A vertical meteorological element prediction method and platform based on a big data framework and web technology

Abstract

The application discloses a vertical meteorological element prediction method and platform based on a big data framework and a Web technology, and comprises the following steps: S1, meteorological data management: adopting a big data processing architecture Hadoop, combining a distributed file system and a distributed database, and performing reliable distributed storage on uploaded meteorological data; S2, meteorological data analysis: performing graphic visual analysis on the meteorological data stored in S1; S3, meteorological data prediction: importing the meteorological data stored in S1 into an SSA-KNN model, a CNN-LSTM model and a mathematical model for vertical prediction, and outputting vertical meteorological prediction results at different altitudes. According to the prediction method, the SSA-KNN and CNN-LSTM meteorological element prediction models and the mathematical model of the correction formula in the standard specification can be used to predict the vertical meteorological conditions of high-rise building positions according to the station observation data submitted by users, and the users can obtain and export the prediction data, thereby assisting in improving the reliability of high-rise building load analysis results.

Description

Technical Field

[0001] This invention belongs to the field of vertical meteorological element prediction, specifically involving a method and platform for vertical meteorological element prediction based on big data framework and Web technology. Background Technology

[0002] Vertical meteorological conditions for high-rise buildings are crucial for building envelope, environmental, and energy management. However, most current meteorological observation stations only provide surface-level meteorological data, lacking upper-air vertical meteorological observation data. This leads to data scarcity and forecasting difficulties in meteorological research and applications related to high-rise buildings. Furthermore, existing meteorological data forecasting systems are mostly built using traditional methods, with limited data processing capabilities, failing to efficiently process massive amounts of meteorological data and accurately predict vertical meteorological elements. Therefore, there is an urgent need for a platform capable of efficiently processing and accurately predicting vertical meteorological data to address the shortcomings of existing technologies. Summary of the Invention

[0003] The purpose of this invention is to provide a method and platform for predicting vertical meteorological elements based on a big data framework and Web technology. This method can efficiently process and accurately predict vertical meteorological element data at different altitudes, solving the problems of insufficient data processing capabilities and lack of accurate prediction methods for meteorological element data at different altitudes in the existing technology.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] A method for predicting vertical meteorological elements based on a big data framework and Web technology includes: S1. Meteorological Data Management: The system adopts the Hadoop big data processing architecture, combined with a distributed file system and a distributed database, to reliably distribute and store the uploaded meteorological data. S2, Meteorological Data Analysis: Perform graphical visualization analysis on the meteorological data stored in S1; S3. Meteorological data prediction: Import the meteorological data stored in S1 into the SSA-KNN model, CNN-LSTM model and mathematical model for vertical prediction, and output the vertical meteorological prediction results at different altitudes.

[0006] Optionally, the calculation process of the SSA-KNN model in S3 is as follows: A flock of sparrows It can be represented as: ; In the formula, Let represent the number of neighbors of the i-th sparrow. Weighting factors in the time-feature distance function The decay parameter of the neighborhood sample weighting function. For the selected distance metric function type, The selected weighting function type; Using a set of parameters represented by each sparrow A model is constructed, and a regression prediction task is run on the test set. The error of the regression result corresponding to each meteorological element prediction model is recorded. In each iteration, all sparrows fly in a multidimensional parameter space, and their position vectors are represented as follows: The specific update format is as follows: Global Discoverer Update: ; In the formula, This represents the position of the i-th sparrow in the nth iteration; This represents the position of the i-th sparrow in the (n+1)-th iteration. It is the currently found globally optimal position, that is, the parameter that performs best across all iterations. It is the position of another sparrow randomly selected in the current iteration. and It is a weighting factor; Local search update: ; In the formula, This represents the position of a sparrow adjacent to the i-th sparrow. It is a weighting factor used for adjustment. right The influence of this, controlling how close the sparrow gets to the nearest optimal solution, Used to generate random numbers in the interval [0,1]; Introduce the time-feature joint distance function as the default distance evaluation metric for constructing neighborhoods: ; In the formula, This represents the time-feature joint distance function, used to comprehensively measure the spatial similarity and temporal consistency between two samples. This represents the weighted ratio controlling for spatial distance and temporal difference, where t represents the time label of the target sample. x This represents the meteorological input of the target prediction sample. The input features represent the m-th candidate neighbor; A distance-based sample weighting strategy is used to construct the regression prediction function: ; In the formula, For the predicted final temperature of the j-th time, the weights are... From SSA , and Control is calculated based on different functions. For the number of neighbors, As the output of the model, these correspond to the temperature measured at the i-th floor.

[0007] Optionally, the computation process of the CNN-LSTM model in S3 is as follows: Transpose the tensor at each time step t: ; R is a high-dimensional space, and B is the size of each batch of samples. This represents the meteorological input of the target prediction sample at time t; Using convolution to extract spatial features of meteorological elements: ; In the formula, C is the number of convolution kernels. The height dimension is the result of convolution; After each convolution calculation, a bias term b is introduced, and an activation function is used to enhance the model's ability to express nonlinear patterns: ; Flattened into vector form: ; This represents the feature map obtained after the convolutional layer; Stack the spatial feature vectors from all time steps, where T is the temperature, to form the sequence input: ; Through the above operations, the meteorological data at each time step is encoded into a high-dimensional spatial feature vector as the input to the LSTM module; After the LSTM layer, a fully connected layer is added to map the temporal features to the final prediction output; The calculation process is as follows: ; In the formula, W is the weight matrix. is the output of the LSTM layer, b is the bias term, and z is the predicted value of meteorological elements.

[0008] Optionally, the mathematical model in S3 is a calculation for vertical correction of meteorological elements in specifications and standards. Based on the vertical correction formulas specified in the standards and specifications of meteorology, architecture and other fields, the model performs height dimension correction calculations for meteorological elements. Its core purpose is to map the meteorological data observed on the ground to different vertical heights through standardized mathematical derivation, and obtain the estimated value of meteorological elements at that height.

[0009] A vertical meteorological element prediction platform based on big data framework and Web technology includes: a data management module, a meteorological data analysis module, and a meteorological prediction module; The data management module adopts the Hadoop big data processing architecture, combined with a distributed file system and a distributed database, to reliably distribute and store the uploaded meteorological data. The meteorological data analysis module enables data preview and graphical visualization analysis; The weather forecasting module imports weather data into the SSA-KNN model, CNN-LSTM model, and mathematical model for vertical forecasting, and outputs vertical weather forecast results at different altitudes.

[0010] Optionally, the calculation process of the SSA-KNN model in the meteorological forecasting module is as follows: A flock of sparrows It can be represented as: ; In the formula, Let represent the number of neighbors of the i-th sparrow. Weighting factors in the time-feature distance function The decay parameter of the neighborhood sample weighting function. For the selected distance metric function type, The selected weighting function type; Using a set of parameters represented by each sparrow A model is constructed, and a regression prediction task is run on the test set. The error of the regression result corresponding to each meteorological element prediction model is recorded. In each iteration, all sparrows fly in a multidimensional parameter space, and their position vectors are represented as follows: The specific update format is as follows: Global Discoverer Update: ; In the formula, This represents the position of the i-th sparrow in the nth iteration; This represents the position of the i-th sparrow in the (n+1)-th iteration. It is the currently found globally optimal position, that is, the parameter that performs best across all iterations. It is the position of another sparrow randomly selected in the current iteration. and It is a weighting factor; Local search update: ; In the formula, This represents the position of a sparrow adjacent to the i-th sparrow. It is a weighting factor used for adjustment. right The influence of this, controlling how close the sparrow gets to the nearest optimal solution, Used to generate random numbers in the interval [0,1]; Introduce the time-feature joint distance function as the default distance evaluation metric for constructing neighborhoods: ; In the formula, This represents the time-feature joint distance function, used to comprehensively measure the spatial similarity and temporal consistency between two samples. This represents the weighted ratio controlling for spatial distance and temporal difference, where t represents the time label of the target sample. x This represents the meteorological input of the target prediction sample. The input features represent the m-th candidate neighbor; A distance-based sample weighting strategy is used to construct the regression prediction function: ; In the formula, For the predicted final temperature of the j-th time, the weights are... From SSA , and Control is calculated based on different functions. For the number of neighbors, As the output of the model, these correspond to the temperature measured at the i-th floor.

[0011] Optionally, the computation process of the CNN-LSTM model in the weather forecasting module is as follows: Transpose the tensor at each time step t: ; R is a high-dimensional space, and B is the size of each batch of samples. This represents the meteorological input of the target prediction sample at time t.

[0012] Using convolution to extract spatial features of meteorological elements:

[0013] In the formula, C is the number of convolution kernels. The height dimension is the height dimension after convolution.

[0014] After each convolution calculation, a bias term b is introduced, and an activation function is used to enhance the model's ability to express nonlinear patterns: ; Flattened into vector form: ; This represents the feature map obtained after the convolutional layer (Conv1D).

[0015] Stack the spatial feature vectors from all time steps, where T is the temperature, to form the sequence input: ; Through the above operations, the meteorological data at each time step is encoded into a high-dimensional spatial feature vector as the input to the LSTM module; After the LSTM layer, a fully connected layer is added to map the temporal features to the final prediction output; The calculation process is as follows:

[0016] In the formula, W is the weight matrix. is the output of the LSTM layer, b is the bias term, and z is the predicted value of meteorological elements.

[0017] Optionally, the mathematical model is a calculation for vertical correction of meteorological elements in specifications and standards. Based on the vertical correction formulas specified in standards and specifications in meteorology, construction and other fields, the model performs height dimension correction calculations for meteorological elements. Its core purpose is to map ground-observed meteorological data to different vertical heights through standardized mathematical derivation, and obtain the estimated value of meteorological elements at that height.

[0018] Optionally, a visualization module is also included to provide an intuitive display of the prediction results, making it easier for users to analyze and download them.

[0019] The beneficial effects of this invention are: (1) It has achieved efficient storage and management of massive meteorological data, solving the problem of limited data scale in traditional technologies; (2) The integration of multiple forecasting models has improved the accuracy and practicality of vertical meteorological element forecasting; (3) It provides a user-friendly interface and an intuitive way of displaying data, which greatly improves the efficiency of data analysis and user experience. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is the overall architecture diagram of the platform of this invention; Figure 2 This is a functional module design diagram of the platform of this invention; Figure 3This is a login sequence diagram of the platform of this invention; Figure 4 This is a data upload diagram of the platform of this invention; Figure 5 This is a schematic diagram of data analysis on the platform of this invention; Figure 6 This is a schematic diagram of vertical meteorological data prediction on the platform of this invention; Figure 7 This is a diagram showing the prediction results of the platform of this invention. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0022] Part One: The Vertical Meteorological Element Prediction Platform for High-Rise Buildings aims to provide forecasting support for meteorological stations lacking meteorological data at specific altitudes. This platform acquires surface meteorological data from meteorological stations and uploads it to the platform. After uploading, the data is automatically entered into a database for easy management and storage. Simultaneously, the platform has a dedicated data analysis module, allowing users to display and analyze the uploaded data and download the results for further research. Furthermore, the platform has a built-in meteorological forecasting function, enabling users to directly access stored meteorological station data, select appropriate forecasting models and vertical height elements, and generate vertical meteorological forecasts for different altitudes. The platform also provides a visual presentation of the forecast data, helping users to understand and apply the meteorological data more intuitively and conveniently.

[0023] This invention presents a locally deployed vertical meteorological element prediction platform that enables unified management of various meteorological observation data uploaded by users. Specifically, the user interface is implemented using HTML5, JavaScript, CSS, and Bootstrap technologies; basic information such as user accounts is managed by a MySQL database; and meteorological data is read and maintained through HDFS and HBase. The system backend is built using Java and integrates open-source frameworks such as Spring Boot, MyBatis, Hadoop, and HBase; it also utilizes Python for graphical data display and analysis. Furthermore, the platform incorporates mathematical models based on the previously researched SSA-KNN and CNN-LSTM meteorological element prediction models and corrected formulas from standard specifications. It can predict vertical meteorological conditions at high-rise building locations based on user-submitted station observation data. Users can obtain and export this predicted data, helping to improve the reliability of high-rise building load analysis results.

[0024] Part Two: The business process design of the platform of this invention is as follows: Figure 1As shown. To achieve comprehensive management and efficient utilization of meteorological data, the vertical meteorological element prediction platform has constructed three main business processes around the complete data processing lifecycle: data management and storage, core business processing logic, and front-end page and user interaction, which together support the stable operation of the platform. Details are as follows: (1) Data management and storage The platform supports users uploading meteorological data obtained from weather stations to the server. To ensure efficient data storage and subsequent processing capabilities, it adopts the Hadoop big data processing architecture, combined with its core components HDFS (distributed file system) and HBase (distributed database), to achieve reliable distributed storage of large-scale meteorological data.

[0025] (2) Core business processing logic Within the system, all critical functionalities, such as file upload and download, data archiving and management, graphical analysis, and vertical weather forecasting, are handled by the backend business logic module. Users initiate various requests through the frontend page. After receiving the instructions, the system backend processes them according to different operation types. Once processing is complete, the platform organizes the final results and returns them to the frontend for display, thus achieving effective linkage between user requests and system functions.

[0026] (3) Front-end page and user interaction The platform's user interface primarily serves as a functional demonstration and user access point. It organizes various module entry points through a visual interface, facilitating browsing, querying, and operation for users. Simultaneously, the front-end page receives and clearly presents the back-end processing results, including weather charts and forecasts, enabling users to intuitively acquire and understand data analysis findings in a graphical manner.

[0027] Part Three: The software architecture of this invention platform can be divided into a presentation layer, a business layer, and a data layer.

[0028] The presentation layer is primarily responsible for user interaction and page display functions. This part undertakes data analysis and visualization tasks for the platform's functional modules, including interactive interface design, chart output, and user input response. In terms of front-end technology, the Bootstrap framework is used, combined with HTML5, CSS, and JavaScript to achieve flexible page layout, fast response speed, and excellent visual effects.

[0029] The business layer, located at the heart of the system architecture, serves as the bridge connecting the front-end pages and back-end data. This layer is responsible for receiving various requests from the front-end, executing business logic processing, and completing service calls and data interactions between modules. It is crucial for realizing the platform's core functions. Operations such as user management, uploading, downloading, and deleting meteorological data files, file storage and analysis, and the prediction and display of vertical meteorological elements are all uniformly coordinated and processed by this layer, ensuring smooth and efficient business processes.

[0030] The data layer provides the platform with underlying data management capabilities, employing HDFS and HBase from the Hadoop ecosystem for distributed storage of large-scale meteorological data. The platform dynamically allocates storage locations based on file size: files larger than 20MB are stored in HDFS to improve data fault tolerance and scalability; while data smaller than 20MB is stored in HBase to optimize access performance for small files. Simultaneously, a MySQL database is used to store metadata information such as user accounts and Bucket configurations, further enhancing the structure and stability of system data access and ensuring the overall data security and efficient operation of the platform.

[0031] This platform achieves the prediction and management of vertical meteorological elements through the organic collaboration of its presentation layer, business layer, and data layer. The presentation layer is primarily responsible for user interaction and data visualization, providing an intuitive interface and graphical displays to enhance the user experience and enable users to view and understand meteorological data more easily. The business layer, as the core of the system, handles the uploading, analysis, and prediction of meteorological data, ensuring the efficiency of each user operation and the system's response speed. The data layer, through Hadoop and HBase technologies, supports the distributed storage and management of large-scale meteorological data, ensuring data security and efficient access. The platform's design fully considers the stability of data processing, the accuracy of prediction, and the scalability of the system. Through Hadoop's distributed storage and HBase's efficient data access mechanism, reliable storage and rapid processing capabilities for large-scale meteorological data are ensured. The platform not only efficiently processes and analyzes meteorological data but also provides technical support for meteorological forecasting and the construction industry.

[0032] Part Four: The modular design of this invention platform is based on actual user application needs and relevant business specifications. The platform comprises several key functional modules. Figure 2 The design framework of each functional module of this platform is shown, and the main modules of the platform are explained.

[0033] (1) User Management Module. This module is used to implement user account creation and authentication functions. When accessing the platform for the first time, users need to complete the account registration process. After successful registration, users can enter their account name, password and verification code information through the login interface. After successful verification, they can enter the system's main page to carry out related operations.

[0034] (2) Data Upload and Download Module. This module manages the storage of meteorological station data uploaded by users. Before using the data analysis and graphical display functions, users must first complete the upload of meteorological files. At the same time, the platform also provides download and deletion functions for stored meteorological data, making it convenient for users to manage and access relevant data resources as needed.

[0035] (3) Meteorological data analysis module. This module provides previews and chart visualization analysis of user-uploaded data. For example, it provides previews and line graph displays of meteorological elements such as temperature and air pressure from uploaded meteorological stations, improving the readability of meteorological documents from different regions and facilitating data interpretation and analysis.

[0036] (4) Meteorological data prediction module. It is mainly responsible for importing the uploaded meteorological element data into the platform's built-in SSA-KNN, CNN-LSTM models and mathematical models for vertical prediction, outputting vertical meteorological prediction results at different altitudes, and intuitively presenting the prediction results of the user-selected model and the corresponding prediction altitude, which can be downloaded and used by the user.

[0037] Part Five: To enable the operation of the platform's core functions and the structured management of data, this invention platform introduces a reasonable database system. The following will introduce the design ideas and structure definitions of the key tables, starting from the uses and structures of the two types of databases.

[0038] (1) In the platform, MySQL is mainly used to store user information, logical resource directories, and metadata related to platform operations. The design process strictly follows the principles of field specifications, primary key uniqueness, and data security to ensure clear relationships between tables and efficient access.

[0039] User Information Table. This table is specifically used to store basic identity information submitted by users during registration, such as user ID, account name, and login password. To enable efficient join queries, the user_id field is set as the primary key and used to join other data tables. Passwords are stored using MD5 encryption to improve account security. The structure is shown in Table 1.

[0040] Table 1 User Information Table

[0041] The Bucket table is used to logically divide user-uploaded weather files for easy grouping, management, and operation. Each user can create multiple Buckets, each with a unique name, creation time, and owner information. Before uploading data, users must select a target Bucket as the storage container. The structure is shown in Table 2.

[0042] Table 2 Bucket Table

[0043] (2) To meet the need for efficient management of meteorological files and their directory path information in the platform, the system adopts HBase database as the main storage engine for unstructured data. The design follows the idea of ​​"separation of path and content", and the directory information and file content are constructed as two logical tables respectively to improve query performance and avoid hot spot problems.

[0044] The path table focuses on recording the storage paths of files within the system. This table uses the path string as the RowKey and has two column families: sub and cf. The sub family stores information about all subdirectories under that path, while the cf family stores path-related metadata. See Table 3 for the structure.

[0045] Table 3 Path Table

[0046] The file table stores the actual file content and related metadata. Its RowKey is composed of the seqid from the path table and the filename, ensuring unique naming and facilitating fast retrieval. The table contains two column families: the c column family stores the file content, while the cf column family records basic information such as the uploader, file size, and type. The structure is shown in Table 4.

[0047] Table 4 Document Table

[0048] Part Six: The implementation of this vertical meteorological element prediction platform mainly includes registration, login, uploading, analysis, and prediction. The specific implementation steps are as follows.

[0049] Step 1: Implementation and result display of the platform of this invention. The login sequence diagram is as follows: Figure 3 As shown, once a user correctly enters their username and password on the login page, the system will complete identity verification and successfully redirect the user. If the information is incorrect, an error message will appear indicating an incorrect password, and login will fail. If a user has not yet registered an account, they can click the "Register" button to enter the registration interface, fill in their username, name, and password, and complete the registration. After successful registration, the user can log in to the platform with their new account.

[0050] Step 2, the data upload function of the platform of the present invention is as follows: Figure 4 As shown, the system's data upload interface is primarily used to import user's local data into the platform, serving as a crucial entry point for the entire platform's data. Taking a file named "4th Floor" as an example, this file contains various meteorological elements for which the user needs to perform vertical forecasting. First, the user must enter the identifier name of the uploaded data in the "Dataset Name" input box for subsequent system classification, management, and indexing. Then, the user selects the business module or analysis task to which the data belongs via a drop-down menu, ensuring the uploaded data falls into the correct task category, thereby improving the standardization of data organization and system query efficiency. The user selects the "4th Floor" format data file to be uploaded from their local machine. The system front-end performs preliminary verification of the file type and format to ensure data consistency and legality. After clicking the "Upload" button, the data file is submitted to the server for data preprocessing and warehousing.

[0051] Step 3, the data analysis module of this invention serves as the core functional area for data display within the platform, such as... Figure 5 As shown, this module primarily enables the visualization and basic statistical analysis of user-uploaded ground meteorological datasets. Currently displayed is a time-series plot of temperature data from the uploaded "4th floor" data. The plot uses a line graph to visually represent the temperature change trend, allowing users to quickly identify periodic changes, abnormal fluctuations, and local extrema in the data. Legends and axis descriptions are provided above the chart, enhancing its readability and professionalism. Through the development of this module, the platform not only improves the interpretability and analysis efficiency of the data but also provides non-professional users with an intuitive and easy-to-use entry point for data understanding, significantly enhancing the platform's data analysis support capabilities and application value.

[0052] Step 4, wherein the calculation process of the SSA-KNN model of the present invention is as follows: Each sparrow corresponds to a set of KNN model parameters to be optimized, including the number of neighbors K, the time-feature distance weight factor algorithm. The types of distance metric functions and neighborhood sample weighting functions are considered. Therefore, the position of each sparrow can be represented by a vector, and a flock of sparrows can be represented as: ; In the formula, Let represent the number of neighbors of the i-th sparrow. Weighting factors in the time-feature distance function The decay parameter of the neighborhood sample weighting function. For the selected distance metric function type, The selected weighting function type.

[0053] Using a set of parameters represented by each sparrow Build a model, run a regression prediction task on the test set, and record the error of the regression results corresponding to each meteorological element prediction model.

[0054] In each iteration, all individual sparrows fly in a multidimensional parameter space, and their position vectors are represented as follows: The specific update format is as follows: Global Discoverer Update: ; In the formula, This represents the position of the i-th sparrow in the nth iteration; This represents the position of the i-th sparrow in the (n+1)-th iteration. It is the currently found globally optimal position, that is, the parameter that performs best across all iterations. It is the position of another sparrow randomly selected in the current iteration. and It is a weighting factor used to control how close the sparrow is to the global optimal solution and random positions.

[0055] Local search update: ; In the formula, This represents the position of a sparrow adjacent to the i-th sparrow. It is a weighting factor used for adjustment. right The influence of this, controlling how close the sparrow gets to the nearest optimal solution, Used to generate random numbers in the [0,1] interval, increasing the randomness of the search and preventing the algorithm from converging too early.

[0056] To improve the model's spatial similarity recognition capability in time series data, a time-feature joint distance function is introduced as the default distance evaluation metric for constructing neighborhoods: ; In the formula, This represents the time-feature joint distance function, used to comprehensively measure the spatial similarity and temporal consistency between two samples. This represents the weighted ratio of controlling spatial distance and temporal difference, where t represents the time label of the target sample, and x represents the meteorological input of the target prediction sample. Let m represent the input features of the m-th candidate neighbor.

[0057] To better describe the contribution relationship of neighborhood samples of different meteorological elements in the prediction, a distance-based sample weighting strategy is adopted to construct a regression prediction function: ; In the formula, For the predicted final temperature of the j-th time, the weights are... From SSA , and Control is calculated based on different functions. For the number of neighbors, As the output of the model, these correspond to the temperature measured at the i-th floor.

[0058] Step 5, the computation flow of the CNN-LSTM model of this invention platform is as follows: In the convolutional layer, by allowing the convolutional kernel to slide across various regions of the input data, convolutional calculations are performed step by step, thereby effectively extracting local feature information from the input. For the tensor at each time step t, the following is performed: ; R is a high-dimensional space, and B is the size of each batch of samples. This represents the meteorological input of the target prediction sample at time t.

[0059] Features extracted from the spatial characteristics of meteorological elements using convolution: ; In the formula, C is the number of convolution kernels. The height dimension is the height dimension after convolution.

[0060] After each convolution calculation, a bias term b is typically introduced, followed by an activation function (ReLU) to enhance the model's ability to express nonlinear patterns. ; Flattened into vector form: ; This represents the feature map obtained after the convolutional layer (Conv1D).

[0061] Stack the spatial feature vectors from all time steps to form the sequence input: ; Through the above operations, the meteorological data at each time step is encoded into a high-dimensional spatial feature vector, reflecting its vertical structure information. This vector sequence will serve as the input to the LSTM module, further modeling the evolutionary relationships over time. The flattened feature vectors are organized according to sequence length to meet the requirements of LSTM in processing time series data, enabling it to capture time dependencies. It is an activation function. It is the hyperbolic tangent activation function. Indicates to hide state and input To splice, It is the output of the forget gate. It is the output of the input gate. It is the output of the output gate. This is new information about the state of the cell to be stored. It is the cell state after time t is updated. The hidden state after time t is updated. , , , These are the weight matrices corresponding to the forget gate, input gate, output gate, and cell state, respectively. , , , These are the bias terms for the corresponding gate and cell states, respectively.

[0062] Forget Gate: ; Input Gate: ; Output Gate: ; Update cell status: ; Cell status update: ; Hidden status update: ; Following the LSTM layer, a fully connected layer is added to map the temporal features to the final predicted output. The computation process is as follows: ; In the formula, W is the weight matrix. is the output of the LSTM layer, b is the bias term, and z is the predicted value of meteorological elements.

[0063] Step 6, the prediction result module of the platform of this invention is a key functional unit that supports the prediction of vertical meteorological elements based on ground observation data, such as... Figure 6 and Figure 7As shown, this module allows users to upload meteorological observation data and, using the built-in SSA-KNN model, CNN-LSTM model, and mathematical model from previous studies, predict target meteorological variables at a specified altitude. The mathematical model is a calculation for vertical correction of meteorological elements as required by standards and specifications. By visually comparing the two on a unified time axis, users can more intuitively see the model's vertical meteorological element prediction results. Through this module, the platform achieves the ability to predict upper-level data from ground-based observation data, solving the problems of difficult and costly upper-level data acquisition in traditional meteorological observations, expanding the dimensions of data application, and providing more refined and reliable data support for building energy consumption simulation.

[0064] Part Seven: Finally, a unified test of each module of the platform of this invention is conducted to verify whether the expected effect is achieved and to check whether the functions are running normally before deployment and operation.

[0065] Table 5. Performance and Expected Results in Specific Test Scenarios

[0066] The test hardware environment consisted of a server-side CPU (Intel Core i7-12700H), 8GB of RAM, a 1TB hard drive, and 30Mbps of intranet bandwidth. The software environment included a server operating system (Windows 11), a client-side environment (Windows), and client software (Chrome). Specific scenarios, expected performance, and test results are detailed in Table 5.

[0067] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A method for predicting vertical meteorological elements based on a big data framework and Web technology, characterized in that, include: S1. Meteorological Data Management: The system adopts the Hadoop big data processing architecture, combined with a distributed file system and a distributed database, to reliably distribute and store the uploaded meteorological data. S2, Meteorological Data Analysis: Perform graphical visualization analysis on the meteorological data stored in S1; S3. Meteorological data prediction: Import the meteorological data stored in S1 into the SSA-KNN model, CNN-LSTM model and mathematical model for vertical prediction, and output the vertical meteorological prediction results at different altitudes.

2. The vertical meteorological element prediction method based on big data framework and Web technology according to claim 1, characterized in that, The calculation process of the SSA-KNN model in S3 is as follows: A flock of sparrows It can be represented as: ； In the formula, Let represent the number of neighbors of the i-th sparrow. Weighting factors in the time-feature distance function The decay parameter of the neighborhood sample weighting function. For the selected distance metric function type, The selected weighting function type; Using a set of parameters represented by each sparrow A model is constructed, and a regression prediction task is run on the test set. The error of the regression result corresponding to each meteorological element prediction model is recorded. In each iteration, all sparrows fly in a multidimensional parameter space, and their position vectors are represented as follows: The specific update format is as follows: Global Discoverer Update: ； In the formula, This represents the position of the i-th sparrow in the nth iteration; This represents the position of the i-th sparrow in the (n+1)-th iteration. It is the currently found globally optimal position, that is, the parameter that performs best across all iterations. It is the position of another sparrow randomly selected in the current iteration. and It is a weighting factor; Local search update: ； In the formula, This represents the position of a sparrow adjacent to the i-th sparrow. It is a weighting factor used for adjustment. right The influence of this, controlling how close the sparrow gets to the nearest optimal solution, Used to generate random numbers in the interval [0,1]; Introduce the time-feature joint distance function as the default distance evaluation metric for constructing neighborhoods: ； In the formula, This represents the time-feature joint distance function, used to comprehensively measure the spatial similarity and temporal consistency between two samples. This represents the weighted ratio controlling for spatial distance and temporal difference, where t represents the time label of the target sample. x This represents the meteorological input of the target prediction sample. The input features represent the m-th candidate neighbor; A distance-based sample weighting strategy is used to construct the regression prediction function: ； In the formula, For the predicted final temperature of the j-th time, the weights are... From SSA , and Control is calculated based on different functions. For the number of neighbors, As the output of the model, these correspond to the temperature measured at the i-th floor.

3. The vertical meteorological element prediction method based on big data framework and O technology according to claim 1 or 2, characterized in that, The computation process of the CNN-LSTM model in S3 is as follows: Transpose the tensor at each time step t: ； R is a high-dimensional space, and B is the size of each batch of samples. This represents the meteorological input of the target prediction sample at time t; Using convolution to extract spatial features of meteorological elements: ； In the formula, C is the number of convolution kernels. The height dimension is the result of convolution; After each convolution calculation, a bias term b is introduced, and an activation function is used to enhance the model's ability to express nonlinear patterns: ； Flattened into vector form: ； This represents the feature map obtained after the convolutional layer; Stack the spatial feature vectors from all time steps, where T is the temperature, to form the sequence input: ； Through the above operations, the meteorological data at each time step is encoded into a high-dimensional spatial feature vector as the input to the LSTM module; After the LSTM layer, a fully connected layer is added to map the temporal features to the final prediction output; The calculation process is as follows: ； In the formula, W is the weight matrix. is the output of the LSTM layer, b is the bias term, and z is the predicted value of meteorological elements.

4. The vertical meteorological element prediction method based on big data framework and Web technology according to claim 1 or 2, characterized in that, The mathematical model in S3 is a calculation for vertical correction of meteorological elements in the norms and standards. Based on the vertical correction formulas specified in the standards and norms of meteorology, construction and other fields, the model performs height dimension correction calculations for meteorological elements. Its core purpose is to map the meteorological data observed on the ground to different vertical heights through standardized mathematical derivation, so as to obtain the estimated value of meteorological elements at that height.

5. A vertical meteorological element forecasting platform based on big data framework and Web technology, characterized in that, include: Data management module, meteorological data analysis module, and meteorological forecasting module; The data management module adopts the Hadoop big data processing architecture, combined with a distributed file system and a distributed database, to reliably distribute and store the uploaded meteorological data. The meteorological data analysis module enables data preview and graphical visualization analysis; The weather forecasting module imports weather data into the SSA-KNN model, CNN-LSTM model, and mathematical model for vertical forecasting, and outputs vertical weather forecast results at different altitudes.

6. The vertical meteorological element prediction platform based on big data framework and Web technology according to claim 5, characterized in that, The calculation process of the SSA-KNN model in the meteorological forecasting module is as follows: A flock of sparrows It can be represented as: ； In the formula, Let represent the number of neighbors of the i-th sparrow. Weighting factors in the time-feature distance function The decay parameter of the neighborhood sample weighting function. For the selected distance metric function type, The selected weighting function type; Using a set of parameters represented by each sparrow A model is constructed, and a regression prediction task is run on the test set. The error of the regression result corresponding to each meteorological element prediction model is recorded. In each iteration, all sparrows fly in a multidimensional parameter space, and their position vectors are represented as follows: The specific update format is as follows: Global Discoverer Update: ； In the formula, This represents the position of the i-th sparrow in the nth iteration; This represents the position of the i-th sparrow in the (n+1)-th iteration. It is the currently found globally optimal position, that is, the parameter that performs best across all iterations. It is the position of another sparrow randomly selected in the current iteration. and It is a weighting factor; Local search update: ； In the formula, This represents the position of a sparrow adjacent to the i-th sparrow. It is a weighting factor used for adjustment. right The influence of this, controlling how close the sparrow gets to the nearest optimal solution, Used to generate random numbers in the interval [0,1]; Introduce the time-feature joint distance function as the default distance evaluation metric for constructing neighborhoods: ； In the formula, This represents the time-feature joint distance function, used to comprehensively measure the spatial similarity and temporal consistency between two samples. This represents the weighted ratio controlling for spatial distance and temporal difference, where t represents the time label of the target sample. x This represents the meteorological input of the target prediction sample. The input features represent the m-th candidate neighbor; A distance-based sample weighting strategy is used to construct the regression prediction function: ； In the formula, For the predicted final temperature of the j-th time, the weights are... From SSA , and Control is calculated based on different functions. For the number of neighbors, As the output of the model, these correspond to the temperature measured at the i-th floor.

7. The vertical meteorological element prediction platform based on big data framework and Web technology according to claim 5, characterized in that, The computation process of the CNN-LSTM model in the weather forecasting module is as follows: Transpose the tensor at each time step t: ； R is a high-dimensional space, and B is the size of each batch of samples. This represents the meteorological input of the target prediction sample at time t. Using convolution to extract spatial features of meteorological elements: In the formula, C is the number of convolution kernels. The height dimension is the height dimension after convolution. After each convolution calculation, a bias term b is introduced, and an activation function is used to enhance the model's ability to express nonlinear patterns: ； Flattened into vector form: ； This represents the feature map obtained after the convolutional layer (Conv1D). Stack the spatial feature vectors from all time steps, where T is the temperature, to form the sequence input: ； Through the above operations, the meteorological data at each time step is encoded into a high-dimensional spatial feature vector as the input to the LSTM module; After the LSTM layer, a fully connected layer is added to map the temporal features to the final prediction output; The calculation process is as follows: In the formula, W is the weight matrix. is the output of the LSTM layer, b is the bias term, and z is the predicted value of meteorological elements.

8. The vertical meteorological element prediction platform based on big data framework and Web technology according to claim 5, characterized in that, The mathematical model described is a calculation for vertical correction of meteorological elements in specifications and standards. Based on the vertical correction formulas specified in standards and specifications in meteorology, construction and other fields, it performs height dimension correction calculations for meteorological elements. Its core purpose is to map ground-observed meteorological data to different vertical heights through standardized mathematical derivation, and obtain the estimated value of meteorological elements at that height.

9. The vertical meteorological element prediction platform based on big data framework and Web technology according to claim 5, characterized in that, It also includes a visualization module that provides an intuitive display of the prediction results, making it easy for users to analyze and download them.

Images