Intelligent processing method, device and electronic equipment for meteorological decision based on agent

Abstract

The application provides a meteorological decision-making intelligent processing method and device based on an agent and electronic equipment, and relates to the field of data processing. The method comprises the following steps: collecting original meteorological data from meteorological observation equipment, satellite data and a numerical prediction model; in response to detecting a user instruction, constructing a user demand model according to a user query request corresponding to the user instruction, user preference settings and user historical behaviors; initializing a meteorological warning decision-making agent according to the user demand model, and configuring a perception module, an inference module and a learning module for the meteorological warning decision-making agent; using the inference module to perform logical inference based on spatiotemporal tensor meteorological data and the user demand model through the meteorological warning decision-making agent, generating meteorological decision-making suggestion data, and displaying the meteorological decision-making suggestion data in a graphical user interface. The method can improve the efficiency of meteorological warning decision-making.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a meteorological decision-making intelligent processing method and device based on an agent and an electronic device. BACKGROUND

[0002] At present, with the increasing influence of climate change, meteorological early warning plays a more prominent role in disaster prevention and mitigation for various departments. The existing meteorological early warning service adopts a traditional media product production method, that is, products are produced in advance according to the specified rules and are published at regular intervals. The early warning information adopts a one-way push mode (such as short message service, APP pop-up window, WeChat group push), and the user needs to manually query the associated industry defense measures, which is time-consuming. Meanwhile, the unified early warning content and the unified early warning service product cannot meet the personalized service needs of different industries, resulting in low efficiency of the existing meteorological early warning decision-making. SUMMARY

[0003] The purpose of the present application is to provide a meteorological decision-making intelligent processing method and device based on an agent and an electronic device to solve the technical problem of low efficiency of meteorological early warning decision-making.

[0004] In a first aspect, the present application provides a meteorological decision-making intelligent processing method based on an agent, which comprises:

[0005] Collecting original meteorological data from meteorological observation equipment, satellite data and numerical prediction models;

[0006] In response to detecting a user instruction, constructing a user demand model according to a user query request corresponding to the user instruction, a user preference setting and a user historical behavior; the user demand model is used to represent the demand and expectation target of the user for meteorological information;

[0007] Initializing a meteorological early warning decision-making agent according to the user demand model, and configuring a perception module, an inference module and a learning module for the meteorological early warning decision-making agent; one end of the perception module is connected to meteorological observation equipment, satellites, radars, numerical prediction models and industry Internet of Things sensors, which is used to obtain the original meteorological data in real time; the other end of the perception module is connected to the inference module, which is used to transmit the spatio-temporal tensor meteorological data obtained by analyzing, boundary checking and exception elimination on the original meteorological data to the inference module;

[0008] Using the inference module to perform logical reasoning based on the spatio-temporal tensor meteorological data and the user demand model through the meteorological early warning decision-making agent, generating meteorological decision-making suggestion data, and displaying the meteorological decision-making suggestion data in a graphical user interface;

[0009] In response to a user feedback operation on the meteorological decision suggestion data in the graphical user interface, user feedback information is obtained according to the user feedback operation, including user modification requirements for the meteorological decision suggestion data, satisfaction evaluation data and feedback opinion data.

[0010] The user feedback information is input into the learning module, so that the learning module iterates and optimizes the behavior strategy decision model of the meteorological warning decision intelligent agent based on the user feedback information through a machine learning algorithm, and obtains an optimized meteorological warning decision intelligent agent.

[0011] In one possible implementation, the execution process of the perception module on the original meteorological data includes:

[0012] The multi-source heterogeneous high-frequency meteorological observation stream data is acquired, cleaned and aligned, and the reliability of the multi-source heterogeneous high-frequency meteorological observation stream data is evaluated, to obtain first meteorological data with reliability;

[0013] The first meteorological data is parsed and boundary checked, and the original meteorological data is detected for abnormality by a variational autoencoder. If the reconstruction error of target record data is greater than a specified standard deviation, the target record data is marked as null, to obtain second meteorological data, so as to avoid noise pollution in subsequent calculation through abnormality elimination;

[0014] All the second meteorological data is transmitted to the same space-time grid, so that the time and space dislocation in the second meteorological data is smoothed through interpolation and error correction, to obtain third meteorological data;

[0015] The third meteorological data is real-time weighted by a dynamic Bayesian belief network. The smaller the error variance of the source in the third meteorological data is, the higher the weight corresponding to the source is, and the weight is updated in time sliding manner, to obtain fourth meteorological data;

[0016] Based on the fourth meteorological data, a configured learning compressor is used to compress radar echoes to a specified compression degree, and the compressed data is losslessly restored in the cloud, to save bandwidth and maintain data accuracy.

[0017] In one possible implementation, the perception module includes a user feedback perception module, and the execution process of the user feedback perception module includes:

[0018] The explicit feedback data and the implicit feedback data of the user in the business scene are collected, semantic extraction and sentiment polarity determination are performed, and learnable user feedback signals are obtained; wherein the explicit feedback data is directly uploaded through an evaluation control or a voice intercom; the implicit feedback data is captured through front-end burying, and the front-end burying includes click operation, stay operation, forwarding operation and whether to perform the operation according to the suggestion;

[0019] The text data and the voice data in the learnable user feedback signals are processed through a unified semantic emotion double-tower model; a first tower model in the semantic emotion double-tower model is used to convert the text in the text data into a vector, a second tower model in the semantic emotion double-tower model is used to convert the sound wave in the voice data into a vector, and the first tower model and the second tower model are fused at the top of the semantic emotion double-tower model to output an intent label and an emotion value;

[0020] An abnormal object in the user is identified by using a graph neural network, abnormal feedback data corresponding to the abnormal object is automatically screened out, filtered user feedback signals are obtained, and the filtered user feedback signals are determined as rewards or punishments of the learning module.

[0021] In one possible implementation, the inference module includes a first inference layer, a second inference layer and a third inference layer;

[0022] The first inference layer is configured to receive the spatio-temporal tensor meteorological data transmitted by the perception module, perform a first round of reasoning by using a neural network subjected to meteorological physical constraints, and obtain probability distribution data of precipitation, temperature and wind speed elements in a specified time period in the future;

[0023] The second inference layer stores cross-industry risk thresholds and specified risk rules; the specified risk rules are checked by a symbolic engine in JSON-LD form; the specified risk rules include at least one of the following: a road surface below a specified temperature triggers a warning, a time precipitation amount exceeding a specified precipitation amount triggers a warning, and a wind speed exceeding a specified speed triggers a remote shutdown of a marine wind farm;

[0024] The third inference layer is configured to transmit the output data of the first inference layer and the second inference layer to a causal Bayesian network, perform a Pareto optimal search from a safety dimension, an economic dimension and a social impact dimension by using the causal Bayesian network, and generate meteorological decision suggestion data that meets physical laws and the specified risk rules and takes into account costs and benefits; the meteorological decision suggestion data includes at least one of the following: a meteorological warning, a disaster weather response suggestion and an agricultural production suggestion.

[0025] In one possible implementation, the method further includes:

[0026] The SHAP interpreter is used to convert key factors of the inference module in the inference process into natural language based on the meteorological decision suggestion data, and the key factors include early warning decision reasons and weights corresponding to the early warning decision reasons;

[0027] In response to a user query request for the meteorological decision suggestion data and / or the natural language, counterfactual explanation data is generated based on the meteorological decision suggestion data, and the counterfactual explanation data is taken as a feedback result of the user query request; wherein the counterfactual explanation data includes a hypothetical event that causes a change in the early warning level in the meteorological decision suggestion data.

[0028] In one possible implementation, the specified risk rules correspond to defense rules of multiple scenarios, and the method further includes:

[0029] A cross-industry knowledge graph is dynamically constructed based on the cross-industry risk threshold and the defense rules of the multiple scenarios, and fusion and processing of multi-source meteorological data are automatically updated according to the cross-industry knowledge graph;

[0030] A MySQ table index hitting the cross-industry knowledge graph is determined using the labels corresponding to the multiple scenarios;

[0031] For structured data in the cross-industry knowledge graph, table fields are directly injected into prompts after SQL hits, and for unstructured data in the cross-industry knowledge graph, ColBERT is used for further fine-grained rearrangement after vector recall paragraphs, to generate a search engine of a dual-path hybrid RAG retrieval enhancement mode for the cross-industry knowledge graph, so as to perform adaptive combination output of text / tables / documents based on a RAG-based multi-modal response engine.

[0032] In one possible implementation, the meteorological decision suggestion data is displayed in the graphical user interface, including:

[0033] In response to a scanning operation of an AR client on a target real scene area, a risk area is determined from the target real scene area by performing risk analysis and meteorological decision on the target real scene area by the meteorological early warning decision intelligent agent, and meteorological decision suggestion data corresponding to the risk area is generated;

[0034] The risk area and the meteorological decision suggestion data corresponding to the risk area are displayed in the graphical user interface.

[0035] In a second aspect, the present application provides a meteorological decision intelligent processing device based on an intelligent agent, including:

[0036] The acquisition module is configured to acquire original meteorological data from meteorological observation equipment, satellite data, and a numerical prediction model;

[0037] a constructing module configured to construct a user demand model according to a user query request corresponding to the user instruction, a user preference setting, and a user historical behavior in response to detecting the user instruction, the user demand model being used to represent a demand and an expectation target of the user for meteorological information;

[0038] a configuring module configured to initialize a meteorological warning decision intelligent agent according to the user demand model, and to configure a perception module, an inference module, and a learning module for the meteorological warning decision intelligent agent, one end of the perception module being connected to meteorological observation equipment, satellites, radars, numerical prediction models, and industry Internet of Things sensors, and the other end of the perception module being connected to the inference module, the perception module being used to transmit spatiotemporal tensor meteorological data obtained by analyzing, boundary checking, and exception removing the original meteorological data to the inference module;

[0039] a generating module configured to generate meteorological decision suggestion data by the meteorological warning decision intelligent agent using the inference module to perform logical inference based on the spatiotemporal tensor meteorological data and the user demand model, and to display the meteorological decision suggestion data in a graphical user interface;

[0040] a determining module configured to determine a modification demand of the user for the meteorological decision suggestion data, satisfaction evaluation data, and feedback opinion data according to a user feedback operation for the meteorological decision suggestion data in the graphical user interface in response to the user feedback operation, and to obtain user feedback information;

[0041] an optimizing module configured to input the user feedback information to the learning module, so that the learning module iterates and optimizes a behavior strategy decision model of the meteorological warning decision intelligent agent based on the user feedback information by using a machine learning algorithm, and obtains an optimized meteorological warning decision intelligent agent.

[0042] In a third aspect, the present application further provides an electronic device, comprising a memory and a processor, the memory storing a computer program capable of running on the processor, and the processor implements the method of the first aspect when executing the computer program.

[0043] In a fourth aspect, the present application further provides a computer readable storage medium storing computer executable instructions, the computer executable instructions causing the processor to run the method of the first aspect when being invoked and run by the processor.

[0044] The present application brings the following beneficial effects:

[0045] The meteorological decision-making intelligent processing method and device based on an agent and the electronic equipment provided by the present application can collect original meteorological data from meteorological observation equipment, satellite data and numerical prediction models; in response to detecting a user instruction, a user demand model is constructed according to a user query request corresponding to the user instruction, a user preference setting and a user historical behavior; the user demand model is used to represent the demand and expected target of the user for meteorological information; a meteorological warning decision-making agent is initialized according to the user demand model, and a perception module, an inference module and a learning module are configured for the meteorological warning decision-making agent; one end of the perception module is connected to meteorological observation equipment, satellites, radars, numerical prediction models and industry Internet of Things sensors, and is used to acquire the original meteorological data in real time; the other end of the perception module is connected to the inference module, and is used to transmit spatio-temporal tensor meteorological data obtained by analyzing, boundary checking and exception removing the original meteorological data to the inference module; the inference module is used to perform logical inference based on the spatio-temporal tensor meteorological data and the user demand model, to generate meteorological decision-making suggestion data, and to display the meteorological decision-making suggestion data in a graphical user interface; in response to a user feedback operation on the meteorological decision-making suggestion data in the graphical user interface, user modification demand, satisfaction evaluation data and feedback opinion data for the meteorological decision-making suggestion data are determined according to the user feedback operation, to obtain user feedback information; the user feedback information is input to the learning module, so that the learning module iterates and optimizes a behavior strategy decision-making model of the meteorological warning decision-making agent based on the user feedback information through a machine learning algorithm, to obtain an optimized meteorological warning decision-making agent; in the present scheme, through the initialization and configuration process of the meteorological warning decision-making agent, the perception, inference and learning modules of the agent are reasonably set, the agent can automatically adapt to the changes in user demand and the dynamic changes in meteorological data through the learning module, to generate more accurate decision-making suggestions; in addition, the user demand model provides meteorological decision-making support tailored for each user, to meet the demand of different users in different scenarios; through multi-source data fusion and intelligent processing, meteorological decision-making suggestions can be quickly generated and optimized, to improve the meteorological warning decision-making efficiency; further, the mechanism of optimizing the meteorological warning decision-making agent through user feedback is improved, the agent is optimized through collecting user feedback and using a machine learning algorithm, to continuously improve the quality of decision-making suggestions and user experience, to realize the optimization of the machine learning algorithm of the learning module of the agent and the continuous improvement of the accuracy and adaptability of decision-making suggestions through user feedback, to further improve the meteorological warning decision-making efficiency.

[0046] In order to make the above objectives, characteristics and advantages of the present application more apparent and easy to understand, a preferred embodiment is described below in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the specific embodiments or the prior art of the present application, the drawings needed to be used in the description of the specific embodiments or the prior art will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0048] Figure 1 A flowchart of an intelligent processing method for meteorological decision based on an agent provided in an embodiment of the present application is shown.

[0049] Figure 2 Another flowchart of an intelligent processing method for meteorological decision based on an agent provided in an embodiment of the present application is shown.

[0050] Figure 3 A structural diagram of an intelligent processing device for meteorological decision based on an agent provided in an embodiment of the present application is shown.

[0051] Figure 4 A structural diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0052] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the present application will be described in detail below with reference to the drawings. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort fall within the scope of the present application.

[0053] The terms "comprise" and "have" and any variations thereof mentioned in the embodiments of the present application are intended to cover the inclusions without exclusivity. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include other steps or units not listed or can optionally include other steps or units inherent to such processes, methods, products or devices.

[0054] Currently, in the prior art, meteorological warning data and defense rules of more than 130 industry scenarios are stored in a scattered manner, and knowledge fragmentation leads to the inability to generate a decision chain; the result presentation only supports text and lacks risk visualization capabilities, such as the inability to generate a road network heat map superimposed with precipitation forecasts, leading to rigid interaction, and the inability to provide accurate meteorological decision support according to the specific needs and preferences of users, lacking personalized services; it is difficult to cope with complex and changing weather conditions and user needs, and the adaptability is poor; moreover, meteorological data cannot be deeply mined and intelligently analyzed, resulting in poor accuracy and timeliness of meteorological services and low intelligence. For example, when facing the diverse meteorological information needs of different users, traditional systems often only provide general weather forecasts, and cannot meet the in-depth needs of different fields such as agricultural production, transportation, disaster prevention, and other fields for meteorological information. At the same time, due to the massive growth and increasing complexity of meteorological data, traditional systems also face the problem of low efficiency in data processing and analysis, making it difficult to provide more accurate and valuable meteorological decision recommendations for users in a timely manner.

[0055] Based on this, the embodiments of the present application provide a meteorological decision intelligent processing method and device based on an agent and an electronic device, which can solve the technical problem of low efficiency of existing meteorological warning decision.

[0056] The embodiments of the present application will be further described below with reference to the accompanying drawings.

[0057] Figure 1 A flowchart of a meteorological decision intelligent processing method based on an agent provided by an embodiment of the present application is shown in FIG. 1. As shown in the figure, the method comprises the following steps. Figure 1

[0058] Step S110, collecting original meteorological data from meteorological observation equipment, satellite data, and numerical prediction models.

[0059] As a possible implementation manner, the system can collect meteorological data from meteorological observation equipment, satellite data, numerical prediction models, and other multi-source channels through a data collection and processing module, and perform cleaning, fusion, and standardization processing on the data to ensure the accuracy and consistency of the data.

[0060] Step S120, in response to detecting a user instruction, constructing a user demand model according to a user query request corresponding to the user instruction, user preference settings, and user historical behavior.

[0061] The user demand model is used to represent the user's demand and expectation for meteorological information. For example, the system can construct a user demand model according to the user's input query request, user preference settings, and user historical behavior records through a user demand analysis module, which is used to describe the user's specific demand and expectation for meteorological information. ​

[0062] Step S130, initializing the weather warning decision-making agent according to the user demand model, and configuring the perception module, the reasoning module and the learning module of the weather warning decision-making agent.

[0063] For example, the system can initialize one or more weather decision-making agents and configure their perception modules, reasoning modules and learning modules through the agent management module according to the user demand model. The perception module is used to obtain meteorological data and user feedback in real time. The perception module is the sensory organ of the warning decision-making agent, which undertakes two core tasks: capturing, cleaning, aligning and credible evaluation of multi-source, heterogeneous and high-frequency meteorological observation streams at the second level; collecting, semantic extraction and sentiment polarity determination of explicit and implicit feedback of users in business scenarios at the minute level, so that the agent can see the most real weather world at any time, and hear the most real voice of the user. It turns complex and large data streams into clean, aligned and credible unified expressions, and also turns the intended or unintended reactions of users into learnable signals.

[0064] In one possible implementation, one end of the perception module is connected to meteorological observation equipment, satellites, radars, numerical prediction models and industry Internet of Things sensors for real-time acquisition of raw meteorological data, and the other end of the perception module is connected to the reasoning module for transmission of spatiotemporal tensor meteorological data after analysis, boundary check and exception elimination of the raw meteorological data to the reasoning module.

[0065] For meteorological data, the perception module can be understood as a conveyor belt that never stops. One end of the perception module is connected to automatic weather stations, satellites, radars, numerical models and industry Internet of Things sensors, and the other end of the perception module delivers the sorted spatiotemporal tensor to the reasoning module. Whether the raw meteorological data is an automatic station message every five seconds or a satellite cloud image every ten minutes, it will be analyzed, boundary checked and exception eliminated immediately.

[0066] As a possible implementation, the execution process of the perception module for the original meteorological data includes: acquiring, cleaning and aligning the multi-source heterogeneous high-frequency meteorological observation stream data, and evaluating the reliability of the multi-source heterogeneous high-frequency meteorological observation stream data to obtain first meteorological data with reliability; analyzing and boundary checking the first meteorological data, and performing anomaly detection on the original meteorological data through a variational autoencoder, if the reconstruction error of the target record data is greater than a specified standard deviation, the target record data is marked as null, and second meteorological data is obtained, so as to avoid noise pollution in subsequent calculation through anomaly elimination; transmitting all the second meteorological data to the same space-time grid, so as to smooth the time and space errors in the second meteorological data through interpolation and error correction, and obtain third meteorological data; real-time weight allocation for the third meteorological data is performed by using a dynamic Bayesian belief network; the smaller the error variance in the third meteorological data, the higher the corresponding allocated weight, and the weight is updated in time sliding, and fourth meteorological data is obtained; based on the fourth meteorological data, the radar echo is compressed to a specified compression degree through a configured learning compressor, and the compressed data is losslessly restored in the cloud, so as to save bandwidth and maintain data accuracy.

[0067] For example, anomaly detection relies on a lightweight but sensitive variational autoencoder: once the reconstruction error of a record exceeds three times the standard deviation, it is marked as null, avoiding noise pollution in subsequent calculations. Subsequently, all data are unified to the same space-time grid, and small errors in time and space are smoothed through interpolation and error correction.

[0068] To solve the problem that observations from different sources often contradict each other, the perception module uses a dynamic Bayesian belief network to assign weights in real time: the smaller the error variance of a source, the higher the weight, and the weight is updated in time sliding, ensuring that the system always favors the more reliable party. In order to run smoothly on a narrowband link, the conveyor belt is also equipped with a learning compressor that can compress radar echoes to one twenty-fourth of the original, and then losslessly restore them in the cloud, saving bandwidth and maintaining accuracy.

[0069] In an optional implementation, the perception module includes a user feedback perception module, and an execution process of the user feedback perception module includes: collecting, semantic extraction and sentiment polarity determination of explicit feedback data and implicit feedback data of a user in a business scenario to obtain learnable user feedback signals; the explicit feedback data is directly uploaded through an evaluation control or voice intercom; the implicit feedback data is captured through front-end burying, and the front-end burying includes click operation, stay operation, forwarding operation and whether to perform the operation according to the suggestion; text data and voice data in the learnable user feedback signals are processed through a unified semantic-sentiment double-tower model; a first tower model in the semantic-sentiment double-tower model is used to convert the text in the text data into a vector, a second tower model in the semantic-sentiment double-tower model is used to convert sound waves in the voice data into a vector, the first tower model and the second tower model are fused at the top of the semantic-sentiment double-tower model to output an intent label and a sentiment value; an abnormal object in the user is identified by using a graph neural network, abnormal feedback data corresponding to the abnormal object is automatically screened out to obtain filtered user feedback signals, and the filtered user feedback signals are determined as rewards or punishments of the learning module.

[0070] For the collection of user feedback, for example, every like or complaint on a user's mobile phone, a voice on the command platform, or even just looking at the warning page for a few seconds will be recorded. Explicit feedback is directly uploaded through a rating button or voice intercom; implicit feedback is captured through front-end burying, such as clicking, staying, forwarding, or whether to perform the operation according to the suggestion. Text and voice are first processed through a unified semantic-sentiment double-tower model: one tower converts text into a vector, and the other tower converts sound waves into a vector. The two towers are fused at the top to output an intent label and a sentiment score. In order to prevent fake reviews or invalid noise, the system uses a graph neural network to identify abnormal groups and automatically filter out suspicious feedback. The filtered signals are fed back in real time and become rewards or punishments of the learning module.

[0071] In step S140, the meteorological warning decision intelligent agent generates meteorological decision suggestion data by using the reasoning module to perform logical reasoning based on the spatio-temporal tensor meteorological data and the user demand model, and displays the meteorological decision suggestion data in a graphical user interface.

[0072] As an optional implementation, the reasoning module includes a first reasoning layer, a second reasoning layer, and a third reasoning layer; the first reasoning layer is configured to receive the spatiotemporal tensor meteorological data transmitted by the perception module, perform a first round of reasoning by using a neural network subjected to meteorological physical constraints, and obtain probability distribution data of elements such as precipitation, temperature, and wind speed in a specified time period; the second reasoning layer stores cross-industry risk thresholds and specified risk rules; the specified risk rules are checked by the symbolic engine in JSON-LD format; the specified risk rules include at least one of the following: a highway starts a warning when the road surface is below a specified temperature, a warning is started when the time precipitation exceeds a specified precipitation, and a marine wind farm remotely stops when the wind speed exceeds a specified speed; the third reasoning layer is configured to transmit the output data of the first reasoning layer and the second reasoning layer to a causal Bayesian network, perform a Pareto optimal search from the safety dimension, the economic dimension, and the social impact dimension by using the causal Bayesian network, and generate meteorological decision suggestion data that meets the physical law and the specified risk rules and takes into account the cost and the benefit; the meteorological decision suggestion data includes at least one of the following: a meteorological warning, a disaster weather response suggestion, and an agricultural production suggestion.

[0073] In the embodiments of the present application, the reasoning module performs logical reasoning based on the meteorological model and the rule base to generate preliminary decision suggestions. For example, the first reasoning layer receives the unified spatiotemporal tensor sent by the perception module, performs a first round of reasoning by using a neural network subjected to meteorological physical constraints, and obtains the probability distribution of elements such as precipitation, temperature, and wind speed in the next few hours; the second reasoning layer is configured to execute rules, which store cross-industry risk thresholds and administrative regulations, for example, a yellow warning must be started when the road surface temperature of a highway drops below zero or the precipitation exceeds three millimeters, or a marine wind farm must be remotely stopped when the ten-meter wind exceeds twenty-five meters per second, and these rules are checked by the symbolic engine in a readable JSON-LD format; the third reasoning layer is configured to weigh, which gives the output of the previous two layers to a causal Bayesian network, performs a Pareto optimal search in the three dimensions of safety, economy, and social impact, and finally generates a decision draft that meets the physical law and the industry regulations and takes into account the cost and the benefit.

[0074] In the embodiments of the present application, the rule base of the agent reasoning module covers rich meteorological models and decision rules, which can support diversified meteorological decision scenarios and make the warning decision more comprehensive and accurate.

[0075] As an optional implementation, the method may further include the following steps: using a SHAP interpreter to convert key factors in the reasoning process of the inference module into natural language based on meteorological decision-making recommendation data; the key factors include the reasons for the warning decision and the corresponding weights of the reasons for the warning decision; in response to user question requests regarding the meteorological decision-making recommendation data and / or natural language, generating counterfactual explanation data based on the meteorological decision-making recommendation data, and using the counterfactual explanation data as feedback results for the user question requests; wherein, the counterfactual explanation data includes hypothetical events that cause a change in the warning level in the meteorological decision-making recommendation data.

[0076] To enable users to understand the rationale behind decision conclusions, the system can invoke the SHAP interpreter to translate key factors into natural language. For example, "78% of the weight of this orange alert comes from the sharp increase in radar echo intensity over the past thirty minutes, 15% from the ground temperature dropping below zero, and 7% from historical accident rate statistics for the same period." If users question the conclusions, the system can also provide immediate counterfactual explanations, such as, "If the temperature drops by only 0.5 degrees, the alert level will be downgraded to yellow."

[0077] For the question-and-answer flow logic of natural language, such as Figure 2 As shown, the system assigns labels based on a user's command, for example, "Give me a color patch map of precipitation in Pudong for the next 3 hours." The system immediately generates three labels: color patch map; location: Pudong; timeframe: next 3 hours. Real-time data is retrieved instantly, and these three labels act as "keys" to query the database: the structured database uses "location + timeframe" to instantly extract gridded precipitation data; if data is missing, it then searches the vector database for similar scenario data. Templates are used for each type of map, with pre-defined templates: color patch map, wind vane, line graph, and bar chart. The newly obtained data is then input into the rendering module according to the template fields.

[0078] In one optional implementation, the specified risk rule corresponds to defense rules for multiple scenarios. The method may further include the following steps: dynamically constructing a cross-industry knowledge graph based on cross-industry risk thresholds and defense rules for multiple scenarios, and automatically updating the fusion and processing of multi-source meteorological data according to the cross-industry knowledge graph; using tags corresponding to multiple scenarios to determine the MySQL table index that hits the cross-industry knowledge graph; for structured data in the cross-industry knowledge graph, after SQL hit, directly injecting the table fields into the prompt; for unstructured data in the cross-industry knowledge graph, after vector recall of paragraphs, using ColBERT to further refine the fine-grained rearrangement, generating a search engine with a dual-path hybrid RAG retrieval enhancement mode for the cross-industry knowledge graph, and performing adaptive combination output of text / charts / documents with a RAG-based multimodal response engine.

[0079] For the construction of cross-industry knowledge base, structured storage of multi-class knowledge base, such as early warning criteria, risk threshold, early warning criteria, industry standard, risk threshold, defense guidelines, case library, regulation library, expert experience and FAQ. Cover multiple scene adaptive output process, for business system API, provide RESTful interface to connect emergency command platform, such as Figure 2 As shown, the generation of defense measures based on knowledge base text. Through three-level lightweight deployment architecture, support APP floating ball to small program to business system API seamless embedded.

[0080] As shown in Figure 2 For keyword routing, L2 scenario tags are used to directly hit MySQL table index (<5ms). For semantic backup, Sentence-BERT 512-dimensional vector recalls Top-5, inner product retrieval <10 ms. For Prompt assembly, template = 「role + scene + defense measures + output format」; Where, the role includes meteorological bureau duty forecaster; The output format is Markdown table combined with risk color bar.

[0081] In order to realize the improvement of decision efficiency, the industry defense scheme generation time is shortened from 30 minutes to 5 seconds, and the knowledge retrieval accuracy is ≥95% (compared with traditional system 70%). For meteorological customized RAG retrieval enhancement generation, retrieval adopts double-way mixed mode, that is, for structured data, SQL hits 1-N table fields directly into prompt, for unstructured data, vector recalls paragraph, and then uses ColBERT for fine-grained rearrangement (nDCG +6.7%).

[0082] On the generation side, for the base model, the model after continuing pre-training on 50G meteorological corpus; For dynamic Few-shot, retrieve multiple historical similar cases from case library as examples to join prompt. For numerical verification, call internal verification API to compare the "precipitation ≥50 mm" in the generated text with the grid real situation, and if it is inconsistent, trigger rewriting. For output control, the length is max_new_tokens = 256, temperature = 0.3; For safety, filter multiple sensitive word libraries combined with regular, delay <1.2 s. Through the multi-modal response engine of RAG, the adaptive combination output of text / table / document is realized (support DOC / HTML export).

[0083] For example, the defense measures in the knowledge base can be understood as an emergency operation manual that can be consulted at any time, which prewrites who should do what according to the "warning level- region-time" three indexes for various weather scenarios such as heavy rain, strong convection, drought, fire danger, etc. For example: Orange rain: school suspension, construction site suspension, underground shopping mall power failure and drainage; High temperature red: sanitation workers suspend outdoor work in the afternoon, community opens cooling points; Forest fire danger level 5: forest area closed, patrol team increased patrol, village gate set up card to collect fire sources. Each measure is equipped with threshold, responsible unit and contact information. After the system receives the user's question, it can extract the corresponding item like a dictionary, and then generate an action instruction that can be done according to the weather situation.

[0084] Through the dynamic construction method of cross-industry risk threshold knowledge graph, the quantitative storage and automatic update technology of multiple scene defense rules, and the fusion and processing technology of multi-source meteorological data, the accuracy and consistency of the data are ensured.

[0085] In some embodiments, the above-mentioned display of meteorological decision suggestion data in the graphical user interface can specifically include the following steps: in response to the scanning operation of the AR client on the target real scene area, the risk area is determined from the target real scene area by the meteorological warning decision intelligent agent through risk analysis and meteorological decision of the target real scene area, and the meteorological decision suggestion data corresponding to the risk area is generated; the risk area and the meteorological decision suggestion data corresponding to the risk area are displayed in the graphical user interface.

[0086] For rendering and output, templates can be input into the Echarts engine, and base64 pictures or html fragments can be generated in 800 milliseconds to be directly returned to the user interface or pushed to the large screen. From the default chart to the extended AR real scene superposition, for example, the risk area of the farmland is displayed by scanning the mobile phone, thereby improving the user experience.

[0087] Step S150, in response to the user feedback operation on the meteorological decision suggestion data in the graphical user interface, the user's modification demand, satisfaction evaluation data and feedback opinion data for the meteorological decision suggestion data are determined according to the user feedback operation, and the user feedback information is obtained.

[0088] For example, the meteorological decision intelligent agent generates meteorological decision suggestions through the reasoning module according to the perceived meteorological data and user demand model, and optimizes them through the learning module. The decision suggestions include but are not limited to meteorological warning, disaster weather response suggestions, agricultural production suggestions, etc.

[0089] For user interaction, the decision suggestions are presented to the user in a visual way, and the user's feedback information is collected. The user can modify the demand, evaluate the satisfaction of the decision suggestions or provide other feedback opinions through the interactive interface.

[0090] Step S160, input the user feedback information to the learning module, so that the learning module iterates and optimizes the behavior strategy decision model of the meteorological warning decision intelligent agent based on the user feedback information through the machine learning algorithm, and obtains the optimized meteorological warning decision intelligent agent.

[0091] In this step, the learning module optimizes the decision model through the machine learning algorithm to improve the decision accuracy. For example, the user feedback information is input to the learning module of the intelligent agent, and the behavior strategy of the intelligent agent is optimized through reinforcement learning or other machine learning algorithms to improve the accuracy and adaptability of subsequent decisions. This module is the cognitive center of the agent (Agent), with "online-offline hybrid learning" as the core, building a closed-loop pipeline from raw feedback to policy iteration, so that the Agent can continuously improve the decision accuracy, environmental adaptability and long-term benefits. The module adopts a "three-layer six-domain" architecture, which ensures that the algorithm is pluggable and interpretable, and also takes into account engineering landing and compliance management.

[0092] In the embodiments of the present application, through the initialization and configuration process of the meteorological warning decision intelligent agent, the perception, reasoning and learning modules of the intelligent agent are reasonably set, the intelligent agent can automatically adapt to the changes of user demand and the dynamic changes of meteorological data through the learning module, generate more accurate decision suggestions, and further provide customized meteorological decision support for each user according to the user demand model, meet the needs of different users in different scenarios, quickly generate and optimize meteorological decision suggestions through multi-source data fusion and intelligent processing, improve service efficiency, and further improve the mechanism of optimizing the meteorological warning decision intelligent agent through user feedback. Through the collection of user feedback and the optimization of the intelligent agent by using the machine learning algorithm, the quality of the decision suggestion and the user experience are continuously improved, the accuracy and adaptability of the decision suggestion are realized through the machine learning algorithm optimization of the intelligent agent learning module and the continuous improvement of the decision suggestion, and the meteorological warning decision efficiency is further improved.

[0093] By introducing the meteorological early warning decision-making agent, the intelligence and individualization of meteorological decision-making service are realized. Compared with the prior art, the present application has the following significant advantages: first, the automatic adaptation capability is strong, the agent can automatically adapt to the changes of user demand and the dynamic changes of meteorological data through the learning module, and generate more accurate decision-making suggestions; second, the individualized service is prominent, and the user demand model is used to provide the meteorological decision-making support for each user, so as to meet the needs of different users in different scenarios; third, the data processing is efficient, through multi-source data fusion and intelligent processing, the meteorological decision-making suggestions can be quickly generated and optimized, and the service efficiency is improved; fourth, the user feedback optimization mechanism is perfect, the agent is optimized by collecting user feedback and using machine learning algorithm, and the quality of decision-making suggestions and user experience are continuously improved. The meteorological decision-making intelligent service method based on the agent technology provided in the embodiment of the present application has the core innovation of introducing the agent technology, and constructs a meteorological service system which can independently perceive the environment, intelligently reason and decide and continuously learn and optimize.

[0094] In practical applications, for the data access layer (Feedback Ingestion Layer), multi-source real-time stream: user explicit rating, implicit behavior log, system indicator, external knowledge graph; semantic aligner: LLM-based semantic parser unifies heterogeneous feedback into "(state, action, reward, next state, meta information)" five-tuple; privacy fence: local differential privacy + federal sampling, ensures compliance, minimum availability principle.

[0095] For the memory and representation layer (Memory&Representation), short-term memory: sliding window experience pool Replay Buffer (supporting priority sampling); long-term memory: vector database + graph database hybrid, supporting fast similarity retrieval and relationship reasoning; state encoder: from discrete / continuous mixed observation to Transformer encoding to unified embedding space (d=512). Metadata (time, user portrait, scene label) as auxiliary vector splicing, alleviate non-stationarity.

[0096] For the engine layer of the learning module mentioned above (Learning Core), the reinforcement learning sub-engine, the main algorithm: PPO + GAE + dynamic entropy regularization, takes into account sample efficiency and policy stability; safety policy optimization: add "cost constraint" Lagrange multiplier to the objective function, realize risk-sensitive reinforcement learning; hierarchical RL: high-level policy does "option selection", low-level policy does fine control, solves the problem of long-time sparse reward. For the supervised / self-supervised sub-engine, offline distillation: train the BC (Behavior Cloning) network with large-scale human demonstration data as policy initialization; inverse reinforcement learning: recover the reward function from expert trajectories to alleviate the reward shaping problem; self-supervised pre-training: use contrastive learning (SimCLR-style) to learn a general world model from user behavior sequences.

[0097] For the online adaptive sub-engine, meta-learning MAML: when the environment distribution drifts, only a small number of new samples are needed to quickly adjust the policy parameters; non-stationary detection: KS statistics + Drift Detector; after triggering, automatically expand the exploration rate and reset the replay buffer.

[0098] For the meteorological warning decision and execution layer (Policy Head&Executor), multi-strategy integration: the main strategy πθ outputs actions through ε-greedy or Thompson Sampling. The safety strategy πsafe is based on the shielding mechanism, which shields high-risk actions in real time. The exploration strategy πexplore uses random network distillation (RND) to maintain continuous novelty. Explainable output: SHAP / LIME interpreter generates "confidence + causal chain + counterfactual" triplets for each action, facilitating personnel supervision. For the evaluation and alignment layer (Evaluation&Alignment), offline indicators are AUC-PR, NDCG, CVR, and long-term value (LTV) simulator; online indicators are A / B experiments + multi-armed Bandit dynamic traffic allocation, reducing experimental costs. For human-machine collaborative alignment, personnel preference modeling (RLHF) fine-tunes the reward model; multi-team confrontation test, find strategy loopholes and ethical risks. For the operation and compliance domain (MLOps&Governance), Feature / Model Store: versioning, bloodline tracking, one-key rollback; automatic retraining pipeline: Airflow + Kubeflow, realize "data drift threshold triggering to model retraining to gray release". For compliance audit: record the full-link logs from "data to feature to model to decision" to meet the GDPR auditable rules.

[0099] Figure 3A structural diagram of an agent-based meteorological decision-making intelligent processing device is provided. As shown in Figure 3 The agent-based meteorological decision-making intelligent processing device 300 includes:

[0100] A collection module 301 is configured to collect original meteorological data from meteorological observation equipment, satellite data, and a numerical prediction model;

[0101] A construction module 302 is configured to construct a user demand model according to a user query request corresponding to a user instruction, a user preference setting, and a user historical behavior in response to detection of the user instruction; the user demand model is used to represent a demand and an expected target of the user for meteorological information;

[0102] A configuration module 303 is configured to initialize a meteorological warning decision-making agent according to the user demand model, and configure a perception module, an inference module, and a learning module for the meteorological warning decision-making agent; one end of the perception module is connected to meteorological observation equipment, satellites, radars, numerical prediction models, and industry Internet of Things sensors, and is configured to acquire original meteorological data in real time; the other end of the perception module is connected to the inference module, and is configured to transmit spatiotemporal tensor meteorological data obtained by analyzing, boundary checking, and exception removing the original meteorological data to the inference module;

[0103] A generation module 304 is configured to generate meteorological decision-making suggestion data by the meteorological warning decision-making agent using the inference module based on the spatiotemporal tensor meteorological data and the user demand model, and display the meteorological decision-making suggestion data in a graphical user interface;

[0104] A determination module 305 is configured to determine a modification demand, a satisfaction evaluation data, and a feedback opinion data of a user for the meteorological decision-making suggestion data according to a user feedback operation for the meteorological decision-making suggestion data in the graphical user interface, and obtain user feedback information;

[0105] An optimization module 306 is configured to input the user feedback information to the learning module, so that the learning module iterates and optimizes a behavior strategy decision-making model of the meteorological warning decision-making agent based on the user feedback information through a machine learning algorithm, and obtains an optimized meteorological warning decision-making agent.

[0106] The agent-based meteorological decision-making intelligent processing device provided by the embodiments of the present application has the same technical features as the agent-based meteorological decision-making intelligent processing method provided by the above embodiments, and can solve the same technical problems and achieve the same technical effects.

[0107] The electronic device provided by the embodiments of the present application is as follows Figure 4As shown, the electronic device 400 includes a processor 402, a memory 401, and the memory stores a computer program executable on the processor, and the processor executes the computer program to implement the steps of the method provided by the above embodiments.

[0108] Referring to Figure 4 , the electronic device further includes a bus 403 and a communication interface 404, and the processor 402, the communication interface 404 and the memory 401 are connected through the bus 403; the processor 402 is used to execute the executable modules stored in the memory 401, such as computer programs.

[0109] The memory 401 can include a high-speed random access memory (RAM) and can also include a non-volatile memory such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 404 (which can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used.

[0110] The bus 403 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 only one bidirectional arrow is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0111] The memory 401 is used to store programs, and the processor 402 executes the programs after receiving execution instructions. The method executed by the device defined by the process disclosed in any embodiment of the present application can be applied to the processor 402 or implemented by the processor 402.

[0112] The processor 402 can be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above method can be completed by the integrated logic circuit of hardware in the processor 402 or the instruction in the form of software. The processor 402 described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. Each method, step and logic block diagram disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as a hardware code processor for execution, or a combination of hardware and software modules in the code processor for execution. The software module can be located in a random access memory, a flash memory, a read only memory, a programmable read only memory or an electrically erasable programmable memory, a register, etc. The storage medium in the art. The storage medium is located in the memory 401, and the processor 402 reads the information in the memory 401, and combines the hardware to complete the steps of the above method.

[0113] Corresponding to the above-mentioned meteorological decision-making intelligent processing method based on agent, the embodiment of the present application further provides a computer readable storage medium, the computer readable storage medium stores computer executable instructions, when the processor calls and runs the computer executable instructions, the computer executable instructions make the processor run the steps of the above-mentioned meteorological decision-making intelligent processing method based on agent.

[0114] The meteorological decision-making intelligent processing device based on agent provided by the embodiment of the present application can be specific hardware on the device or software or firmware installed on the device. The device provided by the embodiment of the present application has the same implementation principle and technical effect as the foregoing method embodiments. For the sake of brevity, the part of the device embodiment not mentioned in the foregoing method embodiments can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can be referred to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0115] In the embodiments of the present application, it should be understood that the disclosed apparatus and method can be implemented in other manners. The embodiments described above are merely specific implementation manners of the present application, and for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation; for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, or the among different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electric, mechanical or other forms.

[0116] For another example, the flowcharts and block diagrams in the drawings show the possible implementation architectures, functions and operations of the apparatus, method and computer program product according to the embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a program segment or a part of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different orders from that shown in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and sometimes they can be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and the combination of blocks in the block diagrams and / or flowcharts, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0117] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., they can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the present embodiment.

[0118] In addition, each functional unit in the embodiments of the present application can be integrated into one processing unit, or each unit can exist physically as a separate unit, or two or more units can be integrated into one unit.

[0119] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the intelligent meteorological decision-making intelligent processing method based on an agent described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0120] It should be noted that: similar reference numbers and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings, in addition, the terms "first", "second", "third" and the like are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0121] Finally, it should be noted that: the above-described embodiments are only specific embodiments of the present application, used to illustrate the technical solutions of the present application, and not to limit them, the protection scope of the present application is not limited thereto, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand: any person skilled in the art within the technical scope disclosed by the present application, they can still modify or easily think of changes to the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to part of the technical features; and these modifications, changes or replacements do not make the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application. All should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An agent-based meteorological decision-making intelligent processing method, characterized in that, The method comprises: Collecting original meteorological data from meteorological observation equipment, satellite data and numerical prediction models; In response to detecting a user instruction, constructing a user demand model according to a user query request corresponding to the user instruction, user preference settings and user historical behavior; the user demand model is used to represent the user's demand and expectation target for meteorological information; Initializing a meteorological warning decision intelligent agent according to the user demand model, and configuring a perception module, an inference module and a learning module for the meteorological warning decision intelligent agent; one end of the perception module is connected to meteorological observation equipment, satellites, radars, numerical prediction models and industry Internet of Things sensors, for real-time acquisition of the original meteorological data, and the other end of the perception module is connected to the inference module, for transmitting the spatio-temporal tensor meteorological data after analyzing, boundary checking and exception removing of the original meteorological data to the inference module; The meteorological warning decision intelligent agent generates meteorological decision suggestion data by using the inference module based on the spatio-temporal tensor meteorological data and the user demand model, and displays the meteorological decision suggestion data in a graphical user interface; In response to a user feedback operation on the meteorological decision suggestion data in the graphical user interface, determining the user's modification demand, satisfaction evaluation data and feedback opinion data for the meteorological decision suggestion data according to the user feedback operation, and obtaining user feedback information; Inputting the user feedback information into the learning module, so that the learning module iterates and optimizes the behavior strategy decision model of the meteorological warning decision intelligent agent based on the user feedback information through a machine learning algorithm, and obtains an optimized meteorological warning decision intelligent agent; The perception module includes a user feedback perception module, and the execution process of the user feedback perception module comprises: collecting, semantic extraction and sentiment polarity determination of explicit feedback data and implicit feedback data of users in a business scenario to obtain learnable user feedback signals; wherein the explicit feedback data is directly uploaded through an evaluation control or a voice intercom; the implicit feedback data is captured through a front-end burying point, and the front-end burying point includes a click operation, a stay operation, a forwarding operation and an operation of whether to execute according to the suggestion; the text data and voice data in the learnable user feedback signals are processed through a unified semantic emotion double-tower model; a first tower model in the semantic emotion double-tower model is used to convert the text in the text data into a vector, a second tower model in the semantic emotion double-tower model is used to convert the sound wave in the voice data into a vector, the first tower model and the second tower model are fused at the top of the semantic emotion double-tower model, and an intent label and an emotion value are output; an abnormal object in the user is identified by using a graph neural network, abnormal feedback data corresponding to the abnormal object is automatically screened out, filtered user feedback signals are obtained, and the filtered user feedback signals are determined as rewards or punishments for the learning module.

2. The method of claim 1, wherein, The execution process of the perception module on the original meteorological data comprises: The multi-source heterogeneous high-frequency meteorological observation stream data is acquired, cleaned and aligned, and the reliability of the multi-source heterogeneous high-frequency meteorological observation stream data is evaluated to obtain first meteorological data with reliability; The first meteorological data is parsed and boundary checked, and the original meteorological data is detected for abnormalities by a variational autoencoder, and if the reconstruction error of the target record data is greater than a specified standard deviation, the target record data is marked as null, to obtain second meteorological data, so as to avoid noise pollution in subsequent calculation through abnormality elimination; All the second meteorological data is transmitted to the same space-time grid, so that the time and space errors in the second meteorological data are smoothed through interpolation and error correction to obtain third meteorological data; The third meteorological data is real-time weighted by a dynamic Bayesian belief network; the smaller the error variance of the third meteorological data is, the higher the weight corresponding to the third meteorological data is, and the weight is updated in time sliding manner to obtain fourth meteorological data; Based on the fourth meteorological data, a configured learning compressor is used to compress radar echoes to a specified compression degree, and the compressed data is losslessly restored in the cloud to save bandwidth and maintain data accuracy.

3. The method of claim 1, wherein, The inference module includes a first inference layer, a second inference layer and a third inference layer; The first inference layer is used to receive the spatio-temporal tensor meteorological data transmitted by the perception module, and a neural network subjected to meteorological physical constraints is used for first round inference to obtain probability distribution data of precipitation, temperature and wind speed elements in a specified time period; The second inference layer stores cross-industry risk thresholds and specified risk rules; the specified risk rules are checked by a symbolic engine in JSON-LD form; the specified risk rules include at least one of starting a warning when a road surface is below a specified temperature, starting a warning when a time precipitation exceeds a specified precipitation, and remotely shutting down a sea wind farm when a wind speed exceeds a specified speed; The third inference layer is used to transmit the output data of the first inference layer and the second inference layer to a causal Bayesian network, and the causal Bayesian network is used for Pareto optimal search from a safety dimension, an economic dimension and a social impact dimension to generate meteorological decision recommendation data conforming to physical laws and the specified risk rules and taking into account costs and benefits; the meteorological decision recommendation data include at least one of a meteorological warning, a disaster weather response suggestion and an agricultural production suggestion.

4. The method of claim 3, wherein, Further comprising: Based on the meteorological decision recommendation data, a SHAP interpreter is used to convert key factors in the inference process of the inference module into natural language; The key factors include warning decision reasons and weights corresponding to the warning decision reasons; In response to a user query request for the meteorological decision recommendation data and / or the natural language, counterfactual explanation data is generated based on the meteorological decision recommendation data, and the counterfactual explanation data is taken as a feedback result of the user query request; wherein the counterfactual explanation data includes a hypothetical event that causes a change in the warning level in the meteorological decision recommendation data.

5. The method of claim 3, wherein, The specified risk rule corresponds to the defense rule of multiple scenarios, and further comprises: Based on the cross-industry risk threshold and the defense rule of multiple scenarios, a cross-industry knowledge graph is dynamically constructed, and the fusion and processing of multi-source meteorological data are automatically updated according to the cross-industry knowledge graph; Using the label corresponding to the multiple scenarios to determine the MySQ table index hitting the cross-industry knowledge graph; For the structured data in the cross-industry knowledge graph, the table field is directly injected into prompt after SQL hit, and for the unstructured data in the cross-industry knowledge graph, ColBERT is used for further fine-grained rearrangement after vector recalling paragraphs, a search engine of double-path mixed RAG retrieval enhancement mode is generated for the cross-industry knowledge graph, and a RAG-based multi-modal response engine is used to perform adaptive combination output of text / table / document.

6. The method of claim 1, wherein, The meteorological decision suggestion data displayed in the graphical user interface comprises: In response to the scanning operation of the AR client for the target real scene area, the risk area is determined from the target real scene area by the meteorological early warning decision intelligent agent through risk analysis and meteorological decision of the target real scene area, and the meteorological decision suggestion data corresponding to the risk area is generated; The risk area and the meteorological decision suggestion data corresponding to the risk area are displayed in the graphical user interface.

7. An agent-based meteorological decision intelligence processing apparatus, characterized by, Comprise: The acquisition module is used for acquiring original meteorological data from meteorological observation equipment, satellite data and numerical prediction model; The construction module is used for constructing a user demand model according to a user query request corresponding to the user instruction, a user preference setting and a user historical behavior in response to detection of a user instruction; The user demand model is used to represent the user's demand and expectation target for meteorological information; The configuration module is used for initializing the meteorological early warning decision intelligent agent according to the user demand model, and configuring a perception module, an inference module and a learning module for the meteorological early warning decision intelligent agent; One end of the perception module is connected to meteorological observation equipment, satellites, radars, numerical prediction models and industry Internet of Things sensors, which are used to acquire the original meteorological data in real time, and the other end of the perception module is connected to the inference module, which is used to transmit the spatio-temporal tensor meteorological data after analyzing, boundary checking and exception elimination of the original meteorological data to the inference module; The generation module is used for generating meteorological decision suggestion data by the meteorological early warning decision intelligent agent using the inference module based on the spatio-temporal tensor meteorological data and the user demand model, and displaying the meteorological decision suggestion data in the graphical user interface; The determination module is used for determining the modification demand, the satisfaction evaluation data and the feedback opinion data of the user for the meteorological decision suggestion data according to the user feedback operation in response to the user feedback operation for the meteorological decision suggestion data in the graphical user interface, and obtaining user feedback information. An optimization module is configured to input the user feedback information into the learning module, so that the learning module iterates and optimizes the behavior strategy decision model of the meteorological warning decision intelligent agent based on the user feedback information through a machine learning algorithm, to obtain an optimized meteorological warning decision intelligent agent. The perception module comprises a user feedback perception module, and an execution process of the user feedback perception module comprises: collecting, semantic extraction and sentiment polarity determination of explicit feedback data and implicit feedback data of users in a business scenario, to obtain learnable user feedback signals; wherein the explicit feedback data is directly uploaded through an evaluation control or a voice intercom; the implicit feedback data is captured through a front-end burying point, and the front-end burying point comprises a click operation, a stay operation, a forwarding operation and an operation of whether to perform according to the suggestion; text data and voice data in the learnable user feedback signals are processed through a unified semantic emotion double-tower model; a first tower model in the semantic emotion double-tower model is configured to convert words in the text data into vectors, a second tower model in the semantic emotion double-tower model is configured to convert sound waves in the voice data into vectors, the first tower model and the second tower model are fused at the top of the semantic emotion double-tower model, and an intent label and a sentiment value are output; a graph neural network is used to identify abnormal objects in the users, abnormal feedback data corresponding to the abnormal objects are automatically screened out, filtered user feedback signals are obtained, and the filtered user feedback signals are determined as rewards or punishments of the learning module.

8. An electronic device comprising a memory, a processor, the memory having stored therein a computer program executable on the processor, characterized in that, The processor executes the computer program to realize the steps of the method in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions, and when the computer executable instructions are called and executed by the processor, the computer executable instructions cause the processor to execute the method in any one of claims 1 to 6.

Images