Disaster risk avoidance method and device based on acoustic fingerprint features, equipment and medium

Abstract

The application relates to the technical field of intelligent decision-making, and discloses a disaster risk avoidance method and device based on acoustic fingerprint features, equipment and a medium, which comprises the following steps: collecting sensor data of a monitoring terminal to determine an open-air state and start acoustic collection; performing time-frequency domain transformation on real-time environmental sound wave signals to extract acoustic fingerprint features, and matching the acoustic fingerprint features with a disaster impact model to identify the type of an impact object; when a plurality of monitoring terminals are time-synchronized and the type of the impact object is consistent, i.e. the impact object is a solid disaster impact, generating disaster center coordinates and a disaster movement vector; determining a disaster movement path according to the disaster center coordinates and the disaster movement vector, searching for a target protection terminal in the path range and in an open-air state; querying available shelter resources around the target protection terminal and generating a risk avoidance navigation path, and sending a risk avoidance instruction. The application can be applied to business scenarios such as financial property insurance, and can realize accurate disaster positioning and path prediction through acoustic sensing and multi-terminal verification, and can link target protection terminals to actively avoid risks, thereby improving response accuracy and timeliness.

Description

Technical Field

[0001] This invention relates to the field of intelligent decision-making technology, and in particular to a disaster avoidance method, apparatus, equipment and medium based on acoustic fingerprint features. Background Technology

[0002] In property insurance, especially auto insurance, catastrophic risk management has long relied on regional disaster warnings issued by meteorological departments. These warnings are typically based on a comprehensive analysis of data from weather radar, satellite cloud images, and weather station observations, often covering areas tens of kilometers or even larger. In financial insurance scenarios, these macro-level warnings are used to trigger risk alerts via SMS, app push notifications, and customer service calls, hoping to reduce customer losses through proactive warnings. However, because weather radar monitoring has a relatively coarse granularity, for disasters with significant localization and suddenness, such as hailstorms, the actual impact point often only covers a very small area of ​​streets or road sections, resulting in many customers not located in the risk area also receiving warning information. Over time, this warning method can easily lead to a situation of "frequent reminders but few hits," reducing customer trust and compliance with warning information, thereby weakening the actual effectiveness of financial institutions' risk warning measures.

[0003] In existing risk warning systems, insurance institutions lack the ability to perceive the true external environment of insured vehicles. Current technology cannot determine whether a vehicle is currently parked in an open-air environment or in a naturally sheltered location such as an underground garage or indoor parking lot. Due to this lack of real-time assessment of the vehicle's environmental conditions, risk warnings are often sent in a uniform manner, failing to differentiate the actual risk exposure levels of different vehicles or provide differentiated and precise warnings for vehicles truly in high-risk environments. This warning method, lacking environmental recognition capabilities, not only increases information delivery costs but also reduces the effective utilization of risk management resources.

[0004] Furthermore, in existing catastrophic claims processes, insurance institutions typically rely on post-disaster reports and claims data to infer the actual affected area of ​​a disaster. Only when a large number of customers file claims can the claims center indirectly infer that a disaster has occurred in a certain area. This approach is a typical post-disaster perception model, unable to quickly pinpoint the location of a disaster at its initial stage, nor can it provide timely evacuation guidance to undamaged vehicles in the vicinity during the disaster's spread. From a financial risk management perspective, this lagging perception mechanism means that risk control always occurs after the loss has occurred, lacking pre-disaster intervention capabilities, leading to a continuous increase in vehicle damage compensation amounts. Summary of the Invention

[0005] The main objective of this invention is to provide a disaster avoidance method, device, equipment, and storage medium based on acoustic fingerprint features, aiming to solve the technical problem that existing technologies cannot achieve real-time and accurate positioning of the disaster landing point based on the actual external environment of the vehicle in the early stage of a disaster, and thereby provide targeted risk avoidance guidance for vehicles in actual risk exposure states.

[0006] To achieve the above objectives, the present invention provides a disaster avoidance method based on acoustic fingerprint features, comprising: The sensor data of the monitoring terminal is collected to determine whether the monitoring terminal is in an open-air state, and the acoustic acquisition mode of the monitoring terminal is activated when the monitoring terminal is in the open-air state; The monitoring terminal acquires real-time environmental acoustic wave signals, performs time-frequency domain transformation on the real-time environmental acoustic wave signals to extract acoustic fingerprint features, and matches the acoustic fingerprint features with a preset disaster impact model to identify the impact object type. The system receives impact object types and corresponding timestamp information uploaded by multiple monitoring terminals located in the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object types are all solid disaster impacts, the system generates disaster center coordinates and disaster movement vectors. Determine the disaster movement path based on the disaster center coordinates and the disaster movement vector, and retrieve the target protection terminal located within the disaster movement path and in an exposed state; The system queries available shelter resources around the target protection terminal, plans a hazard avoidance navigation path from the current location of the target protection terminal to the available shelter resources, and sends a hazard avoidance command containing the hazard avoidance navigation path to the target protection terminal.

[0007] Furthermore, to achieve the above objectives, the present invention provides a disaster avoidance device based on acoustic fingerprint features, comprising: An outdoor state determination module is used to collect sensor data from the monitoring terminal to determine whether the monitoring terminal is in an outdoor state, and to activate the acoustic acquisition mode of the monitoring terminal when the monitoring terminal is in the outdoor state. An acoustic fingerprint recognition module is used to acquire real-time environmental sound wave signals collected by the monitoring terminal, perform time-frequency domain transformation on the real-time environmental sound wave signals to extract acoustic fingerprint features, and match the acoustic fingerprint features with a preset disaster impact model to identify the impact object type. The multi-terminal spatiotemporal fusion positioning module is used to receive the impact object type and corresponding timestamp information uploaded by multiple monitoring terminals located in the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object type is consistent as solid disaster impact, the module generates the disaster center coordinates and disaster movement vector. The risk path prediction and target screening module is used to determine the disaster movement path based on the disaster center coordinates and the disaster movement vector, and to retrieve target protection terminals located within the disaster movement path and in an open-air state. The risk avoidance resource planning and command issuance module is used to query available shelter resources around the target protection terminal, plan a risk avoidance navigation path from the current location of the target protection terminal to the available shelter resources, and send a risk avoidance command containing the risk avoidance navigation path to the target protection terminal.

[0008] Furthermore, to achieve the above objectives, the present invention also provides a computer device, the computer device including a memory, a processor, and a disaster avoidance program based on acoustic fingerprint features stored in the memory and executable on the processor, wherein when the disaster avoidance program based on acoustic fingerprint features is executed by the processor, it implements the steps of the disaster avoidance method based on acoustic fingerprint features as described above.

[0009] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a disaster avoidance program based on acoustic fingerprint features, wherein when the disaster avoidance program based on acoustic fingerprint features is executed by a processor, it implements the steps of the disaster avoidance method based on acoustic fingerprint features as described above.

[0010] Beneficial Effects: This invention relates to the field of intelligent decision-making technology and discloses a disaster avoidance method, device, equipment, and medium based on acoustic fingerprint features. The method includes: collecting sensor data from monitoring terminals to determine the open-air state and activating an acoustic acquisition mode; acquiring real-time environmental sound wave signals and performing time-frequency domain transformation to extract acoustic fingerprint features, matching them with a preset disaster impact model to identify the impact object type; receiving impact object types and timestamp information uploaded by multiple monitoring terminals, and generating disaster center coordinates and disaster movement vectors when the timestamp information meets synchronization conditions and the impact object type is consistently solid disaster impact; determining the disaster movement path based on the disaster center coordinates and disaster movement vectors, retrieving target protection terminals located within the path range and in an open-air state; querying available shelter resources around the target protection terminal and generating an avoidance navigation path, and sending an avoidance command containing the avoidance navigation path. This invention can be applied to business scenarios such as financial property insurance, achieving real-time disaster impact identification through acoustic perception, combining multi-terminal synchronous verification to generate disaster center coordinates and disaster movement vectors, dynamically retrieving target protection terminals within the risk path range and guiding them to avoidance, realizing a closed loop of disaster location, prediction, and proactive defense, improving response accuracy and timeliness, and reducing losses. Attached Figure Description

[0011] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a schematic diagram of an application environment for a disaster avoidance method based on acoustic fingerprint features according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating an embodiment of the disaster avoidance method based on acoustic fingerprint features according to the present invention. Figure 3 This is a schematic diagram of the functional modules of a preferred embodiment of the disaster avoidance device based on acoustic fingerprint features of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 5 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0013] The disaster avoidance method based on acoustic fingerprint features provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server can collect sensor data from monitoring terminals through the client to determine the open-air status and initiate acoustic acquisition mode; acquire real-time environmental sound wave signals and perform time-frequency domain transformation to extract acoustic fingerprint features, matching them with a preset disaster impact model to identify the impact object type; receive impact object type and timestamp information uploaded by multiple monitoring terminals, and generate disaster center coordinates and disaster movement vectors when the timestamp information meets the synchronization condition and the impact object type is consistent as solid disaster impact; determine the disaster movement path based on the disaster center coordinates and disaster movement vectors, and retrieve target protection terminals located within the path range and in an open-air state; query available shelter resources around the target protection terminal and generate a disaster avoidance navigation path, sending a disaster avoidance command containing the disaster avoidance navigation path. This invention can be applied to business scenarios such as financial property insurance, achieving real-time identification of disaster impacts through acoustic perception, combining multi-terminal synchronous verification to generate disaster center coordinates and disaster movement vectors, dynamically retrieving target protection terminals within the risk path range and guiding them to avoid disasters, realizing a closed loop of disaster location, prediction, and proactive defense, improving response accuracy and timeliness, and reducing losses. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will now be described in detail through specific embodiments.

[0014] Please see Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the disaster avoidance method based on acoustic fingerprint features provided by the present invention. It should be noted that although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0015] like Figure 2 As shown, the disaster avoidance method based on acoustic fingerprint features proposed in this invention includes the following steps: S10, collect sensor data from the monitoring terminal to determine whether the monitoring terminal is in an open-air state, and activate the acoustic acquisition mode of the monitoring terminal when the monitoring terminal is in the open-air state; In this embodiment, sensor data from the monitoring terminal is used to determine whether the terminal is in an open-air state, involving the joint perception of the optical environment, electromagnetic propagation environment, and the terminal's own motion state. The ambient light intensity value output by the light sensor reflects the change in incident light flux. In an unobstructed space, this value exhibits high amplitude fluctuations due to natural light, while it shows low amplitude stability when there is a roof or building obstruction. The signal reception strength and signal fluctuation range recorded by the wireless communication module reflect the propagation path of electromagnetic waves in space. In an open environment, the signal propagation path is direct and attenuation is small, while in an enclosed space, the signal is significantly attenuated due to wall reflection and absorption. The triaxial acceleration and angular velocity data output by the inertial sensor are used to identify whether the terminal is in motion, excluding environmental data anomalies caused by vehicle movement or human carrying. The ambient light intensity data is compared with the outdoor light reference threshold, the wireless signal strength data is matched with the open-air signal attenuation model, and the inertial data is used to limit the judgment condition to only be effective in a stationary state. When multiple source data simultaneously meet the unobstructed area judgment result, an open-air state judgment result is generated. This judgment is based on the characteristics of light propagation, the attenuation law of electromagnetic waves, and the application mechanism of inertial measurement in attitude recognition. During implementation, the monitoring terminal periodically samples sensor data, and after filtering and normalization, performs threshold comparison and model matching operations locally, outputting Boolean-type open-air status results.

[0016] When the outdoor condition determination is valid, the acoustic acquisition mode is activated, involving the switching of the acoustic sensor's operating mode and power consumption control strategies. The acoustic sensor maintains a low-power standby mode when not in outdoor conditions. The outdoor condition determination triggers the acoustic sensor to enter a high-sensitivity sampling mode, increasing the sampling frequency and signal gain to capture high-frequency impact sounds. Control commands are written to the sensor configuration register via the embedded control unit, switching sampling parameters and activating the buffer to store continuous sound wave signals. This process relies on the terminal power management mechanism and sensor driving logic to achieve a balance between sampling accuracy and energy consumption. In financial risk control scenarios, vehicles are often parked for extended periods; continuous high-sensitivity acquisition would lead to abnormal power consumption. Triggering acquisition through outdoor condition determination allows for rapid entry into a high-sampling state when a risky environment occurs, while maintaining low-power operation at other times.

[0017] Optical environment data sources can be obtained using photodiodes, ambient camera brightness histograms, or multi-channel illumination sensors. The wireless signal attenuation model can be based on cellular base station signals, or constructed based on WiFi or Bluetooth signal strength. Inertial data can determine the stationary state through acceleration stability or angular velocity variation range. The outdoor environment determination logic can use a fixed threshold comparison method or a binary classification model. The model inputs are illumination values, signal strength values, and inertial stability parameters. The model includes an input layer, two fully connected layers, and an output layer. Training data comes from indoor and outdoor environmental sampling records. The loss function is the cross-entropy function, and training parameters include learning rate, batch size, and iteration rounds. The acoustic acquisition mode can be initiated through local embedded control or triggered by remote configuration strategies. For different terminal battery capacities, the sampling period, buffer length, and gain parameters can be adjusted to adapt to the operating environment.

[0018] This embodiment uses multi-source sensor data to jointly determine the outdoor state and trigger the acoustic acquisition mode, thereby achieving accurate identification of unobstructed environments. At the same time, it maintains the terminal's low-power operation and quickly enters a high-sensitivity sampling state when a risky environment occurs, improving the accuracy of environmental perception and reducing energy consumption.

[0019] S20, acquire the real-time environmental sound wave signal collected by the monitoring terminal, perform time-frequency domain transformation on the real-time environmental sound wave signal to extract acoustic fingerprint features, and match the acoustic fingerprint features with a preset disaster impact model to identify the impact object type; In this embodiment, the real-time ambient sound wave signal acquired by the monitoring terminal comes from the continuous waveform sequence formed by the acoustic sensor in acoustic acquisition mode. The real-time ambient sound wave signal includes metadata such as the time axis sampling point sequence, sampling rate, quantization bit width, and channel identifier. The acquisition process includes analog-to-digital conversion of the analog signal output by the acoustic sensor, buffering and writing the sampled data, and framing and windowing the buffer data to form a stable analysis unit. Time-frequency domain transformation is used to convert the real-time ambient sound wave signal from a single time series expression to a time-frequency representation that simultaneously carries time evolution and frequency distribution. The time-frequency domain transformation can use short-time Fourier transform to obtain the time-frequency amplitude matrix, or continuous wavelet transform to obtain a multi-scale time-frequency coefficient matrix, or constant Q transform to obtain a time-frequency representation that is more sensitive to transient impacts. The frequency band energy distribution of each time slice in the time-frequency representation reflects the transient rise edge, energy concentration frequency band, and attenuation trajectory of the impact sound. Acoustic fingerprint features are used to characterize the distinguishable physical properties of impact events. These features can be formed by combining multiple quantized features extracted from the time-frequency representation. Sources of these quantized features include peak energy distribution, transient duration, band energy decay trend, spectral centroid trajectory, spectral bandwidth variation, kurtosis, and skewness. Combination methods can include vector concatenation, statistical convergence, or sequence encoding, ensuring the comparability of acoustic fingerprint features across different terminal hardware and noise levels. The disaster impact model maps acoustic fingerprint features to impact object types. The matching process can be based on feature distance metrics, which compare the acoustic fingerprint features with the category representation stored in the disaster impact model using Euclidean distance, Mahalanobis distance, or cosine similarity. Alternatively, it can be implemented using a classification network. The classification network takes the acoustic fingerprint features or time-frequency representation as input and outputs the category probability distribution of the impact object type. The identification logic can employ the maximum probability category, threshold filtering, or multi-level discrimination strategies to reduce false alarms. In financial scenarios, the collision object type is used to support the immediate verification and loss reduction decision of risk events. Around objects such as auto insurance insured objects, financial leasing collateral, and fleet asset management, micro-physical evidence is transformed into auditable event labels, which facilitates the formation of consistent data standards in disaster risk management, claims diversion, risk reserve measurement and loss rate tracking.

[0020] The sampling rate of the real-time ambient sound signal can be configured from 8kHz to 48kHz, the quantization bit width can be 16bit or 24bit, and the frame length can be from 20ms to 80ms with an overlap ratio configured to balance time and frequency resolution. After framing, pre-emphasis, bandpass filtering, and noise threshold suppression can be performed to suppress engine idling, wind noise, and human background noise. The time-frequency domain transformation can output either a linear spectrum or a Mel spectrum, and the amplitude scale can be linear or logarithmic to compress the dynamic range. The time-frequency representation can be normalized to adapt to the sensitivity differences of different terminal microphones. Disaster impact models can be implemented in two forms: The first is a prototype-centered model, where the training phase clusters or averages the acoustic fingerprint features labeled as different impact object types to generate category centers, which are then stored along with thresholds. The update phase incrementally updates new samples according to time windows and records the version number. The second is a neural network model, where the network structure includes an input layer, convolutional feature extraction layer, pooling layer, fully connected layer, and output layer. The input is a time-frequency representation matrix or acoustic fingerprint feature vector, and the output is the probability of the impact object type. The training phase includes sample collection, labeling, training and validation set partitioning, loss function setting, parameter iteration, and early stopping strategies. Training parameters can be configured with learning rate, batch size, iteration rounds, and weight decay. Sample augmentation can include random noise addition, random time shifting, frequency band occlusion, and amplitude scaling to improve robustness. Matching deployment can be executed locally on the monitoring terminal to reduce communication latency, or on edge computing nodes to unify model version and threshold configuration. Model version updates can be implemented through canary releases with rollback strategies to avoid recognition errors caused by environmental distribution drift.

[0021] This embodiment converts real-time ambient sound signals into time-frequency representations and extracts acoustic fingerprint features. Then, it uses a disaster impact model to complete matching and identification. This enables the formation of a stable basis for judging the type of impact object in complex noise backgrounds. It transforms impact events from uncertain ambient sounds into structured category results, providing more consistent data input for event verification and loss reduction triggering in disaster risk management. This reduces the resources occupied by invalid early warnings and improves the timeliness of financial risk control response to disaster events.

[0022] S30, receive the impact object type and corresponding timestamp information uploaded by multiple monitoring terminals located in the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object type is consistent as solid disaster impact, generate the disaster center coordinates and disaster movement vector. In this embodiment, multiple monitoring terminals located within the same geographical area represent a set of devices that are spatially close and within the same environmental influence range. The geographical area is uniformly identified through latitude and longitude grid division, cellular cell boundaries, or map partition coding, enabling spatial comparability of data from different terminals. The impact object type is derived from the identification results of real-time environmental acoustic signals and serves as a discrete category label to characterize the physical properties of the current acoustic event. Timestamp information is used to depict the precise time point of the impact event. The timestamp is generated using a unified time reference, and time consistency between terminals is maintained through network time synchronization or satellite time synchronization. The preset synchronization time window represents the maximum range of time differences allowed between events reported by multiple terminals. This time range is set based on the propagation speed of solid disasters in space and the acoustic propagation delay, allowing terminals from different locations to be considered as observation sources of the same disaster event.

[0023] When the timestamp information is within a preset synchronization time window and the impact object type is consistently solid disaster impact, the observation results from multiple terminals show consistency in both time and type dimensions, indicating that the same physical disaster is occurring within the geographic area. The disaster center coordinates are obtained by spatial aggregation calculation of the geographic locations of all monitoring terminals that meet the conditions. Spatial aggregation can employ geometric center calculation, weighted average calculation, or density-based cluster center calculation. The calculation results reflect the spatial core location of the solid disaster at the current time point. The disaster movement vector is used to characterize the changing trend of the disaster center coordinates over time. The disaster movement vector is obtained by vector fitting of the disaster center coordinates generated from multiple consecutive time points. The vector includes two elements: movement direction and movement speed, used to describe the dynamic movement characteristics of solid disasters in geographic space.

[0024] In fintech applications, the coordinates of the disaster center and the disaster movement vector form a quantifiable location of the risk source and the direction of risk propagation, providing precise spatial and temporal basis for identifying risk exposure areas in auto insurance, allocating claims resources, and assessing risk reserves, thus transforming risk identification from macro-regional judgment to micro-locational judgment.

[0025] When uploading the impact object type, the monitoring terminal simultaneously uploads timestamp information and geographic location data. The geographic location data is obtained through satellite positioning or base station positioning. After receiving the data, the system groups it according to geographic region codes and filters out data sets located in the same geographic region. Time synchronization can be maintained through network time protocols, and timestamps are uniformly converted to millisecond-level time formats. The preset synchronization time window can be set from hundreds of milliseconds to several seconds, adjusted according to the typical diffusion rate of solid disasters in geographic space.

[0026] Spatial aggregation calculations can employ a latitude-longitude weighted average algorithm, with weights determined based on the confidence level of the signals reported by the terminals or the acoustic recognition intensity. The calculation of the disaster movement vector can use the least squares method to fit the disaster center coordinates at multiple time points, obtaining the direction vector and displacement per unit time. Alternatively, a time-series prediction model can be constructed, with the disaster center coordinates as the input and the displacement prediction for the next time point as the output. The model structure can include an input layer, a recurrent neural network layer, and a linear output layer. The training process uses historical disaster trajectory data for parameter fitting, with the loss function being the sum of squared position errors. Training parameters include the learning rate, batch size, and number of iterations.

[0027] This embodiment, by determining the consistency of multiple monitoring terminals in the time and space dimensions and performing spatial aggregation and vector fitting, can integrate scattered impact observation results into a disaster expression form with a clear location and movement trend. This enables the location and direction of movement of solid disasters to be accurately quantified in a very short time, providing precise spatial and temporal basis for risk management.

[0028] S40, determine the disaster movement path based on the disaster center coordinates and the disaster movement vector, and retrieve the target protection terminal located within the disaster movement path and in an open-air state; In this embodiment, the disaster center coordinates represent the spatial core location of the solid disaster at the current time point, derived from spatial aggregation calculations of the geographical locations of multiple monitoring terminals. The disaster movement vector describes the direction and speed of the disaster's movement in geographic space, derived from vector fitting of the changes in the disaster center coordinates at continuous time points. The disaster movement path represents the spatial area that the solid disaster may traverse over a future period. The disaster movement path is not a single linear trajectory, but rather a spatial coverage area formed by combining the movement direction, movement speed, and preset warning duration. This area is extended outward in the movement direction and horizontally by a safe bandwidth to form a strip-shaped region, used to cover the uncertainties in the disaster propagation process.

[0029] "Located within the disaster movement path" means the terminal's geographical location is within the extended area along the disaster's movement direction. Geographical location is expressed using latitude and longitude data. Vector projection calculations determine whether the terminal's location is in the forward region of the disaster's movement direction, while spatial distance calculations determine whether it falls within the disaster movement path's coverage area. "Exposed" indicates the terminal is in an unobstructed environment. This state is determined by the terminal's own sensor data, and the result serves as the terminal's current environmental attribute. "Target protection terminal" represents the set of terminals that meet both spatial location and environmental state conditions. This set is selected from all registered terminals and is used for subsequent disaster avoidance guidance.

[0030] In fintech scenarios, target protection terminals represent asset carriers that are in high-risk exposure areas and are actually exposed to the external environment. By combining spatial screening and environmental screening, risk identification is transformed from coarse-grained regional-level judgment to precise device-level judgment.

[0031] The disaster movement path can be calculated using vector extrapolation, extending the disaster center coordinates along the disaster movement vector direction by a distance calculated based on the movement speed and preset warning duration. A lateral buffer distance is added perpendicular to the movement direction, calculated based on historical disaster trajectory offsets. The spatial coverage area is stored in polygon format.

[0032] The geographical locations of all registered terminals are obtained through real-time location data, and the system uses a spatial index structure for rapid retrieval. The system determines whether a terminal is ahead of the disaster's movement direction using vector dot product, and whether it falls within the disaster's movement path coverage area using the shortest distance from the point to the path region. For terminals that meet the spatial conditions, the system retrieves the most recently reported environmental status assessment result to filter out the set of terminals in an exposed state.

[0033] This embodiment uses spatial filtering by combining the direction and speed of disaster movement with the state of the terminal environment. It can accurately identify terminals that are truly in front of the disaster and exposed to the external environment, making risk positioning accurate to the device level and improving the effectiveness of subsequent risk avoidance guidance.

[0034] S50: Query available shelter resources around the target protection terminal, plan a risk avoidance navigation path from the current location of the target protection terminal to the available shelter resources, and send a risk avoidance command containing the risk avoidance navigation path to the target protection terminal.

[0035] In this embodiment, the area surrounding the target protection terminal represents a spatial neighborhood centered on the current location of the target protection terminal. The extent of the spatial neighborhood is defined by a preset radius, a road reachable range, or a preset time reachable range. The current location of the target protection terminal is output from the positioning component of the target protection terminal. The positioning component can output latitude and longitude, map grid codes, or road node identifiers. Latitude and longitude can be generated by satellite positioning, base station positioning, or wireless network positioning. Map grid codes can be obtained by mapping latitude and longitude. Road node identifiers can be obtained by matching latitude and longitude with road topology.

[0036] Available sheltered site resources represent a set of sites that possess both sheltering attributes and accessibility. Sheltering attributes describe a site's physical shielding capacity against solid hazards. These attributes can be represented by site type, structural labels, or roof coverage information. Site types can include underground parking garages, covered parking lots, and covered walkway parking areas. Structural labels can include shelter level, coverage area, and entrance height restrictions. Accessibility conditions describe whether access to a site is permitted at the current moment, including available capacity status, open status, and access constraints. Available capacity status is derived from site vacant parking space data, available parking space counts, or estimated entry flow data. Open status is derived from site operating hours data, gate status data, or platform-published status data. Access constraints include vehicle type restrictions, height restrictions, charging rules, and entry verification requirements. The determination of available sheltered site resources employs a combination of sheltering attribute filtering and accessibility condition filtering to ensure that both sheltering capacity and accessibility are simultaneously satisfied.

[0037] The hazard avoidance navigation path represents a sequence of passable paths from the current location of the target protection terminal to the corresponding location of available shelter resources. This path sequence consists of a sequence of road nodes or road segments. Path generation relies on road topology data, road traffic status data, and estimated travel time data. Road topology data expresses the connection relationships between road nodes and road segments, as well as turning restrictions. Road traffic status data expresses congestion levels, road closures, and traffic restrictions. Estimated travel time data expresses the travel time for different road segments at the current moment. This estimated travel time data can come from real-time traffic data interfaces, historical statistical data, or movement speed data reported by the terminal group. Hazard avoidance navigation path planning employs both accessibility constraints and time safety constraints. Accessibility constraints exclude impassable road segments and inaccessible locations, while time safety constraints ensure arrival at the hazard avoidance point within a preset warning time.

[0038] The evacuation command carries the evacuation navigation path and triggers the target protection terminal to execute evacuation guidance. The evacuation command includes a target evacuation point identifier, evacuation navigation path, path version number, validity period, and guidance presentation strategy. The target evacuation point identifier uniquely points to the selected available sheltered location resource; the path version number distinguishes update frequency; the validity period limits the command's expiration boundary after changes in the disaster situation; and the guidance presentation strategy describes the terminal-side prompting method, which may include a graphical navigation interface, voice prompts, vibration prompts, and pop-up confirmation. The evacuation command is sent via a wireless network, which can be a cellular network, Wi-Fi, or vehicle-to-everything (V2X) communication network. In fintech scenarios, the evacuation command, as part of the risk mitigation reach information, is bound and stored with the target protection terminal identifier for subsequent risk event tracking, compliance documentation, and service quality assessment.

[0039] Available shielded location resource queries can employ a combination of two data sources. The first data source provides basic location information, including location coordinates, entrance coordinates, shielding attribute labels, and access constraint labels. The second data source provides dynamic location status, including available berth data, open status, and gate status. During data fusion, the location identifier is used as the primary key. Basic information is deduplicated first, and then dynamic status is overlaid. Spatial index retrieval is performed on the area surrounding the current location of the target protection terminal. The spatial index can use a grid index or a tree index to reduce the number of candidate locations. During candidate location screening, both shielding attributes and accessibility conditions are verified. Locations with insufficient shielding attributes or unmet open status requirements are eliminated. When available berth data is missing, the most recent valid value with a timestamp is used for verification. If the timestamp exceeds a preset threshold, the location is marked as low-confidence and its priority is reduced.

[0040] Hazard avoidance navigation path planning can employ path search processing based on road topology. Road topology is constructed using nodes and road segments, with segment weights determined by estimated travel time data. This estimated travel time data can be calculated by converting road segment length into real-time speed, which is derived from road condition data interfaces or terminal group speed statistics. The path search outputs candidate paths and candidate travel times to each available sheltered location resource. Target hazard avoidance point selection can utilize multi-condition ranking, with ranking factors including candidate travel time, available parking space data, entrance accessibility, and traffic constraint matching degree. The traffic constraint matching degree is obtained by comparing the target protection terminal equipment type or vehicle attributes with the location's restrictions. In multi-condition ranking, when the candidate travel time meets the preset warning time boundary and the available parking space data meets the preset lower limit, the available sheltered location resource with the highest ranking score is selected as the target hazard avoidance point, and the corresponding candidate path is used as the hazard avoidance navigation path.

[0041] Before sending the evacuation command, the evacuation navigation path is encapsulated, including the target evacuation point coordinates, road node sequence, turning prompts, and validity period. The validity period can be set according to a preset warning duration or the disaster situation update cycle. During transmission, the target protection terminal identifier and message sequence number are bound together; the message sequence number is used for deduplication and confirmation on the terminal side. Upon receiving the evacuation command, the terminal triggers evacuation guidance. The presentation strategy is selected based on the terminal's current interaction state: a combination of pop-up and vibration when the terminal is in silent mode, and a combination of navigation path replacement and voice prompts when the terminal is in navigation mode.

[0042] This implementation dynamically filters the availability of sheltered locations around the target protection terminal and generates a evacuation navigation path to the target evacuation point based on road traffic conditions. The evacuation navigation path is then encapsulated into an evacuation command and sent to the target protection terminal. It can output executable evacuation guidance information when both spatial accessibility and location accessibility are met, thereby improving the feasibility of evacuation guidance and the accuracy of risk reduction.

[0043] In one embodiment, step S10 includes: S101, ambient light intensity data is collected through the light sensor of the monitoring terminal, and wireless signal strength data is collected through the wireless communication module of the monitoring terminal; S102, collects motion state data through the inertial sensor of the monitoring terminal; S103, compare the ambient light intensity data with the preset outdoor light reference threshold; S104, Match the wireless signal strength data with a preset outdoor signal attenuation model to determine the degree of wireless signal penetration attenuation; S105, determine whether the monitoring terminal is in a stationary state based on the motion state data; S106, when the ambient light intensity data is greater than the outdoor light reference threshold and the penetration attenuation of the wireless signal is lower than the threshold defined by the outdoor signal attenuation model, a status determination result indicating that the monitoring terminal is located in an unobstructed area is generated. S107, when the monitoring terminal meets the state determination result and is in the static state, it is determined that the monitoring terminal is in an open-air state; S108, in response to the monitoring terminal being in the open-air state, a start command is sent to the acoustic sensor of the monitoring terminal to drive the acoustic sensor to enter the acoustic acquisition mode.

[0044] In this embodiment, the monitoring terminal is a physical terminal device that carries multiple sensors and communication capabilities. Its deployment form can be a vehicle-mounted system, a mobile communication terminal, a vehicle-mounted network terminal, or an IoT terminal with acoustic sensors and positioning capabilities. Sensor data is a set of observational data characterizing the external environment and the monitoring terminal's own state, including ambient light intensity data, wireless signal strength data, and motion state data. The process for determining whether the monitoring terminal is in an open-air state is constrained by cross-validation of multi-source data to avoid misjudgment due to a single signal. An open-air state corresponds at the data level to a combination of conditions: no obstruction, sufficient external lighting, wireless signal attenuation characteristics conforming to the propagation laws of open spaces, and the terminal being in a stable state.

[0045] A light sensor outputs ambient light intensity data, which can be represented as illuminance, luminous flux, or raw sensor counts, mapped to a unified dimension via calibration coefficients. During data acquisition, sampling time and sampling window are introduced. The sampling time is used for time alignment with wireless signal strength data and motion status data, while the sampling window smooths out short-term fluctuations. In fintech risk control scenarios, the sampling time and sampling window facilitate the creation of auditable status evidence entries, supporting subsequent risk event retrospective analysis and compliance documentation.

[0046] The wireless communication module is used to collect wireless signal strength data, which can be characterized by received signal strength indication, reference signal received power, signal-to-noise ratio, or link quality indication. Wireless signal strength data from different standards need to be normalized. Normalization can be based on a preset mapping table or a piecewise linear function to convert different dimensions into a uniform signal strength level. Introducing carrier frequency band identifiers and cell identifiers when collecting wireless signal strength data helps to distinguish the impact of frequency band differences and base station coverage differences on attenuation, and reduces errors caused by building obstruction and network handover.

[0047] Inertial sensors are used to collect motion state data, which can include acceleration components, angular velocity components, attitude changes, and velocity estimates. Motion state data acquisition also incorporates sampling time and sampling windows. Within the window, peak value, mean, variance, and zero-crossing counts can be calculated. The peak value is used to identify instantaneous jitter, the variance is used to identify continuous vibration, and the zero-crossing count is used to identify periodic swaying. These statistics provide feasible criterion inputs for determining static states, reducing the probability of misjudgments caused by vehicle movement, handheld objects, or tabletop vibrations.

[0048] Outdoor illumination baseline thresholds are used to establish the judgment boundary for ambient light intensity data. Threshold sources can include equipment factory calibration parameters, regional solar radiation model mapping, historical sample statistics, or threshold tables configured by time period. Threshold tables can be divided according to daytime, nighttime, and cloudy / rainy conditions. Threshold entries can include upper and lower limits to suppress interference from direct sunlight and shading. When comparing ambient light intensity data with outdoor illumination baseline thresholds, aggregated values ​​within the same sampling window are used for comparison. Aggregated values ​​can be the median or truncated mean, used to suppress spikes caused by momentary shading and light source flicker.

[0049] The open-air signal attenuation model describes the differences in wireless propagation between open and obstructed spaces. Matching processing maps wireless signal strength data to penetration attenuation levels. The open-air signal attenuation model can employ either a parametric model or a data-driven model. A parametric model consists of a path loss model and an obstruction-related loss term. Inputs include frequency band identifiers, signal strength levels, base station distance estimates, and environmental category labels; the output is the penetration attenuation level. A data-driven model consists of a feature encoding layer and a regression layer. The feature encoding layer encodes the frequency band, signal strength level, and historical stability indicators; the regression layer outputs the penetration attenuation level. Training data can come from labeled open-air and non-open-air sampling records. Labeling sources can include manual inspections, parking lot entrance / exit gate records, or map location label alignment results. The training process may include data cleaning, label consistency verification, training and validation set partitioning, loss function selection, learning rate setting, and early stopping conditions. The output model parameters are fixed in an open-air signal attenuation model configuration file, managed with version numbers during deployment for auditing and rollback. Matching can be done by table lookup matching, model inference matching, or similarity matching. The penetration attenuation level is output as a continuous value or a graded value. The graded value is easy to compare directly with the threshold.

[0050] When determining whether a monitoring terminal is stationary based on motion state data, a stationary state can be defined as a state where the combined amplitude of acceleration and the combined amplitude of angular velocity within the sampling window are both below the corresponding stationary threshold, or the variance within the window is below the stability threshold. The stationary threshold can be derived from terminal calibration, vehicle model configuration differences, or historical behavior statistics. Historical behavior statistics can be grouped by device type to form a threshold set. A continuous constraint can be added to the stationary state determination. This continuous constraint is valid only if multiple consecutive sampling windows meet the stationary condition, reducing misjudgments caused by short pauses.

[0051] The state determination result is used to output the conclusion of the unobstructed area. The generation condition adopts the joint constraint of ambient light intensity data and wireless signal penetration attenuation. The ambient light intensity data is greater than the outdoor light reference threshold to confirm that the light level is consistent with the open or near-open environment. The wireless signal penetration attenuation is lower than the threshold defined by the open-air signal attenuation model to confirm that wireless propagation does not show significant building penetration loss characteristics. This joint condition cross-validates two heterogeneous signal types, light and wireless propagation. Light can reflect top occlusion, and penetration attenuation can reflect perimeter occlusion and indoor wall penetration. When both conditions are met, a state determination result is generated. The determination result can be encoded as a Boolean flag or status code and include the sampling time for easy subsequent recording and auditing.

[0052] When the monitoring terminal meets the status determination result and is stationary, it is determined to be in an exposed state. The stationary state serves as a constraint to eliminate false judgments caused by rapid environmental changes during movement, while the status determination result serves as a constraint to confirm the openness of the environment. After the exposed state is determined, an exposed state identifier can be generated and stored in association with the monitoring terminal identifier. This association storage can be written to a local cache or a remote event log. The event log can carry ambient light intensity data, wireless signal strength data, penetration attenuation level, motion status statistics, and threshold version number, meeting the requirements of financial business for traceability of risk trigger evidence.

[0053] In response to the monitoring terminal being in an outdoor state, a start command is sent to the acoustic sensor to drive it into acoustic acquisition mode. The start command can be sent via the terminal's internal driver interface or device control interface, and includes configurations for sampling rate, gain, acquisition duration, and noise threshold. Acoustic acquisition mode indicates that the acoustic sensor continuously outputs an audio stream or audio clip at a preset sampling rate and writes the audio data to a buffer or upload queue. The sampling rate configuration covers the bandwidth requirements of transient high-frequency components such as hail impacts; the gain configuration balances the dynamic range of weak and strong signals; the acquisition duration configuration controls power consumption and data volume; and the noise threshold configuration delays acquisition or reduces the proportion of invalid data when ambient background noise is too high. Permission verification and privacy configuration verification can be added before sending the start command to ensure that audio acquisition is performed within the scope of authorized and compliance policies, adapting to the governance requirements of financial institutions regarding data acquisition and usage boundaries.

[0054] This embodiment generates an outdoor state determination result by combining ambient light intensity data, wireless signal strength data, and motion state data. When the outdoor state is established, an acoustic acquisition mode is triggered, which can reduce the probability of indoor false triggering and mobile false triggering, reduce the burden of invalid acoustic data acquisition and transmission, and form a traceable determination record with time stamp and threshold version information, which is convenient for the retention of risk access evidence and compliance auditing in financial business scenarios.

[0055] In one embodiment, step S20 above includes: S201, The wavelet transform algorithm is used to perform multi-scale decomposition on the real-time ambient sound wave signal to extract high-frequency sub-band signals; S202, determine the peak energy density, pulse duration, and spectral attenuation slope in the high-frequency sub-band signal, and combine the peak energy density, pulse duration, and spectral attenuation slope into an acoustic fingerprint feature; S203, input the acoustic fingerprint feature into a preset disaster impact model, and determine the feature distance between the acoustic fingerprint feature and the feature center of the pre-stored solid disaster sample in the preset disaster impact model; S204, when the feature distance is less than the preset judgment threshold, the impact object type is determined to be solid disaster impact.

[0056] In this embodiment, the real-time ambient sound signal is time-series data output by the acoustic sensor of the monitoring terminal. The data format can be a single-channel audio stream, a dual-channel audio stream, or a multi-channel audio stream with a multi-microphone array. The acquisition process includes, at the implementation level, sampling rate configuration, quantization bit width configuration, gain configuration, buffer queue configuration, and timestamp writing. Sampling rate configuration covers the high-frequency components of the impact transient signal; quantization bit width configuration covers the dynamic range of weak and strong impacts; gain configuration maintains amplitude stability under different environmental noise levels; buffer queue configuration supports continuous acquisition and segmented processing; and timestamp writing establishes an alignment basis with subsequent geographic area aggregation and synchronization time window determination. Before the acoustic sensor output enters processing, bandpass filtering and amplitude normalization can be performed. Bandpass filtering suppresses the interference of low-frequency wind noise and engine noise on impact characteristics, and amplitude normalization reduces feature drift caused by differences in the sensitivity of different terminal hardware. Segmented processing can extract sound wave segments by a fixed-length sliding window and generate a segment time stamp for each segment. The sliding window length and step size are controlled by a preset parameter table, which can be configured according to terminal type, installation location, and scene noise level.

[0057] Time-frequency domain transformation is used to map time series data to a time-frequency representation to extract the transient structure and energy distribution of impact events. The transformation can be implemented using wavelet transform operations to perform multi-scale decomposition on real-time ambient sound signals. The computational unit for multi-scale decomposition consists of a scale selection module, a filter bank module, and a reconstruction module. The scale selection module determines the number of decomposition levels and the center frequency band for each scale. The filter bank module performs layer-by-layer decomposition on the input segment and outputs sub-band signals at different scales. The reconstruction module performs time-domain alignment on the sub-band signals at the selected scales for feature calculation. High-frequency sub-band signals are the set of sub-bands located in the high-frequency range in the multi-scale decomposition results. The extraction process includes band selection rules and threshold rules at the implementation level. Band selection rules can determine the high-frequency range based on a preset frequency band table, while threshold rules can filter out low-contribution sub-bands based on their energy proportion. The preset frequency band table and threshold parameters can be established according to the frequency response characteristics of the terminal hardware and the typical impact spectrum distribution.

[0058] Peak energy density is used to characterize the maximum energy concentration of a high-frequency subband signal within a short time window. The calculation process involves first generating a short-time energy density sequence from the high-frequency subband signal. This energy density sequence can be obtained by integrating the squares of the amplitudes. The peak value can be approximated by the maximum value or quantile of the energy density sequence, and the peak occurrence time can be included to assist pulse localization. Pulse duration characterizes the time span from the appearance of the impact pulse to its decay. The calculation process involves setting trigger and fallback thresholds based on the energy density sequence. The trigger threshold determines the pulse start time, and the fallback threshold determines the pulse end time. The duration is the difference between these two thresholds. The trigger and fallback thresholds can be obtained from noise floor estimation, which is generated based on the energy statistics of the silent region within the segment. Spectral attenuation slope characterizes the attenuation trend of the high-frequency subband signal's energy in the frequency domain. The calculation process involves generating a logarithmic amplitude sequence from the subband spectral amplitude sequence and performing a linear fit on the frequency range to obtain the slope. The fitted frequency range can be given by a preset frequency band table. To reduce the influence of outliers, the fitting can employ truncation or robust fitting. Acoustic fingerprint features are a combination of energy density peak, pulse duration, and spectral attenuation slope. The combination can be a concatenated vector, a discrete encoding after bucketing, or a scalar set after weighted fusion. Normalization and dimensional alignment can be added during combination. The normalization parameters can be obtained from historical sample statistics and managed according to the model version number to avoid distance calculation offset caused by changes in the distribution of data in different batches.

[0059] The disaster impact model maps acoustic fingerprint features to impact object type determination results. Internally, the model includes a feature access module, a feature standardization module, a sample feature center library, a distance metric module, and a threshold determination module. The feature access module receives acoustic fingerprint features and verifies their dimensions and value ranges. The feature standardization module standardizes the input based on the mean and variance parameters fixed during training. The sample feature center library stores solid disaster sample feature centers, which are aggregated representations of solid disaster impact samples in the feature space. These centers can be obtained by clustering or mean calculation of the features labeled as solid disaster impacts during training. When multiple solid disaster impact subtypes exist, the sample feature center library can store multiple centers with subtype labels. The distance metric module calculates the feature distance between the acoustic fingerprint features and the solid disaster sample feature centers. The distance can be Euclidean, Mahalanobis, or cosine distance. If Mahalanobis distance is used, the covariance matrix or diagonal covariance vector output during training is required as a parameter. The preset judgment threshold is used to map the feature distance to a binary judgment. The threshold can be obtained by selecting the target false alarm rate on the validation set during the training phase, or it can be configured according to the business risk preference and bound to the model version number. When the feature distance is less than the preset judgment threshold, the threshold judgment module outputs the impact object type of solid disaster impact, and can also output the distance value and confidence level as audit fields at the same time.

[0060] The input data for the training phase consists of acoustic sample data and label data. Acoustic sample data can be sourced from field collection, vehicle recording data, event segments reported by terminals, and simulated sound source playback data. Label data is used to identify solid and non-solid impacts. The training process includes sample cleaning, segment segmentation, time-frequency transformation, multi-scale decomposition, high-frequency subband extraction, acoustic fingerprint feature calculation, sample feature center calculation, threshold selection, and model parameter fixing. Sample cleaning removes clipped, silent, and strong wind noise-dominated segments. Segment segmentation parameters include window length, step size, and minimum effective energy threshold. Sample feature center calculation parameters include aggregation method, number of centers, and number of iterations. Threshold selection parameters include target false positive rate, target false negative rate, and cost weight. The training output includes a sample feature center library, standardized parameters, distance metric parameters, and preset judgment thresholds. Output files are managed by version number and archived in the fintech scenario in conjunction with the event audit system to meet the requirement of traceable risk trigger evidence. The model's input in fintech scenarios is real-time ambient acoustic wave signals reported by monitoring terminals or their derived acoustic fingerprint features. The output is the impact object type and confidence field. The output is a compliance decision record that can be used for disaster event identification, risk exposure assessment, and early warning delivery, but it does not introduce action constraints related to subsequent stages.

[0061] This embodiment extracts high-frequency subband signals through multi-scale decomposition and constructs acoustic fingerprint features composed of energy density peak, pulse duration, and spectral attenuation slope. Then, based on the sample feature center and feature distance, threshold determination is completed. This can form quantifiable and traceable impact object type identification results on the terminal side, reduce identification drift caused by environmental noise and hardware differences, and provide stable data support for disaster trigger evidence retention and risk event auditing in financial business scenarios.

[0062] In one embodiment, step S30 above includes: S301, receives the impact object type, corresponding timestamp information and geographical location data uploaded by multiple monitoring terminals located in the same geographical area; S302, Select monitoring terminals from the plurality of monitoring terminals whose impact object type is consistent with solid disaster impact; S303, determine the maximum time difference between the timestamp information corresponding to the selected monitoring terminals; S304, when the maximum time difference is less than or equal to the preset synchronization time window, the selected monitoring terminal is defined as a valid verification terminal. S305, count the number of valid verification terminals, and when the number reaches or exceeds a preset regional verification threshold, extract the geographical location data of all valid verification terminals; S306, Determine the geographic center point based on the geographic location data of all valid verified terminals, and use the geographic center point as the coordinates of the disaster center; S307, the disaster center coordinates are added to the historical disaster trajectory sequence in the order of timestamps, and the coordinate changes of multiple consecutive time points in the historical disaster trajectory sequence are vector fitted to generate a disaster movement vector that includes the direction and speed of movement.

[0063] In this embodiment, the data uploaded by multiple monitoring terminals consists of impact object type, timestamp information, and geographic location data. The monitoring terminals can be vehicle-mounted systems, mobile phones, in-vehicle hardware, or mobile devices with acoustic acquisition and positioning capabilities. The upload process can be completed via cellular or wireless networks. The impact object type expresses the monitoring terminal's classification result for a single impact event, originating from the monitoring terminal's identification output of real-time environmental sound signals. The field format can be an enumerated value or a string marker. The timestamp information expresses the time of occurrence of the event corresponding to the impact object type, originating from the monitoring terminal's local clock or time synchronization module. During generation, a time zone identifier and a time precision identifier must be written, with time precision reaching millisecond levels to adapt to the aggregation and judgment of sudden impact events. The geographic location data expresses the spatial location of the monitoring terminal at the time of the event. The source can be satellite positioning results, base station positioning results, inertial navigation fusion results, or map matching results. The field format can be latitude and longitude, planar coordinates, or three-dimensional coordinates with elevation. To support subsequent spatial aggregation, the geographic location data can include a precision radius and positioning method marker. The precision radius can be used to eliminate terminal records with excessive positioning drift. In fintech scenarios, uploaded records can simultaneously include risk event numbers, policy identifiers, asset identifiers, or account identifiers for auditing and risk control record keeping, without changing the composition of the three types of inputs: impact object type, timestamp information, and geolocation data.

[0064] The same geographic region is used to limit the set of monitoring terminals participating in aggregation. The source of the geographic region can be administrative grids, geofences, raster index units, or business partitions. The division method can be configured according to a fixed raster size or dynamically configured according to road networks and population density. Before entering the filtering process, the receiving end can perform regional consistency verification, mapping the geographic location data of the monitoring terminals to geographic region identifiers. Only data with consistent regional identifiers enters subsequent processing, avoiding erroneous aggregation of cross-regional data. The receiving end can maintain a regional data buffer, which stores data in buckets according to geographic region identifiers. When each record is written, it retains the impact object type, timestamp information, geographic location data triplet, and terminal identifier, which facilitates deduplication, counting, and backtracking.

[0065] Selecting monitoring terminals from multiple monitoring terminals that consistently identify solid disaster impacts is a type consistency filtering process. Solid disaster impact, as one possible impact object type, originates from an enumeration mapping output by the monitoring terminal. The receiving end maintains an impact object type mapping table to accommodate value differences between different terminal versions. This table merges synonymous values ​​reported by different terminals under the unified label of solid disaster impact. After filtering, a candidate monitoring terminal set is obtained. Each member in the candidate monitoring terminal set retains timestamp information and geographic location data, providing input for synchronization determination and spatial calculations.

[0066] The maximum time difference is used to measure the dispersion of event times within the candidate monitoring terminal set. The calculation process extracts timestamp information from the candidate monitoring terminal set to obtain the minimum and maximum timestamps; the maximum time difference equals the maximum timestamp minus the minimum timestamp. To reduce the impact of individual terminal clock drift, timestamp information can be clock-corrected before calculation. Correction sources can include network time synchronization offset, clock deviation information reported by terminals, or server-side time alignment tables. Alternatively, quantile truncation can be used when calculating the maximum time difference to remove extreme timestamps that deviate excessively from the main group before calculating the maximum time difference. A preset synchronization time window is used to express the maximum allowable dispersion range. Its source can be the business configuration center or a risk control parameter table, and the unit can be milliseconds or seconds. Parameters can be configured hierarchically according to terminal density and network latency levels in geographical areas. When terminal density is low, the synchronization time window can be appropriately widened to retain sufficient samples; when terminal density is high, the synchronization time window can be tightened to improve positioning accuracy.

[0067] When the maximum time difference is less than or equal to the preset synchronization time window, the candidate monitoring terminal set is defined as the valid verification terminal set. Valid verification terminals are a subset of terminals that meet both type consistency and time synchronization conditions. The defined action can be implemented by writing a valid flag, which can be written to a memory set, a database table, or a message queue record. The valid flag may contain a verification batch number, which is generated by combining a geographical region identifier and the start and end information of the time window, used to bind the input and output of a single aggregation calculation. In fintech scenarios, after the valid verification terminal set is formed, a risk event aggregation record can be generated. This risk event aggregation record can be used as the audit basis for subsequent risk exposure assessment, loss mitigation reach, and claims warning.

[0068] The number of validly verified terminals reflects the consensus strength of simultaneous solid disaster impact events within the same geographical area. The counting action deduplicates terminal identifiers from the validly verified terminal set and then calculates the number. Deduplication rules can be based on terminal identifier, account identifier, or device fingerprint to avoid duplicate reporting by the same device, which could lead to an inflated number of terminals. The regional verification threshold expresses the minimum required number of valid terminals. Its source can be a business parameter table or risk control strategy configuration. The regional verification threshold can be set based on the terminal deployment density, historical false alarm rate, and seasonal disaster level of the geographical area. Once the number reaches or exceeds the regional verification threshold, a geographic location data extraction action is triggered. This action reads the geographic location data of each member in the validly verified terminal set and forms a set of location points. The location point set can include a precision radius for subsequent weighted processing.

[0069] The geographic center point is used to represent the spatial aggregation result of a set of location points. The calculation can employ arithmetic mean, weighted average, or robust center estimation. Arithmetic mean is suitable for scenarios with similar positioning accuracy, averaging the longitude and latitude of the location point set to obtain the center point. Weighted average is suitable for scenarios with significant differences in positioning accuracy; the weights can be generated from the reciprocal of the positioning accuracy radius or the signal quality score, with higher-accuracy location points having greater weights. Robust center estimation is suitable for scenarios with outliers; spatial outlier detection can be performed on the location point set before center point calculation. Outlier detection can be based on the distance distribution from the initial center, removing location points with excessive deviations before calculating the center point. Using the geographic center point as the disaster center coordinates is an output assignment action. The disaster center coordinates can use the same coordinate system and field format as the geographic location data, facilitating integration with existing map services and risk control geographic engines.

[0070] Historical disaster trajectory sequences are used to store the time-varying coordinates of disaster centers within the same geographic area. Adding an action writes the disaster center coordinates into the sequence in timestamp order. The timestamp order can be based on the window timestamp corresponding to the generation of the disaster center coordinates. The window timestamp can be the minimum, maximum, or median timestamp of the candidate monitoring terminal set, and is written into the trajectory sequence entries. The trajectory sequences can be stored in a time-series database, an in-memory queue, or a key-value store. The key can consist of a geographic area identifier and a disaster event number, facilitating backtracking by region and event.

[0071] Vector fitting is used to estimate the direction and speed of movement from historical disaster trajectory sequences. The fitting process takes the timestamps and corresponding disaster center coordinates in the trajectory sequence as input, and outputs a disaster movement vector consisting of the direction and speed. The direction of movement can be obtained by normalizing the coordinate difference vector between adjacent time points, or by the overall trend direction of multiple time points; the speed can be obtained by dividing the displacement length by the time difference, or by the average speed of multiple time points. To suppress noise and jump points, a smoothing window can be introduced into the fitting process. The length of the smoothing window is configured by a preset parameter table. The smoothing window performs sliding aggregation on the trajectory sequence before calculating the direction and speed. The disaster movement vector can include fields such as direction angle, two-dimensional or three-dimensional components, and a speed scalar; the fields can also include a fitting confidence score, derived from trajectory point consistency measures, such as direction variance and residual magnitude, facilitating the stratification of early warning levels and response strategies in financial risk management.

[0072] This embodiment forms the disaster center coordinates through type consistency filtering, time synchronization determination, quantity threshold verification, and geographic center point calculation. It also performs vector fitting on the historical disaster trajectory sequence to obtain the disaster movement vector. This can aggregate events reported by scattered terminals into auditable spatial location and movement trend information within the same geographical area, reduce the impact of false alarms from single terminals on risk event identification, improve the stability of disaster location and trend estimation, and provide consistent data basis for risk triggering records and subsequent decision-making in fintech scenarios.

[0073] In one embodiment, step S40 above includes: S401, taking the disaster center coordinates as the starting point, determining the disaster coverage area within the preset warning time according to the movement direction and speed indicated by the disaster movement vector, and using the disaster coverage area as the disaster movement path; S402, Obtain the location information of multiple registered associated protection terminals, wherein the associated protection terminals include the monitoring terminal; S403, filter out associated protection terminals whose location information is located within the disaster movement path; S404, Collect sensor data from the selected associated protection terminals; S405, Based on the sensor data, determine whether the associated protection terminal is in an open-air state; S406 identifies associated protection terminals that are determined to be in an open-air state as target protection terminals.

[0074] In this embodiment, the disaster center coordinates are used to characterize the spatial convergence location of solid disaster impacts, and the disaster movement vector is used to characterize the movement trend of this convergence location in the time dimension. Both serve as inputs for path determination and must satisfy coordinate system consistency and time reference consistency. Coordinate system consistency includes consistent geographic coordinate representation, maintaining uniformity in latitude and longitude or planar coordinate representation, and, when necessary, converting location information from different sources to the same geographic reference system. Time reference consistency includes consistent dimensions of the disaster movement vector's velocity and the preset warning duration. Velocity is expressed in distance per unit of time, and the preset warning duration is expressed in seconds or milliseconds, facilitating the multiplication of the movement velocity and the preset warning duration to obtain the advancing distance. The movement direction describes the directional component of the disaster movement vector, and can be expressed as a direction angle, a unit direction vector, or a component pair in the coordinate system. The movement velocity describes the magnitude component of the disaster movement vector, and can be expressed as a scalar velocity or a component velocity. The disaster coverage area represents the spatial extent to which a disaster may reach within a preset warning duration. The coverage area is determined using the disaster center coordinates as the spatial starting point, the direction of movement as the main axis direction, and the movement speed and preset warning duration as the main axis length. The shape of the coverage area can be a corridor region, a fan-shaped region, or an elliptical region. The width of the corridor region can be derived from the lateral distribution of historical disaster trajectory sequences, the geographic grid scale, or a risk control parameter table. The angle of the fan-shaped region can be derived from the uncertainty of the movement direction, and the major and minor axes of the elliptical region can be derived from speed fluctuations and positioning errors. The disaster movement path carries the output representation of the coverage area. It can directly use the geometric description of the coverage area as the path representation, or the coverage area can be discretized into multiple path segments or polygon vertex sequences for retrieval and calculation. The path representation needs to record the time validity period of the coverage area. The start of the time validity period corresponds to the time of coverage area generation, and the end point corresponds to the generation time plus the preset warning duration, facilitating risk control access window control and audit record keeping in fintech scenarios.

[0075] The term "associated protection terminals" refers to a set of terminals with location reporting and sensor data acquisition capabilities that are already under management. "Registered" defines the source and credibility boundaries of this set. The source of registration can be device binding, account binding, asset binding, or business identifier binding completed on the terminal side. Registration records can include terminal identifier, account identifier, asset identifier, permission identifier, and last online time. The terminal identifier uniquely locates the device instance; the account identifier is used to associate with the financial account system; the asset identifier is used to associate with vehicle insurance or financial leasing assets; and the permission identifier limits data reporting and command receiving capabilities. Monitoring terminals are a subset of associated protection terminals. This subset relationship expresses the different functional roles the same physical device plays at different times. Monitoring terminals focus more on the collection and reporting of disaster impact events, while associated protection terminals focus more on inclusion in the candidate protection object pool. Location information expresses the geographical location status of the associated protection terminals. Sources can include satellite positioning, base station positioning, inertial navigation fusion, or map matching. Location information fields can include latitude and longitude, positioning accuracy, speed and heading, and update time. Positioning accuracy is used for subsequent filtering and weighting, and update time is used to align with the validity period of the disaster movement path. Location information can be obtained through periodic reporting, event-triggered reporting, or server-side retrieval. Periodic reporting pushes location updates at preset intervals, event-triggered reporting pushes updates when location changes exceed a threshold, and retrieval is achieved by the server requesting the terminal to return the current location through a communication channel. In fintech scenarios, the acquisition action can be linked to the account risk level, and a higher update frequency can be set for terminals bound to high-value assets to improve timeliness.

[0076] The location information filtering within the disaster movement path is used to converge associated protection terminals from the full set to a spatially relevant set. The filtering implementation includes spatial inclusion judgment and temporal inclusion judgment. Spatial inclusion judgment determines whether the terminal coordinates fall within the geometric range of the coverage area, while temporal inclusion judgment determines whether the terminal's location update time falls within the validity period of the coverage area. In corridor-shaped areas, spatial inclusion judgment is completed by projecting the terminal coordinates onto the main axis of the movement direction and calculating the lateral distance. If the lateral distance is less than the corridor width threshold and the axial projection falls within the length range of the main axis, the terminal is considered included. In fan-shaped areas, spatial inclusion is completed by calculating the azimuth and radial distance of the terminal relative to the disaster center coordinates. If the azimuth falls within the fan-shaped angle range and the radial distance is less than the advancing distance, the terminal is considered included. In polygonal representations, the selection is completed by determining whether a point is within the polygon. The filtering results form a spatial candidate set, which still retains the terminal identifier and location information, facilitating subsequent sensor data collection and status determination. To reduce false filtering, a positioning accuracy threshold can be introduced. Terminals with positioning accuracy exceeding the threshold are removed from the spatial candidate set or downweighted. The threshold can be derived from a business parameter table or historical misjudgment statistics.

[0077] Sensor data is collected from selected associated protective terminals to support the determination of outdoor conditions. This sensor data expresses the degree of environmental exposure and obstruction. Data sources can include light sensors, signal strength readings from wireless communication modules, barometric pressure sensors, temperature and humidity sensors, magnetic field sensors, and inertial sensors. Sensor data acquisition can be achieved through local reading and reporting at the terminal or by triggering a acquisition request on the server side. The acquisition request can include sampling duration, sampling frequency, and sampling timestamp. Sampling duration covers short-term fluctuations, and sampling frequency balances real-time performance and power consumption. During acquisition, raw readings can be standardized, including unit unification, missing value imputation, outlier truncation, and time alignment. Time alignment maps multi-source sensor readings to the same sampling window. In fintech scenarios, sensor data can be combined with asset risk tags. For example, for assets parked in high-risk areas, a longer sampling duration can be set to reduce false positives; for terminals with low battery, a lower sampling frequency can be set to control energy consumption.

[0078] The determination of whether a target protection terminal is in an exposed state based on sensor data serves as the final confirmation condition. An exposed state indicates that the terminal is in an unobstructed or poorly obstructed environment, increasing the risk of disaster impact and maximizing its hazard avoidance value. The determination can be implemented using threshold determination, scoring determination, or model determination. Threshold determination compares light intensity, signal strength, air pressure fluctuations, temperature and humidity changes with their respective thresholds to obtain multi-dimensional determination results, then outputs the exposed state based on a combination of conditions. Scoring determination maps each dimension to a score and performs a weighted sum, with weights derived from historical data statistics or operational parameter tables. Model determination can use classification or regression models, with inputs being standardized multi-dimensional sensor data and location-derived features, and outputs being the exposed probability or exposed label. The model structure can employ a multilayer perceptron network, gradient boosting tree, or lightweight neural network, with network layers including an input layer, several hidden layers, and an output layer. Hidden layer connections are fully connected or residual connections, and the output layer provides a binary classification output. Model training can utilize labeled outdoor and non-outdoor sample data. Sample data sources can include sensor data collected by the terminal in different scenarios and manually labeled data. The training process includes sample cleaning, feature normalization, training set and validation set partitioning, loss function definition, and iterative updates. Iterative updates can employ a mini-batch update method, with batch size, learning rate, and number of iterations configured by a parameter table. After training, the model parameters are distributed to the terminal or deployed on the server side for inference. To adapt to compliance audits of fintech businesses, the judgment output can simultaneously generate a judgment basis field. This field records the sensor dimensions involved in the judgment, key numerical ranges, and judgment timestamp, facilitating subsequent interpretation and traceability of risk-related actions.

[0079] Identifying associated protection terminals determined to be in an exposed state as target protection terminals involves set convergence and role assignment. Target protection terminals represent the set of terminals that need to enter the subsequent evacuation guidance range. The determination action can write the terminal identifier into the target set and record the associated disaster movement path identifier, determination timestamp, and exposed state determination result. The target set can be stored in memory cache or persistent storage to support subsequent calls. To avoid the same terminal repeatedly entering the target set within a short period, a deduplication window can be introduced. The deduplication window length can be derived from a preset alert duration or a business parameter table. To reduce the impact of misjudgments, a secondary confirmation strategy can be introduced. This strategy repeatedly collects sensor data and verifies the exposed state within a short period. If the verification results are consistent, the terminal enters the target set; otherwise, entry is delayed or canceled. In fintech scenarios, after the target protection terminal is determined, a risk event detail record can be generated simultaneously. The record includes the terminal identifier, account identifier, asset identifier, geographical area identifier, and determination timestamp, supporting subsequent risk control reports, compliance reporting, and pre-claims verification.

[0080] This embodiment generates the disaster coverage area and forms the disaster movement path by using the disaster center coordinates and disaster movement vector. It then completes spatial screening by combining the location information of the registered associated protection terminals. Based on sensor data, it outputs the open-air status and determines the target protection terminals. This can converge risk identification from the macro-regional level to a set of terminals that are consistent with the disaster's direction of advancement and time window, reducing the reach of irrelevant terminals and improving the spatial and status relevance of target screening. This provides a more controllable data foundation for risk alerts and loss reduction in fintech businesses.

[0081] In one embodiment, step S50 above includes: S501, using the real-time positioning point of the target protection terminal as the center, retrieve a list of location information including underground garages and covered parking lots within a preset radius; S502, obtain the available berth data of each location in the location information list through the real-time data interface, and determine the locations with available berths as available sheltered location resources based on the available berth data. S503, determine the travel time of the target protection terminal to each available shelter resource, and determine the estimated arrival time of the disaster to each available shelter resource based on the disaster movement vector; S504, select the available sheltered location resource with the shortest travel time and the shorter travel time as the target shelter point, and plan the shelter navigation path from the current location of the target protection terminal to the target shelter point; S505, the risk avoidance navigation path is encapsulated into a risk avoidance command, and the risk avoidance command is pushed to the target protection terminal via a wireless network to trigger the target protection terminal to start emergency risk avoidance guidance.

[0082] In this embodiment, the spatial boundary around the target protection terminal is used to define the resource retrieval. The spatial boundary is expressed by real-time positioning points, which represent the geographical location of the target protection terminal at a certain moment. The data source can be satellite positioning, base station positioning, inertial navigation fusion, or map matching. The output includes latitude and longitude, positioning accuracy, and update time. Positioning accuracy is used to control the effectiveness of the retrieval radius. When the positioning accuracy is high, the preset radius range can be enlarged proportionally or a positioning verification can be triggered. Positioning verification reduces false resource detections caused by positioning drift by collecting positioning readings again and making consistency judgments on multiple readings. The preset radius range is used to parameterize the retrieval range. The value source can be a risk control parameter table, the average traffic capacity of urban roads, or the configuration of capital loss reduction strategies. In a fintech scenario, the radius can be linked to asset risk stratification. Assets with higher risk levels correspond to smaller radii to increase the probability of reaching the account, while assets with lower risk levels correspond to larger radii to increase the resource hit rate.

[0083] The location information list serves as a collection of candidate locations available for evacuation. Underground parking garages and covered parking lots represent the types of shelter structures, which are used to block impacts from solid debris. The location information list can be obtained from map services, parking service platforms, property management systems, and urban traffic data platforms. The data structure can include location identifiers, location coordinates, entrance coordinates, opening hours, access restrictions, vehicle height restrictions, charging rules, and location type tags. Location identifiers are used for subsequent data association, entrance coordinates are used for navigation endpoints, opening hours and access restrictions are used for compliance and accessibility filtering, and charging rules can serve as a supplementary decision-making dimension, supporting service tiers based on customer benefits. Search actions can be performed within a preset radius for spatial queries. Spatial queries can use grid indexes or geofence indexes. Index construction divides the city map into grid cells and maps location coordinates to these cells. During a query, only the set of covered grid cells is scanned, reducing computational overhead and improving real-time performance, making it suitable for high-concurrency risk access scenarios.

[0084] The real-time data interface is used to associate the list of venue information with dynamic availability information. Available parking space data indicates the current number of vehicles a venue can accommodate. Data sources can include parking gate counters, magnetic parking space sensors, video recognition counting, and platform aggregated data. Interface interaction can be via pull or subscription. Pulling involves the server requesting available parking space data in batches based on venue identifiers. Subscription involves the server maintaining subscription relationships and receiving change notifications, which can include timestamps and data version numbers for idempotent processing. A positive value for available parking space data forms the criterion for determining available shaded venue resources. Available shaded venue resources are used to support a set of venues that meet the shaded type and availability conditions. To reduce misjudgments caused by data fluctuations, a stability check can be introduced for available parking space data. This check performs consistency checks on available parking space data across multiple consecutive sampling times, triggering a re-pull or delayed confirmation when short-term fluctuations occur. In fintech scenarios, available shaded venue resources can also be tagged with service availability labels, such as cooperation benefit labels, transparent pricing labels, and invoicing labels, facilitating subsequent customer service and settlement reconciliation.

[0085] Travel time is used to express the time cost for a target protection terminal to reach a certain available sheltered location from its current location. Determining travel time requires road network data, real-time traffic conditions, turning restrictions, and entrance coordinates. Data sources can include navigation services, traffic data platforms, and historical traffic statistics. Travel time can be calculated by accumulating the time consumed in segments. Segment time consumption is estimated from segment length and speed. Speed ​​estimates can be derived from real-time traffic conditions or recent statistics. The accumulated segment time is then overlaid with estimated entrance queuing times, which can be derived from gate access records or platform statistics. If vehicle height or access restrictions exist, accessibility filtering can be applied to available sheltered location resources before calculating travel time. The filtering conditions are determined by the asset attributes bound to the target protection terminal, which can include vehicle height, vehicle type, and access permissions.

[0086] The estimated arrival time of a disaster at each available sheltered location is used to represent the time window for the disaster to advance to the vicinity of each candidate location. The input is the disaster movement vector and the location coordinates of each available sheltered location. The disaster movement vector includes the movement direction and movement speed. The movement direction is used to determine the directional reference of the disaster's advance ray or advance zone, and the movement speed is used to map spatial distance to temporal distance. The estimated arrival time can be determined by converting projected distance and speed. The projected distance is obtained by projecting the location coordinates relative to the disaster center coordinates onto the movement direction. A positive projected distance indicates that the location is ahead of the advance direction, and a negative projected distance indicates that the location is behind the advance direction. When the projected distance is negative, the estimated arrival time can be marked as risk-free or set to a maximum value to avoid being selected. To cover the uncertainty of disaster spread, an uncertainty margin can be added to the estimated arrival time. The uncertainty margin can be sourced from deviation statistics of historical disaster trajectory sequences or risk control parameter tables. The margin can be added by reducing the estimated arrival time or tightening the safety threshold, thereby making the selection of locations more conservative. In fintech scenarios, the comparison between estimated arrival time and travel time can generate interpretable fields. These fields include projected distance, speed, margin value, and calculation timestamp, facilitating audit tracking and dispute resolution.

[0087] The target hazard avoidance point is used to select one or a small number of preferred locations from available sheltered location resources. Selection criteria include travel time shorter than the estimated arrival time and the shortest travel time. A travel time shorter than the estimated arrival time ensures time accessibility, while the shortest travel time reduces exposure time and increases the probability of execution. If multiple locations have similar travel times, a secondary sorting key can be introduced. This secondary sorting key can be based on available parking spaces, entrance capacity, location type priority, or the friendliness of charging rules. The source of the secondary sorting key is determined by business configuration, facilitating strategy switching between cost reduction, customer experience, and cooperative settlement. After the target hazard avoidance point is determined, an hazard avoidance navigation path is generated. This path represents the route from the current location to the target hazard avoidance point's entrance coordinates. The path representation can include multiple path coordinate sequences, a set of turning prompts, estimated travel time, and remaining distance. Path generation can be returned by the navigation service or generated by the proprietary road network engine. The proprietary road network engine needs to maintain the road topology, road segment weights, and traffic restriction rules. Weights can be dynamically updated according to road conditions, with updates sourced from the traffic data platform or historical statistics.

[0088] The risk avoidance command is used to send executable guidance information to the target protection terminal. The encapsulated action combines the risk avoidance navigation path with the target risk avoidance point identifier, validity period, and verification field into a command payload. The validity period is used to control the timeliness of the command, and its value can be associated with a preset alert duration. The verification field is used for integrity verification and anti-tampering, and the verification field can be a hash value, signature value, or serial number. The wireless network is used to carry the command push channel, which can be a mobile data network, a vehicle network channel, or a message push channel. A delivery receipt request can be attached when pushing the command. The delivery receipt is used to confirm the target protection terminal's reception status and record the arrival timestamp. The arrival record can be written into the risk control event ledger. The ledger fields can include account identifier, terminal identifier, target risk avoidance point identifier, risk avoidance navigation path version number, and delivery status, which meets the compliance record-keeping and subsequent claims loss reduction analysis requirements of financial business. Emergency evacuation guidance is used to trigger terminal-side display and reminders. The presentation can include map navigation interface, voice prompts, vibration prompts, and lock screen pop-ups. The trigger expression can be implemented through command type code or priority field. The priority field is used to ensure that the evacuation command gets higher processing priority when multiple types of messages compete for it.

[0089] This embodiment completes spatial retrieval of the site information list by using real-time positioning points and a preset radius range. It then obtains available berth data and determines available sheltered site resources by combining real-time data interfaces. Finally, it selects the target evacuation point and generates an evacuation navigation path by comparing the travel time with the estimated arrival time. The evacuation navigation path is then encapsulated into an evacuation command and pushed to the target protection terminal. Under time constraints, it can unify resource availability, access feasibility, and disaster advancement window into the same decision-making process, reducing ineffective guidance and resource unavailability guidance, and improving the executability and traceability of risk access.

[0090] In one embodiment, after step S505 above, the method further includes: S506, Monitor the geographical location changes of the target protection terminal; S507, when the geographical location of the target protection terminal enters the target refuge point and the displacement within a preset time period is less than a preset threshold, it is confirmed that the target protection terminal has successfully avoided danger. S508, generate a successful risk avoidance confirmation message, associate the successful risk avoidance confirmation message with the identifier of the target protection terminal and store it, and provide incentive feedback to the account associated with the target protection terminal.

[0091] In this embodiment, encapsulating the hazard avoidance navigation path into the hazard avoidance command belongs to the information payload organization and executable command generation process. The hazard avoidance navigation path includes fields such as path coordinate sequence, turn prompt set, estimated travel time, and target hazard avoidance point identifier. The encapsulation action combines the above fields with command type, command validity period, priority flag, and integrity verification field to form the hazard avoidance command payload. The command validity period is used to limit the executable time window of emergency hazard avoidance guidance. The value can be linked with a preset duration to prevent expired paths from being executed. The integrity verification field is used to identify whether the payload has been tampered with or truncated during transmission. The verification field can be composed of a payload hash value and a server signature value. The signature value is generated by signing the payload digest with a key and attaching a timestamp. After receiving the signature value, the target protection terminal verifies the signature value and checks whether the timestamp falls within the validity period. Wireless network push is used to realize the delivery of commands. The push channel can be a mobile data network message channel, a vehicle network message channel, or a push service channel. The push message includes terminal identifier, session identifier, retransmission sequence number, and acknowledgment receipt flag. The terminal side returns a delivery receipt and carries the reception timestamp and payload verification result. Delivery receipts are written to the event log. Event log fields can include account identifier, target protection terminal identifier, target evacuation point identifier, evacuation instruction sequence number, delivery status, and receipt timestamp. These are used for risk control auditing, claims loss reduction verification, and customer dispute resolution. Emergency evacuation guidance is triggered on the terminal side by instruction type and priority. Terminal display formats can include a top navigation screen, a lock screen pop-up, voice prompts, and vibration alerts. The displayed content is generated based on the evacuation navigation path. The terminal side also records the guidance start timestamp and current geographical location, providing a data foundation for subsequent evacuation success assessment.

[0092] Geographic location change monitoring is used to acquire the spatial movement trajectory of the target protection terminal after push notification. Data sources can include periodic reporting from the positioning module, terminal-side positioning subscription, and background positioning retrieval. The monitoring period and positioning accuracy jointly determine the trajectory resolution. The monitoring period can be dynamically adjusted according to the risk level and network quality. When the network quality is poor, the reporting frequency can be reduced and location caching can be enabled for batch reporting. The location cache contains multiple positioning points and corresponding timestamps. Geographic location changes are extracted from a continuous sequence of positioning points. The positioning point sequence is used to calculate displacement and velocity. Displacement calculation can use geographic coordinate distance conversion and map matching correction. Correction is achieved by projecting the positioning points to reasonable roads or near the entrance of the location through road topology and driving accessibility constraints, reducing misjudgments caused by positioning drift. The target avoidance point serves as the center of a geofence or a polygonal fence object. The fence range can be constructed from the location entrance coordinates and the location boundary. Entry judgment is triggered when the positioning point falls into the fence range. After triggering, monitoring continues to execute to determine whether the dwell state is valid.

[0093] The preset duration describes the observation window after entering the target evacuation point. Its value can be associated with risk reach strategies, customer rights levels, and location access characteristics. A displacement less than a preset threshold defines the evacuation state. This threshold absorbs positioning noise and short-term micro-movements and can be dynamically adjusted based on positioning accuracy. When positioning accuracy is poor, the threshold is relaxed accordingly, and multiple sampling consistency checks are introduced. The evacuation judgment performs displacement statistics on multiple positioning points within the preset duration. The statistical objects can be the displacement of adjacent positioning points or the displacement of the positioning point relative to the entry point. A successful evacuation state is formed when the displacement is less than the preset threshold. The successful evacuation confirmation information is a structured record expressing this successful evacuation state. Fields can include evacuation command sequence number, target protection terminal identifier, account identifier, target evacuation point identifier, entry timestamp, confirmation timestamp, positioning evidence summary, and verification result. The positioning evidence summary can consist of a positioning point sequence summary and a fence judgment result summary. Summary generation supports subsequent audit review and model training data accumulation. Association storage establishes a searchable relationship between successful risk mitigation confirmation information and the target protection terminal identifier. The storage format can be a relational table partitioned by terminal identifier, or an event stream written to an index database with a combined index based on terminal identifier and account identifier. Index fields are used for quick retrieval of evidence chains in claims processing, risk verification, and customer service work orders. Incentive feedback provides positive outreach to accounts. Outreach content can take the form of reward points, service limits, level growth values, or exemption qualifications. Incentive feedback is generated based on the confirmation timestamp and evidence summary in the successful risk mitigation confirmation information. The calculation process can incorporate different reward strategies configured in the risk control rule table. Strategy parameters include trigger thresholds, reward types, reward limits, and validity periods. Incentive feedback is written to the account's reward account and synchronized to the message center. The message center sends a feedback notification to the target protection terminal. Notification fields include reward type, arrival time, usage conditions, and query entry, meeting the traceability and reconciliation requirements of financial transactions.

[0094] For example, in a catastrophic risk management scenario for auto insurance, the terminal corresponding to the insured vehicle maintains low-intervention operation for an extended period. During a certain period, localized severe convective weather occurs, and dense impact sounds begin to be heard on some road sections. Monitoring terminals located in parking lots or on the roadside collect sensor data to determine whether the monitoring terminal is in an exposed state. Sensor data includes ambient light-related data, wireless signal strength-related data, and motion-related data. Ambient light-related data reflects overhead shading, wireless signal strength-related data reflects spatial penetration attenuation characteristics, and motion-related data eliminates environmental judgment biases caused by movement. When the exposed state is determined, the monitoring terminal activates acoustic acquisition mode. The acoustic sensors enter continuous sampling mode and output real-time ambient sound signals. The sampled data carries timestamp information and is written to a buffer, facilitating subsequent cross-terminal time alignment and aggregation on the server side.

[0095] The server receives the real-time environmental acoustic signal processing results uploaded by the monitoring terminals, performs time-frequency domain transformation on the real-time environmental acoustic signals, and extracts acoustic fingerprint features. The acoustic fingerprint features include a combination of energy distribution, persistence features, and attenuation features related to the impact transient, used to improve comparability under different terminal hardware and background noise conditions. The acoustic fingerprint features are then matched against a preset disaster impact model to output the impact object type. This impact object type is used as a risk event label in the financial risk control semantics and entered into the risk control event ledger, facilitating consistent reporting in subsequent claims management and loss reduction statistics. When multiple monitoring terminals report impact object type and timestamp information almost simultaneously within the same geographical area, the server filters the reported records within the same geographical area, retaining the set of records where the impact object type is consistently solid disaster impact. The server extracts the corresponding timestamp information and calculates the maximum time difference. When the maximum time difference meets a preset synchronization time window, the corresponding monitoring terminal is defined as a valid verification terminal. Once the number of valid verification terminals reaches or exceeds the regional verification threshold, the server reads the geographical location data of the valid verification terminals, determines the geographical center point based on the geographical location data, and uses it as the disaster center coordinates. The coordinates of the disaster center are written into the historical disaster trajectory sequence in the order of timestamps. The coordinate changes of multiple consecutive time points in the historical disaster trajectory sequence are used for vector fitting to generate a disaster movement vector that includes the direction and speed of movement. This transforms the risk source from a macro-regional indication into a quantifiable expression of the disaster center coordinates and disaster movement vector.

[0096] Risk identification enters the adjacent risk hedging phase. The server determines the disaster coverage area based on the disaster center coordinates and disaster movement vector, using this area as the disaster movement path. The server obtains location information from registered associated protection terminals, which cover accessible objects such as vehicle-mounted terminals bound to insured vehicles, active terminals of vehicle owners, and monitoring terminals. The server filters associated protection terminals whose location information is within the disaster movement path, collects sensor data from the filtered associated protection terminals, and determines whether the associated protection terminals are in an exposed state. Associated protection terminals determined to be in an exposed state are identified as target protection terminals. This process converges the risk objects from the registered terminal pool to a set of terminals related to the disaster movement path and exposed to the open, avoiding frequent access to terminals already located in underground parking garages or roof-covered areas, and reducing the risk of decreased compliance due to ineffective access.

[0097] Once the target protection terminal is determined, the server searches for available sheltered locations around the terminal's current location. Using the terminal's real-time location as the center, the server searches for underground parking garages and covered parking lots within a preset radius to form a list of locations. It obtains the available parking space data for each location in the list via a real-time data interface, identifying locations with more than zero available parking spaces as available sheltered locations. The server determines the travel time from the target protection terminal to each available sheltered location and, based on the disaster movement vector, determines the estimated arrival time of the disaster. It selects the available sheltered location with the shortest travel time (less than the estimated arrival time) as the target evacuation point and generates an evacuation navigation path from the terminal's current location to the target evacuation point. This evacuation navigation path is encapsulated into an evacuation command and pushed to the target protection terminal via a wireless network. The terminal triggers an emergency evacuation guidance presentation, which can be a combination of a high-priority pop-up, voice broadcast, vibration alerts, and navigation interface injection, providing the user with actionable evacuation navigation guidance within the risk window.

[0098] After the risk avoidance command is issued, the process enters the behavior verification and incentive feedback phase. The server monitors the geographical location changes of the target protection terminal. When the geographical location of the target protection terminal enters the target risk avoidance point and the displacement within a preset time period is less than a preset threshold, the risk avoidance of the target protection terminal is confirmed to be successful. The server generates risk avoidance success confirmation information, associates and stores the risk avoidance success confirmation information with the identifier of the target protection terminal, and forms a traceable record of loss reduction evidence. The recorded content can be used for risk event auditing, loss reduction statistics, and pre-claim verification. Based on the risk avoidance success confirmation information, the server provides incentive feedback to the account associated with the target protection terminal. The incentive feedback can be manifested as a notification of the issuance of rights points, service limits, car wash benefits, or premium deductions. The delivery channel can be completed by push notifications from the message center, thereby connecting risk reach, execution verification, data recording, and positive incentives into a closed loop of financial risk control, improving the customer's response efficiency and execution stability in catastrophic scenarios.

[0099] This embodiment forms an executable reach record by encapsulating the risk avoidance command payload, pushing it through a wireless network, and recording the receipt. It combines geographical location change monitoring, target risk avoidance point entry judgment, and displacement threshold constraints within a preset time period to determine the risk avoidance success status. Then, it associates and stores the risk avoidance success confirmation information with the target protection terminal identifier and provides incentive feedback to the account. This can realize closed-loop support from command delivery, execution evidence to result recording, improve the verifiability of the loss reduction process and the compliance record-keeping capability of financial business, and enhance the continuity of customers' emergency risk avoidance guidance.

[0100] In one embodiment, a disaster avoidance device based on acoustic fingerprint features is provided, which corresponds one-to-one with the disaster avoidance methods based on acoustic fingerprint features described in the above embodiments. (Refer to...) Figure 3 , Figure 3 This is a schematic diagram of the functional modules of a preferred embodiment of the disaster avoidance device based on acoustic fingerprint features of the present invention. The modules include: an open-air state determination module 10, an acoustic fingerprint recognition module 20, a multi-terminal spatiotemporal fusion positioning module 30, a risk path prediction and target screening module 40, and a disaster avoidance resource planning and command issuance module 50. Detailed descriptions of each functional module are as follows: The outdoor state determination module 10 is used to collect sensor data from the monitoring terminal to determine whether the monitoring terminal is in an outdoor state, and to activate the acoustic acquisition mode of the monitoring terminal when the monitoring terminal is in the outdoor state. The acoustic fingerprint recognition module 20 is used to acquire the real-time environmental sound wave signal collected by the monitoring terminal, perform time-frequency domain transformation on the real-time environmental sound wave signal to extract acoustic fingerprint features, and match the acoustic fingerprint features with a preset disaster impact model to identify the impact object type. The multi-terminal spatiotemporal fusion positioning module 30 is used to receive the impact object type and corresponding timestamp information uploaded by multiple monitoring terminals located in the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object type is consistent as solid disaster impact, the module generates the disaster center coordinates and disaster movement vector. The risk path prediction and target screening module 40 is used to determine the disaster movement path based on the disaster center coordinates and the disaster movement vector, and to retrieve target protection terminals located within the disaster movement path and in an open-air state. The risk avoidance resource planning and instruction issuance module 50 is used to query available shelter resources around the target protection terminal, plan a risk avoidance navigation path from the current location of the target protection terminal to the available shelter resources, and send a risk avoidance instruction containing the risk avoidance navigation path to the target protection terminal.

[0101] Specific limitations regarding disaster avoidance devices based on acoustic fingerprint features can be found in the aforementioned limitations on disaster avoidance methods based on acoustic fingerprint features, and will not be repeated here. Each module in the aforementioned disaster avoidance device based on acoustic fingerprint features can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0102] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides determination and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a disaster mitigation method based on acoustic fingerprint characteristics on the server side.

[0103] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides determination and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements client-side functions or steps of a disaster mitigation method based on acoustic fingerprint characteristics.

[0104] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: The sensor data of the monitoring terminal is collected to determine whether the monitoring terminal is in an open-air state, and the acoustic acquisition mode of the monitoring terminal is activated when the monitoring terminal is in the open-air state; The monitoring terminal acquires real-time environmental acoustic wave signals, performs time-frequency domain transformation on the real-time environmental acoustic wave signals to extract acoustic fingerprint features, and matches the acoustic fingerprint features with a preset disaster impact model to identify the impact object type. The system receives impact object types and corresponding timestamp information uploaded by multiple monitoring terminals located in the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object types are all solid disaster impacts, the system generates disaster center coordinates and disaster movement vectors. Determine the disaster movement path based on the disaster center coordinates and the disaster movement vector, and retrieve the target protection terminal located within the disaster movement path and in an exposed state; The system queries available shelter resources around the target protection terminal, plans a hazard avoidance navigation path from the current location of the target protection terminal to the available shelter resources, and sends a hazard avoidance command containing the hazard avoidance navigation path to the target protection terminal.

[0105] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, and a computer program is stored thereon, which, when executed by a processor, performs the following steps: The sensor data of the monitoring terminal is collected to determine whether the monitoring terminal is in an open-air state, and the acoustic acquisition mode of the monitoring terminal is activated when the monitoring terminal is in the open-air state; The monitoring terminal acquires real-time environmental acoustic wave signals, performs time-frequency domain transformation on the real-time environmental acoustic wave signals to extract acoustic fingerprint features, and matches the acoustic fingerprint features with a preset disaster impact model to identify the impact object type. The system receives impact object types and corresponding timestamp information uploaded by multiple monitoring terminals located in the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object types are all solid disaster impacts, the system generates disaster center coordinates and disaster movement vectors. Determine the disaster movement path based on the disaster center coordinates and the disaster movement vector, and retrieve the target protection terminal located within the disaster movement path and in an exposed state; The system queries available shelter resources around the target protection terminal, plans a hazard avoidance navigation path from the current location of the target protection terminal to the available shelter resources, and sends a hazard avoidance command containing the hazard avoidance navigation path to the target protection terminal.

[0106] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0107] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0108] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0109] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0110] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A disaster avoidance method based on acoustic fingerprint features, characterized in that, Includes the following steps: The sensor data of the monitoring terminal is collected to determine whether the monitoring terminal is in an open-air state, and the acoustic acquisition mode of the monitoring terminal is activated when the monitoring terminal is in the open-air state; The monitoring terminal acquires real-time environmental acoustic wave signals, performs time-frequency domain transformation on the real-time environmental acoustic wave signals to extract acoustic fingerprint features, and matches the acoustic fingerprint features with a preset disaster impact model to identify the impact object type. The system receives impact object types and corresponding timestamp information uploaded by multiple monitoring terminals located in the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object types are all solid disaster impacts, the system generates disaster center coordinates and disaster movement vectors. Determine the disaster movement path based on the disaster center coordinates and the disaster movement vector, and retrieve the target protection terminal located within the disaster movement path and in an exposed state; The system queries available shelter resources around the target protection terminal, plans a hazard avoidance navigation path from the current location of the target protection terminal to the available shelter resources, and sends a hazard avoidance command containing the hazard avoidance navigation path to the target protection terminal.

2. The disaster avoidance method based on acoustic fingerprint features as described in claim 1, characterized in that, Collect sensor data from the monitoring terminal to determine whether the monitoring terminal is in an open-air state, and activate the acoustic acquisition mode of the monitoring terminal when the monitoring terminal is in the open-air state, including: Ambient light intensity data is collected by the light sensor of the monitoring terminal, and wireless signal strength data is collected by the wireless communication module of the monitoring terminal. Motion state data is collected through the inertial sensors of the monitoring terminal; The ambient light intensity data is compared with a preset outdoor light intensity threshold. The wireless signal strength data is matched with a preset outdoor signal attenuation model to determine the degree of wireless signal penetration attenuation. Based on the motion state data, determine whether the monitoring terminal is in a stationary state; When the ambient light intensity data is greater than the outdoor illumination reference threshold and the penetration attenuation of the wireless signal is lower than the threshold defined by the outdoor signal attenuation model, a status determination result indicating that the monitoring terminal is located in an unobstructed area is generated. When the monitoring terminal meets the state determination result and is in the stationary state, it is determined that the monitoring terminal is in an open-air state; In response to the monitoring terminal being in the open-air state, a start command is sent to the acoustic sensor of the monitoring terminal to drive the acoustic sensor into the acoustic acquisition mode.

3. The disaster avoidance method based on acoustic fingerprint features as described in claim 1, characterized in that, The process involves acquiring real-time environmental acoustic signals collected by the monitoring terminal, performing time-frequency domain transformation on the real-time environmental acoustic signals to extract acoustic fingerprint features, and matching the acoustic fingerprint features with a preset disaster impact model to identify the impact object type, including: The real-time ambient sound signal is decomposed into multiple scales using a wavelet transform algorithm to extract high-frequency subband signals. Determine the peak energy density, pulse duration, and spectral attenuation slope in the high-frequency subband signal, and combine the peak energy density, pulse duration, and spectral attenuation slope into an acoustic fingerprint feature. The acoustic fingerprint features are input into a preset disaster impact model to determine the feature distance between the acoustic fingerprint features and the feature centers of pre-stored solid disaster samples in the preset disaster impact model. When the feature distance is less than a preset threshold, the impact object type is determined to be a solid disaster impact.

4. The disaster avoidance method based on acoustic fingerprint features as described in claim 1, characterized in that, The system receives impact object types and corresponding timestamp information uploaded by multiple monitoring terminals located within the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object type is consistently solid disaster impact, it generates disaster center coordinates and disaster movement vectors, including: Receives impact object type, corresponding timestamp information and geographical location data uploaded by multiple monitoring terminals located in the same geographical area; From the multiple monitoring terminals, those whose impact object type is consistently solid disaster impact are selected; Determine the maximum time difference between the timestamp information corresponding to the selected monitoring terminals; When the maximum time difference is less than or equal to the preset synchronization time window, the selected monitoring terminal is defined as a valid verification terminal. The number of valid verification terminals is counted, and when the number reaches or exceeds a preset regional verification threshold, the geographical location data of all valid verification terminals is extracted. The geographic center point is determined based on the geographic location data of all valid verification terminals, and the geographic center point is used as the coordinates of the disaster center. The disaster center coordinates are added to the historical disaster trajectory sequence in the order of timestamps, and the coordinate changes of multiple consecutive time points in the historical disaster trajectory sequence are vector fitted to generate a disaster movement vector that includes the direction and speed of movement.

5. The disaster avoidance method based on acoustic fingerprint features as described in claim 1, characterized in that, Based on the disaster center coordinates and the disaster movement vector, the disaster movement path is determined, and target protection terminals located within the disaster movement path and in an exposed state are retrieved, including: Starting from the coordinates of the disaster center, the disaster coverage area within the preset warning time is determined according to the direction and speed of movement indicated by the disaster movement vector, and the disaster coverage area is used as the disaster movement path; Obtain the location information of multiple registered associated protection terminals, including the monitoring terminal; Filter out associated protection terminals whose location information is located within the disaster movement path; Collect sensor data from the selected associated protection terminals; Based on the sensor data, determine whether the associated protection terminal is in an open-air state; The associated protection terminals that are determined to be in an open-air state are identified as the target protection terminals.

6. The disaster avoidance method based on acoustic fingerprint features as described in claim 1, characterized in that, The system queries available shelter resources around the target protection terminal, plans an evacuation navigation path from the target protection terminal's current location to the available shelter resources, and sends an evacuation command containing the evacuation navigation path to the target protection terminal, including: Centered on the real-time location point of the target protection terminal, a list of location information including underground garages and covered parking lots is retrieved within a preset radius. The available berth data of each location in the location information list is obtained through a real-time data interface, and the locations with available berths indicated by the available berth data are identified as available sheltered location resources. Determine the travel time of the target protection terminal to each available shelter resource, and determine the estimated arrival time of the disaster to each available shelter resource based on the disaster movement vector; Select the available sheltered location resource with the shortest travel time (less than the estimated arrival time) as the target hazard avoidance point, and plan a hazard avoidance navigation path from the current location of the target protection terminal to the target hazard avoidance point; The hazard avoidance navigation path is encapsulated into a hazard avoidance command, which is then pushed to the target protection terminal via a wireless network to trigger the target protection terminal to initiate emergency hazard avoidance guidance.

7. The disaster avoidance method based on acoustic fingerprint features as described in claim 6, characterized in that, After encapsulating the evasion navigation path into an evasion command and pushing the evasion command to the target protection terminal via a wireless network to trigger the target protection terminal to initiate emergency evasion guidance, the method further includes: Monitor changes in the geographical location of the target protection terminal; When the geographical location of the target protection terminal enters the target refuge point and the displacement within a preset time period is less than a preset threshold, the target protection terminal is confirmed to have successfully avoided the danger. Generate a successful risk avoidance confirmation message, associate the successful risk avoidance confirmation message with the identifier of the target protection terminal and store it, and provide incentive feedback to the account associated with the target protection terminal.

8. A disaster avoidance device based on acoustic fingerprint features, characterized in that, The disaster avoidance device based on acoustic fingerprint features includes: An outdoor state determination module is used to collect sensor data from the monitoring terminal to determine whether the monitoring terminal is in an outdoor state, and to activate the acoustic acquisition mode of the monitoring terminal when the monitoring terminal is in the outdoor state. An acoustic fingerprint recognition module is used to acquire real-time environmental sound wave signals collected by the monitoring terminal, perform time-frequency domain transformation on the real-time environmental sound wave signals to extract acoustic fingerprint features, and match the acoustic fingerprint features with a preset disaster impact model to identify the impact object type. The multi-terminal spatiotemporal fusion positioning module is used to receive the impact object type and corresponding timestamp information uploaded by multiple monitoring terminals located in the same geographical area. When the timestamp information of the multiple monitoring terminals is within a preset synchronization time window and the impact object type is consistent as solid disaster impact, the module generates the disaster center coordinates and disaster movement vector. The risk path prediction and target screening module is used to determine the disaster movement path based on the disaster center coordinates and the disaster movement vector, and to retrieve target protection terminals located within the disaster movement path and in an open-air state. The risk avoidance resource planning and command issuance module is used to query available shelter resources around the target protection terminal, plan a risk avoidance navigation path from the current location of the target protection terminal to the available shelter resources, and send a risk avoidance command containing the risk avoidance navigation path to the target protection terminal.

9. A computer device, characterized in that, The computer device includes a memory, a processor, and a disaster avoidance program based on acoustic fingerprint features stored in the memory and executable on the processor. When executed by the processor, the disaster avoidance program based on acoustic fingerprint features implements the steps of the disaster avoidance method based on acoustic fingerprint features as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a disaster avoidance program based on acoustic fingerprint features. When the disaster avoidance program based on acoustic fingerprint features is executed by the processor, it implements the steps of the disaster avoidance method based on acoustic fingerprint features as described in any one of claims 1-7.

Images