be adopted for the integration process according to specific requirements. The following is an example of a possible integration strategy and its application scenario: A common integration strategy is direct concatenation. Suppose the second extreme climate data representation vector is a three-dimensional vector containing rainfall, wind speed, and air pressure, such as [rainfall: 150mm, wind speed: 30m / s, air pressure: 960hPa]. At the same time, the fourth extreme climate data representation vector is also a three-dimensional vector, but it may contain different features or values of the same feature under different conditions, such as [rainfall increase: 50mm, wind speed increase: 10m / s, air pressure decrease: 20hPa]. By directly concatenating these two vectors, a six-dimensional fifth extreme climate data representation vector containing more features or more detailed feature descriptions can be obtained, such as [rainfall: 150mm, rainfall increase: 50mm, wind speed: 30m / s, wind speed increase: 10m / s, air pressure: 960hPa, air pressure decrease: 20hPa].

[0125] This integration strategy is simple and intuitive and can retain all the information in the original vectors. However, in some cases, direct concatenation may lead to too high a vector dimension, increasing the complexity of subsequent processing. To solve this problem, feature selection or dimensionality reduction techniques can be used to screen out the most important features or reduce the dimension of the vector.

[0126] Another more complex integration strategy is to use a feature fusion network. A feature fusion network is a deep learning model that takes multiple representation vectors as input and performs feature fusion and transformation through a series of network layers (such as fully connected layers, convolutional layers, etc.), and finally outputs a fused representation vector. In this scenario, the feature fusion network can take the second extreme climate data representation vector and the fourth extreme climate data representation vector as input, and by learning the interactions and correlations between different features, generate a more comprehensive and accurate fifth extreme climate data representation vector.

[0127] Regardless of which integration strategy is adopted, the goal of step S30III is to combine extreme climate data representation vectors from different sources and generation methods to generate a more rich and diverse fifth extreme climate data representation vector. These vectors will be used as part of the training data for the training and optimization of the extreme climate pre-identification neural network, thereby improving the prediction accuracy and generalization ability of the model.

[0128] As an implementation manner, the method further includes:

[0129] Step S30IV: Obtain an extreme representation vector construction network to be trained, the first non-extreme climate data representation vector of the first non-extreme climate training data, and construct the fourth enhancement information.

[0130] In step S30IV of the embodiment of the present application, the computer system performs multiple key operations, including obtaining an extreme representation vector construction network to be trained, extracting the first non-extreme climate data representation vector of the first non-extreme climate training data, and constructing the fourth enhancement information.

[0131] The computer system first needs an extreme representation vector construction network to be trained. This network is usually a deep learning model designed to generate representation vectors with extreme climate characteristics based on the input non-extreme climate data. For example, this network can be the generator part of a generative adversarial network (GAN) or a variational autoencoder (VAE). These models can simulate the climate change process from non-extreme to extreme by learning the distribution laws of a large amount of meteorological data. Next, the computer system needs to extract the first non-extreme climate data representation vector from the first non-extreme climate training data. These training data contain common non-extreme climate conditions in the target acquisition area, such as normal rainfall, temperature, wind speed, etc. Through feature extraction algorithms or machine learning models, the system can extract key meteorological parameters from these data to form a multi-dimensional representation vector. This vector highly summarizes the characteristics of non-extreme climate data and is the basis for generating extreme climate data representation vectors subsequently.

[0132] For example, assume that the first non-extreme climate training data contains the daily rainfall data of a certain city for one year. The system can extract the feature vectors of these data, such as the average rainfall, the maximum rainfall, the number of rainy days, etc., through statistical methods or machine learning algorithms. This feature vector will be used as the input data and passed to the extreme feature vector construction network.

[0133] To simulate the transition process from non-extreme climate to extreme climate, the system needs to construct the fourth enhancement information. These enhancement information describes the extreme changes that non-extreme climate data may undergo under specific conditions, such as a sharp increase in rainfall, a sharp increase in wind speed, etc. The construction of the enhancement information can be based on the change rules of historical meteorological data, expert knowledge, or another machine learning model.

[0134] Taking the change in rainfall as an example, the change trend of rainfall from non-extreme to extreme in historical meteorological data can be analyzed, and statistical information such as the amplitude and frequency of rainfall increase can be extracted. Then, based on this statistical information, the system can generate a series of enhancement factors as the fourth enhancement information. These enhancement factors will be used to simulate the transition process from non-extreme rainfall data to extreme rainfall data.

[0135] Step S30V: Based on the first non-extreme climate data feature vector and the fourth enhancement information, construct the sixth extreme climate data feature vector through the to-be-trained extreme feature vector construction network.

[0136] In the embodiment of the present application, step S30V is based on the first non-extreme climate data feature vector and the fourth enhancement information, and generates the sixth extreme climate data feature vector through the to-be-trained extreme feature vector construction network. In step S30IV, the computer system has obtained the to-be-trained extreme feature vector construction network, the first non-extreme climate data feature vector of the first non-extreme climate training data, and constructed the fourth enhancement information. These elements provide the necessary data and model basis for step S30V. In step S30V, the core task of the computer system is to use the to-be-trained extreme feature vector construction network, take the first non-extreme climate data feature vector and the fourth enhancement information as the input, and generate the sixth extreme climate data feature vector. This network is usually a deep learning model, such as the generator part of a generative adversarial network (GAN) or a variational autoencoder (VAE). By learning the distribution rules of a large amount of meteorological data, it can simulate the climate change process from non-extreme to extreme.

[0137] Specifically, the computer system first transmits the first non-extreme climate data characterization vector and the fourth enhancement information to the extreme characterization vector construction network to be trained. In this process, the fourth enhancement information plays a key role. It simulates the potential changes of non-extreme climate data under certain specific conditions, and these changes may point to extreme climate events. By combining the enhancement information with the non-extreme climate data characterization vector, the system can simulate the data characteristics closer to real extreme climate events.

[0138] Then, the extreme characterization vector construction network to be trained will receive these input data and perform a series of feature transformation and fusion operations through its internal complex network structure and parameter settings. These operations are aimed at transforming the non-extreme climate data characterization vector into a characterization vector with extreme climate characteristics. In this process, the network can learn the interactions and correlations between various meteorological parameters, as well as their variation laws under different conditions. Finally, after the processing and transformation of the network, the system will output the sixth extreme climate data characterization vector. This vector simulates the characteristics of extreme climate events transformed from non-extreme climate data and contains the specific values of multiple key meteorological parameters such as rainfall, wind speed, and air pressure. These values not only reflect the main characteristics of extreme climate events but also have a certain degree of diversity and uncertainty, which helps to improve the generalization ability and accuracy of the extreme climate pre-identification neural network.

[0139] Step S30VI: Identify the sixth extreme climate data characterization vector through the characterization vector identification network to obtain the characterization vector identification result, and the characterization vector identification result is used to represent the probability that the sixth extreme climate data characterization vector is an actual extreme climate data characterization vector.

[0140] In the embodiment of the present application, step S30VI uses the characterization vector identification network to identify the sixth extreme climate data characterization vector to evaluate the similarity or authenticity of these simulated vectors with the real extreme climate data characterization vector. In step S30V, the computer system generates the sixth extreme climate data characterization vector through the extreme characterization vector construction network to be trained. These vectors simulate the characteristics of extreme climate events transformed from non-extreme climate conditions. However, the quality of the generated vectors and whether they truly reflect the characteristics of extreme climate events need to be verified through an independent evaluation mechanism.

[0141] In step S30VI, the computer system introduces a characterization vector identification network to perform this evaluation task. The characterization vector identification network is a pre-trained machine learning model designed to identify the input extreme climate data characterization vectors and determine whether they are real extreme climate data characterization vectors or calculate the probability that they belong to real extreme climate data. This network may be a classifier model, such as a support vector machine (SVM) or a deep neural network (DNN), which can learn the characteristics of real extreme climate data characterization vectors and classify or score the input vectors accordingly.

[0142] Specifically, in step S30VI, the computer system passes the sixth extreme climate data characterization vectors as input to the characterization vector identification network. The network will process and analyze these vectors, extract the key features therein, and compare them with the features of real extreme climate data characterization vectors learned by itself. Through a series of calculations and judgments, the network will output a characterization vector identification result. This result is usually a probability value indicating the likelihood that the input sixth extreme climate data characterization vectors are real extreme climate data characterization vectors. By identifying the sixth extreme climate data characterization vectors through the characterization vector identification network, the computer system can obtain a quantitative evaluation result, thereby understanding the quality and authenticity of the generated vectors. This helps to further optimize the extreme characterization vector construction network, improve the accuracy and reliability of the generated vectors, and thus enhance the performance of the extreme climate pre-identification neural network.

[0143] Step S30VII: Based on the actual extreme climate data characterization vectors and the characterization vector identification result, determine the network debugging error of the extreme characterization vector construction network to be trained.

[0144] In step S30VII of the embodiment of the present application, the computer system determines the network debugging error of the extreme characterization vector construction network to be trained based on the actual extreme climate data characterization vectors and the characterization vector identification result. The actual extreme climate data characterization vectors refer to the feature vectors extracted from real meteorological data and representing extreme climate events. These vectors are directly obtained from extreme climate events through feature extraction algorithms and reflect the true characteristics of extreme climate. The characterization vector identification result is the output obtained after the characterization vector identification network identifies the simulated extreme climate data characterization vectors, which represents the similarity or authenticity between the simulated vectors and the real extreme climate data characterization vectors.

[0145] In step S30VII, the computer system first obtains the actual extreme climate data representation vectors and the representation vector identification results. Then, using these data as a reference, the system compares and analyzes them with the simulated extreme climate data representation vectors generated by the network for constructing the extreme representation vectors to be trained. By comparing the differences between the simulated vectors and the real vectors, the system can evaluate the performance of the network when generating the extreme climate data representation vectors.

[0146] Specifically, an error metric method, such as the mean squared error (MSE) or the mean absolute error (MAE), can be used to calculate the differences between the simulated vectors and the real vectors. These error metric methods can quantitatively reflect the accuracy of the data generated by the network. At the same time, the system also combines the representation vector identification results and considers the performance of the simulated vectors in terms of authenticity. If the identification results show that the simulated vectors have a high similarity to the real vectors, the error can be correspondingly reduced; conversely, if the similarity is low, the error can be increased.

[0147] For example, assume that the extreme climate data representation vectors for heavy rain events are of concern. In step S30VII, the computer system first obtains a set of actual extreme climate data representation vectors for heavy rain, which represent the key features of real heavy rain events. Then, the system uses the network for constructing the extreme representation vectors to be trained to generate a set of simulated extreme climate data representation vectors for heavy rain and identifies these simulated vectors through the representation vector identification network. Next, the error between the simulated vectors and the real vectors is calculated. Assume that the mean squared error (MSE) is used as the error metric method. The system compares the differences between the simulated vectors and the real vectors in terms of key features such as rainfall amount and rainfall duration and calculates the average of the sum of the squares of these differences. At the same time, the system also considers the representation vector identification results. If the identification results show that the simulated vectors have a high authenticity (such as a high similarity score), the error can be reduced to a certain extent. By calculating the network debugging error, the computer system can objectively evaluate the performance of the network for constructing the extreme representation vectors to be trained. If the error is large, it indicates that there are large deviations or deficiencies in the network when generating the extreme climate data representation vectors and further debugging and optimization are required. Conversely, if the error is small, it indicates that the network already has good performance and can be used for subsequent meteorological data analysis tasks.

[0148] Step S30VIII: Based on the network debugging error, debug the network for constructing the extreme representation vectors to be trained to obtain the network for constructing the extreme representation vectors.

[0149] In the embodiments of the present application, step S30VIII is a key step in optimizing and improving the generation process of extreme climate data representation vectors. In this step, the computer system debugs the network constructed for the extreme representation vectors to be trained based on the network debugging error, aiming to improve the accuracy of generating extreme climate data representation vectors by adjusting the network parameters and structure.

[0150] In step S30VII, the computer system has calculated the network debugging error based on the actual extreme climate data representation vectors and the representation vector identification results. This error quantifies the performance gap of the network constructed for the extreme representation vectors to be trained when generating extreme climate data representation vectors, providing a basis for subsequent debugging and optimization. In step S30VIII, the computer system starts to debug the network constructed for the extreme representation vectors to be trained. The debugging process generally involves the following key steps:

[0151] Parameter adjustment: The computer system first checks the parameter settings of the network, including the learning rate, weight decay, batch size, etc. These parameters have an important impact on the training speed and performance of the network. Based on the network debugging error, the system can try to adjust these parameters to find a better configuration. For example, if the network debugging error is large, the system can try to decrease the learning rate or increase the batch size to improve the convergence speed and stability of the network. Structure optimization: In addition to parameter adjustment, the system can also consider optimizing the network structure. This includes increasing or decreasing the number of network layers, changing the network layer type (such as using different activation functions or pooling methods), etc. The optimization of the network structure aims to improve the network's ability to process complex meteorological data and enable it to generate extreme climate data representation vectors more accurately.

[0152] Through continuous debugging and optimization, the computer system can finally obtain a network for constructing extreme representation vectors with excellent performance. This network can accurately generate high-quality extreme climate data representation vectors from non-extreme climate data, providing more valuable training samples for the training of the extreme climate pre-identification neural network, thereby improving the accuracy and reliability of meteorological data analysis.

[0153] For the above-mentioned extreme representation vector construction network to be trained in S30IV, it can be pre-constructed. For example, the extreme representation vector construction network to be trained includes a mutual attention component and two residual network components. Each residual network includes a convolutional network, a fully connected network, and an activation network. The generation method of the fourth enhanced information can refer to the construction process of the third enhanced information, which will not be elaborated here. For step S30V, based on the first non-extreme climate data representation vector and the fourth enhanced information, the sixth extreme climate data representation vector is generated through the extreme representation vector construction network to be trained. For step S30VI, based on the actual extreme climate data representation vector, the sixth extreme climate data representation vector is identified through the representation vector identification network to obtain the representation vector identification result. For step S30VII, first, the actual extreme climate data representation vector is obtained, and then based on the actual extreme climate data representation vector and the representation vector identification result, the network debugging error of the extreme representation vector construction network to be trained is determined; thus, in step S30VIII, based on the network debugging error, the network parameters of the extreme representation vector construction network to be trained are updated to debug the extreme representation vector construction network to be trained to obtain the extreme representation vector construction network.

[0154] As an implementation manner, step S30III, integrating the second extreme climate data representation vector and the fourth extreme climate data representation vector to obtain the fifth extreme climate data representation vector, may include:

[0155] Step S30IIIa: Obtain the first influence coefficient of the second extreme climate data representation vector and obtain the second influence coefficient of the fourth extreme climate data representation vector.

[0156] In the embodiments of the present application, step S30IIIa determines the weights, that is, influence coefficients, of the second extreme climate data representation vector and the fourth extreme climate data representation vector during the integration process. These influence coefficients reflect the contribution degrees of different representation vectors to the finally generated fifth extreme climate data representation vector.

[0157] In step S30IIIa, the primary task of the computer system is to obtain the first influence coefficient of the second extreme climate data representation vector and the second influence coefficient of the fourth extreme climate data representation vector. These influence coefficients are determined based on multiple factors, including but not limited to the statistical characteristics of historical meteorological data, expert knowledge, and the output of machine learning models. For the second extreme climate data representation vector (obtained by enhancing the original extreme climate data), its first influence coefficient may rely more on the statistical characteristics of historical data. For example, if historical data shows that in the target acquisition area, the representation vectors obtained by enhancing the original extreme climate data have a very high correlation with real extreme climate events, then the system can assign higher first influence coefficients to these vectors. This means that in the integration process, these vectors will occupy a greater weight and have a greater impact on the finally generated fifth extreme climate data representation vector.

[0158] For the fourth extreme climate data representation vector (generated based on non-extreme climate data and enhancement information), the determination of its second influence coefficient may be more complex. On the one hand, the system can consider the performance of these vectors in simulating the transition from non-extreme climate to extreme climate. If the simulated generated representation vectors can accurately reflect this transition process and have a high similarity with the real extreme climate data representation vectors, then the system can assign higher second influence coefficients to them. On the other hand, the system can also use machine learning models to predict or calculate these influence coefficients. For example, a regression model or a decision tree model can be used, taking some key features of the representation vectors (such as rainfall, wind speed, etc.) as inputs and outputting the corresponding influence coefficients. In this way, the system can dynamically adjust the weights of different representation vectors according to specific meteorological conditions and requirements. In practical applications, the determination process of these influence coefficients may be an iterative and optimization process. The computer system can first set the initial values of the influence coefficients based on initial settings or historical experience, and then make dynamic adjustments during the actual data integration and model training process. Through continuous attempts and verifications, the system can find the optimal combination of influence coefficients, making the integrated fifth extreme climate data representation vector most accurately reflect the characteristics of real extreme climate events.

[0159] Step S30IIIb: Based on the first influence coefficient and the second influence coefficient, adjust the second extreme climate data representation vector and the fourth extreme climate data representation vector to obtain the fifth extreme climate data representation vector.

[0160] In the embodiment of the present application, in step S30IIIb, the second extreme climate data representation vector and the fourth extreme climate data representation vector are weighted and adjusted by applying the first influence coefficient and the second influence coefficient, so as to generate the fifth extreme climate data representation vector. In step S30IIIa, the computer system has determined the first influence coefficient of the second extreme climate data representation vector and the second influence coefficient of the fourth extreme climate data representation vector. These influence coefficients are calculated based on various factors (such as the statistical characteristics of historical meteorological data, expert knowledge, or the output of a machine learning model), and they reflect the importance or contribution degree of different representation vectors in the integration process. In step S30IIIb, the computer system will use these influence coefficients to weight and adjust the second extreme climate data representation vector and the fourth extreme climate data representation vector. Specifically, the system will calculate according to the following formula:

[0161] V5 = a·V1 + b·V2;

[0162] Where V5 is the fifth extreme climate data representation vector to be generated, V1 is the second extreme climate data representation vector, V2 is the fourth extreme climate data representation vector, a is the first influence coefficient, and b is the second influence coefficient. The values of a and b are between 0 and 1, and a + b = 1, to ensure that the weighted and adjusted vector is still within a reasonable numerical range.

[0163] In this way, the computer system can combine the extreme climate data representation vectors from different sources to generate a fifth extreme climate data representation vector that not only contains the characteristics of real typhoon events but also reflects the simulated transition process. This vector will be used as part of the training data for the training and optimization of the extreme climate pre-identification neural network, thereby improving the prediction accuracy and generalization ability of the model. In practical applications, by adjusting the values of the first influence coefficient and the second influence coefficient, the contribution degree of different representation vectors in the integration process can be flexibly controlled to meet different meteorological data analysis requirements.

[0164] As an implementation manner, after step S30, which enhances the first extreme climate data representation vector corresponding to the set centroid of each representation vector grouping set to obtain the second extreme climate data representation vector corresponding to each first extreme climate data representation vector, the method may further include:

[0165] Step S30a: Evaluate multiple second extreme climate data representation vectors respectively to obtain the evaluation result of each second extreme climate data representation vector, and the evaluation result is used to represent the commonality measurement result between the second extreme climate data representation vector and the actual extreme climate data representation vector.

[0166] In the embodiment of the present application, in step S30a, each of the multiple second extreme climate data characterization vectors generated through enhancement processing is evaluated one by one to quantify the commonalities or similarities between these simulated vectors and the real extreme climate data characterization vectors. In step S30, the computer system has performed enhancement processing on the first extreme climate data characterization vector corresponding to the set centroid of the set of characterization vector groups, generating multiple second extreme climate data characterization vectors. These vectors simulate the characteristics of new extreme climate events that have changed from the original extreme climate data. However, the quality of these simulated vectors and whether they truly reflect the characteristics of extreme climate events need to be verified through an independent evaluation mechanism.

[0167] In step S30a, the task of the computer system is to construct an evaluation mechanism to evaluate multiple second extreme climate data characterization vectors. The purpose of the evaluation is to obtain a commonality measurement result between each simulated vector and the real extreme climate data characterization vector. This commonality measurement result can be a similarity score, a distance value, or other quantitative indicators used to characterize the similarity or difference degree between the simulated vector and the real data.

[0168] The evaluation process can be implemented based on multiple methods. A common method is to use a similarity measurement function to calculate the similarity between the simulated vector and the real vector. For example, the cosine similarity function can be used to calculate the cosine value of the angle between two vectors in terms of direction, thereby quantifying their similarity. The closer the cosine value is to 1, the more similar the two vectors are; the closer it is to -1, the more opposite the two vectors are; and being close to 0 indicates that the two vectors are orthogonal and have no similarity. Another method is to use a machine learning model for evaluation. This model can be a pre-trained characterization vector identification network that can classify or score the input extreme climate data characterization vectors. By inputting the simulated vector into this network, the system can obtain an output value as the evaluation result, which reflects the commonalities or similarities between the simulated vector and the real extreme climate data characterization vector. For example, if the output value is a probability value, it represents the likelihood that the simulated vector is the real extreme climate data characterization vector. Through the evaluation process of step S30a, the computer system can quantify the quality of the simulated extreme climate data characterization vectors, providing valuable reference information for subsequent data screening and neural network training.

[0169] Step S30b: Based on the evaluation results of each second extreme climate data characterization vector, determine the target extreme climate data characterization vectors whose evaluation results meet the evaluation requirements from the multiple second extreme climate data characterization vectors.

[0170] In the embodiments of the present application, step S30b screens out high-quality target extreme climate data characterization vectors that meet specific evaluation requirements from numerous simulated generated vectors based on the evaluation results of each second extreme climate data characterization vector obtained in step S30a. In step S30a, the computer system has evaluated each second extreme climate data characterization vector and obtained a quantitative evaluation result, which reflects the commonality or similarity between the simulated vector and the true extreme climate data characterization vector. This evaluation result may be a similarity score, a probability value, or other quantitative indicators, specifically depending on the choice of the evaluation method.

[0171] In step S30b, the computer system will use these evaluation results to screen the target extreme climate data characterization vectors. First, the system needs to set an evaluation requirement or threshold, which can be determined according to specific application scenarios and requirements. For example, it can be required that the similarity score of the simulated vector is higher than a specific value, or it can be required that the probability value output by the characterization vector identification network is greater than a certain threshold. These requirements or thresholds ensure that the selected vectors have high quality in terms of commonality and authenticity. Next, all second extreme climate data characterization vectors are traversed and judged according to their evaluation results. If the evaluation result of a certain vector meets the preset evaluation requirement or threshold, then it will be marked as the target extreme climate data characterization vector; otherwise, it will be excluded. This process ensures that only those simulated vectors that are closest to the true extreme climate events are selected for subsequent neural network training and optimization.

[0172] Through the screening process of step S30b, the computer system can ensure that only high-quality and high-authenticity simulated extreme climate data characterization vectors are used for the training of the neural network. This helps to improve the prediction accuracy and generalization ability of the neural network, enabling it to more accurately identify and predict extreme climate events in the real world.

[0173] As an implementation manner, after step S20, which performs a grouping and clustering operation on multiple extreme climate data characterization vectors to obtain one or more sets of characterization vector groupings, the method may further include:

[0174] Step S20A: In one or more sets of characterization vector groupings, determine a first set of characterization vector groupings, and determine the spatial coefficients between each second set of characterization vector groupings and the first set of characterization vector groupings, where the second set of characterization vector groupings is a set of characterization vector groupings different from the first set of characterization vector groupings in one or more sets of characterization vector groupings.

[0175] In the embodiment of the present application, in step S20A, a reference set is determined from one or more sets of grouped characterization vectors, and a spatial coefficient (such as vector distance) between other sets and the reference set is calculated. The purpose of this step is to quantify the relative positional relationship of different sets of grouped characterization vectors in the feature space, providing a basis for subsequent data processing and enhancement. In step S20, the computer system has performed clustering operations (clustering and grouping) on multiple extreme climate data characterization vectors, obtaining one or more sets of grouped characterization vectors. Each grouped set contains characterization vectors of a class of extreme climate event data with similar characteristics. In step S20A, the computer system selects one of these grouped sets as the first set of grouped characterization vectors, which is usually used as a reference point. The selection basis can be the size of the grouped set, the tightness of the internal vectors, or other statistical characteristics. For example, the grouped set with the largest number of vectors or the smallest distance between vectors can be selected as the first set of grouped characterization vectors.

[0176] Next, the spatial coefficient between all other grouped sets (the second set of grouped characterization vectors) and this first set of grouped characterization vectors is determined. The spatial coefficient can be measured by vector distance, which reflects the relative positional relationship of the two grouped sets in the feature space. Specifically, the Euclidean distance or cosine similarity between the centroid vector of each second set of grouped characterization vectors and the centroid vector of the first set of grouped characterization vectors can be calculated as the spatial coefficient. The centroid vector is usually the average or median of all vectors in a grouped set, representing the central position or main features of the set.

[0177] Through the calculation in step S20A, the computer system can obtain a spatial coefficient matrix, which contains the spatial coefficient values between each second set of grouped characterization vectors and the first set of grouped characterization vectors. This matrix will provide important reference information for subsequent data processing and enhancement, helping the computer system to formulate differential processing strategies based on the spatial relationships of different grouped sets.

[0178] Step S20B: In one or more sets of grouped characterization vectors, determine the third set of grouped characterization vectors with a spatial coefficient greater than the preset spatial coefficient, and the fourth set of grouped characterization vectors with a spatial coefficient not greater than the preset spatial coefficient.

[0179] In the embodiment of the present application, step S20B is a continuation of step S20A. It further classifies the set of characterization vector groups based on the calculated spatial coefficients. The purpose is to identify the set of groups that are far and near to the reference set in the feature space, so that subsequent data processing and enhancement strategies can perform differential processing for different types of extreme climate events. In step S20A, the computer system has determined a first set of characterization vector groups as the reference point and calculated the spatial coefficients between all other sets of groups (the second set of characterization vector groups) and this reference set. These spatial coefficients quantify the relative positional relationships of different sets of groups in the feature space.

[0180] In step S20B, the computer system sets a preset spatial coefficient threshold. This threshold can be determined according to specific application scenarios and requirements, and is used to distinguish between sets of groups with larger and smaller spatial coefficients. Then, the system traverses all the second sets of characterization vector groups and compares their spatial coefficients with the preset threshold.

[0181] If the spatial coefficient of a certain second set of characterization vector groups is greater than the preset threshold, then it will be classified as the third set of characterization vector groups. These sets are far from the first set of characterization vector groups (the reference set) in the feature space and may represent relatively rare or special types of extreme climate events. Since their quantity in the dataset is small, more attention may be needed in subsequent data enhancement processing to simulate more similar extreme climate events, thereby enriching the training dataset. On the contrary, if the spatial coefficient of a certain second set of characterization vector groups is not greater than the preset threshold, then it will be classified as the fourth set of characterization vector groups. These sets are close to the reference set in the feature space and may represent more common or typical types of extreme climate events. Since their quantity in the dataset is large, relatively fewer simulated data may need to be generated in subsequent data enhancement processing to avoid problems such as data redundancy and model overfitting. Through the classification in step S20B, the computer system can provide clear guidance for subsequent data enhancement processing. For the third set of characterization vector groups, the system can generate more simulated data to simulate these rare events; while for the fourth set of characterization vector groups, the system may need to generate relatively fewer simulated data to avoid data redundancy and model overfitting problems. This differential processing strategy helps to improve the generalization ability and accuracy of the extreme climate pre-identification neural network.

[0182] Based on this, in step S30, the first extreme climate data characterization vectors corresponding to the set centroids of each set of characterization vector groups are enhanced to obtain the second extreme climate data characterization vectors corresponding to each first extreme climate data characterization vector, which may specifically include:

[0183] Step S31a: Enhance the first extreme climate data representation vector corresponding to the set centroid of each set of third representation vector groups to obtain the second extreme climate data representation vector corresponding to each set of third representation vector groups.

[0184] In the embodiments of the present application, the goal of step S31a is to simulate more similar extreme climate data representation vectors through enhancement processing for those extreme climate event types that are far from the reference set in the feature space (i.e., relatively rare or special), so as to enrich the training dataset and improve the generalization ability of the model.

[0185] In step S20, the system groups and clusters multiple extreme climate data representation vectors through a clustering algorithm to obtain one or more sets of representation vector groups. Then, in steps S20A and S20B, the system determines a reference set and divides the other grouped sets into a set of third representation vector groups (far from the reference set) and a set of fourth representation vector groups (close to the reference set) based on a spatial coefficient (such as vector distance). Step S31a focuses on the set of third representation vector groups. These sets represent relatively rare or special extreme climate event types, and their quantity in the dataset may be small, so enhancement processing is needed to simulate more similar data. In step S31a, the computer system enhances the first extreme climate data representation vector corresponding to the set centroid of each set of third representation vector groups. The set centroid is usually the average or median of all vectors in the grouped set, and it represents the main features of the set.

[0186] The specific method of enhancement processing can be determined according to specific application scenarios and requirements. A common method is to add random perturbations or noises to the original data representation vector to simulate the uncertainty in the real world. These perturbations can be generated based on a certain probability distribution (such as a normal distribution) to ensure that the enhanced data not only retains the main features of the original data but also has a certain degree of diversity.

[0187] Through the enhancement processing of step S31a, more data representation vectors similar to rare or special extreme climate events can be simulated. These enhanced data will be used as part of the training data for the training and optimization of the extreme climate pre-identification neural network. Since they simulate relatively rare event types, they help improve the model's recognition ability and generalization performance for these events.

[0188] Step S32a: Enhance the first extreme climate data representation vectors corresponding to the set centroids of each set of fourth representation vector groups to obtain the second extreme climate data representation vectors corresponding to each set of fourth representation vector groups; wherein, the number of the second extreme climate data representation vectors corresponding to each set of third representation vector groups is less than the number of the second extreme climate data representation vectors corresponding to each set of fourth representation vector groups.

[0189] In the embodiment of the present application, step S32a focuses on enhancing the data in the set of fourth representation vector groups. These grouped sets are relatively close to the reference set in the feature space and usually represent more common or typical types of extreme climate events. Through the enhancement process, the computer system can simulate more similar data representation vectors, thereby further enriching the training data set and improving the generalization ability and accuracy of the extreme climate pre-recognition neural network. In step S32a, the computer system enhances the first extreme climate data representation vectors corresponding to the set centroids of each set of fourth representation vector groups. The set centroid usually represents the main features of the grouped set, so enhancing based on it can ensure that the generated second extreme climate data representation vectors maintain the main characteristics of the original data. Different from step S31a, since the set of fourth representation vector groups represents common or typical types of extreme climate events, their quantity in the data set is relatively large. Therefore, in step S32a, the system will generate more second extreme climate data representation vectors. The purpose of this is to further improve the recognition accuracy of these common events and ensure that the model can fully learn the features of these events during the training process.

[0190] The specific method of the enhancement process can be determined according to the specific application scenario and requirements. A common method is to add random perturbations or noises to the original data representation vectors to simulate the uncertainties in the real world. These perturbations can be generated based on a certain probability distribution (such as a normal distribution) to ensure that the enhanced data not only maintains the main features of the original data but also has a certain degree of diversity.

[0191] In step S32a, the computer system enhances the first extreme climate data representation vectors corresponding to the set centroids of each set of fourth representation vector groups (i.e., common or typical typhoon events). By adding random perturbations based on a normal distribution, the system generates multiple second extreme climate data representation vectors, which simulate extreme climate situations that are similar to but slightly different from the original common typhoon events. Since the event types represented by the set of fourth representation vector groups are relatively common, the system will generate a relatively large number of second extreme climate data representation vectors to ensure that the model can fully learn the features of these events during the training process.

[0192] Through the enhancement process of step S32a, the computer system can simulate more data representation vectors similar to common extreme climate events and generate a dataset containing rich and diverse extreme climate situations. This dataset, as part of the training data, is used for the training and optimization of the extreme climate pre-identification neural network. Since it contains more simulated data of common events, it helps to improve the recognition accuracy and generalization ability of the model for these events.

[0193] Figure 2 The following is a schematic diagram of the hardware entity of a computer system provided by an embodiment of the present application. As Figure 2 shown, the hardware entity of the computer system 1000 includes: a processor 1001 and a memory 1002. Among them, the memory 1002 stores a computer program that can run on the processor 1001, and when the processor 1001 executes the program, it implements the steps in the method of any of the above embodiments.

Claims

1. A method for generating weather reports based on deep learning, characterized in that, The method includes: Obtaining target meteorological acquisition data; Loading the target meteorological acquisition data into a pre - debugged extreme climate pre - recognition neural network for processing to obtain an extreme climate pre - recognition result; Generating a meteorological report based on the predicted extreme climate pre - recognition result; Wherein, the debugging process of the extreme climate pre - recognition neural network includes the following steps: Extracting extreme climate data characterization vectors corresponding to multiple extreme climate training data respectively; Wherein, the extreme climate training data corresponds to a target acquisition area. When performing extreme climate pre - recognition on the target acquisition area according to the extreme climate training data, the obtained extreme climate pre - recognition result is used to indicate that the target acquisition area has extreme climate conditions; Performing a grouping and clustering operation on the multiple extreme climate data characterization vectors to obtain one or more sets of characterization vector groupings; Among the one or more sets of characterization vector groupings, determining a first set of characterization vector groupings, and determining the spatial coefficients between each second set of characterization vector groupings and the first set of characterization vector groupings, where the second set of characterization vector groupings is the set of characterization vector groupings different from the first set of characterization vector groupings among the one or more sets of characterization vector groupings; Among the one or more sets of characterization vector groupings, determining a third set of characterization vector groupings with a spatial coefficient greater than a preset spatial coefficient, and a fourth set of characterization vector groupings with a spatial coefficient not greater than the preset spatial coefficient; The enhancing the first extreme climate data characterization vector corresponding to the centroid of each set of characterization vector groupings to obtain a second extreme climate data characterization vector corresponding to each first extreme climate data characterization vector includes: Enhancing the first extreme climate data characterization vector corresponding to the centroid of each third set of characterization vector groupings to obtain a second extreme climate data characterization vector corresponding to each third set of characterization vector groupings; Enhancing the first extreme climate data characterization vector corresponding to the centroid of each fourth set of characterization vector groupings to obtain a second extreme climate data characterization vector corresponding to each fourth set of characterization vector groupings; Wherein, the number of second extreme climate data characterization vectors corresponding to each third set of characterization vector groupings is less than the number of second extreme climate data characterization vectors corresponding to each fourth set of characterization vector groupings; Enhancing the first extreme climate data characterization vector corresponding to the centroid of each set of characterization vector groupings to obtain a second extreme climate data characterization vector corresponding to each first extreme climate data characterization vector; Iteratively debugging the extreme climate pre - recognition neural network based on the second extreme climate data characterization vectors, and stopping the debugging when reaching a preset convergence condition to obtain the debugged extreme climate pre - recognition neural network, which is used to perform extreme climate pre - recognition on target meteorological acquisition data; Before enhancing the first extreme climate data representation vectors corresponding to the set centroids of each of the representation vector grouping sets to obtain the second extreme climate data representation vectors corresponding to each of the first extreme climate data representation vectors, the method further includes: Construct one or more first enhancement information, where each of the first enhancement information is different from each other; The enhancing the first extreme climate data representation vectors corresponding to the set centroids of each of the representation vector grouping sets to obtain the second extreme climate data representation vectors corresponding to each of the first extreme climate data representation vectors includes: For each of the representation vector grouping sets, perform the following operations: Integrate the first enhancement information into the first extreme climate data representation vectors corresponding to the set centroids of the representation vector grouping sets respectively to obtain one or more second extreme climate data representation vectors corresponding to the first extreme climate data representation vectors; The constructing one or more first enhancement information includes: Perform one or more sparse extractions on the data satisfying the first data dispersion situation to obtain one or more sparse extraction results, and use each of the sparse extraction results as each of the first enhancement information, where each of the sparse extraction results is different from each other; Or; Obtain the data satisfying one or more second data dispersion situations, and for each of the second data dispersion situations, perform sparse extraction on the data satisfying the second data dispersion situation to obtain the first enhancement information, where each of the second data dispersion situations is different from each other.

2. The method according to claim 1, characterized in that, Before enhancing the first extreme climate data representation vectors corresponding to the set centroids of each of the representation vector grouping sets to obtain the second extreme climate data representation vectors corresponding to each of the first extreme climate data representation vectors, the method further includes: Construct the second enhancement information corresponding to each of the representation vector grouping sets; The enhancing the first extreme climate data representation vectors corresponding to the set centroids of each of the representation vector grouping sets to obtain the second extreme climate data representation vectors corresponding to each of the first extreme climate data representation vectors includes: For each of the representation vector grouping sets, perform the following operations: Integrate the second enhancement information corresponding to the representation vector grouping set into the first extreme climate data representation vector corresponding to the set centroid of the representation vector grouping set to obtain the second extreme climate data representation vector corresponding to the first extreme climate data representation vector.

3. The method according to claim 2, characterized in that, The constructing the second enhancement information corresponding to each of the representation vector grouping sets includes: For each of the representation vector grouping sets, perform the following operations: Determine the vector probability density distribution of the representation vector grouping set; Construct the data satisfying the vector probability density distribution of the representation vector grouping set; Perform sparse extraction on the data satisfying the vector probability density distribution of the representation vector grouping set to obtain the second enhancement information corresponding to the representation vector grouping set.

4. The method according to claim 1, wherein After performing enhancement processing on the first extreme climate data representation vectors corresponding to the set centroids of each of the grouped sets of the representation vectors to obtain second extreme climate data representation vectors corresponding to each of the first extreme climate data representation vectors, the method further includes: Obtain supplementary training data that does not originate from the target acquisition area; Extract supplementary training data representation vectors of the supplementary training data; Integrate the second extreme climate data representation vectors and the supplementary training data representation vectors to obtain third extreme climate data representation vectors; and / or; use the supplementary training data representation vectors as the third extreme climate data representation vectors; After performing enhancement processing on the first extreme climate data representation vectors corresponding to the set centroids of each of the grouped sets of the representation vectors to obtain second extreme climate data representation vectors corresponding to each of the first extreme climate data representation vectors, the method further includes: Extract non-extreme climate data representation vectors of non-extreme climate training data and construct third enhancement information; Wherein, the non-extreme climate training data corresponds to the target acquisition area, and when pre-identifying extreme climate in the target acquisition area based on the non-extreme climate training data, the obtained extreme climate pre-identification result is used to indicate that there is no extreme climate condition in the target acquisition area; Based on the non-extreme climate data representation vectors and the third enhancement information, generate fourth extreme climate data representation vectors through an extreme representation vector construction network; Integrate the second extreme climate data representation vectors and the fourth extreme climate data representation vectors to obtain fifth extreme climate data representation vectors.

5. The method according to claim 4, characterized in that, The method further includes: Obtain an extreme representation vector construction network to be trained, first non-extreme climate data representation vectors of first non-extreme climate training data, and construct fourth enhancement information; Based on the first non-extreme climate data representation vectors and the fourth enhancement information, construct sixth extreme climate data representation vectors through the extreme representation vector construction network to be trained; Identify the sixth extreme climate data representation vectors through a representation vector identification network to obtain a representation vector identification result, where the representation vector identification result is used to represent the probability that the sixth extreme climate data representation vectors are actual extreme climate data representation vectors; Based on the actual extreme climate data representation vectors and the representation vector identification result, determine the network debugging error of the extreme representation vector construction network to be trained; Based on the network debugging error, debug the extreme representation vector construction network to be trained to obtain the extreme representation vector construction network; The integrating the second extreme climate data representation vectors and the fourth extreme climate data representation vectors to obtain fifth extreme climate data representation vectors includes: Obtain a first influence coefficient of the second extreme climate data representation vectors and obtain a second influence coefficient of the fourth extreme climate data representation vectors; Based on the first influence coefficient and the second influence coefficient, the second extreme climate data representation vector and the fourth extreme climate data representation vector are adjusted to obtain the fifth extreme climate data representation vector.

6. The method according to claim 1, wherein After enhancing the first extreme climate data representation vector corresponding to the set centroid of each set of the representation vector groups to obtain the second extreme climate data representation vector corresponding to each of the first extreme climate data representation vectors, the method further includes: Evaluating each of the multiple second extreme climate data representation vectors to obtain an evaluation result for each of the second extreme climate data representation vectors, where the evaluation result is used to represent the commonality measurement result between the second extreme climate data representation vector and the actual extreme climate data representation vector; Based on the evaluation result of each of the second extreme climate data representation vectors, a target extreme climate data representation vector whose evaluation result meets the evaluation requirements is determined from the multiple second extreme climate data representation vectors.

7. A computer system, comprising a memory and a processor, the memory storing a computer program that can run on the processor, characterized in that, When the processor executes the program, the steps in the method according to any one of claims 1 to 6 are implemented.

Images