Geological disaster remote sound-light alarm and data acquisition system and method

Abstract

The invention relates to the technical field of geological disaster alarm, and discloses a geological disaster remote sound-light alarm and data acquisition system and method, which realize intelligent judgment of early warning level through an early warning rule and a trend model on the basis of supporting acquisition of multi-source geological and environmental data, and realize remote sound-light alarm and data acquisition of geological disasters by introducing a regional linkage mechanism. And the linkage coordination capability of early warning accuracy and alarm response is effectively improved. Besides, according to factors such as monitoring point risk levels, geographic topology and key facility distribution, response priorities of acousto-optic devices in an area are dynamically allocated, multi-point multi-level differential alarm control is realized, post-disaster traceability analysis is realized by using alarm state feedback operation, and the risk level of the monitoring point is greatly improved. And the alarm reliability and the emergency response efficiency in a complex geological scene are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of geological disaster alarm technology, and in particular to a geological disaster remote sound and light alarm and data acquisition system and method. Background Art

[0002] Geological disasters such as landslides, debris flows, and mountain collapses are often characterized by sudden onset, short warning times, and wide-ranging impacts. Once they occur, they pose a significant threat to life and property. To this end, various geological disaster early warning systems are becoming increasingly widespread. Some systems deploy sensors to collect geological parameters, transmit this data to back-end platforms for processing, and trigger audible and visual alarms when specific warning conditions are met. However, current systems generally suffer from rigid response mechanisms and simple alarm judgment rules, leading to high false alarm rates, delayed responses, and poor tracing. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a geological disaster remote sound and light alarm and data acquisition system and method to solve the technical problems existing in the existing geological disaster alarm, such as high false alarm rate, delayed response and poor tracing effect.

[0004] The first aspect of the present invention discloses a geological disaster remote sound and light alarm and data acquisition system, which includes a monitoring terminal, an alarm control center, a sound and light alarm device and a remote platform; wherein,

[0005] The monitoring terminal is used to collect multi-source monitoring data related to geological disasters and transmit it to the alarm control center in real time;

[0006] The alarm control center includes an intelligent determination unit and a regional linkage unit;

[0007] The intelligent judgment unit is used to evaluate the warning level based on multi-source monitoring data, historical disaster information and environmental factors, and output the corresponding alarm trigger signal;

[0008] The regional linkage unit is used to generate a linkage control strategy when it detects that the target monitoring point is at a medium or high warning level, combining the spatial geographical distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices of the geographical area to which the monitoring point belongs;

[0009] The sound and light alarm device is used to receive the alarm trigger signal and linkage control strategy, and perform corresponding sound and light alarm operations;

[0010] The remote platform is used to receive and store alarm records and alarm execution results reported by the alarm control center and the sound and light alarm devices for subsequent disaster tracing and analysis.

[0011] Furthermore, the monitoring terminal includes a collection unit and a processing unit;

[0012] The acquisition unit is used to synchronously collect surface deformation, environmental vibration, image information and rain intensity data as multi-source monitoring data; the acquisition unit includes a strain sensor, an inclination sensor, an acceleration sensor, an image acquisition device and a rain gauge;

[0013] The processing unit is used to perform time alignment, spatial consistency detection and feature fusion operations on multi-source monitoring data to form a multimodal feature vector and transmit it to the alarm control center.

[0014] Furthermore, the processing unit is further configured to sequentially perform abnormal segment identification, data compression, and data hierarchical storage operations on the collected data;

[0015] Among them, abnormal fragments are obtained through normal state reference model identification. After the abnormal data fragments are identified, the abnormal fragments are distinguished from the normal segment data, and the mutation time index and geographic location information are embedded when compressing the abnormal fragment data. According to the abnormal level of the abnormal fragment, the compressed data is stored in a hierarchical manner and transmitted to the remote platform in a hierarchical manner.

[0016] Furthermore, the intelligent decision unit includes an early warning rule engine, a trend modeling subunit and a weight adjustment subunit; wherein,

[0017] The early warning rule engine is used to form a combination of early warning indicators based on historical multi-source monitoring data and environmental factors, set corresponding thresholds for each early warning indicator, and determine multi-dimensional threshold matching rules based on the set thresholds;

[0018] The trend modeling subunit is used to construct a risk evolution trend curve based on the multimodal feature vectors of continuous time periods, and to determine trend labels of rising risk levels, increased volatility, or high-level stagnation based on the risk evolution trend curve;

[0019] The dynamic weight adjustment unit is used to periodically update the weights of each warning indicator in the multidimensional threshold matching rule by combining the historical false alarm rate, sensor signal-to-noise ratio and risk evolution trend output by the trend modeling subunit.

[0020] Furthermore, the intelligent determination unit further includes a rule determination subunit, a model determination subunit and a warning level assessment subunit;

[0021] Among them, the rule judgment subunit is used to determine the current warning indicator value based on the current multimodal feature vector and current environmental factors, and judge the current warning indicator value according to the updated multidimensional threshold matching rule to form a basic risk warning result;

[0022] The model judgment subunit is used to identify and classify the current multimodal feature vector based on the trained disaster identification model and output a disaster risk identification label;

[0023] The warning level assessment subunit is used to perform a weighted fusion operation based on the output results of the rule judgment subunit and the model judgment subunit to generate a comprehensive warning level label, and generate a corresponding alarm trigger signal based on the comprehensive warning level label.

[0024] Furthermore, the process of constructing the risk evolution trend curve based on the multimodal feature vectors of the continuous time period by the trend modeling subunit specifically includes:

[0025] The trend modeling subunit performs a time series analysis operation on the multimodal feature vectors within a continuous time period based on the trend model to generate a risk evolution trend curve;

[0026] The trend model is constructed by combining residual autoregressive analysis with a change point detection algorithm, and is used to identify trend mutation points, periodic risk turning points, and risk mitigation cycles.

[0027] Furthermore, the regional linkage unit includes a regional topology construction subunit and an early warning priority determination subunit;

[0028] The regional topology construction subunit is used to construct a multi-region linkage topology map based on the spatial geographic distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices in the geographical area to which the monitoring points belong. The nodes in the topology map represent geographical sub-regions, and the edges are used to describe the linkage relationship between regions with the possibility of risk propagation.

[0029] The warning priority determination subunit is used to calculate the linkage response priority score of the area when the warning level label of the target monitoring point is medium or high, based on the geological sensitivity, population density, key facility density and current warning level factor of the area where the target monitoring point is located in the topological map;

[0030] The priority scoring result is used to guide the sound and light alarm device to execute differentiated activation sequences and alarm intensity control strategies in multiple adjacent or overlapping impact areas.

[0031] Furthermore, the sound and light alarm device includes an environment sensing unit and a response adjustment unit; wherein,

[0032] The environmental perception unit is used to detect the ambient noise level, visual distance and density of surrounding sound and light alarm devices in the alarm area as environmental perception results;

[0033] The response adjustment unit adaptively adjusts the sound intensity level, light flash frequency, and alarm duration parameters based on the environmental perception results.

[0034] Furthermore, the sound and light alarm device further includes a state feedback unit;

[0035] The process of the alarm execution result reported by the sound and light alarm device specifically includes:

[0036] The status feedback unit records and reports the response status, execution delay, and abnormal operation of the sound and light alarm device during each alarm process to the alarm control center.

[0037] The second aspect of the present invention discloses a method for remote sound and light alarm and data collection of geological disasters, which is applied to the system disclosed in the first aspect. The method comprises:

[0038] Collect multi-source monitoring data related to geological hazards;

[0039] Evaluate warning levels based on multi-source monitoring data, historical disaster information, and environmental factors, and output corresponding alarm trigger signals;

[0040] When a target monitoring point is detected to be at a medium or high warning level, a linkage control strategy is generated based on the spatial geographical distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices of the geographical area to which the monitoring point belongs;

[0041] Based on the alarm trigger signal and linkage control strategy, execute corresponding sound and light alarm operations;

[0042] Obtain alarm records and alarm execution results, and perform subsequent disaster tracing and analysis operations based on the alarm records and alarm execution results.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] This invention, based on the collection of multi-source geological and environmental data, uses early warning rules and trend models to intelligently determine warning levels. By introducing a regional linkage mechanism, it effectively improves the accuracy of early warnings and the coordinated coordination of alarm responses. Furthermore, the invention dynamically assigns response priorities to acousto-optic devices within a region based on factors such as the risk level of monitoring points, geographic topology, and the distribution of key facilities, enabling multi-point, multi-level, differentiated alarm control. It also utilizes alarm status feedback to enable post-disaster source analysis, significantly improving alarm reliability and emergency response efficiency in complex geological scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of the application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0046] Figure 1 This is a structural diagram of a geological disaster remote sound and light alarm and data acquisition system disclosed in Example 1 of the present invention. DETAILED DESCRIPTION

[0047] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0048] Example 1

[0049] The first aspect of the present invention discloses a geological disaster remote sound and light alarm and data acquisition system, please refer to Figure 1 , Figure 1 This is a structural diagram of a geological disaster remote sound and light alarm and data acquisition system disclosed in an embodiment of the present invention. The system includes a monitoring terminal, an alarm control center, a sound and light alarm device, and a remote platform; wherein,

[0050] The monitoring terminal is used to collect multi-source monitoring data related to geological disasters and transmit it to the alarm control center in real time;

[0051] The alarm control center includes an intelligent determination unit and a regional linkage unit;

[0052] The intelligent judgment unit is used to evaluate the warning level based on multi-source monitoring data, historical disaster information and environmental factors, and output the corresponding alarm trigger signal;

[0053] The regional linkage unit is used to generate a linkage control strategy when it detects that the target monitoring point is at a medium or high warning level, combining the spatial geographical distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices of the geographical area to which the monitoring point belongs;

[0054] The sound and light alarm device is used to receive the alarm trigger signal and linkage control strategy, and perform corresponding sound and light alarm operations;

[0055] The remote platform is used to receive and store alarm records and alarm execution results reported by the alarm control center and the sound and light alarm devices for subsequent disaster tracing and analysis.

[0056] Furthermore, the monitoring terminal includes a collection unit and a processing unit. The collection unit is used to simultaneously collect surface deformation, environmental vibration, image information, and rain intensity data as multi-source monitoring data. The collection unit includes, but is not limited to, strain sensors, tilt sensors, accelerometers, image acquisition devices, and rain gauges. It should be understood that the aforementioned multi-source monitoring data and the specific contents of the collection unit are merely preferred examples of the present invention and are not intended to limit its scope.

[0057] The processing unit is used to perform time alignment, spatial consistency detection and feature fusion operations on multi-source monitoring data to form a multimodal feature vector and transmit it to the alarm control center.

[0058] Furthermore, the processing unit is further configured to sequentially perform abnormal segment identification, data compression, and data hierarchical storage operations on the collected data;

[0059] Among them, abnormal fragments are obtained through normal state reference model identification. After the abnormal data fragments are identified, the abnormal fragments are distinguished from the normal segment data, and the mutation time index and geographic location information are embedded when compressing the abnormal fragment data. According to the abnormal level of the abnormal fragment, the compressed data is stored in a hierarchical manner and transmitted to the remote platform in a hierarchical manner.

[0060] In an embodiment of the present invention, after completing the time alignment and feature fusion of multi-source monitoring data, the processing unit of the monitoring terminal further undertakes the functions of abnormal segment identification, compression encoding and data management of the collected data, with the aim of improving the warning sensitivity and remote transmission efficiency.

[0061] Specifically, when performing the abnormal segment identification operation, the present invention introduces a normal state reference model as a basic control. This reference model is constructed by training historical data continuously collected in the target area under disaster-free conditions, covering typical daily deformation variables, vibration levels, rainfall changes, and image stability features. For example, these historical data are modeled based on a sliding window statistical model to extract the feature distribution and behavior trajectory of each data channel under normal mode. The newly collected data will be compared with the reference model point by point or sliding window. If some of its features deviate from the normal state range or undergo a sudden mutation, it will be identified as an "abnormal segment."

[0062] Identified anomalous segments are segmented along with the normal segments preceding and following them. A structured compression method that preserves key feature location information is used for compression of anomalous segments. The precise mutation time index and corresponding geographic location information (latitude and longitude + monitoring point ID) are embedded in the data packet, ensuring that subsequent remote platforms can accurately locate the time and spatial location of the anomaly during analysis or traceback.

[0063] In order to further optimize data management efficiency and bandwidth utilization, the embodiment of the present invention also performs hierarchical management on abnormal fragments according to their abnormality levels. The abnormality level is determined by a comprehensive score of the characteristic mutation amplitude, duration and joint response of multiple channels of the abnormal fragment. For example, a multi-factor scoring formula can be used to weight and sum factors such as displacement jump value, strain rate surge ratio, vibration peak value, and map them to level labels (such as L1-general abnormality, L2-medium abnormality, L3-severe abnormality). Depending on the abnormality level, differentiated local storage priorities and transmission strategies (such as immediate push or delayed upload) are adopted for the compressed data.

[0064] Finally, by implementing hierarchical storage and transmission of the completed abnormal fragment data to the remote platform according to its level, including immediate upload of severe-level data, batch upload of medium-level data when bandwidth permits, and initial storage of general-level data in the local cache and transmission during periodic synchronization, this ensures the timely upload of key disaster information while reducing the processing load on the remote platform.

[0065] Through the above operations of the processing unit, on the one hand, the response and transmission efficiency to sudden anomalies are improved, ensuring that key changes are promptly fed back to the remote platform for disaster analysis; on the other hand, by compressing and embedding location information and timestamps and executing a hierarchical transmission strategy, the pressure on the communication link is effectively reduced and the "data value first" transmission mode is realized, thereby improving the coordinated response efficiency of the overall intelligent early warning.

[0066] Furthermore, the intelligent decision unit includes an early warning rule engine, a trend modeling subunit and a weight adjustment subunit; wherein,

[0067] The early warning rule engine is used to form a combination of early warning indicators based on historical multi-source monitoring data and environmental factors, set corresponding thresholds for each early warning indicator, and determine multi-dimensional threshold matching rules based on the set thresholds;

[0068] The trend modeling subunit is used to construct a risk evolution trend curve based on the multimodal feature vectors of continuous time periods, and to determine trend labels of rising risk levels, increased volatility, or high-level stagnation based on the risk evolution trend curve;

[0069] The dynamic weight adjustment unit is used to periodically update the weights of each warning indicator in the multidimensional threshold matching rule by combining the historical false alarm rate, sensor signal-to-noise ratio and risk evolution trend output by the trend modeling subunit.

[0070] In the embodiment of the present invention, the intelligent determination unit is mainly responsible for the key tasks of identifying disaster risks from multi-source monitoring data and triggering early warnings.

[0071] Among them, the early warning rule engine is used to build a multi-source early warning indicator system and formulate corresponding threshold judgment rules. The so-called early warning indicator refers to the change characteristics of physical quantities that are sensitive to disaster precursors and extracted from historical multi-source monitoring data and environmental background parameters. In the present invention, early warning indicators include but are not limited to strain rate, surface displacement slope, vibration intensity increment and rainfall change rate. Among them, the strain rate represents the change in strain value per unit time, reflecting the internal force evolution trend of the material; the surface displacement slope represents the local change rate of the surface displacement curve, which is used to identify chronic slip or sudden displacement; the vibration intensity increment is obtained by the acceleration sensor, reflecting the degree of amplitude change during surface micro-vibration or disturbance, and is used to identify abnormal vibration phenomena before the start of the landslide; the rainfall change rate represents the fluctuation value of short-term rainfall intensity.

[0072] Specifically, a unified historical monitoring dataset is constructed by collecting data from multiple sensors, including historical strain, tilt, acceleration, imagery, and rainfall, and then performing time alignment, anomaly removal, and normalization on this data. Furthermore, environmental contextual information, such as topographic parameters, rock and soil types, groundwater distribution, and vegetation coverage, in historical disaster-affected areas is incorporated as spatial auxiliary factors. Time series feature extraction is performed on the monitoring data to analyze the changing trends of specific physical quantities before, during, and after a disaster. For example, surface displacement often accelerates before a landslide, strain values ​​exhibit both slow accumulation and sudden jumps, and rainfall intensity can rise sharply within 24 hours before a disaster. These changing patterns are statistically summarized and sensitivity evaluated, calculating their response delay and stability under different geological environments. This results in the identification of a combination of indicators with high predictive value for disaster occurrence. These indicators have been validated in multiple historical cases to be sensitive to disaster precursors and effectively support multidimensional threshold matching.

[0073] Through this extraction method, the early warning indicators are not only observable and quantifiable, but also integrate historical big data and regional geological environment characteristics, improving the reliability of responses to complex geological disaster scenarios.

[0074] After obtaining the early warning indicators, combined with actual engineering geological conditions (such as slope and other factors), one or more risk-sensitive thresholds (such as single-point thresholds, upper and lower limits, or abnormal windows) are set for each indicator. The rule set consisting of multiple indicators and their thresholds is called a multi-dimensional threshold matching rule. This rule performs multi-dimensional feature matching on the input data at each moment. If several indicators trigger the threshold at the same time, it is identified as a warning signal for a potential risk event.

[0075] Secondly, the trend modeling subunit outputs a risk evolution trend curve by performing time series modeling on the multimodal feature vectors of continuous time periods. During the analysis process, trend labels such as rising risk levels, intensified volatility, and high-level retention are identified and output. Among them, rising risk levels refer to indicators that continue to approach or exceed preset thresholds and show an increasing trend over a period of time; intensified volatility refers to frequent and large jumps in relevant indicators in the time dimension; and high-level retention refers to the fact that although the risk level has not increased significantly, key indicators remain in a high-risk range for a long time, indicating the possibility of "delayed disasters" or "hidden risks." Based on the output results of trend labels, areas that have not completely exceeded the limit but have the potential for risk evolution are identified in advance, thereby improving the foresight of the overall warning.

[0076] Furthermore, the dynamic weight adjustment subunit is responsible for adaptively updating the judgment weights of each warning indicator based on trend label feedback, combined with historical false alarms and the health status of each sensor device in the current monitoring terminal acquisition unit. Specifically, in the historical warning records recorded by the remote platform, alarms that have not actually evolved into disaster events are marked as "false alarms", and the contribution ratio of each type of warning indicator to false alarms is counted. Secondly, the signal stability of each sensor device in the recent period is regularly evaluated, and the weight of the corresponding indicator of high-noise equipment is set to be reduced. If the trend modeling subunit continuously outputs the "risk level increased" or "high-level retention" label, the corresponding warning indicator (such as strain rate, displacement slope) will obtain an enhanced weight. The above-mentioned evaluation result fusion update is performed at the set update period (such as every 24 hours or after each warning event) to adjust the relative weight factor of each indicator in the multi-dimensional threshold matching rule.

[0077] Through the above operations, the scientificity and flexibility of early warning triggering are improved, which not only reduces the false alarm rate, but also increases the sensitivity to slowly evolving disasters (such as landslides and crack expansion), which is significantly better than the traditional single-threshold hard triggering method.

[0078] Furthermore, the process of constructing the risk evolution trend curve based on the multimodal feature vectors of the continuous time period by the trend modeling subunit specifically includes:

[0079] The trend modeling subunit performs time series analysis on the multimodal feature vectors in a continuous time period based on the trend model to generate a risk evolution trend curve;

[0080] Among them, the trend model is constructed by combining residual autoregressive analysis with change point detection algorithm, which is used to identify trend mutation points, periodic risk turning points and risk mitigation cycles.

[0081] In this embodiment of the present invention, the core task of the trend modeling subunit is to construct a risk evolution trend curve, which is used to dynamically track the evolution characteristics of geological hazard warning indicators over time. This is based on modeling multimodal feature vector sequences. By using a sliding window of continuous time periods as the unit, the multimodal feature sequence within that period is input into the trend model, and the risk evolution trend curve is output.

[0082] To ensure that the modeling results can sensitively capture sudden, periodic, and phased changes, the trend model adopted in the embodiment of the present invention combines residual autoregression analysis (R-AR) with change point detection algorithm (CPD) to achieve accurate perception of risk signal details.

[0083] The residual autoregression analysis component primarily establishes a dynamic dependency between the current risk state and its historical state. Specifically, an AR model is constructed for each warning indicator time series and the forecast residual is calculated at each moment. Sudden increases in residual changes are often a precursor to abnormal behavior. Compared to traditional smooth trend lines, the R-AR structure focuses more on nonlinear deviations and is therefore more sensitive to non-stationary changes in disaster precursors.

[0084] On this basis, a change point detection algorithm is introduced to identify mutation nodes, risk growth inflection points, and mitigation signal intervals in trend sequences. Furthermore, a Bayesian Information Criterion (BIC) algorithm is used to monitor structural change points in trend curves in real time, marking them as "change points."

[0085] Through these operations, the trend model not only outputs a complete risk evolution trend curve but also automatically labels key time nodes. Trend mutation points represent nodes where risk values ​​experience a sharp jump or drop within a short period of time; phased risk turning points are key inflection points where risk status shifts from low to high, or from high to stable; and the risk mitigation cycle indicates whether a significant mitigation signal segment appears after a period of high risk. Ultimately, based on the trend curve shape and the distribution of change points, corresponding trend labels are generated, such as "rising risk," "increasing volatility," "staying at a high level," or "easing and stabilizing."

[0086] Furthermore, the intelligent judgment unit also includes a rule judgment subunit, a model judgment subunit and a warning level assessment subunit;

[0087] Among them, the rule judgment subunit is used to determine the current warning indicator value based on the current multimodal feature vector and current environmental factors, and judge the current warning indicator value according to the updated multidimensional threshold matching rule to form a basic risk warning result;

[0088] The model judgment subunit is used to identify and classify the current multimodal feature vector based on the trained disaster identification model and output a disaster risk identification label;

[0089] The warning level assessment subunit is used to perform a weighted fusion operation based on the output results of the rule judgment subunit and the model judgment subunit to generate a comprehensive warning level label, and generate a corresponding alarm trigger signal based on the comprehensive warning level label.

[0090] Specifically, the logical rules of the rule judgment subunit are suitable for rapid response to clear abnormal states, and have certain limitations when dealing with complex scenarios such as fuzzy boundaries and multi-source interactions. In order to make up for the above shortcomings, the present invention introduces a model judgment subunit, the core of which is a disaster identification model. The model is constructed using a multi-channel deep neural network structure based on an attention mechanism, which can perform feature encoding on monitoring data of different channels (such as strain, displacement, vibration, rainfall, etc.) respectively, and focus on short-term mutation features through the attention mechanism, thereby improving the classification and identification capabilities of geological disaster precursor states. In the model training stage, supervised learning is performed using monitoring sample data with historical risk labels. The sample data includes historical multimodal feature vectors, historical environmental factors, and verified disaster event labels. The training goal is to minimize the multi-category cross entropy loss function between the predicted results and the actual labels, and to improve the classification accuracy of the model for states of different risk levels.

[0091] As a preferred implementation scheme of Example 1 of the present invention, the disaster identification model based on deep learning constructed by the model judgment subunit in the intelligent judgment unit includes multi-channel time series input, trend residual modeling branch, intra-channel time attention mechanism, trend modulation gating mechanism, inter-channel self-attention fusion structure and multi-target output module, which is used to realize the joint identification and prediction of geological disaster risk level and trend rate.

[0092] Specifically, suppose the channel set of multi-source monitoring data is X = {X (1) ,X (2) ,…,X (M)},in represents the observation data of the mth sensor channel in the time window, T is the length of the time window, d m is the channel dimension.

[0093] First, the disaster recognition model encodes the input sequence for each channel m through the gated recurrent unit (GRU), and combines it with trend residual modeling to obtain the feature sequence in the channel

[0094]

[0095] in, represents the trend estimate of the mth channel at time t, obtained by exponentially weighted moving average (EMA); β (m) is the residual modulation coefficient, which is used to control the response intensity to short-term mutations; is the observation data of the mth sensing channel at time t.

[0096] Then, the intra-channel temporal attention mechanism is used to assign weights to each time slice:

[0097]

[0098] in, is the temporal attention weight; Z (m) is the temporal feature aggregation result of channel m; is the query vector in the attention mechanism; W h is a linear transformation matrix used to transform the channel features after GRU encoding Projection into attention space; b h is the bias vector.

[0099] In order to adapt to the dynamic changes of risk trends, the model introduces a trend modulation gating mechanism, which uses the risk trend curve to Modulate the channel representation:

[0100]

[0101] z (m)＇ =γ m ·Z (m)

[0102] in, is the trend modulation weight matrix, which is used to transform the risk trend curve R t The learnable parameters mapped to the channel modulation space represent the trend control ability of the mth channel; is the trend modulation bias vector corresponding to channel m; σ is the activation function, preferably the Sigmoid function, which is used to output a gating coefficient γ in the range [0,1] m ;z (m)＇ It is the channel representation after trend modulation.

[0103] On this basis, all modulated channels are represented as z (m)＇ The data is concatenated into a matrix, and global information is fused through a self-attention mechanism between channels. After fusion, the final global disaster feature representation is output. Furthermore, by setting up a dual-task head of classification and regression at the output layer, the disaster level classification result and the trend expansion rate prediction value are output respectively.

[0104] Through the construction process of the above-mentioned disaster identification model, the model can extract key time series characteristics and risk change patterns from complex historical disaster evolution data, have the ability to perform high-confidence identification of current multimodal inputs, and can quickly output disaster risk level labels, significantly improving the intelligent judgment effect and response speed of the alarm control center.

[0105] Furthermore, in this preferred embodiment, the disaster identification model, in addition to outputting a disaster risk level label, also outputs a predicted trend expansion rate value, which is used to assess the potential rate and scope of disaster development. This predicted value, calculated through the model's regression task head, reflects the evolution rate of the disaster situation over a short period of time in the future, serving as an auxiliary decision-making basis for regional coordinated scheduling and sound and light alarm adjustment. In subsequent processing, the predicted trend expansion rate value is used to dynamically adjust the alarm intensity, duration, or coordinated coverage of the sound and light alarm devices, thereby achieving a more refined and forward-looking emergency response mechanism.

[0106] Furthermore, the regional linkage unit includes a regional topology construction subunit and an early warning priority determination subunit;

[0107] The regional topology construction subunit is used to construct a multi-region linkage topology map based on the spatial geographic distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices in the geographical area to which the monitoring points belong. The nodes in the topology map represent geographical sub-regions, and the edges are used to describe the linkage relationship between regions with the possibility of risk propagation.

[0108] The warning priority determination subunit is used to calculate the linkage response priority score of the area when the warning level label of the target monitoring point is medium or high, based on the geological sensitivity, population density, key facility density and current warning level factor of the area where the target monitoring point is located in the topological map;

[0109] The priority scoring result is used to guide the sound and light alarm device to execute differentiated activation sequences and alarm intensity control strategies in multiple adjacent or overlapping impact areas.

[0110] In an embodiment of the present invention, in order to achieve regional coordination and hierarchical response of geological disaster alarm strategies, a regional topology construction subunit and an early warning priority determination subunit are respectively set in the regional linkage unit to build a multi-regional linkage relationship and dynamically formulate response levels and scheduling sequences according to risk influence.

[0111] The task of the regional topology construction sub-unit is to establish a regional linkage topology map that reflects the potential for geological disaster propagation. Specifically, a preliminary set of geographical sub-regions is constructed based on the geographical administrative unit or geomorphological division to which each monitoring point belongs, and the spatial geographical distribution of each region (such as adjacent bordering relations), topographic and geomorphological information (such as ridge connectivity, river valley diffusion paths), historical disaster propagation records (such as debris flow direction, landslide propagation path) and the deployment density of surrounding sound and light alarm devices are combined to calculate the probability of linkage influence between any two regions. When the probability is higher than the set threshold, an edge is added to the topological map, indicating that the two geographical sub-regions corresponding to the edge have a risk propagation linkage relationship. In the final topological map structure, each node represents a sub-region, and the edge represents its potential risk linkage path with other regions.

[0112] The warning priority determination subunit is used to score the response level of each area based on the constructed topology map and the status information of the current target area, and determine the priority of its linkage activation. The scoring model introduces multiple weight factors, including but not limited to:

[0113] The geological sensitivity factor reflects the probability and vulnerability of the area to past disasters;

[0114] Population density factor, reflecting the number of residents that may be affected if a disaster occurs;

[0115] Key facility density factor, which assesses the concentration of important facilities (such as schools, power stations, and hospitals) in the region;

[0116] The current warning level factor is the risk level label currently output by the intelligent judgment unit.

[0117] By weighting and integrating the above factors, a coordinated response priority score is calculated for each region. When warnings are triggered simultaneously in multiple adjacent or overlapping impact areas, the scoring results determine the activation sequence, duration, and intensity settings of the sound and light alarms, ensuring that high-risk areas receive priority response and reducing the possibility of secondary disasters spreading.

[0118] Through the design of the above-mentioned regional linkage units, the present invention not only achieves the spatial coordination of disaster warning, but also improves the resource allocation efficiency and response accuracy of the sound and light alarm strategy through risk sensitivity classification, effectively supporting the intelligent scheduling capability under multi-region concurrent risk conditions.

[0119] Furthermore, the sound and light alarm device includes an environment sensing unit and a response adjustment unit; wherein,

[0120] The environmental perception unit is used to detect the ambient noise level, visual distance and density of surrounding sound and light alarm devices in the alarm area as environmental perception results;

[0121] The response adjustment unit adaptively adjusts the sound intensity level, light flash frequency, and alarm duration parameters based on the environmental perception results.

[0122] Furthermore, the sound and light alarm device further includes a state feedback unit;

[0123] The process of the alarm execution result reported by the sound and light alarm device specifically includes:

[0124] The status feedback unit records and reports the response status, execution delay, and abnormal operation of the sound and light alarm device during each alarm process to the alarm control center.

[0125] Furthermore, in this invention, the remote platform serves as the back-end processing and analysis core of the system, undertaking key tasks such as data aggregation, disaster record archiving, alarm execution auditing, and disaster trend retrospective analysis. The data it receives primarily comes from two sources: reports from the alarm control center and audible and visual alarm devices, and abnormal fragment data identified by monitoring terminals.

[0126] Specifically, the first type of input to the remote platform is alarm records and alarm execution results. Alarm records are reported by the alarm control center after each alarm is triggered. They include the warning level label, trigger time, identification of the monitoring point involved, a summary of the judgment basis, and the parameters of the generated linkage control strategy.

[0127] The alarm execution result includes the status feedback unit from the sound and light alarm device, which includes the response startup status (whether it is successful), execution delay (time from trigger to actual action), whether the sound and light output meets the standards (such as whether the sound intensity and light flash frequency are abnormal) and equipment operation abnormality records (such as power failure, signal loss, etc.).

[0128] The coordinated aggregation of these two parts of information facilitates the remote platform to conduct closed-loop analysis of the integrity and accuracy of the early warning response chain. On the one hand, it can be used for system performance evaluation and improvement, and on the other hand, it also provides a basis for post-event responsibility tracing and emergency response summary.

[0129] The second type of input for the remote platform comes from abnormal fragment data identified by the monitoring terminal processing unit. In the data processing logic running on the monitoring terminal side, once the sensor data is identified as a sudden change segment (such as a sudden and drastic change in strain, a cliff-like jump in displacement, an extreme increase in rainfall, etc.), the segment is compressed and embedded with the mutation time index, geographic location information, and anomaly level label. After receiving these segments, the remote platform compares and analyzes them with the warning level and alarm response records at the time to identify any missed warnings, untimely responses, or false alarms. It also supports the construction of an abnormal behavior database for subsequent model training and empirical rule optimization.

[0130] By receiving and fusing multi-path data from the alarm control center, audible and visual alarm devices, and monitoring terminals, the remote platform builds an analytical hub that supports historical archiving, closed-loop response tracking, and continuous iteration of intelligent models. In particular, the structured access and hierarchical processing of abnormal fragment data significantly enhances the system's ability to identify early signs of minor disasters and trace risk evolution trends.

[0131] Example 2

[0132] A second aspect of the present invention discloses a method for remote sound and light alarm and data collection of geological disasters, the method comprising:

[0133] Collect multi-source monitoring data related to geological hazards;

[0134] Evaluate warning levels based on multi-source monitoring data, historical disaster information, and environmental factors, and output corresponding alarm trigger signals;

[0135] When a target monitoring point is detected to be at a medium or high warning level, a linkage control strategy is generated based on the spatial geographical distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices of the geographical area to which the monitoring point belongs;

[0136] Based on the alarm trigger signal and linkage control strategy, execute corresponding sound and light alarm operations;

[0137] Obtain alarm records and alarm execution results, and perform subsequent disaster tracing and analysis operations based on the alarm records and alarm execution results.

[0138] It should be noted that the specific implementation process of Example 2 is similar to that of Example 1 and will not be repeated in this embodiment.

[0139] Finally, it should be noted that the above-mentioned embodiments include multiple parallel implementation methods of the present invention, and deleting or otherwise adjusting one or more of the implementation methods will not affect the implementation of the scheme. In addition, the geological disaster remote sound and light alarm and data acquisition system and method disclosed in the embodiments of the present invention are only preferred embodiments of the present invention, which are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the above embodiments, it should be understood by those skilled in the art that the technical solutions described in the above embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A geological disaster remote sound and light alarm and data acquisition system, characterized in that: The system includes a monitoring terminal, an alarm control center, an audible and visual alarm device, and a remote platform; wherein, The monitoring terminal is used to collect multi-source monitoring data related to geological disasters and transmit it to the alarm control center in real time; The alarm control center includes an intelligent determination unit and a regional linkage unit; The intelligent judgment unit is used to evaluate the warning level based on multi-source monitoring data, historical disaster information and environmental factors, and output the corresponding alarm trigger signal; The regional linkage unit is used to generate a linkage control strategy when it detects that the target monitoring point is at a medium or high warning level, combining the spatial geographical distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices of the geographical area to which the monitoring point belongs; The sound and light alarm device is used to receive the alarm trigger signal and linkage control strategy, and perform corresponding sound and light alarm operations; The remote platform is used to receive and store alarm records and alarm execution results reported by the alarm control center and the sound and light alarm devices for subsequent disaster tracing and analysis.

2. The geological disaster remote sound and light alarm and data acquisition system according to claim 1 is characterized in that: The monitoring terminal includes a collection unit and a processing unit; The acquisition unit is used to synchronously collect surface deformation, environmental vibration, image information and rain intensity data as multi-source monitoring data; the acquisition unit includes a strain sensor, an inclination sensor, an acceleration sensor, an image acquisition device and a rain gauge; The processing unit is used to perform time alignment, spatial consistency detection and feature fusion operations on multi-source monitoring data to form a multimodal feature vector and transmit it to the alarm control center.

3. The geological disaster remote sound and light alarm and data acquisition system according to claim 2 is characterized in that: The processing unit is further configured to sequentially perform abnormal segment identification, data compression, and data hierarchical storage operations on the collected data; Among them, abnormal fragments are obtained through normal state reference model identification. After the abnormal data fragments are identified, the abnormal fragments are distinguished from the normal segment data, and the mutation time index and geographic location information are embedded when compressing the abnormal fragment data. According to the abnormal level of the abnormal fragment, the compressed data is stored in a hierarchical manner and transmitted to the remote platform in a hierarchical manner.

4. The geological disaster remote sound and light alarm and data acquisition system according to claim 2 is characterized in that: The intelligent decision unit includes an early warning rule engine, a trend modeling subunit and a weight adjustment subunit; wherein, The early warning rule engine is used to form a combination of early warning indicators based on historical multi-source monitoring data and environmental factors, set corresponding thresholds for each early warning indicator, and determine multi-dimensional threshold matching rules based on the set thresholds; The trend modeling subunit is used to construct a risk evolution trend curve based on the multimodal feature vectors of continuous time periods, and to determine trend labels of rising risk levels, increased volatility, or high-level stagnation based on the risk evolution trend curve; The dynamic weight adjustment unit is used to periodically update the weights of each warning indicator in the multidimensional threshold matching rule by combining the historical false alarm rate, sensor signal-to-noise ratio and risk evolution trend output by the trend modeling subunit.

5. The geological disaster remote sound and light alarm and data acquisition system according to claim 4 is characterized in that: The intelligent judgment unit also includes a rule judgment subunit, a model judgment subunit and a warning level assessment subunit; Among them, the rule judgment subunit is used to determine the current warning indicator value based on the current multimodal feature vector and current environmental factors, and judge the current warning indicator value according to the updated multidimensional threshold matching rule to form a basic risk warning result; The model judgment subunit is used to identify and classify the current multimodal feature vector based on the trained disaster identification model and output a disaster risk identification label; The warning level assessment subunit is used to perform a weighted fusion operation based on the output results of the rule judgment subunit and the model judgment subunit to generate a comprehensive warning level label, and generate a corresponding alarm trigger signal based on the comprehensive warning level label.

6. The geological disaster remote sound and light alarm and data acquisition system according to claim 5 is characterized in that: The process of the trend modeling subunit constructing the risk evolution trend curve based on the multimodal feature vectors of the continuous time period specifically includes: The trend modeling subunit performs a time series analysis operation on the multimodal feature vectors within a continuous time period based on the trend model to generate a risk evolution trend curve; The trend model is constructed by combining residual autoregressive analysis with a change point detection algorithm, and is used to identify trend mutation points, periodic risk turning points, and risk mitigation cycles.

7. The geological disaster remote sound and light alarm and data acquisition system according to claim 1 is characterized in that: The regional linkage unit includes a regional topology construction subunit and an early warning priority determination subunit; The regional topology construction subunit is used to construct a multi-region linkage topology map based on the spatial geographic distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices in the geographical area to which the monitoring points belong. The nodes in the topology map represent geographical sub-regions, and the edges are used to describe the linkage relationship between regions with the possibility of risk propagation. The warning priority determination subunit is used to calculate the linkage response priority score of the area when the warning level label of the target monitoring point is medium or high, based on the geological sensitivity, population density, key facility density and current warning level factor of the area where the target monitoring point is located in the topological map; The priority scoring result is used to guide the sound and light alarm device to execute differentiated activation sequences and alarm intensity control strategies in multiple adjacent or overlapping impact areas.

8. The geological disaster remote sound and light alarm and data acquisition system according to claim 1 is characterized in that: The sound and light alarm device includes an environment sensing unit and a response adjustment unit; wherein, The environmental perception unit is used to detect the ambient noise level, visual distance and density of surrounding sound and light alarm devices in the alarm area as environmental perception results; The response adjustment unit adaptively adjusts the sound intensity level, light flash frequency, and alarm duration parameters based on the environmental perception results.

9. The geological disaster remote sound and light alarm and data acquisition system according to claim 8, characterized in that: The sound and light alarm device also includes a state feedback unit; The process of the alarm execution result reported by the sound and light alarm device specifically includes: The status feedback unit records and reports the response status, execution delay, and abnormal operation of the sound and light alarm device during each alarm process to the alarm control center.

10. A method for remote sound and light alarm and data collection for geological disasters, said method being applied to the system according to any one of claims 1 to 9, characterized in that: The method comprises: Collect multi-source monitoring data related to geological hazards; Evaluate warning levels based on multi-source monitoring data, historical disaster information, and environmental factors, and output corresponding alarm trigger signals; When a target monitoring point is detected to be at a medium or high warning level, a linkage control strategy is generated based on the spatial geographical distribution, topographic information, historical disaster propagation records, and the deployment of surrounding sound and light alarm devices of the geographical area to which the monitoring point belongs; Based on the alarm trigger signal and linkage control strategy, execute corresponding sound and light alarm operations; Obtain alarm records and alarm execution results, and perform subsequent disaster tracing and analysis operations based on the alarm records and alarm execution results.

Images