Slope collapse rockfall monitoring method

Abstract

The invention discloses a slope collapse rockfall monitoring method, and relates to the technical field of slope rockfall monitoring, and the method comprises the following steps: arranging multiple types of monitoring sensors in a slope region, and covering a key monitoring region in combination with a camera to form a multi-source data collection network; according to the method, false alarms caused by natural interference and non-dangerous interference are effectively reduced through real-time monitoring of a multi-source sensor and a signal filtering technology based on an environmental noise model, and the authenticity of risk signals is verified from multiple dimensions by combining sensor data with camera images through a multi-modal data fusion algorithm, so that the reliability of the system is improved. In combination with a dynamically updated slope collapse feature library and an intelligent algorithm, the system can accurately identify key parameters of slope instability and predict risks, flexibly responds based on a graded early warning mechanism, improves the disaster prevention and control capability, reduces resource waste caused by false alarm, reduces threats of disasters to life and property, and improves the safety and reliability of the system. And sufficient time is provided for managers to take measures.

Description

Technical Field

[0001] The present invention relates to the technical field of slope rockfall monitoring, and particularly to a method for monitoring slope collapse and rockfall. Background Art

[0002] Slope collapse and rockfall monitoring refers to the process of using professional monitoring equipment and technical means to conduct real-time or periodic observation, analysis, and early warning of possible collapse or rockfall behaviors in slope areas. Its core purpose is to detect potential collapse risks in advance by monitoring geological changes, stress changes, or other abnormal activities of slopes, so as to take appropriate preventive measures to reduce or avoid threats to personnel, property, and the environment. This kind of monitoring usually covers multiple aspects such as sensor arrangement, data collection and processing, early warning mechanism, and risk assessment, involving the comprehensive application of geology, engineering, and information technology.

[0003] The existing technology has the following deficiencies:

[0004] During the monitoring process of slope collapse and rockfall in the existing technology, the problems of false triggering and false alarms of sensor data may lead to serious consequences. Due to the complex field environment, such as interference factors like extreme weather, animal activities, or mechanical vibrations, the monitoring sensors may trigger alarms due to non-hazardous factors, resulting in frequent alarms of the system. This situation may weaken the trust of monitoring personnel in the system, increase resource waste, and delay the identification and response to real disaster signals. In slope disaster monitoring, long-term false alarms may interfere with the judgment of managers on the actual situation, causing the real risks to be ignored, thus leading to major safety accidents. This problem reveals the insufficient environmental adaptability of existing sensors and the need for improvement in signal filtering technology, and it is urgent to solve it through optimized algorithms or multi-dimensional data fusion technology.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, so it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for monitoring slope collapse and rockfall. Through real-time monitoring by multi-source sensors and signal filtering technology based on an environmental noise model, false alarms caused by natural interference and non-hazardous interference are effectively reduced. The multi-modal data fusion algorithm combines sensor data with camera images to verify the authenticity of risk signals from multiple dimensions, improving the reliability of the system. Combining a dynamically updated slope collapse feature library and intelligent algorithms, the system can accurately identify the key parameters of slope instability and predict risks, flexibly respond based on a hierarchical early warning mechanism, improve the disaster prevention and control ability, reduce resource waste caused by false alarms, and at the same time reduce the threat of disasters to life and property, providing sufficient time for managers to take measures to solve the problems in the above background art.

[0007] To achieve the above object, the present invention provides the following technical solution: A monitoring method for slope collapse and falling rocks, characterized by comprising the following steps:

[0008] Arrange multiple types of monitoring sensors in the slope area, and combine cameras to cover key monitoring areas to form a multi-source data acquisition network;

[0009] Perform real-time preprocessing on the raw data collected by multiple types of sensors, remove outliers, and synchronize multiple data in time to form a time-series monitoring data stream;

[0010] Based on a preset environmental noise model, identify and classify interference sources for sensor data, including natural interference and non-dangerous interference, and eliminate irrelevant interference signals;

[0011] Construct a slope collapse behavior feature library, and set key feature parameters and triggering conditions for slope instability in combination with geological conditions, historical data, and expert experience;

[0012] Apply an improved machine learning algorithm to perform real-time analysis on multi-source sensor data, learn the complex associations between feature parameters through model training, and dynamically adjust the alarm threshold to reduce false alarms;

[0013] Apply a multi-modal data fusion algorithm to fuse and analyze sensor and camera data, generate a comprehensive slope state evaluation result, and generate a precise warning signal based on a weight allocation algorithm.

[0014] Preferably, the specific steps of arranging multiple types of monitoring sensors in the slope area and combining cameras to cover key monitoring areas to form a multi-source data acquisition network are as follows:

[0015] Delimit the monitoring area according to geological exploration and risk assessment to ensure coverage of key high-risk points;

[0016] Select sensors according to requirements and arrange them scientifically to ensure a wide monitoring coverage range and stable signal transmission;

[0017] Improve the stability and environmental tolerance of the sensors through reinforcement and protection measures to ensure data accuracy;

[0018] Configure a power and data transmission system to ensure the long-term stable operation of the sensors and real-time data upload.

[0019] Preferably, the specific steps of performing real-time preprocessing on the raw data collected by multiple types of sensors, removing outliers, and synchronizing multiple data in time to form a time-series monitoring data stream are as follows:

[0020] Collect acceleration, strain, infrared, and image data by type, and store them classified to ensure data independence and integrity;

[0021] Identify and remove outliers in sensor data through statistical methods to improve data quality and reliability;

[0022] Perform time synchronization and format standardization on various data to ensure seamless integration and unified processing of cross-type data;

[0023] Integrate the preprocessed multi-source data into a time-ordered time series monitoring data stream to provide an accurate input basis for subsequent analysis.

[0024] Preferably, based on a preset environmental noise model, identify and classify interference sources in sensor data, including natural interference and non-hazardous interference. The specific steps for removing irrelevant interference signals are as follows:

[0025] Extract common interference characteristics from historical data, construct a noise model and use it for signal identification;

[0026] Extract the characteristics of the monitoring data and compare them point by point with the noise model to identify the interference type;

[0027] Remove irrelevant signals according to the interference category to optimize data quality and pertinence;

[0028] Dynamically optimize the noise model with new data to improve the accuracy and adaptability of interference identification.

[0029] Preferably, construct a characteristic library of slope collapse behavior, and the specific steps for setting key characteristic parameters and triggering conditions for slope instability in combination with geological conditions, historical data and expert experience are as follows:

[0030] Summarize the regional geological conditions and historical monitoring data as the basic information input for the characteristic library;

[0031] Screen significant characteristic parameters from the monitoring data that are closely related to slope instability;

[0032] Based on the parameter characteristics, expert experience and experimental results, set accurate triggering conditions for slope instability;

[0033] Continuously optimize the characteristic library with new data to enhance its prediction ability and adaptability to slope instability.

[0034] Preferably, use an improved machine learning algorithm to perform real-time analysis on multi-source sensor data, learn the complex correlations between characteristic parameters through model training, and dynamically adjust the alarm threshold to reduce false alarms. The specific steps are as follows:

[0035] In the multi-source sensor data collected in real time, extract key characteristic parameters, through standardization and normalization processing, convert different physical quantities into dimensionless forms, smooth the outliers in the time series to ensure data quality, and generate a characteristic matrix for subsequent analysis. Its definition formula is as follows:

[0036] ,

[0037] wherein, A t is the original vibration data collected by the acceleration sensor at time t, S t is the stress change data of the rock stratum or soil recorded by the strain sensor at time t, T t is the temperature change data of the slope surface or environment captured by the infrared sensor at time t, C t is the texture complexity feature extracted from the camera image, μ A , μ S , μ T and μ C are the historical means of the acceleration sensor data, strain sensor data, infrared sensor data, and the texture complexity of the camera image respectively, σ A , σ S , σ T and σ C are the historical standard deviations of the acceleration sensor data, strain sensor data, infrared sensor data, and the texture complexity of the camera image respectively, F t is the standardized feature matrix;

[0038] Using the weighted multi-modal analysis method, dynamic weights are assigned to different features, and the feature weight formula is defined as follows:

[0039] W t = softmax(F t ·V + b),

[0040] wherein, V is the linear transformation matrix of the feature weights, b is the bias vector used to control the offset of the model, and the softmax function ensures the normalization of the feature weights W t such that the sum of all weights is 1, and W t is the dynamic feature weight.

[0041] Preferably, the standardized feature matrix F t is trained through a long short-term memory network to learn the complex temporal correlation relationships between features and output a prediction score. The prediction model is defined by the following formula:

[0042] P t = σ(W t ·LSTM(F t-n:t ) + b p ),

[0043] wherein, LSTM(F t-n:t ) models the feature data sequence for n time steps, and b pis the bias term of the prediction model, and σ is the sigmoid function, which is used to limit the prediction score between [0, 1], and P t is the risk prediction score;

[0044] Using the risk prediction value of the previous time step and the current feature weights, the alarm threshold is dynamically calculated. The dynamic threshold formula is defined as follows:

[0045] ,

[0046] where T t is the dynamic alarm threshold at the current time step, T t-1 is the alarm threshold at the previous time step, α is the smoothing coefficient, which is used to balance the weights of the historical threshold and the current calculation result, W t，i is the weight of the i-th feature, and F t，i is the normalized value of the i-th feature.

[0047] Preferably, the multi-modal data fusion algorithm is applied to fuse and analyze the sensor and camera data, generate a comprehensive slope state evaluation result, and generate a precise early warning signal based on the weight allocation algorithm. The specific steps are as follows:

[0048] In order to realize the fusion analysis of different types of sensor data, first, the multi-modal data needs to be normalized to eliminate the influence of unit differences on the calculation. The formula is as follows:

[0049] ,

[0050] where X j is the original data collected by the j-th type of sensor, X min is the minimum value of the collected original data, X max is the maximum value of the collected original data, and X j，n is the normalized data value of the j-th type of sensor;

[0051] Based on the normalized data value X j，n , the multi-modal data is combined into a comprehensive slope state evaluation index using the weighted fusion model. Considering the importance and reliability of the sensors, weights are assigned to each type of data. The calculation expression is as follows:

[0052] ,

[0053] where N is the number of sensor categories, w j is the weight of the j-th type of sensor data, and S is the comprehensive slope state evaluation index.

[0054] Preferably, based on the key parameters and triggering conditions of slope instability in the feature library, it is determined whether the comprehensive evaluation index reaches the warning threshold. The judgment formula for the warning condition is as follows: E = S - T, where E is the abnormal state index, indicating the degree of deviation of the current state from the warning threshold, and T is the dynamically adjusted warning threshold, which is dynamically updated through a machine learning algorithm according to the changes in the monitoring environment and historical data. The calculation formula is as follows: T = T0 + ω·Δ t , where T0 is the initial threshold, and ω is the environmental adjustment coefficient, indicating the impact of the environment on the warning condition, Δ t is the time variation of the characteristic parameter;

[0055] Combining the abnormal state index E and the weights of various sensor data, a precise warning signal is generated through a multi-modal fusion algorithm based on weight allocation. The intensity of the final warning signal is calculated by the following formula:

[0056] ,

[0057] where P is the warning signal intensity, β is the amplification coefficient of abnormal deviation, and γ is the contribution coefficient of multi-source data.

[0058] In the above technical solution, the technical effects and advantages provided by the present invention are as follows:

[0059] Through real-time monitoring by multi-source sensors and interference signal filtering technology based on the environmental noise model, the present invention effectively reduces false alarms caused by natural interferences (such as wind and rain) and non-dangerous interferences (such as animal activities). The system uses a multi-modal data fusion algorithm to combine and analyze the physical signals of sensors with the real-time images of cameras to verify the authenticity of risk signals from multiple dimensions. For example, when the acceleration sensor detects abnormal vibrations, the system further confirms through the camera whether there are falling rocks to ensure that only real danger signals trigger the alarm. In addition, the dynamic update of the noise model enables the system to adapt to different environmental conditions and continuously optimize the signal filtering effect. The reduction of false alarms not only improves the reliability of the system but also enhances the trust of management personnel in the alarms, avoiding resource waste caused by false alarms and delaying the response time to real risks.

[0060] Through the construction of a behavior feature library for slope collapses and optimization in combination with intelligent algorithms, the system can dynamically identify the key parameters of slope instability and predict potential risks. For example, the long-term cumulative anomalies of strain data, the sudden enhancement of vibration signals, and the combined anomalies of multiple parameters can all be used as warning trigger conditions for the system. Based on the dynamically updated feature library and the deep learning prediction model, the system can accurately evaluate the slope stability state, identify potential risks in advance, and issue warnings, providing sufficient time for managers to take prevention and control measures. In addition, the hierarchical warning mechanism enables the system to flexibly adjust the response strategy according to the risk level, ensuring that the monitoring system can not only cope with sudden major disasters but also effectively manage daily risk changes. This accurate warning ability greatly enhances the prevention and control ability of slope disasters and reduces the threat of disasters to life and property. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0062] Figure 1 It is a method flow chart of a method for monitoring slope collapse and falling rocks of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] Now, the exemplary embodiments will be described more fully with reference to the accompanying drawings. However, the exemplary embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these exemplary embodiments are provided so that the present disclosure will be more thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art.

[0064] The present invention provides a method for monitoring slope collapse and falling rocks as shown in Figure 1 the following, which includes the following steps:

[0065] Arrange multiple types of monitoring sensors in the slope area, including acceleration sensors, strain sensors, and infrared sensors, and combine cameras to cover key monitoring areas to form a multi-source data acquisition network;

[0066] The specific steps of arranging multiple types of monitoring sensors in the slope area, including acceleration sensors, strain sensors, and infrared sensors, and combining cameras to cover key monitoring areas to form a multi-source data acquisition network are as follows:

[0067] Delimit the monitoring area according to geological exploration and risk assessment to ensure coverage of key high-risk points;

[0068] First, conduct geological surveys and risk assessments on the target slope area to identify key areas and dangerous points that need to be monitored. These areas include high-risk areas where landslides, collapses or rockfalls may occur, such as the foot of the slope, the top of the slope, areas with concentrated cracks, and historical disaster-prone areas. On this basis, formulate an overall plan for the layout of monitoring equipment. Draw a schematic diagram of the sensor layout based on the terrain conditions to ensure that the layout plan covers every key area. Through this scientific planning, the integrity of the monitoring network can be guaranteed and the occurrence of monitoring blind spots can be avoided.

[0069] Select sensors according to needs and arrange them scientifically to ensure wide monitoring coverage and stable signal transmission;

[0070] Select the appropriate sensor type according to monitoring needs. Accelerometers are used to detect vibration signals from falling rocks, strain sensors are used to monitor stress changes inside the slope, and infrared sensors can monitor temperature changes and thermal radiation from moving objects. Cameras are used to record real-time images of the slope to provide intuitive evidence for data analysis. When arranging, the effective monitoring range, environmental adaptability, and data transmission requirements of different sensors need to be considered. For example, the selection of sensor installation points should try to avoid areas that are easily affected by wind, rain, or animal interference, while considering signal coverage and transmission stability between deployment points, so as to establish a coherent monitoring network.

[0071] Improve the stability and environmental tolerance of sensors through reinforcement and protection measures to ensure data accuracy;

[0072] To prevent the sensors from being disturbed by external factors and affecting the accuracy of monitoring data, special attention should be paid to stability during installation. Accelerometers and strain sensors should be tightly fitted to key slope surfaces or rock cracks, and reinforced with special waterproof sealants or steel fixing brackets. Infrared sensors should be hung at a height that can cover the monitoring area to prevent their line of sight from being blocked or misdirected to non-target areas. Cameras need to be equipped with sun visors and waterproof housings, and adjusted at appropriate angles to ensure coverage of key areas. All sensors should be resistant to high temperatures and corrosion to adapt to the harsh outdoor environment of the slope.

[0073] Configure power and data transmission systems to ensure long-term stable operation of sensors and real-time data upload;

[0074] After the hardware layout is completed, a stable power supply and data transmission channel need to be provided for the sensors. Long-term power support can be provided for the sensors through solar panels and energy storage batteries, and at the same time, long-distance real-time data transmission can be achieved by combining wireless communication modules (such as LoRa, 5G). For wired-connected sensors, corrosion-resistant and waterproof cables and data lines should be laid to ensure their long-term use in harsh environments. In addition, data relay stations or integrated communication terminals can be set up in the area to centrally receive sensor data and transmit it to the remote monitoring center. Finally, by building a perfect transmission and power supply network, the long-term reliable operation of multi-source data acquisition is guaranteed.

[0075] Perform real-time preprocessing on the raw data collected by multiple types of sensors, remove outliers, and synchronize multiple data in time to form a time-series monitoring data stream;

[0076] The specific steps for performing real-time preprocessing on the raw data collected by multiple types of sensors, removing outliers, and synchronizing multiple data in time to form a time-series monitoring data stream are as follows:

[0077] Collect acceleration, strain, infrared, and image data by type and store them classified to ensure data independence and integrity;

[0078] After the sensors are installed and connected to the data transmission network, the system starts to collect various types of raw data in real time. The acceleration sensor records the slope vibration information, the strain sensor obtains the internal stress change data, the infrared sensor captures the temperature and the thermal radiation signal of moving objects, and the camera continuously transmits video image data. Since the output forms and accuracies of various sensors are different, the data first needs to be classified and stored in their respective cache modules to ensure the integrity of the acquisition process and the independence of each type of data. The purpose of classified storage is to provide a clear structure for subsequent processing and avoid conflicts between different sensor data during the acquisition stage.

[0079] Identify and remove outliers in the sensor data through statistical methods to improve data quality and reliability;

[0080] Based on classified storage, outlier detection is performed on various types of sensor data. Outliers may be caused by equipment failures, sudden external environmental interferences (such as strong winds or heavy rains), or accidental mistriggers. Statistical methods such as the Z-value standard deviation method, dynamic range filtering, or moving average method are used to detect abnormal data points that deviate significantly from the normal range. For example, if an acceleration sensor records an abnormally high amplitude value without an actual rockfall event, this point will be determined as an outlier and removed. By removing outliers, the data quality can be significantly improved, and the impact of interference signals on subsequent analysis can be reduced.

[0081] Synchronize multiple data in time and standardize the format to ensure seamless integration and unified processing of cross-type data;

[0082] Since the sampling frequency and timestamp accuracy of different sensors may be different, the data needs to be time synchronized. Align the data of each sensor based on a unified time axis, such as completing or adjusting mismatched time points through timestamp interpolation or linear interpolation methods. At the same time, the data format also needs to be standardized, such as converting camera image data into a specific pixel value matrix and converting acceleration and strain data into a unified physical unit. The purpose of time synchronization and format standardization is to ensure that multi-source data can be seamlessly integrated in subsequent analysis and realize linkage processing across data types.

[0083] Integrate pre-processed multi-source data into a time-series monitoring data stream sorted by time, providing an accurate input basis for subsequent analysis;

[0084] After the outlier removal and time synchronization are completed, the preprocessed data of multiple types of sensors are combined into a unified time series monitoring data stream. This data stream is sorted by time axis and contains synchronization information of different data types, such as vibration value, stress change value, thermal radiation information and corresponding image frames at a certain time point. This time series monitoring data stream not only provides continuous and accurate information on the dynamic change of the slope, but also provides an integrated input basis for subsequent algorithm analysis and early warning systems. Ultimately, through this structured time series data stream, it is ensured that the monitoring system can quickly and accurately capture slope risk signals in complex environments.

[0085] Based on the preset environmental noise model, the sensor data is identified and classified for interference sources, including natural interference (such as wind and rain) and non-dangerous interference (such as small animal activities), and irrelevant interference signals are eliminated;

[0086] Based on the preset environmental noise model, the sensor data is subjected to interference source identification and classification, including natural interference (such as wind, rain, etc.) and non-dangerous interference (such as small animal activities). The specific steps for eliminating irrelevant interference signals are as follows:

[0087] Extract common interference features from historical data, build noise models and use them for signal recognition;

[0088] Based on the environmental characteristics and historical data of the slope area, an environmental noise model is constructed that includes common natural interference (such as wind, rain, temperature fluctuations) and non-dangerous interference (such as small animal activities and mechanical vibrations). The model generates specific parameter standards such as frequency, amplitude, and duration by extracting features and summarizing patterns of interference signals in historical monitoring data. For example, vibrations caused by wind usually appear as high frequencies and low amplitudes, while infrared signals caused by animal activities usually have irregular intermittent characteristics. After the model is built, it is imported into the system as a reference for identifying interference signals to ensure the scientificity and pertinence of the classification.

[0089] Extract the characteristics of the monitoring data and compare them point by point with the noise model to identify the type of interference;

[0090] Perform point-by-point comparative analysis on the time-series monitoring data collected in real time and the environmental noise model. Specifically, the system extracts characteristic parameters from the monitoring data through algorithms (such as the frequency and amplitude of acceleration signals, the fluctuation characteristics of infrared signals, etc.) and matches them with the standard parameter ranges in the model. For example, when the acceleration sensor detects a continuous signal with low amplitude and high frequency, the system identifies it as a wind interference signal; while if the infrared sensor detects a short-term and irregular thermal radiation signal, it may be determined as an animal activity interference. Through this real-time comparison, it can quickly determine whether the monitoring signal belongs to irrelevant interference.

[0091] Eliminate irrelevant signals according to the interference category to optimize the data quality and pertinence;

[0092] For the signals identified as interference, the system classifies and eliminates them. Natural interference signals (such as wind and rain) are usually characterized by changes in fluctuation frequency and amplitude, and these signals will be marked as low priority and removed from the time-series monitoring data stream. For non-dangerous interference signals (such as animal activities and mechanical vibrations), further verification is carried out by combining multi-source data. For example, when the infrared signal shows the movement of a heat source but the camera does not detect relevant images, it can be determined that the signal is an animal interference and eliminated. The data after classification processing is more targeted, reducing the impact of interference on the analysis results.

[0093] Dynamically optimize the noise model with new data to improve the accuracy and adaptability of interference recognition;

[0094] During the data processing process, the system will dynamically adjust and optimize the environmental noise model according to the newly collected data. For example, when a new type of interference signal is detected (such as the specific frequency vibration generated by equipment operation), its characteristic parameters can be added to the noise model through manual annotation or machine learning, and the classification rules of the model can be updated. This dynamic update mechanism ensures the adaptability of the model to environmental changes, enabling it to continuously improve the accuracy of interference recognition and classification, thus providing a more reliable basis for subsequent analysis and early warning.

[0095] Construct a characteristic library of slope collapse behavior, and set the key characteristic parameters and triggering conditions for slope instability in combination with geological conditions, historical data, and expert experience;

[0096] The specific steps for constructing a characteristic library of slope collapse behavior and setting the key characteristic parameters and triggering conditions for slope instability in combination with geological conditions, historical data, and expert experience are as follows:

[0097] Summarize the regional geological conditions and historical monitoring data and input them as the basic information of the characteristic library;

[0098] Based on the geological characteristics of the slope monitoring area, collect relevant geological condition data, including soil type, rock layer structure, slope, rainfall, groundwater level, etc. In addition, combine historical data to obtain dynamic information related to slope instability, such as ground deformation rate, time and intensity of collapses, stress changes, etc. After screening and classifying these data, they will be used as the basic input for the feature library. Combining with the previous results of removing interference signals can ensure the high quality of the input feature data, avoid the influence of interference information on the construction of the feature library, and thus ensure the accuracy and applicability of the feature library.

[0099] Screen significant feature parameters from the monitoring data that are closely related to slope instability;

[0100] Based on the integrated data, extract key feature parameters directly related to slope instability. For example, abnormal amplitude in acceleration data may be a precursor to rockfall, long-term cumulative changes in strain data may reflect the internal geological instability trend, and local temperature increase in infrared data may indicate landslide activity. By comparing with the preprocessed data from which interference signals have been removed, screen out significant parameters that are closely related to instability and eliminate secondary features that have no actual impact on instability. This step aims to provide representative parameters for the feature library to ensure its ability to effectively describe the triggering conditions of slope instability.

[0101] Set accurate triggering conditions for slope instability based on parameter characteristics, expert experience, and experimental results;

[0102] By consulting geological engineering experts and combining the results of slope instability simulation experiments, set reasonable triggering thresholds for the extracted feature parameters. For example, specific conditions such as continuous rainfall exceeding a certain time or intensity, stress change rate reaching a certain threshold, or simultaneous anomalies of multiple parameters can be set as the instability trigger points. These triggering conditions need to be optimized through theoretical verification and comparison with actual data to ensure that they neither miss risks nor generate false alarms in actual monitoring. This step is directly related to the previous noise model identification and classification results, avoiding the influence of noise interference on the setting of triggering conditions.

[0103] Continuously optimize the feature library using new data to enhance its prediction ability and adaptability to slope instability;

[0104] As the slope monitoring continues, new monitoring data is input into the feature library in real time, and the parameters and trigger conditions in the library are dynamically updated through machine learning or statistical analysis methods. After a new slope collapse event occurs, it is necessary to verify by comparing the parameter matching results in the feature library, and adjust the threshold or add new features to improve the applicability and prediction ability of the library. For example, if the system repeatedly identifies a strong correlation between a certain new feature (such as a special stress change pattern) and instability, it will be incorporated into the feature library to enrich the trigger conditions. This dynamic update mechanism not only improves the accuracy of the feature library but also ensures its applicability in long-term monitoring.

[0105] Use an improved machine learning algorithm to perform real-time analysis on multi-source sensor data, learn the complex correlations between feature parameters through model training, and dynamically adjust the alarm threshold to reduce false alarms;

[0106] The specific steps of using an improved machine learning algorithm to perform real-time analysis on multi-source sensor data, learning the complex correlations between feature parameters through model training, and dynamically adjusting the alarm threshold to reduce false alarms are as follows:

[0107] Extract key feature parameters from the real-time collected multi-source sensor data, including acceleration, strain, infrared temperature, and the texture complexity of camera images. Through standardization and normalization processing, convert different physical quantities into dimensionless forms, smooth the outliers in the time series to ensure data quality, and generate a feature matrix for subsequent analysis. Its definition formula is as follows:

[0108] ,

[0109] In the formula, A t is the original vibration data collected by the acceleration sensor at time t, which is usually used to detect vibration activities on the slope, such as precursors of rockfalls or landslides. S t is the stress change data of the rock formation or soil recorded by the strain sensor at time t, which is used to detect the deformation or instability trend inside the slope. T t is the temperature change data of the slope surface or environment captured by the infrared sensor at time t, which is used to detect potential risks caused by groundwater flow, collapse, or other thermal changes. C t is the texture complexity feature extracted from the camera image, which is usually calculated through image processing techniques (such as gradient intensity analysis, Laplace transform), and reflects the degree of change on the slope surface. μ A , μ S , μ T and μ C are the historical means of the acceleration sensor data, strain sensor data, infrared sensor data, and the texture complexity of the camera image respectively. σ A , σ S , σT and σ C are the historical standard deviations of the acceleration sensor data, strain sensor data, infrared sensor data, and the texture complexity of the camera images respectively. F t is the standardized feature matrix, which is the feature matrix formed after normalizing acceleration, strain, infrared temperature, and image texture complexity;

[0110] The standardized F t ensures that the feature parameters have the same scale and are suitable for input into the machine learning model.

[0111] Using the weighted multi-modal analysis method, dynamic weights are assigned to different features to represent the importance of each feature at a specific time point. The feature weight formula is defined as follows:

[0112] W t = softmax(F t ·V + b),

[0113] where V is the linear transformation matrix of the feature weights, representing the internal relationship between features, b is the bias vector used to control the offset of the model, and the softmax function ensures the normalization of the feature weights W t such that the sum of all weights is 1 and it is easy to interpret and compare. W t is the dynamic feature weight;

[0114] The obtained W t can be used to emphasize the important feature signals in the current environment. For example, during heavy rain, the weight of strain data is increased and the influence of image texture is reduced.

[0115] The standardized feature matrix F t is trained through a long short-term memory network (LSTM) to learn the complex temporal correlation relationships between features and output a prediction score. The prediction model is defined by the following formula:

[0116] P t = σ(W t ·LSTM(F t-n:t ) + b p ),

[0117] where LSTM(F t-n:t ) models the feature data sequence for n time steps, b p is the bias term of the prediction model, σ is the sigmoid function used to limit the prediction score between [0, 1], and P t is the risk prediction score;

[0118] The output P t represents the real-time risk value of slope instability. A value close to 1 indicates high risk, and a value close to 0 indicates low risk.

[0119] Using the risk prediction value of the previous time step and the current feature weights, the alarm threshold is dynamically calculated, and the dynamic threshold formula is defined as follows:

[0120] ,

[0121] where T t is the dynamic alarm threshold of the current time step, T t-1 is the alarm threshold of the previous time step, α is the smoothing coefficient, with a value in [0, 1], used to balance the weights of the historical threshold and the current calculation result, W t，i is the weight of the i-th feature, and F t，i is the normalized value of the i-th feature.

[0122] By dynamically adjusting T t , the alarm trigger condition can be adaptively controlled according to environmental changes, significantly reducing the occurrence of false alarms.

[0123] Apply the multi-modal data fusion algorithm to fuse and analyze the sensor and camera data, generate a comprehensive slope state assessment result, and generate a precise early warning signal based on the weight allocation algorithm;

[0124] The specific steps to apply the multi-modal data fusion algorithm to fuse and analyze the sensor and camera data, generate a comprehensive slope state assessment result, and generate a precise early warning signal are as follows:

[0125] To achieve the fusion analysis of different types of sensor data, first, the multi-modal data (including acceleration sensor, strain sensor, infrared sensor, and camera data) needs to be normalized. The purpose of normalization is to map various data to a unified numerical range, eliminating the influence of unit differences on calculations. The formula is as follows:

[0126] ,

[0127] where X j is the original data collected by the j-th type of sensor, X min is the minimum value of the collected original data, X max is the maximum value of the collected original data, and X j，n is the normalized data value of the j-th type of sensor;

[0128] The normalized X j，n value is used for the subsequent fusion analysis steps to ensure the comparability and consistency of various data.

[0129] For the normalized data value X t，nBased on this, a weighted fusion model is used to combine multi-modal data into a comprehensive slope state evaluation index. Considering the importance and reliability of sensors, weights are assigned to each type of data, and the calculation formula is as follows:

[0130] ,

[0131] where N is the number of sensor categories (for example, there are 4 categories including acceleration, strain, infrared, and camera, N = 4), w j is the weight of the data of the j-th type of sensor, S is the comprehensive slope state evaluation index, which is the result obtained through weighted fusion and represents the current comprehensive stability state of the slope;

[0132] The initial value of the weight w j can be set according to the monitoring accuracy and historical performance of the sensor, and is subsequently adaptively adjusted by the model. The calculated value of S characterizes the current comprehensive stability state of the slope.

[0133] Based on the key parameters and triggering conditions of slope instability in the feature library, it is judged whether the comprehensive evaluation index reaches the warning threshold. The judgment formula for the warning condition is as follows: E = S - T, where E is the abnormal state index, indicating the deviation degree between the current state and the warning threshold, T is the dynamically adjusted warning threshold, which is dynamically updated through machine learning algorithms according to the changes in the monitoring environment and historical data. The calculation formula is as follows: T = T0 + ω·Δ t , where T0 is the initial threshold, ω is the environmental adjustment coefficient, indicating the influence of the environment on the warning condition, and Δ t is the time change amount of the characteristic parameter, such as the difference between the latest data and the historical mean;

[0134] When E > 0, the system triggers a warning signal; otherwise, it is in a stable state.

[0135] Combining the abnormal state index E and the weights of various types of sensor data, a precise warning signal is generated through a multi-modal fusion algorithm based on weight allocation. The intensity of the final warning signal is calculated by the following formula:

[0136] ,

[0137] where P is the warning signal intensity, the higher the value, the greater the risk. β is the amplification coefficient of abnormal deviation, emphasizing the influence of E, and γ is the contribution coefficient of multi-source data, indicating the correction effect of comprehensive data on the warning.

[0138] The system matches the P value with the warning level mapping table to generate the corresponding warning level (such as green, yellow, red alarms) and outputs a complete slope state evaluation report.

[0139] Embodiment 1: The first step in slope collapse monitoring is to construct a comprehensive multi-source sensor monitoring network to achieve real-time data collection and dynamic risk assessment. Multiple types of sensors are arranged in the slope area, including acceleration sensors, strain sensors, infrared sensors, and cameras, and each sensor has its specific monitoring function. The acceleration sensor is used to detect the vibration signals of falling rocks or precursors of landslides, the strain sensor is used to monitor the internal stress changes of soil or rock layers, the infrared sensor can capture temperature anomalies or the thermal radiation of moving objects, and the camera provides real-time visual information to verify the detection results of physical sensors. These sensors are reasonably arranged according to the topographic characteristics and distribution of dangerous points of the slope to ensure the monitoring coverage and sensitivity. The sensor data is transmitted to the data processing center through wireless or wired networks for centralized management and processing.

[0140] In the preprocessing stage of sensor data, the data is first classified and stored and preliminarily cleaned. Since the sensors are installed in a complex field environment, the data often contains interference signals, such as outliers caused by strong winds, rain, or animal activities. To ensure the accuracy and usability of the data, the system uses statistical methods (such as Z-value analysis, moving average method) and dynamic range filtering technology to screen the original data. By removing outliers, the interference of non-related signals to the monitoring results can be effectively reduced. At the same time, the sampling frequencies of different sensors may be different, so it is necessary to synchronize the data in time, align all the data on the same time axis, so as to form a unified time-series monitoring data stream and provide a reliable basis for subsequent in-depth analysis.

[0141] To further improve the accuracy and applicability of the monitoring data, the system introduces a preset environmental noise model. Based on the environmental characteristics and historical data of the slope area, the model predefines the characteristics of common interference signals, such as the high-frequency and low-amplitude signals of wind, the random pulse characteristics of rain, and the irregular infrared signals of animal activities. By comparing the real-time data with the noise model point by point, the system can quickly identify and classify interference signals. For example, when the infrared sensor detects the movement of a heat source but the camera does not capture the relevant picture, it can be determined that the signal is an animal interference and be excluded. The joint analysis of multi-source data not only reduces the false alarm rate but also can more accurately capture the real slope instability signals.

[0142] Finally, through this method of real-time monitoring of multi-source sensors and filtering of interference signals, the system can provide high-precision data support in a complex slope environment. Its advantages are that the complementarity of multiple types of sensors significantly improves the comprehensiveness of monitoring, while noise filtering and data synchronization processing greatly enhance the reliability and analysis efficiency of the data, laying a solid foundation for the subsequent construction of the feature library and the operation of the early warning system.

[0143] Embodiment 2: On the basis of multi-source data collection and filtering, the construction of the feature library is a key link to achieve accurate early warning of slope collapse. The establishment of the feature library is based on geological conditions, historical monitoring data, and expert experience. The purpose is to extract and summarize the key feature parameters that can reflect the slope instability risk and set reasonable triggering conditions for these parameters. For example, the abnormal increase in slope deformation rate, the stress change reaching the critical value, and the soil saturation caused by continuous rainfall may all be precursors of slope instability. The extraction of these parameters not only requires the support of historical data but also needs to be optimized through theoretical analysis and field verification to ensure the scientificity and practicality of the parameters.

[0144] The triggering conditions of the key parameters in the feature library are set based on comprehensive analysis from multiple sources. For example, historical data may show that a landslide occurred in a certain slope area after the continuous rainfall reached 100 mm, then the triggering threshold of rainfall can be set close to this value. At the same time, through simulation experiments and numerical calculations, the interaction relationship between feature parameters is further verified. For example, the abnormal stress change rate is often related to the abnormal ground vibration signal, and the system can use the joint abnormality of multiple parameters as one of the triggering conditions. By setting such refined triggering conditions, false alarms caused by single-parameter fluctuations can be effectively reduced, and the accuracy of early warning can be improved.

[0145] During the operation of the system, the real-time collected monitoring data will be dynamically compared with the triggering conditions in the feature library. The system adopts a hierarchical early warning mechanism to grade and evaluate the stability state of the slope according to the degree of satisfaction of the triggering conditions. When the triggering conditions are fully met, the system immediately issues a high-level alarm to notify the management to take emergency measures. For example, if the strain sensor detects that the stress accumulation in the key area exceeds the set threshold, and at the same time the acceleration sensor detects an abnormal vibration signal, the early warning mechanism can be immediately activated. The advantage of hierarchical early warning is that it can provide sufficient response time for managers and avoid waste of resources caused by over-sensitivity.

[0146] In addition, the feature library has the ability to be dynamically updated. When the system identifies new instability patterns or discovers deficiencies in existing triggering conditions during operation, the feature library can be optimized through machine learning algorithms. For example, new monitoring data may reveal a strong correlation between a previously undetected stress change pattern and instability, and the system can add this pattern to the feature library and adjust the relevant triggering conditions. Through this dynamic update mechanism, the feature library can continuously improve its applicability and prediction accuracy, so as to maintain high efficiency and accuracy in long-term monitoring.

[0147] Implementation method 3: Multimodal data fusion is a technical means of combining multiple sensor data types for joint analysis. Its goal is to improve the reliability and intelligence level of slope monitoring through the collaborative processing of different data sources. In the slope monitoring system, acceleration, strain, infrared and camera data provide vibration, stress, temperature and image information respectively. These information may have limitations when analyzed separately, but fusion processing can make up for the defects of a single data source. For example, when the acceleration sensor detects abnormal vibration, the system can confirm whether there is rockfall in combination with the real-time image of the camera to avoid false alarms. In addition, the thermal radiation information captured by the infrared sensor can assist in identifying potential landslide areas and provide evidence for abnormal signals in acceleration and strain data.

[0148] The implementation of data fusion relies on weight allocation algorithms and machine learning techniques. During the fusion process, the system dynamically adjusts the weights according to the importance of the data source. For example, in severe weather conditions, infrared sensors may be disturbed and their weights need to be appropriately reduced, while the importance of strain and acceleration sensor data is relatively increased. The dynamic adjustment of weights ensures that the system can accurately identify risk signals in various complex environments. By training machine learning models, the system can automatically extract complex associations between multi-source data, thereby improving the ability to perceive potential risks. For example, the system can identify small fluctuation patterns between multiple parameters and match them with historical collapse events, thereby issuing an early alarm.

[0149] The role of intelligent algorithms in multimodal data fusion cannot be ignored. In addition to dynamically adjusting weights, intelligent algorithms can also optimize the system's early warning mechanism through adaptive learning. For example, a time series prediction model based on deep learning can predict the future stability of the slope, thereby identifying potential risks in advance. At the same time, the algorithm can also continuously update the priority and trigger conditions of feature parameters, allowing the system to adapt to new environments and data changes. The introduction of intelligent algorithms not only improves the degree of automation of the monitoring system, but also significantly reduces the false alarm rate and missed alarm rate.

[0150] By combining multimodal data fusion with intelligent algorithms, the slope monitoring system can handle complex risk scenarios more efficiently and provide accurate real-time warnings. The advantage of this technical approach is that it can fully tap the potential of different data sources, maximize the value of data, and provide managers with comprehensive and reliable risk assessment information, so that effective preventive measures can be taken before disasters occur.

[0151] Through real-time monitoring by multi-source sensors and interference signal filtering technology based on an environmental noise model, the present invention effectively reduces false alarms caused by natural interferences (such as wind and rain) and non-dangerous interferences (such as animal activities). The system utilizes a multi-modal data fusion algorithm to combine and analyze the physical signals of sensors with the real-time images of cameras, verifying the authenticity of risk signals from multiple dimensions. For example, when the acceleration sensor detects abnormal vibrations, the system further confirms through the camera whether there are falling rocks, ensuring that only real danger signals trigger alarms. In addition, the dynamic update of the noise model enables the system to adapt to different environmental conditions and continuously optimize the signal filtering effect. The reduction of false alarms not only improves the reliability of the system but also enhances the trust of management personnel in the alarms, avoiding resource waste caused by false alarms and delaying the response time to real risks.

[0152] By constructing a characteristic library of slope collapse behaviors and optimizing with intelligent algorithms, the system of the present invention can dynamically identify the key parameters of slope instability and predict potential risks. For example, the long-term cumulative anomalies of strain data, the sudden enhancement of vibration signals, and the combined anomalies of multiple parameters can all be used as the warning trigger conditions of the system. Based on the dynamically updated characteristic library and deep learning prediction model, the system can accurately evaluate the slope stability state, identify potential risks in advance and issue warnings, providing sufficient time for managers to take prevention and control measures. In addition, the hierarchical warning mechanism enables the system to flexibly adjust the response strategy according to the risk level, ensuring that the monitoring system can not only cope with sudden major disasters but also effectively manage daily risk changes. This precise warning ability greatly enhances the prevention and control ability of slope disasters and reduces the threat of disasters to life and property.

[0153] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0154] Only some exemplary embodiments of the present invention have been described by way of illustration above. Undoubtedly, for those of ordinary skill in the art, without departing from the spirit and scope of the present invention, the described embodiments can be modified in various different ways. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

[0155] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0156] It should be understood that in various embodiments of the present application, the magnitude of the sequence numbers of the above processes does not mean the order of execution, and the order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0157] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0158] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0159] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0160] In addition, in various embodiments of the present application, the functional units can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0161] As described above, this is only the specific implementation manner of the present application. However, the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

[0162] Only some exemplary embodiments of the present invention have been described above by way of illustration. Undoubtedly, for those of ordinary skill in the art, the described embodiments can be modified in various different ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the protection scope of the claims of the present invention.

Claims

1. A monitoring method for slope collapse and falling rocks, characterized in that The following steps are involved: Arrange multiple types of monitoring sensors in the slope area and combine them with cameras to cover key monitoring areas to form a multi-source data collection network; Real-time preprocessing of raw data collected by multiple types of sensors to remove outliers, and time synchronization of multiple data to form a time series monitoring data stream; Based on the preset environmental noise model, the sensor data is identified and classified for interference sources, including natural interference and non-dangerous interference, and irrelevant interference signals are eliminated; Construct a slope collapse behavior feature library, and set key characteristic parameters and trigger conditions for slope instability based on geological conditions, historical data, and expert experience; Use improved machine learning algorithms to analyze multi-source sensor data in real time, learn the complex relationships between feature parameters through model training, and dynamically adjust alarm thresholds to reduce false alarms; A multimodal data fusion algorithm is used to fuse and analyze sensor and camera data to generate comprehensive slope status assessment results, and accurate warning signals are generated based on a weight distribution algorithm.

2. The slope collapse rockfall monitoring method according to claim 1, characterized in that The specific steps of deploying multiple types of monitoring sensors in the slope area and combining them with cameras to cover key monitoring areas to form a multi-source data collection network are as follows: Delineate monitoring areas based on geological surveys and risk assessments to ensure coverage of key high-risk points; Select sensors according to needs and arrange them scientifically to ensure wide monitoring coverage and stable signal transmission; Improve the stability and environmental tolerance of sensors through reinforcement and protection measures to ensure data accuracy; Configure power and data transmission systems to ensure long-term stable operation of sensors and real-time data uploading.

3. The slope collapse rockfall monitoring method according to claim 1, wherein The specific steps for real-time preprocessing of raw data collected by multiple types of sensors, removing outliers, and time-synchronizing multiple data to form a time series monitoring data stream are as follows: Collect acceleration, strain, infrared and image data by type and store them in categories to ensure data independence and integrity; Identify and remove outliers in sensor data through statistical methods to improve data quality and reliability; Synchronize time and standardize formats of various data to ensure seamless integration and unified processing of cross-type data; Integrate the preprocessed multi-source data into a time-series monitoring data stream sorted by time, providing an accurate input basis for subsequent analysis.

4. A method for monitoring rockfalls on a slope according to claim 1, characterized in that, Based on the preset environmental noise model, the sensor data is subjected to interference source identification and classification, including natural interference and non-dangerous interference, and the specific steps for eliminating irrelevant interference signals are as follows: Extract common interference features from historical data, build noise models and use them for signal recognition; Extract monitoring data features and compare them point by point with the noise model to identify interference types; Eliminate irrelevant signals according to interference categories to optimize data quality and pertinence; Dynamically optimize the noise model with new data to improve the accuracy and adaptability of interference identification.

5. A method for monitoring slope collapse and falling rocks according to claim 1, characterized in that, The specific steps for constructing a slope collapse behavior feature library and setting key characteristic parameters and triggering conditions for slope instability based on geological conditions, historical data, and expert experience are as follows: Summarize regional geological conditions and historical monitoring data as basic information input for the feature library; Screen out significant characteristic parameters that are closely related to slope instability from monitoring data; Set accurate triggering conditions for slope instability based on parameter characteristics, expert experience, and experimental results; Continuously optimize the feature library with new data to enhance its prediction ability and adaptability for slope instability.

6. A method for monitoring rockfalls on a slope according to claim 1, characterized in that, The specific steps for real-time analysis of multi-source sensor data using an improved machine learning algorithm, learning the complex correlations between feature parameters through model training, and dynamically adjusting the alarm threshold to reduce false alarms are as follows: In the multi-source sensor data collected in real-time, extract key feature parameters, through standardization and normalization processing, convert different physical quantities into dimensionless forms, smooth the outliers in the time series to ensure data quality, and generate a feature matrix for subsequent analysis. Its definition formula is as follows: ， Where, A t is the original vibration data collected by the acceleration sensor at time t, S t is the stress change data of the rock formation or soil recorded by the strain sensor at time t, T t is the temperature change data of the slope surface or environment captured by the infrared sensor at time t, C t is the texture complexity feature extracted from the camera image, μ A 、μ S 、μ T and μ C are the historical means of the acceleration sensor data, strain sensor data, infrared sensor data, and the texture complexity of the camera image respectively, σ A 、σ S 、σ T and σ C are the historical standard deviations of the acceleration sensor data, strain sensor data, infrared sensor data, and the texture complexity of the camera image respectively, F t is the standardized feature matrix; Using the weighted multi-modal analysis method, assign dynamic weights to different features. The feature weight formula is defined as follows: W t = softmax(F t ·V + b), Wherein, V is a linear transformation matrix of feature weights, b is a bias vector used to control the offset of the model, and the softmax function ensures the normalization of the feature weights W t such that the sum of all weights is 1, and W t are dynamic feature weights.

7. A slope collapse rockfall monitoring method according to claim 6, characterized in that, Through the long short-term memory network for the standardized feature matrix F t Perform training, learn the complex temporal correlation relationships between features, and output prediction scores. The definition formula of the prediction model is as follows: P t = σ(W t ·LSTM(F t-n:t ) + b p ), where, LSTM(F t-n:t ) models the feature data sequence for n time steps, b p is the bias term of the prediction model, σ is the sigmoid function used to limit the prediction score between [0, 1], and P t is the risk prediction score; Using the risk prediction value of the previous time step and the current feature weights, dynamically calculate the alarm threshold. The dynamic threshold formula is defined as follows: ， where T t is the dynamic alarm threshold of the current time step, T t-1 is the alarm threshold of the previous time step, α is the smoothing coefficient used to balance the weights of the historical threshold and the current calculation result, W t，i is the weight of the i-th feature, F t，i is the normalized value of the i-th feature.

8. A method for monitoring slope collapse and falling rocks according to claim 1, characterized in that, The specific steps for applying the multi-modal data fusion algorithm to fuse and analyze sensor and camera data, generate a comprehensive slope state assessment result, and generate a precise early warning signal based on the weight assignment algorithm are as follows: To achieve the fusion analysis of different types of sensor data, first, normalize the multi-modal data to eliminate the influence of unit differences on calculations. The formula is as follows: ， where X j is the original data collected by the j-th type of sensor, X min is the minimum value of the collected original data, X max is the maximum value of the collected original data, X j，n is the data value after normalization of the j-th type of sensor; Based on the normalized data value X j，n On this basis, a weighted fusion model is used to combine multi-modal data into a comprehensive slope state evaluation index. Considering the importance and reliability of sensors, weights are assigned to each type of data, and the calculation expression is as follows: ， where N is the number of sensor categories, w j is the weight of the data of the j-th type of sensor, and S is the comprehensive slope state evaluation index.

9. A method for monitoring slope collapse and falling rocks according to claim 8, characterized in that, Based on the key parameters and triggering conditions of slope instability in the feature library, determine whether the comprehensive evaluation index reaches the warning threshold. The judgment formula for the warning condition is as follows: E = S - T, where E is the abnormal state index, indicating the degree of deviation of the current state from the warning threshold, T is the dynamically adjusted warning threshold, which is dynamically updated according to the changes in the monitoring environment and historical data through machine learning algorithms. The calculation formula is as follows: T = T0 + ω·Δ t , where T0 is the initial threshold, ω is the environmental adjustment coefficient, indicating the impact of the environment on the warning condition, Δ t is the time variation of the characteristic parameter; Combining the abnormal state index E and the weights of various types of sensor data, generate a precise early warning signal through the multi-modal fusion algorithm based on weight assignment. The intensity of the final early warning signal is calculated by the following formula: ， In the formula, P is the intensity of the early warning signal, β is the amplification coefficient of abnormal deviation, and γ is the contribution coefficient of multi-source data.

Images