Earthquake early warning analysis system and method based on intelligent equipment

Abstract

The invention relates to the technical field of earthquake monitoring and early warning, in particular to an earthquake early warning analysis system and method based on intelligent equipment. The system comprises a terminal sensing and initial judgment module which collects data through a sensor array, identifies P-wave features by using a dynamic weight adjustment model and generates an encrypted initial alarm; the adaptive network transmission module is used for establishing a dynamic self-organizing network based on a D2D protocol and realizing multi-hop relay transmission of alarms by electing relay nodes; the edge aggregation and verification module is used for executing space-time matching and cross verification on a plurality of alarms in a specific area, judging the authenticity of an event and generating graded early warning information; and the cloud control fusion and actuarial module is used for fusing the regional event data and the professional network data, executing high-precision positioning and earthquake magnitude calculation, and issuing a final early warning instruction. According to the invention, a high-density sensing network is constructed by using massive intelligent devices, the early warning blind area is effectively eliminated, and faster, more accurate and more robust earthquake early warning is realized.

Description

Technical Field

[0001] This invention relates to the field of earthquake monitoring and early warning technology, and more specifically, to an earthquake early warning analysis system and method based on intelligent devices. Background Technology

[0002] An earthquake early warning analysis system is a technical system that monitors seismic wave propagation in real time, rapidly calculates source parameters, and issues early warning information to users before destructive seismic waves reach the target area. The essence of this system is not to predict whether an earthquake will occur, but rather to "monitor earthquakes that have already occurred and issue early warnings before the arrival of destructive waves," helping target areas to take emergency measures in advance and reduce casualties and property damage.

[0003] Current mainstream earthquake early warning analysis technologies can be divided into two core methods based on differences in "data source" and "analysis speed": The first is the rapid early warning method based on a single seismic station. This method uses the initial P-wave signal received from a single seismic station to quickly estimate the source parameters. It relies solely on data from a single station, eliminating the need to wait for data aggregation from multiple stations, resulting in extremely fast analysis speed, making it suitable for monitoring near-field earthquakes.

[0004] The precision early warning method based on "multi-station networking" aggregates the initial motion signals of P-waves from multiple seismic stations, determines the epicenter location through "triangulation," and corrects the magnitude by combining amplitude data from multiple stations. This method offers higher analysis accuracy, but requires waiting for data transmission and aggregation. It is suitable for monitoring "tele-earthquakes" or "strong earthquakes."

[0005] The existing technology has the following shortcomings, specifically:

[0006] 1. Fatal blind spots in early warning: Existing earthquake early warning systems rely entirely on a limited number of professional earthquake monitoring stations in fixed locations. This sparse deployment means that before the arrival of destructive S-waves, the epicenter area, which most needs early warning, is unable to receive any effective early warning time because it is located in the "blank" center of the station network, creating an unavoidable early warning blind spot and causing failure in the most critical areas.

[0007] 2. The system communication is extremely vulnerable and highly dependent: The data transmission of traditional systems relies heavily on centralized public communication networks (such as cellular networks and the Internet). However, strong earthquakes can easily damage the infrastructure of these networks (such as optical cables and base stations), causing the entire early warning system to be paralyzed at critical moments due to "information silos". Early warning information cannot be collected and distributed, and the survivability and reliability of the system face severe challenges.

[0008] 3. Insufficient reliability and accuracy of early warning: Due to the single and limited data sources, the existing system has poor filtering ability for local interference (such as large vehicles and mechanical vibration), resulting in a high risk of false alarms. At the same time, for complex source rupture processes, relying solely on a small amount of initial P-wave data can easily underestimate the magnitude of mega-earthquakes (i.e., the "magnitude saturation" problem), causing missed reports or inaccurate warning ranges, which seriously affects the actual effectiveness and credibility of early warning. Summary of the Invention

[0009] The purpose of this invention is to provide an earthquake early warning analysis system and method based on intelligent devices to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention aims to provide an earthquake early warning analysis system based on intelligent devices, comprising: a terminal sensing and preliminary judgment module: used to continuously collect local vibration data and environmental data through a sensor array, perform real-time data fusion and processing, identify earthquake P-wave characteristics through a dynamic weight adjustment model, and generate and output an encrypted preliminary alarm containing confidence level and characteristic parameters.

[0011] Adaptive network transmission module: connected to the terminal sensing and preliminary judgment module, built based on the D2D communication protocol, used to automatically build a dynamic self-organizing network among smart devices, form a multi-hop transmission link by electing relay nodes, and perform differential encoding, aggregation and encrypted relay transmission of the preliminary alarm.

[0012] Edge aggregation and verification module: connected to the adaptive network transmission module, used to receive multiple preliminary alarms from a specific geographical area within a preset time window, execute spatiotemporal matching and multi-node cross-validation algorithms to determine whether a regional earthquake event is established. If a regional earthquake event is established, it immediately generates graded early warning information containing the estimated magnitude and affected area and releases it outside the threat area.

[0013] The cloud-controlled fusion and precision calculation module communicates with the edge aggregation and verification module to aggregate and fuse regional event data and professional seismic network data. Based on multi-source information, it performs high-precision earthquake location and magnitude calculation, determines the final three elements of an earthquake, and issues precise early warning instructions and linkage control signals to the public emergency system.

[0014] As a further improvement to this technical solution, the dynamic weight adjustment model is specifically implemented as follows: extracting the data collected by the sensor array and dividing the data collected by the sensor array into a core perception dimension and an auxiliary correction dimension. The core perception dimension includes vibration waveform data and device posture data, while the auxiliary correction dimension includes environmental noise spectrum data and environmental air pressure fluctuation data.

[0015] Historical earthquake sample data is obtained from the database, and offline training is performed based on the historical earthquake sample data. The core perception dimension is assigned an initial weight ratio higher than that of the auxiliary correction dimension, among which vibration waveform data is assigned the highest initial weight as the core basis for P-wave identification.

[0016] A reinforcement learning framework is used to optimize the weights. The state space is based on the environmental noise level, the type of historical false alarms of the device, and the quality of sensor data. The action space is based on the preset step size increase or decrease of the feature weights of each dimension. The positive reward is the improvement of the P-wave recognition accuracy, and the negative penalties are false alarms and missed alarms. The weight matrix is ​​iteratively updated through a temporal difference algorithm. The weight matrix is ​​stored with low precision quantization. Only the feature channels that contribute more than a preset contribution threshold to P-wave recognition are retained, resulting in a dynamic weight adjustment model.

[0017] As a further improvement to this technical solution, the automatic formation of a dynamic self-organizing network among smart devices is specifically implemented as follows: after a smart device starts up, it broadcasts a probe frame containing the device's unique identifier and hardware feature code through the D2D communication protocol. After receiving the frame, a neighboring smart device verifies the legitimacy of the sender's identifier. If the verification is successful, it returns a response frame containing its own identifier and communication capabilities, and the two parties establish a temporary communication link.

[0018] Based on the signal strength parameters and relative location information in the response frame, each intelligent device constructs a priority list of neighboring devices, forming a star-shaped subnet centered on the core device. Each subnet achieves full network interconnection through cross-connected intelligent devices.

[0019] Smart devices periodically send status heartbeat packets to connected nodes, including current battery level, remaining bandwidth, and location changes. If a receiving node does not receive heartbeat packets for a continuous period or its signal strength drops below a threshold, it determines that the connection has failed and automatically selects the next smart device from the priority list to rebuild the connection.

[0020] When a smart device detects that its location has moved beyond a preset distance, it actively sends location update information to the original connection node and broadcasts a new detection frame to discover new smart devices in the vicinity. Based on the movement speed and direction, it predicts the connection that may be lost and establishes a preliminary connection with the newly discovered smart device in advance.

[0021] The relay forwarding volume of each node is counted in real time. When the forwarding volume of a node exceeds the preset forwarding volume threshold, it automatically sends a load balancing request to the neighboring nodes. The neighboring nodes respond to the request according to their remaining resources and share part of the relay task.

[0022] As a further improvement to this technical solution, the election relay node forms a multi-hop transmission link. The specific implementation method is as follows: after the intelligent device accesses the self-organizing network, it automatically detects its own hardware capabilities and operating status. Devices that meet the preset basic conditions actively broadcast candidate nomination information containing their own capability parameters to the network.

[0023] After receiving candidate nomination information, each intelligent device in the network quantifies and scores the candidate nodes based on a preset evaluation model. The evaluation dimensions include communication coverage, historical forwarding reliability, resource redundancy, and location stability.

[0024] Each smart device sends a vote confirmation to the candidate node with the highest priority based on the scoring results. When the number of valid votes received by a candidate node reaches a preset proportion of the total number of devices in its local network, it automatically becomes a formal relay node and broadcasts the election declaration to the entire network.

[0025] The term of a relay node is set to a dynamic duration. During its term, it continuously sends status heartbeats to surrounding smart devices. Before the end of its term, the smart devices within the current coverage area are re-evaluated based on its performance during the term. If it meets the conditions for re-election, its term is automatically extended; otherwise, a new round of election is initiated.

[0026] When a relay node experiences a sudden failure or its performance drops below a threshold, the smart devices directly connected to it immediately initiate a temporary election to quickly select a replacement node from the backup candidate nodes.

[0027] As a further improvement to this technical solution, the specific implementation method of the spatiotemporal matching and multi-node cross-validation algorithm is as follows: analyze multiple preliminary alarms received within a preset time window and a specific geographical area, and extract the P-wave arrival time, smart device geographical coordinates, vibration characteristic parameters and confidence values ​​contained in each alarm.

[0028] Using the arrival time of the P-wave of the first valid alarm in the region as a benchmark, calculate the difference between the arrival time of the P-wave of other alarms and the benchmark time, and calculate the standard deviation of all differences. If the standard deviation is less than the preset time threshold, it is determined to be a time series match; otherwise, alarms that exceed the time deviation threshold are marked as suspicious data.

[0029] Based on the geographical coordinates of each smart device and the arrival time of the P wave, the back projection algorithm is used to calculate the theoretical epicenter candidate area. The spatial distance between the actual location of each smart device and the theoretical epicenter candidate area is statistically analyzed, and the average distance and the coefficient of variation are calculated. If the coefficient of variation is less than the preset spatial threshold, it is determined to be a spatial distribution match.

[0030] For alarms that pass spatiotemporal matching, calculate the cosine similarity and peak deviation rate of vibration characteristic parameters, and retain alarms with similarity higher than a preset feature threshold and peak deviation rate lower than a preset deviation threshold to form an effective sample set.

[0031] The confidence values ​​of the valid sample set are weighted and summed. If the weighted result is higher than the preset confidence threshold and the proportion of the number of valid samples to the total number of initial alarms exceeds the preset threshold, then the regional earthquake event is determined to be valid. Otherwise, it is determined to be a non-earthquake event.

[0032] As a further improvement to this technical solution, the specific implementation method of the hierarchical early warning information is as follows:

[0033] Based on regional event data obtained through spatiotemporal matching and multi-node cross-validation, a finite fault model is used to calculate and determine the predicted epicenter location, magnitude, and seismic wave propagation velocity parameters. Combined with pre-set regional geological structure data and site effect parameters, the predicted intensity values ​​at different locations within the affected area are calculated using the ground motion prediction equation. Based on the predicted intensity values, the affected area is divided into three warning levels, and each warning level is matched with a corresponding warning instruction text and response suggestion.

[0034] As a further improvement to this technical solution, the high-precision earthquake positioning based on multi-source information is specifically implemented as follows: time synchronization calibration and outlier removal are performed on regional event data and professional seismic network data, retaining valid data with timestamp errors within a preset range, constructing a multi-source data fusion positioning model, assigning weights to smart device data and seismic network data according to spatial distribution density, using iterative weighted least squares method to solve for epicenter coordinates, taking the distance residuals from each smart device to the theoretical epicenter as the optimization target, and reducing the sum of squared residuals through multiple iterations to obtain high-precision epicenter coordinates.

[0035] As a further improvement to this technical solution, the magnitude calculation is specifically implemented as follows: based on data from intelligent devices, the equivalent magnitude of the region is calculated using empirical formulas by utilizing peak vibration parameters and device distribution radius; based on data from professional seismic networks, the surface wave magnitude calculation method is adopted, and the network magnitude is determined by combining the waveform amplitude recorded by the stations with the epicentral distance correction value; the two types of magnitude results are weighted and fused, and a deviation correction factor for historical earthquake events is introduced to generate the final magnitude result.

[0036] The second aspect of the present invention provides a method for an earthquake early warning analysis system based on intelligent devices, comprising: S1, terminal perception and preliminary judgment: for continuously collecting local vibration data and environmental data through a sensor array, fusing and processing the data in real time, identifying earthquake P-wave characteristics through a dynamic weight adjustment model, and generating and outputting an encrypted preliminary alarm containing confidence level and characteristic parameters.

[0037] S2, Adaptive Network Transmission: Based on the D2D communication protocol, it is used to automatically build a dynamic self-organizing network among smart devices. By electing relay nodes, it forms a multi-hop transmission link and performs differential coding, aggregation, and encrypted relay transmission of the initial alarm.

[0038] S3, Edge Aggregation and Verification: This function receives multiple preliminary alerts from a specific geographic area within a preset time window, executes spatiotemporal matching and multi-node cross-validation algorithms to determine whether a regional earthquake event is confirmed. If a regional earthquake event is confirmed, it immediately generates graded early warning information containing the estimated magnitude and affected area and releases it outside the threat area.

[0039] S4, Cloud Control Fusion and Precision Calculation: Used to aggregate and merge regional event data and professional seismic network data, perform high-precision earthquake location and magnitude calculation based on multi-source information, determine the final three elements of an earthquake, and issue precise early warning instructions and linkage control signals to the public emergency system.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] 1. This invention innovatively utilizes a large number of ordinary smart devices (such as smartphones) to construct a high-density, widely distributed earthquake sensing network. This enables a large number of devices gathered in the epicenter area to become the first detectors, realizing "early warning from the epicenter". This fundamentally eliminates the early warning blind spots of traditional technologies and provides the maximum possible time for all areas surrounding the epicenter to take shelter.

[0042] 2. A highly resilient and self-healing communication lifeline has been constructed: By introducing dynamic self-organizing network technology based on the D2D protocol, the system does not rely on any potentially damaged fixed communication infrastructure. Smart devices can automatically network and transmit data through multi-hop relay, ensuring that even in the most extreme disaster situations, early warning information can still be reliably transmitted from the epicenter like a relay race, greatly enhancing the robustness and survivability of the entire system.

[0043] 3. This invention reduces the false alarm rate to an extremely low level through a dual intelligent filtering mechanism of "terminal intelligent identification (dynamic weight model)" and "edge group cross-validation". At the same time, the cloud overcomes the limitations of a single data source by integrating massive amounts of intelligent device data and professional network data, realizing the calculation of source parameters from rapid preliminary estimation to final high-precision determination, ensuring that the early warning information is both "fast" and "accurate", and significantly improving the overall reliability of the system. Attached Figure Description

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

[0045] Figure 1 This is a schematic diagram of the system structure connection of the present invention.

[0046] Figure 2 This is a schematic diagram of the implementation steps of the method of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example: Please refer to Figure 1 As shown, an earthquake early warning analysis system based on intelligent devices is provided, including: a terminal sensing and preliminary judgment module: used to continuously collect local vibration data and environmental data through a sensor array, perform real-time fusion and processing of the data, identify earthquake P-wave characteristics through a dynamic weight adjustment model, and generate and output an encrypted preliminary alarm containing confidence level and characteristic parameters.

[0049] In one specific embodiment, the dynamic weight adjustment model is implemented by extracting data collected by the sensor array and dividing the data collected by the sensor array into a core perception dimension and an auxiliary correction dimension. The core perception dimension includes vibration waveform data and device posture data, while the auxiliary correction dimension includes environmental noise spectrum data and environmental air pressure fluctuation data.

[0050] Accelerometers collect vibration waveform data (for extracting P-wave time / frequency domain features), gyroscopes collect equipment attitude data (for eliminating non-seismic shaking interference from the equipment), microphones collect environmental noise spectrum data (to distinguish between natural vibrations and man-made noise), and barometers collect environmental air pressure fluctuation data (to help determine whether air pressure anomalies are caused by geological activity).

[0051] Historical earthquake sample data is obtained from the database, and offline training is performed based on the historical earthquake sample data. The core perception dimension is assigned an initial weight ratio higher than that of the auxiliary correction dimension, among which vibration waveform data is assigned the highest initial weight as the core basis for P-wave identification.

[0052] Vibration waveform data has the highest weight (as the core basis for P-wave identification), followed by equipment attitude data (used to filter out non-seismic interference), and environmental noise and air pressure data are assigned basic weights as correction terms. The initial weight matrix is ​​generated offline through training with historical earthquake sample data.

[0053] A reinforcement learning framework is used to optimize the weights. The state space is based on the environmental noise level, the type of historical false alarms of the device, and the quality of sensor data. The action space is based on the preset step size increase or decrease of the feature weights of each dimension. The positive reward is the improvement of the P-wave recognition accuracy, and the negative penalties are false alarms and missed alarms. The weight matrix is ​​iteratively updated through a temporal difference algorithm. The weight matrix is ​​stored with low precision quantization. Only the feature channels that contribute more than a preset contribution threshold to P-wave recognition are retained, resulting in a dynamic weight adjustment model.

[0054] It should be noted that the preset step size is set by professionals. For example, it is set differently based on the importance of feature dimensions. The step size of the core perception dimension (such as vibration waveform) is set to a smaller value for fine-tuning, while the step size of the auxiliary correction dimension is set to a larger value for rapid adaptation. At the same time, it is dynamically adjusted in the early stage of model training and the convergence stage. The step size is slightly larger in the early stage to accelerate iteration, and the step size is reduced in the later stage for accurate convergence.

[0055] Adaptive network transmission module: connected to the terminal sensing and preliminary judgment module, built based on the D2D communication protocol, used to automatically build a dynamic self-organizing network among smart devices, form a multi-hop transmission link by electing relay nodes, and perform differential encoding, aggregation and encrypted relay transmission of the preliminary alarm.

[0056] In one specific embodiment, the automatic formation of a dynamic self-organizing network among smart devices is implemented as follows: after a smart device starts up, it broadcasts a probe frame containing the device's unique identifier and hardware feature code through the D2D communication protocol. After receiving the frame, a neighboring smart device verifies the legitimacy of the sender's identifier (based on a pre-stored device whitelist or temporary certificate). If the verification is successful, it returns a response frame containing its own identifier and communication capabilities, and the two parties establish a temporary communication link.

[0057] Based on the signal strength parameters and relative location information in the response frame, each intelligent device constructs a priority list of neighboring devices, forming a star-shaped subnet centered on the core device. Each subnet achieves full network interconnection through cross-connected intelligent devices.

[0058] Smart devices periodically (the period is dynamically adjusted according to network load) send status heartbeat packets to connected nodes, which include the current battery level, remaining bandwidth, and location changes. If a receiving node does not receive heartbeat packets for a continuous period or the signal strength drops below the threshold, it is determined that the connection has failed and the next smart device in the priority list is automatically selected to rebuild the connection.

[0059] It should be noted that the signal strength threshold is set by professionals, for example, based on communication quality requirements and differences in device performance, with the core objective of ensuring that the data transmission error rate is lower than the preset error rate threshold. The basic threshold is determined in conjunction with the hardware specifications of the smart device's communication module. At the same time, the threshold is dynamically adjusted in conjunction with the network load. Under high load, the threshold is appropriately increased to prioritize the transmission of critical data, and under low load, the threshold is decreased to expand the connection coverage.

[0060] When a smart device detects that its location has moved beyond a preset distance, it actively sends location update information to the original connection node and broadcasts a new detection frame to discover new smart devices in the vicinity. Based on the movement speed and direction, it predicts the connection that may be lost and establishes a preliminary connection with the newly discovered smart device in advance.

[0061] It should be noted that the preset distance is set by professionals, for example, with the core setting of ensuring connection stability and network adjustment efficiency, and the basic value is determined by combining the device's communication coverage radius and the distance data of connection failure caused by historical movement; at the same time, it is dynamically adapted to the movement speed, with a smaller value set for high-speed movement to trigger adjustment in advance, and a larger value set for low-speed movement to reduce unnecessary network overhead.

[0062] The relay forwarding volume of each node is counted in real time. When the forwarding volume of a node exceeds the preset forwarding volume threshold, it automatically sends a load balancing request to the neighboring nodes. The neighboring nodes respond to the request based on their remaining resources (power, bandwidth) and share part of the relay task.

[0063] It should be noted that the preset forwarding threshold is set by professionals. For example, it is set with the node's resource carrying capacity and network transmission stability as the core, and the basic threshold is determined by combining the node's remaining power, bandwidth limit and hardware forwarding performance to ensure that the threshold is lower than the node's overload threshold. At the same time, the overall network load is dynamically adjusted. When the network is under high load, the threshold is appropriately reduced to divert traffic in advance, and when the network is under low load, the threshold is increased to reduce the overhead of frequent load scheduling.

[0064] Through the four steps of “device discovery - topology initialization - dynamic adjustment - self-healing optimization”, the technical implementation of “automatic construction of dynamic self-organizing network” is fully defined. It is both dependent on the sovereign adaptive network transmission module and its dynamic characteristics that distinguish it from traditional fixed networks (such as automatic adaptation to device movement and real-time link repair).

[0065] In one specific embodiment, the election relay node forms a multi-hop transmission link. The specific implementation method is as follows: after the smart device accesses the self-organizing network, it automatically detects its own hardware capabilities (communication module power, data processing performance) and operating status (remaining power, current load). Devices that meet the preset basic conditions actively broadcast candidate nomination information containing their own capability parameters to the network.

[0066] After receiving candidate nomination information, each intelligent device in the network quantifies and scores the candidate nodes based on a preset evaluation model. The evaluation dimensions include communication coverage, historical forwarding reliability, resource redundancy, and location stability.

[0067] The dimensions include communication coverage, calculated based on signal strength attenuation curves; historical forwarding reliability, calculated based on recent period data success forwarding rate; resource redundancy, calculated based on remaining power and idle bandwidth ratio; and location stability, calculated based on recent displacement changes.

[0068] Each smart device sends a vote confirmation to the candidate node with the highest priority based on the scoring results. When the number of valid votes received by a candidate node reaches a preset proportion of the total number of devices in its local network, it automatically becomes a formal relay node and broadcasts the election declaration to the entire network.

[0069] The term of relay nodes is set to a dynamic duration (adjusted according to network load fluctuations). During the term, they continuously send status heartbeats to surrounding smart devices. Before the end of the term, the smart devices within the current coverage area are re-evaluated based on their performance during the term (forwarding delay, packet loss rate). If they meet the conditions for re-election, their term is automatically extended; otherwise, a new round of election is initiated.

[0070] When a relay node experiences a sudden failure (such as power depletion or communication interruption) or its performance drops below a threshold, the smart devices directly connected to it immediately initiate a temporary election to quickly select a replacement node from the backup candidate nodes.

[0071] Edge aggregation and verification module: connected to the adaptive network transmission module, used to receive multiple preliminary alarms from a specific geographical area within a preset time window, execute spatiotemporal matching and multi-node cross-validation algorithms to determine whether a regional earthquake event is established. If a regional earthquake event is established, it immediately generates graded early warning information containing the estimated magnitude and affected area and releases it outside the threat area.

[0072] "Preset time window" is a fixed time interval set based on the propagation speed of seismic waves (especially P waves) on the Earth's surface. Its core purpose is to ensure that multiple initial alerts received "originate from the same seismic event," rather than independent interference from different times (such as accidental equipment shaking or traffic vibrations). "Specific geographic area" is a circular / polygonal geographic range defined by the edge server based on the "location of the device that triggered the first initial alert." Its core purpose is to ensure that the received initial alerts "come from the core area affected by the earthquake," eliminating false alarms from distant, unrelated devices. "Multiple initial alerts" refers to receiving at least a preset number (e.g., 3 or more) of initial alerts under the constraints of the above "time window + geographic area." Its core purpose is to reduce the risk of false alarms from a single device through "cross-verification by multiple devices" (e.g., a device triggering an alert due to a drop or collision).

[0073] In one specific embodiment, the execution of the spatiotemporal matching and multi-node cross-validation algorithm is specifically implemented by: parsing multiple preliminary alarms received within a preset time window and a specific geographical area, and extracting the P-wave arrival time, smart device geographical coordinates, vibration characteristic parameters, and confidence values ​​contained in each alarm.

[0074] Using the arrival time of the P-wave of the first valid alarm in the region as a benchmark, calculate the difference between the arrival time of the P-wave of other alarms and the benchmark time, and calculate the standard deviation of all differences. If the standard deviation is less than the preset time threshold, it is determined to be a time series match; otherwise, alarms that exceed the time deviation threshold are marked as suspicious data.

[0075] It should be noted that the preset time threshold is set by professionals. For example, it is set based on the propagation law of seismic waves and the distribution of regional equipment. The basic threshold is calculated by combining the propagation speed of P waves in the earth's crust (3-6 km / s) and the average distribution density of smart devices in the region to ensure that the alarm time difference caused by the same earthquake is within a reasonable range. At the same time, it is calibrated with reference to the standard deviation data of the arrival time of P waves of multiple smart devices in historical earthquake events to take into account the need to reduce misjudgment and accurately screen co-seismic alarms.

[0076] Based on the geographical coordinates of each smart device and the arrival time of the P wave, the back projection algorithm is used to calculate the theoretical epicenter candidate area. The spatial distance between the actual location of each smart device and the theoretical epicenter candidate area is statistically analyzed, and the average distance and the coefficient of variation are calculated. If the coefficient of variation is less than the preset spatial threshold, it is determined to be a spatial distribution match.

[0077] It should be noted that the preset spatial threshold is set by professionals, for example, with the core objective of ensuring the accuracy of epicenter location and the effectiveness of spatial matching. The baseline value is determined by combining the location error characteristics of the back-projection algorithm, the distribution density of regional smart devices, and the dispersion coefficient data of historical earthquake epicenter locations. Simultaneously, it is dynamically calibrated in conjunction with the estimated magnitude; for high magnitudes, the threshold is appropriately lowered to rigorously screen for accurate matching data, while for low magnitudes, the threshold is slightly increased to adapt to the location error range. The back-projection algorithm is existing technology and will not be described in detail here.

[0078] For alarms that pass spatiotemporal matching, calculate the cosine similarity and peak deviation rate of vibration characteristic parameters, and retain alarms with similarity higher than a preset feature threshold and peak deviation rate lower than a preset deviation threshold to form an effective sample set.

[0079] It should be noted that the preset feature threshold and preset deviation threshold are set by professionals. For example, the preset feature threshold (cosine similarity) is set based on the consistency of vibration feature parameters, combined with the cosine similarity distribution data of historical earthquake P-wave waveforms, to ensure that alarms with highly matched waveform features are selected. The preset deviation threshold (peak deviation rate) is set based on the reasonable fluctuation range of historical earthquake vibration peak values, while also taking into account the calibration deviation of the equipment sensor accuracy. Both take into account the feature matching accuracy and the tolerance for minor equipment errors. The calculation of the cosine similarity and peak deviation rate of vibration feature parameters is existing technology and will not be elaborated on here.

[0080] The confidence values ​​of the valid sample set are weighted and summed. If the weighted result is higher than the preset confidence threshold and the proportion of the number of valid samples to the total number of initial alarms exceeds the preset threshold, then the regional earthquake event is determined to be valid. Otherwise, it is determined to be a non-earthquake event.

[0081] It should be noted that the preset confidence threshold and preset quantity threshold are set by professionals, for example, based on statistical analysis of massive amounts of historical earthquake event data and interference event data. The core objective is to achieve the optimal balance between maximizing the success rate of early warnings and minimizing the risk of false alarms. Specific values ​​are ultimately determined after repeated testing and optimization in different geographical regions and network environments, and can be dynamically adjusted based on system operation feedback.

[0082] In one specific embodiment, the graded early warning information is implemented as follows: based on regional event data that has been matched and cross-validated by multiple nodes, a finite fault model is used to calculate and determine the estimated epicenter location, magnitude, and seismic wave propagation velocity parameters. Combined with preset regional geological structure data and site effect parameters, the estimated intensity values ​​of different locations in the affected area are calculated through the seismic motion prediction equation. Based on the estimated intensity values, the affected area is divided into three early warning levels, and a corresponding early warning instruction text and response suggestion are matched for each early warning level.

[0083] It should be noted that the calculation and determination of the predicted epicenter location first involves extracting the P-wave arrival times and device coordinates from multiple intelligent devices in the regional event data; then, based on the regional geological data, pre-setting geometric parameters such as the strike, dip, and slip angle of faults, constructing a fault grid and dividing it into several sub-faults; finally, minimizing the residual between the observed values ​​and model calculation values ​​of the P-wave arrival times of each device using an inversion algorithm to determine the equivalent rupture initiation point on the fault, the projection of this point onto the Earth's surface being the predicted epicenter location; the calculation and determination of the predicted magnitude first involves calculating the fault rupture area based on the fault parameters and width obtained from the inversion; then, extracting the peak ground acceleration (PGA) of each device from the regional event data, and combining this with the distance between the device and the predicted epicenter, correcting it using an empirical formula to obtain the equivalent PGA at the hypocenter; finally, based on the fault rupture area and equivalent PGA, substituting them into the magnitude calculation formula to obtain the predicted magnitude; the calculation and determination of the predicted seismic wave propagation velocity parameters utilizes the P-wave arrival time difference of multiple intelligent devices and the distance between the devices to invert the propagation velocity of seismic waves (mainly P-waves) in the regional medium. First, two devices with known distances are selected, and their P-wave arrival time difference is calculated. Then, assuming that the seismic wave propagates in a homogeneous medium, the propagation velocity is initially estimated. Finally, combined with regional geological structure data (such as the distribution of strata lithology), the velocity is corrected by layer (the velocity is different for strata at different depths) to obtain the final seismic wave propagation velocity parameters.

[0084] The preset regional geological structure data mainly refers to key information such as the known distribution of active faults, fault types (such as strike-slip and thrust), and regional stress fields, which are used to constrain the possible rupture modes of the finite fault model.

[0085] The preset site effect parameters refer to the site amplification effect coefficients plotted based on local soil structure and geological conditions (such as soft soil thickness and Vs30 wave velocity), which are used to correct the amplification or attenuation effect of bedrock ground motion propagating to the ground surface.

[0086] Level 1 Warning (Minor Impact): When the estimated magnitude M < 3.0, or 3.0 ≤ M < 4.0 and the distance D > 30 km, the warning information includes "minor tremor reminder" and "no emergency evacuation required";

[0087] Level 2 warning (moderate impact): When the magnitude is 4.0 ≤ M < 5.0 and the distance to D ≤ 30 km, or when the magnitude is 5.0 ≤ M < 6.0 and the distance to D > 50 km, the warning information includes "moderate tremor warning", "take shelter nearby (e.g., stay away from windows and furniture)" and "estimated arrival time of the tremor".

[0088] Level 3 Warning (Severe Impact): When the estimated magnitude M ≥ 5.0 and the distance D ≤ 50 km, or M ≥ 6.0, the warning information includes "Emergency Evacuation Alert", "Immediate Evacuation to an Open Area (or the Base of a Load-Bearing Wall)", "Estimated Time of Earthquake Arrival", and "Location of the Nearest Emergency Shelter".

[0089] Based on the real-time geographic location of smart devices, the corresponding level of early warning information is only pushed to devices "outside the threat area and possibly affected by the earthquake"—Level 1 warnings are pushed to within 50 kilometers of the epicenter, Level 2 warnings to within 100 kilometers of the epicenter, and Level 3 warnings to within 200 kilometers of the epicenter, avoiding the push of invalid information to unaffected areas.

[0090] The cloud-controlled fusion and precision calculation module communicates with the edge aggregation and verification module to aggregate and fuse regional event data and professional seismic network data. Based on multi-source information, it performs high-precision earthquake location and magnitude calculation, determines the final three elements of an earthquake, and issues precise early warning instructions and linkage control signals to the public emergency system.

[0091] In one specific embodiment, the method for performing high-precision earthquake positioning based on multi-source information is as follows: time synchronization calibration and outlier removal are performed on regional event data and professional seismic network data, valid data with timestamp errors within a preset range are retained, a multi-source data fusion positioning model is constructed, the data from smart devices and the network data are weighted according to spatial distribution density, the epicenter coordinates are solved using the iterative weighted least squares method, the residual distance from each smart device to the theoretical epicenter is used as the optimization objective, and the sum of squared residuals is reduced through multiple iterations to obtain high-precision epicenter coordinates.

[0092] It should be noted that the preset range is determined based on the statistical distribution characteristics of typical synchronization errors between smart device clocks and national standard time, as well as network transmission delays. By analyzing the timestamp error distribution of valid earthquake events and interference events in historical data, the preset range is set as a reasonable interval (e.g., ±500 milliseconds to ±2 seconds) that can cover the vast majority of real events while excluding obviously abnormal data. This range can be configured differently according to the communication conditions and device performance in different regions, and can be dynamically optimized and adjusted as data accumulates during system operation.

[0093] In one specific embodiment, the magnitude calculation is implemented as follows: based on data from intelligent devices, the equivalent magnitude of the region is calculated using empirical formulas by utilizing peak vibration parameters and device distribution radius; based on data from professional seismic networks, the surface wave magnitude calculation method is adopted, and the network magnitude is determined by combining the waveform amplitude recorded by the stations with the epicentral distance correction value; the two types of magnitude results are weighted and fused, and a deviation correction factor for historical earthquake events is introduced to generate the final magnitude result.

[0094] It should be noted that the empirical formulas and surface wave magnitude calculation methods mentioned are existing technologies and will not be elaborated upon here.

[0095] See Figure 2 As shown, a method for an earthquake early warning analysis system based on intelligent devices is provided, including: S1, terminal perception and preliminary judgment: used to continuously collect local vibration data and environmental data through a sensor array, perform real-time fusion and processing of the data, identify earthquake P-wave characteristics through a dynamic weight adjustment model, and generate and output an encrypted preliminary alarm containing confidence level and characteristic parameters.

[0096] S2, Adaptive Network Transmission: Based on the D2D communication protocol, it is used to automatically build a dynamic self-organizing network among smart devices. By electing relay nodes, it forms a multi-hop transmission link and performs differential coding, aggregation, and encrypted relay transmission of the initial alarm.

[0097] S3, Edge Aggregation and Verification: This function receives multiple preliminary alerts from a specific geographic area within a preset time window, executes spatiotemporal matching and multi-node cross-validation algorithms to determine whether a regional earthquake event is confirmed. If a regional earthquake event is confirmed, it immediately generates graded early warning information containing the estimated magnitude and affected area and releases it outside the threat area.

[0098] S4, Cloud Control Fusion and Precision Calculation: Used to aggregate and merge regional event data and professional seismic network data, perform high-precision earthquake location and magnitude calculation based on multi-source information, determine the final three elements of an earthquake, and issue precise early warning instructions and linkage control signals to the public emergency system.

[0099] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. An earthquake early warning and analysis system based on intelligent devices, characterized in that, include. Terminal perception and preliminary judgment module: It is used to continuously collect local vibration data and environmental data through sensor array, perform real-time fusion and processing of data, identify earthquake P-wave characteristics through dynamic weight adjustment model, and generate and output encrypted preliminary alarm containing confidence level and characteristic parameters. Adaptive network transmission module: connected to the terminal sensing and preliminary judgment module, built based on the D2D communication protocol, used to automatically build a dynamic self-organizing network among smart devices, form a multi-hop transmission link by electing relay nodes, and perform differential coding, aggregation and encrypted relay transmission of the preliminary alarm; Edge aggregation and verification module: connected to the adaptive network transmission module, used to receive multiple preliminary alarms from a specific geographical area within a preset time window, execute spatiotemporal matching and multi-node cross-validation algorithms to determine whether a regional earthquake event is established; if a regional earthquake event is established, immediately generate graded early warning information containing the estimated magnitude and affected area, and release it outside the threat area. The cloud-controlled fusion and precision calculation module communicates with the edge aggregation and verification module to aggregate and fuse regional event data and professional seismic network data. Based on multi-source information, it performs high-precision earthquake location and magnitude calculation, determines the final three elements of an earthquake, and issues precise early warning instructions and linkage control signals to the public emergency system.

2. The earthquake early warning and analysis system based on intelligent devices according to claim 1, characterized in that, The dynamic weight adjustment model is specifically implemented as follows: Extract the data collected by the sensor array and divide the data collected by the sensor array into core perception dimension and auxiliary correction dimension. The core perception dimension includes vibration waveform data and equipment attitude data, while the auxiliary correction dimension includes environmental noise spectrum data and environmental air pressure fluctuation data. Historical earthquake sample data is obtained from the database, and offline training is performed based on the historical earthquake sample data. The core perception dimension is assigned an initial weight ratio higher than that of the auxiliary correction dimension, among which vibration waveform data is assigned the highest initial weight as the core basis for P-wave identification. A reinforcement learning framework is used to optimize the weights. The state space is based on the environmental noise level, the type of historical false alarms of the device, and the quality of sensor data. The action space is based on the preset step size increase or decrease of the feature weights of each dimension. The positive reward is the improvement of the P-wave recognition accuracy, and the negative penalties are false alarms and missed alarms. The weight matrix is ​​iteratively updated through a temporal difference algorithm. The weight matrix is ​​stored with low precision quantization. Only the feature channels that contribute more than a preset contribution threshold to P-wave recognition are retained, resulting in a dynamic weight adjustment model.

3. The earthquake early warning and analysis system based on intelligent devices according to claim 1, characterized in that, The specific method for automatically establishing a dynamic self-organizing network among smart devices is as follows: After a smart device is started, it broadcasts a probe frame containing the device's unique identifier and hardware feature code via the D2D communication protocol. When a nearby smart device receives the frame, it verifies the legitimacy of the sender's identifier. If the verification is successful, it returns a response frame containing its own identifier and communication capabilities, and the two parties establish a temporary communication link. Based on the signal strength parameters and relative location information in the response frame, each intelligent device constructs a priority list of neighboring devices, forming a star-shaped subnet centered on the core device. Each subnet achieves full network interconnection through cross-connected intelligent devices. Smart devices periodically send status heartbeat packets to connected nodes, including current battery level, remaining bandwidth, and location changes. If a receiving node does not receive heartbeat packets for a continuous period or its signal strength drops below the threshold, it is determined that the connection has failed and automatically selects the next smart device from the priority list to rebuild the connection. When a smart device detects that its position has moved beyond a preset distance, it actively sends location update information to the original connection node and broadcasts a new detection frame to discover new smart devices in the vicinity. Based on the speed and direction of movement, it predicts the connection that may be lost and establishes a preliminary connection with the newly discovered smart device in advance. The relay forwarding volume of each node is counted in real time. When the forwarding volume of a node exceeds the preset forwarding volume threshold, it automatically sends a load balancing request to the neighboring nodes. The neighboring nodes respond to the request according to their remaining resources and share part of the relay task.

4. The earthquake early warning and analysis system based on intelligent devices according to claim 1, characterized in that, The election of relay nodes forms a multi-hop transmission link, and the specific implementation method is as follows: After a smart device connects to an ad hoc network, it automatically detects its own hardware capabilities and operating status. Devices that meet the preset basic conditions will proactively broadcast candidate nomination information containing their own capability parameters to the network. After receiving candidate nomination information, each intelligent device in the network quantifies and scores the candidate nodes based on a preset evaluation model. The evaluation dimensions include communication coverage, historical forwarding reliability, resource redundancy, and location stability. Each smart device sends a vote confirmation to the candidate node with the highest priority based on the scoring results. When the number of valid votes received by a candidate node reaches a preset proportion of the total number of devices in its local network, it automatically becomes a formal relay node and broadcasts the election declaration to the entire network. The term of relay nodes is set to a dynamic duration. During the term, they continuously send status heartbeats to surrounding smart devices. Before the end of the term, the smart devices within the current coverage area are re-evaluated based on their performance during the term. If they meet the conditions for re-election, their term is automatically extended; otherwise, a new round of election is initiated. When a relay node experiences a sudden failure or its performance drops below a threshold, the smart devices directly connected to it immediately initiate a temporary election to quickly select a replacement node from the backup candidate nodes.

5. The earthquake early warning and analysis system based on intelligent devices according to claim 1, characterized in that, The specific implementation method of the spatiotemporal matching and multi-node cross-validation algorithm is as follows: Multiple preliminary alarms received within a preset time window and a specific geographical area are analyzed to extract the P-wave arrival time, smart device geographical coordinates, vibration characteristic parameters, and confidence values ​​contained in each alarm. Using the arrival time of the P-wave of the first valid alarm in the area as the benchmark, calculate the difference between the arrival time of the P-wave of other alarms and the benchmark time, and calculate the standard deviation of all differences. If the standard deviation is less than the preset time threshold, it is determined to be a time series match; otherwise, alarms that exceed the time deviation threshold are marked as suspicious data. Based on the geographical coordinates of each smart device and the arrival time of the P wave, the back projection algorithm is used to calculate the theoretical epicenter candidate area. The spatial distance between the actual location of each smart device and the theoretical epicenter candidate area is statistically analyzed, and the average distance and the coefficient of variation are calculated. If the coefficient of variation is less than the preset spatial threshold, it is determined to be a spatial distribution match. For alarms that are matched in time and space, the cosine similarity and peak deviation rate of vibration feature parameters are calculated. Alarms with similarity higher than a preset feature threshold and peak deviation rate lower than a preset deviation threshold are retained to form an effective sample set. The confidence values ​​of the valid sample set are weighted and summed. If the weighted result is higher than the preset confidence threshold and the proportion of the number of valid samples to the total number of initial alarms exceeds the preset threshold, then the regional earthquake event is determined to be valid. Otherwise, it is determined to be a non-earthquake event.

6. The earthquake early warning and analysis system based on intelligent devices according to claim 1, characterized in that, The specific implementation method of the hierarchical early warning information is as follows: Based on regional event data obtained through spatiotemporal matching and multi-node cross-validation, a finite fault model is used to calculate and determine the predicted epicenter location, magnitude, and seismic wave propagation velocity parameters. Combined with pre-set regional geological structure data and site effect parameters, the predicted intensity values ​​at different locations within the affected area are calculated using the ground motion prediction equation. Based on the predicted intensity values, the affected area is divided into three warning levels, and each warning level is matched with a corresponding warning instruction text and response suggestion.

7. The earthquake early warning and analysis system based on intelligent devices according to claim 1, characterized in that, The specific implementation method for performing high-precision seismic positioning based on multi-source information is as follows: Time synchronization calibration and outlier removal are performed on regional event data and professional seismic network data. Valid data with timestamp errors within a preset range are retained. A multi-source data fusion positioning model is constructed. Smart device data and seismic network data are weighted according to spatial distribution density. The epicenter coordinates are solved using the iterative weighted least squares method. The residual distance from each smart device to the theoretical epicenter is used as the optimization objective. Through multiple iterations, the sum of squared residuals is reduced to obtain high-precision epicenter coordinates.

8. The earthquake early warning and analysis system based on intelligent devices according to claim 1, characterized in that, The specific method for calculating the magnitude is as follows: Based on data from smart devices, the equivalent magnitude of the region is calculated using empirical formulas by utilizing peak vibration parameters and device distribution radius. Based on data from professional seismic networks, the surface wave magnitude calculation method is used, and the network magnitude is determined by combining the waveform amplitude recorded by the stations with the epicentral distance correction value. The two types of magnitude results are weighted and fused, and a deviation correction factor for historical earthquake events is introduced to generate the final magnitude result.

9. A method for implementing the earthquake early warning analysis system based on intelligent devices according to any one of claims 1-8, characterized in that, include: S1. Terminal perception and preliminary judgment: It is used to continuously collect local vibration data and environmental data through sensor array, perform real-time fusion and processing of data, identify earthquake P-wave characteristics through dynamic weight adjustment model, and generate and output encrypted preliminary alarm containing confidence level and characteristic parameters. S2, Adaptive Network Transmission: Based on the D2D communication protocol, it is used to automatically build a dynamic self-organizing network among smart devices. By electing relay nodes, it forms a multi-hop transmission link and performs differential coding, aggregation and encrypted relay transmission of the initial alarm. S3, Edge Aggregation and Verification: Used to receive multiple preliminary alerts from a specific geographic area within a preset time window, execute spatiotemporal matching and multi-node cross-validation algorithms to determine whether a regional earthquake event is confirmed. If a regional earthquake event is confirmed, it immediately generates graded early warning information containing the estimated magnitude and affected area and releases it outside the threat area. S4, Cloud Control Fusion and Precision Calculation: Used to aggregate and merge regional event data and professional seismic network data, perform high-precision earthquake location and magnitude calculation based on multi-source information, determine the final three elements of an earthquake, and issue precise early warning instructions and linkage control signals to the public emergency system.

Images