Geological disaster emergency evacuation path intelligent planning system based on edge calculation

Abstract

The invention relates to the technical field of geological disaster emergency, and discloses a geological disaster emergency evacuation path intelligent planning system based on edge calculation. The system comprises a terrain monitoring layer, a risk assessment layer, a path deduction layer and an evacuation control layer. The terrain monitoring layer collects a multi-dimensional terrain data flow of a disaster area in real time through an edge computing node; the risk assessment layer fuses the historical catastrophe model and the multi-dimensional topographic data flow to carry out geological stability calculation, and generates a dynamic risk distribution map; the path deduction layer generates an initial evacuation path network through a self-adaptive optimization algorithm in combination with the dynamic risk distribution map and a preset evacuation constraint condition; and the evacuation control layer converts the initial evacuation path network into a navigation instruction set and drives terminal execution equipment to implement evacuation guidance. According to the system, the terrain change of a disaster area can be mastered in real time, the risk is accurately evaluated, an evacuation path fitting the reality is generated, personnel evacuation is effectively guided, and the efficiency and safety of emergency evacuation of geological disasters are improved.

Description

Technical Field

[0001] This invention relates to the field of geological disaster emergency technology, specifically to an intelligent planning system for geological disaster emergency evacuation routes based on edge computing. Background Technology

[0002] Geological disasters are characterized by their suddenness and destructive power, often posing a serious threat to people's lives and property. In geological disaster emergency response, the planning of emergency evacuation routes is a crucial measure to reduce casualties and property losses. Traditional methods of geological disaster emergency evacuation route planning mostly rely on pre-established static plans, which are usually based on historical data and experience-based judgments, making them ill-suited to the complex and ever-changing on-site conditions during a disaster. With the development of technology, some computer-based path planning methods have been gradually applied to the field of geological disaster emergency response, but many limitations still exist. For example, some systems rely on cloud computing for data processing and path deduction. Since geological disasters are often accompanied by communication interruptions or network congestion, the response speed of cloud computing is greatly affected, making it impossible to generate effective evacuation routes in a timely manner. At the same time, existing terrain monitoring methods are relatively simple, mostly focusing on monitoring a specific terrain parameter, which makes it difficult to comprehensively and accurately reflect the real-time terrain changes in the disaster area, resulting in significant safety hazards in evacuation routes generated based on this monitoring data. Most existing risk assessment models are based on historical disaster data and lack the integration and analysis of real-time terrain data streams. This results in significant discrepancies between the risk assessment results and the actual conditions at the disaster site, failing to provide a reliable basis for evacuation route planning. Moreover, the route simulation process often fails to fully consider various constraints during evacuation, such as the number of evacuees and the carrying capacity of evacuation routes. Consequently, the generated evacuation routes are difficult to implement effectively in practice, impacting the efficiency and effectiveness of emergency evacuation. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent planning system for emergency evacuation routes in geological disasters based on edge computing, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, this invention provides an intelligent planning system for emergency evacuation routes in geological disasters based on edge computing, the system comprising: The terrain monitoring layer is used to collect multi-dimensional terrain data streams of disaster areas in real time through edge computing nodes; The risk assessment layer is used to integrate historical disaster models with the multidimensional terrain data stream to perform geological stability calculations and generate a dynamic risk distribution map. The path deduction layer is used to combine the dynamic risk distribution map with the preset evacuation constraints and generate an initial evacuation path network through an adaptive optimization algorithm. The evacuation control layer is used to convert the initial evacuation path network into a navigation instruction set and drive the terminal execution device to implement evacuation guidance.

[0005] Preferably, the terrain monitoring layer includes: Multi-source sensing units are deployed at edge nodes of disaster sites to acquire real-time data on surface displacement, groundwater level, and meteorological data. The terrain modeling unit performs spatiotemporal alignment processing on the surface displacement data, groundwater level data, and meteorological data to construct a three-dimensional terrain change matrix. The data compression unit uses a wavelet transform algorithm to perform layered compression on the three-dimensional terrain change matrix to generate the multi-dimensional terrain data stream.

[0006] Preferably, the risk assessment layer includes: The model loading unit calls up the geological structure parameters and disaster triggering thresholds from the historical disaster database. The real-time analysis unit inputs the multidimensional terrain data stream and the geological structure parameters into a convolutional recurrent neural network and outputs a geological deformation trend vector. The map generation unit classifies and labels the geological deformation trend vector according to the disaster triggering threshold, forming the dynamic risk distribution map with risk levels.

[0007] Preferably, the path deduction layer includes: The constraint parsing unit parses the path safety threshold and the upper limit of traffic capacity in the preset evacuation constraints. The network construction unit divides safe passage areas and high-risk restricted areas based on the dynamic risk distribution map; The path generation unit applies an ant colony optimization algorithm within the safe passage area, calculates the path passage weights based on the upper limit of passage capacity, and generates the weighted initial evacuation path network.

[0008] Preferably, the evacuation control layer includes: The instruction conversion unit maps the path nodes in the initial evacuation path network into a navigation coordinate sequence; The device adaptation unit converts the navigation coordinate sequence into device-parseable instructions based on the positioning accuracy and communication protocol of the terminal execution device. The execution drive unit sends an encrypted instruction data packet to the terminal execution device to activate the audio-visual guidance device.

[0009] Preferably, the system further includes a feedback optimization layer, the feedback optimization layer further including: The performance evaluation unit compares the deviation between personnel evacuation status data and preset evacuation efficiency indicators. The dynamic correction unit triggers the path inference layer to recalculate the local path network when the deviation exceeds the fault tolerance threshold.

[0010] Preferably, the feedback optimization layer further includes: The status tracking unit analyzes the movement speed and gathering density in the personnel evacuation status data in real time; The bottleneck detection unit identifies path nodes whose movement speed is lower than a preset speed threshold; The topology reconfiguration unit adjusts the path network topology weights based on the cluster density distribution and generates path update instructions.

[0011] Preferably, the dynamic correction unit includes: The incremental calculation unit performs local neighborhood path replanning based on the path update instruction. The conflict resolution unit verifies the resource usage conflict between the newly planned path and the existing path; The instruction iteration unit merges the newly planned paths after conflict resolution into the initial evacuation path network.

[0012] Preferably, the system further includes: The communication relay unit uses a multi-hop transmission protocol to synchronize the dynamic risk distribution map between edge computing nodes.

[0013] Preferably, the system further includes: The encryption gateway unit encrypts the navigation instruction set using quantum key encryption and then distributes it to the terminal execution device.

[0014] Compared with the prior art, the beneficial effects of the present invention are: By utilizing edge computing nodes to collect multi-dimensional terrain data streams from disaster areas in real time through a terrain monitoring layer, rich and real-time terrain information can be obtained quickly and easily. This avoids the latency issues in data transmission and processing inherent in traditional cloud computing, ensuring timely understanding of terrain changes in disaster areas. This real-time monitoring data provides a solid foundation for subsequent risk assessment and route planning, enabling the entire system to closely align with the actual conditions at the disaster site. The risk assessment layer integrates historical disaster models with multidimensional topographic data streams to perform geological stability calculations and generate dynamic risk distribution maps, breaking the limitations of traditional risk assessments that rely solely on historical data. Historical disaster models contain the experience and patterns of past disasters, while real-time multidimensional topographic data streams reflect the current disaster situation. The combination of the two makes risk assessment more comprehensive and accurate, dynamically displaying risk changes in disaster areas, allowing planners to clearly understand the degree of danger in different areas, and providing direction for evacuation route planning. The path deduction layer combines a dynamic risk distribution map with preset evacuation constraints, and generates an initial evacuation path network through an adaptive optimization algorithm, fully considering the actual constraints at the disaster site. Preset evacuation constraints, such as the distribution of evacuees and the condition of passageways, are key factors affecting evacuation effectiveness. The adaptive optimization algorithm can intelligently adjust path planning strategies based on these conditions and the dynamic risk distribution, resulting in an initial evacuation path network that better meets actual needs, improving the feasibility and safety of evacuation paths. The evacuation control layer transforms the initial evacuation route network into a navigation instruction set and drives the terminal execution devices to implement evacuation guidance, achieving a seamless connection from route planning to actual evacuation action. The navigation instruction set is clear and unambiguous, effectively guiding evacuees to leave along the planned routes. The driving of the terminal execution devices ensures the effective execution of the instructions, avoiding chaos and blind spots during the evacuation process, guaranteeing the orderly conduct of emergency evacuation work, and helping to evacuate people to safe areas in the shortest possible time, reducing casualties and property losses caused by disasters. Attached Figure Description

[0015] Figure 1 This is a timing diagram of the intelligent planning system for geological disaster emergency evacuation routes based on edge computing as described in this invention. Figure 2 Flowchart for working on the terrain monitoring layer; Figure 3 A flowchart illustrating the work of the path deduction layer; Figure 4 A flowchart for the feedback optimization layer work (extended part); Figure 5 A flowchart illustrating the operation of the dynamic correction unit. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1 This invention provides an intelligent planning system for emergency evacuation routes in geological disasters based on edge computing. The system includes a terrain monitoring layer, a risk assessment layer, a route simulation layer, and an evacuation control layer, and optionally a feedback optimization layer, a communication relay unit, and an encryption gateway unit.

[0018] The terrain monitoring layer collects multi-dimensional terrain data streams of the disaster area in real time. The risk assessment layer integrates historical disaster models and data streams to generate a dynamic risk distribution map. The path deduction layer combines the map with constraints to generate an initial evacuation path network. The evacuation control layer transforms the path network into navigation commands and drives terminal execution devices to implement guidance. Each layer works collaboratively through edge computing nodes, supporting real-time data processing and rapid response, adapting to the needs of dynamically changing geological disaster scenarios.

[0019] Example 1: See Figure 2 The terrain monitoring layer, serving as the perception foundation of the entire system, primarily consists of multi-source sensing units, terrain modeling units, and data compression units working collaboratively to achieve real-time acquisition, processing, and efficient transmission of multi-dimensional terrain data from disaster areas. Multi-source sensing units are deployed at edge nodes of the disaster site. These nodes are typically selected at key locations in disaster-prone areas, such as the leading edge of landslides, flood-prone areas along rivers, and transition zones between steep slopes and gentle terrain, ensuring coverage of the main areas affected by the disaster. The unit contains various types of sensors. Surface displacement data is acquired via a GNSS receiver, which is mounted on reinforced bedrock or a stable surface structure to avoid interference with the measurement results due to its own displacement. Its sampling frequency is 1 Hz, capable of recording the planar coordinates and elevation changes of surface points with millimeter-level accuracy. Groundwater level data is collected by a piezometer, which is buried in a groundwater level monitoring well. The well depth is set from 5 to 20 meters depending on geological conditions. The sensor indirectly obtains the groundwater level by measuring pore water pressure, with a sampling frequency of 0.5 Hz and centimeter-level accuracy. Meteorological data is acquired by an anemometer and a rain gauge. The anemometer is mounted on an open, unobstructed mountaintop or high-altitude support, while the rain gauge uses a tipping bucket structure. Both collect data at a frequency of 2 Hz, recording wind speed, wind direction, and rainfall, respectively. These sensors are connected to edge computing nodes via wired or wireless means, transmitting the collected raw data in real time to the storage module within the node for temporary storage.

[0020] As data continues to be imported, the terrain modeling unit begins spatiotemporal alignment of the multi-source data. Time alignment is a crucial step. Due to potential slight deviations in the clocks of different sensors, the edge computing nodes incorporate high-precision real-time clock modules and synchronize with a remote time server via Network Time Protocol (NTP) to ensure that the timestamp error of all sensor data is controlled within 10 milliseconds. For occasional clock drift, the unit automatically detects and adjusts. For example, if the timestamp of a GNSS receiver deviates from the node's master clock by more than 50 milliseconds, the system triggers a correction procedure, using an interpolation algorithm to align the timestamps of subsequent data to the master clock reference. Spatial alignment is equally important. All sensors undergo coordinate measurement before deployment, using RTK (Real-Time Kinematic) positioning technology to obtain their precise latitude, longitude, and altitude in the WGS84 geographic coordinate system, and storing this coordinate information in the sensor's configuration file. During data acquisition, each data point is automatically appended with corresponding coordinate information. For coordinate missing issues caused by sensor failure or signal loss, the terrain modeling unit uses linear interpolation to complete the mapping. For example, if a piezometer is missing data from the 100th to the 105th second, the system will perform linear interpolation based on the coordinate data from the 99th and 106th seconds to estimate the coordinate values ​​for the missing time period. After completing temporal and spatial alignment, the unit will fuse the scattered surface displacement data, groundwater level data, and meteorological data to construct a three-dimensional terrain change matrix. This matrix has time as its first dimension, with each time point corresponding to a two-dimensional plane matrix. The rows and columns of the plane matrix correspond to longitude and latitude, respectively, and each matrix element stores the surface displacement and groundwater level change at that location.

[0021] To meet the real-time transmission requirements of edge computing nodes, the data compression unit needs to efficiently compress the 3D terrain change matrix. The unit employs wavelet transform as the core compression tool, which can significantly reduce data volume while preserving key data features. Specifically, the time series processing performs discrete wavelet decomposition on the surface displacement and groundwater level time series data for each latitude and longitude location. A single-level decomposition using Haar wavelet basis functions is selected, decomposing the original time series into high-frequency detail coefficients and low-frequency approximation coefficients. High-frequency detail coefficients capture abrupt changes within a short period, such as sudden small landslides or rapid water level rises caused by heavy rain; low-frequency approximation coefficients reflect slow, long-term trends, such as groundwater level rises caused by slow infiltration due to continuous rainfall. Spatial dimension processing involves performing two-dimensional wavelet decomposition in both longitude and latitude directions. Taking the surface displacement matrix as an example, the two-dimensional decomposition divides the matrix into four sub-matrices: horizontal low-frequency-vertical low-frequency (LL), horizontal low-frequency-vertical high-frequency (LH), horizontal high-frequency-vertical low-frequency (HL), and horizontal high-frequency-vertical high-frequency (HH). The LL sub-matrix contains global, slowly changing information, while the LH and HL sub-matrices reflect local abrupt changes in the horizontal and vertical directions, respectively. The HH sub-matrix captures drastically changing areas on the two-dimensional plane (such as the boundaries of landslide bodies). After completing the multidimensional wavelet decomposition, the data compression unit filters the coefficients according to a preset compression ratio (usually set to 1:8), prioritizing the retention of low-frequency approximation coefficients containing key information and significant portions of high-frequency detail coefficients. The resulting compressed multidimensional terrain data stream has a data volume reduced by approximately 80% compared to the original data, significantly reducing transmission bandwidth requirements. This enables real-time transmission even with limited computing resources at edge computing nodes, ensuring that the subsequent risk assessment layer can promptly obtain the latest terrain data for disaster analysis.

[0022] The core function of the risk assessment layer is to combine the multi-dimensional topographic data stream provided by the topographic monitoring layer with historical disaster models to generate a dynamic risk distribution map reflecting the current geological stability. This process is completed collaboratively by the model loading unit, the real-time analysis unit, and the map generation unit. The model loading unit pre-stores a historical disaster database, which contains a large amount of relevant data on historical disaster events, mainly including two types of information: geological structural parameters and disaster triggering thresholds. Geological structural parameters cover basic parameters reflecting regional geological conditions, such as rock hardness (obtained through uniaxial compressive strength testing, in megapascals), fault distribution density (number of faults per square kilometer), and soil shear strength (internal friction angle and cohesion, in degrees and kilopascals, respectively). These parameters are accumulated over a long period through methods such as geological drilling, geophysical exploration, and laboratory testing. The disaster triggering threshold is a critical value derived from statistical analysis of historical disaster events. For example, when the surface displacement rate exceeds 5 mm / h, the groundwater level rise rate exceeds 10 cm / h, or the slope exceeds 30°, the risk of landslides or debris flows increases significantly.

[0023] Upon receiving the multidimensional terrain data stream from the terrain monitoring layer, the real-time analysis unit extracts the current surface displacement rate (calculated by dividing the displacement difference between two adjacent time points by the time interval, in millimeters per hour), the groundwater level rise rate (similarly, in centimeters per hour), and real-time slope information (calculated by the ratio of the elevation difference to the horizontal distance between adjacent grids, in degrees, from the elevation data in the three-dimensional terrain change matrix). The unit combines these real-time parameters with the geological structure parameters from the model loading unit and inputs them into a convolutional recurrent neural network (CRNN) for processing. The CRNN's network structure design considers both spatial and temporal feature extraction: the convolutional layers use 3×3 convolutional kernels to perform sliding window operations on the two-dimensional terrain matrix, extracting features such as local spatial displacement hotspots and abrupt slope changes; the recurrent layers employ a long short-term memory (LSTM) network to model the displacement rate and groundwater level rise rate over time, capturing trend changes between consecutive time points, such as identifying whether the displacement rate shows a continuous increasing trend. The outputs of the convolutional and recurrent layers are merged by the feature fusion layer and then passed to the fully connected layer, where a geological deformation trend vector is output through the Softmax activation function.

[0024] The map generation unit classifies and labels geological deformation trend vectors based on trigger thresholds in the historical disaster database, forming a dynamic risk distribution map. Specifically, for each risk indicator in the trend vector, if its probability value exceeds 0.7, it is marked as high-risk (corresponding to red); if the probability value is between 0.4 and 0.7, it is marked as medium-risk (corresponding to yellow); and if the probability value is below 0.4, it is marked as low-risk (corresponding to green). This risk level information is overlaid on the Geographic Information System (GIS) base map to form a visualized dynamic risk distribution map. The map update frequency is synchronized with the terrain data stream acquisition frequency, typically once per second, ensuring real-time reflection of changes in geological stability. For example, when the rate of groundwater level rise in a certain area suddenly increases from 8 cm / h to 12 cm / h, the probability value of the water level increase risk calculated by the real-time analysis unit increases from 0.5 to 0.8. The map generation unit will update the risk level of that area from yellow to red and mark the area in a prominent red on the GIS map, alerting relevant personnel that the risk of disaster in that area is high.

[0025] To ensure that the dynamic risk distribution map accurately reflects actual geological conditions, the risk assessment layer also needs to consider the spatial resolution and noise impact of the terrain data. The spatial resolution of the three-dimensional terrain change matrix collected by the terrain monitoring layer is 10 meters × 10 meters, meaning that each matrix element corresponds to a 10-meter × 10-meter area on the actual ground. This resolution is sufficient for most disaster scenarios, but it may miss minor local terrain changes. Therefore, the real-time analysis unit performs spatial smoothing on the three-dimensional terrain change matrix, using a 3×3 mean filter to average each element and its eight neighboring elements to eliminate the influence of random noise and retain the main terrain change features. In addition, the historical disaster database in the model loading unit is updated regularly. Whenever a new disaster event occurs, the system automatically collects relevant data for that event (including pre-disaster terrain change data, triggering conditions at the time of the disaster, and the post-disaster impact range) and adds it to the database to optimize the training parameters of the convolutional recurrent neural network and improve the accuracy of risk prediction. For example, when a small landslide occurs in a certain area, but the risk level predicted by the original model is medium, the system will use the terrain data, triggering conditions and actual results of this event as new training samples to adjust the network weights.

[0026] The generation of dynamic risk distribution maps relies not only on real-time topographic data and historical models, but also on the influence of other environmental factors, such as precipitation intensity and vegetation cover. While these factors are not explicitly included in the data stream of the topographic monitoring layer, they can be obtained through external data interfaces (such as real-time precipitation data interfaces from meteorological departments and remote sensing data interfaces for vegetation cover from forestry departments) and input into the real-time analysis unit for comprehensive processing. For example, during periods of heavy precipitation, even if the surface displacement rate does not reach the trigger threshold, the accelerated rise in groundwater level can lead to soil saturation, reducing shear strength and thus increasing the risk of landslides.

[0027] Example 2: See Figure 3 The core function of the path deduction layer is to generate an initial evacuation path network that meets both safety and efficiency requirements by combining a dynamic risk distribution map with preset evacuation constraints. This process is completed collaboratively by the constraint parsing unit, the network construction unit, and the path generation unit. The constraint parsing unit receives evacuation constraints preset by the system or manually input, which contain two types of key information: path safety thresholds and upper limits of traffic capacity.

[0028] The network building unit delineates safe passage areas and high-risk restricted areas based on a dynamic risk distribution map. The dynamic risk distribution map uses a Geographic Information System (GIS) base map as its carrier, overlaying the risk levels (high, medium, and low) and specific risk elements of each region. The network building unit extracts low-risk and medium-risk areas (i.e., safe passage areas) from the map and identifies the connectivity of these areas using a region growing algorithm: selecting population gathering points within the disaster area as initial seed points, and progressively expanding adjacent grids (spatial resolution of 10m × 10m) that meet safety thresholds (slope ≤ 25 degrees, non-river buffer zones, non-landslide areas) to form continuous connected blocks, each of which represents a safe passage area. For high-risk restricted areas, the network building unit marks them as impassable areas on the GIS map based on high-risk areas in the risk distribution map and prohibited units in the path safety threshold. The boundaries of these areas are precisely defined using polygon vector data to ensure that they are not crossed during path generation. In addition, the network building unit also integrates other physical attributes of the terrain: the slope data obtained through the terrain monitoring layer (calculated by the ratio of the elevation difference between adjacent grids to the horizontal distance by the three-dimensional terrain change matrix) is used to mark the slope value on each grid in the safe passage area; the road width information (such as 1.2 meters for paved roads and 0.8 meters for dirt roads) is obtained through road vector data (basic geographic information from the civil affairs department) and these attributes are associated with and stored with the safe area.

[0029] The path generation unit applies an ant colony optimization algorithm to generate an initial evacuation path network within the safe passage area. The ant colony optimization algorithm simulates the behavior of ants searching for food paths, utilizing the positive feedback mechanism of pheromones to find the optimal path. Specifically, the evacuation start and end points are determined: the start point is a gathering point of people within the disaster area, and the end point is a pre-defined safe refuge location; these locations are pre-marked using a GIS system. Each start point is initialized as an "anthill," and each "ant" represents an evacuation individual, randomly selecting adjacent safe nodes (i.e., connected grid points within the safe passage area) from the anthill to move. During movement, the ants calculate the selection probability based on multiple path attributes: path length (Euclidean distance, shorter is better), risk level (obtained from a dynamic risk distribution map; lower risk, higher probability), and remaining passage capacity (the difference between the currently used capacity and the upper limit of passage capacity; larger remaining capacity, higher probability). For example, when an ant is at a certain node, there are three adjacent safe nodes: Path A is 100 meters long, low risk level (probability weight 0.8), and has a remaining capacity of 200 people / hour; Path B is 150 meters long, medium risk level (probability weight 0.5), and has a remaining capacity of 300 people / hour; Path C is 80 meters long, high risk level (probability weight 0.3), and has a remaining capacity of 50 people / hour. The ant is most likely to choose Path A (overall probability = length weight × risk weight × capacity weight). After each ant completes a path selection, the pheromone concentration on the path is updated according to the total cost of the path (length cost × 0.4 + risk cost × 0.6): the lower the cost of the path, the more pheromone is added (for example, a path with a cost of 10 adds 2 units of pheromone, and a path with a cost of 20 adds 1 unit of pheromone). After multiple iterations (e.g., 100 times), the distribution of pheromone on the paths gradually stabilizes. At this point, the top 10% of paths with the highest pheromone concentration are selected as the initial evacuation path network. These paths are stored as weighted network structures, with weights representing the overall cost value, and are used to generate subsequent navigation instructions.

[0030] The core function of the evacuation control layer is to transform the initial evacuation path network into navigation commands recognizable by the terminal execution devices and drive the devices to implement evacuation guidance. This process is completed collaboratively by the command conversion unit, the device adaptation unit, and the execution drive unit. The command conversion unit first converts the topology of the initial evacuation path network into a navigation coordinate sequence. The initial evacuation path network consists of nodes (latitude and longitude coordinates) and edges (connections between nodes). Nodes include the starting point, intermediate turning points, and the ending point. To reduce data volume and improve navigation efficiency, the command conversion unit uses the Douglas-Pock algorithm to simplify the path: it traverses the nodes in the path, calculates the deviation of the current node from the lines connecting it to the nodes before and after it, and removes the node if the deviation is less than a preset threshold (e.g., 5 meters); otherwise, it retains the node. For example, a path containing 20 nodes may, after processing by the Douglas-Pock algorithm, retain only 5 key nodes (the starting point, two major turning points, and the ending point), with the coordinate accuracy of these nodes controlled within 5 meters to ensure navigation accuracy. The simplified path is converted into a coordinate sequence in the format [start coordinates, intermediate node 1 coordinates, intermediate node 2 coordinates, ..., end coordinates]. Each coordinate includes longitude, latitude, and altitude (optional, for high-precision navigation).

[0031] The device adaptation unit needs to adjust navigation commands according to the type and performance of the terminal execution device to ensure that different devices can receive and use navigation information normally. For smartphones, the positioning accuracy is usually 5-10 meters (when the GPS signal is good), and they support receiving data in JSON format. Therefore, the device adaptation unit converts the coordinate sequence into the GCJ-02 coordinate system (China National Bureau of Surveying and Mapping encrypted coordinate system) and encapsulates it in JSON format, including a coordinate list, suggested speed (e.g., walking speed 1.2 m / s), and a list of voice prompts (each node corresponds to one voice prompt). For in-vehicle navigation devices, the positioning accuracy is higher (usually 2-5 meters), and they are connected to the vehicle control system via CAN bus. Therefore, the commands are converted into the WGS84 coordinate system (Global Positioning System original coordinate system) and transmitted via a binary protocol (such as the CAN bus protocol), including path coordinates, turning commands (left turn, right turn, straight), and speed limit information (e.g., speed limit of 30 km / h for the road section). For wearable devices such as smart bracelets, their small screen size and limited processing power mean they can only display simple navigation information. Therefore, the device adaptation unit extracts key nodes from the coordinate sequence (one node every 500 meters) and simplifies voice prompts into short texts (such as "turn right") or vibration commands (short vibrations indicate turning, long vibrations indicate reaching the node) to ensure the information is concise and easy to understand. Furthermore, the device adaptation unit also considers differences in terminal communication protocols: for example, older mobile phones may only support GPRS communication, so commands are compressed into binary format to reduce data transmission; while smartphones support 4G / 5G communication and can use JSON or XML formats to transmit more detailed information.

[0032] The execution drive unit is responsible for securely and reliably sending the adapted navigation instruction set to the terminal execution device and activating its audio-visual guidance device. The instruction set needs to be encrypted to prevent malicious attacks or data tampering. The system uses the AES-256 symmetric encryption algorithm. The key is generated by XORing the system master key with the terminal device's unique ID (such as IMEI number), ensuring that each terminal's key is unique and dynamically changes. The encryption process converts the navigation instruction set (including coordinate sequences, voice prompts, suggested speeds, etc.) into ciphertext. The ciphertext length is the same as the original text, and the encryption time is controlled within milliseconds to ensure real-time performance. The encrypted instruction data packet is sent to the terminal execution device via wireless communication methods such as 4G / 5G, Wi-Fi, or Bluetooth: for smartphones, it is usually transmitted via 4G / 5G networks; for in-vehicle navigation systems, it can be transmitted via vehicle-to-everything (V2X) technology; for smart bracelets, it can be indirectly received after connecting to a mobile phone via Bluetooth Low Energy (BLE) technology. After receiving the encrypted data packet, the terminal device uses a pre-stored private key (associated with the system master key) to decrypt it and recover the original navigation instructions.

[0033] The decrypted navigation commands are parsed by the navigation module of the terminal execution device, activating the audio-visual guidance system. The smartphone's navigation interface highlights the planned route on the map, marks completed and incomplete sections with different colors, and plays voice prompts through the speaker. The in-vehicle navigation system displays the route on the in-vehicle display screen, simultaneously broadcasting turn commands via a voice synthesis module and activating the vehicle's lighting system. The smart bracelet uses a vibration motor to generate short vibrations (approximately 0.5 seconds) to indicate a turn is needed, and long vibrations (approximately 1.5 seconds) to indicate arrival at a path node. The screen displays simple arrow icons (up for going straight, left / right arrows for turning), ensuring users can obtain navigation information without looking at the screen. Furthermore, the execution drive unit monitors the terminal device's reception status in real time: if a terminal does not receive commands for an extended period (more than 30 seconds), the system resends encrypted data packets and notifies the user to check the device status via SMS or voice call (through the carrier interface).

[0034] Example 3: See Figure 4 The feedback optimization layer, as the system's adaptive adjustment module, dynamically adjusts the path network to improve the adaptability of evacuation guidance by tracking the evacuation process status in real time, evaluating evacuation efficiency, and identifying bottlenecks. Its core consists of a performance evaluation unit, a dynamic correction unit, a status tracking unit, a bottleneck detection unit, and a topology reconstruction unit. These units work collaboratively to achieve closed-loop optimization of the evacuation paths.

[0035] The performance evaluation unit first acquires personnel evacuation status data. This data originates from the positioning information of the terminal execution devices. The movement trajectory of each evacuee is collected in real time through positioning modules such as GPS and Beidou, including information such as timestamps (accurate to milliseconds), latitude and longitude coordinates (accuracy better than 5 meters), and altitude (optional). After being aggregated by edge computing nodes, a global evacuation status database is formed. Preset evacuation efficiency indicators are set according to the disaster type, regional scale, and shelter capacity. For example, for evacuation scenarios in mountainous villages, the preset indicators might be: evacuate 30% of the target population within 10 minutes of evacuation starting, 80% within 30 minutes, and complete all evacuation within one hour. Actual evacuation rate. The calculation method is the ratio of the number of people who have arrived at the evacuation site at a certain moment to the total number of people who need to be evacuated. The number of people who have arrived is obtained by matching the terminal location information with the geographical range of the evacuation site (when the terminal coordinates enter the evacuation site buffer zone, it is marked as evacuated). Target evacuation rate Then it is retrieved by looking up a table based on the time progress, for example, the 10th minute corresponds to... The 30th minute corresponds to .

[0036] Deviation The calculation formula is:

[0037] in: Indicates the degree of deviation. Indicates the actual evacuation rate. Indicates the target evacuation rate. When... When the fault tolerance threshold (e.g., 15%) is exceeded, the performance evaluation unit triggers the dynamic correction unit to start local path replanning.

[0038] The dynamic correction unit first identifies deviation areas: using the spatial query function of a Geographic Information System (GIS), it filters out areas where the actual evacuation rate is significantly lower than the target value (e.g., the actual evacuation rate of a certain settlement is only 10%, far below the target of 30%). For this area, the unit extracts the current dynamic risk distribution map and personnel distribution data. Simultaneously, the unit obtains real-time constraint updates for this area, such as reduced traffic capacity due to flooding of some routes (from 500 people / hour to 200 people / hour), or temporary closure of a route due to newly added small landslides. This information is obtained through real-time data streams from the terrain monitoring layer and feedback from on-site rescue personnel, and input into the constraint parsing unit of the route extrapolation layer to regenerate the constraint database for this area.

[0039] Based on the updated constraints, the network construction unit re-divides the safe passage zone: using a region growing algorithm, it expands connected components that meet the new safety thresholds (such as avoiding flooded areas and landslides) using the locations of the remaining evacuees as seed points, forming a new safe passage zone. The path generation unit recalculates paths within this zone using an improved ant colony optimization algorithm: during initialization, the locations of the remaining evacuees are used as new ant colonies, with each "ant" representing an evacuee. The calculation of movement probability considers not only path length, risk level, and remaining capacity, but also adds a time-sensitivity weight (paths closer to the target refuge have a higher time weight). After multiple iterations (e.g., 50 times), the top 20% of optimal paths are selected after pheromone convergence, generating a locally corrected path network. Update instructions (including newly added path coordinates and adjustments to the original path weights) are then sent to the evacuation control layer.

[0040] The status tracking unit analyzes the movement speed and cluster density in the personnel evacuation status data in real time. The movement speed is calculated based on the time-series data of the terminal positioning: for each terminal, three consecutive positioning points are selected. , , Calculate the time difference between the first two points. and distance To obtain instantaneous velocity Similarly, calculate the instantaneous velocities of the last two points. The final value is taken as the average moving speed of the terminal during that time period. The cluster density is calculated using kernel density estimation: a search radius is set with the location of each terminal as the center. (e.g., 50 meters), count the number of all terminals within this radius. Then the cluster density of the region The unit is people per square meter.

[0041] The bottleneck detection unit identifies path bottleneck nodes by analyzing movement speed and clustering density. Path nodes refer to intermediate turning points in the initial evacuation path network (such as the intersection of two paths), and each node is associated with its upstream path (from the starting point to the node) and downstream path (from the node to the destination). The unit traverses the movement trajectories of all terminals, calculates the arrival and departure times of each node, and determines the dwell time. If the average dwell time of a node exceeds a preset threshold (e.g., 1 minute) and the average movement speed of its downstream path is below a safe threshold (e.g., 0.5 m / s), then the node is marked as a bottleneck node. For example, if a node is located on the connecting path between two residential areas, and multiple groups of people gather there due to the narrow road, with a dwell time of up to 2 minutes and a downstream path speed of only 0.3 m / s, this node will be identified as a bottleneck.

[0042] The topology reconfiguration unit adjusts the path network topology weights based on clustering density and generates path update instructions. For high-clustering-density areas (such as...), The unit increases the weight of its adjacent paths (reducing travel costs) to guide the evacuation of people to suboptimal paths. For example, if an area has a high population density, the unit reduces the weight of the connecting paths between that area and surrounding safe areas by 15%, making subsequent people more likely to choose these paths. For bottleneck nodes, the unit adds a detour path: using Dijkstra's algorithm, it finds the shortest path from upstream to downstream of the bottleneck node within the safe passage area (avoiding the original bottleneck node), and increases the weight of the original path by 20% (increasing travel costs), reducing the probability of subsequent people choosing the original path. After the topology reconstruction is completed, the unit generates a path update instruction, which includes the coordinate sequence of the new path (simplified by the Douglas-Puk algorithm), the weight adjustment value of the original path, and the effective timestamp, and sends it to the terminal execution device through the evacuation control layer.

[0043] After receiving the update command, the terminal execution device dynamically adjusts the route display via the navigation module: terminals approaching the bottleneck will receive a voice prompt, "Road ahead is congested, please take the right-hand side road," and the new route will be highlighted on the map; terminals not yet reaching the bottleneck will directly update their navigation route and move along the optimized path. Simultaneously, the status tracking unit continuously monitors the updated evacuation status; if the deviation... If the evacuation rate still falls below the fault tolerance threshold, the dynamic correction unit will trigger path replanning again, forming a closed-loop optimization mechanism of "evaluation-correction-tracking-re-correction" to ensure that the evacuation process always proceeds in an efficient and safe direction.

[0044] Example 4: See Figure 5 The dynamic correction unit, as the core execution module of the feedback optimization mechanism, is mainly responsible for handling specific problems caused by changes in the disaster environment or local congestion during evacuation. The following section, using an evacuation scenario of a landslide disaster in a mountain village as an example, details its working mechanism.

[0045] A small-scale landslide was triggered by continuous rainfall in a mountain village, blocking a main evacuation route (numbered P03) and reducing the accessibility of some nodes in the original path network. At this point, the dynamic correction unit was triggered, and the local path replanning problem began to be addressed.

[0046] The incremental calculation unit first identifies the local area requiring replanning. Based on feedback, the node affected by the landslide is N12 (an intermediate node on the original path P03), and the traffic capacity of its downstream nodes N15 and N18 is reduced by approximately 40% due to the blockage at P03. The unit selects three nodes upstream and three nodes downstream of N12 (upstream N08, N10, N11; downstream N13, N14, N15) to form a local subgraph, ensuring coverage of the affected core area. The subgraph contains seven edges: N08-N10 (edge ​​E01), N10-N11 (E02), N11-N12 (E03), N12-N13 (E04), N13-N14 (E05), N14-N15 (E06), and N15-N18 (E07).

[0047] The incremental calculation unit updates the weights of each edge in the subgraph. The edge weight is determined by the path length, risk level, and remaining capacity, and the calculation rule is: Weight = Length × 0.3 + Risk Level × 0.4 + (1 / Remaining Capacity) × 0.3. For example, the original edge E03 (N11-N12) has a length of 200 meters, a risk level of medium (corresponding to value 2), and a remaining capacity of 500 people / hour. The original weight = 200 × 0.3 + 2 × 0.4 + (1 / 500) × 0.3 = 60 + 0.8 + 0.0006 ≈ 60.8. Because the downstream path P03 of N12 is blocked, the remaining capacity of E03 decreases to 200 people / hour. The new weight = 200 × 0.3 + 2 × 0.4 + (1 / 200) × 0.3 = 60 + 0.8 + 0.0015 ≈ 60.8015. Similarly, edge E04 (N12-N13) has its risk level raised to high (corresponding to value 3) due to the addition of a temporary access road (length increased by 50 meters to 250 meters), and its remaining passage capacity is 300 people / hour. The new weight = 250×0.3 + 3×0.4 + (1 / 300)×0.3 = 75 + 1.2 + 0.001 ≈ 76.201. The weights of other edges are adjusted according to the actual situation to form an updated edge weight table, see Table 1.

[0048] Table 1: Updated edge weight table.

[0049] Side numbering Original weights Updated weights Reasons for adjustment E01 45.2 45.2 No impact E02 38.6 38.6 No impact E03 60.8 60.8015 Downstream traffic capacity decline E04 52.1 76.201 The addition of a new access road increases its length and poses greater risks. E05 28.4 28.4 No impact E06 32.7 32.7 No impact E07 41.5 41.5 No impact Based on the updated edge weights, the incremental computation unit applies Dijkstra's algorithm to recalculate the shortest paths from each node in the local subgraph to the destination. The total weight of the original path N11-N12-N13-N14-N15 is 60.8+76.201+28.4+32.7=198.101. After replanning, it is found that the total weight of the new path N11-N10-N08-N17-N18 (bypassing N12) is 38.6+45.2+55.3+40.1=179.2, which is lower than the original path. Therefore, the subsequent paths of N11 are adjusted to N10-N08-N17-N18.

[0050] The conflict resolution unit needs to verify the resource occupation conflict between the newly planned route and the existing route. At this time, the evacuation status data shows that 20 people are already moving along the original route N11-N12-N13 (expected to reach N12 in 10 minutes), while the newly planned route N11-N10-N08-N17-N18 overlaps with another emergency passage P05 in the N08-N17 section (this section is only 3 meters wide and has a capacity of 400 people / hour).

[0051] The conflict resolution unit first detects capacity conflicts: the original route N11-N12-N13 has a remaining capacity of 200 people / hour, with 20 people currently using it, leaving a remaining capacity of 180 people / hour; the new route N08-N17 has a remaining capacity of 400-150=250 people / hour (150 people are already planning to use it). There is no direct capacity conflict between the two routes, but the spatial overlap may lead to a decrease in travel speed (the narrow road section is prone to congestion).

[0052] To address the spatial conflict, the unit implemented staggered travel time measures: a 10-minute delay notification was sent to the 20 people traveling along the original route, while a notification was also sent to users on the newly planned route. Furthermore, the unit adjusted the weights of the two routes: the weight of the original route N11-N12-N13 was increased by 10% (to 198.101 × 1.1 ≈ 217.9), while the weight of the new route N11-N10-N08-N17-N18 was decreased by 5% (to 179.2 × 0.95 ≈ 170.2), guiding subsequent travelers to prioritize the new route.

[0053] The instruction iteration unit generates and sends the path update instruction after conflict resolution. The instruction consists of three parts: first, the edge weight adjustment value of the local subgraph; second, the coordinate sequence of the newly added alternative path (the specific latitude and longitude of N11-N10-N08-N17-N18); and third, the update strategy of the terminal device (e.g., users on the original path depart 10 minutes later, and users on the new path depart 5 minutes later).

[0054] Upon receiving the instruction, the navigation module immediately updates its display: terminals approaching N12 display a message stating "Road ahead is closed, please detour via N11-N10," and highlight the new route on the map; terminals not yet reaching N11 directly switch to the new route, which is displayed as a blue dashed line (indicating the optimized route). Simultaneously, the execution drive unit sends encrypted location synchronization instructions to all affected terminals via the 4G network, ensuring consistent navigation data for all users.

[0055] Example 5: The communication relay unit acts as a data transmission bridge between edge computing nodes, undertaking the task of real-time synchronization of dynamic risk distribution maps in complex terrain environments at disaster sites. Disaster sites typically have complex terrain, with features such as mountain obstructions and road interruptions, making it difficult for a single communication link to cover all edge computing nodes. Therefore, the system employs a multi-hop transmission protocol to construct a mesh network, with nodes deployed at base stations, monitoring poles, rescue vehicles, etc., forming a communication network with a coverage radius of 10 kilometers. Each edge computing node has a built-in wireless communication module supporting multi-hop forwarding. When a node cannot communicate directly with the central node, it automatically selects a nearby node with a signal strength ≥ -85dBm as a relay, transmitting data through multi-hop relay.

[0056] The implementation of multi-hop transmission protocols relies on packet structure design and time slot allocation mechanisms. A packet contains four core fields: source node ID, destination node ID, hop count counter, and payload (dynamic risk distribution map data). The hop count counter is initialized to 0 and increments by 1 for each relay node. When the hop count exceeds the maximum limit (e.g., 10 hops), the packet is discarded to avoid network congestion. To prevent data collisions, TDMA (Time Division Multiple Access) technology is used to allocate communication time slots between nodes: the system dynamically adjusts the frame length based on the number of active nodes. For example, with 10 nodes, each time slot is 10ms, for a total frame length of 100ms; when the number of nodes increases to 20, the time slots are shortened to 5ms, while the total frame length remains unchanged at 100ms. Each node transmits data within its allocated time slot, while other nodes are in a listening state, ensuring that only one node transmits data at a time.

[0057] The dynamic risk distribution map needs to be compressed before transmission to adapt to the low bandwidth characteristics of multi-hop transmission. The system uses the Protobuf protocol to serialize the map data. This protocol defines a binary format for data structures (such as risk level, coordinate range, and timestamp), removes redundant fields, and uses variable-length encoding to compress the data size to less than one-third of the original JSON format. For example, map data containing 100 risk areas, the original JSON file size is about 500KB, but after Protobuf serialization, it is only about 150KB, significantly reducing transmission latency. The compressed data packets are forwarded node by node through multi-hop links. Each relay node verifies the data integrity (using CRC checksums) before forwarding. If an error is found, it requests the source node to retransmit, ensuring the accuracy of the map data.

[0058] Dynamic network topology adjustment is a crucial function of communication relay units. When secondary disasters at a disaster site cause node failures, the system re-elects a cluster head node via broadcast beacon frames. The beacon frame contains information such as node ID, location coordinates, remaining battery power, and signal strength. After receiving the beacon frame, all nodes vote to elect a new cluster head node (the node with the strongest signal and the most remaining battery power) based on signal strength (priority) and remaining battery power (secondary). The cluster head node is responsible for managing the communication time slot allocation of its subordinate nodes and adjusting the TDMA frame structure to adapt to the new network topology. For example, if a rescue vehicle node experiences communication interruption due to evacuation, and the original cluster head node fails, the system will re-elect a new cluster head node within 10 seconds via beacon frame broadcast. A nearby base station node will become the new cluster head, adjusting the time slot allocation and bringing the five nodes previously managed by the original cluster head under its control, ensuring uninterrupted synchronization of the dynamic risk distribution map.

[0059] The encryption gateway unit is responsible for the secure transmission of navigation command sets. Its core function is to generate an absolutely secure key through quantum key distribution (QKD) and combine it with symmetric encryption algorithms to protect the command data. The quantum key distribution module is deployed in the system's central node computer room and connected to the edge computing nodes via single-mode fiber optic links. The QKD process is based on the quantum no-cloning theorem: the quantum light source at the central node generates random quantum states (such as the polarization state of photons) and transmits them to the edge nodes via fiber optics; the edge nodes use polarization beam splitters to measure the quantum states, and the measurement results are transmitted back to the central node via classical channels (such as the internet). Both parties compare the measurement results and select a consistent random number as a shared key. This key cannot be intercepted by a third party (any eavesdropping will alter the quantum state, leading to inconsistencies in the measurement results).

[0060] Before encryption, the navigation command set undergoes data integrity verification. The system first calculates the SHA-256 hash value of the command data. This hash value is uniquely determined by the command content (coordinate sequence, voice prompts, suggested speed, etc.) and is used to detect whether the data has been tampered with during transmission. After verification, a one-time pad (OTP) encryption mode is used to XOR the command data: the quantum key is XORed bitwise with the command data to generate ciphertext. The key length is exactly the same as the command data length, ensuring the encryption independence of each bit, theoretically making it impossible to crack using known-plaintext attacks. For example, a navigation command containing 1000 bytes can be XORed with a 1000-byte random number generated by the quantum key to obtain 1000 bytes of ciphertext. Decryption can be performed by XORing the same quantum key again to recover the original command.

[0061] For different types of terminal devices, the encryption gateway unit employs differentiated encryption strategies. Devices with strong computing capabilities, such as smartphones and in-vehicle navigation systems, directly use quantum key encryption (OTP mode). For older phones and smart bracelets with limited processing power, the system switches to AES-256 symmetric encryption mode. AES keys are distributed via a pre-shared key method: each terminal has a unique pre-shared key (associated with the system master key) pre-installed at the factory. During encryption, this key is used to encrypt the instruction data using AES-256, with the ciphertext length matching the original text, and the encryption time controlled within milliseconds. After receiving the data, the terminal uses the pre-stored pre-shared key to decrypt and recover the original instruction. This differentiated strategy ensures that terminals with varying performance can securely receive navigation instructions, while also balancing encryption efficiency and device compatibility.

[0062] The encrypted instruction data packets are sent to the terminal execution device via various wireless communication methods: smartphones receive them via 4G / 5G networks, in-vehicle navigation systems receive them via vehicle-to-everything (V2X) technology, and smart bracelets receive them indirectly after connecting to the phone via Bluetooth Low Energy (BLE) technology. Upon receiving the ciphertext, the terminal device first verifies the hash value to ensure data integrity: it calculates the SHA-256 hash value of the received data and compares it with the original hash value carried in the ciphertext. If they do not match, the data is discarded and a retransmission is requested. If the hash verification is successful, the terminal uses the corresponding key (quantum key or AES key) to decrypt and recover the navigation instruction set.

[0063] The terminal executes the instructions parsed and decrypted by the device's navigation module, activating the audio-visual guidance device. For example, a smartphone's navigation interface highlights the planned route on the map, marks passed and inaccessible sections with different colors, and plays voice prompts through the speaker; a car navigation system displays the route on the screen while simultaneously coordinating with the vehicle's lighting system (such as activating hazard lights to alert other vehicles); a smart bracelet uses a vibration motor to generate short vibrations (about 0.5 seconds) to indicate turning, and long vibrations (about 1.5 seconds) to indicate arrival at a node, with the screen displaying simple arrow icons to guide evacuation.

[0064] The encryption gateway unit continuously monitors the reception status of terminal devices. If a terminal fails to acknowledge receipt for an extended period (more than 30 seconds), the system regenerates the encrypted instruction data packet and sends it via a backup communication link (such as satellite communication). Simultaneously, it notifies the user via SMS or voice call to check the device status, ensuring that all evacuees can obtain safe navigation information in a timely manner. This multi-layered secure transmission mechanism, combining the absolute security of quantum keys with the efficiency of symmetric encryption, provides dual protection for the reliable transmission of evacuation instructions.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0066] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart planning system for emergency evacuation routes in geological disasters based on edge computing, characterized in that, include: The terrain monitoring layer is used to collect multi-dimensional terrain data streams of disaster areas in real time through edge computing nodes; The risk assessment layer is used to integrate historical disaster models with the multidimensional terrain data stream to perform geological stability calculations and generate a dynamic risk distribution map. The path deduction layer is used to combine the dynamic risk distribution map with the preset evacuation constraints and generate an initial evacuation path network through an adaptive optimization algorithm. The evacuation control layer is used to convert the initial evacuation path network into a navigation instruction set and drive the terminal execution device to implement evacuation guidance.

2. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 1, characterized in that, The terrain monitoring layer includes: Multi-source sensing units are deployed at edge nodes of disaster sites to acquire real-time data on surface displacement, groundwater level, and meteorological data. The terrain modeling unit performs spatiotemporal alignment processing on the surface displacement data, groundwater level data, and meteorological data to construct a three-dimensional terrain change matrix. The data compression unit uses a wavelet transform algorithm to perform layered compression on the three-dimensional terrain change matrix to generate the multi-dimensional terrain data stream.

3. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 2, characterized in that, The risk assessment layer includes: The model loading unit calls up the geological structure parameters and disaster triggering thresholds from the historical disaster database. The real-time analysis unit inputs the multidimensional terrain data stream and the geological structure parameters into a convolutional recurrent neural network and outputs a geological deformation trend vector. The map generation unit classifies and labels the geological deformation trend vector according to the disaster triggering threshold, forming the dynamic risk distribution map with risk levels.

4. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 3, characterized in that, The path deduction layer includes: The constraint parsing unit parses the path safety threshold and the upper limit of traffic capacity in the preset evacuation constraints. The network construction unit divides safe passage areas and high-risk restricted areas based on the dynamic risk distribution map; The path generation unit applies an ant colony optimization algorithm within the safe passage area, calculates the path passage weights based on the upper limit of passage capacity, and generates the weighted initial evacuation path network.

5. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 4, characterized in that, The evacuation control layer includes: The instruction conversion unit maps the path nodes in the initial evacuation path network into a navigation coordinate sequence; The device adaptation unit converts the navigation coordinate sequence into device-parseable instructions based on the positioning accuracy and communication protocol of the terminal execution device. The execution drive unit sends an encrypted instruction data packet to the terminal execution device to activate the audio-visual guidance device.

6. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 5, characterized in that, It also includes a feedback optimization layer, which further includes: The performance evaluation unit compares the deviation between personnel evacuation status data and preset evacuation efficiency indicators. The dynamic correction unit triggers the path inference layer to recalculate the local path network when the deviation exceeds the fault tolerance threshold.

7. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 6, characterized in that, The feedback optimization layer also includes: The status tracking unit analyzes the movement speed and gathering density in the personnel evacuation status data in real time; The bottleneck detection unit identifies path nodes whose movement speed is lower than a preset speed threshold; The topology reconfiguration unit adjusts the path network topology weights based on the cluster density distribution and generates path update instructions.

8. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 7, characterized in that, The dynamic correction unit includes: The incremental calculation unit performs local neighborhood path replanning based on the path update instruction. The conflict resolution unit verifies the resource usage conflict between the newly planned path and the existing path; The instruction iteration unit merges the newly planned paths after conflict resolution into the initial evacuation path network.

9. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 1, characterized in that, Also includes: The communication relay unit uses a multi-hop transmission protocol to synchronize the dynamic risk distribution map between edge computing nodes.

10. The intelligent planning system for emergency evacuation routes for geological disasters based on edge computing according to claim 1, characterized in that, Also includes: The encryption gateway unit encrypts the navigation instruction set using quantum key encryption and then distributes it to the terminal execution device.

Images