CN-122024444-A - Collapse early warning model and visual early warning platform based on artificial intelligence

CN122024444ACN 122024444 ACN122024444 ACN 122024444ACN-122024444-A

Abstract

The invention discloses a collapse early warning model and a visual early warning platform based on artificial intelligence, and belongs to the technical field of geological disaster monitoring and early warning. According to the method, an LSTM early warning model is customized according to loess collapsible geological characteristics, the mechanism law of collapse under the geological environment is deeply attached, dynamic characteristics of changes of monitoring indexes along with time can be accurately captured, the attention mechanism is combined with the depth of the loess collapsible characteristics, so that the model can better cope with extreme data scenes, the problem of characteristic distortion caused by the extreme data is effectively avoided, the influence of abnormal fluctuation of the data on a prediction result is reduced, the model can keep stable prediction performance in a complex and changeable geological monitoring environment, collapse risk early warning requirements under different working conditions are met, risk classification is clear, multi-channel early warning information is automatically pushed by medium and high risks, prevention and control suggestions are attached, and the visualization platform intuitively displays data, weight distribution and prediction results in various modes, so that workers can accurately control risks.

Inventors

TONG JINGSHENG
ZHANG WEIQIANG
Zhang Dixia
ZHAO GUORUI
TIAN YU
ZHANG DAI
LI DONG
ZHAO LIPING

Assignees

中国市政工程西北设计研究院有限公司

Dates

Publication Date: 20260512
Application Date: 20260206

Claims (10)

1. An artificial intelligence based collapse early warning model, comprising: The data input module is used for receiving the multi-dimensional monitoring data of the collapse-prone area, preprocessing the obtained multi-dimensional monitoring data, constructing an effective feature set highly related to collapse, and generating standardized input data; The LSTM prediction module is used for acquiring standardized input data, extracting time sequence characteristics of the standardized input data based on the long and short memory network, predicting the change trend of key indexes in a period of time in the future of the monitoring area based on the time sequence characteristics, and outputting a preliminary time sequence prediction result and extracted time sequence characteristics; The attention mechanism optimizing module is used for carrying out weight distribution and dynamic adjustment on the extracted time sequence features, carrying out attention strengthening on key time sequence features related to collapse occurrence and carrying out interference suppression on irrelevant time sequence features; The risk output module is used for carrying out fusion calculation on the weight distribution and the time sequence characteristics after dynamic adjustment, combining a preset risk assessment index system, generating and outputting collapse occurrence probability, risk grade and trend prediction information, simultaneously calculating the confidence coefficient of the model reasoning result in real time, and carrying out rechecking on the prediction result based on the confidence coefficient calculation result.
2. The artificial intelligence based collapse early warning model of claim 1, wherein the multi-dimensional monitoring data comprises geological environment data, deformation monitoring data, environmental disturbance data and basic data: The geological environment data comprise stratum lithology, loess moisture content, pore water pressure, groundwater level burial depth, physical and mechanical parameters of a rock-soil body, dry density and collapse coefficient; the deformation monitoring parameters comprise earth surface settlement, soil horizontal displacement and settlement rate, underground structure displacement and foundation pit peripheral soil strain; the environmental disturbance data comprise peripheral engineering construction load, underground water pumping and drainage intensity, rainfall/snow melting amount, earthquake microseism data and air temperature; the base data includes the geographical coordinates of the monitored area, topography and surrounding building distribution.
3. The artificial intelligence based collapse early warning model of claim 1, wherein the attention mechanism optimization module comprises: the feature activation unit is used for activating the time sequence features, enhancing the characterization capability of the key time sequence features, inhibiting the invalid output of the redundant features and screening out feature components with obvious distinction; The weight calculation unit is used for initializing basic weights of key time sequence features and irrelevant time sequence features based on loess collapse characteristics and collapse occurrence mechanisms, carrying out dynamic weight assignment on feature components by combining historical collapse case marking data, distributing high weights to the key time sequence features before collapse occurs, and distributing low weights to the irrelevant time sequence features; And the feature fusion unit is used for receiving the weighted time sequence features, carrying out normalization fusion processing on each weighted time sequence feature, filtering irrelevant features with the weight ratio lower than a threshold value, and outputting the optimized target time sequence features.
4. The collapse early warning model based on artificial intelligence according to claim 1 is characterized in that the collapse early warning model adopts a cloud end and edge end dual deployment mode, multi-dimensional monitoring data and preprocessed standardized input data transmitted by the data input module are obtained, the deployment mode is adapted in a division mode based on data processing scale and prediction requirements, the cloud end deployment is used for obtaining a large amount of multi-dimensional monitoring data and standardized input data, integrating and processing a large amount of data, and conducting long-term collapse risk prediction, and the edge end deployment is used for obtaining on-site real-time monitoring data, conducting real-time data rapid analysis and preprocessing and conducting short-term collapse risk prediction.
5. The visual early warning platform based on the artificial intelligence collapse early warning model is applied to the collapse early warning model based on the artificial intelligence as claimed in claim 1, and is characterized by comprising the following components: The data acquisition module is used for acquiring field monitoring data and basic data of a target monitoring area in real time, and carrying out standardized processing of an adaptation model processing format on the acquired field monitoring data to generate standardized model input data; the model prediction module is used for inputting standardized model input data into the collapse early warning model, and carrying out risk prediction based on the collapse early warning model to obtain collapse occurrence probability, risk level, trend prediction information and prediction confidence value; The visual display module is used for displaying loess collapse related parameters, model input data, risk prediction results and historical data comparison in real time in the form of map labeling, curve charts and numerical panels, and dynamically simulating and displaying a target monitoring area according to the risk prediction results; and the early warning pushing module is used for pushing early warning information to a preset intelligent terminal when the collapse risk level output by the collapse early warning model is equal to or higher than the middle risk, wherein the early warning information comprises risk level, prediction probability, an influence area and prevention and control suggestions.
6. The visual early warning platform based on the artificial intelligence collapse early warning model according to claim 5, wherein the data acquisition module comprises: combining the preset monitoring index requirement of the collapse early warning model, determining target monitoring data for collapse risk monitoring, and simultaneously distributing unique data identification for each type of target monitoring data; Inputting the data identification of the target monitoring data into a preset monitoring database for matching, and outputting a basic data type corresponding to the target monitoring data based on a matching result; Matching the target monitoring data with the basic data types one by one, classifying the target monitoring data based on the matching result, and generating a first target monitoring data segment and a second target monitoring data segment based on the classification result; Based on the subdivision category of the basic data type, carrying out second classification on the data in the second target monitoring data segment to generate sub-monitoring data sets corresponding to geological environment, deformation monitoring and environment disturbance, wherein each sub-monitoring data set corresponds to a unique monitoring dimension identification; Randomly extracting data in a plurality of first target monitoring data segments to serve as first data samples, and randomly extracting data in each sub monitoring data set to serve as second data samples respectively; Reading and analyzing the first data sample, determining abnormal geological data feature points of the monitoring data, constructing a geological data abnormal correlation architecture according to the abnormal geological data feature points, and deploying the first target monitoring data in the geological data abnormal correlation architecture according to the abnormal type and the geological partition; Based on the deployment characteristics of the first target monitoring data in the geological data anomaly association architecture, determining an anomaly grade and an association influence range corresponding to the anomaly data, and acquiring the data anomaly risk tendency of the target monitoring area; Based on the basic data type and the historical normal monitoring data, a geological data normal association framework is constructed, a second data sample is input into the geological data normal association framework to simulate geological data rules, and the normal fluctuation rules and the safety threshold ranges of the monitoring indexes under different geological partitions of the target monitoring area are determined; Acquiring a first weight corresponding to the abnormal risk tendency of the data and a second weight corresponding to the normal fluctuation rule, scoring the abnormal risk tendency of the data based on the first weight to determine a first score, and scoring the normal fluctuation rule based on the second weight to determine a second score; calculating a comprehensive data abnormal risk score of the target monitoring area by combining the first score and the second score, comparing the comprehensive data abnormal risk score with a preset risk threshold, and judging whether the target monitoring area has data abnormal risk; when the comprehensive data abnormal risk score is equal to or greater than a preset risk threshold, judging that the target monitoring area has data abnormal risk, and triggering a data alarm linkage mechanism; When the abnormal risk score of the comprehensive data is smaller than a preset risk threshold value, judging that the target monitoring area is in a safe state, and continuously collecting the field detection data in real time; When judging that the target monitoring area has abnormal data risk, the data acquisition module automatically generates an abnormal data report and acquires corresponding monitoring point numbers, monitoring time and abnormal index details; Based on the monitoring point number, the responsible person corresponding to the monitoring point is matched, and the abnormal data report is synchronously pushed to the terminal where the responsible person is located.
7. The visual early warning platform based on the artificial intelligence collapse early warning model according to claim 6, wherein basic data types related to collapse early warning are preset in the preset monitoring database, and include geological environment basic data types, deformation monitoring basic data types, environment disturbance basic data types and safety early warning related data types.
8. The visual early warning platform based on the artificial intelligence collapse early warning model according to claim 6, wherein the data acquisition module is further used for synchronously storing abnormal data reports, comprehensive data abnormal risk scores and abnormal data characteristic analysis results into a cloud database, and performing association comparison with prediction results of the collapse early warning model to continuously optimize construction accuracy of a geological data abnormal association framework and a geological data normal association framework.
9. The visual early warning platform based on the artificial intelligent collapse early warning model according to claim 5, wherein the visual display module further comprises a weight distribution condition of the attention mechanism to the loess collapse related key monitoring indexes, and simultaneously, prediction precision difference comparison parameters of the collapse early warning model before and after introduction of the attention mechanism are synchronously presented.
10. The visual early warning platform based on the artificial intelligence collapse early warning model according to claim 5, wherein the early warning pushing mode of the early warning pushing module comprises one or more of short messages, APP pushing, webpage popup and audible and visual warning.

Description

Collapse early warning model and visual early warning platform based on artificial intelligence Technical Field The invention relates to the technical field of geological disaster monitoring and early warning, in particular to an artificial intelligence-based collapse early warning model and a visual early warning platform. Background The traditional collapse early warning means depend on manual monitoring or single index analysis, and have the problems of insufficient data dimension, inaccurate time sequence characteristic capture and the like, so that early warning is delayed, and the misjudgment rate is high. For example, china patent application with publication number of CN119578652A discloses an Internet of things method and system for monitoring and early warning collapsible loess foundation collapse, and the method comprises a loess foundation partition module and a loess foundation monitoring and early warning module. According to the method, the minimum monitoring unit is built, the precipitation data, the ground surface data and the precipitation prediction evaluation model are combined, the CNN and LSTM models are utilized to carry out space-time depth fusion on the precipitation data, accuracy of precipitation prediction is improved, accurate monitoring and early warning on collapsible loess foundation areas are achieved, the monitoring areas can be divided into a plurality of early warning areas by combining the precipitation conditions and the ground surface data through a deep clustering technology, accuracy and response speed of early warning are improved, disasters caused by collapsible loess foundation collapse are effectively prevented, and fine management on the loess foundation areas is achieved. However, although the above patent realizes the regional monitoring of collapsible loess foundation areas and the risk early warning related to precipitation, the attention mechanism is only applied to the feature matching link of precipitation risk prediction, customized weight optimization is not performed aiming at the unique geological characteristics of loess collapse, key time sequence features before collapse occur cannot be subjected to pertinence strengthening, irrelevant feature interference is easy, the confidence rechecking mechanism of model reasoning results is lacking, and the problem of false early warning and missing early warning caused by insufficient extreme data or feature characterization is difficult to avoid. Disclosure of Invention The invention aims to provide an artificial intelligence based collapse early warning model and a visual early warning platform, which are used for fusing an LSTM network and an attention mechanism, accurately extracting collapse related time sequence characteristics, improving the prediction precision by more than 20%, effectively avoiding false early warning and missing early warning, and adapting to extreme data scenes so as to solve the problems in the background technology. In order to achieve the above purpose, the present invention provides the following technical solutions: an artificial intelligence based collapse early warning model comprising: The data input module is used for receiving the multi-dimensional monitoring data of the collapse-prone area, preprocessing the obtained multi-dimensional monitoring data, constructing an effective feature set highly related to collapse, and generating standardized input data; The LSTM prediction module is used for acquiring standardized input data, extracting time sequence characteristics of the standardized input data based on the long and short memory network, predicting the change trend of key indexes in a period of time in the future of the monitoring area based on the time sequence characteristics, and outputting a preliminary time sequence prediction result and extracted time sequence characteristics; The attention mechanism optimizing module is used for carrying out weight distribution and dynamic adjustment on the extracted time sequence features, carrying out attention strengthening on key time sequence features related to collapse occurrence and carrying out interference suppression on irrelevant time sequence features; The risk output module is used for carrying out fusion calculation on the weight distribution and the time sequence characteristics after dynamic adjustment, combining a preset risk assessment index system, generating and outputting collapse occurrence probability, risk grade and trend prediction information, simultaneously calculating the confidence coefficient of the model reasoning result in real time, and carrying out rechecking on the prediction result based on the confidence coefficient calculation result. Further, the multi-dimensional monitoring data comprises geological environment data, deformation monitoring data, environment disturbance data and basic data: The geological environment data comprise stratum lithology, loess moisture content, pore water pr