Urban Subway Flood Early Warning System Based on Knowledge Domain Language Model

Abstract

The application provides a city subway water disaster early warning system based on a knowledge field language model, which has the following characteristics: a data acquisition module for acquiring real-time data of subway pipe networks and surrounding environments; a preprocessing module for preprocessing the real-time data to obtain preprocessed data; a prediction module for generating a prediction result of subway water disasters according to the preprocessed data; a warning judgment module for judging whether to generate a warning signal according to the prediction result and real-time water levels in the real-time data, and if yes, generating a corresponding level of warning signal, and if no, no warning; a warning text generation module including a knowledge field language model for generating water disaster warning information corresponding to different objects according to the real-time data, the warning signal and the prediction result; and an information sending module for sending the water disaster warning information to the corresponding objects. In summary, the method can automatically warn subway water disasters and generate water disaster warning information for each group of people.

Description

Technical Field

[0001] This invention relates to an early warning system, specifically to an urban subway flood early warning system based on a knowledge domain language model. Background Technology

[0002] In the rapid process of urbanization, the subway system, as the backbone of urban public transportation, is crucial for the normal functioning of city life. However, with the increasing frequency of extreme weather events, such as torrential rains and floods, urban subway networks face an increasingly severe threat of flooding. As underground transportation facilities, subways operate in a relatively enclosed environment. Once flooded, not only may trains be suspended and passengers trapped, but irreversible damage to subway facilities may also occur, seriously affecting urban public transportation order and the safety of residents' lives and property.

[0003] Traditional subway flood warning methods mainly rely on manual patrols, fixed-point sensor monitoring, and experience-based judgment. These methods have significant limitations. Manual patrols are limited by time and manpower costs, making it difficult to achieve all-weather, all-round monitoring. While fixed-point sensor monitoring can acquire data for some areas, its monitoring range is limited, and the data update speed may not keep up with the rapid changes in water accumulation under extreme weather conditions. Experience-based judgment often relies on historical data and experience, lacking real-time analysis support, making it difficult to accurately predict flood risks. Historical rainfall and flood information is often voluminous, difficult to automatically identify, and requires manual processing. Even after processing in the traditional model, the data still needs to be manually converted into readable information by professionals.

[0004] Therefore, existing subway flood warning methods still have problems such as untimely warnings and the inability to automatically generate warning information. Summary of the Invention

[0005] This invention is made to solve the above-mentioned problems, and its purpose is to provide an urban subway flood early warning system based on a knowledge domain language model.

[0006] This invention provides an urban subway flood early warning system based on a knowledge domain language model, characterized by the following features: a data acquisition module for collecting real-time data of the subway network and surrounding environment; a preprocessing module for preprocessing the real-time data to obtain preprocessed data; a prediction module for generating a subway flood prediction result based on the preprocessed data; an early warning judgment module for determining whether to generate an early warning signal based on the prediction result and the real-time water level in the real-time data; if so, generating an early warning signal of the corresponding level; otherwise, not issuing an early warning; an early warning text generation module, including a knowledge domain language model, for generating flood early warning information corresponding to different objects based on the real-time data, the early warning signal, and the prediction result; and an information sending module for sending the flood early warning information to the corresponding objects, wherein the real-time data includes real-time water level, rainfall data, and water accumulation images.

[0007] The urban subway flood early warning system based on a knowledge domain language model provided by this invention may also have the following features: the prediction result includes the future rainfall intensity, the early warning judgment module stores the warning water level, the guaranteed water level, the first rainfall intensity, and the second rainfall intensity, and the early warning judgment module generates the early warning signal according to the prediction result and water level data in the following way: when the real-time water level is lower than the warning water level and the future rainfall intensity is less than the first rainfall intensity, the flood warning information is "passing is permitted"; when the real-time water level is higher than the warning water level and lower than the guaranteed water level, or the future rainfall intensity is greater than the first rainfall intensity and less than the second rainfall intensity, the flood warning information is "passing with caution"; when the real-time water level is higher than the guaranteed water level, or the future rainfall intensity is greater than the second rainfall intensity, the flood warning information is "passing is prohibited".

[0008] The urban subway flood early warning system based on a knowledge domain language model provided by this invention may also have the following features: the prediction results include the predicted range of subway floods, the flood warning information includes the predicted range of subway floods, the flood warning information and suggested measures, and the objects are divided by user group characteristics, including the subway flood risk level.

[0009] The urban subway flood early warning system based on a knowledge domain language model provided by this invention may also have the following features: the information sending module sends flood early warning information wirelessly and via wired means, including broadcasting, television, the Internet and SMS.

[0010] The urban subway flood early warning system based on a knowledge domain language model provided by this invention may also have the following features: the process of training the knowledge domain language model includes the following steps: Step S1, constructing a training dataset based on water depth data, drainage system data, subway line station data, historical early warning information, and historical impact range; Step S2, using an existing pre-trained large language model as the initial model; Step S3, fine-tuning the initial model using the training dataset; Step S4, fine-tuning the initial model using text commands; Step S5, determining whether the initial model has reached the preset performance. If yes, the knowledge domain language model is obtained; otherwise, step S3 is executed.

[0011] The urban subway flood early warning system based on a knowledge domain language model provided by this invention may also have the following feature: in step S3, fine-tuning is performed using the LoRA method.

[0012] The urban subway flood early warning system based on a knowledge domain language model provided by this invention may also include the following features: a train early warning module, which stores tunnel line data and is used to obtain a train early warning result based on the tunnel line data, real-time water level, and predicted water level in the prediction results. The train early warning result includes whether the train can reach a station or not; and a safe position calculation module, which is used to calculate a safe position based on the real-time position of the train and the tunnel line data when the train early warning result is that the train cannot reach a station. The information sending module is also used to send the safe position to the train. When the train cannot reach a station, the train driver will drive the train to the safe position.

[0013] The role and effect of invention

[0014] According to the knowledge domain language model-based urban subway flood early warning system of the present invention, firstly, a prediction module generates prediction results based on real-time data; secondly, an early warning judgment module obtains corresponding level early warning signals based on multiple different levels of traffic judgment conditions, combined with the prediction results and real-time water levels; and thirdly, an early warning text generation module generates flood early warning information containing relevant information in different expression forms using the knowledge domain language model based on the early warning signals, prediction results, and implementation data, and sends it to the corresponding groups through the information sending module. Therefore, the knowledge domain language model-based urban subway flood early warning system of the present invention can automatically issue early warnings for subway floods and generate flood early warning information for various groups. Attached Figure Description

[0015] Figure 1 This is a block diagram of an urban subway flood early warning system according to an embodiment of the present invention;

[0016] Figure 2This is a schematic diagram of the process of constructing a knowledge domain language model in an embodiment of the present invention;

[0017] Figure 3 This is a schematic diagram of the subway flood warning process in an embodiment of the present invention. Detailed Implementation

[0018] To make the technical means, creative features, objectives and effects of this invention easier to understand, the following embodiments, in conjunction with the accompanying drawings, provide a detailed description of the urban subway flood early warning system based on a knowledge domain language model.

[0019] This embodiment provides an urban subway flood early warning system based on a knowledge domain language model, hereinafter referred to as the urban subway flood early warning system.

[0020] Figure 1 This is a block diagram of an urban subway flood early warning system according to an embodiment of the present invention.

[0021] like Figure 1 As shown, the urban subway flood early warning system 100 includes a data acquisition module 11, a preprocessing module 12, a prediction module 13, an early warning judgment module 14, an early warning text generation module 15, an information sending module 16, a train early warning module 17, a safe position calculation module 18, an emergency response module 19, and a control module 20 that controls the above modules.

[0022] The data acquisition module 11 is used to collect real-time data on the subway network and surrounding environment. This real-time data includes real-time water levels, rainfall data, and images of accumulated water.

[0023] In this embodiment, the data acquisition module 11 includes a variety of sensors such as a water level sensor, a rain gauge, a camera, a flow sensor, and a gas concentration sensor, and is installed in and around the subway pipeline network.

[0024] The preprocessing module 12 is used to preprocess real-time data to obtain preprocessed data. In this embodiment, preprocessing includes data cleaning, data transformation, and data format unification. Data cleaning removes invalid data such as noise and outliers. Data transformation converts text data into data information using a language model.

[0025] The prediction module 13 is used to generate prediction results for subway flooding based on preprocessed data. The prediction results include the future rainfall intensity and the predicted range of subway flooding.

[0026] The early warning judgment module 14 is used to determine whether to generate an early warning signal based on the prediction results and the real-time water level in the real-time data. If yes, an early warning signal of the corresponding level is generated; otherwise, no early warning is generated.

[0027] The early warning judgment module 14 stores the warning water level, the guaranteed water level, the first rainfall intensity, and the second rainfall intensity.

[0028] The early warning judgment module 14 generates early warning signals based on the prediction results and water level data in the following ways: when the real-time water level is lower than the warning water level and the future rainfall intensity is less than the first rainfall intensity, the flood warning information is "passing is permitted"; when the real-time water level is higher than the warning water level but lower than the guaranteed water level, or when the future rainfall intensity is greater than the first rainfall intensity but less than the second rainfall intensity, the flood warning information is "passing with caution"; when the real-time water level is higher than the guaranteed water level, or when the future rainfall intensity is greater than the second rainfall intensity, the flood warning information is "passing is prohibited".

[0029] The early warning text generation module 15 includes a knowledge domain language model, used to generate flood early warning information corresponding to different objects based on real-time data, early warning signals, and prediction results. The flood early warning information includes the predicted flood range for the subway, flood early warning information, and recommended measures. In this embodiment, the early warning signals, from low to high, are: yellow warning, orange warning, and red warning.

[0030] Figure 2 This is a schematic diagram of the process of constructing a knowledge domain language model in an embodiment of the present invention.

[0031] like Figure 2 As shown, the process of constructing a language model for a knowledge domain includes the following steps:

[0032] Step S1: Construct a training dataset based on water depth data, drainage system data, subway line and station data, historical early warning information, and historical impact range.

[0033] Step S2: Use the existing pre-trained large language model as the initial model. In this embodiment, the existing pre-trained large language model can generate corresponding text output based on the input text data.

[0034] Step S3: Fine-tune the initial model using the training dataset.

[0035] In step S3, the LoRA method is used for fine-tuning, thereby adapting the model to a specific task while keeping most of the initial model parameters unchanged, using a small number of trainable parameters. LoRA training parameters are configured, including learning rate, batch size, and number of training epochs, to ensure that the training process is both effective and efficient.

[0036] Step S4: Fine-tune the initial model using text instructions. In this embodiment, the text instructions are constructed based on guidance suggestions for different objects. Then, the output generated by the text instructions is analyzed based on the accuracy, readability, and relevance of the initial model to obtain analysis results. The initial model is then fine-tuned based on the analysis results.

[0037] Step S5: Determine whether the initial model has reached the preset performance. If yes, obtain the knowledge domain language model; otherwise, proceed to step S3.

[0038] In this embodiment, the flood warning information for different objects is a warning text that conforms to language habits and is easy to understand. Furthermore, the text description is adjusted for different objects, i.e. different groups, in order to improve the targeting and dissemination efficiency of the information.

[0039] The users are categorized based on their group characteristics, including the risk of subway flooding. In this embodiment, the user group is divided into two main categories based on the risk of subway flooding: emergency contact users and general contact users. Emergency contact users include subway emergency personnel and people on trains at risk of flooding, while general contact users include subway staff, ordinary passengers, and other personnel.

[0040] Furthermore, in this embodiment, the major categories of objects are further subdivided into multiple minor categories based on age, occupation, and cultural background, so that people in each minor category can obtain the most relevant and easily understood content based on the corresponding warning information text, thereby enabling them to carry out subway flood prevention measures in an orderly manner or to quickly implement subway flood emergency measures.

[0041] In other embodiments, user group characteristics can be set based on actual subway flood emergency response plans to segment the population and achieve subway flood prevention and emergency response.

[0042] The information sending module 16 is used to send flood warning information to corresponding recipients. The information sending module 16 sends flood warning information wirelessly and via wired means. Wireless methods include broadcasting, television, the internet, and SMS. In this embodiment, the information sending module 16 also sends flood warning information to display devices, such as LED screens, installed at key locations like subway stations and tunnel entrances, providing relevant information to pedestrians and vehicles near these locations.

[0043] The train warning module 17 stores tunnel line data and is used to obtain train warning results based on the tunnel line data, real-time water level and predicted water level in the prediction results. The train warning results include stations that the train can reach and stations that the train cannot reach.

[0044] The safe position calculation module 18 is used to calculate a safe position based on the train's real-time position and tunnel track data when the train warning result indicates that the train cannot reach the station. In this embodiment, the safe position calculation module 18 calculates a relatively safe and higher-altitude position that the train can reach based on algorithms such as gradient calculation, integral method, and dynamic simulation.

[0045] If the train cannot reach the station, the information sending module 16 sends a safe location information to the train, and the train driver then drives the train to the safe location. In this embodiment, the information sending module 16 generates rescue information based on the rescue request and notifies the passengers inside the train at the safe location of the rescue information.

[0046] The emergency response module 19 includes an emergency handling model built based on an existing knowledge domain language model. It is used to generate emergency rescue and disaster relief plans based on real-time data, early warning signals and prediction results, including specific measures in multiple aspects such as personnel evacuation, equipment repair and drainage operations.

[0047] The control module 20 stores the control program that controls the operation of each module.

[0048] The following description, in conjunction with the attached diagram, illustrates the process of using the Urban Subway Flood Early Warning System 100 for subway flood early warning.

[0049] Figure 3 This is a schematic diagram of the subway flood warning process in an embodiment of the present invention.

[0050] like Figure 3 As shown, the process of using the urban subway flood early warning system 100 to issue a subway flood early warning includes the following steps:

[0051] Step T1: Use data acquisition module 11 to collect real-time data of the subway pipeline network and surrounding environment.

[0052] Step T2: The real-time data is preprocessed using the preprocessing module 12 to obtain preprocessed data.

[0053] Step T3: The prediction module 13 generates the prediction results of the subway flood based on the preprocessed data.

[0054] In step T4, the early warning judgment module 14 determines whether to generate an early warning signal based on the prediction results and the real-time water level in the real-time data. If yes, an early warning signal of the corresponding level is generated; otherwise, no early warning is generated.

[0055] Step T5: The warning text generation module 15 generates flood warning information corresponding to different objects based on real-time data, warning signals and prediction results.

[0056] Step T6: The flood warning information is sent to the corresponding target using the information sending module 16.

[0057] The role and effect of the embodiments

[0058] According to the urban subway flood early warning system based on a knowledge domain language model involved in this embodiment, firstly, the prediction module generates prediction results based on real-time data; secondly, the early warning judgment module obtains corresponding level early warning signals based on multiple different levels of traffic judgment conditions, combined with the prediction results and real-time water levels; thirdly, the early warning text generation module generates flood early warning information containing relevant information in different expression forms based on the early warning signals, prediction results, and implementation data, using a knowledge domain language model, and sends it to the corresponding groups through the information sending module. In summary, this method can automatically provide early warnings of subway floods and generate flood early warning information for various groups.

[0059] Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A city subway flood early warning system based on a knowledge field language model, characterized in that, The system comprises: a data acquisition module for acquiring real-time data of a subway pipe network and a surrounding environment; a preprocessing module for preprocessing the real-time data to obtain preprocessed data; a prediction module for generating a prediction result of a subway water disaster based on the preprocessed data; a warning judgment module for judging whether to generate a warning signal based on the prediction result and real-time water level in the real-time data, and if so, generating a corresponding level of warning signal, and if not, no warning; a warning text generation module comprising a knowledge domain language model, for generating water disaster warning information corresponding to different objects based on the real-time data, the warning signal and the prediction result; an information sending module for sending the water disaster warning information to the corresponding object, a train warning module storing tunnel line data, for obtaining a train warning result based on the tunnel line data, the real-time water level and the predicted water level in the prediction result, the train warning result comprising a station that a train can reach and a station that the train cannot reach; a safe position calculation module for calculating a safe position based on the real-time position of the train and the tunnel line data when the train warning result is a station that the train cannot reach, wherein the information sending module is further configured to send the safe position to the train, when the train cannot reach the station, the driver of the train drives the train to the safe position, the real-time data comprises the real-time water level, rainfall data and water accumulation images, the process of training the knowledge domain language model comprises the following steps: Step S1: constructing a training data set based on water accumulation depth data, drainage system data, subway line station data, historical warning information and historical influence range; Step S2: taking an existing pre-trained large language model as an initial model; Step S3: fine-tuning the initial model through the training data set; Step S4: fine-tuning the initial model through text instructions; Step S5: judging whether the initial model reaches a preset performance, and if so, obtaining the knowledge domain language model, and if not, executing the step S3.

2. The urban subway water disaster warning system based on the knowledge domain language model according to claim 1, wherein: wherein the prediction result comprises future rainfall intensity, the warning judgment module stores warning water level, guarantee water level, first rainfall intensity and second rainfall intensity, the warning judgment module generates a warning signal based on the prediction result and the water level data in the following manner: when the real-time water level is lower than the warning water level and the future rainfall intensity is less than the first rainfall intensity, the water disaster warning information is for allowing passage; when the real-time water level is higher than the warning water level and less than the guarantee water level, or the future rainfall intensity is greater than the first rainfall intensity and less than the second rainfall intensity, the water disaster warning information is for cautious passage; when the real-time water level is higher than the guarantee water level, or the future rainfall intensity is greater than the second rainfall intensity, the water disaster warning information is for prohibiting passage.

3. The urban subway water disaster early warning system based on the knowledge field language model according to claim 1, wherein: the prediction result comprises a subway water disaster prediction range, wherein, the water disaster early warning information comprises the subway water disaster prediction range, the water disaster early warning information and the recommended measures, the objects are divided by user group characteristics, the user group characteristics comprise a subway water disaster risk degree.

4. The urban subway water disaster early warning system based on the knowledge field language model according to claim 1, wherein: the information sending module sends the water disaster early warning information through wireless and wired ways, wherein, the wireless way comprises broadcasting, television, the Internet and short message.

5. The urban subway water disaster early warning system based on the knowledge field language model according to claim 1, wherein: in the step S3, fine tuning is performed by a LoRA method. wherein ​

Images