Data prediction method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention belongs to the technical field of artificial intelligence, and relates to an artificial intelligence-based data prediction method, which comprises the steps of receiving typhoon name information input by a user; collecting related text information of a target typhoon corresponding to the typhoon name information, and performing abstract generation on the related text information based on the target large model to obtain an abstract text; encoding the collected numerical data related to the target typhoon to obtain target numerical data; integrating the abstract text and the target numerical data to obtain multi-modal information; analyzing the multi-modal information based on a typhoon loss prediction model to obtain a prediction result; and outputting a prediction result. The invention further provides a data prediction device based on artificial intelligence, computer equipment and a storage medium. In addition, the prediction result can be stored in the block chain. The method can be applied to a typhoon loss prediction scene in the field of financial science and technology, the processing accuracy of typhoon loss prediction is improved, and the accuracy of the generated prediction result is ensured.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology and can be applied to fields such as fintech and digital healthcare, particularly to data prediction methods, devices, computer equipment, and storage media based on artificial intelligence. Background Technology

[0002] In the insurance sector, the scale of losses caused by natural disasters, especially typhoons, is enormous, placing significant pressure on insurance companies' payouts. Effectively reducing payout risks and accurately predicting typhoon losses have become critical issues that the insurance industry urgently needs to address, with accurate typhoon loss prediction being a particularly crucial aspect. Currently, the mainstream technical approach for typhoon loss prediction in the insurance field primarily employs methods based on traditional machine learning and deep learning, which are also the most widely used technical paths in the industry. This approach typically uses historical typhoon meteorological data (such as maximum wind speed, central pressure, movement speed, and path) and geospatial and socioeconomic data (such as population density, GDP, and infrastructure distribution) as input features, training regression or classification models to predict the amount of loss or the level of risk.

[0003] However, such methods have several drawbacks and shortcomings: First, they suffer from a limited information modality and insufficient data utilization. These models are essentially "numerical models," with their inputs strictly limited to structured numerical tables. In practical applications, a large amount of valuable unstructured information, especially textual information from internet news, social media, government reports, and expert analyses, is completely excluded from the model. This results in the model's understanding of typhoon impacts remaining superficial, failing to construct a comprehensive and in-depth "typhoon profile," and making it difficult to accurately grasp the comprehensive impact of typhoons in various aspects. Second, they lack static analysis capabilities and trend insights. Traditional models learn and predict based on static historical data, essentially a "static correlation" rather than "dynamic deduction." During typhoon development, risk factors are in a dynamic evolutionary state, from "warning issuance" to "disaster landfall" and then to "post-disaster impact," the entire process containing rich trend information. However, traditional models struggle to understand and model these dynamic evolutionary patterns, failing to effectively capture the trend information in the entire chain. Therefore, they are weak in trend prediction and forward-looking analysis, leading to low accuracy in typhoon loss predictions and failing to meet the insurance industry's demand for precise forecasts.

[0004] Therefore, it is necessary to develop a new technology to overcome the shortcomings of existing technologies and improve the accuracy and reliability of typhoon loss prediction. Summary of the Invention

[0005] The purpose of this application is to propose a data prediction method, apparatus, computer device, and storage medium based on artificial intelligence, in order to solve the technical problem that existing methods for predicting typhoon losses based on traditional machine learning and deep learning have low accuracy.

[0006] Firstly, an artificial intelligence-based data prediction method is provided, including: Receive typhoon name information input by the user; Collect relevant text information of the target typhoon corresponding to the typhoon name information, and perform summary generation processing on the relevant text information based on the preset target big model to obtain the corresponding summary text; Collect numerical data related to the target typhoon; The numerical data is encoded based on a preset encoding strategy to obtain the corresponding target numerical data. The summary text and the target numerical data are integrated and processed to obtain the corresponding multimodal information; The multimodal information is analyzed and processed based on a preset typhoon loss prediction model to obtain the corresponding prediction results; The prediction results are then processed for output.

[0007] Secondly, an artificial intelligence-based data prediction device is provided, comprising: The receiving module is used to receive the typhoon name information input by the user; The generation module is used to collect relevant text information of the target typhoon corresponding to the typhoon name information, and perform summary generation processing on the relevant text information based on the preset target big model to obtain the corresponding summary text; The data collection module is used to collect numerical data related to the target typhoon; The encoding module is used to encode the numerical data based on a preset encoding strategy to obtain the corresponding target numerical data. An integration module is used to integrate the summary text and the target numerical data to obtain corresponding multimodal information; The analysis module is used to analyze and process the multimodal information based on a preset typhoon loss prediction model to obtain the corresponding prediction results. The output module is used to process the prediction results.

[0008] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described artificial intelligence-based data prediction method.

[0009] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned data prediction method based on artificial intelligence.

[0010] In the aforementioned scheme implemented by the artificial intelligence-based data prediction method, device, computer equipment, and storage medium, the following steps are taken: First, typhoon name information input by the user is received, and relevant text information of the target typhoon corresponding to the typhoon name information is collected. Then, the relevant text information is processed to generate a summary based on a preset target big model to obtain the corresponding summary text. Next, numerical data related to the target typhoon is collected, and the numerical data is encoded based on a preset encoding strategy to obtain the corresponding target numerical data. Subsequently, the summary text and the target numerical data are integrated to obtain the corresponding multimodal information. Further, the multimodal information is analyzed and processed based on a preset typhoon loss prediction model to obtain the corresponding prediction result. Finally, the prediction result is output. Based on the above processing flow, this application collects relevant textual information about the target typhoon corresponding to the typhoon name information, and generates summary text by using a target large-scale model. It also collects numerical data related to the target typhoon, encodes the numerical data using an encoding strategy to obtain target numerical data, and then integrates the summary text and target numerical data to obtain multimodal information. This multimodal information is then analyzed and processed using a typhoon loss prediction model, and the resulting prediction results are output. Thus, by integrating the summary text corresponding to the target typhoon generated based on the target large-scale model with the target numerical data obtained by encoding the numerical data related to the target typhoon using an encoding strategy to obtain multimodal information, and then analyzing and processing this multimodal information using a typhoon loss prediction model, this application can automatically and accurately complete the loss prediction processing for the target typhoon, improving the accuracy of typhoon loss prediction and ensuring the accuracy of the generated prediction results. Attached Figure Description

[0011] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of an embodiment of the artificial intelligence-based data prediction method according to this application; Figure 3 This is a schematic diagram of a structure of an embodiment of the artificial intelligence-based data prediction device according to this application; Figure 4 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0013] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0014] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0015] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0017] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0018] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0019] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0020] It should be noted that the data prediction method based on artificial intelligence provided in the embodiments of this application is generally executed by a server / terminal device, and correspondingly, the data prediction device based on artificial intelligence is generally set in the server / terminal device.

[0021] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0022] Continue to refer to Figure 2 The flowchart illustrates an embodiment of the AI-based data prediction method according to this application. The order of steps in the flowchart can be changed, and some steps can be omitted, depending on different needs. The AI-based data prediction method provided in this application can be applied to any scenario requiring product recommendation, and thus can be applied to products in these scenarios, such as product recommendations in the financial insurance field. The AI-based data prediction method includes the following steps: Step S201: Receive the typhoon name information input by the user.

[0023] In this embodiment, the artificial intelligence-based data prediction method operates on an electronic device (e.g., Figure 1The server / terminal device shown can acquire the image to be checked via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future-developed wireless connection methods. This application can be applied to typhoon loss prediction scenarios in the insurance sector within the fintech field. The implementing entity of this application is specifically a data prediction system, which can be simply referred to as the system. After the system is started, the user can input the typhoon name information of the target typhoon according to actual needs.

[0024] Step S202: Collect relevant text information of the target typhoon corresponding to the typhoon name information, and perform summary generation processing on the relevant text information based on the preset target big model to obtain the corresponding summary text.

[0025] In this embodiment, the specific implementation process of collecting relevant text information of the target typhoon corresponding to the typhoon name information and performing summary generation processing on the relevant text information based on the preset target big model to obtain the corresponding summary text will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0026] Step S203: Collect numerical data related to the target typhoon.

[0027] In this embodiment, the system can collect various numerical data related to the target typhoon based on the typhoon name information, such as the typhoon's wind speed, air pressure, latitude and longitude coordinates on its movement path, and radius of influence.

[0028] Step S204: Encode the numerical data based on a preset encoding strategy to obtain the corresponding target numerical data.

[0029] In this embodiment, the specific implementation process of encoding the numerical data based on the preset encoding strategy to obtain the corresponding target numerical data will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0030] Step S205: Integrate the summary text and the target numerical data to obtain the corresponding multimodal information.

[0031] In this embodiment, the specific implementation process of integrating the summary text and the target numerical data to obtain the corresponding multimodal information will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0032] Step S206: Analyze and process the multimodal information based on the preset typhoon loss prediction model to obtain the corresponding prediction results.

[0033] In this embodiment, the pre-trained typhoon loss prediction model analyzes and processes the input multimodal information based on its internalized "risk grammar." "Risk grammar" is the model's understanding of the relationship between numerical sequences and textual descriptions learned during pre-training. The typhoon loss prediction model comprehensively considers the typhoon's physical characteristics reflected in the numerical data and the real-time contextual information contained in the summary text description, such as the relationship between the typhoon's intensity, direction of movement, and the geographical, demographic, and economic conditions of the potentially affected areas. Through this comprehensive analysis, the typhoon loss prediction model can uncover hidden patterns and regularities related to economic losses within the data. Furthermore, after complex calculations and analysis within the typhoon loss prediction model, it finally outputs a specific predicted economic loss value, i.e., the prediction result. This value represents the amount of potential economic loss predicted by the typhoon loss prediction model based on the input typhoon-related data; the unit can be tens of thousands, hundreds of millions, etc., depending on the data scale and actual application requirements.

[0034] The specific construction process of the aforementioned typhoon loss prediction model will be described in more detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0035] Step S207: Output the prediction results.

[0036] In this embodiment, the generated target business prediction results can be presented to the corresponding business personnel by means of data transmission (such as email or SMS) or interface display, thereby completing the output processing of the generated prediction results.

[0037] This application first receives typhoon name information input by the user and collects relevant text information about the target typhoon corresponding to the typhoon name information. Then, it performs summary generation processing on the relevant text information based on a preset target big data model to obtain corresponding summary text. Next, it collects numerical data related to the target typhoon and encodes the numerical data based on a preset encoding strategy to obtain corresponding target numerical data. Subsequently, it integrates the summary text and the target numerical data to obtain corresponding multimodal information. Further, it analyzes and processes the multimodal information based on a preset typhoon loss prediction model to obtain corresponding prediction results. Finally, it outputs the prediction results. Based on the above processing flow, this application collects relevant text information about the target typhoon corresponding to the typhoon name information, performs summary generation processing on the relevant text information based on the use of a target big data model to obtain summary text, collects numerical data related to the target typhoon, encodes the numerical data based on the use of an encoding strategy to obtain target numerical data, integrates the summary text and target numerical data to obtain multimodal information, analyzes and processes the multimodal information based on the use of a typhoon loss prediction model, and outputs the obtained prediction results. Thus, this application integrates the summary text corresponding to the target typhoon generated based on the target large model with the target numerical data obtained by encoding the numerical data related to the target typhoon based on the encoding strategy to obtain multimodal information. Then, based on the use of the typhoon loss prediction model, the multimodal information is analyzed and processed, thereby realizing the automatic and accurate completion of the loss prediction processing for the target typhoon, improving the processing accuracy of typhoon loss prediction, and ensuring the accuracy of the generated prediction results.

[0038] In some alternative implementations, step S202 includes the following steps: The typhoon name information is crawled based on a preset query expansion strategy to obtain the corresponding related text information.

[0039] In this embodiment, the specific implementation process of crawling the typhoon name information based on the preset query expansion strategy to obtain the corresponding related text information will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0040] The relevant text information is cleaned based on a preset cleaning strategy to obtain the corresponding target text information.

[0041] In this embodiment, the above-mentioned cleaning process also refers to purification, including: Text vector transformation: Using the Sentence-BERT text embedding model, each crawled text information is converted into a high-dimensional vector. The Sentence-BERT model can capture the semantic information of the text, mapping the text to a high-dimensional space, so that the vector distance of semantically similar texts in this space is relatively close. Similarity calculation and deduplication: The cosine similarity between vectors is used to measure the degree of similarity between texts. The value of cosine similarity is between [-1, 1], and the closer the value is to 1, the more similar the two texts are. The locality-sensitive hashing deduplication algorithm is applied. This algorithm can efficiently find text pairs with similarity exceeding a certain threshold, and remove duplicate and highly similar reports, retaining only representative text information.

[0042] Since data crawled from multiple sources may contain duplicate or highly similar content, the information cleansing step aims to remove this redundant information, ensuring that the data entering the summary generation stage is diverse and of high quality, and avoiding the impact of duplicate information on the accuracy and effectiveness of the summary.

[0043] Based on the preset prompt template, the target large model is used to perform summary generation processing on the target text information to obtain the corresponding summary information.

[0044] In this embodiment, the above-mentioned summary generation process includes: designing a prompt template: designing a structured prompt template rich in domain knowledge. This template not only contains explicit task instructions, such as "Please generate a summary based on the following relevant text information," but also incorporates multiple high-quality summary examples through few-shot learning. These examples can provide a reference pattern for the large model to generate summaries, making the generated summaries more compliant with requirements. Generating summaries using the large model: utilizing the core capabilities of a large language model (target large model), a summary is generated based on the designed prompt template. The target large model can understand the instructions and examples in the template, and, combined with the input text information, generate a structured summary containing key typhoon information (such as typhoon intensity, impact range, and damage caused). The selection of the target large model is not specifically limited; any general-purpose large model can be selected based on actual business needs.

[0045] The summary information is optimized based on a preset optimization strategy to obtain the corresponding target summary information.

[0046] In this embodiment, the specific implementation process of optimizing the summary information based on the preset optimization strategy to obtain the corresponding target summary information will be further described in detail in subsequent specific embodiments of this application, and will not be elaborated on here.

[0047] The target summary information is used as the summary text.

[0048] Based on the above processing flow, this application crawls typhoon name information to obtain text information using a query expansion strategy, then cleans the text information using a cleaning strategy to obtain target text information. Next, based on a preset prompt template, a target large model is used to generate a summary of the target text information, and an optimization strategy is used to optimize the summary information. The generated target summary information is then used as the final summary text. This process transforms text information into a precise, concise, and structured text summary rich in key risk information, providing a concise and important text input for downstream typhoon loss prediction models, enabling these models to better understand typhoon-related information.

[0049] In some optional implementations of this embodiment, the step of crawling the typhoon name information based on a preset query expansion strategy to obtain the corresponding related text information includes the following steps: The typhoon name information is input into a preset web crawler to obtain query data related to the target typhoon from the internet.

[0050] In this embodiment, the input typhoon name information can be entered into a web crawler to obtain multiple sets of query data related to the target typhoon, such as the top 50 sets, from the web.

[0051] Generate a specified number of query terms based on the query data.

[0052] In this embodiment, the obtained query data can be input into a general model to generate a certain number (e.g., 10) of efficient query terms based on the query data. For example, if the typhoon name is input as "Typhoon Mangkhut", possible generated query terms include "Typhoon Mangkhut aliases", "Typhoon Mangkhut occurrence in 2018", "Typhoon Mangkhut international number", etc.

[0053] Based on the query terms, the preset data source is crawled to obtain the corresponding crawled information.

[0054] In this embodiment, web crawling is performed again using the generated query terms. During the crawling process, data is preferentially obtained from preset high-quality and highly authoritative information sources (data sources). These information sources may include official agencies such as the China Meteorological Administration, the Japan Meteorological Agency, and the Joint Typhoon Warning Center, whose data is highly accurate and authoritative; they also include reports from authoritative news media and emergency management departments, which can provide information such as the real-time dynamics and impact range of typhoons. In this way, it is ensured that the collected data (crawled information) is comprehensive and reliable.

[0055] The crawled information is used as the text information.

[0056] This application obtains query data related to the target typhoon from the internet by inputting the typhoon name information into a preset web crawler. Then, it generates a specified number of query terms based on the query data. Next, it performs web crawling processing on a preset data source based on the query terms to obtain corresponding crawled information. This crawled information is then used as the text information. Based on this processing flow, this application can automatically and accurately collect rich information related to the target typhoon from the data source, providing a sufficient data foundation for generating accurate summaries. Furthermore, through multiple rounds of query term optimization, it can more comprehensively cover various types of information about the target typhoon, thereby improving the completeness and accuracy of information acquisition.

[0057] In some optional implementations, optimizing the summary information based on a preset optimization strategy to obtain the corresponding target summary information includes the following steps: The summary information is subjected to accuracy optimization processing to obtain the correct first summary information.

[0058] In this embodiment, the accuracy optimization process includes: 1. Fact Check: Establishing an authoritative information source database: Collecting and organizing authoritative information sources related to typhoons, such as reports issued by official meteorological departments and research results from authoritative scientific research institutions. The model-generated summary is compared one by one with these authoritative information sources to check the accuracy of key facts such as typhoon intensity, movement path, affected areas, and damage caused. Knowledge graph verification: Constructing a knowledge graph in the typhoon field, presenting various attributes of typhoons (such as name, formation time, intensity level, etc.) and their interrelationships (such as the relationship between typhoons and affected areas, and the relationship between typhoons and the resulting disasters) in graph form. Mapping the information in the summary onto the knowledge graph to verify whether it conforms to known knowledge and logic. 2. Data Correction: Identifying erroneous data: If data in the summary is found to be inconsistent with authoritative information during the fact check process, these erroneous data need to be marked. For example, the summary mentions that a certain typhoon has a wind speed of 200 kilometers per hour, but the authoritative meteorological report shows that its wind speed is 180 kilometers per hour. Correct and record: Correct erroneous data based on authoritative information and record the details and basis of the correction. Simultaneously, analyze the reasons why the model generated erroneous data, such as whether it was due to noise in the input data or a bias in the model's understanding of certain data features, in order to improve the model in the future.

[0059] The first summary information is subjected to integrity optimization processing to obtain the corresponding second summary information.

[0060] In this embodiment, the above-mentioned integrity optimization process includes: 1. Key information supplementation: Determining a key information list: Based on the needs of typhoon loss prediction, a key information list is determined, including the typhoon's formation time, location, intensity level, movement path, impact range, types of disasters caused (such as heavy rain, storm surge, strong winds, etc.), casualties, and property damage. Checking and supplementing: Checking whether the summary generated by the model contains all key information against the key information list. If some key information is found to be missing, relevant content needs to be extracted from the original multi-source heterogeneous information and supplemented into the summary. For example, if the summary does not mention the casualties caused by the typhoon, relevant information can be obtained from news reports, government announcements, etc., and added. 2. Contextual relationship supplementation: Analyzing contextual relationships: Typhoon information is often not isolated; it is closely related to surrounding meteorological environment, geographical features, socio-economic conditions, and other factors. Analyzing the contextual relationships between the typhoon information mentioned in the summary and other relevant factors, such as the impact of the topography of the typhoon landfall area on the degree of disaster, and the impact of the local economic structure on property damage, etc. Supplementing Relevant Information: Based on the analysis results, relevant contextual information is added to the abstract to provide a more comprehensive picture of the typhoon event. For example, if the abstract mentions that a typhoon made landfall in a coastal area, information such as the coastline characteristics and population distribution of that area can be added to better understand the potential impact of the typhoon.

[0061] The second summary information is optimized for simplicity to obtain the corresponding third summary information.

[0062] In this embodiment, the above-mentioned conciseness optimization process includes: 1. Redundant information removal: Identifying redundant content: Carefully reading the summary generated by the model to identify redundant content such as repeated expressions, irrelevant details, and redundant modifiers. For example, the summary repeatedly mentions the "powerful force" of the typhoon, or describes some unrelated meteorological phenomena in detail during the typhoon's formation. Deleting and simplifying: Deleting the identified redundant content to simplify the summary. At the same time, ensuring that the removal of redundant information does not affect the key information conveyed and semantic integrity of the summary. For example, changing the repeated expression of "powerful force" to a single concise description, or deleting descriptions of meteorological phenomena unrelated to typhoon loss prediction. 2. Sentence conciseness: Simplifying sentence structure: Checking the sentence structure in the summary, breaking complex sentences into simple and clear short sentences, or merging multiple related short sentences into a more compact sentence. For example, simplifying "During the movement of the typhoon, due to its high wind speed, trees along the way were blown down, and the roofs of many houses were blown off" to "The typhoon moved rapidly, blowing down trees along the way and blowing off the roofs of many houses." Using concise vocabulary: Selecting more concise and accurate vocabulary to express the same meaning. Avoid using overly complex or obscure vocabulary to improve the readability and comprehension of the summary. For example, change "extremely destructive" to "extremely destructive".

[0063] The third summary information is logically optimized to obtain the corresponding fourth summary information.

[0064] In this embodiment, the above-mentioned logical optimization process includes: 1. Information sorting adjustment: Determining the logical order: Based on the development process of the typhoon event and the importance of the information, determine the reasonable logical order of the information in the summary. Generally, it can be arranged in the order of the typhoon's formation, development, landfall, and impact. Adjusting the information position: If the arrangement order of the information in the summary is found to be illogical, adjust the position of the information to make it more in line with people's cognitive habits and the development logic of the event. For example, move the description of property damage caused by the typhoon mentioned earlier in the summary to after the description of the typhoon's landfall and impact range. 2. Logical relationship sorting: Clarifying causal relationships: Check whether the causal relationships between the various pieces of information mentioned in the summary are clear. For example, is there a reasonable causal link between the changes in the typhoon's wind speed and air pressure and the degree of disaster caused? If the causal relationship is unclear, it needs to be further analyzed and clearly stated. Supplementing logical connectors: Appropriately supplement the summary with logical connectors, such as "because," "therefore," "however," "at the same time," etc., to enhance the logic and coherence of the information. For example, when describing two different disaster situations caused by the typhoon, "at the same time" can be used to connect them, making it easier for readers to understand the relationship between the two.

[0065] The fourth summary information is then optimized for readability to obtain the corresponding fifth summary information.

[0066] In this embodiment, the above-mentioned readability optimization processing includes: 1. Language style adjustment: conforming to domain language habits: ensuring the language style of the abstract conforms to the professional language habits of the typhoon field, using accurate terminology and expressions. For example, using professional terms such as "tropical cyclone," "eye," and "pressure gradient" to describe the relevant characteristics of typhoons. Easy to understand: while ensuring professionalism, strive to make the language of the abstract easy to understand, avoiding overly obscure or difficult vocabulary and sentence structures. For some professional terms, appropriate explanations or clarifications can be provided so that non-professionals can also understand them. For example, "pressure gradient" can be simply explained as "the degree of pressure change per unit distance." 2. Format optimization: paragraphing and headings: dividing the abstract into reasonable paragraphs and adding subheadings to each paragraph to make the structure of the abstract clearer and easier for readers to quickly browse and understand. For example, the abstract can be divided into paragraphs such as "Basic Typhoon Information," "Affected Area," and "Caused Disasters," with corresponding subheadings added to each. Punctuation and layout: paying attention to the correct use of punctuation marks to make the tone and semantic expression of sentences more accurate. At the same time, proper layout, such as setting appropriate line spacing and character spacing, can improve the visual readability of the abstract.

[0067] The fifth summary information is used as the target summary information.

[0068] This application optimizes the accuracy of the summary information to obtain a first summary; then optimizes the completeness of the first summary information to obtain a second summary; next, optimizes the conciseness of the second summary information to obtain a third summary; subsequently, optimizes the logic of the third summary information to obtain a fourth summary; further optimizes the readability of the fourth summary information to obtain a fifth summary; finally, the fifth summary information is used as the target summary information. Based on the above processing flow, this application can effectively improve the data quality of the generated target summary information by optimizing the summary information in terms of accuracy, completeness, conciseness, logic, and readability.

[0069] In some alternative implementations, step S204 includes the following steps: The numerical data is normalized to obtain the corresponding first processed data.

[0070] In this embodiment, the normalization process includes: for a time series x1: C+H = {x1, ..., xC+H}, the first C time steps constitute the historical sequence, and the subsequent H time steps constitute the prediction sequence. Min-max normalization is used to map unbounded real values ​​to a bounded range of -0.1 to 0.1.

[0071] The first processed data is discretized to obtain the corresponding second processed data.

[0072] In this embodiment, the discretization process includes: uniformly dividing the interval [-1, 1] into 10,000 bins, taking the center of each bin and retaining 4 decimal places to obtain discrete real values. The normalized real values ​​are then mapped to the corresponding bins, with the center value of each bin being the discrete value of that real value. For example, if a normalized value is 0.2835, it will be mapped to the bin containing that value, with the center value of the bin serving as its discrete value.

[0073] The second processed data is serialized to obtain the corresponding third processed data.

[0074] In this embodiment, the serialization process includes adding the marker character "@@" before and after the discrete value to form a "foreign language word" in the form of @@0.2835@@, and expanding the tokenizer vocabulary (1w + 1 <NaN>). This allows discrete numerical values ​​to be represented in a sequence form that the model can process, facilitating subsequent fusion with text data.

[0075] The third processed data is used as the target numerical data.

[0076] In this embodiment, numerical time-series data and summary text data differ in form and nature. In order for the large model to uniformly understand and process these heterogeneous data, the numerical time-series data needs to be transformed into a "foreign language" sequence that the large model can recognize. This step prepares for the subsequent construction of multimodal "risk grammar" sequences, enabling numerical data to participate in model training and prediction together with text data.

[0077] This application normalizes the numerical data to obtain corresponding first processed data; then discretizes the first processed data to obtain corresponding second processed data; subsequently, it serializes the second processed data to obtain corresponding third processed data; and finally, the third processed data is used as the target numerical data. Based on the above processing flow, this application automatically and accurately completes the encoding processing of numerical data by normalizing, discretizing, and serializing the numerical data, thereby converting heterogeneous numerical time-series data into a foreign language sequence that the model can uniformly understand and process, improving the accuracy and standardization of the generated target numerical data.

[0078] In some optional implementations of this embodiment, step S205 includes the following steps: The summary text is converted using a preset text segmenter to obtain the corresponding text sequence.

[0079] In this embodiment, the selection of the text segmenter is not specifically limited and can be determined according to actual business needs. For example, the text segmenter built into the aforementioned base model can be used. Specifically, the text segmenter can convert the summary text into a standard text token sequence, i.e., the aforementioned text sequence. Specifically, the text segmenter can segment the summary text into meaningful tokens according to certain rules (such as based on spaces, punctuation marks, etc.). These tokens are the basic units for the model to perform semantic understanding and processing.

[0080] Obtain the preset splicing strategy.

[0081] In this embodiment, the above-mentioned concatenation strategy includes concatenating the numerical sequence (i.e., the target numerical data), the text token sequence (i.e., the text sequence), and special instruction tokens to form a complete model input rich in multimodal information. Its structure is as follows: [INST] Please predict the final economic loss based on the following typhoon data and summary: [ / INST][NUM][ / NUM][TXT][ / TXT]. Here, [NUM] and [TXT] are newly added special tokens used to clearly distinguish modal boundaries and guide the model to establish cross-modal associations between numerical "foreign languages" and natural language. By recognizing these special tokens, the model can distinguish between numerical data and text data and understand the relationships between them.

[0082] The text sequence and the target numerical data are concatenated based on the concatenation strategy to obtain the corresponding concatenated data.

[0083] In this embodiment, the splicing process of the above text sequence and target numerical data can be performed based on the strategy content of the splicing strategy, and the resulting spliced ​​data can be used as the required multimodal information.

[0084] The spliced ​​data is used as the multimodal information.

[0085] In this embodiment, the purpose of the fusion step is to integrate numerical time-series data and text summary information to form a complete model input rich in multimodal information. In this way, the model can better understand the multimodal characteristics of typhoon risk, including numerical trends and relevant information in text descriptions, thereby improving the accuracy of predictions.

[0086] This application converts the abstract text using a preset text segmenter to obtain a corresponding text sequence; then, it obtains a preset concatenation strategy; subsequently, it concatenates the text sequence with the target numerical data based on the concatenation strategy to obtain corresponding concatenated data; finally, it uses the concatenated data as the multimodal information. Based on the above processing flow, this application converts the abstract text using a text segmenter to obtain a text sequence, then concatenates the text sequence with the target numerical data using a concatenation strategy, and uses the resulting concatenated data as the final multimodal information. This allows for efficient and accurate integration of the target numerical data and the abstract text, ensuring the completeness and richness of the obtained multimodal information.

[0087] In some optional implementations of this embodiment, before step S206, the electronic device may further perform the following steps: Obtain pre-collected historical typhoon data.

[0088] In this embodiment, the process of collecting historical typhoon data includes: collecting numerical time-series data: collecting massive amounts of historical typhoon numerical time-series data from multiple reliable meteorological data sources, including typhoon wind speed, air pressure, latitude and longitude coordinates along the movement path, and radius of influence. This data needs to be recorded at uniform time intervals for subsequent analysis and processing. Collecting summary text data: collecting textual descriptive information related to historical typhoons, such as meteorological reports, news reports, and academic papers. These textual descriptions should include key information such as typhoon intensity, affected areas, and damage caused. Simultaneously, the collected text is cleaned and organized to remove irrelevant information and noise, ensuring the quality and accuracy of the text.

[0089] The historical typhoon data is processed to construct samples, resulting in corresponding sample data.

[0090] In this embodiment, the above sample construction process includes: Data integration: Integrating the collected numerical time-series data and corresponding summary text data to construct an unsupervised pre-training dataset. During the integration process, it is necessary to ensure that the correspondence between numerical data and text descriptions is accurate. For example, for each typhoon event, its numerical time-series data throughout its entire life cycle is associated with related text descriptions to form a complete data sample. Data annotation: If further optimization of model performance is needed, annotation of some data can be considered. For example, annotation of key information in the text description, such as the typhoon intensity level and the range of affected areas. Annotated data can be used for supervised training to help the model better understand the semantic information in the data.

[0091] Call the preset base model.

[0092] In this embodiment, while the general-purpose large model possesses powerful language processing capabilities, it lacks specific understanding of tasks in the typhoon domain (such as loss prediction). The purpose of continuous pre-training is to enable the general-purpose large model to understand the "foreign language" (numerical sequences) of the typhoon domain and its relationship with textual descriptions, accurately aligning its capabilities to the specific task of "loss prediction," stimulating its potential dynamic trend inference capabilities, and enabling the model to process multimodal information and make accurate predictions. Therefore, the goals and task requirements for model construction include: Defining the goal: Constructing a continuously pre-trained model capable of understanding the relationship between the "foreign language" (numerical sequences) of the typhoon domain and textual descriptions, and accurately aligning it to the "loss prediction" task. Task analysis: Clarifying that the model needs to process multimodal information (numerical time-series data and summary text) to achieve accurate prediction of typhoon economic losses. This requires the model not only to understand the changing patterns of numerical data but also to combine key information from textual descriptions with numerical data for comprehensive analysis and prediction.

[0093] The selection process for the base model included: Model Evaluation: Existing general-purpose large models in the market were evaluated, taking into account factors such as language processing capabilities, parameter size, and scalability. After evaluation, the qwen3-235b model was selected as the base model. This model possesses powerful language understanding and generation capabilities, providing a solid foundation for subsequent continuous pre-training. Model Adaptation: The selected initial base model underwent preliminary adaptation to ensure it could adapt to subsequent training tasks and data processing workflows. This included adjusting the model's input and output interfaces to enable it to receive and process numerical sequences and textual descriptions of typhoon-related data.

[0094] Furthermore, the selected initial base model is used for model architecture design to obtain the required base model. Core structure is retained: Based on the basic architecture of the qwen3-235b model, its core structures such as the transformer layer, embedding layer, and LM Head layer are retained. These structures are the foundation for language processing and generation, possessing powerful feature extraction and representation capabilities. Parameter adjustment: The model parameters are appropriately adjusted to adapt to the needs of the typhoon loss prediction task. For example, parameters such as the number of transformer layers and the dimension of hidden layers can be adjusted to increase the model's capacity and expressive power. Simultaneously, the embedding layer is optimized to better map numerical sequences and text descriptions into a unified semantic space. Multimodal fusion design: A multimodal fusion module is designed to effectively fuse numerical sequences and text descriptions. This module can employ techniques such as attention mechanisms and gating mechanisms to dynamically adjust the weights of different modal information based on the features and importance of the input data, achieving information complementarity and enhancement.

[0095] Based on a preset model training strategy, the sample data is used to train and optimize the base model until a specified model that meets the construction requirements is obtained.

[0096] In this embodiment, the transformer layer, embedding layer, and LMHearn layer of the base model can be trained. During training, the model continuously adjusts its parameters, learns the correlation between numerical sequences and textual descriptions, and how to predict economic losses based on this information. Through extensive data and repeated training, the model gradually acquires the ability to process multimodal information and make accurate predictions.

[0097] Specifically, the above model training strategy includes: Unsupervised pre-training: First, the model is pre-trained using a pre-constructed unsupervised pre-training dataset. During pre-training, the model learns the inherent connections and patterns between numerical sequences and textual descriptions through self-supervised learning. For example, the model can learn the correlation between typhoon wind speed and pressure changes and typhoon intensity in the textual description, as well as the correspondence between the movement path in the numerical sequence and the affected areas in the textual description. Supervised fine-tuning (optional): If data annotation has been performed, supervised fine-tuning can be conducted using the labeled data based on the unsupervised pre-training. Through supervised fine-tuning, the model can further learn how to accurately predict economic losses based on numerical sequences and textual descriptions. During fine-tuning, optimization objectives such as the cross-entropy loss function can be used to continuously adjust the model's parameters, making its prediction results more accurate. Training process monitoring: During model training, the training process needs to be monitored in real time, including metrics such as the model's loss function value, accuracy, and recall. By monitoring these metrics, problems that occur during training, such as overfitting and underfitting, can be identified in a timely manner, and corresponding measures can be taken to adjust them. For example, if overfitting is detected, it can be addressed by increasing training data or adjusting regularization parameters. Model evaluation and optimization: During training, the model should be evaluated periodically using independent test datasets to verify its performance. Based on the evaluation results, the model should be optimized and adjusted. For example, if the model is found to be biased in predicting economic losses from certain types of typhoons, the reasons can be analyzed, and the model architecture or training strategy can be adjusted accordingly to improve the model's generalization ability and prediction accuracy.

[0098] Among them, the training strategy based on the above model can be used to train and optimize the base model using sample data until an optimized specified model that meets the construction requirements is obtained, and this specified model is used as the required typhoon loss prediction model.

[0099] The specified model is used as the typhoon loss prediction model, and the typhoon loss prediction model is deployed.

[0100] In this embodiment, after model training and optimization, the trained typhoon loss prediction model is exported in a deployable format, such as ONNX. This facilitates the deployment and operation of the model on different hardware platforms. The deployment environment includes servers, storage devices, and network devices. It is ensured that the deployment environment meets the model's operational requirements, such as computing power, memory capacity, and network bandwidth. Furthermore, an application interface for the model is developed, allowing other systems or applications to easily call the model for prediction. The application interface can adopt RESTful APIs or similar methods, providing a simple and easy-to-use interface specification for easy integration and use by developers. Subsequently, the deployed typhoon loss prediction model is applied to actual typhoon loss prediction scenarios, and user feedback is collected based on the actual application. Based on the feedback, the model is further optimized and improved, continuously enhancing its performance and usability.

[0101] Based on the above processing flow, this application constructs sample data from pre-collected historical typhoon data, and then uses the sample data to train and optimize the base model based on the model training strategy. The obtained model that meets the construction requirements is then used as the required typhoon loss prediction model. This enables the efficient and accurate construction of a typhoon loss prediction model that can process multimodal information and make accurate predictions, improving the construction efficiency of the typhoon loss prediction model and ensuring the model performance of the generated typhoon loss prediction model.

[0102] In some alternative implementations, the user information obtained is subject to user consent and complies with relevant laws and policies.

[0103] Furthermore, any software tools or components not belonging to our company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0104] Furthermore, this application has the following innovative features: 1. Intelligent Information Collection and Summary Generation for Typhoon Field: This application does not simply involve web crawling, but rather achieves the automatic generation of high-quality, high-density risk-oriented summaries from multi-source heterogeneous information through query expansion, information cleansing based on semantic embedding, and domain knowledge injection based on structured prompts. It systematically solves the core deficiency of traditional numerical models in utilizing unstructured text information, extracting massive amounts of messy web information into high-value, model-understandable "risk briefings" through an automated process, providing crucial textual modal contextual information for accurate prediction.

[0105] 2. The concept of "risk grammar" was proposed, and a complete training paradigm was designed to enable general-purpose large models to grasp the dynamic evolution of typhoon risks. Based on this, a continuous pre-training method was used to internalize the basic "vocabulary and syntax" of the typhoon domain into the large model and precisely align its capabilities to the loss prediction task, thereby stimulating its trend inference ability and achieving a leap from static analysis to dynamic prediction. This fundamentally overcomes the inherent weakness of large models in numerical reasoning, enabling them to analyze dynamic sequences of data such as wind speed and air pressure like understanding text, thus achieving for the first time deep semantic fusion and joint reasoning of numerical signals and natural language within the model.

[0106] 3. This solution successfully transforms large-scale models from "direct prediction tools" into "risk grammar parsers with specialized domain knowledge" for the first time. This solution not only deeply integrates numerical data and textual information, overcoming the drawbacks of traditional models' single information modality, but also endows large-scale models with professional temporal reasoning and dynamic inference capabilities through specialized "risk grammar" training, solving the problem of poor reliability in direct prediction. Ultimately, this dedicated model can understand the evolutionary "narrative" of typhoon risk over time, much like parsing a language, achieving accurate and forward-looking predictions of economic losses and providing insurance companies with unprecedented risk management capabilities.

[0107] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0108] It should be emphasized that, to further ensure the privacy and security of the above prediction results, the prediction results can also be stored in a blockchain node.

[0109] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0110] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0111] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0112] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0113] Further reference Figure 3 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of an artificial intelligence-based data prediction device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0114] like Figure 3 As shown, the artificial intelligence-based data prediction device 300 described in this embodiment includes: a receiving module 301, a generating module 302, a collecting module 303, an encoding module 304, an integrating module 305, an analysis module 306, and an output module 307. Wherein: The receiving module 301 is used to receive the typhoon name information input by the user; The generation module 302 is used to collect relevant text information of the target typhoon corresponding to the typhoon name information, and perform summary generation processing on the relevant text information based on the preset target big model to obtain the corresponding summary text; Collection module 303 is used to collect numerical data related to the target typhoon; The encoding module 304 is used to encode the numerical data based on a preset encoding strategy to obtain the corresponding target numerical data. The integration module 305 is used to integrate the summary text and the target numerical data to obtain the corresponding multimodal information; The analysis module 306 is used to analyze and process the multimodal information based on a preset typhoon loss prediction model to obtain the corresponding prediction results. The output module 307 is used to process the prediction results.

[0115] In some optional implementations of this embodiment, the generation module 302 includes: The crawling submodule is used to crawl the typhoon name information based on a preset query expansion strategy to obtain the corresponding related text information; The cleaning submodule is used to clean the relevant text information based on a preset cleaning strategy to obtain the corresponding target text information. The generation submodule is used to perform summary generation processing on the target text information based on the preset prompt template and the target large model to obtain the corresponding summary information; The optimization submodule is used to optimize the summary information based on a preset optimization strategy to obtain the corresponding target summary information; The first determining submodule is used to use the target summary information as the summary text.

[0116] In some optional implementations of this embodiment, the crawling submodule includes: The acquisition unit is used to input the typhoon name information into a preset web crawler in order to obtain query data related to the target typhoon from the network. The generation unit is used to generate a specified number of query terms based on the query data; The crawling unit is used to perform web crawling processing on a preset data source based on the query terms to obtain the corresponding crawling information; The first determining unit is used to treat the crawled information as the text information.

[0117] In some optional implementations of this embodiment, the optimized submodule includes: The first processing unit is used to perform accuracy optimization processing on the summary information to obtain appropriate first summary information; The second processing unit is used to perform integrity optimization processing on the first summary information to obtain the corresponding second summary information; The third processing unit is used to perform conciseness optimization processing on the second summary information to obtain the corresponding third summary information; The fourth processing unit is used to perform logical optimization processing on the third digest information to obtain the corresponding fourth digest information; The fifth processing unit is used to perform readability optimization processing on the fourth summary information to obtain the corresponding fifth summary information; The second determining unit is used to use the fifth summary information as the target summary information.

[0118] In some optional implementations of this embodiment, the encoding module 304 includes: The first processing submodule is used to normalize the numerical data to obtain the corresponding first processed data. The second processing submodule is used to discretize the first processed data to obtain the corresponding second processed data. The third processing submodule is used to serialize the second processed data to obtain the corresponding third processed data; The second determining submodule is used to use the third processed data as the target numerical data.

[0119] In some optional implementations of this embodiment, the integration module 305 includes: The conversion submodule is used to convert the summary text based on a preset text segmenter to obtain the corresponding text sequence; The `get` submodule is used to retrieve preset splicing strategies; The splicing submodule is used to splice the text sequence and the target numerical data based on the splicing strategy to obtain the corresponding spliced ​​data; The third determining submodule is used to use the spliced ​​data as the multimodal information.

[0120] In some optional implementations of this embodiment, the artificial intelligence-based data prediction device further includes: The acquisition module is used to acquire pre-collected historical typhoon data; The construction module is used to perform sample construction processing on the historical typhoon data to obtain corresponding sample data; The calling module is used to call the preset base model; The training module is used to train and optimize the base model using the sample data based on a preset model training strategy until a specified model that meets the construction requirements is obtained. The deployment module is used to use the specified model as the typhoon loss prediction model and to deploy the typhoon loss prediction model.

[0121] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0122] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected via a system bus. It should be noted that only the computer device 4 with components 41-43 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0123] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0124] The memory 41 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 41 may be an internal storage unit of the computer device 4, such as the hard disk or memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 4. Of course, the memory 41 may also include both the internal storage unit and its external storage device of the computer device 4. In this embodiment, the memory 41 is typically used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions for data prediction methods based on artificial intelligence. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output.

[0125] In some embodiments, the processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to execute computer-readable instructions stored in the memory 41 or to process data, for example, to execute computer-readable instructions for the artificial intelligence-based data prediction method.

[0126] The network interface 43 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 4 and other electronic devices.

[0127] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the artificial intelligence-based data prediction method described above.

[0128] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0129] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A data prediction method based on artificial intelligence, characterized in that, Includes the following steps: Receive typhoon name information input by the user; Collect relevant text information of the target typhoon corresponding to the typhoon name information, and perform summary generation processing on the relevant text information based on the preset target big model to obtain the corresponding summary text; Collect numerical data related to the target typhoon; The numerical data is encoded based on a preset encoding strategy to obtain the corresponding target numerical data. The summary text and the target numerical data are integrated and processed to obtain the corresponding multimodal information; The multimodal information is analyzed and processed based on a preset typhoon loss prediction model to obtain the corresponding prediction results; The prediction results are then processed for output.

2. The data prediction method based on artificial intelligence according to claim 1, characterized in that, The steps of collecting relevant text information of the target typhoon corresponding to the typhoon name information, and performing summary generation processing on the relevant text information based on a preset target big model to obtain the corresponding summary text, specifically include: The typhoon name information is crawled based on a preset query expansion strategy to obtain the corresponding related text information; The relevant text information is cleaned based on a preset cleaning strategy to obtain the corresponding target text information; Based on the preset prompt template, the target large model is used to perform summary generation processing on the target text information to obtain the corresponding summary information; The summary information is optimized based on a preset optimization strategy to obtain the corresponding target summary information; The target summary information is used as the summary text.

3. The data prediction method based on artificial intelligence according to claim 2, characterized in that, The step of crawling information from the typhoon name information based on a preset query expansion strategy to obtain the corresponding relevant text information specifically includes: The typhoon name information is input into a preset web crawler to obtain query data related to the target typhoon from the Internet; Generate a specified number of query terms based on the query data; Based on the query terms, the preset data source is crawled to obtain the corresponding crawled information; The crawled information is used as the text information.

4. The data prediction method based on artificial intelligence according to claim 2, characterized in that, The step of optimizing the summary information based on a preset optimization strategy to obtain the corresponding target summary information specifically includes: The summary information is subjected to accuracy optimization processing to obtain the corresponding first summary information; The first summary information is subjected to integrity optimization processing to obtain the corresponding second summary information; The second summary information is optimized for simplicity to obtain the corresponding third summary information; Logical optimization is performed on the third summary information to obtain the corresponding fourth summary information; The fourth summary information is optimized for readability to obtain the corresponding fifth summary information; The fifth summary information is used as the target summary information.

5. The data prediction method based on artificial intelligence according to claim 1, characterized in that, The step of encoding the numerical data based on a preset encoding strategy to obtain the corresponding target numerical data specifically includes: The numerical data is normalized to obtain the corresponding first processed data; The first processed data is discretized to obtain the corresponding second processed data; The second processed data is serialized to obtain the corresponding third processed data; The third processed data is used as the target numerical data.

6. The data prediction method based on artificial intelligence according to claim 1, characterized in that, The step of integrating the summary text and the target numerical data to obtain the corresponding multimodal information specifically includes: The summary text is converted based on a preset text segmenter to obtain the corresponding text sequence; Obtain the preset splicing strategy; The text sequence and the target numerical data are concatenated based on the concatenation strategy to obtain the corresponding concatenated data; The spliced ​​data is used as the multimodal information.

7. The data prediction method based on artificial intelligence according to claim 1, characterized in that, Before the step of analyzing and processing the multimodal information based on the preset typhoon loss prediction model to obtain the corresponding prediction results, the method further includes: Acquire pre-collected historical typhoon data; The historical typhoon data is processed to construct a sample, resulting in corresponding sample data. Call the preset base model; Based on a preset model training strategy, the sample data is used to train and optimize the large base model until a specified model that meets the construction requirements is obtained. The specified model is used as the typhoon loss prediction model, and the typhoon loss prediction model is deployed.

8. A data prediction device based on artificial intelligence, characterized in that, include: The receiving module is used to receive the typhoon name information input by the user; The generation module is used to collect relevant text information of the target typhoon corresponding to the typhoon name information, and perform summary generation processing on the relevant text information based on the preset target big model to obtain the corresponding summary text; The data collection module is used to collect numerical data related to the target typhoon; The encoding module is used to encode the numerical data based on a preset encoding strategy to obtain the corresponding target numerical data. An integration module is used to integrate the summary text and the target numerical data to obtain corresponding multimodal information; The analysis module is used to analyze and process the multimodal information based on a preset typhoon loss prediction model to obtain the corresponding prediction results. The output module is used to process the prediction results.

9. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the data prediction method based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the data prediction method based on artificial intelligence as described in any one of claims 1 to 7.

Images