Emergency rescue decision-making method and system based on AI super command brain

Abstract

The embodiment of the invention provides an emergency rescue decision-making method and system based on an AI super command brain, and relates to the technical field of emergency rescue, and the method comprises the steps: carrying out the receiving of multi-source heterogeneous disaster data through a distributed data gateway; building damage identification is carried out, and a distributed disaster area identifier is output; performing state transition prediction, and outputting time sequence disaster prediction data; carrying out risk evolution analysis, and outputting a distributed prediction region identifier; after the real-time traffic topology data of the affected area is called, rescue strategy optimization is carried out, and a rescue optimization scheme is output; and after grading trace-leaving approval is performed on the rescue optimization scheme based on the block chain, multi-type rescue equipment collaborative scheduling is executed. According to the method, the technical problem that the post-disaster loss is increased due to the influence on effective allocation of post-disaster resources and scientific planning of a rescue path because the emergency war-taking decision-making method in the prior art often depends on a single data source and is difficult to comprehensively and accurately reflect the real-time dynamic state of a disaster area is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of emergency rescue, in particular to an emergency battle and rescue decision-making method and system based on an AI super command brain. BACKGROUND

[0002] With the increasing frequency of disasters and the increasing demand for emergency management, the traditional emergency battle and rescue decision-making method has been unable to meet the demand for efficient and accurate decision-making in modern complex disaster situations. The traditional emergency battle decision-making method often relies on a single data source, such as ground sensors, manual surveys or satellite images, etc. Although these methods can provide disaster information to some extent, the data sources are relatively single, and the update frequency is low. In particular, in the event of a large-scale disaster, a single data source is difficult to fully and accurately reflect the real-time dynamics of the disaster area. Due to the lack of sufficient multi-dimensional information for disaster assessment, the decision-makers do not have a comprehensive understanding of the disaster area, which affects the effective allocation of post-disaster resources and the scientific planning of rescue routes, resulting in waste or delay of rescue resources and increasing post-disaster losses.

[0003] It should be noted that the information disclosed in this background section is only intended to increase the understanding of the overall background of the present application, and should not be considered as acknowledging or implying in any form that this information constitutes prior art known to those skilled in the art. SUMMARY

[0004] In view of the above defects or improvement needs of the prior art, the present application provides an emergency battle and rescue decision-making method and system based on an AI super command brain, which solves the technical problem that the existing emergency battle decision-making method often relies on a single data source, which is difficult to fully and accurately reflect the real-time dynamics of the disaster area, affecting the effective allocation of post-disaster resources and the scientific planning of rescue routes, resulting in waste or delay of rescue resources and increasing post-disaster losses. The specific technical solutions are as follows:

[0005] According to a first aspect of the present application, an emergency battle rescue decision method based on an AI super-command brain is provided, the method comprising: receiving multi-source heterogeneous disaster data through a distributed data gateway to obtain double-time satellite data, real-time disaster data and real-time Internet of Things sensing data; identifying building damage based on pre-disaster remote sensing data and post-disaster remote sensing data in the double-time satellite data to output distributed disaster-affected area identification; predicting state transition of the real-time disaster data to output time-series disaster prediction data; performing risk evolution analysis of the distributed disaster-affected area identification according to the time-series disaster prediction data to output distributed predicted area identification; projecting real-time traffic topology data of the disaster-affected area to the distributed predicted area identification after calling the real-time traffic topology data to perform rescue strategy optimization and output a rescue optimization scheme; and distributing rescue tasks according to the rescue optimization scheme after hierarchical trace approval of the rescue optimization scheme based on a block chain to perform multi-type rescue equipment collaborative scheduling.

[0006] In one embodiment, the building damage identification based on the pre-disaster remote sensing data and the post-disaster remote sensing data in the double-time satellite data to output the distributed disaster-affected area identification comprises:

[0007] spatial-temporal alignment of the pre-disaster remote sensing data and the post-disaster remote sensing data is performed using a geographic information system; double-time building outline baseline extraction of the pre-disaster remote sensing data and the post-disaster remote sensing data is performed using an edge detection algorithm to output a double-time building distribution vector map; pixel-level gray difference calculation is performed by traversing the double-time building distribution vector map to output a building gray difference vector map; state probability prediction is performed by traversing the double-time building distribution vector map to output a building state probability vector map; and building collapse determination is performed by fusing the building gray difference vector map and the building state probability vector map to output the distributed disaster-affected area identification.

[0008] In one embodiment, the state transition prediction of the real-time disaster data to output the time-series disaster prediction data comprises:

[0009] continuous time-series disaster feature extraction is performed on the real-time disaster data to obtain a time-series disaster multi-dimensional feature vector; the time-series disaster multi-dimensional feature vector is input into a disaster evolution prediction model constructed based on a double-layer attention LSTM network to perform disaster state transition prediction to output time-series risk probability data and time-series disaster intensity data; after the time-series risk probability data and the time-series disaster intensity data are marked with high-risk area spatial boundaries, structured data storage is performed to output the time-series disaster prediction data.

[0010] In an embodiment, after obtaining the real-time traffic topology data of the disaster area, the real-time traffic topology data is projected to the distributed prediction area identifier, rescue strategy optimization is performed, and a rescue optimization scheme is output, including:

[0011] The real-time traffic topology data is projected to the spatial boundary of the distributed prediction area identifier, traffic risk superposition pruning is performed, and a feasible traffic road network graph is output. The distributed area rescue demand of the distributed prediction area identifier is matched based on the attributes of the disaster area, where the attributes of the disaster area include the number of trapped people, the building collapse rate, and the proportion of special groups. The distributed area rescue demand and the feasible traffic road network graph are used as optimization inputs, multi-objective optimization algorithm is used for rescue path planning, and a distributed rescue strategy is output to constitute the rescue optimization scheme.

[0012] In an embodiment, the distributed area rescue demand and the feasible traffic road network graph are used as optimization inputs, multi-objective optimization algorithm is used for rescue path planning, and a distributed rescue strategy is output to constitute the rescue optimization scheme, including:

[0013] The first material shelf life of the first area rescue demand is obtained interactively, and the first material shelf life is used as a rescue timeliness constraint. The rescue road slope constraint and the rescue resource consumption constraint are predefined. The rescue timeliness constraint, the rescue road slope constraint, and the rescue resource consumption constraint are used as hard constraint conditions, and rescue path planning is performed in the feasible traffic road network graph with the first disaster prediction area as the rescue destination, and a first rescue strategy is output, where the first rescue strategy includes an evacuation path and a resource allocation list.

[0014] In an embodiment, the state probability prediction is performed by traversing the double-time-phase building distribution vector graph to output a building state probability vector graph, and before that, including:

[0015] A plurality of sample satellite remote sensing images are obtained interactively, and building outline coordinates and building state label identifiers are identified from the plurality of sample satellite remote sensing images to obtain a plurality of sample labeled data sets. The plurality of sample satellite remote sensing images and the plurality of sample labeled data sets are used as training data to perform parameter optimization of a pre-disaster feature extraction channel and a post-disaster feature extraction channel constructed based on a DCNN model. After the pre-disaster feature extraction channel and the post-disaster feature extraction channel are connected in parallel, a consistency verification layer is cascaded at the output end of the pre-disaster feature extraction channel and the post-disaster feature extraction channel to complete the construction of a double-channel DCNN model.

[0016] In an embodiment, the state probability prediction is performed by traversing the double-time-phase building distribution vector graph to output a building state probability vector graph, including:

[0017] Based on a preset segmentation scale, the dual-temporal building distribution vector map is rasterized to obtain multiple sets of dual-temporal building distribution domain maps with multiple raster indices. The first pre-disaster building distribution domain map and the first post-disaster building distribution domain map in the first set of dual-temporal building distribution domain maps are input in parallel into the pre-disaster feature extraction channel and the post-disaster feature extraction channel of the dual-channel DCNN model to perform state probability prediction and output the first pre-disaster state probability vector and the first post-disaster state probability vector. After the first pre-disaster state probability vector and the first post-disaster state probability vector are spatiotemporally aligned at the consistency verification layer, the collapse probability difference value is calculated and the first local state probability vector map is output. By analogy, the dual-channel DCNN model is driven to analyze the multiple sets of dual-temporal building distribution domain maps to obtain multiple local state probability vector maps. The multiple local state probability vector maps are spliced ​​according to the multiple raster index spaces to output the building state probability vector map.

[0018] In one implementation, after the consistency verification layer spatiotemporally aligns the first pre-disaster state probability vector and the first post-disaster state probability vector, it performs collapse probability difference value calculation and outputs a first local state probability vector map, including:

[0019] Based on the first pre-disaster state probability vector and the first post-disaster state probability vector, calculate the first collapse probability difference value of the first building; if the first collapse probability difference value is greater than a preset difference threshold, calculate the first image block structure similarity index of the first pre-disaster building layout domain map and the first post-disaster building layout domain map; if the first image block structure similarity index is less than a preset structure change threshold, output the first post-disaster state probability vector as the first local state probability vector map; if the first image block structure similarity index is greater than or equal to the structure change threshold, weightedly fuse the first pre-disaster state probability vector and the first post-disaster state probability vector to output the first local state probability vector map; if the first collapse probability difference value is less than or equal to the difference threshold, directly output the first post-disaster state probability vector as the first local state probability vector map.

[0020] According to a second aspect of the present invention, an emergency response and rescue decision-making system based on an AI super command brain is provided. The system includes: a data receiving module, used to receive multi-source heterogeneous disaster data through a distributed data gateway to obtain dual-temporal satellite data, real-time disaster data, and real-time IoT sensor data; a damage identification module, used to identify building damage based on pre-disaster remote sensing data and post-disaster remote sensing data in the dual-temporal satellite data, and output distributed disaster-affected area identifiers; a state transition prediction module, used to perform state transition prediction on the real-time disaster data, and output time-series disaster prediction data; a risk evolution analysis module, used to perform risk evolution analysis on the distributed disaster-affected area identifiers based on the time-series disaster prediction data, and output distributed prediction area identifiers; a rescue strategy optimization module, used to retrieve real-time traffic topology data of the disaster-affected area, project the real-time traffic topology data onto the distributed prediction area identifiers, perform rescue strategy optimization, and output rescue optimization schemes; and an equipment collaborative scheduling module, used to perform hierarchical and traceable approval of the rescue optimization schemes based on blockchain, allocate rescue tasks according to the rescue optimization schemes, and execute collaborative scheduling of multiple types of rescue equipment.

[0021] Beneficial effects of the embodiments of the present invention:

[0022] By accessing multi-source heterogeneous disaster data through a distributed data gateway, including dual-temporal satellite data, real-time disaster data, and IoT sensor data, information about disaster areas can be collected and analyzed in real time. This data fusion method provides high-precision and comprehensive basic data support for disaster assessment. By comparing pre-disaster and post-disaster remote sensing data, the damage to buildings can be accurately identified, thereby determining the specific location of the affected area. This method is more efficient and can cover a wider area than traditional ground surveys. Using real-time disaster data for state transition prediction can predict post-disaster development trends, providing early warnings for post-disaster emergency management. It can help decision-makers predict future changes in disaster risks and avoid blind responses and misjudgments. By retrieving real-time traffic topology data and combining it with distributed predicted area identifiers to optimize rescue routes, optimal rescue route plans can be generated. This method maximizes the conservation of rescue resources, ensuring that rescue forces can reach the most severely affected areas as quickly as possible. The approval and task allocation of rescue optimization plans based on blockchain technology enables a tiered, transparent, and traceable management system, which not only enhances the credibility of the rescue process but also effectively prevents information tampering and resource waste. Through the coordinated scheduling of multiple types of rescue equipment, it ensures the coordinated cooperation of different types of rescue forces, maximizing the utilization of various resources for post-disaster recovery. The coordinated scheduling of rescue tasks helps reduce resource conflicts, improves the collaborative efficiency of rescue operations, and ensures that each task can be completed in the shortest possible time. Of course, implementing any product or method of this invention does not necessarily require achieving all of the above advantages simultaneously. Attached Figure Description

[0023] 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 these drawings without creative effort.

[0024] Figure 1 This invention illustrates a flowchart of an emergency response and rescue decision-making method based on an AI super command brain, provided by the present invention.

[0025] Figure 2 The diagram shows the structure of the emergency response and rescue decision-making system based on AI super command brain provided by the present invention.

[0026] Explanation of reference numerals in the attached diagram: Data receiving module 10, damage identification module 20, state transition prediction module 30, risk evolution analysis module 40, rescue strategy optimization module 50, equipment collaborative scheduling module 60. Detailed Implementation

[0027] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein; rather, these embodiments are provided so that the disclosure of the invention will be more thorough and complete.

[0028] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0029] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0030] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0031] The emergency response and rescue decision-making method and system based on AI super command brain provided by this invention is used to solve the technical problem that existing emergency response and rescue decision-making methods often rely on a single data source, which makes it difficult to comprehensively and accurately reflect the real-time dynamics of the disaster area, affecting the effective allocation of post-disaster resources and the scientific planning of rescue routes, thereby causing waste or delays in rescue resources and increasing post-disaster losses.

[0032] Example 1: See Figure 1 The flowchart of the emergency response and rescue decision-making method based on AI super command brain provided in this embodiment of the invention includes:

[0033] Y100: Receives multi-source heterogeneous disaster data through a distributed data gateway, obtaining dual-temporal satellite data, real-time disaster data, and real-time IoT sensor data.

[0034] Distributed data gateways are used to receive and integrate data from multiple different sources. They connect via a network and ensure real-time data transmission. The gateway's role is to convert and process data of different formats and protocols, ensuring unified access to the decision-making system. This includes acquiring pre- and post-disaster remote sensing imagery data, obtained through satellite sensors, providing wide-area ground imagery and change information; real-time disaster data, including on-site data from the disaster area, such as reports from rescue departments, real-time disaster video, weather monitoring, and post-disaster investigation data; and real-time IoT sensor data acquired through sensors deployed in the disaster area, including seismometers, meteorological sensors, and temperature sensors, providing real-time environmental monitoring data such as temperature, humidity, and air pressure.

[0035] Y200: Based on the pre-disaster and post-disaster remote sensing data in the dual-temporal satellite data, identify building damage and output distributed disaster-affected area identifiers.

[0036] Spatiotemporal alignment of pre- and post-disaster remote sensing images is performed to ensure spatial correspondence between images from two different time points. This is because satellite images may have some discrepancies, such as differences in shooting angle and altitude, thus requiring image calibration and alignment. Edge detection algorithms or image difference analysis methods are used to analyze the morphological changes of buildings before and after the disaster, extracting the building outlines and distribution. By comparing changes in building shape, damaged areas can be identified. By calculating grayscale or structural differences in buildings and comparing pre- and post-disaster images, areas with significant changes, such as collapsed or severely damaged buildings, are detected. These changes are then converted into distributed disaster area identifiers to mark damaged buildings or areas.

[0037] Y300: Performs state transition prediction on the real-time disaster data and outputs time-series disaster prediction data.

[0038] Key time-series disaster characteristics, such as the rate of change, scope of impact, and severity, are extracted from real-time disaster data. Based on historical patterns of disaster evolution, a predictive model is constructed. This model analyzes the evolution trend of disasters and, through training on historical data, can predict the development trend of disasters in the next few hours or days. The model outputs time-series disaster prediction data, which includes information such as the probability of disaster risk and changes in disaster intensity. For example, it predicts whether a region will experience a larger-scale disaster or whether the intensity of an existing disaster will intensify.

[0039] Y400: Based on the time-series disaster prediction data, perform risk evolution analysis on the distributed disaster-affected area identifiers and output distributed prediction area identifiers.

[0040] Using time-series disaster forecasting data, the evolution of disasters is analyzed, i.e., the risk changes from the current state to the future state. Risk evolution analysis includes: disaster intensity evolution analysis, assessing whether the intensity of the disaster will change in the future and whether it will worsen, such as earthquakes or fire spread; and risk area evolution analysis, identifying which areas have increased risk based on disaster changes, for example, if a region is about to experience secondary disasters such as floods or landslides, the risk in that region will increase. Using risk assessment models, the evolution of disaster-affected areas is predicted based on time-series data. Distributed prediction area identification refers to predicting and marking areas that may be affected by disasters in the future based on the results of risk evolution analysis. This area identification not only reflects the current state of the disaster but also includes the possible future expansion of the disaster.

[0041] Y500: After retrieving real-time traffic topology data of the disaster-stricken area, the real-time traffic topology data is projected onto the distributed prediction area identifier to optimize the rescue strategy and output the rescue optimization plan.

[0042] Real-time traffic topology data refers to road network information within the disaster area, including the location, traffic conditions, traffic flow, traffic congestion, and traffic accidents of each road. This data is obtained through traffic sensors, GPS information, and drone monitoring to capture real-time traffic conditions in the disaster area. Projection involves mapping the real-time traffic topology data onto previously output distributed prediction area markers. Here, projection means associating the traffic network with the affected areas of the disaster zone, assessing which traffic routes can successfully reach high-risk areas and which may be impassable due to road damage, traffic congestion, or other reasons. In this way, feasible road networks are identified—those transportation routes that can still be used for rescue operations.

[0043] Multi-objective optimization algorithms, such as genetic algorithms and particle swarm optimization, are used to address the different requirements of the rescue mission as constraints, such as time efficiency, resource allocation, and the transportation capacity of personnel and materials. The algorithm seeks the optimal path and action plan within a feasible transportation network. Based on the transportation network and the distribution of the disaster area, the optimization algorithm formulates the optimal rescue route to ensure that rescue resources can reach the disaster area quickly and efficiently, ultimately outputting an optimized rescue plan.

[0044] Y600: After the rescue optimization plan is approved and recorded in a hierarchical manner based on blockchain, rescue tasks are allocated according to the rescue optimization plan, and multi-type rescue equipment is coordinated and dispatched.

[0045] By leveraging blockchain technology, every step of the optimized rescue plan is recorded on the blockchain, ensuring transparency and traceability. The immutable nature of the blockchain provides a robust record, guaranteeing the legitimacy and validity of every rescue decision, especially crucial for complex rescue operations and resource allocation. Based on the optimized rescue plan's details, such as rescue routes, material needs, and disaster situation, specific rescue tasks are assigned according to priority, resource allocation, and feasibility. Each task is assigned to the most suitable rescue team or resource. An intelligent dispatch platform coordinates different types of rescue equipment, such as medical equipment, transport vehicles, excavators, and drones, automatically scheduling them for efficient collaboration based on task requirements, ensuring the smooth operation of the rescue mission.

[0046] In one implementation, building damage identification is performed based on pre-disaster and post-disaster remote sensing data from the dual-temporal satellite data, and a distributed disaster-affected area identifier is output, including:

[0047] Y210: Apply a geographic information system to perform spatiotemporal alignment of the pre-disaster and post-disaster remote sensing data; Y220: Use an edge detection algorithm to extract the dual-temporal building outline baselines from the pre-disaster and post-disaster remote sensing data, and output a dual-temporal building distribution vector map; Y230: Traverse the dual-temporal building distribution vector map to perform pixel-level grayscale difference calculation, and output a building grayscale difference vector map; Y240: Traverse the dual-temporal building distribution vector map to perform state probability prediction, and output a building state probability vector map; Y250: Fuse the building grayscale difference vector map and the building state probability vector map to determine building collapse, and output the distributed disaster-affected area identifier.

[0048] Spatiotemporal alignment refers to the spatial and temporal consistency calibration of pre-disaster and post-disaster remote sensing data to ensure that the two are accurately aligned in terms of geographical location and time. Since the images were taken at different times before and after the disaster, the Earth's surface may have shifted. For example, due to ground deformation caused by earthquakes or other disasters, the spatial coordinates of the images may differ. Therefore, spatiotemporal alignment can ensure that the remote sensing data from these two points in time can be compared for the same geographical location.

[0049] Edge detection is used to extract edge information from images. In remote sensing images, edges represent the outlines of ground features, such as the shape of buildings. In pre-disaster and post-disaster remote sensing data, edge detection algorithms are applied to extract building outlines for each time phase. Then, through vectorization—that is, converting edge lines into vector form—these outlines are transformed into building distribution vector maps. The vector maps represent the spatial distribution and shape characteristics of buildings. Dual-temporal building distribution vector maps include pre-disaster and post-disaster building distribution vector maps.

[0050] Buildings in pre- and post-disaster remote sensing data will visually change, especially when buildings are damaged or collapsed. Grayscale difference refers to the difference in grayscale values ​​of pixels at the same location in two temporal images. In pre- and post-disaster images, buildings may show significant changes in grayscale values. For example, collapsed buildings can cause drastic changes in grayscale values. By traversing the dual-temporal building distribution vector map, comparing the grayscale values ​​of every pixel in the pre- and post-disaster images, calculating the grayscale difference for each pixel, and generating a building grayscale difference vector map, this image can highlight areas where buildings have changed, such as collapsed or severely damaged buildings.

[0051] In the aftermath of a disaster, buildings can be in various states, such as completely collapsed, partially collapsed, damaged, or undamaged. Based on pre- and post-disaster image data, models predict the state of each building. These state prediction models can employ machine learning algorithms, such as support vector machines, random forests, and convolutional neural networks, to classify the degree of damage to buildings. By inputting pre- and post-disaster image features, the model can calculate the probability of collapse for each building based on its pre- and post-disaster grayscale differences, contour changes, and other information. The final output is a building state probability vector map, which uses color or numerical values ​​to represent the probability of collapse or the state of damage for each building, providing a basis for post-disaster assessment and rescue decisions.

[0052] The pre- and post-disaster grayscale difference map shows the morphological changes of buildings, while the state probability map reflects the probability of building collapse. Integrating these two maps allows for a comprehensive analysis of visual changes and collapse risk. Methods such as weighted averaging and image overlay are used to combine the grayscale difference map and the state probability map. For example, a certain threshold is set; if the probability of a building's collapse exceeds this threshold and the grayscale difference is significant, the building is considered to have collapsed. The final output, a distributed disaster-affected area identifier, provides an identifier for each area or building, indicating whether the area was affected by the disaster and the extent of damage.

[0053] In one implementation, state transition prediction is performed on the real-time disaster data to output time-series disaster prediction data, including:

[0054] Y310: Extract continuous time-series disaster features from the real-time disaster data to obtain a time-series disaster multidimensional feature vector; Y320: Input the time-series disaster multidimensional feature vector into a disaster evolution prediction model constructed based on a two-layer attention LSTM network to predict the disaster state transition and output time-series risk probability data and time-series disaster intensity data; Y330: After identifying the time-series risk probability data and time-series disaster intensity data using high-risk area spatial boundaries, perform structured data storage and output the time-series disaster prediction data.

[0055] Continuous temporal feature extraction refers to extracting features that change over time from real-time disaster data. Disaster data is usually sequential in time, and its features also change over time. Therefore, it is necessary to capture these dynamic temporal features, including changes in disaster intensity, speed of disaster spread, changes in affected areas, and multi-dimensional features of the disaster, such as external factors affecting the disaster, such as weather changes, wind speed changes, and precipitation changes. Temporal feature extraction methods extract different disaster variables in chronological order, forming a multi-dimensional feature vector of the disaster data over time.

[0056] LSTM is a deep learning model capable of processing time-series data, suitable for time-series data with long-term dependencies. LSTM effectively captures long-term trends and short-term fluctuations in time-series data through gating mechanisms. Building upon traditional LSTM, an attention mechanism is introduced to enhance the model's ability to focus on important moments. This step employs a two-layer attention mechanism, selectively focusing on two levels: temporal attention, which focuses on key moments at different time steps, enhancing the model's attention to data at important moments; and feature-level attention, which strengthens the model's focus on important features within the input temporal feature vector, especially in different disaster data where certain features may be more critical than others.

[0057] The obtained multi-dimensional feature vector of time-series disaster is input into a two-layer attention LSTM network. The network learns the patterns of historical disaster data to predict the future development trend of the disaster, that is, the state transition of the disaster. In the output results, the time-series risk probability data is the probability distribution of disaster risk at future time, such as the probability of a disaster occurring in a certain area in the next two hours, and the possibility of the disaster intensifying; the time-series disaster intensity data is the trend of disaster intensity change, such as predicting the intensity of an aftershock of an earthquake or the change of flood level in the next few hours.

[0058] Based on the obtained time-series risk probability data and time-series disaster intensity data, further analysis is conducted to identify areas with higher disaster risks. For example, high-risk areas can be identified using geographic information systems or other spatial analysis methods. The spatial boundaries of high-risk areas can be defined by setting thresholds. These identified high-risk areas will become the key monitoring areas for the disaster prediction model. The time-series prediction data of high-risk areas is stored in a structured format, which can be stored as a time series, recording the corresponding disaster risk and intensity at each time point. The output time-series disaster prediction data includes a comprehensive prediction of future disaster conditions, providing the risk probability, disaster intensity, and boundary and key prediction data of high-risk areas at each time point.

[0059] In one implementation, after retrieving real-time traffic topology data of the disaster-stricken area, the real-time traffic topology data is projected onto the distributed prediction area identifier to optimize the rescue strategy and output an optimized rescue plan, including:

[0060] Y510: Project the real-time traffic topology data onto the spatial boundary of the distributed prediction area identifier, perform traffic risk overlay pruning, and output a feasible traffic network map; Y520: Match the distributed area rescue needs of the distributed prediction area identifier based on the disaster-stricken area attributes, wherein the disaster-stricken area attributes include the number of trapped people, building collapse rate, and proportion of special groups; Y530: Use the distributed area rescue needs and feasible traffic network map as optimization inputs, perform rescue path planning based on a multi-objective optimization algorithm, output a distributed rescue strategy, and constitute the rescue optimization scheme.

[0061] This process combines real-time traffic topology data with distributed predictive area identifiers. Specifically, the predictive area identifiers define the spatial extent of the disaster area and its high-risk zones. This step maps the traffic network to the spatial boundaries of the disaster area, ensuring that traffic data only considers areas relevant to the disaster. Projection operations include spatial alignment and registration, enabling precise matching between traffic network data and disaster area boundary data, avoiding interference from any irrelevant areas.

[0062] Risk overlay refers to superimposing different risk factors on different roads to comprehensively assess the traffic risk of each road. Traffic risks include road damage, traffic congestion, and traffic accidents. Pruning refers to filtering usable routes by removing roads that do not meet rescue needs. For example, completely destroyed roads or high-risk traffic areas can be excluded using pruning algorithms, retaining feasible roads. Through this overlay and pruning operation, a feasible traffic network map is obtained, which is a traffic network that can be used in the disaster area and is suitable for rescue.

[0063] Disaster area attributes refer to various key characteristics of a disaster area, used to assess the rescue needs of the disaster area. These attributes can be obtained through remote sensing data, IoT sensors, and on-site surveys. Among them, the number of people trapped refers to the number of people who are unable to save themselves or evacuate due to the disaster; the building collapse rate refers to the degree of collapse of buildings in the disaster area. This information helps to assess the urgency and complexity of the rescue mission; the proportion of special groups refers to the proportion of special groups such as the elderly, children, pregnant women, and people with disabilities. These groups usually require higher priority rescue resources.

[0064] By using distributed predictive regional identifiers, the severity of the disaster in different areas within the disaster zone can be determined. Based on the disaster situation in each area, the attributes of the affected areas are matched with the disaster data to form the distributed regional rescue needs for each disaster zone. For example, if a certain area has a high building collapse rate and a large number of people trapped, it can be inferred that the rescue needs of that area are more urgent.

[0065] Multi-objective optimization algorithms are a class of methods for solving optimization problems, suitable for situations requiring simultaneous consideration of multiple objectives. In rescue route planning, multiple factors need to be optimized, such as rescue timeliness, resource consumption, rescue route safety, and traffic flow control. Multi-objective optimization algorithms include genetic algorithms, particle swarm optimization algorithms, and ant colony optimization algorithms. This paper takes distributed regional rescue needs and feasible road network maps as input, and uses multi-objective optimization algorithms to plan rescue routes. The planning process needs to generate optimal routes based on factors such as traffic conditions, terrain features, resource needs, and time urgency in the disaster area, ensuring that rescue tasks can be completed efficiently and with priority, ultimately generating a distributed rescue strategy. The overall rescue optimization scheme consists of rescue routes, resource allocation, and priority ranking, ensuring maximum efficiency in the use of rescue resources and timely completion of rescue tasks.

[0066] In one implementation, the distributed regional rescue needs and feasible traffic network map are used as optimization inputs. A multi-objective optimization algorithm is used for rescue route planning, and a distributed rescue strategy is output, constituting the rescue optimization scheme, including:

[0067] Y531: Interact to obtain the shelf life of the first material needed for the first area's rescue, and use the shelf life of the first material as a constraint on the timeliness of rescue; Y532: Predefine the slope constraint of the rescue road and the consumption constraint of the rescue resources; Y533: Use the constraints on the timeliness of rescue, the slope constraint of the rescue road, and the consumption constraint of the rescue resources as hard constraints, execute the rescue route planning with the first disaster prediction area as the rescue destination on the feasible traffic network map, and output the first rescue strategy, wherein the first rescue strategy includes the disaster avoidance route and the resource allocation list.

[0068] By interacting with the disaster area's on-site rescue command system, material dispatch center, and post-disaster assessment team, we obtain the initial rescue needs of the region, including the types, quantities, and shelf lives of various urgently needed supplies, as well as possible storage conditions and transportation requirements. The shelf life of supplies refers to the effective period before use; as time passes, the effectiveness of some supplies may decrease, therefore, the shelf life needs to be considered in the rescue plan. Incorporating the shelf life of supplies as a constraint into the rescue route planning ensures that all supplies with shelf lives can be delivered to the disaster area within their validity period.

[0069] Rescue road gradient refers to the angle of inclination or steepness of a road. In disaster relief, especially in mountainous or earthquake-stricken areas, roads with steep gradients may be unsuitable for heavy rescue vehicles or significantly reduce transport speed. Predefined road gradient constraints refer to setting certain restrictions on the gradient of different roads. For example, it may stipulate that heavy equipment is not allowed to pass on roads with gradients exceeding a certain angle, or that roads with higher gradients require additional time or resources to handle.

[0070] Resource consumption constraints include the amount of resources used in the rescue process, such as fuel, vehicles, and personnel. Predefined resource consumption constraints are mainly to ensure that rescue routes do not lead to excessive resource consumption. For example, some routes may consume more fuel and manpower due to poor road conditions, long distances, or the need for detours.

[0071] Hard constraints refer to conditions that must be strictly met. In path planning, timeliness constraints, road gradient constraints, and resource consumption constraints are all considered hard constraints. Path planning must satisfy these conditions; otherwise, the solution is infeasible. In the feasible traffic network map, a multi-objective optimization algorithm is used to plan paths based on all hard constraints. During the path planning process, the optimal path is generated based on the state of roads in the disaster area, rescue needs, and the aforementioned constraints, and the first rescue strategy is output. Among these, disaster avoidance paths refer to planned paths that bypass high-risk areas, avoiding areas with secondary disasters within the disaster area; the resource allocation list refers to the resources required for each path, such as rescue supplies, personnel, and vehicles.

[0072] In one implementation, the state probability is predicted by traversing the dual-temporal building distribution vector map and outputting the building state probability vector map. Prior to this, the following steps are included:

[0073] Y241: Interactively acquire multiple sample satellite remote sensing images, and obtain multiple sample labeled data sets by identifying building outline coordinates and building status labels on the multiple sample satellite remote sensing images; Y242: Use the multiple sample satellite remote sensing images and multiple sample labeled data sets as training data to perform parameter tuning and optimization of the pre-disaster feature extraction channel and the post-disaster feature extraction channel based on the DCNN model; Y243: After connecting the pre-disaster feature extraction channel and the post-disaster feature extraction channel in parallel, cascade a consistency verification layer at the output of the pre-disaster feature extraction channel and the post-disaster feature extraction channel to complete the construction of the dual-channel DCNN model.

[0074] The sample satellite remote sensing images provide pre- and post-disaster geographic information, covering a large disaster area. By collecting a large amount of image data from satellites or other remote sensing platforms, the ground conditions at different time points before and after the disaster are captured. Building outlines are extracted from multiple sample satellite remote sensing images, primarily using image segmentation or edge detection techniques to extract building boundaries. Building outline coordinates refer to the edge positions of buildings in the image, represented by a set of polygon coordinates that accurately describe the outline of each building. Building status labels assign a status label to each building or area of ​​a building, indicating the degree of damage. Through the above processing of multiple sample satellite remote sensing images, multiple sample labeled data sets are obtained; each image corresponds to a set of building outline coordinates and building status labels. This dataset will serve as the data foundation for model training.

[0075] DCNN is a deep convolutional neural network used for tasks such as image classification and object detection. DCNN automatically extracts features from images through multiple convolutional layers, typically eliminating the need for manually designed feature extractors. The pre-disaster feature extraction channel is responsible for extracting features from pre-disaster remote sensing images, such as the shape, location, and color of buildings, helping the model understand the pre-disaster building structure. The post-disaster feature extraction channel, through parameter tuning and optimization, is responsible for extracting features from post-disaster remote sensing images, such as the degree of damage to buildings, collapsed areas, and damaged parts, used to analyze the post-disaster building condition.

[0076] The acquired sample satellite remote sensing images and labeled sample data are used together as the training set and input into the DCNN model. The model aims to learn the features of buildings before and after a disaster from this data and to automatically identify the damage status of buildings based on post-disaster images. Parameter tuning and optimization refer to adjusting the hyperparameters of the DCNN model, such as kernel size, number of layers, and learning rate, to achieve optimal model performance. The parameter tuning process relies on methods such as cross-validation, grid search, or random search. During training, these parameters are continuously adjusted to optimize the model's ability to extract features from pre- and post-disaster images and improve the accuracy of building damage identification.

[0077] The pre-disaster feature extraction channel and the post-disaster feature extraction channel are connected in parallel, meaning they extract features independently but share a portion of the model structure. The consistency verification layer's role is to fuse and verify the features from pre-disaster and post-disaster images, including: fusing the output features from the pre-disaster and post-disaster channels so the model can comprehensively consider the influence of both; verifying whether the features extracted from pre-disaster and post-disaster images are consistent and how they are related, especially in determining the condition of buildings after the disaster. After connecting the pre-disaster and post-disaster feature extraction channels through the consistency verification layer, a dual-channel DCNN model is finally formed. This model can process both types of images, including pre-disaster and post-disaster images, and by fusing information from both types of images, it can achieve more accurate building damage identification.

[0078] In one implementation, the state probability is predicted by traversing the dual-temporal building distribution vector map and outputting the building state probability vector map, including:

[0079] Y244: Based on a preset segmentation scale, rasterize the dual-temporal building distribution vector map to obtain multiple sets of dual-temporal building distribution domain maps with multiple raster indices; Y245: Input the first pre-disaster building distribution domain map and the first post-disaster building distribution domain map from the first set of dual-temporal building distribution domain maps into the pre-disaster feature extraction channel and post-disaster feature extraction channel of the dual-channel DCNN model in parallel, perform state probability prediction, and output the first pre-disaster state probability vector and the first post-disaster state probability vector; Y246: After spatiotemporally aligning the first pre-disaster state probability vector and the first post-disaster state probability vector in the consistency verification layer, perform collapse probability difference value calculation and output the first local state probability vector map; Y247: By analogy, drive the dual-channel DCNN model to analyze the multiple sets of dual-temporal building distribution domain maps to obtain multiple local state probability vector maps; Y248: Based on the multiple raster index spaces, stitch together the multiple local state probability vector maps and output the building state probability vector map.

[0080] The preset segmentation scale divides the building distribution vector map into grid cells of a certain size, such as 10m*10m or 20m*20m. Each grid cell contains building information within a specific area, simplifying subsequent analysis. After rasterization, multiple raster indices are obtained, with each grid cell representing the building distribution of a region. These indices allow for quick location and analysis of building changes within a specific area.

[0081] The first pre-disaster building layout map and the first post-disaster building layout map from the first set of dual-temporal building layout maps are input in parallel into a dual-channel DCNN model. The pre-disaster building layout map provides structural information about the buildings before the disaster, while the post-disaster building layout map provides information about the damage to the buildings after the disaster. The model has two channels: a pre-disaster feature extraction channel and a post-disaster feature extraction channel. These two channels simultaneously extract features from both images to compare the changes in buildings before and after the disaster. In each channel, state probability prediction is performed, outputting the first pre-disaster state probability vector and the first post-disaster state probability vector.

[0082] The consistency verification layer performs spatiotemporal alignment on the state probability vectors from pre-disaster and post-disaster images. Pre-disaster and post-disaster images may exhibit temporal variations and spatial discrepancies, necessitating spatiotemporal alignment to ensure data consistency. This alignment process involves techniques such as timestamp matching, spatial coordinate overlap, and scale matching to ensure a reasonable combination of pre-disaster and post-disaster data. After spatiotemporal alignment, the collapse probability difference value is calculated, representing the difference between the pre-disaster and post-disaster states. This difference value reflects the degree of change in the likelihood of building collapse after the disaster. For example, if the pre-disaster state probability vector shows a building intact, but the post-disaster state probability vector shows severe damage, a large difference value indicates that the building may have collapsed. The output first local state probability vector map is an image covering the entire disaster area, where the state probability of each local region represents the damage status of buildings within that region.

[0083] Similarly, by feeding new input data one by one into the pre-trained dual-channel DCNN model for inference, each input data is processed by the model to obtain a prediction result. Each local state probability vector map assigns a state probability value to the distribution of buildings in a specific grid cell. These values ​​reflect information such as the probability of damage and collapse of buildings in that area.

[0084] Multiple local state probability vector maps are spatially stitched together according to the raster index. The spatial stitching process is to synthesize multiple scattered local probability vector maps in space according to the raster index to form a complete building state probability map covering the entire disaster area. This map shows the probability of damage to buildings in the entire disaster area.

[0085] In one implementation, after the consistency verification layer spatiotemporally aligns the first pre-disaster state probability vector and the first post-disaster state probability vector, it performs collapse probability difference value calculation and outputs a first local state probability vector map, including:

[0086] Y2461: Calculate the first collapse probability difference value of the first building based on the first pre-disaster state probability vector and the first post-disaster state probability vector; Y2462: If the first collapse probability difference value is greater than a preset difference threshold, calculate the first image block structure similarity index of the first pre-disaster building layout domain map and the first post-disaster building layout domain map; Y2463: If the first image block structure similarity index is less than a preset structure change threshold, output the first post-disaster state probability vector as the first local state probability vector map; Y2464: If the first image block structure similarity index is greater than or equal to the structure change threshold, weightedly fuse the first pre-disaster state probability vector and the first post-disaster state probability vector to output the first local state probability vector map; Y2465: If the first collapse probability difference value is less than or equal to the difference threshold, directly output the first post-disaster state probability vector as the first local state probability vector map.

[0087] The first collapse probability difference value is obtained by comparing the first pre-disaster state probability vector and the first post-disaster state probability vector. Specifically, this difference value quantifies the change of the building between the pre-disaster and post-disaster states. This difference value is used to determine whether the building has undergone significant changes. For example, if the post-disaster state probability of the building shows that it has been severely damaged or collapsed, while the pre-disaster state still shows that the building is intact, then the collapse probability difference value is large, indicating that the building may collapse.

[0088] The system determines whether the first collapse probability difference value exceeds a preset difference threshold. This threshold is set based on changes in post-disaster building damage. For example, if the collapse probability difference value exceeds the threshold, it indicates severe post-disaster building damage, requiring further verification of structural changes. The image patch structural similarity index measures the structural changes of buildings before and after the disaster. It assesses the similarity between pre- and post-disaster building distribution image patches, reflecting whether significant structural changes have occurred after the disaster. The calculation method involves using image matching techniques such as the structural similarity index or normalized mutual information to calculate the structural similarity between two images.

[0089] If the structural similarity index of the first image block is less than the preset structural change threshold, it indicates that the structural changes of the buildings before and after the disaster are significant, meaning that the damage to the buildings after the disaster is more severe. In this case, the first post-disaster state probability vector is directly used as the first local state probability vector. The post-disaster state probability vector can reflect the current status of the buildings, such as collapse or severe damage, and therefore can be directly used as the final damage assessment result.

[0090] If the structural similarity index of the first image block is greater than or equal to the preset structural change threshold, it indicates that the structural changes of buildings before and after the disaster are not significant. Therefore, the data before and after the disaster can be further fused. Weighted fusion means combining the state probability vectors before and after the disaster according to certain weights to obtain a comprehensive evaluation result. After weighted fusion, the output first local state probability vector map is a comprehensive building state evaluation map that can consider both pre-disaster and post-disaster information, providing a more comprehensive reference for the final post-disaster building damage assessment.

[0091] If the first collapse probability difference value is less than or equal to the preset difference threshold, it indicates that the changes before and after the disaster are not significant, meaning that the building's condition changes little before and after the disaster. Therefore, the post-disaster state probability vector can be directly used as the result of the building damage assessment. In this case, the pre-disaster state has not changed significantly, so the post-disaster assessment result is more accurate, and the post-disaster state probability vector is directly output as the first local state probability vector map.

[0092] Example 2: Based on the same inventive concept as the emergency response and rescue decision-making method based on an AI super command brain in the foregoing examples, this invention provides an emergency response and rescue decision-making system based on an AI super command brain. See [link to example]. Figure 2 As shown, the system includes:

[0093] The data receiving module 10 is used to receive multi-source heterogeneous disaster data through a distributed data gateway, obtaining dual-temporal satellite data, real-time disaster data, and real-time IoT sensor data; the damage identification module 20 is used to identify building damage based on pre-disaster remote sensing data and post-disaster remote sensing data in the dual-temporal satellite data, and output distributed disaster-affected area identifiers; the state transition prediction module 30 is used to predict the state transition of the real-time disaster data and output time-series disaster prediction data; the risk evolution analysis module 40 is used to perform risk evolution analysis of the distributed disaster-affected area identifiers based on the time-series disaster prediction data, and output distributed prediction area identifiers; the rescue strategy optimization module 50 is used to retrieve real-time traffic topology data of the disaster-affected area, project the real-time traffic topology data onto the distributed prediction area identifiers, perform rescue strategy optimization, and output rescue optimization schemes; the equipment collaborative scheduling module 60 is used to allocate rescue tasks according to the rescue optimization schemes after hierarchical and traceable approval based on blockchain, and execute collaborative scheduling of multiple types of rescue equipment.

[0094] In one implementation, the damage identification module 20 is used to perform the following operation steps:

[0095] The system employs a geographic information system to perform spatiotemporal alignment of pre-disaster and post-disaster remote sensing data; it uses an edge detection algorithm to extract the dual-temporal building outline baselines from the pre-disaster and post-disaster remote sensing data, outputting a dual-temporal building distribution vector map; it traverses the dual-temporal building distribution vector map to perform pixel-level grayscale difference calculation, outputting a building grayscale difference vector map; it traverses the dual-temporal building distribution vector map to perform state probability prediction, outputting a building state probability vector map; and it fuses the building grayscale difference vector map and the building state probability vector map to determine building collapse, outputting the distributed disaster-affected area identifier.

[0096] In one implementation, the state transition prediction module 30 is used to perform the following steps:

[0097] The real-time disaster data is subjected to continuous time-series disaster feature extraction to obtain a time-series disaster multidimensional feature vector; the time-series disaster multidimensional feature vector is input into a disaster evolution prediction model constructed based on a two-layer attention LSTM network to predict the disaster state transition and output time-series risk probability data and time-series disaster intensity data; after the time-series risk probability data and time-series disaster intensity data are identified by the spatial boundary of high-risk areas, structured data storage is performed to output the time-series disaster prediction data.

[0098] In one implementation, the rescue strategy optimization module 50 is used to perform the following steps:

[0099] The real-time traffic topology data is projected onto the spatial boundary of the distributed predicted area identifier, and traffic risk overlay and pruning are performed to output a feasible traffic network map. Based on the disaster-stricken area attributes, the distributed area rescue needs of the distributed predicted area identifier are matched, wherein the disaster-stricken area attributes include the number of trapped people, building collapse rate, and proportion of special groups. Using the distributed area rescue needs and feasible traffic network map as optimization inputs, rescue path planning is performed based on a multi-objective optimization algorithm to output a distributed rescue strategy, which constitutes the rescue optimization scheme.

[0100] In one implementation, the rescue strategy optimization module 50 is used to perform the following steps:

[0101] The system interactively obtains the shelf life of the first supplies needed for rescue in the first area and uses the shelf life of the first supplies as a constraint on the timeliness of rescue; it predefines the slope constraint of the rescue road and the consumption constraint of the rescue resources; it uses the constraints on the timeliness of rescue, the slope constraint of the rescue road and the consumption constraint of the rescue resources as hard constraints, and performs rescue route planning with the first disaster prediction area as the rescue destination on the feasible traffic network map, and outputs the first rescue strategy, wherein the first rescue strategy includes disaster avoidance route and resource allocation list.

[0102] In one implementation, the damage identification module 20 is used to perform the following operation steps:

[0103] Multiple sample satellite remote sensing images are obtained interactively, and multiple sample labeled data sets are obtained by identifying building outline coordinates and building status labels on the multiple sample satellite remote sensing images. The multiple sample satellite remote sensing images and multiple sample labeled data sets are used as training data to perform parameter tuning and optimization of pre-disaster feature extraction channels and post-disaster feature extraction channels based on DCNN models. After the pre-disaster feature extraction channels and post-disaster feature extraction channels are connected in parallel, a consistency verification layer is cascaded at the output of the pre-disaster feature extraction channels and post-disaster feature extraction channels to complete the construction of a dual-channel DCNN model.

[0104] In one implementation, the damage identification module 20 is used to perform the following operation steps:

[0105] Based on a preset segmentation scale, the dual-temporal building distribution vector map is rasterized to obtain multiple sets of dual-temporal building distribution domain maps with multiple raster indices. The first pre-disaster building distribution domain map and the first post-disaster building distribution domain map in the first set of dual-temporal building distribution domain maps are input in parallel into the pre-disaster feature extraction channel and the post-disaster feature extraction channel of the dual-channel DCNN model to perform state probability prediction and output the first pre-disaster state probability vector and the first post-disaster state probability vector. After the first pre-disaster state probability vector and the first post-disaster state probability vector are spatiotemporally aligned at the consistency verification layer, the collapse probability difference value is calculated and the first local state probability vector map is output. By analogy, the dual-channel DCNN model is driven to analyze the multiple sets of dual-temporal building distribution domain maps to obtain multiple local state probability vector maps. The multiple local state probability vector maps are spliced ​​according to the multiple raster index spaces to output the building state probability vector map.

[0106] In one implementation, the damage identification module 20 is used to perform the following operation steps:

[0107] Based on the first pre-disaster state probability vector and the first post-disaster state probability vector, calculate the first collapse probability difference value of the first building; if the first collapse probability difference value is greater than a preset difference threshold, calculate the first image block structure similarity index of the first pre-disaster building layout domain map and the first post-disaster building layout domain map; if the first image block structure similarity index is less than a preset structure change threshold, output the first post-disaster state probability vector as the first local state probability vector map; if the first image block structure similarity index is greater than or equal to the structure change threshold, weightedly fuse the first pre-disaster state probability vector and the first post-disaster state probability vector to output the first local state probability vector map; if the first collapse probability difference value is less than or equal to the difference threshold, directly output the first post-disaster state probability vector as the first local state probability vector map.

[0108] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0109] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. An emergency response and rescue decision-making method based on an AI super command brain, characterized in that: The method includes: Multi-source heterogeneous disaster data is received through a distributed data gateway, resulting in dual-temporal satellite data, real-time disaster data, and real-time IoT sensor data. Based on the pre-disaster and post-disaster remote sensing data in the dual-temporal satellite data, building damage is identified, and distributed disaster-affected area identifiers are output. Perform state transition prediction on the real-time disaster data and output time-series disaster prediction data; Based on the time-series disaster prediction data, perform risk evolution analysis on the distributed disaster-affected area identifiers and output distributed predicted area identifiers; After retrieving real-time traffic topology data of the disaster-stricken area, the real-time traffic topology data is projected onto the distributed prediction area identifier to optimize the rescue strategy and output the rescue optimization plan. After the rescue optimization plan is approved and recorded in a tiered manner based on blockchain, rescue tasks are assigned according to the rescue optimization plan, and multi-type rescue equipment is coordinated and dispatched.

2. The emergency response and rescue decision-making method based on an AI super command brain as described in claim 1, characterized in that, Based on the pre-disaster and post-disaster remote sensing data from the aforementioned dual-temporal satellite data, building damage is identified, and distributed disaster-affected area identifiers are output, including: The pre-disaster remote sensing data and the post-disaster remote sensing data are spatiotemporally aligned using a geographic information system. An edge detection algorithm is used to extract the building outline baselines from the pre-disaster and post-disaster remote sensing data in two time periods, and output a two-time period building distribution vector map. The pixel-level grayscale difference is calculated by traversing the dual-temporal building distribution vector map, and the building grayscale difference vector map is output. The state probability prediction is performed by traversing the dual-temporal building distribution vector map, and the building state probability vector map is output. The building collapse determination is made by fusing the building grayscale difference vector map and the building state probability vector map, and the distributed disaster area identifier is output.

3. The emergency response and rescue decision-making method based on an AI super command brain as described in claim 1, characterized in that, Perform state transition prediction on the real-time disaster data and output time-series disaster prediction data, including: Continuous time-series disaster features are extracted from the real-time disaster data to obtain a multi-dimensional feature vector of time-series disasters; The multi-dimensional feature vector of the time-series disaster is input into the disaster evolution prediction model constructed based on a two-layer attention LSTM network to predict the disaster state transition and output time-series risk probability data and time-series disaster intensity data. After identifying the temporal risk probability data and temporal disaster intensity data using the spatial boundary of high-risk areas, structured data storage is performed, and the temporal disaster prediction data is output.

4. The emergency response and rescue decision-making method based on an AI super command brain as described in claim 1, characterized in that, After retrieving real-time traffic topology data of the disaster-stricken area, the real-time traffic topology data is projected onto the distributed prediction area identifier to optimize the rescue strategy and output an optimized rescue plan, including: The real-time traffic topology data is projected onto the spatial boundary of the distributed prediction area, and traffic risk overlay and pruning are performed to output a feasible traffic network map. Distributed regional rescue needs are matched based on the disaster-affected area attributes with the distributed predicted area identifier, wherein the disaster-affected area attributes include the number of people trapped, the building collapse rate, and the proportion of special groups. Using the distributed regional rescue needs and feasible traffic network map as optimization inputs, rescue route planning is performed based on a multi-objective optimization algorithm, and a distributed rescue strategy is output, which constitutes the rescue optimization scheme.

5. The emergency response and rescue decision-making method based on an AI super command brain as described in claim 4, characterized in that, Using the distributed regional rescue needs and feasible traffic network map as optimization inputs, a multi-objective optimization algorithm is used for rescue route planning, outputting a distributed rescue strategy to constitute the rescue optimization scheme, including: The system interactively obtains the shelf life of the first supplies needed for rescue in the first area and uses the shelf life of the first supplies as a constraint on the timeliness of rescue efforts. Predefine the slope constraints of the rescue road and the consumption constraints of rescue resources; Using the constraints of rescue timeliness, rescue road gradient, and rescue resource consumption as hard constraints, rescue route planning with the first disaster prediction area as the rescue destination is performed on the feasible traffic network map, and a first rescue strategy is output. The first rescue strategy includes disaster avoidance routes and resource allocation lists.

6. The emergency response and rescue decision-making method based on an AI super command brain as described in claim 2, characterized in that, The state probability prediction is performed by traversing the dual-temporal building distribution vector map, and the building state probability vector map is output. Prior to this, the following steps are included: Multiple sample satellite remote sensing images are obtained interactively, and multiple sample labeled data groups are obtained by identifying building outline coordinates and building status labels on the multiple sample satellite remote sensing images; The multiple sample satellite remote sensing images and multiple sample labeled data sets were used as training data to perform parameter tuning and optimization of the pre-disaster feature extraction channel and the post-disaster feature extraction channel based on the DCNN model. After connecting the pre-disaster feature extraction channel and the post-disaster feature extraction channel in parallel, a consistency verification layer is cascaded at the output of the pre-disaster feature extraction channel and the post-disaster feature extraction channel to complete the construction of the dual-channel DCNN model.

7. The emergency response and rescue decision-making method based on an AI super command brain as described in claim 6, characterized in that, The state probability is predicted by traversing the dual-temporal building distribution vector map and outputting the building state probability vector map, including: Based on the preset segmentation scale, the dual-temporal building distribution vector map is rasterized to obtain multiple sets of dual-temporal building distribution domain maps with multiple raster indexes; The first pre-disaster building layout domain map and the first post-disaster building layout domain map in the first set of dual-temporal building layout domain maps are input in parallel into the pre-disaster feature extraction channel and the post-disaster feature extraction channel of the dual-channel DCNN model, and state probability prediction is performed to output the first pre-disaster state probability vector and the first post-disaster state probability vector. After the consistency verification layer spatiotemporally aligns the first pre-disaster state probability vector and the first post-disaster state probability vector, it performs the collapse probability difference value calculation and outputs the first local state probability vector map. By analogy, the dual-channel DCNN model is driven to analyze the multiple sets of dual-temporal building layout domain maps to obtain multiple local state probability vector maps. The building state probability vector map is output by stitching together the multiple local state probability vector maps based on the multiple raster index spaces.

8. The emergency response and rescue decision-making method based on an AI super command brain as described in claim 7, characterized in that, After the consistency verification layer spatiotemporally aligns the first pre-disaster state probability vector and the first post-disaster state probability vector, it performs collapse probability difference value calculation and outputs the first local state probability vector map, including: Based on the first pre-disaster state probability vector and the first post-disaster state probability vector, calculate the first collapse probability difference value of the first building; If the first collapse probability difference value is greater than the preset difference threshold, then calculate the first image block structure similarity index of the first pre-disaster building layout domain map and the first post-disaster building layout domain map. If the structural similarity index of the first image block is less than the preset structural change threshold, then the first post-disaster state probability vector is output as the first local state probability vector. If the structural similarity index of the first image block is greater than or equal to the structural change threshold, then the first pre-disaster state probability vector and the first post-disaster state probability vector are weighted and fused to output the first local state probability vector map. If the first collapse probability difference value is less than or equal to the difference threshold, then the first post-disaster state probability vector is directly output as the first local state probability vector map.

9. An emergency response and rescue decision-making system based on an AI super command brain, characterized in that: The system is used to implement the emergency response and rescue decision-making method based on an AI super command brain as described in any one of claims 1-8, the system comprising: The data receiving module is used to receive multi-source heterogeneous disaster data through a distributed data gateway, and obtain dual-temporal satellite data, real-time disaster data and real-time IoT sensor data. The damage identification module is used to identify building damage based on pre-disaster remote sensing data and post-disaster remote sensing data in the dual-temporal satellite data, and output distributed disaster-affected area identifiers. The state transition prediction module is used to perform state transition prediction on the real-time disaster data and output time-series disaster prediction data. The risk evolution analysis module is used to perform risk evolution analysis on the distributed disaster-affected area identifier based on the time-series disaster prediction data, and output the distributed prediction area identifier. The rescue strategy optimization module is used to retrieve real-time traffic topology data of the disaster-stricken area, project the real-time traffic topology data onto the distributed prediction area identifier, perform rescue strategy optimization, and output an optimized rescue plan. The equipment collaborative scheduling module is used to allocate rescue tasks according to the rescue optimization plan after the rescue optimization plan is approved and recorded based on blockchain, and to perform collaborative scheduling of multiple types of rescue equipment.

Images