Method and system for generating meteorological disaster prevention and mitigation plans based on big data

Abstract

The present application relates to the field of data processing technology, and in particular provides a method and system for generating meteorological disaster prevention and mitigation plans based on big data, which adopts the strategy of two meteorological state recognition algorithms, and improves the accuracy of state recognition by comparing and analyzing the results of the two algorithms. At the same time, the embodiment of the present application also innovatively utilizes the abnormal jump thermal distribution in the historical meteorological monitoring log, providing a new perspective for meteorological data analysis. Combining these technologies, it is possible to generate more scientific, timely and effective disaster emergency response plans, significantly improving the ability of disaster prevention and mitigation.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method and system for generating meteorological disaster prevention and mitigation plans based on big data. Background Art

[0002] In the field of meteorological monitoring and disaster warning, accurate meteorological state recognition is the basis for formulating effective emergency plans. However, traditional meteorological state recognition methods often rely on a single algorithm, which limits the accuracy and comprehensiveness of recognition to a certain extent. Due to the differences in optimization strategies of different algorithms, a single algorithm may find it difficult to capture all meteorological features, thus affecting the accuracy of state recognition.

[0003] In addition, traditional meteorological data analysis often ignores the abnormal jump information in historical meteorological monitoring logs, which is crucial for understanding the complexity and variability of meteorological conditions. The lack of in-depth analysis and utilization of these abnormal data limits the comprehensive understanding of meteorological conditions and the accurate prediction of future trends.

[0004] In terms of emergency plan generation, existing methods often lack customized plans based on accurate meteorological state identification, which results in emergency responses not being quick and accurate enough when facing disasters, and failing to minimize the impact of disasters. Summary of the invention

[0005] In order to improve the above problems, the present application provides a method and system for generating meteorological disaster prevention and mitigation plans based on big data.

[0006] The present application embodiment provides a method for generating a meteorological disaster prevention and mitigation plan based on big data, which is applied to a text plan generation system. The method includes:

[0007] According to the meteorological state vector mining information of the historical meteorological monitoring log by the first meteorological state recognition algorithm and the second meteorological state recognition algorithm, the state recognition consistency error is determined, and there is a difference in the algorithm weight optimization strategy of the first meteorological state recognition algorithm and the second meteorological state recognition algorithm;

[0008] Using the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log, the meteorological monitoring data set in the historical meteorological monitoring log is marked to obtain a meteorological monitoring annotated data set, and the abnormal jump thermal distribution is used to reflect the abnormal jump bias coefficients of different meteorological monitoring data sets in the historical meteorological monitoring log;

[0009] Processing the meteorological monitoring annotated data set by using the first meteorological state recognition algorithm and the emergency plan generation algorithm to obtain a disaster emergency response plan;

[0010] The first meteorological state recognition algorithm is debugged by using the disaster prevention and reduction monitoring matching error between the disaster emergency response plan and the historical meteorological monitoring log, as well as the state recognition consistency error.

[0011] Optionally, the step of determining the abnormal jump thermal distribution includes at least one of the following:

[0012] Determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log by using the potential abnormal jump dense semantics obtained by mining the historical meteorological monitoring log using the abnormal jump mining branch in the second meteorological state recognition algorithm;

[0013] The abnormal jump mining branch in the first meteorological state recognition algorithm is used to mine the historical meteorological monitoring log to obtain the potential abnormal jump dense semantics, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log.

[0014] Optionally, the second meteorological state recognition algorithm includes an x-order abnormal jump mining branch, where x is a positive integer;

[0015] The method of using the abnormal transition mining branch in the second meteorological state recognition algorithm to mine the historical meteorological monitoring log for the potential abnormal transition dense semantics to determine the abnormal transition thermal distribution corresponding to the historical meteorological monitoring log includes:

[0016] For each order abnormal jump mining branch in the x-order abnormal jump mining branch, the potential abnormal jump dense semantics obtained by mining the historical meteorological monitoring log by using the abnormal jump mining branch are used to determine the local abnormal jump thermal relationship network corresponding to the abnormal jump mining branch;

[0017] The x local abnormal jump thermal relationship networks corresponding to the x-order abnormal jump mining branch are integrated to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log.

[0018] Optionally, for each order abnormal jump mining branch in the x-order abnormal jump mining branch, using the potential abnormal jump dense semantics obtained by mining the historical meteorological monitoring log by the abnormal jump mining branch, determining the local abnormal jump thermal relationship network corresponding to the abnormal jump mining branch, including:

[0019] Obtaining the u-th state transition dense semantics obtained by mining the historical meteorological monitoring logs by the u-th order abnormal transition mining branch in the second meteorological state recognition algorithm, where u is less than x;

[0020] Based on the semantic focus adjustment result of the u-th state jump dense semantics, obtaining the u+1-th abnormal jump bias coefficient of each meteorological monitoring data set;

[0021] Based on the u+1th abnormal jump bias coefficient of each meteorological monitoring data set, generating the local abnormal jump thermal relationship network corresponding to the u+1th order abnormal jump mining branch;

[0022] The integration of the x local abnormal jump thermal relationship networks corresponding to the x-order abnormal jump mining branch to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log includes: summing the abnormal jump bias coefficients corresponding to the same meteorological monitoring data set in different local abnormal jump thermal relationship networks to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log.

[0023] Optionally, on the basis that the abnormal jump thermal distribution is determined based on two ideas, the method includes:

[0024] On the basis that the current debugging cycle belongs to the first debugging cycle, the potential abnormal jump dense semantics obtained by mining the historical meteorological monitoring log using the abnormal jump mining branch in the first meteorological state recognition algorithm are used to determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log;

[0025] On the basis that the current debugging cycle belongs to the second debugging cycle, the abnormal jump mining branch in the second meteorological state recognition algorithm is used to mine the historical meteorological monitoring log to obtain the dense semantics of the potential abnormal jump, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log;

[0026] The first debugging cycle and the second debugging cycle are debugging cycles that are cyclically performed during the debugging process.

[0027] Optionally, the use of the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log to mark the meteorological monitoring data set in the historical meteorological monitoring log to obtain the meteorological monitoring annotated data set includes:

[0028] On the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the bias coefficient threshold, the target meteorological monitoring data set is marked to obtain the meteorological monitoring annotated data set.

[0029] Optionally, on the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than a bias coefficient threshold, the target meteorological monitoring data set is marked to obtain the meteorological monitoring annotated data set, including:

[0030] Acquire a marking guide feature, where the marking guide feature is used to locate a meteorological monitoring data set to be marked that meets a marking condition in the historical meteorological monitoring log;

[0031] On the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the bias coefficient threshold, and the target meteorological monitoring data set belongs to the meteorological monitoring data set to be marked indicated by the marking guidance feature, the target meteorological monitoring data set is marked to obtain the meteorological monitoring labeled data set.

[0032] Optionally, before the meteorological monitoring data set in the historical meteorological monitoring log is marked by using the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log to obtain the meteorological monitoring annotated data set, the method further includes:

[0033] The bias coefficient threshold used in the current debugging cycle is determined according to the current debugging cycle, wherein the bias coefficient threshold used in a debugging cycle with a higher sorting priority is greater than the bias coefficient threshold used in a debugging cycle with a lower sorting priority.

[0034] Optionally, the historical meteorological monitoring log is a historical meteorological monitoring message, and the method further includes:

[0035] Filter the first number of first to-be-processed meteorological monitoring messages from the historical meteorological monitoring message stream according to the filtering rule;

[0036] Disassembling the first to-be-processed meteorological monitoring messages of the first number into a plurality of meteorological monitoring message sets, and respectively selecting the second to-be-processed meteorological monitoring messages of the second number from each meteorological monitoring message set;

[0037] The screened second meteorological monitoring message to be processed is determined as the historical meteorological monitoring message, and the historical meteorological monitoring message is preprocessed, wherein the preprocessed meteorological monitoring message information matches the format conditions of the first meteorological state identification algorithm and the second meteorological state identification algorithm.

[0038] Optionally, the algorithm weight optimization strategy of the second meteorological state recognition algorithm is a soft threshold variable optimization strategy, and the mining information of the meteorological state vector of the historical meteorological monitoring log according to the first meteorological state recognition algorithm and the second meteorological state recognition algorithm is determined to determine the state recognition consistency error, including:

[0039] The historical meteorological monitoring log is input into the first meteorological state recognition algorithm to obtain a first historical meteorological state vector;

[0040] The historical meteorological monitoring log is input into the second meteorological state recognition algorithm to obtain a second historical meteorological state vector;

[0041] Obtaining a meteorological state vector debugging example from a debugging data set, wherein the meteorological state vector debugging example is a meteorological state vector obtained by mining the meteorological state vectors of the remaining historical meteorological monitoring logs by the second meteorological state recognition algorithm, and the debugging data set is continuously adjusted during the pre-debugging process;

[0042] The state identification consistency error is determined using the first historical meteorological state vector, the second historical meteorological state vector and the meteorological state vector debugging example, wherein the state identification consistency error has a first quantitative relationship with the quantitative difference between the first historical meteorological state vector and the second historical meteorological state vector, and has a second quantitative relationship with the quantitative difference between the first historical meteorological state vector and the meteorological state vector debugging example.

[0043] Optionally, the using the disaster prevention and mitigation monitoring matching error between the disaster emergency response plan and the historical meteorological monitoring log, and the state recognition consistency error, to debug the first meteorological state recognition algorithm includes:

[0044] According to the current debugging cycle, determine the plan matching confidence feature corresponding to the disaster prevention and reduction monitoring matching error and the consistency confidence feature corresponding to the state identification consistency error used in the current debugging cycle, wherein the consistency confidence feature used in the debugging cycle with a higher sorting priority is greater than the consistency confidence feature used in the debugging cycle with a lower sorting priority, and the plan matching confidence feature used in the debugging cycle with a lower sorting priority is greater than the plan matching confidence feature used in the debugging cycle with a higher sorting priority;

[0045] Determine a global training error using the disaster prevention and reduction monitoring matching error, the plan matching confidence feature, the state identification consistency error, and the consistency confidence feature;

[0046] The first meteorological state recognition algorithm is debugged using the global training error.

[0047] An embodiment of the present application provides a text plan generation system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the above method.

[0048] An embodiment of the present application provides a computer-readable storage medium on which a computer program is stored. The computer program implements the above method when running.

[0049] The embodiment of the present application adopts the strategy of two meteorological state recognition algorithms, and improves the accuracy of state recognition by comparing and analyzing the results of the two algorithms. At the same time, the embodiment of the present application also innovatively utilizes the abnormal jump thermal distribution in the historical meteorological monitoring log, providing a new perspective for meteorological data analysis. Combining these technologies, a more scientific, timely and effective disaster emergency response plan can be generated, significantly improving the ability of disaster prevention and mitigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a method for generating a meteorological disaster prevention and mitigation plan based on big data provided in an embodiment of the present application.

[0051] Figure 2 A structural diagram of a text plan generation system 200 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to better understand the above technical scheme, the technical scheme of the present application is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, rather than limitations on the technical scheme of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0053] Figure 1 A method for generating a meteorological disaster prevention and mitigation plan based on big data is shown, which is applied to a text plan generation system. The method includes the following steps 110 to 140.

[0054] Step 110, the text plan generation system mines information on the meteorological state vector of the historical meteorological monitoring log based on the first meteorological state recognition algorithm and the second meteorological state recognition algorithm to determine the state recognition consistency error. There are differences in the algorithm weight optimization strategies of the first meteorological state recognition algorithm and the second meteorological state recognition algorithm.

[0055] Step 120, the text plan generation system uses the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log to mark the meteorological monitoring data set in the historical meteorological monitoring log to obtain a meteorological monitoring annotated data set. The abnormal jump thermal distribution is used to reflect the abnormal jump bias coefficient of different meteorological monitoring data sets in the historical meteorological monitoring log.

[0056] Step 130: The text plan generation system processes the meteorological monitoring annotated data set through the first meteorological state recognition algorithm and the emergency plan generation algorithm to obtain a disaster emergency response plan.

[0057] Step 140: The text plan generation system uses the disaster prevention and mitigation monitoring matching error between the disaster emergency response plan and the historical meteorological monitoring log, as well as the state recognition consistency error, to debug the first meteorological state recognition algorithm.

[0058] In a possible application scenario, the text plan generation system first obtains a large number of historical meteorological monitoring logs. The historical meteorological monitoring logs record various meteorological data, such as temperature, humidity, wind speed, air pressure, etc. These meteorological data are crucial for analyzing and predicting meteorological conditions.

[0059] In step 110, the text plan generation system first uses two different meteorological state recognition algorithms: a first meteorological state recognition algorithm and a second meteorological state recognition algorithm to mine the meteorological state vectors in the historical meteorological monitoring log. Since the weight optimization strategies of the two algorithms are different, their processing results for the same log data may be different. By comparing the recognition results of the two algorithms, the system determines a state recognition consistency error, which reflects the consistency of the two algorithms in meteorological state recognition.

[0060] Next, in step 120, the text plan generation system uses the abnormal jump thermal distribution information in the historical meteorological monitoring log to mark the meteorological monitoring data set in the historical meteorological monitoring log. This abnormal jump thermal distribution can reflect the abnormal jump bias coefficients of different data sets in the historical meteorological monitoring log, that is, which data sets are more likely to have abnormal changes under what circumstances. Through this marking, the system obtains a meteorological monitoring annotated data set, which provides important reference information for subsequent processing.

[0061] In step 130, the text plan generation system uses the first meteorological state recognition algorithm and the emergency plan generation algorithm to process the marked meteorological monitoring data set. In this process, the system generates a disaster emergency response plan based on the results of the first meteorological state recognition algorithm and the emergency plan generation algorithm. This plan includes emergency measures and response strategies that should be taken under different meteorological conditions.

[0062] Finally, in step 140, the text plan generation system uses the disaster prevention and reduction monitoring matching error between the generated disaster emergency response plan and the historical meteorological monitoring log, as well as the previously calculated state recognition consistency error, to debug the first meteorological state recognition algorithm. This process is to improve the accuracy and reliability of the algorithm so that it can more accurately predict and respond to various situations in future meteorological monitoring and emergency response. In this way, the text plan generation system can continuously optimize its meteorological state recognition capabilities and provide more scientific and effective support for disaster emergency response.

[0063] Based on the above application scenarios, the following are three different meteorological disaster examples to introduce the workflow and application of the text plan generation system.

[0064] Example 1: Typhoon disaster: Before the typhoon arrives, the text plan generation system first obtains a large number of historical typhoon meteorological monitoring logs. In step 110, the system uses two different meteorological state recognition algorithms to analyze these logs and determine the state recognition consistency error to evaluate the algorithm's recognition accuracy of the typhoon meteorological state. In step 120, the system marks the data sets with abnormal jumps in the historical logs, especially the sharp changes in wind speed and air pressure, which are crucial for predicting the intensity and path of the typhoon. Entering step 130, the system combines the first meteorological state recognition algorithm and the emergency plan generation algorithm to generate an emergency response plan for the typhoon. This plan may include strategies such as evacuation routes, the setting of shelters, and the distribution of emergency supplies. In step 140, the system uses historical data to compare with the generated plan and debugs the algorithm to improve the accuracy of future typhoon forecasts.

[0065] Example 2: Heavy rain and floods: For heavy rain and flood disasters, the text plan generation system also mines data from historical meteorological monitoring logs. In step 110, the system identifies the meteorological state of the heavy rain through two algorithms and calculates the consistency error. In step 120, the system pays attention to abnormal jumps in key data such as rainfall and water level, and marks these data sets. In step 130, the system generates an emergency response plan for heavy rain and floods based on the meteorological state identification results, which may include the scheduling of drainage systems, the activation of flood control facilities, evacuation instructions, etc. Finally, in step 140, the system adjusts and optimizes the meteorological state identification algorithm according to the degree of match between the plan and the actual disaster situation.

[0066] Example 3: Drought disaster: In the face of drought disaster, the text plan generation system focuses on analyzing key data such as precipitation and soil moisture. In step 110, two meteorological state recognition algorithms will evaluate the severity and duration of the drought. In step 120, the system marks abnormal data sets related to drought in historical data, such as continued low precipitation or a sudden drop in soil moisture. In step 130, based on the meteorological state of drought, the system generates corresponding emergency plans, which may include water resource allocation, adjustment of irrigation plans, changes in crop planting structure, etc. In step 140, the system compares historical drought data with the generated plan, and continuously debugs and optimizes the algorithm to improve the ability to predict and respond to future drought conditions.

[0067] Through the above three examples, we can see that the text plan generation system can play an important role in different types of meteorological disasters. Through data analysis and algorithm optimization, it can provide scientific and effective support for disaster emergency response.

[0068] Next, taking typhoon disaster as an application scenario, the technical solutions recorded in steps 110 to 140 are introduced in detail.

[0069] In step 110, the first meteorological state recognition algorithm is an algorithm for identifying and analyzing meteorological states. In the application scenario of typhoon disasters, the first meteorological state recognition algorithm can be designed to pay special attention to meteorological characteristics related to typhoons, such as wind speed, wind direction, air pressure changes, etc. By processing and analyzing these data, the first meteorological state recognition algorithm can determine key information such as the formation, intensity, and movement path of the typhoon. The first meteorological state recognition algorithm can use machine learning or deep learning technology to improve its recognition accuracy through training with a large amount of historical data.

[0070] Similar to the first meteorological state recognition algorithm, the second meteorological state recognition algorithm is also an algorithm for analyzing meteorological states. However, its algorithm weight optimization strategy is different from the first algorithm, and it can focus more on different meteorological parameters or use different mathematical models. In the typhoon disaster scenario, the second meteorological state recognition algorithm can pay more attention to certain specific meteorological conditions or provide different prediction angles. The diversity of this algorithm helps to provide a more comprehensive meteorological state analysis and increase the reliability and accuracy of the prediction.

[0071] Historical meteorological monitoring logs refer to long-term accumulated records of meteorological monitoring data, including temperature, humidity, wind speed, air pressure and other meteorological parameters. Before a typhoon arrives, these logs are valuable resources because they contain information about similar events in the past and can be used to predict and assess the potential impact of future typhoons. By analyzing these logs, we can understand the historical trajectory of typhoons, changes in intensity and specific impacts on the local environment.

[0072] Meteorological state vector mining information refers to key information extracted from historical meteorological monitoring logs, which represents different aspects of meteorological state in the form of vectors. For example, in a typhoon disaster scenario, the meteorological state vector may include multiple dimensions such as wind speed, wind direction, precipitation, and air pressure changes. By mining and analyzing these vectors, we can more accurately understand the development trend and potential impact of typhoons.

[0073] The state recognition consistency error refers to the difference between the recognition results of the first meteorological state recognition algorithm and the second meteorological state recognition algorithm when analyzing the historical meteorological monitoring logs. The state recognition consistency error reflects the consistency of the two algorithms when identifying the same meteorological state. In the typhoon disaster scenario, reducing the state recognition consistency error helps to improve the accuracy and reliability of typhoon forecasts.

[0074] The algorithm weight optimization strategy refers to the strategy of adjusting its internal weights in order to improve the performance of the algorithm. Different meteorological state recognition algorithms can adopt different optimization strategies to better adapt to specific meteorological conditions and forecasting needs. In the application of typhoon disasters, the optimization strategy can include adjusting the weights of different meteorological parameters, improving the training method of the model, etc., to improve the algorithm's sensitivity to typhoon characteristics and recognition accuracy.

[0075] It can be understood that step 110 is one of the key steps for the text plan generation system to process typhoon disaster information. In step 110, the system first conducts in-depth mining and analysis of the meteorological state vectors in the historical meteorological monitoring log based on the first meteorological state recognition algorithm and the second meteorological state recognition algorithm. Although the two algorithms have the same goal of identifying meteorological states, their algorithm weight optimization strategies are different. This difference can be to capture the different characteristics of typhoons more comprehensively, or to provide different prediction perspectives. For example, the first algorithm can pay more attention to the changes in wind speed and wind direction, while the second meteorological state recognition algorithm can focus more on the analysis of air pressure and precipitation. During the mining process, the system extracts key meteorological state vectors related to typhoons, such as wind speed, wind direction, air pressure, etc., and conducts a detailed analysis of these vectors. By analyzing the changing trends of these vectors in historical data, the system can more deeply understand the evolution process of typhoons and their possible impacts. Subsequently, the system compares the recognition results of the two algorithms for the historical meteorological monitoring logs to determine the state recognition consistency error between them, which reflects the degree of difference between the two algorithms in identifying the same meteorological state. If the error is large, it means that the two algorithms have a large difference in identifying the typhoon state, which may require further adjustment of the algorithm's weight optimization strategy or introduction of more training data to improve the accuracy of recognition. Therefore, step 110 is an important part of the text plan generation system when processing typhoon disaster information. It provides key data support for subsequent disaster prediction and emergency plan generation by mining and analyzing the meteorological state vectors in the historical meteorological monitoring logs.

[0076] In step 120, the abnormal jump thermal distribution refers to the abnormal, large or sudden numerical changes in the time series of each meteorological data point in the historical meteorological monitoring log, which may be caused by extreme weather events such as typhoons. The thermal distribution shows the frequency and intensity of these abnormal jumps in different data sets and time points. For example, before a typhoon arrives, the wind speed and air pressure data may change dramatically, and these changes will be clearly marked on the thermal distribution map, helping researchers quickly identify which data points are affected by the typhoon.

[0077] Meteorological monitoring datasets refer to a series of meteorological-related data sets collected from various meteorological monitoring sites, including but not limited to wind speed, wind direction, temperature, humidity, air pressure, etc. In the application scenario of typhoon disasters, meteorological monitoring datasets are the key information source for recording the impact of typhoons. By analyzing these datasets, we can understand the intensity, movement path, and impact of typhoons on the surrounding environment.

[0078] The meteorological monitoring annotated dataset is a dataset obtained by additionally annotating the original meteorological monitoring dataset. Annotations include, for example, the identification, classification, and location of anomalies or important events in the data. In the case of a typhoon disaster, this may involve marking the data changes before and after the typhoon arrives, or marking the data segments most significantly affected by the typhoon. Such annotated datasets are very valuable for subsequent data analysis and model training because they provide additional contextual information.

[0079] The abnormal jump bias coefficient is an indicator used to quantify the degree of abnormal jumps in meteorological monitoring data. It reflects the degree of deviation of a data set from its normal range of variation under specific meteorological conditions. In typhoon disasters, this coefficient can be used to measure the abnormal fluctuations of data such as wind speed or air pressure under the influence of typhoons. A high bias coefficient may mean that the data set has been greatly affected and needs to be paid special attention.

[0080] In detail, the system will use the abnormal jump thermal distribution information recorded in the historical meteorological monitoring log based on step 120 to carefully mark the original meteorological monitoring data set. First, the system reviews and analyzes the historical meteorological monitoring logs, especially those during or before and after the typhoon. The abnormal jump thermal distribution in these logs provides valuable information about which data sets have abnormal fluctuations and when. For example, when a typhoon approaches, the wind speed and air pressure data sets may have significant abnormal jumps. Next, the system uses this abnormal jump information to mark the original meteorological monitoring data set. The marking process includes identifying those data points that are directly related to the impact of the typhoon and marking them as abnormal or important events. In this way, in subsequent data analysis and model training, researchers can quickly identify these key data points, so as to more accurately understand the impact of the typhoon on meteorological data. In addition, the system will also calculate the abnormal jump bias coefficient to quantify the degree of deviation of each data set under the influence of the typhoon. This coefficient helps researchers determine which data sets have the largest abnormal fluctuations during the typhoon, and then focus on these data sets to obtain more detailed information about the impact of the typhoon. Finally, through the processing of step 120, the system obtains a meteorological monitoring annotated data set, which not only contains the original meteorological monitoring data, but also adds the label and bias coefficient information related to the typhoon impact, which makes the subsequent data analysis and emergency plan generation more accurate and efficient.

[0081] In step 130, the emergency plan generation algorithm is used to automatically or semi-automatically generate emergency response plans for specific disaster situations. The algorithm generates a set of scientific and effective emergency plans by processing and analyzing various disaster-related data (such as meteorological data, geographic information, population distribution, resource reserves, etc.), combining the knowledge and experience of disaster management experts, and the response strategies of historical disaster cases. In the typhoon disaster scenario, the emergency plan generation algorithm will consider factors such as the typhoon's path, intensity, and expected impact range to formulate appropriate evacuation, rescue, and material deployment plan measures.

[0082] A disaster emergency response plan refers to a detailed plan prepared in advance before a disaster occurs in order to effectively respond to and reduce the impact of a disaster. This plan includes, for example, the issuance of warning signals, planning of evacuation routes, preparation and deployment of emergency supplies, deployment of rescue forces, arrangements for medical assistance, etc. In typhoon disasters, disaster emergency response plans are particularly important because they can guide disaster prevention and mitigation agencies to respond quickly and accurately before and when a typhoon arrives, thereby protecting people's lives and property to the greatest extent possible.

[0083] Further, step 130 involves converting the meteorological monitoring annotated data set into an actual disaster emergency response plan. The text plan generation system will first call the first meteorological state recognition algorithm to conduct an in-depth analysis of the data in the meteorological monitoring annotated data set. The first meteorological state recognition algorithm can identify key information such as the intensity, movement path, and expected impact range of the typhoon, which is the basis for formulating an emergency plan. Next, the system uses the emergency plan generation algorithm to process these annotated meteorological data. The emergency plan generation algorithm combines the professional knowledge and historical experience of disaster management, and can intelligently generate a set of suitable emergency response plans based on the specific situation of the current typhoon. This plan will plan in detail the evacuation route, emergency material allocation plan, rescue force deployment strategy, etc., to ensure that it can respond quickly and effectively when the typhoon comes. Finally, through the processing of step 130, the text plan generation system can output a complete disaster emergency response plan, which is not only based on real-time meteorological data, but also integrates the wisdom of experts and historical experience, providing strong decision-making support for disaster prevention and mitigation agencies, and helping to protect people's lives and property safety in typhoon disasters.

[0084] In step 140, the disaster prevention and mitigation monitoring matching error in the typhoon disaster management scenario refers to the difference between the disaster prevention and mitigation measures recommended by the disaster emergency response plan and the actual meteorological monitoring situation. This error can come from many aspects, including but not limited to the inconsistency between the prediction model in the plan and the actual situation, the accuracy of the monitoring data, the timeliness of the execution of the plan, etc. In order to reduce this error, the plan needs to be continuously debugged and optimized to ensure that it can more accurately respond to the actual typhoon disaster situation.

[0085] In detail, step 140 aims to debug and optimize the first meteorological state recognition algorithm by analyzing and comparing the disaster prevention and reduction monitoring matching errors between the disaster emergency response plan and the historical meteorological monitoring log, as well as the state recognition consistency error calculated previously. First, the system analyzes in detail the disaster prevention and reduction monitoring matching errors between the disaster emergency response plan and the actual meteorological monitoring log, which involves comparing the disaster prevention and reduction measures recommended in the plan with the actual monitored meteorological data to see whether the measures in the plan have been effectively applied in actual situations and whether the expected results have been achieved. If there is a large matching error, it means that some measures in the plan may not be applicable to the current meteorological conditions and need to be adjusted accordingly. Secondly, the system will also conduct a comprehensive analysis in combination with the state recognition consistency error calculated previously. The state recognition consistency error reflects the accuracy of the first meteorological state recognition algorithm in identifying meteorological states. If this error is large, it means that the algorithm may have problems in identifying meteorological states and needs to be debugged and improved. Finally, based on the analysis results of the above two errors, the system debugs the first meteorological state recognition algorithm, which can include adjusting the algorithm parameters, optimizing the model structure, improving the recognition method, etc., to improve the algorithm's recognition accuracy of meteorological states, thereby reducing the matching error of disaster prevention and reduction monitoring, and making the disaster emergency response plan more in line with the actual meteorological conditions. Through such a debugging process, the text plan generation system can continuously improve its ability and effectiveness in responding to typhoon disasters.

[0086] In the above technical solution, the first meteorological state recognition algorithm is an algorithm designed to identify and classify meteorological states. It identifies the current meteorological state by analyzing meteorological data in historical meteorological monitoring logs, such as temperature, humidity, wind speed, air pressure, etc. The algorithm can use machine learning or deep learning methods to learn how to accurately identify different meteorological states by training a large amount of historical data. In addition, the algorithm also takes into account the characteristics of time series data and can capture the changing trend of meteorological data over time, so as to more accurately predict future meteorological states. The first meteorological state recognition algorithm plays an important role in the typhoon disaster management system, and its accuracy directly affects the formulation of disaster emergency response plans.

[0087] The second meteorological state recognition algorithm is similar to the first meteorological state recognition algorithm and is also an algorithm for identifying and classifying meteorological states. However, unlike the first algorithm, the second algorithm has differences in optimization strategies and can focus more on certain specific meteorological features or adopt different model architectures. This difference enables the two algorithms to provide different perspectives and focuses when identifying meteorological states, thereby increasing the robustness and accuracy of the entire system. The second meteorological state recognition algorithm can also use machine learning or deep learning technology and improve its recognition performance through a unique weight optimization strategy.

[0088] The emergency plan generation algorithm is an algorithm that automatically generates disaster emergency response plans based on current meteorological conditions and forecast results. The algorithm combines multiple data sources, including meteorological monitoring data, geographic information, population distribution, etc., as well as disaster management expertise and historical experience. By analyzing and processing these data, the emergency plan generation algorithm can assess the potential impact of typhoon disasters and formulate corresponding response measures, which may include evacuation route planning, rescue force deployment, material allocation, etc. The core goal of the algorithm is to quickly and accurately generate disaster emergency response plans that conform to actual conditions in order to minimize the losses caused by disasters.

[0089] In more detail, the first meteorological state recognition algorithm can be exemplarily embodied as a deep learning model based on a long short-term memory network (LSTM). LSTM is a neural network suitable for processing and predicting complex patterns with very long intervals and delays in time series data. In this algorithm, LSTM is trained to identify long-term dependencies of key meteorological parameters such as temperature, humidity, wind speed, and air pressure in historical meteorological monitoring logs. By capturing the subtle differences in how these parameters change over time, the model can accurately distinguish different meteorological states, such as sunny days, rainy days, cloudy days, snowy days, etc. In addition, the algorithm also incorporates an attention mechanism, which enables the model to adaptively focus on the historical data segments that are most relevant to the current meteorological state, thereby improving the accuracy of recognition.

[0090] The second meteorological state recognition algorithm can use an improved convolutional neural network (CNN) model, which is particularly suitable for processing data with spatial correlation. In this algorithm, CNN is used to extract local features in meteorological data, such as wind direction shear, air pressure gradient, etc. These features are crucial for identifying complex meteorological states. Unlike the first algorithm, the second algorithm can focus more on capturing the spatial distribution pattern of meteorological data to improve the recognition sensitivity of some specific meteorological states (such as the sudden drop in air pressure before a typhoon). In addition, the algorithm can also incorporate an adaptive weight adjustment strategy, so that the model can dynamically adjust its learning focus according to the difficulty of identifying different meteorological states.

[0091] The emergency plan generation algorithm can be built on a deep generative model, such as a generative adversarial network (GAN) or a variational autoencoder (VAE), which can learn the potential representation of disaster emergency response plans from a large amount of historical data. By inputting data such as the current meteorological status, geographic information, and population distribution, the model can generate a series of emergency response plan candidates that conform to the actual situation. These plans can include the planning of evacuation routes, the deployment strategy of rescue forces, and the optimization plan for material allocation. In order to ensure that the generated plans are both scientific and practical, the knowledge and experience of disaster management experts are also incorporated into the algorithm. The generated plans are screened and optimized through a rule-based post-processing method, and finally a set of efficient and feasible disaster emergency response plans are output.

[0092] In addition, in the above technical solution, the state recognition consistency error (or state recognition consistency loss) can be obtained by calculating the difference between the meteorological state recognition results of the first meteorological state recognition algorithm and the second meteorological state recognition algorithm for the same historical meteorological monitoring log. The meteorological state vector identified by the first algorithm is (S_1), and the meteorological state vector identified by the second algorithm is (S_2). The state recognition consistency error (E_{consistency}) can be expressed by the following formula: [E_{consistency}=|S_1-S_2|_p];

[0093] Among them, (|*|_p) represents the (p)-norm. For example, (p=2) (i.e., Euclidean distance) can be selected as a standard to measure the difference between two state vectors. This error reflects the consistency of the two algorithms when identifying the same meteorological data. The smaller the error, the closer the recognition results of the two algorithms are.

[0094] The disaster prevention and mitigation monitoring matching error (or disaster prevention and mitigation monitoring matching loss) measures the degree of matching between the disaster emergency response plan and the historical meteorological monitoring log. The meteorological state vector corresponding to the disaster prevention and mitigation measures recommended in the disaster emergency response plan is (P), while the meteorological state vector in the actual historical meteorological monitoring log is (M). The disaster prevention and mitigation monitoring matching error (E_{matching}) can be calculated using the following formula: [E_{matching}=|PM|_q];

[0095] Similarly, (|*|_q) in the embodiment of the present application represents the (q)-norm, and (q=2) is taken as an example. This error is used to evaluate the degree of conformity between the meteorological conditions in the plan and the actual monitored meteorological conditions; the smaller the error, the more consistent the plan is with the actual situation, and the disaster prevention and mitigation measures can be more effective.

[0096] The calculation of the above two errors provides quantitative indicators for the optimization of the typhoon disaster management system, helping to adjust and improve relevant algorithms, thereby improving the overall performance and accuracy of the system.

[0097] By using the embodiment of the present application, by adopting the first meteorological state recognition algorithm and the second meteorological state recognition algorithm, the difference in the weight optimization strategy of the two algorithms makes the state recognition more comprehensive and accurate. By comparing the two algorithms to mine the meteorological state vector information of the historical meteorological monitoring log, the consistency error of the state recognition can be accurately determined, thereby effectively evaluating the performance of the algorithm. In addition, by using the abnormal jump thermal distribution in the historical meteorological monitoring log, the meteorological monitoring data set can be accurately marked to generate a meteorological monitoring annotated data set. This method not only reveals the abnormal jump situation in the data set, but also provides a deeper data analysis perspective through the abnormal jump bias coefficient.

[0098] Furthermore, by processing the labeled data set with the first meteorological state recognition algorithm and the emergency plan generation algorithm, a disaster emergency response plan can be quickly generated. This plan is based on accurate meteorological state recognition and abnormal data analysis, and is therefore more targeted and effective. Finally, by comparing the disaster emergency response plan and the disaster prevention and mitigation monitoring matching error between the historical meteorological monitoring log, as well as the aforementioned state recognition consistency error, the first meteorological state recognition algorithm can be fine-tuned to continuously improve its accuracy and adaptability. This series of processes and strategies not only improves the accuracy of meteorological state recognition, but also provides a more scientific, timely and effective plan for disaster emergency response, significantly enhancing the ability of disaster prevention and mitigation.

[0099] In some preferred embodiments, the step of determining the abnormal jump thermal distribution includes at least one of the following.

[0100] The first determination step: using the abnormal jump mining branch in the second meteorological state recognition algorithm to mine the historical meteorological monitoring log to obtain the potential abnormal jump dense semantics, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log.

[0101] The second determination step: using the abnormal jump mining branch in the first meteorological state recognition algorithm to mine the historical meteorological monitoring log to obtain the potential abnormal jump dense semantics, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log.

[0102] In certain preferred embodiments of the text plan generation system, the determination of abnormal jump thermal distribution is performed using the following detailed steps, which is crucial for improving the accuracy of meteorological state identification and the pertinence of emergency plans.

[0103] First, in the first determination step, the text plan generation system uses the abnormal jump mining branch in the second meteorological state recognition algorithm to conduct in-depth mining of historical meteorological monitoring logs. This mining process focuses on identifying the potential abnormal jump dense semantics in the logs. These abnormal jumps can be manifested as sharp changes in meteorological parameters such as temperature, wind speed, and air pressure, which are points that need special attention in meteorological analysis. Through the special optimization strategy of the second meteorological state recognition algorithm, the system can capture these anomalies more accurately and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring logs. This distribution map intuitively shows in which time periods and regions the abnormal jump phenomenon is most concentrated, providing an important reference for subsequent disaster warnings and emergency plan formulation.

[0104] Next, in the second determination step, the text plan generation system turns to the abnormal jump mining branch in the first meteorological state recognition algorithm. Unlike the second algorithm, the first algorithm can use different weights and optimization strategies when mining abnormal jumps, thus providing another perspective to examine potential anomalies in historical meteorological monitoring logs. Similarly, through the in-depth mining of this algorithm, the system can identify the dense semantics of abnormal jumps in the logs and generate the corresponding abnormal jump thermal distribution. This step complements the first step and together provides the system with a more comprehensive and multi-angle abnormal jump analysis.

[0105] After completing these two steps, the text plan generation system has detailed information about abnormal jumps in historical meteorological monitoring logs, which not only helps to improve the accuracy of meteorological state identification, but also provides valuable data support for the generation of emergency plans. The system can formulate targeted disaster prevention and mitigation measures based on the thermal distribution of these abnormal jumps to ensure timely and effective responses before disasters occur.

[0106] In this way, the text plan generation system can more accurately identify and analyze abnormal jumps in historical meteorological monitoring logs, and then generate more scientific and practical disaster emergency response plans, which not only improves the efficiency and accuracy of disaster prevention and mitigation, but also provides a more solid guarantee for the safety of life and property of the public.

[0107] In the following example, the second meteorological state identification algorithm includes an x-order abnormal jump mining branch, where x is a positive integer; then, the abnormal jump mining branch in the second meteorological state identification algorithm is used to mine the historical meteorological monitoring log to obtain the dense semantics of potential abnormal jumps, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log, including: for each order of abnormal jump mining branches in the x-order abnormal jump mining branch, the abnormal jump mining branch is used to mine the historical meteorological monitoring log to obtain the dense semantics of the potential abnormal jumps, and determine the local abnormal jump thermal relationship network corresponding to the abnormal jump mining branch; integrate the x local abnormal jump thermal relationship networks corresponding to the x-order abnormal jump mining branch to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log.

[0108] In a specific embodiment of the text plan generation system, a second meteorological state recognition algorithm is introduced. The special feature of this algorithm is that it contains an x-order abnormal jump mining branch, where x is a positive integer, which means that the algorithm can deeply mine potential abnormal jump information in historical meteorological monitoring logs from multiple levels and dimensions.

[0109] First, for each order of the x-order abnormal jump mining branch, the system independently uses these branches to perform deep mining on historical meteorological monitoring logs. Each order of the branch focuses on identifying different types of potential abnormal jump dense semantics, which may include but are not limited to rapid changes in temperature, sudden increases or decreases in wind speed, or abnormal fluctuations in air pressure. Through the fine mining of each order of branches, the system can capture these subtle but important abnormal jump signals.

[0110] Next, for each order of abnormal jump mining branch, the system determines the local abnormal jump thermal relationship network corresponding to the branch based on the dense semantics of the potential abnormal jumps mined. These relationship networks are actually a form of data visualization that can intuitively show the concentration and mutual relationship of abnormal jumps in a specific time period and location.

[0111] Finally, the system integrates the x local abnormal jump thermal relationship networks obtained by all x-order abnormal jump mining branches. By superimposing and integrating this information, the system finally generates a comprehensive and detailed abnormal jump thermal distribution map, which not only reveals the overall situation and trend of abnormal jumps in historical meteorological monitoring logs, but also provides a scientific basis for subsequent disaster warning and emergency plan formulation.

[0112] Through this series of precise and meticulous operations, the text plan generation system can more accurately identify and analyze abnormal jumps in meteorological data, thereby improving the accuracy of meteorological state recognition and the pertinence of emergency plans. This multi-level abnormal jump mining method not only enhances the system's ability to parse data, but also provides more reliable data support for disaster prevention and mitigation work. Therefore, by introducing the x-order abnormal jump mining branch and integrating the local abnormal jump thermal relationship network, the text plan generation system can generate a more accurate and comprehensive abnormal jump thermal distribution map, which not only improves the system's sensitivity and recognition accuracy to meteorological anomalies, but also provides strong support for the formulation of more scientific and effective disaster emergency response plans. In this way, the efficiency and accuracy of disaster prevention and mitigation work can be greatly improved, making important contributions to protecting the lives and property of the public.

[0113] It is worth mentioning that the abnormal jump mining branch is a core innovation in the embodiment of this application. It is specially designed to find and locate jump points that are significantly different from the normal data pattern in data sets such as historical meteorological monitoring logs. These jump points can indicate some abnormal meteorological events or data recording errors, which are of great value for meteorological analysis, early warning systems, and the formulation of emergency plans.

[0114] Take Isolation Forest (iForest) as an example, which is a machine learning algorithm suitable for anomaly detection. In the abnormal jump mining branch, the Isolation Forest algorithm can be used to model the distribution of normal data points and identify outliers by calculating the anomaly score of each data point. The Isolation Forest algorithm constructs multiple isolation trees by randomly selecting features and split values ​​until each leaf node contains only one data point or reaches the maximum depth. In this process, anomalies are often isolated at a shallow level because they have a different distribution pattern from normal data.

[0115] Through this method, the abnormal jump mining branch can effectively identify the abnormal jump points in the data set and provide accurate information for subsequent analysis and early warning.

[0116] The dense semantics of potential abnormal jumps is another key concept in the embodiments of the present application, which refers to the deep semantic information related to abnormal jumps hidden in the data. This information may not be easy to observe directly, but can be mined and analyzed through specific algorithms and techniques. For example, there is a historical meteorological monitoring data set, in which each data point is a multidimensional feature vector, including information in multiple dimensions such as temperature, humidity, and wind speed. In a certain period of time, it was observed that the temperature characteristic value fluctuated abnormally, accompanied by corresponding changes in humidity and wind speed. The correlation and change pattern between these feature vectors constitute part of the dense semantics of potential abnormal jumps.

[0117] To illustrate this more concretely, consider an example where the following sequence of feature vectors is observed in three consecutive days of meteorological monitoring data:

[0118] Day 1: [Temperature: 20℃, Humidity: 60%, Wind speed: 5m / s];

[0119] Day 2: [Temperature: 25°C, Humidity: 50%, Wind speed: 7m / s];

[0120] Day 3: [Temperature: 35℃, humidity: 30%, wind speed: 12m / s].

[0121] In this example, the data on the third day showed a significant jump compared to the previous two days, especially in temperature and wind speed. This jump can indicate an abnormal meteorological event (such as a heat wave or the arrival of a storm). By analyzing and mining the dense semantics of these potential abnormal jumps, we can better understand the inherent laws and abnormal patterns of the data, thereby making more accurate predictions and decisions.

[0122] In summary, the abnormal jump mining branch and the potential abnormal jump dense semantics are two core inventions in the embodiments of this application. They use specific algorithms and technologies to identify and analyze abnormal patterns and deep semantic information in data, providing strong support for meteorological analysis, early warning systems, and the formulation of emergency plans.

[0123] In the next step, for each order abnormal jump mining branch in the x-order abnormal jump mining branch, the potential abnormal jump dense semantics obtained by mining the historical meteorological monitoring logs by the abnormal jump mining branch are used to determine the local abnormal jump thermal relationship network corresponding to the abnormal jump mining branch, including: obtaining the u-th state jump dense semantics obtained by mining the historical meteorological monitoring logs by the u-th order abnormal jump mining branch in the second meteorological state recognition algorithm, where u is less than x; based on the semantic focusing adjustment result of the u-th state jump dense semantics, obtaining the u+1-th abnormal jump bias coefficient of each meteorological monitoring data set; based on the u+1-th abnormal jump bias coefficient of each meteorological monitoring data set, generating the local abnormal jump thermal relationship network corresponding to the u+1-th order abnormal jump mining branch.

[0124] Then, the x local abnormal jump thermal relationship networks corresponding to the x-order abnormal jump mining branch are integrated to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log, including: summing the abnormal jump bias coefficients corresponding to the same meteorological monitoring data set in different local abnormal jump thermal relationship networks to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log.

[0125] During the operation of the text plan generation system, when the system processes the x-order abnormal jump mining branch to mine the dense semantics of potential abnormal jumps in historical meteorological monitoring logs and determine the thermal distribution of abnormal jumps, the following detailed steps will be executed.

[0126] First, the system obtains the u-th state jump dense semantics obtained by mining the historical meteorological monitoring logs by the u-th order abnormal jump mining branch in the second meteorological state recognition algorithm. u in the embodiment of the present application is a positive integer less than x, representing the order of the current processing. The u-th state jump dense semantics refers to the deep semantic information related to the abnormal jump identified in the u-th order mining branch.

[0127] Next, the system performs semantic focus adjustment on the dense semantics of the u-th state jump. The purpose of this step is to more accurately locate the characteristics and patterns of abnormal jumps. Through semantic focus adjustment, the system can identify which data points or data sets are more likely to have abnormal jumps under specific meteorological conditions, and calculate a u+1-th abnormal jump bias coefficient for each meteorological monitoring data set. This bias coefficient reflects the tendency of the data set to have abnormal jumps in the next-level mining.

[0128] Then, based on the u+1th abnormal jump bias coefficient of each meteorological monitoring data set, the system generates a local abnormal jump thermal relationship network corresponding to the u+1th abnormal jump mining branch. This relationship network shows the correlation and strength of abnormal jumps between various data sets in a visual way.

[0129] Finally, when the system has completed the processing of all x-order abnormal jump mining branches, it will integrate the local abnormal jump thermal relationship network. In this step, the system sums the abnormal jump bias coefficients corresponding to the same meteorological monitoring data set in different local abnormal jump thermal relationship networks. The purpose of this is to obtain a comprehensive map that reflects the abnormal jump thermal distribution in the entire historical meteorological monitoring log.

[0130] In this way, the text plan generation system can generate an accurate and detailed abnormal jump thermal distribution map. This map not only reveals the overall situation and trend of abnormal jumps in historical meteorological monitoring logs, but also provides a scientific basis for subsequent disaster warning and emergency plan formulation. It can be seen that by introducing the x-order abnormal jump mining branch and accurately calculating the bias coefficients of each order abnormal jump, the text plan generation system can generate a more accurate and comprehensive abnormal jump thermal distribution map. This not only improves the system's sensitivity and recognition accuracy to meteorological anomalies, but also provides strong support for the formulation of more scientific and effective disaster emergency response plans.

[0131] It is worth mentioning that the semantic focus adjustment result is a core invention of the text plan generation system, which combines the attention mechanism to achieve precise adjustment of semantics. The exemplary implementation of the semantic focus adjustment process is as follows.

[0132] 1) Data preprocessing: First, the system preprocesses historical meteorological monitoring logs or other related data to extract key information and features. This information can include data in multiple dimensions such as temperature, humidity, wind speed, and their changing trends over time.

[0133] 2) Introduction of attention mechanism: The key to the semantic focus adjustment result is the introduction of the attention mechanism. The attention mechanism allows the model to give different attention to different parts when processing input data, so as to highlight important information and suppress irrelevant information.

[0134] 3) Calculate attention weights: The system calculates attention weights based on the characteristics of the data. For example, in meteorological data, certain outliers or mutation points may be given higher weights. These weights reflect the importance of each part of the data to the overall semantics.

[0135] 4) Semantic focus adjustment: Using the calculated attention weights, the system makes weighted adjustments to the original semantics. In this way, important semantic features will be strengthened, while unimportant features will be relatively weakened.

[0136] 5) Output results: After semantic focus adjustment, the system outputs an adjusted semantic representation. This representation is more focused on key information, which is helpful for subsequent data analysis and decision making.

[0137] It can be seen that by introducing the attention mechanism, semantic focus adjustment can more accurately capture the key information and features in the data. This is crucial for tasks such as identifying abnormal jumps and predicting meteorological trends. Traditional data processing methods often treat all data equally, while semantic focus adjustment can highlight key information. This makes the system more efficient and accurate when processing complex data. Semantic focus adjustment can be flexibly adjusted according to specific tasks and data characteristics. By adjusting the calculation method of attention weights, it can adapt to different application scenarios and needs. Since the attention mechanism can clearly point out which parts of the data receive more attention, the results of semantic focus adjustment also have a certain degree of interpretability. This is of great significance for understanding the model decision process and results. In summary, the results of semantic focus adjustment achieve accurate capture and enhancement of key semantics by introducing the attention mechanism, thereby improving the accuracy and efficiency of data processing. This invention has significant novelty and practical value.

[0138] In some other possible embodiments, on the basis that the abnormal jump thermal distribution is determined based on two ideas, the method includes: on the basis that the current debugging cycle belongs to the first debugging cycle, the abnormal jump mining branch in the first meteorological state recognition algorithm is used to mine the historical meteorological monitoring log to obtain the dense semantics of the potential abnormal jump, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log; on the basis that the current debugging cycle belongs to the second debugging cycle, the abnormal jump mining branch in the second meteorological state recognition algorithm is used to mine the historical meteorological monitoring log to obtain the dense semantics of the potential abnormal jump, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log; wherein, the first debugging cycle and the second debugging cycle are debugging cycles that are cyclically performed during the debugging process.

[0139] In some embodiments of the text plan generation system, the determination of abnormal jump thermal distribution is based on two different debugging cycles. These two debugging cycles can be understood as a first debugging cycle and a second debugging cycle, which are cyclically performed during the system debugging process.

[0140] When the system is in the first debugging cycle, the text plan generation system uses the abnormal jump mining branch in the first meteorological state recognition algorithm to perform in-depth mining of historical meteorological monitoring logs. This mining process aims to identify and extract the potential abnormal jump dense semantics in the logs. These dense semantics can hide abnormal patterns or trends in meteorological data, which is crucial for understanding and predicting future meteorological conditions.

[0141] After the first debugging cycle, the system determines the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log based on the mined potential abnormal jump dense semantics. This process involves in-depth analysis of the data to reveal which areas or time periods have more concentrated abnormal jumps, thereby helping decision makers better understand the dynamic changes of meteorological data and formulate corresponding countermeasures accordingly.

[0142] When the system enters the second debugging cycle, the text plan generation system will use the abnormal jump mining branch in the second meteorological state recognition algorithm to mine the historical meteorological monitoring logs. Different from the first algorithm, the second meteorological state recognition algorithm can use different models, parameters or processing methods to extract different potential abnormal jump dense semantics from the logs.

[0143] Similarly, after the second debugging cycle, the system determines the thermal distribution of abnormal jumps in the historical meteorological monitoring logs based on the mining results. By comparing the thermal distributions obtained under the two debugging cycles, the system can have a more comprehensive understanding of the abnormal jumps in the meteorological data and discover abnormal patterns that can be ignored by a single algorithm.

[0144] The circulation of these two debugging cycles enables the text plan generation system to examine the historical meteorological monitoring logs from multiple angles, thereby obtaining a more accurate and comprehensive abnormal jump thermal distribution. This not only improves the system's ability to detect meteorological anomalies, but also provides more reliable data support for subsequent meteorological warnings and emergency plan formulation. In this way, by introducing two different debugging cycles and corresponding meteorological state recognition algorithms, the text plan generation system can more accurately determine the abnormal jump thermal distribution in the historical meteorological monitoring logs. This technical solution not only enhances the system's ability to process complex meteorological data, but also provides more solid technical support for meteorological warnings and emergency responses, which helps to improve the efficiency and accuracy of disaster prevention and mitigation.

[0145] In other preferred design ideas, the meteorological monitoring data set in the historical meteorological monitoring log is marked by using the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log to obtain the meteorological monitoring labeled data set, including: on the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the bias coefficient threshold, the target meteorological monitoring data set is marked to obtain the meteorological monitoring labeled data set.

[0146] In other preferred design ideas of the text plan generation system, the system will mark the meteorological monitoring data set in the historical meteorological monitoring log using the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log, thereby obtaining the meteorological monitoring annotated data set. This process is crucial for subsequent data analysis and model training.

[0147] First, the generated abnormal jump thermal distribution is systematically reviewed. This distribution diagram shows in detail the abnormal jumps of various meteorological monitoring data sets in different time periods or regions. The abnormal jump bias coefficient is a key indicator, which reflects the tendency or strength of the data set in terms of abnormal jumps.

[0148] Next, the system sets a bias coefficient threshold. This threshold is an important reference standard for determining which meteorological monitoring data sets have abnormal jumps that require special attention. The threshold is set based on historical data, expert experience, and actual needs to ensure that truly meaningful abnormal jumps can be screened out.

[0149] Then, the system starts to traverse each meteorological monitoring data set in the abnormal jump thermal distribution and compares its abnormal jump bias coefficient with the set threshold value. When the abnormal jump bias coefficient of a target meteorological monitoring data set is greater than the threshold, it means that the data set has significant abnormal jumps and needs to be specially marked.

[0150] Finally, the system marks the meteorological monitoring data sets that exceed the threshold to form a meteorological monitoring annotated data set. This annotated data set will contain data with significant abnormal jump characteristics, which is of great value for subsequent data mining, model training, and abnormal warning tasks.

[0151] In this way, the text plan generation system can automatically identify and mark key abnormal data in historical meteorological monitoring logs, greatly improving the efficiency and accuracy of data processing. At the same time, this labeling method also provides more accurate data input for subsequent data analysis and early warning systems, which helps to improve the performance and reliability of the entire system. In this way, by using abnormal jump thermal distribution to mark meteorological monitoring data sets, the text plan generation system not only optimizes the data processing process, but also provides more accurate data support for subsequent meteorological anomaly warnings and emergency plan formulation. This technical solution significantly improves the system's intelligence level and ability to cope with complex meteorological conditions, and has made important contributions to protecting the safety of public life and property and social stability.

[0152] In the next step, on the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the bias coefficient threshold, the target meteorological monitoring data set is marked to obtain the meteorological monitoring labeled data set, including: obtaining a marking guidance feature, the marking guidance feature is used to locate the meteorological monitoring data set to be marked that meets the marking conditions in the historical meteorological monitoring log; on the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the bias coefficient threshold, and the target meteorological monitoring data set belongs to the meteorological monitoring data set to be marked indicated by the marking guidance feature, the target meteorological monitoring data set is marked to obtain the meteorological monitoring labeled data set.

[0153] In the next step, the text plan generation system will continue to process the data in the abnormal jump thermal distribution, especially for those meteorological monitoring data sets whose abnormal jump bias coefficients exceed the set threshold. The key to this process is to accurately mark these data sets for subsequent data analysis and model training.

[0154] First, the system obtains the tagging guidance features. The tagging guidance features are designed based on historical data, expert knowledge and system requirements. The tagging guidance features help the system accurately locate the meteorological monitoring data sets to be marked in the historical meteorological monitoring logs that meet specific tagging conditions.

[0155] Next, the system combines the abnormal jump thermal distribution and the marking guidance feature to mark the data set. When the system finds that the abnormal jump bias coefficient corresponding to a target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the set bias coefficient threshold, it means that the data set has significant abnormal jumps. However, at this time, the system will not mark it immediately, but will further check whether this data set belongs to the meteorological monitoring data set to be marked indicated by the marking guidance feature.

[0156] The system will only mark the target meteorological monitoring data set if it satisfies both the condition that the abnormal jump bias coefficient is greater than the threshold and the indication of the marking guidance feature. This double verification mechanism ensures that the marked data set not only has significant abnormal jump characteristics, but is also a data set that the system needs to pay special attention to.

[0157] Finally, after this series of precise and rigorous steps, the system obtains a meteorological monitoring annotated dataset. This dataset will contain all the labeled target meteorological monitoring datasets, which have significant abnormal jump characteristics and properties that the system pays special attention to, and are of great value for subsequent data mining, model training, and abnormal warning tasks.

[0158] In this way, the text plan generation system can more accurately identify and mark key meteorological monitoring data sets, providing strong support for subsequent data analysis and processing. At the same time, this marking method based on abnormal jump thermal distribution and marking guidance features also greatly improves the efficiency and accuracy of data processing, enabling the system to respond more quickly to complex and changeable meteorological conditions. It can be seen that the text plan generation system combines abnormal jump thermal distribution and marking guidance features to accurately mark meteorological monitoring data sets, which not only optimizes the data processing process, but also provides more accurate data support for subsequent meteorological anomaly warnings and emergency plan formulation.

[0159] It is worth mentioning that the marking guidance feature is another core invention point of the embodiment of the present application, which is mainly used to guide the system to accurately mark key meteorological monitoring data sets. The marking guidance feature can be a series of numerical indicators, which are carefully designed based on historical meteorological data, expert experience and system requirements. For example, in a specific implementation case, the marking guidance feature can include the following numerical values: Data abnormality: The degree of deviation of the data set calculated by the algorithm from the normal mode can be represented by a specific numerical value, such as 0.8 (ranging between 0 and 1, the larger the value, the higher the abnormality); Time period importance: Data in certain time periods can be more critical, such as data in the rainy season. This feature can be represented by a weighted numerical value, such as 1.5 in the rainy season and 1 in other seasons; Geographical location importance: For certain specific geographical locations, such as flood-prone areas, the importance of data can be higher. This feature can also be represented by a specific numerical value, such as 2 for flood-prone areas and 1 for non-prone areas.

[0160] The acquisition of labeling guidance features exemplarily involves the following steps: data collection: first, the system needs to collect a large amount of historical meteorological monitoring data, including temperature, humidity, wind speed and other information; feature extraction: based on the collected data, the system will use algorithms to automatically extract features related to abnormal jumps, such as the abnormality of the data, the importance of time period and geographical location, etc.; expert evaluation: experts will evaluate and adjust the extracted features to ensure that these features can accurately reflect the abnormality and importance of the data; feature application: finally, these verified features will be applied to the system as labeling guidance features to guide the labeling of the data set.

[0161] By introducing the marking guidance feature, the embodiment of the present application can more accurately identify and mark key meteorological monitoring data sets and reduce false alarms and missed alarms. The marking guidance feature can help the system optimize resource allocation, focus more computing and analysis resources on key data sets, and improve overall processing efficiency. Since the marking guidance feature is designed based on historical data and expert experience, it can be flexibly adjusted according to actual conditions to adapt to different application scenarios and needs. In summary, the introduction of the marking guidance feature is an important innovation in the text plan generation system. It not only improves the accuracy and efficiency of the system in processing data, but also provides more accurate data support for subsequent data analysis and early warning.

[0162] In other alternative embodiments, the method further includes: determining the bias coefficient threshold used in the current debugging cycle according to the current debugging cycle, wherein the bias coefficient threshold used in the debugging cycle with a higher sorting priority is greater than the bias coefficient threshold used in the debugging cycle with a lower sorting priority, before marking the meteorological monitoring data set in the historical meteorological monitoring log using the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log to obtain the meteorological monitoring annotated data set.

[0163] In some alternative embodiments of the text plan generation system, before marking the meteorological monitoring data set using the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log, the system will first determine a key parameter based on the current debugging cycle: the bias coefficient threshold. This threshold value plays a vital role in the data marking process, and it determines which data sets have a large enough degree of abnormal jump and need to be specially marked.

[0164] The debugging cycle is an important stage in the system development and optimization process, and each debugging cycle has its specific goals and priorities. In this embodiment, the system sets different bias coefficient thresholds according to the sorting priority of the debugging cycle. Specifically, a relatively high bias coefficient threshold is used for a debugging cycle with a high sorting priority, while a relatively low bias coefficient threshold is used for a debugging cycle with a low sorting priority.

[0165] This design is based on the following considerations: In the early debugging cycle, the system can pay more attention to those data sets with very obvious abnormal jumps, so as to quickly discover and fix potential problems. Therefore, setting a higher bias coefficient threshold can help the system filter out those less significant abnormal jumps and focus on more serious situations. As the debugging progresses, the system may need to pay attention to more details and edge cases. At this time, lowering the bias coefficient threshold can allow more data sets to be marked for further analysis and optimization by developers. For example, the system has three debugging cycles, namely initial debugging, mid-term debugging, and late debugging. In the early debugging stage, the system can set a higher bias coefficient threshold, such as 0.9, to ensure that only those data sets with very obvious abnormal jumps are marked. After entering the mid-term debugging stage, the system can reduce the threshold to 0.7 to expand the range of marked data sets and capture more abnormal situations. In the late debugging stage, the threshold can be further reduced to 0.5 so that the system can pay attention to more details and potential problems.

[0166] In this way, the text plan generation system can flexibly adjust the bias coefficient threshold according to the different needs of the debugging cycle, so as to more effectively use the abnormal jump thermal distribution in the historical meteorological monitoring log to mark the key meteorological monitoring data set. This not only improves the pertinence and efficiency of data processing, but also provides a more accurate data basis for subsequent data analysis and model training. It can be seen that by dynamically adjusting the bias coefficient threshold according to the debugging cycle, the text plan generation system can more flexibly respond to different debugging needs and improve the accuracy and efficiency of data marking. This technical solution provides strong support for the development and optimization of the system, and helps to improve the performance and stability of the entire system.

[0167] In an optional embodiment, the historical meteorological monitoring log is a historical meteorological monitoring message, and the method further includes: filtering a first number of first meteorological monitoring messages to be processed from the historical meteorological monitoring message stream according to a filtering rule; disassembling the first number of the first meteorological monitoring messages to be processed into multiple meteorological monitoring message sets, and filtering a second number of second meteorological monitoring messages to be processed from each meteorological monitoring message set; determining the filtered second meteorological monitoring messages to be processed as the historical meteorological monitoring messages, and preprocessing the historical meteorological monitoring messages, wherein the preprocessed meteorological monitoring message information matches the format conditions of the first meteorological state identification algorithm and the second meteorological state identification algorithm.

[0168] In an optional embodiment of the text plan generation system, when the historical meteorological monitoring log exists in the form of historical meteorological monitoring messages, the system will take a series of steps to process these messages and ultimately convert them into a data format suitable for processing by the first meteorological state recognition algorithm and the second meteorological state recognition algorithm.

[0169] First, the system will filter out the first number of first meteorological monitoring messages to be processed from the historical meteorological monitoring message stream according to the preset screening rules. The "first number" in the embodiment of the present application refers to a quantity value determined by the system according to the amount of data to be processed or a specific screening condition. The screening rules may include multiple conditions such as time range, geographical location, meteorological parameters, etc. to ensure that the selected messages meet the needs of subsequent processing.

[0170] Next, the system will disassemble the first meteorological monitoring message to be processed into multiple meteorological monitoring message sets. The purpose of this step is to group a large number of messages for subsequent refined processing. The disassembly can be based on the attributes of the message such as the sending time and monitoring site.

[0171] Then, the system will filter out the second number of second to-be-processed meteorological monitoring messages from each meteorological monitoring message set. The "second number" in the embodiment of the present application is also a number determined according to processing requirements or screening conditions. The screening of this step can be more refined to ensure that the selected message set is more representative or contains more key information.

[0172] Finally, the system determines the second meteorological monitoring messages to be processed as historical meteorological monitoring messages and preprocesses these messages. The purpose of preprocessing is to convert the data format into a format that matches the first meteorological state recognition algorithm and the second meteorological state recognition algorithm. This may involve operations such as data cleaning, conversion, and standardization to ensure that subsequent algorithms can correctly identify and analyze these data.

[0173] Through this processing method, the text plan generation system can efficiently process a large number of historical meteorological monitoring messages, extract key information, and convert it into a format suitable for algorithm analysis. This not only improves the efficiency of data processing, but also ensures the accuracy and reliability of subsequent analysis. In this way, through a series of steps such as screening, disassembly, re-screening, and preprocessing, the text plan generation system can effectively extract key data from historical meteorological monitoring messages and convert it into a format suitable for algorithm analysis. This not only optimizes the data processing process, but also provides a high-quality data foundation for subsequent meteorological status identification and analysis.

[0174] In some preferred embodiments, the algorithm weight optimization strategy of the second meteorological state recognition algorithm is a soft threshold variable optimization strategy, and the state recognition consistency error is determined based on the meteorological state vector mining information of the historical meteorological monitoring log by the first meteorological state recognition algorithm and the second meteorological state recognition algorithm, including: passing the historical meteorological monitoring log into the first meteorological state recognition algorithm to obtain a first historical meteorological state vector; passing the historical meteorological monitoring log into the second meteorological state recognition algorithm to obtain a second historical meteorological state vector; obtaining a meteorological state vector debugging example from a debugging data set, wherein the meteorological state vector debugging example is a meteorological state vector obtained by the second meteorological state recognition algorithm mining the meteorological state vector of the remaining historical meteorological monitoring logs, and the debugging data set is continuously adjusted during the pre-debugging process; using the first historical meteorological state vector, the second historical meteorological state vector and the meteorological state vector debugging example, the state recognition consistency error is determined, wherein the state recognition consistency error has a first quantitative relationship with the quantitative difference between the first historical meteorological state vector and the second historical meteorological state vector, and has a second quantitative relationship with the quantitative difference between the first historical meteorological state vector and the meteorological state vector debugging example.

[0175] In some preferred embodiments, the text plan generation system adopts a specific algorithm weight optimization strategy: soft threshold variable optimization strategy to optimize the performance of the second meteorological state recognition algorithm. This strategy allows the algorithm to adjust the weight more flexibly when processing data to adapt to different meteorological conditions and data characteristics, thereby improving the accuracy of state recognition.

[0176] In order to determine the consistency error of state identification, the system first passes the historical meteorological monitoring log into the first meteorological state identification algorithm and the second meteorological state identification algorithm respectively. Through the processing of the first algorithm, the system obtains the first historical meteorological state vector, which describes a representation of the meteorological state in the log. Similarly, through the processing of the second algorithm, the system obtains the second historical meteorological state vector, which is another state description of the same log.

[0177] In addition, the system also obtains meteorological state vector debugging examples from the debugging data set. These examples are the results obtained by the second meteorological state recognition algorithm mining the meteorological state vectors of the remaining historical meteorological monitoring logs, and they are continuously adjusted during the pre-commissioning process to ensure their accuracy and representativeness.

[0178] Next, the system uses the first historical meteorological state vector, the second historical meteorological state vector, and the meteorological state vector debugging example to determine the state recognition consistency error. This error reflects the degree of consistency of the results of the two algorithms when identifying the same historical meteorological monitoring log. Specifically, the state recognition consistency error has a first quantitative relationship with the quantitative difference between the first historical meteorological state vector and the second historical meteorological state vector, which indicates the difference in state recognition between the two algorithms. At the same time, the error has a second quantitative relationship with the quantitative difference between the first historical meteorological state vector and the meteorological state vector debugging example, which reflects the degree of conformity between the algorithm recognition result and the debugging example.

[0179] In this way, the text plan generation system can accurately evaluate the differences and consistency of the two meteorological state recognition algorithms when processing the same data, thus providing an important basis for optimizing algorithm performance and improving recognition accuracy. It can be seen that by adopting the soft threshold variable optimization strategy and the accurate state recognition consistency error evaluation method, the text plan generation system can more effectively use the historical meteorological monitoring log for state recognition, which not only improves the accuracy of recognition, but also provides a solid foundation for subsequent meteorological analysis and plan generation. This technical solution significantly enhances the system's adaptability to complex and changeable meteorological conditions, and brings substantial improvements to meteorological monitoring and early warning work.

[0180] It is worth mentioning that the soft threshold variable optimization strategy is a flexible algorithm adjustment method, which allows the threshold of the algorithm to be dynamically adjusted according to the characteristics of the data. In meteorological state recognition, this strategy can help the algorithm to more accurately capture the characteristics of meteorological data, thereby improving the accuracy of recognition. In an embodiment of the present application, the soft threshold variable optimization strategy is applied to the second meteorological state recognition algorithm, and by dynamically adjusting the threshold of the algorithm, the algorithm has better adaptability and accuracy when processing different meteorological data.

[0181] The first historical meteorological state vector and the second historical meteorological state vector are the results obtained after processing the historical meteorological monitoring logs by the first meteorological state recognition algorithm and the second meteorological state recognition algorithm, respectively. They are both quantitative descriptions of meteorological states, but they may differ due to different algorithms. By comparing these two vectors, the accuracy and consistency of the two algorithms in meteorological state recognition can be evaluated. At the same time, these two vectors are also the basis for subsequent calculation of state recognition consistency errors.

[0182] In order to more intuitively introduce the first historical meteorological state vector and the second historical meteorological state vector, a specific example may be used for illustration.

[0183] For example, there is a historical meteorological monitoring log that records multiple meteorological parameters such as wind speed, wind direction, temperature, humidity, etc. in a certain area. This log is input into the first meteorological state recognition algorithm and the second meteorological state recognition algorithm respectively.

[0184] When the historical meteorological monitoring log is input into the first meteorological state recognition algorithm, the algorithm will analyze the meteorological parameters in the log and generate a numerical feature vector to describe the current meteorological state. For example, this vector can contain the quantized values ​​of meteorological parameters such as wind speed, wind direction, temperature, and humidity. For example, the first historical meteorological state vector is: [3.5, 180, 22, 60]. This vector represents:

[0185] The wind speed is 3.5 m / s;

[0186] The wind direction is 180 degrees (i.e. south wind);

[0187] The temperature is 22 degrees Celsius;

[0188] The humidity is 60%.

[0189] Similarly, when the same historical meteorological monitoring log is input into the second meteorological state recognition algorithm, the algorithm will also generate a numerical feature vector to describe the meteorological state. Due to different algorithms, this vector may be different from the first historical meteorological state vector.

[0190] For example, the second historical meteorological state vector is: [3.8, 175, 23, 58]. This vector represents:

[0191] The wind speed is 3.8 m / s;

[0192] The wind direction is 175 degrees (i.e. southerly, but with a slight deviation);

[0193] The temperature is 23 degrees Celsius;

[0194] The humidity is 58%.

[0195] By comparing these two vectors, we can see the differences between the first meteorological state recognition algorithm and the second meteorological state recognition algorithm when processing the same data. These differences can be caused by different algorithm designs, parameter selections, or ways of interpreting data. In order to evaluate the accuracy and consistency of the two algorithms, we can further calculate the state recognition consistency error between them, thereby guiding the improvement and optimization of the algorithms.

[0196] The first quantitative relationship refers to the negative correlation between the quantitative difference between the first historical meteorological state vector and the second historical meteorological state vector and the state recognition consistency error. In simple terms, the greater the difference between the meteorological state vectors obtained by the two algorithms, the greater the state recognition consistency error. This negative correlation reminds us that when optimizing the algorithm, we should minimize the difference between the two algorithms to improve the accuracy of state recognition.

[0197] The second quantitative relationship refers to the positive correlation between the quantitative similarity between the first historical meteorological state vector and the meteorological state vector debugging example and the state recognition consistency error. In other words, the more similar the first historical meteorological state vector is to the debugging example, the smaller the consistency error of state recognition is. This positive correlation shows that the accuracy of state recognition can be improved by adjusting the algorithm to make it closer to the debugging example. At the same time, this also provides a method for evaluating and optimizing algorithm performance.

[0198] In summary, the soft threshold variable optimization strategy, the first historical meteorological state vector, the second historical meteorological state vector and the two quantitative relationships are mutually influential and interrelated. They together constitute the core invention of the embodiment of the present application and provide strong support for improving the accuracy of meteorological state recognition.

[0199] In other embodiments, the use of the disaster prevention and reduction monitoring matching error between the disaster emergency response plan and the historical meteorological monitoring log, and the state identification consistency error, to debug the first meteorological state identification algorithm includes: determining, according to the current debugging cycle, the plan matching confidence feature corresponding to the disaster prevention and reduction monitoring matching error used in the current debugging cycle, and the consistency confidence feature corresponding to the state identification consistency error, wherein the consistency confidence feature used in the debugging cycle with a higher sorting priority is greater than the consistency confidence feature used in the debugging cycle with a lower sorting priority, and the plan matching confidence feature used in the debugging cycle with a lower sorting priority is greater than the plan matching confidence feature used in the debugging cycle with a higher sorting priority; determining the global training error using the disaster prevention and reduction monitoring matching error, the plan matching confidence feature, the state identification consistency error, and the consistency confidence feature; and debugging the first meteorological state identification algorithm using the global training error.

[0200] In other embodiments, the text plan generation system uses the disaster prevention and reduction monitoring matching error between the disaster emergency response plan and the historical meteorological monitoring log, as well as the state recognition consistency error, to debug the first meteorological state recognition algorithm. This process is to improve the accuracy of the algorithm in identifying meteorological states, so as to better support disaster prevention and reduction work.

[0201] First, the system will determine the plan matching confidence features corresponding to the disaster prevention and reduction monitoring matching errors and the consistency confidence features corresponding to the state identification consistency errors used in the current debugging cycle. It should be explained here that different debugging cycles will use different confidence features, and the weights of the confidence features will change with the sorting priority of the debugging cycle. Specifically, the debugging cycle with a higher sorting priority will pay more attention to the consistency confidence features, that is, it will pay more attention to the consistency of the two algorithms in state identification; while the debugging cycle with a lower sorting priority will pay more attention to the plan matching confidence features, that is, it will pay more attention to the degree of matching between the algorithm identification results and the plan.

[0202] Next, the system will use the disaster prevention and mitigation monitoring matching error, plan matching confidence features, state recognition consistency error and consistency confidence features to determine the global training error. This global training error is an error indicator that combines multiple factors and reflects the performance of the algorithm in multiple aspects.

[0203] Finally, the system will debug the first meteorological state recognition algorithm based on this global training error. The goal of debugging is to reduce the global training error, thereby improving the accuracy and stability of the algorithm in meteorological state recognition. Specific debugging methods may include adjusting the parameters of the algorithm, optimizing the structure of the algorithm, etc.

[0204] In this way, the text plan generation system can continuously optimize the performance of the first meteorological state recognition algorithm, so that it can better adapt to complex and changeable meteorological conditions, and provide more accurate and reliable support for disaster prevention and reduction work. In this way, by comprehensively considering the disaster prevention and reduction monitoring matching error and the state recognition consistency error, and adjusting the weight of the confidence feature according to the debugging cycle, the text plan generation system can effectively debug and optimize the performance of the first meteorological state recognition algorithm. This not only improves the accuracy of algorithm recognition, but also provides a more solid technical foundation for subsequent disaster warning and emergency plan generation. This technical solution significantly enhances the system's adaptability to complex meteorological conditions and brings substantial improvements to disaster prevention and reduction work.

[0205] It is worth mentioning that the calculation formula of the global training error can be designed according to the actual application scenario and requirements, but it usually involves a weighted combination of disaster prevention and mitigation monitoring matching error, plan matching confidence features, state recognition consistency error and consistency confidence features. The following is an exemplary global training error calculation formula:

[0206] (E_{global}=w1*E_{match}*F_{plan}+w2*E_{consistency}*F_{consistency});

[0207] in:

[0208] (E_{global}) is the global training error.

[0209] (E_{match}) is the disaster prevention and mitigation monitoring matching error, which indicates the mismatch between the disaster emergency response plan and the historical meteorological monitoring log.

[0210] (F_{plan}) is the plan matching confidence feature, which indicates the importance attached to the plan matching error in different debugging cycles.

[0211] (E_{consistency}) is the state recognition consistency error, which indicates the inconsistency of the results of the first meteorological state recognition algorithm and the second meteorological state recognition algorithm when recognizing the same historical meteorological monitoring log.

[0212] (F_{consistency}) is the consistency confidence feature, which indicates the importance attached to the consistency error of state identification in different debugging cycles.

[0213] (w1) and (w2) are weighted coefficients used to balance the contribution of disaster prevention and mitigation monitoring matching error and state recognition consistency error in the global training error. These coefficients can be adjusted according to specific needs and experience.

[0214] In the above formula, the plan matching confidence feature and consistency confidence feature are adjusted according to the sorting priority of the debugging cycle. The debugging cycle with a higher sorting priority will increase the weight of the consistency confidence feature (i.e., increase (F_{consistency})), while the debugging cycle with a lower sorting priority will increase the weight of the plan matching confidence feature (i.e., increase (F_{plan})).

[0215] By calculating the global training error (E_{global}), the text plan generation system can quantitatively evaluate the performance of the first meteorological state recognition algorithm and debug and optimize it accordingly. This global training error not only takes into account the matching degree between the algorithm recognition result and the plan, but also takes into account the consistency between algorithms, thus more comprehensively reflecting the performance of the algorithm.

[0216] By applying the above embodiment, a global training error (E_{global}) is innovatively proposed. This indicator can comprehensively reflect the influence of multiple key factors, including disaster prevention and mitigation monitoring matching error (E_{match}), plan matching confidence feature (F_{plan}), state recognition consistency error (E_{consistency}) and consistency confidence feature (F_{consistency}). Through this comprehensive approach, the performance of the meteorological state recognition algorithm can be more comprehensively evaluated and optimized.

[0217] The weighted coefficients (w1) and (w2) in the formula, as well as the plan matching confidence features and consistency confidence features (F_{plan}) and (F_{consistency}), can be dynamically adjusted according to the sorting priority of the debugging cycle. This dynamism enables the algorithm to focus on different performance indicators at different debugging stages, thereby achieving more refined optimization.

[0218] In addition, the above formula not only considers the matching degree between the algorithm recognition result and the plan (reflected by (E_{match}) and (F_{plan})), but also considers the consistency between algorithms (reflected by (E_{consistency}) and (F_{consistency})). This dual consideration makes the evaluation more comprehensive and provides more flexibility for algorithm optimization.

[0219] By calculating the global training error (E_{global}), the text plan generation system can quantitatively evaluate the performance of the first meteorological state recognition algorithm. This quantitative evaluation is not only objective and repeatable, but also provides a clear direction and goal for subsequent debugging and optimization.

[0220] In addition, the above formula can be designed and adjusted according to the actual application scenarios and requirements, and the weighting coefficients and confidence features can be fine-tuned according to the specific situation to adapt to different application environments and demand changes. At the same time, the formula also has a certain degree of scalability, and other related evaluation indicators can be easily introduced to further enrich and improve the evaluation system.

[0221] In summary, the calculation formula of the global training error demonstrates its innovation in terms of comprehensive consideration of multiple key factors, dynamic adjustment of weights, practicality and flexibility, quantitative evaluation and optimization, adaptability and scalability, etc.

[0222] Further, Figure 2 Schematic diagram of the structure of a text plan generation system 200 provided in an embodiment of the present application. Figure 2 The text plan generation system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0223] Alternatively, if Figure 2 As shown, the text plan generation system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0224] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0225] Alternatively, if Figure 2 As shown, the text plan generation system 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, may send information or data to other devices, or receive information or data sent by other devices.

[0226] Optionally, the text plan generation system 200 can implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0227] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by the hardware integrated logic circuit or software instructions in the processor. The above processor can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams 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 the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor are combined and performed. The software module can be located in a mature storage medium in the field such as 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 is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0228] It can be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0229] It should be understood that the above-mentioned memory is exemplary but not restrictive. For example, the memory in the embodiments of the present application may also be static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM), etc. That is to say, the memory in the embodiments of the present application is intended to include but not limited to these and any other suitable types of memory.

[0230] Based on the above, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program implements the above method when running.

[0231] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations.

Claims

1. A method for generating meteorological disaster prevention and mitigation plans based on big data, characterized in that: Applied to a text plan generation system, the method comprises: According to the meteorological state vector mining information of the historical meteorological monitoring log by the first meteorological state recognition algorithm and the second meteorological state recognition algorithm, the state recognition consistency error is determined, and there are differences in the algorithm weight optimization strategies of the first meteorological state recognition algorithm and the second meteorological state recognition algorithm; the first meteorological state recognition algorithm is a deep learning model based on the long short-term memory network LSTM, and the second meteorological state recognition algorithm is a convolutional neural network CNN. The state recognition consistency error refers to the degree of difference between the recognition results of the first meteorological state recognition algorithm and the second meteorological state recognition algorithm when the historical meteorological monitoring log is analyzed by using the first meteorological state recognition algorithm and the second meteorological state recognition algorithm; Wherein, the algorithm weight optimization strategy of the second meteorological state recognition algorithm is a soft threshold variable optimization strategy, and the state recognition consistency error is determined according to the meteorological state vector mining information of the historical meteorological monitoring log by the first meteorological state recognition algorithm and the second meteorological state recognition algorithm, including: passing the historical meteorological monitoring log to the first meteorological state recognition algorithm to obtain the first historical meteorological state vector; passing the historical meteorological monitoring log to the second meteorological state recognition algorithm to obtain the second historical meteorological state vector; obtaining a meteorological state vector debugging example from a debugging data set, wherein the meteorological state vector debugging example is a meteorological state vector obtained by the second meteorological state recognition algorithm mining the meteorological state vector of the remaining historical meteorological monitoring log, and the debugging data set is continuously adjusted during the pre-debugging process; using the first historical meteorological state vector, the second historical meteorological state vector and the meteorological state vector debugging example, the state recognition consistency error is determined, wherein the state recognition consistency error has a first quantitative relationship with the quantitative difference between the first historical meteorological state vector and the second historical meteorological state vector, and has a second quantitative relationship with the quantitative difference between the first historical meteorological state vector and the meteorological state vector debugging example; Using the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log, the meteorological monitoring data set in the historical meteorological monitoring log is marked to obtain a meteorological monitoring annotated data set, wherein the abnormal jump thermal distribution is used to reflect the abnormal jump bias coefficients of different meteorological monitoring data sets in the historical meteorological monitoring log; the abnormal jump thermal distribution refers to the abnormal, large or sudden numerical changes of each meteorological data point in the time series in the historical meteorological monitoring log; the meteorological monitoring annotated data set is a data set obtained by additionally annotating the original meteorological monitoring data set, and the annotation includes the identification, classification and positioning of abnormalities or important events in the data; The step of determining the abnormal jump thermal distribution includes at least one of the following: Determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log by using the potential abnormal jump dense semantics obtained by mining the historical meteorological monitoring log using the abnormal jump mining branch in the second meteorological state recognition algorithm; Determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log by using the potential abnormal jump dense semantics obtained by mining the historical meteorological monitoring log using the abnormal jump mining branch in the first meteorological state recognition algorithm; The second meteorological state recognition algorithm includes an x-order abnormal jump mining branch, where x is a positive integer; the use of the abnormal jump mining branch in the second meteorological state recognition algorithm to mine the historical meteorological monitoring log to obtain the potential abnormal jump dense semantics to determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log, including: for each order of abnormal jump mining branch in the x-order abnormal jump mining branch, using the abnormal jump mining branch to mine the historical meteorological monitoring log to obtain the potential abnormal jump dense semantics, determine the local abnormal jump thermal relationship network corresponding to the abnormal jump mining branch; integrate the x local abnormal jump thermal relationship networks corresponding to the x-order abnormal jump mining branch to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log; The method of using the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log to mark the meteorological monitoring data set in the historical meteorological monitoring log to obtain the meteorological monitoring annotated data set includes: marking the target meteorological monitoring data set on the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the bias coefficient threshold to obtain the meteorological monitoring annotated data set; Wherein, on the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the bias coefficient threshold, the target meteorological monitoring data set is marked to obtain the meteorological monitoring annotated data set, including: obtaining a marking guidance feature, the marking guidance feature is used to locate the meteorological monitoring data set to be marked that meets the marking condition in the historical meteorological monitoring log; on the basis that the abnormal jump bias coefficient corresponding to the target meteorological monitoring data set in the abnormal jump thermal distribution is greater than the bias coefficient threshold, and the target meteorological monitoring data set belongs to the meteorological monitoring data set to be marked indicated by the marking guidance feature, the target meteorological monitoring data set is marked to obtain the meteorological monitoring annotated data set; Wherein, before the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log is used to mark the meteorological monitoring data set in the historical meteorological monitoring log to obtain the meteorological monitoring annotated data set, the method further includes: determining the bias coefficient threshold used in the current debugging cycle according to the current debugging cycle, wherein the bias coefficient threshold used in the debugging cycle with a higher sorting priority is greater than the bias coefficient threshold used in the debugging cycle with a lower sorting priority; Processing the meteorological monitoring annotated data set by using the first meteorological state recognition algorithm and the emergency plan generation algorithm to obtain a disaster emergency response plan; The first meteorological state recognition algorithm is debugged by using the disaster prevention and reduction monitoring matching error between the disaster emergency response plan and the historical meteorological monitoring log, as well as the state recognition consistency error.

2. The method according to claim 1, characterized in that For each order abnormal jump mining branch in the x-order abnormal jump mining branch, the potential abnormal jump dense semantics obtained by mining the historical meteorological monitoring log by the abnormal jump mining branch are used to determine the local abnormal jump thermal relationship network corresponding to the abnormal jump mining branch, including: Obtaining the u-th state jump dense semantics obtained by mining the historical meteorological monitoring logs by the u-th order abnormal jump mining branch in the second meteorological state recognition algorithm, where u is less than x; Based on the semantic focus adjustment result of the u-th state jump dense semantics, obtaining the u+1-th abnormal jump bias coefficient of each meteorological monitoring data set; Based on the u+1th abnormal jump bias coefficient of each meteorological monitoring data set, generating the local abnormal jump thermal relationship network corresponding to the u+1th order abnormal jump mining branch; The integration of the x local abnormal jump thermal relationship networks corresponding to the x-order abnormal jump mining branch to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log includes: summing the abnormal jump bias coefficients corresponding to the same meteorological monitoring data set in different local abnormal jump thermal relationship networks to obtain the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log.

3. The method according to claim 1, characterized in that On the basis that the abnormal jump thermal distribution is determined based on two ideas, the method includes: On the basis that the current debugging cycle belongs to the first debugging cycle, the abnormal jump mining branch in the first meteorological state recognition algorithm is used to mine the historical meteorological monitoring log to obtain the dense semantics of the potential abnormal jump, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log; On the basis that the current debugging cycle belongs to the second debugging cycle, the abnormal jump mining branch in the second meteorological state recognition algorithm is used to mine the historical meteorological monitoring log to obtain the dense semantics of the potential abnormal jump, and determine the abnormal jump thermal distribution corresponding to the historical meteorological monitoring log; The first debugging cycle and the second debugging cycle are debugging cycles that are cyclically performed during the debugging process.

4. The method according to any one of claims 1 to 3, characterized in that: The historical meteorological monitoring log is a historical meteorological monitoring message, and the method further includes: Filter the first number of first to-be-processed meteorological monitoring messages from the historical meteorological monitoring message stream according to the filtering rule; Disassembling the first to-be-processed meteorological monitoring messages of the first number into a plurality of meteorological monitoring message sets, and respectively selecting the second to-be-processed meteorological monitoring messages of the second number from each meteorological monitoring message set; The screened second meteorological monitoring message to be processed is determined as the historical meteorological monitoring message, and the historical meteorological monitoring message is preprocessed, wherein the preprocessed meteorological monitoring message information matches the format conditions of the first meteorological state identification algorithm and the second meteorological state identification algorithm.

5. The method according to any one of claims 1 to 3, characterized in that: The method of debugging the first meteorological state recognition algorithm by using the disaster prevention and reduction monitoring matching error between the disaster emergency response plan and the historical meteorological monitoring log, and the state recognition consistency error, comprises: According to the current debugging cycle, determine the plan matching confidence feature corresponding to the disaster prevention and reduction monitoring matching error and the consistency confidence feature corresponding to the state identification consistency error used in the current debugging cycle, wherein the consistency confidence feature used in the debugging cycle with a higher sorting priority is greater than the consistency confidence feature used in the debugging cycle with a lower sorting priority, and the plan matching confidence feature used in the debugging cycle with a lower sorting priority is greater than the plan matching confidence feature used in the debugging cycle with a higher sorting priority; Determine a global training error using the disaster prevention and reduction monitoring matching error, the plan matching confidence feature, the state identification consistency error, and the consistency confidence feature; The first meteorological state recognition algorithm is debugged using the global training error.

6. A text plan generation system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 5.

Images