Underground infrastructure flood evacuation analysis method and system based on large model

Abstract

The application discloses a kind of underground infrastructure flood evacuation analysis method and system based on large model, method includes: obtaining flood image information, through large model flood analysis mode training system The disaster situation description and risk level corresponding to the flood image information output by;Through the integration system of large model and digital twin platform, the data output by the large model flood analysis mode training system is input into the large model-digital twin real-time deduction system for calling;Through the large model-digital twin real-time deduction system, data deduction is carried out, and analysis result is output;The application proposes a new underground infrastructure flood evacuation analysis method, improves the decision efficiency of flood evacuation and emergency.

Description

Technical Field

[0001] This invention relates to the field of digital twin technology, and in particular to a method and system for analyzing flood evacuation of underground infrastructure based on a large model. Background Technology

[0002] Current methods for monitoring and assessing flood risks in underground facilities mainly focus on monitoring flood data based on the Internet of Things (IoT) and using various factor analysis methods to determine risk indicators to assess flood risks. However, in determining these risk indicators, only monitoring data of underground facilities on water conditions is used, without considering the impact of evacuation routes and the effectiveness of evacuation measures on these indicators.

[0003] Therefore, existing technologies still need improvement. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for analyzing flood evacuation of underground infrastructure based on a large model, in order to address the shortcomings of existing technologies and solve the technical problem that traditional flood analysis systems can only perform single-function water situation monitoring.

[0005] The technical solution adopted by this invention to solve the technical problem is as follows:

[0006] In a first aspect, the present invention provides a method for flood evacuation analysis of underground infrastructure based on a large model, comprising:

[0007] Acquire flood image information, and output the disaster description and risk level corresponding to the flood image information through a large-scale model flood analysis pattern training system;

[0008] The data output from the large model flood analysis model training system is input into the large model-digital twin real-time simulation system for use through the integrated system of large model and digital twin platform;

[0009] Data simulation is performed using the large-scale model-digital twin real-time simulation system, and analysis results are output.

[0010] In one implementation, acquiring flood image information and outputting the disaster description and risk level corresponding to the flood image information through a large-scale flood analysis model training system includes:

[0011] Acquire the flood disaster image information and perform object segmentation on the water, people, equipment, and underground structures in the flood disaster image information;

[0012] The large-scale flood analysis model training system describes the disaster situation of the segmented images and assigns them corresponding risk levels. The images are then grouped according to the risk levels and output as a dataset in the form of images and descriptions.

[0013] The datasets with different risk levels are scored, and the training system for the large model flood analysis mode is optimized by the score difference between the first preset description information and the second preset description information.

[0014] The optimized large-scale flood analysis model training system outputs an optimized disaster description and corresponding risk level, and the disaster description and risk level of the same flood image information are overwritten with the optimized disaster description and corresponding risk level.

[0015] In one implementation, acquiring flood image information and segmenting water, people, equipment, and underground structures within the flood image information includes:

[0016] Acquire flood video datasets and process them into slices according to different flood spread stages to obtain flood image information for the corresponding stages.

[0017] The flood image information and the preset partition prompts are input into the large model flood analysis mode training system;

[0018] By combining image segmentation masking technology with transfer learning on the flood disaster image information, object segmentation of water, people, equipment and underground structures in the flood disaster image information can be achieved.

[0019] In one implementation, the process of training the segmented images using the large-scale flood analysis model training system to describe the disaster situation and assign corresponding risk levels, grouping them according to the risk levels, and outputting the dataset in image-description format includes:

[0020] The segmented image is acquired, and the artificial water situation description corresponding to the segmented image is retrieved. The risk level corresponding to the artificial water situation description is obtained by using factor analysis.

[0021] The large-scale flood disaster analysis model training system is trained based on the artificial flood disaster description and the corresponding risk level, and the disaster description and risk level of all segmented images are output through the large-scale flood disaster analysis model training system.

[0022] The data is grouped according to the risk level, and output as a dataset in the form of an image-description, and stored in the database.

[0023] In one implementation, scoring datasets with different risk levels and optimizing the large-scale flood analysis model training system based on the score difference between a first preset descriptive information and a second preset descriptive information includes:

[0024] Retrieve datasets of the same risk level from the database, label the order of the disaster descriptions by partial order pairs, and assign an initial score to each set of image-description datasets by preset scoring prompts.

[0025] The large-scale flood analysis model training system is optimized by maximizing the difference between good and bad descriptions within the same risk level and automatically scoring them.

[0026] In one implementation, the step of performing data extrapolation through the large model-digital twin real-time extrapolation system and outputting analysis results includes:

[0027] The large model-digital twin real-time inference system standardizes the data input from the integrated system of the large model and the digital twin platform, encodes the standardized data, and uses a neural network to infer the data to obtain the inference results.

[0028] Determine whether the risk index derived from data analysis is higher than the critical value;

[0029] If the risk index is higher than the critical value, a warning message will be issued and an emergency plan will be output in conjunction with the disaster description of the large model-digital twin real-time simulation system.

[0030] If the risk index is not higher than the critical value, the flood image information of the next area will be input into the large model-digital twin real-time simulation system for analysis, and the analysis results will be output.

[0031] In one implementation, the data input to the integrated system of the large model and the digital twin platform is standardized through the large model-digital twin real-time inference system. The standardized data is then encoded and processed through a neural network to obtain the data inference result, including:

[0032] Each prediction index is set using a multi-dimensional vector on the standardized data, the standardized data is embedded in the multi-dimensional vector, the embedded multi-dimensional vector is encoded, and the encoded data is input into multiple neural network layers.

[0033] The encoded data undergoes residual connection and normalization, multi-dimensional correction of the attention model, secondary residual connection and normalization, forward propagation, and multi-round neuron loops with loss function verification feedback.

[0034] The data output by the neuron after looping is mapped to a preset interval, converted into a probability distribution, and the data vector with the highest probability is selected as the predicted data output.

[0035] The predicted data is input into the BIM model, weighted, and combined with the corresponding risk level to output the risk rating of the flood-prone area.

[0036] Secondly, the present invention also provides a large-scale model-based underground infrastructure flood evacuation analysis system, comprising:

[0037] The training module is used to acquire flood image information and, through the large-scale flood analysis model training system, output the disaster description and risk level corresponding to the flood image information.

[0038] The integration module is used to input the data output by the large model flood analysis mode training system into the large model-digital twin real-time simulation system for use through the integration system of the large model and digital twin platform;

[0039] The simulation module is used to perform data simulation through the large model-digital twin real-time simulation system and output analysis results.

[0040] In a second aspect, the present invention also provides a terminal, comprising: a processor and a memory, the memory storing a large-model-based underground infrastructure flood evacuation analysis program, which, when executed by the processor, is used to implement the operation of the large-model-based underground infrastructure flood evacuation analysis method as described in the first aspect.

[0041] Thirdly, the present invention also provides a computer-readable storage medium storing a large-model-based underground infrastructure flood evacuation analysis program, which, when executed by a processor, is used to implement the operation of the large-model-based underground infrastructure flood evacuation analysis method as described in the first aspect.

[0042] The present invention, by employing the above technical solution, has the following effects:

[0043] This invention acquires flood image information and outputs corresponding disaster descriptions and risk levels through a large-scale flood analysis model training system. This system learns and establishes a flood analysis model, enabling it to output a specific flood disaster description and risk level for each input flood image. Through an integrated system of the large-scale model and a digital twin platform, the data output from the large-scale model flood analysis model training system can be input into a large-scale model-digital twin real-time simulation system. Finally, the large-scale model-digital twin real-time simulation system performs data simulation and outputs analysis results. This system performs a series of processes, including region segmentation, image understanding, and granular control of region description, and then outputs disaster descriptions and risk levels for real-time flood images. This invention proposes a new method for analyzing flood evacuation of underground infrastructure, improving the efficiency of flood evacuation and emergency response decisions. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0045] Figure 1 This is a flowchart of a flood evacuation analysis method for underground infrastructure based on a large model, as described in one implementation of the present invention.

[0046] Figure 2 This is a schematic diagram of a system module in one implementation of the present invention.

[0047] Figure 3 This is a schematic diagram of the flood image deconstruction process in one implementation of the present invention.

[0048] Figure 4 This is an architecture diagram of a large model-digital twin real-time simulation system in one implementation of the present invention.

[0049] Figure 5 This is a flowchart of sensor data acquisition in one implementation of the present invention.

[0050] Figure 6 This is an overall roadmap for the large model-digital twin real-time inference system in one implementation of the present invention.

[0051] Figure 7 This is a simulation roadmap of the BIM module in one implementation of the present invention.

[0052] Figure 8This is a risk level display diagram in one implementation of the present invention.

[0053] Figure 9 This is a system technology roadmap in one implementation of the present invention.

[0054] Figure 10 This is a structural diagram of the system implementation in one embodiment of the present invention.

[0055] Figure 11 This is a functional schematic diagram of the terminal in one implementation of the present invention.

[0056] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0057] This invention provides a method, system, terminal, and storage medium for flood evacuation analysis of underground infrastructure based on a large model. To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0058] Exemplary methods

[0059] Current methods for monitoring and assessing the risks of flooding in underground infrastructure primarily focus on monitoring flood data based on the Internet of Things (IoT) and using various factor analyses to determine risk indicators. However, the determination of these risk indicators relies solely on monitoring data of water conditions within underground facilities, neglecting the impact of evacuation routes and the effectiveness of evacuation measures. Therefore, designing a new algorithm that integrates water condition monitoring and real-time evacuation data in the analysis of flood evacuation in underground infrastructure represents a new need in the monitoring and risk assessment of flooding in underground facilities.

[0060] To address the aforementioned technical problems, this invention provides a method for analyzing flood evacuation of underground infrastructure based on a large model. The aim is to propose a new method for analyzing flood evacuation of underground infrastructure and improve the efficiency of flood evacuation and emergency response decisions.

[0061] like Figure 1 As shown, this embodiment of the invention provides a method for flood evacuation analysis of underground infrastructure based on a large model, including the following steps:

[0062] Step S100: Obtain flood image information, and output the disaster description and risk level corresponding to the flood image information through the large model flood analysis mode training system.

[0063] In this embodiment, the large-model-based underground infrastructure flood evacuation analysis method is applied to a terminal, which includes, but is not limited to, devices such as computers and mobile terminals; the terminal is equipped with a training and transfer platform for the large-model-based underground infrastructure flood evacuation analysis model.

[0064] This embodiment proposes a method for analyzing flood evacuation in underground facilities. This method includes: a large-scale flood analysis pattern training system, an integration system of the large-scale model and a digital twin platform, and a large-scale model-digital twin real-time simulation system. The large-scale model flood analysis pattern training system includes a pre-training module, an instruction adjustment module, a reward scoring module, and a reinforcement learning module. The integration system of the large-scale model and the digital twin platform defines a specific port for connecting the large-scale model and the digital twin platform. The large-scale model-digital twin real-time simulation system inputs real-time monitoring images and equipment data into the large-scale model, calling upon the large-scale model's analytical capabilities to perform real-time simulations. This allows for the estimation of flood spread speed, potential future direction of spread, and the density of evacuated populations by analyzing flood scene images. Furthermore, by combining the BIM (Building Information Modeling) model of the affected area and statistical population evacuation data, further evacuation plans can be provided in real-time to assist decision-making, significantly improving the efficiency of flood evacuation and emergency response.

[0065] In this embodiment, the underground facility flood evacuation analysis system is mainly divided into 6 modules, such as... Figure 2 As shown, the modules are: data acquisition module, large model module, BIM module, trend analysis module, risk assessment module, and contingency plan analysis module, and can provide services to users through the application system.

[0066] Specifically, in one implementation of this embodiment, step S100 includes the following steps:

[0067] Step S101: Obtain the flood image information and perform object segmentation on the water, people, equipment and underground structures in the flood image information;

[0068] Step S102: The segmented images are described and assigned corresponding risk levels by the large model flood analysis mode training system. The images are then grouped according to the risk levels and output as a dataset in the form of images-descriptions.

[0069] Step S103: Score the datasets of different risk levels, and optimize the large model flood analysis mode training system by the score difference between the first preset description information and the second preset description information;

[0070] Step S104: Use the dataset with higher scores as the training set to further optimize the large model flood analysis mode training system, and output the disaster description and risk level corresponding to the flood image information.

[0071] In this embodiment, the large-scale model flood analysis pattern training system is a pre-analysis module used for real-time flood simulation. Its main function is to enable the large model to learn and establish a flood analysis pattern, so that when flood image information is input, the large model can output a description of the flood situation and risk level for that image. This training system includes a pre-training module, an instruction adjustment module, a reward scoring module, and a reinforcement learning module.

[0072] In this embodiment, the first preset description information represents a good description, and the second preset description information represents a poor description. A good description is grammatically correct, contains basic disaster information, highlights key population groups, and contains no irrelevant information, while a poor description is the opposite. The large-scale flood disaster analysis training system is optimized by maximizing the difference between good and poor descriptions.

[0073] Specifically, in one implementation of this embodiment, step S101 includes the following steps:

[0074] Step S101a: Obtain the flood video dataset, and process the slices according to different flood spread stages to obtain flood image information for the corresponding stage;

[0075] Step S101b: Input the flood image information and the preset partition prompt words into the large model flood analysis mode training system;

[0076] Step S101c: Combine image segmentation masking technology to perform transfer learning on the flood image information to achieve object segmentation of water, people, equipment and underground structures in the flood image information.

[0077] In this embodiment, the first step is to acquire a flood video dataset and preprocess it. Then, the preprocessed images are used to pre-train the large-scale flood analysis model training system, improving its image processing capabilities. Finally, the processed images are segmented and output.

[0078] In this embodiment, the pre-training module is the image preprocessing part of the training system. Its main function is to retrieve historical flood monitoring videos or flood monitoring videos of similar underground facilities stored in the digital twin platform from surveillance cameras at different heights in different areas of underground facilities, and process them into multiple image sequences according to different flood spread stages. This sequence, along with the partition prompt words, is input into the large model, and the flood images are transferred to the model using image segmentation masking technology, thereby achieving object segmentation of water, people, equipment, and underground structural facilities in the flood images.

[0079] Specifically, in one implementation of this embodiment, step S102 includes the following steps:

[0080] Step S102a: Obtain the segmented image, retrieve the artificial water disaster description corresponding to the segmented image, and use factor analysis to obtain the risk level corresponding to the artificial water disaster description;

[0081] Step S102b: Train the large-scale flood disaster analysis model training system according to the artificial flood disaster description and the corresponding risk level, and output the disaster description and risk level of all segmented images through the large-scale flood disaster analysis model training system.

[0082] Step S102c: Group the data according to the risk level, output the dataset in the form of an image-description, and store it in the database.

[0083] In this embodiment, after obtaining the segmented images, images with high segmentation accuracy are selected as training and evaluation samples through algorithm or manual evaluation. The risk level is then classified according to the flood situation description input by the algorithm or manual input. The large-scale flood analysis model training system is then trained using the manually input flood situation description and corresponding risk levels. The trained large-scale flood analysis model training system outputs disaster descriptions for all images and, based on grouping, outputs them as a dataset in the database in image-description format.

[0084] In this embodiment, the instruction adjustment module is the main training module of the training system. Its main function is to retrieve the segmented images from the pre-training module, select images with high segmentation accuracy, and manually describe the flood situation, including flood coverage area, water level, facility submersion, population density, evacuation route accessibility, and public sentiment. Factor analysis is then used to comprehensively consider these factors, assigning a risk level (I, II, III, and IV) to each image. The disaster descriptions and risk levels are used as the original training corpus. The system's ability to continue writing and reasoning is trained using a large-scale flood analysis model. The trained model outputs disaster descriptions for more images, and the data is grouped according to risk level and stored in the database using a one-to-one image-description mapping.

[0085] Specifically, in one implementation of this embodiment, step S103 includes the following steps:

[0086] Step S103a: Retrieve datasets of the same risk level from the database, label the order of the disaster descriptions by partial order pairs, and assign an initial score to each set of image-description datasets by preset scoring prompts.

[0087] Step S103b: The large-scale flood analysis model training system is used to maximize the difference between good and bad descriptions in the same risk level and automatically score them to obtain the optimized large-scale flood analysis model training system.

[0088] In this embodiment, a flood image dataset of the same risk level is retrieved and sorted according to the disaster descriptions of the images using a partial order pairing method. Within a set of descriptions, the partial order pairing determines that at least one description is better or worse than others. Each image-description pair is assigned an initial score using preset scoring prompts. This score can be used as an automatic score for training the large-scale flood analysis model training system. By inputting more sets of image-description data, the large-scale flood analysis model training system is trained to maximize the score difference between good and poor descriptions, ensuring that both the mean and variance of the score difference between good and poor descriptions show an increasing trend, thereby optimizing the flood image analysis accuracy of the large-scale flood analysis model training system.

[0089] In this embodiment, the reward scoring module is the adjustment and correction part of the training system, mainly used to optimize the accuracy of flood image analysis of the large model. Its main process is as follows: Flood images and their descriptions of the same risk level are retrieved from the instruction adjustment module database as a training set. Then, the order of the flood descriptions is labeled using a partial order pair pattern. Initial scores are assigned to each image-description dataset using manual scoring prompts. More sets of data are then input, allowing the large model to learn automatic scoring by maximizing the difference between good and bad descriptions. This results in an increasing trend in both the mean and variance of the score difference as the loss function decreases, until the scoring of flood images of the same risk level is completed. The loss function is shown in Formula 1. After this, the above automatic scoring process is performed on data groups of different risk levels one by one.

[0090]

[0091] Where x represents the ranking of the flood descriptions, y i Given a description of a flood, r θ This represents the scoring function.

[0092] Specifically, in one implementation of this embodiment, step S104 includes the following steps:

[0093] Step S104a: Input the flood image information into the optimized large model flood analysis mode training system to obtain the optimized disaster description and the corresponding risk level;

[0094] Step S104b: Replace the original disaster description and corresponding risk level with the optimized disaster description and corresponding risk level to obtain the updated database.

[0095] In this embodiment, the reinforcement learning module is a post-processing module of the reward scoring module. Its main purpose is to enhance the generalization ability of the large-scale flood analysis model trained above. Its main functions are: to retrieve descriptions with higher scores from the reward scoring module, and to replace the original disaster description for the same flood image, thereby overwriting the instruction adjustment module database. Then, using this new database, the original model is retrained, allowing the large-scale model to learn the language logic of excellent descriptions. Finally, it outputs flood descriptions that meet the conditions and provides the corresponding risk level.

[0096] like Figure 1 As shown, in one implementation of this invention, the method for flood evacuation analysis of underground infrastructure based on a large model further includes the following steps:

[0097] Step S200: The data output by the large model flood analysis mode training system is input into the large model-digital twin real-time simulation system for use through the integrated system of the large model and digital twin platform.

[0098] In this embodiment, the integration system of the large-scale model and the digital twin platform mainly integrates the interface of the large-scale model into the flood analysis module of the digital twin platform. Its main function is to input the comprehensive data from the digital twin platform into the large-scale model for use. Integrating the data output from the large-scale model flood analysis training system into the large-scale model-digital twin real-time simulation system enables real-time flood simulation and extrapolation, which is of great significance for disaster emergency response, prediction, and decision support. Through the integration of the large-scale model and the digital twin platform, real-time, dynamic, and visualized flood analysis can be achieved, providing more accurate and comprehensive information and tools for disaster emergency response and decision support. This helps improve the efficiency and effectiveness of disaster response, reduce disaster losses, and protect the safety of people's lives and property.

[0099] like Figure 1 As shown, in one implementation of this invention, the method for flood evacuation analysis of underground infrastructure based on a large model further includes the following steps:

[0100] Step S300: Perform data simulation using the large model-digital twin real-time simulation system and output the analysis results.

[0101] In this embodiment, the large-scale model-digital twin real-time simulation system is a comprehensive management system for flood monitoring, flood spread and population evacuation prediction, and emergency response effect monitoring and negative feedback. The main functions of the large-scale model-digital twin real-time simulation system are: when a flood occurs, it retrieves real-time flood monitoring videos stored in the digital twin platform from surveillance cameras at different heights in different flood-prone areas of underground facilities. The system then processes these videos into multiple image sequences through arithmetic slicing of the video frames. These image sequences are then input into a large-scale model trained on a flood analysis model via an integrated port for a series of processing steps, including region segmentation, image understanding, and granular control of region description. Finally, it outputs a disaster description and initial risk level for the real-time flood images. The flood image deconstruction process is as follows: Figure 3 As shown.

[0102] In this embodiment, as Figure 4As shown, the large-scale model-digital twin real-time simulation system can be divided into a foundation layer, a data layer, a component layer, an application layer, and a service layer. The foundation layer connects the digital twin platform with the large-scale model, while the data layer mainly consists of the large-scale model's training database, the BIM model's geometric database, sensor data, and equipment log data. Various data are input into the major modules within the core components, enabling real-time disaster analysis, flood spread trend prediction, flood area risk assessment, and emergency evacuation plans. Administrators can observe the actual flood situation to understand the basic disaster conditions, then watch the flood spread simulation animation to judge the flood's development, and quickly locate severely affected areas by highlighting high-risk areas. Combined with the large-scale model's recommendations, they can make decisions and manage population evacuation.

[0103] Specifically, in one implementation of this embodiment, step S300 includes the following steps:

[0104] Step S301: The data input to the integrated system of the large model and the digital twin platform is standardized through the large model-digital twin real-time inference system. The standardized data is encoded and inferred through a neural network to obtain the data inference result.

[0105] Step S302: Determine whether the risk index derived from the data analysis is higher than the critical value;

[0106] Step S303: If the risk index is higher than the critical value, a warning message is issued and an emergency plan is output in conjunction with the disaster description of the large model-digital twin real-time simulation system.

[0107] Step S304: If the risk index is not higher than the critical value, the flood image information of the next area is input into the large model-digital twin real-time simulation system for analysis, and the analysis results are output.

[0108] In this embodiment, the data input to the integrated system of the large model and the digital twin platform is standardized using a large model-digital twin real-time simulation system. For example... Figure 5 As shown, the data standardization process is as follows: Data such as temperature, water pressure, flow velocity, pedestrian speed, and crowd density at the flood site are acquired using sensors on various devices. This data is then transmitted to a RabbitMQ cluster via the MQTT data transmission protocol. Real-time processing and transformation are performed using Flink data streams, and the data is written to databases such as MySQL, InfluxDB, and Redis. Finally, the standardized data is output to the interface of the large-scale model-digital twin real-time simulation system. The standardized data is encoded and processed through a neural network to obtain simulation results, which can then be input into the BIM module for simulation display.

[0109] In this embodiment, a risk index is extracted from the data extrapolation results. Weights are assigned to various data extrapolation results using the analytic hierarchy process (AHP), and then superimposed with the risk index corresponding to the initial risk level to output the risk rating of the flood-affected area. A weighted average of the risk indices is used to establish a risk threshold. The large-scale model-digital twin real-time extrapolation system then determines whether the risk index exceeds the threshold. If it does, a warning message is issued on the evacuation management interface, and a suggested emergency plan is provided based on the disaster description in the large-scale model, prompting managers to take further flood control and evacuation measures. If the risk index does not exceed the threshold, the monitoring image information of the next area is input into the large-scale model for analysis.

[0110] Specifically, in one implementation of this embodiment, step S301 includes the following steps:

[0111] Step S301a: Use a multi-dimensional vector to set each prediction index for the standardized data, embed the standardized data into the multi-dimensional vector, encode the embedded multi-dimensional vector, and input the encoded data into multiple neural network layers;

[0112] Step S301b involves performing residual connection and normalization, multi-dimensional correction of the attention model, secondary residual connection and normalization, forward propagation, and multi-round neuron loops for loss function verification feedback on the encoded data.

[0113] Step S301c: Map the data output by the neuron after looping to a preset interval, convert it into a probability distribution, and select the data vector with the highest probability as the predicted data output.

[0114] Step S301d: Input the predicted data into the BIM model, assign weights to the predicted data, and output the risk rating of the flood-prone area in combination with the corresponding risk level.

[0115] In this embodiment, after receiving standardized data from the sensor module, the real-time data interface of the large-scale model-digital twin real-time simulation system first embeds each prediction indicator using multi-dimensional vectors. Then, each vector is encoded for subsequent parallel processing and input into multiple neural network layers. These layers undergo residual connection and normalization, multi-dimensional correction of the attention model, secondary residual connection and normalization, forward propagation, and loss function verification feedback through multiple rounds of neural network loops. Finally, the outputs of multiple neurons, including the water level height in the next second and the population distribution in the next second, are mapped to the (0,1) interval, thus converting the prediction results into a set of probability distributions. The data vector with the highest probability is output as the prediction data, which can then be input into the BIM module for simulation and simulation display. Furthermore, by weighting the output data vectors and combining them with the initial risk level, a risk rating for the flood-affected area can be output. This rating can be highlighted when input into the display interface. The overall simulation path of the large-scale model-digital twin real-time simulation system is as follows: Figure 6 As shown.

[0116] In this embodiment, after obtaining the simulation data results from the large-scale model-digital twin real-time simulation system, a flood simulation animation can be demonstrated using the BIM module of the digital twin platform. After setting up the database environment using SQL statements in the BIM module of the digital twin platform, IFC BIM data is imported using functions, where components, geometric objects, and material textures are stored as table objects. After importing the flood prediction data from the large-scale model-digital twin real-time simulation system, a simulation model is generated using the BIM module, allowing for the demonstration of the flood simulation animation. The process is as follows: Figure 7 As shown.

[0117] In this embodiment, as Figure 8 As shown, based on the risk rating of the large-scale model-digital twin real-time simulation system, different risk levels of flood-prone areas can be matched in the BIM simulation model and then highlighted. Then, by synthesizing the analysis results, a risk threshold is established, and the disaster-stricken areas are automatically labeled with risk indices and highlighted with colors in the digital twin virtual model.

[0118] In this embodiment, within the risk area display module of the flood evacuation analysis system, users can clearly distinguish the severity of different disaster-stricken areas through the different colors highlighted and displayed risk indices in the entire underground space BIM model. Clicking on a corresponding area allows access to the area's cameras to view the real-time water conditions and crowd evacuation status, including the water level reaching specific body parts, water flow rate, and the congestion level of stairs and escalators along evacuation routes. Furthermore, the large model for each risk area provides possible evacuation routes to assist managers in adopting more effective evacuation measures. In addition, users can access the top three or five areas with the highest risk indices in the entire flood-affected area through the evacuation management interface to determine the priority of rescue and evacuation.

[0119] like Figure 9 As shown, the main implementation method of the flood evacuation analysis system is as follows: First, an interface is established for the digital twin platform to call the large model. Then, a large number of flood monitoring images are collected and compiled into a dataset. This dataset is then stored in the large model training database module built into the digital twin platform. Next, the interface is integrated with the database. After integration, the BIM model data interface and sensor data interface in the digital twin platform are connected to the output interface of the large model, enabling the large model to receive image information in real time and display the analysis results on the BIM model. Finally, the data from the large model needs to be transferred to the evacuation management interface for storage and display.

[0120] This embodiment achieves the following technical effects through the above technical solution:

[0121] This embodiment acquires flood image information and outputs the corresponding disaster description and risk level through a large-scale flood analysis model training system. This system learns and establishes a flood analysis model, enabling it to output a specific flood disaster description and risk level for each input flood image. Through the integration of the large-scale model with a digital twin platform, the data output from the large-scale flood analysis model training system can be input into the large-scale model-digital twin real-time simulation system. Finally, the large-scale model-digital twin real-time simulation system performs data simulation and outputs analysis results. This involves a series of processes, including region segmentation, image understanding, and granular control of region description, before outputting a disaster description and risk level for real-time flood images. This embodiment proposes a new method for analyzing flood evacuation of underground infrastructure, improving the efficiency of flood evacuation and emergency response decisions. Compared to existing flood analysis systems that rely solely on water condition monitoring, this embodiment focuses on combining water condition monitoring with real-time population evacuation data, including population density, flow speed, and emotions at key evacuation entrances and routes. These factors are beyond the reach of traditional sensor monitoring; only through descriptions of monitoring image information can effective information be extracted. This approach essentially shifts the focus of evacuation decisions from solely relying on water condition data to a comprehensive approach that considers both water conditions and population dynamics. This not only improves the scientific rigor and effectiveness of decision-making but also reflects the human-centered design principle of the decision support system.

[0122] Exemplary device

[0123] Based on the above embodiments, such as Figure 10 As shown, the present invention also provides a large-scale model-based underground infrastructure flood evacuation analysis system, including:

[0124] Training module 100 is used to acquire flood image information and, through the large model flood analysis mode training system, output the disaster description and risk level corresponding to the flood image information.

[0125] The integration module 200 is used to input the data output by the large model flood analysis mode training system into the large model-digital twin real-time simulation system for use through the integration system of the large model and digital twin platform;

[0126] The simulation module 300 is used to perform data simulation through the large model-digital twin real-time simulation system and output analysis results.

[0127] In one implementation, the training module 100 includes:

[0128] The pre-training module is used to acquire the flood image information and perform object segmentation on water, people, equipment and underground structures in the flood image information;

[0129] The instruction adjustment module is used to describe the disaster situation and assign corresponding risk levels to the segmented images through the large model flood analysis mode training system, group them according to the risk levels and output them as a dataset in the form of image-description;

[0130] The reward scoring module is used to score datasets with different risk levels and optimize the large model flood analysis mode training system by using the score difference between the first preset description information and the second preset description information.

[0131] The reinforcement learning module is used to train the system to output optimized disaster descriptions and corresponding risk levels through the optimized large-scale flood analysis model, and to overwrite the disaster descriptions and risk levels of the same flood image information with the optimized disaster descriptions and corresponding risk levels.

[0132] In one implementation, the pre-training module includes:

[0133] The first acquisition unit is used to acquire flood video datasets and obtain flood image information for the corresponding stage by slicing according to different flood spread stages.

[0134] The input unit is used to input the flood image information and preset partition prompts into the large model flood analysis mode training system;

[0135] The transfer unit is used to perform transfer learning on the flood image information by combining image segmentation masking technology, so as to achieve object segmentation of water, people, equipment and underground structures in the flood image information.

[0136] In one implementation, the instruction adjustment module includes:

[0137] The second acquisition unit is used to acquire the segmented image, retrieve the artificial water situation description corresponding to the segmented image, and use factor analysis method to obtain the risk level corresponding to the artificial water situation description.

[0138] The training unit is used to train the large-scale flood disaster analysis mode training system based on the artificial flood disaster description and the corresponding risk level, and to output the disaster description and risk level of all segmented images through the large-scale flood disaster analysis mode training system.

[0139] A grouping unit is used to group data according to the risk level, output the dataset in the form of an image-description based on the grouping, and store it in a database.

[0140] In one implementation, the reward scoring module includes:

[0141] The assignment unit is used to retrieve datasets of the same risk level from the database, mark the order of the disaster descriptions by the pattern of partial order pairs, and assign an initial score to each set of image-description datasets by preset scoring prompt words.

[0142] The scoring unit is used to automatically score the large model flood analysis mode training system by maximizing the score difference between good and bad descriptions in the same risk level, so as to obtain the optimized large model flood analysis mode training system.

[0143] In one implementation, the deduction module 300 includes:

[0144] The data standardization module is used to standardize the data input from the integrated system of the large model and the digital twin platform through the large model-digital twin real-time inference system, encode the obtained standardized data and infer it through a neural network to obtain the data inference result;

[0145] The judgment module is used to determine whether the risk index obtained from data extrapolation is higher than the critical value; if the risk index is higher than the critical value, a warning message is issued and an emergency plan is output in conjunction with the disaster description of the large model-digital twin real-time extrapolation system; if the risk index is not higher than the critical value, the flood image information of the next area is input into the large model-digital twin real-time extrapolation system for analysis, and the analysis results are output.

[0146] In one implementation, the data standardization module includes:

[0147] A vector unit is used to set each prediction index using a multi-dimensional vector on the standardized data, embed the standardized data into the multi-dimensional vector, encode the embedded multi-dimensional vector, and input the encoded data into multiple neural network layers.

[0148] The neuron recurrent unit is used to perform residual connection and normalization, multi-dimensional correction of the attention model, secondary residual connection and normalization, forward propagation, and multi-round neuron recursion for loss function verification feedback on the encoded data.

[0149] The mapping unit is used to map the data output by the neuron after looping to a preset interval, convert it into a probability distribution, and select the data vector with the highest probability as the predicted data output.

[0150] The risk rating unit is used to input the predicted data into the BIM model, assign weights to the predicted data, and output the risk rating of the flood-prone area in combination with the corresponding risk level.

[0151] Furthermore, the present invention also provides a mobile terminal, such as... Figure 11 As shown, it includes a central processing unit 10 adapted to implement various instructions; and a memory 20 connected to the central processing unit, the memory storing a computer program, which, when executed by the central processing unit 10, implements the mobile terminal data storage method as described in any of the above schemes.

[0152] Preferably, the mobile terminal may include, but is not limited to, smartphones, tablets, laptops, and hybrid devices.

[0153] In some embodiments, the memory 20 may be an internal storage unit of the mobile terminal, such as the hard drive or memory of the mobile terminal. In other embodiments, the memory 20 may also be an external storage device of the mobile terminal, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on the mobile terminal.

[0154] Furthermore, the memory 20 may include both internal storage units and external storage devices of the mobile terminal. The memory 20 is used to store application software and various types of data installed on the device. The memory 20 can also be used to temporarily store data that has been output or will be output.

[0155] The present invention also provides a terminal, comprising: a processor, a memory, an interface, a display screen, and a communication module connected via a system bus; wherein, the processor is used to provide computing and control capabilities; the memory includes a storage medium and internal memory; the storage medium stores an operating system and computer programs; the internal memory provides an environment for the operation of the operating system and computer programs in the storage medium; the interface is used to connect to external devices, such as mobile terminals and computers; the display screen is used to display corresponding information; and the communication module is used to communicate with a cloud server or a mobile terminal.

[0156] When the computer program is executed by the processor, it is used to implement an operation of a flood evacuation analysis method for underground infrastructure based on a large model.

[0157] It will be understood by those skilled in the art that Figure 11 The schematic diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the terminal to which the present invention is applied. A specific terminal may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0158] In one embodiment, a terminal is provided, comprising: a processor and a memory, the memory storing a large-model-based underground infrastructure flood evacuation analysis program, which, when executed by the processor, is used to implement the operation of the large-model-based underground infrastructure flood evacuation analysis method as described above.

[0159] In one embodiment, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a large-model-based underground infrastructure flood evacuation analysis program, which, when executed by the processor, is used to implement the operation of the large-model-based underground infrastructure flood evacuation analysis method as described above.

[0160] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory.

[0161] In summary, this invention provides a method and system for analyzing flood evacuation of underground infrastructure based on a large model. The method includes: acquiring flood image information; outputting the disaster description and risk level corresponding to the flood image information through a large model flood analysis mode training system; inputting the data output by the large model flood analysis mode training system into a large model-digital twin real-time simulation system for use through an integrated system of the large model and digital twin platform; performing data simulation through the large model-digital twin real-time simulation system and outputting the analysis results. This invention proposes a new method for analyzing flood evacuation of underground infrastructure, improving the decision-making efficiency of flood evacuation and emergency response.

[0162] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

[0163] Possible design modifications to this invention primarily involve expanding the output of the large-scale flood analysis model during training. For example, the risk rating component in the instruction adjustment module could utilize different factor analysis methods to determine other indicator outputs for risk evaluation. In the reward scoring module, other optimization functions could be employed to optimize the flood situation description of the large-scale model, or the description could be simplified by extracting keywords. Furthermore, for the overall flood analysis system, avoidance design solutions could utilize deep learning, machine learning, or other decision tree algorithms to replace the large-scale model's analysis.

Claims

1. A flood evacuation analysis method for underground infrastructure based on a large model, characterized in that, include: Acquire flood image information, and output the disaster description and risk level corresponding to the flood image information through a large-scale model flood analysis pattern training system; The data output from the large model flood analysis model training system is input into the large model-digital twin real-time simulation system for use through the integrated system of large model and digital twin platform; Data simulation is performed using the large-scale model-digital twin real-time simulation system, and analysis results are output. The process of acquiring flood image information and outputting the corresponding disaster description and risk level through a large-scale flood analysis model training system includes: Acquire the flood disaster image information and perform object segmentation on the water, people, equipment, and underground structures in the flood disaster image information; The large-scale flood analysis model training system describes the disaster situation of the segmented images and assigns them corresponding risk levels. The images are then grouped according to the risk levels and output as a dataset in the form of images and descriptions. The datasets with different risk levels are scored, and the training system for the large model flood analysis mode is optimized by the score difference between the first preset description information and the second preset description information. The optimized large-scale flood analysis model training system outputs an optimized disaster description and corresponding risk level, and the disaster description and risk level of the same flood image information are overwritten with the optimized disaster description and corresponding risk level.

2. The method for flood evacuation analysis of underground infrastructure based on a large model as described in claim 1, characterized in that, The acquisition of flood image information, and the segmentation of objects such as water, people, equipment, and underground structures in the flood image information, include: Acquire flood video datasets and process them into slices according to different flood spread stages to obtain flood image information for the corresponding stages. The flood image information and the preset partition prompts are input into the large model flood analysis mode training system; By combining image segmentation masking technology with transfer learning on the flood disaster image information, object segmentation of water, people, equipment and underground structures in the flood disaster image information can be achieved.

3. The method for flood evacuation analysis of underground infrastructure based on a large model as described in claim 1, characterized in that, The system trains the segmented images using the large-scale flood analysis model to describe the disaster situation and assign corresponding risk levels. The images are then grouped according to these risk levels and output as a dataset in image-description format, including: The segmented image is acquired, and the artificial water situation description corresponding to the segmented image is retrieved. The risk level corresponding to the artificial water situation description is obtained by using factor analysis. The large-scale flood disaster analysis model training system is trained based on the artificial flood disaster description and the corresponding risk level, and the disaster description and risk level of all segmented images are output through the large-scale flood disaster analysis model training system. The data is grouped according to the risk level, and output as a dataset in the form of an image-description, and stored in the database.

4. The method for flood evacuation analysis of underground infrastructure based on a large model as described in claim 3, characterized in that, The process of scoring datasets at different risk levels and optimizing the large-scale flood analysis model training system based on the score difference between the first and second preset descriptive information includes: Retrieve datasets of the same risk level from the database, label the order of the disaster descriptions by partial order pairs, and assign an initial score to each set of image-description datasets by preset scoring prompts. The optimized large-scale flood analysis model training system is obtained by maximizing the difference between good and bad descriptions within the same risk level through automatic scoring.

5. The method for flood evacuation analysis of underground infrastructure based on a large model according to claim 1, characterized in that, The process of performing data extrapolation through the large-scale model-digital twin real-time extrapolation system and outputting analysis results includes: The large model-digital twin real-time inference system standardizes the data input from the integrated system of the large model and the digital twin platform, encodes the standardized data, and uses a neural network to infer the data to obtain the inference results. Determine whether the risk index derived from data analysis is higher than the critical value; If the risk index is higher than the critical value, a warning message will be issued and an emergency plan will be output in conjunction with the disaster description of the large model-digital twin real-time simulation system. If the risk index is not higher than the critical value, the flood image information of the next area will be input into the large model-digital twin real-time simulation system for analysis, and the analysis results will be output.

6. The method for flood evacuation analysis of underground infrastructure based on a large model as described in claim 5, characterized in that, The process involves standardizing the data input from the integrated system of the large model and the digital twin platform through the large model-digital twin real-time inference system, encoding the standardized data, and processing it through a neural network to obtain data inference results, including: Each prediction index is set using a multi-dimensional vector on the standardized data, the standardized data is embedded in the multi-dimensional vector, the embedded multi-dimensional vector is encoded, and the encoded data is input into multiple neural network layers. The encoded data undergoes residual connection and normalization, multi-dimensional correction of the attention model, secondary residual connection and normalization, forward propagation, and multi-round neuron loops with loss function verification feedback. The data output by the neuron after looping is mapped to a preset interval, converted into a probability distribution, and the data vector with the highest probability is selected as the predicted data output. The predicted data is input into the BIM model, weighted, and combined with the corresponding risk level to output the risk rating of the flood-prone area.

7. A large-model-based underground infrastructure flood evacuation analysis system, used to implement the large-model-based underground infrastructure flood evacuation analysis method as described in any one of claims 1-6, characterized in that, include: The training module is used to acquire flood image information and, through the large-scale flood analysis model training system, output the disaster description and risk level corresponding to the flood image information. The integration module is used to input the data output by the large model flood analysis mode training system into the large model-digital twin real-time simulation system for use through the integration system of the large model and digital twin platform; The simulation module is used to perform data simulation through the large model-digital twin real-time simulation system and output analysis results.

8. A terminal, characterized in that, include: The processor and memory, wherein the memory stores a large-model-based underground infrastructure flood evacuation analysis program, which, when executed by the processor, is used to implement the operation of the large-model-based underground infrastructure flood evacuation analysis method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a large-model-based underground infrastructure flood evacuation analysis program, which, when executed by a processor, is used to implement the operation of the large-model-based underground infrastructure flood evacuation analysis method as described in any one of claims 1-6.

Images