Intelligent monitoring system and method of geological disasters based on multiple sensors

Abstract

The present invention relates to the field of intelligent monitoring technology for geological disasters, specifically to an intelligent monitoring system and method for geological disasters based on multiple sensors; the method steps are: using multiple sensors to cover the monitoring area, collecting and summarizing environmental data; using a spatiotemporal adaptive decomposition method to separate effective signals and noise, combining a dynamic threshold of the signal-to-noise ratio and a local weighted regression smoothing trend residual to complete data denoising, and then aligning time and dimensions through linear interpolation and standardization, and integrating multi-sensor data based on an adaptive weighted fusion method; generating and reviewing the fused data through pending mark generation to ensure data integrity; extracting key features, constructing a hybrid model based on LSTM and random forest to assess disaster risks, and setting risk level division rules; finally outputting fused data and analysis results to generate early warning signals and decision-making recommendations for potential geological disasters. The present invention realizes accurate monitoring and risk assessment of geological environmental changes.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring of geological disasters, and in particular to a multi-sensor based intelligent monitoring system and method for geological disasters. Background Art

[0002] With global climate change and the intensification of human activities, the frequency and severity of geological disasters have shown a significant upward trend, posing a major threat to human life, property and the ecological environment. Traditional geological disaster monitoring methods rely on single sensors or manual observations, and have problems such as limited monitoring range, poor real-time performance and insufficient data accuracy, making it difficult to meet the needs of geological disaster prediction and assessment in complex environments.

[0003] A Chinese invention patent with announcement number CN114399210B discloses a geological disaster emergency management and decision-making system, method and readable storage medium. The geological disaster emergency management and decision-making system includes: a pre-disaster prevention system for preventing and monitoring geological disasters; an emergency rescue command and dispatch system for generating emergency decisions; a post-processing system for summarizing the response to sudden geological disasters; the pre-disaster prevention system includes a geological disaster hazard monitoring and early warning subsystem, a geological disaster inspection subsystem, a plan library dynamic management subsystem and a publicity and education training subsystem; the emergency rescue command and dispatch system includes a disaster report subsystem, an emergency decision-making subsystem and an emergency command and dispatch subsystem; the post-processing system includes a summary and evaluation subsystem and a post-disaster reconstruction subsystem; the present invention realizes intelligent decision support based on emergency plans, effectively improving the scientific and technological level of emergency decision support for sudden geological disasters.

[0004] However, existing traditional geological disaster monitoring methods are insufficient in terms of real-time, accuracy and data processing capabilities, especially the difficulty in efficiently executing monitoring tasks in large-scale and complex environments; in existing technologies, geological disaster warnings rely on a single data source or traditional data analysis methods, resulting in delayed monitoring information and difficulty in accurately predicting the time, location and severity of disasters; in addition, a single sensor or monitoring equipment is easily affected by external interference when facing changing environmental conditions, resulting in instability in monitoring results and increased errors; and the existing system also lacks effective data processing and intelligent analysis mechanisms, and is unable to respond to the needs of handling sudden disaster events in a timely manner. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and to propose a multi-sensor based intelligent monitoring system and method for geological disasters.

[0006] The technical solution of the present invention is a multi-sensor based intelligent monitoring method for geological disasters, comprising the following specific implementation steps:

[0007] S1. Several types of sensors deployed in the target area collect environmental data according to a preset sampling frequency and summarize the environmental data sent by the sensors;

[0008] S2. Based on the time-space adaptive decomposition and reconstruction method, the sensor data is decomposed into multiple components through wavelet transform and empirical mode decomposition, and the effective signal is screened by spatial correlation and signal-to-noise ratio. The low-frequency residual is locally weighted regression smoothed, and the data of different time and dimension are aligned through linear interpolation and normalization. Then, the adaptive weighted fusion method is used to combine the sensor accuracy, data quality and historical performance to fuse the data and generate fused data. Then, a pending mark is generated for the fused data.

[0009] S3. Check whether the data is missing, use first-order and second-order auxiliary review parameters to confirm, extract key features through convolutional neural network and use principal component analysis to reduce the dimension of data, then build a risk assessment model combining LSTM and random forest to capture time series characteristics and conduct disaster risk assessment, and output disaster risk analysis results, which include disaster risk probability and risk level;

[0010] S4. Automatically activate emergency response strategies based on disaster risk analysis results: If the disaster risk probability exceeds the advanced warning threshold, initiate a comprehensive emergency response; if it is in the medium risk range, issue a warning and strengthen monitoring; if it is below the threshold, recommend regular monitoring; then use genetic algorithm optimization to schedule resources, with the goal of minimizing response time and prioritizing important resources; after the disaster, evaluate the emergency response effect, analyze the gap between actual and expected losses, optimize the disaster prediction model and response strategy, and output resource allocation results and post-disaster assessment results;

[0011] S5. Real-time display of fused data, disaster risk analysis results, resource allocation results and post-disaster assessment results of the monitoring area through a visual interface.

[0012] Preferably, the implementation process based on the spatiotemporal adaptive decomposition and reconstruction method is as follows:

[0013] S21. The input data is a time series signal x(t)=s(t)+n(t). Combining wavelet transform and empirical mode decomposition, x(t) is decomposed into multiple components:

[0014] ;

[0015] In the formula, c i (t) represents the component signal, that is, the components of different frequencies and scales obtained after time-space decomposition; r(t) represents the residual, that is, the trend part that cannot be decomposed; k represents the number of components obtained by the decomposition method, that is, the number of sub-signals into which the signal is decomposed;

[0016] S22. Introduce spatial correlation analysis and filter out abnormal components through mutual correlation coefficients:

[0017] ;

[0018] In the formula, x i (t), x j (t) represents the data of the i-th and j-th sensors; ρ ij Represents the spatial correlation coefficient, which is used to determine the abnormality of the component; and denote the mean values ​​of the i-th and j-th sensor signals respectively;

[0019] S23. Calculate the signal-to-noise ratio (SNR) of each component according to the decomposed components. i :

[0020] ;

[0021] In the formula, s i (t) represents the effective signal of the i-th component; n i (t) represents the noise of the i-th component;

[0022] Based on this, a dynamic threshold is set based on the signal-to-noise ratio: ;

[0023] In the formula, α represents the adjustment factor; δ i represents the standard deviation of the i-th component, that is, the volatility of the component;

[0024] S24, for each component c i (t), the following rules apply:

[0025] If |c i (t)|≥T i ,but =c i (t);

[0026] If |c i (t)| <T i ,but =0;

[0027] in, Represents the denoised component;

[0028] S25, after filtering, the effective signals of each component are retained, but there is still nonlinear trend noise in the low-frequency residual r(t). The residual is smoothed by local weighted regression to obtain ;

[0029] S26, reconstruct all denoised components and trend residuals to obtain the final denoised signal :

[0030] ;

[0031] The denoised data .

[0032] Preferably, the fusion process of fusion data is as follows:

[0033] S31, setting the adaptive weight of the sensor data, the weight calculation formula is:

[0034] ;

[0035] In the formula, w i Represents the weight of sensor i, which determines its contribution to the final fusion result; α i represents the accuracy coefficient of sensor i, that is, the stability and measurement error of the sensor; β i represents the data quality coefficient of sensor i, reflecting the credibility of the current data of the sensor; γ i represents the historical performance coefficient of sensor i;

[0036] S32. Use the weighted average method to fuse the data of each sensor. The fusion formula is as follows:

[0037] ;

[0038] In the formula, x fused represents the fused sensor data, that is, the optimal prediction value obtained after multi-sensor fusion; x i represents the preprocessed data of sensor i; n represents the total number of sensors;

[0039] S33: Update the adaptive weight of the sensor data and output the fused data X.

[0040] Preferably, the generation process of the pending review mark is as follows:

[0041] S41, select the marker generating code Cg∈[1,q-1], and calculate the marker parsing code Ca=[Cg]×G;

[0042] Where G is the elliptic curve E:y 2 =(x 3 +ax+b) mod p on the q-order base point, E is the finite field F defined above p The elliptic curve on E, a, b, p and q represent the parameters of the elliptic curve E;

[0043] S42, select a random number r∈[1,q-1], and calculate the first-order auxiliary identification parameter Pl 1 =[r]G;

[0044] S43, calculate the short information I of the fused data X = H (Pl 1 ,BX);

[0045] Wherein, H represents a hash function; BX represents the binary string form of the fused data X;

[0046] S44. Calculate the second-order auxiliary identification parameter Pl 2 =(I×Cg+r) mod q, and satisfies Pl 2 ≠0, otherwise return to step S42;

[0047] S45, generate a pending mark L={Pl 1 , Pl 2}.

[0048] Preferably, the review process for checking whether the data is missing is as follows:

[0049] S51, inspection Pl 2 ∈[1,q-1] whether:

[0050] If it is established, then calculate the first-order auxiliary review parameter Pa 1 =[Pl 2 ]×G;

[0051] If it is not true, it means that the fused data X is missing;

[0052] Among them, Pl 2 Represents the second-order auxiliary identification parameter; G is the elliptic curve E:y 2 =(x 3 +ax+b) mod p on the q-order base point, E is the finite field F defined above p The elliptic curve on E, a, b, p and q represent the parameters of the elliptic curve E;

[0053] S52. Calculate the second-order auxiliary review parameter Pa 2 =[H(Pl 1 ,BX)]×Ca+Pl 1 ;

[0054] Wherein, Ca represents the tag parsing code; H represents the hash function; BX represents the binary string form of the fused data X; Pl 1 represents the first-order auxiliary identification parameter;

[0055] S53, if Pa 1 =Pa 2 , it means that the fused data X is complete; otherwise, it means that the fused data X is missing.

[0056] Preferably, the risk assessment model is constructed as follows:

[0057] S61. Use the LSTM network to learn historical data of time series features and predict future trends. The basic formula of the LSTM model is:

[0058] ;

[0059] In the formula, h t represents the hidden state of LSTM, that is, the network output at time step t; x t Indicates the input data at the current moment; h t-1 Represents the hidden state of the previous moment, that is, the previous network memory; W h , U h represents the weight matrix, that is, the influence of the input data and the previous hidden state on the current hidden state; b h Represents the bias term, which adjusts the offset of the hidden state;

[0060] S62. After the LSTM neural network models the temporal characteristics of the data, the random forest model is introduced to improve the stability of the model and its ability to classify complex data;

[0061] The random forest model trains multiple decision trees by randomly selecting sample subsets and feature subsets, and then determines the final prediction result through a voting mechanism:

[0062] ;

[0063] In the formula, , , …, Represents the prediction output of multiple decision trees; Represents the final prediction result of random forest;

[0064] S63, integrating the LSTM neural network with the random forest model, firstly, performing time series modeling on the multi-sensor data through the LSTM network to obtain a feature vector containing time series information; then, the feature vector is passed as input to the random forest model to perform classification prediction of disaster risks;

[0065] The formula of the fused model is:

[0066] ;

[0067] In the formula, f LSTM (X) represents the output predicted by the LSTM model for the input data X, i.e., the features extracted from the time series data; f RF () represents the random forest model, which predicts the features output by LSTM and obtains the final disaster risk probability .

[0068] Preferably, the optimization process of resource scheduling using genetic algorithm optimization is: optimizing resource allocation according to disaster type, severity and response time requirements;

[0069] The objective function of the genetic algorithm is:

[0070] ;

[0071] In the formula, represents the weight of the i-th resource, that is, the importance of the resource; t i represents the response time assigned to the i-th resource, that is, the urgency of the resource response; m represents the total number of resources;

[0072] Based on this: The optimization goal of the objective function is to minimize the emergency response time while giving priority to the scheduling of important resources to ensure that the losses can be minimized when a disaster occurs.

[0073] Preferably, the updating process of the adaptive weights of the sensor data is as follows:

[0074] S81. Introduce error estimation. Error estimation is based on the difference between predicted and actual data, and calculates the error range of fused data:

[0075] ;

[0076] In the formula, ε i represents the error value of sensor i, that is, the difference between the predicted value and the actual measured value of the sensor data; x pred represents the sensor data of fusion prediction; x meas Indicates the value actually measured by the sensor;

[0077] S82. When it is detected that the data error of a certain sensor is greater than the set threshold, the weight of the sensor is reduced. The weight calculation formula after adjustment is:

[0078] ;

[0079] In the formula, λ' represents the adjustment factor, which is used to control the influence of the error on the weight adjustment; ε i Represents the error value of sensor i; α i represents the accuracy coefficient of sensor i, that is, the stability and measurement error of the sensor; β i represents the data quality coefficient of sensor i, reflecting the credibility of the current data of the sensor; γ i represents the historical performance coefficient of sensor i.

[0080] The technical solution of the present invention is a multi-sensor based intelligent monitoring system for geological disasters, which is used to execute the above-mentioned multi-sensor based intelligent monitoring method for geological disasters, including:

[0081] Multi-sensor data acquisition module, which integrates several sensors and is used to collect geological disaster environment data in real time;

[0082] A data processing module is used to pre-process the collected geological disaster environment data;

[0083] Data analysis module, used to perform pattern recognition, anomaly detection and trend analysis on sensor data and output analysis results;

[0084] Intelligent early warning and decision-making module, which is used to issue early warning signals according to the analysis results, and provide decision-making support after the disaster occurs by combining risk assessment results and disaster response strategies;

[0085] The human-computer interaction and feedback module is used to provide a visual interface, display real-time monitoring data, early warning information and decision support, and continuously optimize the monitoring strategy based on user feedback.

[0086] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0087] The present invention designs a multi-sensor based intelligent geological disaster monitoring system and method, which shows significant advantages in data collection, processing, analysis and risk assessment, and significantly improves the accuracy, reliability and intelligence level of geological disaster monitoring:

[0088] (1) Comprehensive collection and processing of multi-dimensional environmental data: By deploying various types of sensors such as seismic sensors, meteorological sensors, displacement sensors and gas sensors to cover the target monitoring area, comprehensive environmental information related to geological disasters can be collected, thereby improving the comprehensiveness of monitoring and the reliability of the data base;

[0089] (2) Enhance data quality and information credibility: We introduced a variety of data preprocessing techniques such as data cleaning, noise reduction, standardization, and time alignment, combined with the spatiotemporal adaptive decomposition and reconstruction (STADR) method, to effectively remove noise and outliers, improve the quality and credibility of sensor data, and lay a solid foundation for subsequent analysis;

[0090] (3) Realize fusion analysis of multi-source heterogeneous data: Adopt an adaptive weighted fusion algorithm to comprehensively consider the accuracy, data quality and historical performance of sensors, dynamically adjust weights, and generate high-quality fused data. At the same time, the error estimation unit optimizes the sensor weight settings in real time to ensure the accuracy and stability of the fused data.

[0091] (4) Improving the accuracy and reliability of disaster risk identification: Extracting key features based on convolutional neural networks (CNNs) and reducing the dimension of high-dimensional data through principal component analysis (PCA) effectively simplifies the data structure and highlights the core information; combining the time series modeling capabilities of long short-term memory (LSTM) networks and the classification capabilities of random forests to build a hybrid model to accurately assess the risk level and potential impact of geological disasters;

[0092] (5) Ensure data integrity and transmission security: Through the hash tag generation mechanism, the integrity and security of the fused data during transmission and storage are guaranteed, eliminating the possibility of data loss or tampering, and further improving the reliability of the system;

[0093] (6) Strong real-time and decision-making support capabilities: Real-time aggregation and analysis of multi-sensor data, combined with intelligent early warning and decision-making modules, can provide accurate early warnings before disaster risks occur. At the same time, the timeliness and accuracy of information transmission are improved through human-computer interaction and feedback modules, providing strong technical support for geological disaster prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] Figure 1 This is a system architecture diagram of a multi-sensor based intelligent monitoring system for geological disasters proposed by the present invention;

[0095] Figure 2 This is a method flow chart of a multi-sensor based intelligent monitoring method for geological disasters proposed by the present invention. DETAILED DESCRIPTION

[0096] Embodiment 1, as Figure 1 As shown, the present invention proposes a multi-sensor based intelligent monitoring system for geological disasters, which includes: a multi-sensor data acquisition module, a data processing module, a data analysis module, an intelligent early warning and decision-making module, and a human-computer interaction and feedback module.

[0097] The multi-sensor data acquisition module integrates several sensors, including but not limited to seismic sensors, meteorological sensors (including but not limited to temperature and humidity, air pressure, precipitation), soil moisture sensors, displacement sensors, gas sensors (including but not limited to methane and carbon dioxide concentrations), to collect geological disaster environmental data in real time;

[0098] The data processing module pre-processes the collected geological disaster environment data;

[0099] The data analysis module performs pattern recognition, anomaly detection and trend analysis on sensor data and outputs the analysis results;

[0100] The intelligent early warning and decision-making module issues early warning signals based on the analysis results, and provides decision-making support after the disaster occurs, including but not limited to emergency dispatch, resource allocation and personnel evacuation suggestions, in combination with risk assessment results and disaster response strategies;

[0101] The human-computer interaction and feedback module provides a visual interface to display real-time monitoring data, warning information and decision support, supports multi-level user permissions and emergency response operations, and continuously optimizes monitoring strategies based on user feedback to enhance warning accuracy.

[0102] Embodiment 2, as Figure 2 As shown, the multi-sensor based intelligent monitoring method for geological disasters proposed in the present invention is applied to the multi-sensor based intelligent monitoring system for geological disasters proposed in Example 1, and its specific implementation steps are as follows:

[0103] S1. Deploy various types of sensors, including but not limited to seismic sensors, meteorological sensors (including but not limited to soil moisture, temperature, air pressure, precipitation), displacement sensors and gas sensors (including but not limited to methane and carbon dioxide concentrations), covering the target monitoring area;

[0104] Each sensor collects environmental data according to a preset sampling frequency and transmits the environmental data to the multi-sensor data acquisition module in real time;

[0105] The multi-sensor data acquisition module aggregates the data sent by the sensors through wireless communication technology (including but not limited to LoRa, 5G or Wi-Fi) 1 ,x 2 ,…,x i ,…,x n};

[0106] Among them, x i Represents the environmental data collected by the i-th sensor.

[0107] S2. Perform preliminary cleaning of the collected sensor data, remove null values, duplicate values ​​and obvious abnormal values ​​to improve the data quality, apply denoising algorithm to reduce the noise of sensor data, eliminate the influence of equipment noise or environmental interference, and use multi-sensor data fusion algorithm to jointly analyze different types of sensor data to restore the geological environment change characteristics of the monitoring area from multiple angles. The specific implementation process is as follows:

[0108] S21. Based on spatiotemporal adaptive decomposition and reconstruction (STADR), combined with the spatiotemporal characteristics of the signal, multi-scale decomposition and adaptive learning mechanism are used to separate the noise from the effective signal in the sensor data. The specific implementation process is as follows:

[0109] S2101. Assuming that the input data is a time series signal x(t)=s(t)+n(t), combine wavelet transform and empirical mode decomposition (EMD) to decompose x(t) into multiple components:

[0110] ;

[0111] In the formula, c i (t) represents the component signal, that is, the components of different frequencies and scales obtained after time-space decomposition; r(t) represents the residual, that is, the trend part that cannot be decomposed; k represents the number of components obtained by the decomposition method, that is, the number of sub-signals into which the signal is decomposed;

[0112] S2102. For multi-sensor signal data, spatial correlation analysis is introduced to filter abnormal components through mutual correlation coefficients:

[0113] ;

[0114] In the formula, x i (t), x j (t) represents the data of the i-th and j-th sensors; ρ ij Represents the spatial correlation coefficient, which is used to determine the abnormality of the component; and denote the mean values ​​of the i-th and j-th sensor signals respectively;

[0115] S2103. Calculate the signal-to-noise ratio (SNR) of each component according to the decomposed components:

[0116] ;

[0117] In the formula, s i (t) represents the effective signal of the i-th component; n i (t) represents the noise of the i-th component;

[0118] Based on this, a dynamic threshold is set based on the signal-to-noise ratio: ;

[0119] In the formula, α represents the adjustment factor; δ i represents the standard deviation of the i-th component, indicating the volatility of the component;

[0120] S2104, for each component c i (t), the following rules apply:

[0121] If |c i (t)|≥T i ,but =ci (t);

[0122] If |c i (t)| <T i ,but =0;

[0123] in, Represents the denoised component;

[0124] S2105. After filtering, the effective signals of each component are retained, but there is still nonlinear trend noise in the low-frequency residual r(t). The residual is smoothed using local weighted regression (LOWESS) to obtain ;

[0125] S2106, reconstruct all denoised components and trend residuals to obtain the final denoised signal :

[0126] ;

[0127] The denoised data ;

[0128] S22, use linear interpolation method to align time:

[0129] ;

[0130] In the formula, x aligned (t) Alignment data value at target time point t;

[0131] S23. Standardize data of different dimensions for unified processing:

[0132] ;

[0133] In the formula, x represents the original data value; x max and x min Respectively represent the minimum and maximum values ​​of the current sensor data;

[0134] S24. Based on the adaptive weighted fusion method, the sensors are fused. The fusion process is as follows:

[0135] S2401, setting the adaptive weight of sensor data, the weight calculation formula is:

[0136] ;

[0137] In the formula, w i Represents the weight of sensor i, which determines its contribution to the final fusion result; α i represents the accuracy coefficient of sensor i, that is, the stability and measurement error of the sensor; βi represents the data quality coefficient of sensor i, reflecting the credibility of the current data of the sensor; γ i represents the historical performance coefficient of sensor i;

[0138] S2402, using the weighted average method to fuse the data of each sensor, the fusion formula is as follows:

[0139] ;

[0140] In the formula, x fused represents the fused sensor data, that is, the optimal prediction value obtained after multi-sensor fusion; x i represents the preprocessed data of sensor i; n represents the total number of sensors;

[0141] S2403. In order to improve the stability and accuracy of the fusion results, error estimation is introduced. The error estimation is based on the difference between the predicted and actual data, and the error range of the fusion data is calculated:

[0142] ;

[0143] In the formula, ε i represents the error value of sensor i, that is, the difference between the predicted value and the actual measured value of the sensor data; x pred represents the sensor data of fusion prediction; x meas Indicates the value actually measured by the sensor;

[0144] Accordingly: when it is detected that the data error of a certain sensor is greater than the set threshold, the weight of the sensor is reduced to ensure that the fusion result will not be excessively affected by a single unreliable sensor. The adjusted weight calculation formula is:

[0145] ;

[0146] In the formula, λ' represents the adjustment factor, which is used to control the influence of the error on the weight adjustment; ε i Represents the error value of sensor i; α i represents the accuracy coefficient of sensor i, that is, the stability and measurement error of the sensor; β i represents the data quality coefficient of sensor i, reflecting the credibility of the current data of the sensor; γ i represents the historical performance coefficient of sensor i;

[0147] S25, is the fusion data X={x 1 ,x 2 ,…,x i ,…,x n Generate a pending review mark, the generation process is as follows:

[0148] S2501, select the mark generation code Cg∈[1,q-1], and calculate the mark analysis code Ca=[Cg]×G;

[0149] Where G is the elliptic curve E:y 2 =(x 3 +ax+b) mod p on the q-order base point, E is the finite field F defined above p The elliptic curve on E, a, b, p and q represent the parameters of the elliptic curve E;

[0150] S2502, select a random number r∈[1,q-1], and calculate the first-order auxiliary identification parameter Pl 1 =[r]G;

[0151] S2503, calculate the short information I of the fused data X = H (Pl 1 ,BX);

[0152] Wherein, H represents a hash function; BX represents the binary string form of the fused data X;

[0153] S2504. Calculate the second-order auxiliary identification parameter Pl 2 =(I×Cg+r) mod q, and satisfies Pl 2 ≠0, otherwise return to step S2502;

[0154] S2505, generate a pending review mark L={Pl 1 , Pl 2};

[0155] S26. Transmit {L, X} to the data analysis module.

[0156] S3, the data analysis module receives {L, X}, L = {Pl 1 , Pl 2}, analyze the fused data, identify potential disaster risks, and assess the possibility and severity of disasters. The analysis process is as follows:

[0157] S31. Check whether the received fusion data X is missing. The checking process is as follows:

[0158] S3101, Inspection Pl 2 ∈[1,q-1] whether:

[0159] If it is established, then calculate the first-order auxiliary review parameter Pa 1 =[Pl 2 ]×G;

[0160] If it is not true, it means that the received fusion data X is missing;

[0161] Among them, Pl2 Represents the second-order auxiliary identification parameter; G is the elliptic curve E:y 2 =(x 3 +ax+b) mod p on the q-order base point, E is the finite field F defined above p The elliptic curve on E, a, b, p and q represent the parameters of the elliptic curve E;

[0162] S3102. Calculate the second-order auxiliary review parameter Pa 2 =[H(Pl 1 ,BX)]×Ca+Pl 1 ;

[0163] Wherein, Ca represents the tag parsing code; H represents the hash function; BX represents the binary string form of the fused data X; Pl 1 represents the first-order auxiliary identification parameter;

[0164] S3103, if Pa 1 =Pa 2 , it means that the fused data X is received completely; otherwise, it means that the received fused data X is missing;

[0165] S32, extracting key features through a convolutional neural network (CNN), the key features including but not limited to ground displacement, groundwater level, meteorological changes, and soil moisture;

[0166] S33. Use the principal component analysis (PCA) method to reduce the dimensionality of high-dimensional data. The PCA formula is:

[0167] ;

[0168] Where Z represents the data matrix after dimensionality reduction, which contains the key features extracted from the fused data. Each new column (principal component) represents the projection of the original data in the new feature space. X' represents the input multidimensional original data matrix, which contains various observation data from multiple sensors. Each column represents a different sensor variable, and each row represents a time step or data sampling point. P represents the principal component matrix, which is composed of the eigenvectors in the principal component analysis (PCA) process. Each principal component is a linear combination of the original data, which is used to explain the most important direction of change in the data. Each column is a principal component, which is the basis of the new feature space after the original data is linearly transformed.

[0169] S34. Construct a risk assessment model. First, use the long short-term memory (LSTM) network to model the time series data to capture the time series dependencies in the sensor data. Then combine the features output by the LSTM network with the random forest to construct a hybrid model for the final assessment of disaster risk. The construction process is as follows:

[0170] S3401. Use the LSTM network to learn historical data of time series features (including but not limited to earthquake waves and rainfall changes) and predict future trends. The basic formula of the LSTM model is:

[0171] ;

[0172] In the formula, h t represents the hidden state of LSTM, that is, the network output at time step t; x t Indicates the input data at the current moment; h t-1 Represents the hidden state of the previous moment, that is, the previous network memory; W h , U h represents the weight matrix, that is, the influence of the input data and the previous hidden state on the current hidden state; b h Represents the bias term, which adjusts the offset of the hidden state;

[0173] S3402. After the LSTM neural network models the temporal characteristics of the data, the random forest model is introduced to improve the stability of the model and its ability to classify complex data;

[0174] The random forest model trains multiple decision trees by randomly selecting sample subsets and feature subsets, and then determines the final prediction result through a voting mechanism:

[0175] ;

[0176] In the formula, , , …, Represents the prediction output of multiple decision trees; Represents the final prediction result of random forest;

[0177] S3403, integrating the LSTM neural network with the random forest model, firstly, performing time series modeling on the multi-sensor data through the LSTM network to obtain a feature vector containing time series information; then, the feature vector is passed as input to the random forest model to perform classification prediction of disaster risks;

[0178] The formula of the fused model is:

[0179] ;

[0180] In the formula, f LSTM (X) represents the output predicted by the LSTM model for the input data X, i.e., the features extracted from the time series data; f RF () represents the random forest model, which predicts the features output by LSTM and obtains the final disaster risk probability ;

[0181] S35. Set the risk level L classification rules:

[0182] like <0.3, it is a low risk level;

[0183] If 0.3≤ <0.7, it is a medium risk level;

[0184] like ≥0.7, it is a high risk level;

[0185] S36, merge the fused data X and the analysis result = {disaster risk probability , risk level L} is transmitted to the intelligent warning and decision-making module and the human-computer interaction and feedback module.

[0186] S4, the intelligent early warning and decision-making module issues early warning signals based on the analysis results, and provides decision support after the disaster occurs in combination with the risk assessment results and disaster response strategies. The specific implementation process is as follows:

[0187] S41, based on the analysis results = {disaster risk probability , risk level L}, automatically activate the corresponding emergency response strategy:

[0188] If the probability of disaster risk If the level of infection continues to rise and exceeds the set advanced warning threshold, a comprehensive emergency response plan will be initiated to automatically deploy resources and dispatch emergency teams.

[0189] If the probability of disaster risk If the risk is medium, a warning will be issued and relevant personnel will be advised to strengthen monitoring and protection;

[0190] If the probability of disaster risk If it is below a predetermined threshold, regular monitoring is recommended to ensure that disaster risks are minimized;

[0191] S42, a resource scheduling model based on genetic algorithm, optimizes resource allocation according to disaster type, severity and response time requirements;

[0192] The objective function of the genetic algorithm is:

[0193] ;

[0194] In the formula, represents the weight of the i-th resource, that is, the importance of the resource; t i represents the response time assigned to the i-th resource, that is, the urgency of the resource response; m represents the total number of resources;

[0195] Based on this: The optimization goal of the objective function is to minimize the emergency response time and prioritize the dispatch of important resources to ensure that the losses can be minimized when a disaster occurs;

[0196] S43. After a disaster occurs, conduct a comprehensive assessment of the post-disaster situation, analyze the effectiveness of the emergency response and the efficiency of the response measures. The post-disaster assessment results will serve as feedback input to further optimize the disaster prediction model and emergency response strategy;

[0197] The evaluation formula is:

[0198] ;

[0199] In the formula, E eff Represents the emergency response efficiency; R actual Indicates the actual reduction in post-disaster losses; R expected It represents the expected reduction in post-disaster losses;

[0200] S44, transmitting the resource allocation results and post-disaster assessment results to the human-computer interaction and feedback module.

[0201] S5. Human-computer interaction and feedback module displays the geological data of the monitoring area in real time through a visual interface. X. Analysis results = {disaster risk probability , risk level L}, resource allocation results and post-disaster assessment results, so as to facilitate coordinated dispatch of emergency command personnel.

[0202] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A multi-sensor based intelligent monitoring method for geological disasters, characterized in that: The specific implementation steps include the following: S1. Collect and summarize environmental data; S2, based on the time-space adaptive decomposition and reconstruction method, the sensor data is processed by wavelet transform and empirical mode decomposition, the effective signal is screened and the low-frequency residual is smoothed, and the adaptive weighted fusion method is used to generate the fused data after aligning the data, and then the pending mark is generated for the fused data; The generation process of pending review marks is as follows: S41, select the marker generating code Cg∈[1,q-1], and calculate the marker parsing code Ca=[Cg]×G; Where G is the elliptic curve E:y 2 =(x 3 +ax+b) mod p on the q-order base point, E is the finite field F defined above p The elliptic curve on E, a, b, p and q represent the parameters of the elliptic curve E; S42, select a random number r∈[1,q-1], and calculate a first-order auxiliary identification parameter Pl1=[r]G; S43, calculate the short information I=H(Pl1,BX) of the fused data X; Wherein, H represents a hash function; BX represents the binary string form of the fused data X; S44, calculate the second-order auxiliary identification parameter Pl2 = (I × Cg + r) mod q, and satisfy Pl2≠0, otherwise return to step S42; S45, generate a pending review mark L={Pl1,Pl2}; S3. Check whether the data is missing, extract key features through convolutional neural network and use principal component analysis to reduce the data dimension. Then build a risk assessment model combining LSTM and random forest to capture time series characteristics and conduct disaster risk assessment, and output disaster risk analysis results, which include disaster risk probability and risk level; The review process for checking whether the data is missing is as follows: S51. Check whether Pl2∈[1,q-1] holds: If it holds, then calculate the first-order auxiliary review parameter Pa1=[Pl2]×G; If it is not true, it means that the fused data X is missing; S52, calculate the second-order auxiliary review parameter Pa2=[H(Pl1,BX)]×Ca+Pl1; Wherein, BX represents the binary string form of the fused data X; S53. If Pa1=Pa2, it means that the fused data X is complete; otherwise, it means that the fused data X is missing; S4. Automatically activate emergency response based on disaster risk analysis: comprehensive response for high risk, warning and enhanced monitoring for medium risk, regular monitoring for low risk, and then use genetic algorithm to optimize resource scheduling, minimize response time and prioritize important resources, evaluate the effect after the disaster, optimize the prediction model and response strategy, and output resource allocation and post-disaster evaluation results; S5. Real-time display of fused data, disaster risk analysis results, resource allocation results and post-disaster assessment results of the monitoring area through a visual interface.

2. The multi-sensor based intelligent monitoring method for geological disasters according to claim 1 is characterized in that: The implementation process based on the spatiotemporal adaptive decomposition and reconstruction method is as follows: S21. The input data is a time series signal x(t)=s(t)+n(t). Combining wavelet transform and empirical mode decomposition, x(t) is decomposed into multiple components: ； In the formula, c i (t) represents the component signal, that is, the components of different frequencies and scales obtained after time-space decomposition; r(t) represents the residual, that is, the trend part that cannot be decomposed; k represents the number of components obtained by the decomposition method, that is, the number of sub-signals into which the signal is decomposed; S22. Introduce spatial correlation analysis and filter out abnormal components through mutual correlation coefficients: ； In the formula, x i (t), x j (t) represents the data of the i-th and j-th sensors; ρ ij Represents the spatial correlation coefficient, which is used to determine the abnormality of the component; and denote the mean values ​​of the i-th and j-th sensor signals respectively; S23. Calculate the signal-to-noise ratio (SNR) of each component according to the decomposed components. i : ； In the formula, s i (t) represents the effective signal of the i-th component; n i (t) represents the noise of the i-th component; Based on this, a dynamic threshold is set based on the signal-to-noise ratio: ; In the formula, α represents the adjustment factor; δ i represents the standard deviation of the i-th component, that is, the volatility of the component; S24, for each component c i (t), the following rules apply: If |c i (t)|≥T i ,but =c i (t); If |c i (t)| <T i ,but =0; in, Represents the denoised component; S25, after filtering, the effective signals of each component are retained, but there is still nonlinear trend noise in the low-frequency residual r(t). The residual is smoothed by local weighted regression to obtain ; S26, reconstruct all denoised components and trend residuals to obtain the final denoised signal : ； The denoised data .

3. The multi-sensor based intelligent monitoring method for geological disasters according to claim 1 is characterized in that: The fusion process of fused data is as follows: S31, setting the adaptive weight of the sensor data, the weight calculation formula is: ； In the formula, w i Represents the weight of sensor i, which determines its contribution to the final fusion result; α i represents the accuracy coefficient of sensor i, that is, the stability and measurement error of the sensor; β i represents the data quality coefficient of sensor i, reflecting the credibility of the current data of the sensor; γ i represents the historical performance coefficient of sensor i; S32. Use the weighted average method to fuse the data of each sensor. The fusion formula is as follows: ； In the formula, x fused represents the fused sensor data, that is, the optimal prediction value obtained after multi-sensor fusion; x i represents the preprocessed data of sensor i; n represents the total number of sensors; S33: Update the adaptive weight of the sensor data and output the fused data X.

4. The multi-sensor based intelligent monitoring method for geological disasters according to claim 1, characterized in that: The process of building the risk assessment model is as follows: S61. Use the LSTM network to learn historical data of time series features and predict future trends. The basic formula of the LSTM model is: ； In the formula, h t represents the hidden state of LSTM, that is, the network output at time step t; x t Indicates the input data at the current moment; h t-1 Represents the hidden state of the previous moment, that is, the previous network memory; W h , U h represents the weight matrix, that is, the influence of the input data and the previous hidden state on the current hidden state; b h Represents the bias term, which adjusts the offset of the hidden state; S62. After the LSTM neural network models the temporal characteristics of the data, the random forest model is introduced to improve the stability of the model and its ability to classify complex data; The random forest model trains multiple decision trees by randomly selecting sample subsets and feature subsets, and then determines the final prediction result through a voting mechanism: ； In the formula, , , …, Represents the prediction output of multiple decision trees; Represents the final prediction result of random forest; S63, integrating the LSTM neural network with the random forest model, firstly, performing time series modeling on the multi-sensor data through the LSTM network to obtain a feature vector containing time series information; then, the feature vector is passed as input to the random forest model to perform classification prediction of disaster risks; The formula of the fused model is: ； In the formula, f LSTM (X) represents the output predicted by the LSTM model for the input data X, i.e., the features extracted from the time series data; f RF () represents the random forest model, which predicts the features output by LSTM and obtains the final disaster risk probability .

5. The multi-sensor based intelligent monitoring method for geological disasters according to claim 1 is characterized in that: The optimization process of resource scheduling using genetic algorithm optimization is as follows: optimizing resource allocation according to the disaster type, severity and response time requirements; The objective function of the genetic algorithm is: ； In the formula, represents the weight of the i-th resource, that is, the importance of the resource; t i represents the response time assigned to the i-th resource, that is, the urgency of the resource response; m represents the total number of resources; Based on this: The optimization goal of the objective function is to minimize the emergency response time while giving priority to the scheduling of important resources to ensure that the losses can be minimized when a disaster occurs.

6. The multi-sensor based intelligent monitoring method for geological disasters according to claim 3 is characterized in that: The updating process of the adaptive weights for updating sensor data is as follows: S81. Introduce error estimation. Error estimation is based on the difference between predicted and actual data, and calculates the error range of fused data: ； In the formula, ε i represents the error value of sensor i, that is, the difference between the predicted value and the actual measured value of the sensor data; x pred represents the sensor data of fusion prediction; x meas Indicates the value actually measured by the sensor; S82. When it is detected that the data error of a certain sensor is greater than the set threshold, the weight of the sensor is reduced. The weight calculation formula after adjustment is: ； In the formula, λ' represents the adjustment factor, which is used to control the influence of the error on the weight adjustment; ε i Represents the error value of sensor i; α i represents the accuracy coefficient of sensor i, that is, the stability and measurement error of the sensor; β i represents the data quality coefficient of sensor i, reflecting the credibility of the current data of the sensor; γ i represents the historical performance coefficient of sensor i.

7. A multi-sensor based intelligent geological disaster monitoring system, which is used to execute a multi-sensor based intelligent geological disaster monitoring method according to any one of claims 1 to 6, characterized in that: include: Multi-sensor data acquisition module, which integrates several sensors and is used to collect geological disaster environment data in real time; A data processing module is used to pre-process the collected geological disaster environment data; Data analysis module, used to perform pattern recognition, anomaly detection and trend analysis on sensor data and output analysis results; Intelligent early warning and decision-making module, which is used to issue early warning signals according to the analysis results, and provide decision-making support after the disaster occurs by combining risk assessment results and disaster response strategies; The human-computer interaction and feedback module is used to provide a visual interface, display real-time monitoring data, early warning information and decision support, and continuously optimize the monitoring strategy based on user feedback.

Images