Landslide disaster identification method and system based on multi-source data fusion

Abstract

The invention provides a landslide disaster recognition method and system based on multi-source data fusion, and relates to the technical field of landslide disaster recognized.The landslide disaster recognition method comprises the steps that firstly, sensors are evenly arranged in a landslide area to obtain soil humidity, underground water level, horizontal displacement and vertical displacement, and meteorological data at the same moment are recorded; thirdly, preprocessing the collected data, extracting a feature matrix by applying principal component analysis, then constructing a landslide risk model based on a deep learning model, and performing training by taking a landslide risk index as a label; and on the basis, analyzing a landslide risk index and meteorological data to obtain a landslide risk probability. And finally, monitoring slope body data in real time, preprocessing the slope body data, inputting the preprocessed slope body data into the trained model, obtaining a current landslide risk index, calculating a risk probability, and comparing the risk probability with a preset threshold value to judge a landslide safety state, thereby providing a scientific basis for landslide disaster early warning.

Description

Technical Field

[0001] The invention relates to the technical field of landslide hazard identification, and in particular to a landslide hazard identification method and system based on multi-source data fusion. Background Art

[0002] Landslide disasters are geological disasters caused by multiple factors such as geology, meteorology and human activities, which often cause serious casualties and property losses. With the frequent occurrence of global climate change and extreme weather events, the risk of landslides is increasing, especially in rainy and earthquake-prone areas. The background technology of landslide hazard identification mainly involves the comprehensive application of multiple disciplines such as remote sensing technology, geological exploration, topographic mapping and geographic information system. Through the analysis of high-resolution satellite images, drone aerial photography, radar interferometry and seismic monitoring data, it can quickly and accurately identify the scope, activity characteristics and potential risk areas of landslides, so as to achieve timely early warning and scientific post-disaster assessment. The coordinated use of these technologies makes the prediction and management of landslide disasters more accurate and efficient. Traditional landslide monitoring methods mainly rely on manual investigation and simple physical indicator monitoring, which cannot reflect the dynamic changes in complex geological environments in real time.

[0003] The prior art has the following deficiencies:

[0004] Traditional landslide hazard identification technology mainly relies on a single data source, such as remote sensing images or meteorological data. This method often has limitations in dealing with variable geological and climatic conditions, resulting in frequent omissions or false alarms. In addition, the timeliness of the data is poor, making it difficult to achieve timely warnings, resulting in a slow response when a disaster occurs, thereby increasing the risk of loss of life and property. Therefore, how to effectively integrate multi-source real-time data to improve the identification accuracy and response speed of landslide disasters has become a technical problem that needs to be solved urgently.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a landslide hazard identification method and system based on multi-source data fusion to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides a landslide hazard identification method based on multi-source data fusion, the specific steps of which include:

[0008] Step 1: Arrange relevant sensors at multiple locations of the landslide to obtain slope-related data, including soil moisture, groundwater level, horizontal displacement and vertical displacement data, and obtain meteorological data at the corresponding time according to the timestamp, wherein the meteorological data includes rainfall, temperature and wind speed data;

[0009] Step 2: Preprocess the collected slope-related data and meteorological data, then use principal component analysis to extract features from the preprocessed data to obtain the covariance matrix, and perform eigenvalue decomposition to obtain a new feature matrix;

[0010] Step 3: Build a landslide risk model based on the deep learning model, input the new feature matrix into the model, obtain the landslide risk index as a label, and train the landslide risk model;

[0011] Step 4: Analyze the landslide risk index and the meteorological data of the corresponding timestamp to obtain the landslide risk probability;

[0012] Step 5: Collect slope-related data and meteorological data in real time, execute step 2 to obtain the real-time feature matrix, input the real-time feature matrix into the trained landslide risk model, obtain the landslide risk index output by the model, calculate the landslide risk probability at the current moment, and compare it with the preset risk threshold to judge the current landslide safety status.

[0013] Furthermore, the specific logic for collecting slope-related data and meteorological data is as follows:

[0014] The landslide area is divided into multiple monitoring grid areas according to the size of Sm×Sm. A humidity sensor and a displacement meter are arranged at the center of each grid to obtain the soil moisture, horizontal displacement and vertical displacement data of the monitoring point. A water level sensor is installed in the groundwater level monitoring well at the bottom edge of the landslide to obtain the groundwater level data. The collected slope-related data are marked with a timestamp, and the meteorological data at the same timestamp is recorded.

[0015] Furthermore, the specific logic for obtaining a new feature matrix is ​​as follows:

[0016] The collected slope-related data and meteorological data were cleaned, and the missing data were processed by the missing value interpolation method. The data with the same timestamp were integrated into a set of observation vectors X. The principal component analysis was used to extract features from the preprocessed data. The formula for obtaining the covariance matrix of the data was as follows:

[0017]

[0018] Where C is the calculated covariance matrix, n is the total number of data samples, X is the observation vector composed of the collected data, and μ is the mean vector of the data;

[0019] Perform eigenvalue analysis on the calculated covariance matrix, and the formula for obtaining the eigenvector is:

[0020] Cv=λv

[0021] Among them, λ is the eigenvalue and v is the corresponding eigenvector;

[0022] The calculated eigenvalues ​​are sorted from large to small, and the eigenvectors corresponding to the first K eigenvalues ​​are selected to form a new matrix. The formula for projecting the new characteristic matrix is:

[0023] Y=XV k

[0024] Among them, Y is the new feature matrix obtained by projection, V k A new matrix consisting of the eigenvectors corresponding to the first K selected eigenvalues.

[0025] Furthermore, the specific logic for constructing the landslide risk model is as follows:

[0026] A landslide risk model is constructed based on the combination of convolutional neural network and long short-term memory network. The new feature matrix obtained after principal component analysis is input into the landslide risk model. After convolution operation through the CNN layer, the state is updated through the LSTM layer. The input layer uses the softmax activation function to output the landslide risk index predicted by the model as the label. The loss function selects monitoring verification loss. When the verification loss function drops below 0.01 in five consecutive training cycles, the model training is completed.

[0027] Furthermore, the specific logic for obtaining the landslide risk probability is as follows:

[0028] The landslide risk index of each grid area is obtained through the model, and the average of the landslide risk index of all grid areas is calibrated as the landslide risk index of the slope. The landslide risk index of the slope and the rainfall, temperature and wind speed data in the same time period are analyzed, and the formula for obtaining the landslide risk probability of the slope is:

[0029]

[0030] Among them, Q is the landslide risk index of the slope, m is the number of grid areas, and q i is the landslide risk index of the i-th grid area output by the model;

[0031]

[0032] Among them, P is the landslide risk probability of the slope, R is the rainfall, T is the temperature, W is the wind speed, e is the nonlinear parameter, k is the adjustment parameter, β1, β2, β3, and β4 are the weight coefficients of each item, β1>β2>β3>β4>0, and β1+β2+β3+β4=1.

[0033] Furthermore, the specific logic for judging the current landslide safety status is as follows:

[0034] Collect soil moisture, groundwater level, horizontal displacement and vertical displacement data of the landslide at the current moment, perform data preprocessing, obtain the real-time feature matrix, input the real-time feature matrix into the model, output the landslide risk index at the current moment, and combine the rainfall, temperature and wind speed data at this moment to obtain the landslide risk probability at the current moment, compare the landslide risk probability at the current moment with the preset risk threshold, if the landslide risk probability at the current moment is higher than the preset risk threshold, issue a landslide risk warning, if the landslide risk probability at the current moment is lower than the preset risk threshold, continue monitoring.

[0035] The present invention further provides a landslide hazard identification system based on multi-source data fusion, wherein the landslide hazard identification system based on multi-source data fusion is used to implement the above-mentioned landslide hazard identification method based on multi-source data fusion, and comprises:

[0036] A data acquisition module is used to arrange relevant sensors at multiple locations of the landslide to obtain slope-related data, including soil moisture, groundwater level, horizontal displacement and vertical displacement data, and obtain meteorological data at the corresponding time according to the timestamp, and the meteorological data includes rainfall, temperature and wind speed data;

[0037] The data preprocessing and feature extraction module is used to preprocess the collected slope-related data and meteorological data, and then use principal component analysis to extract features from the preprocessed data to obtain the covariance matrix, and perform eigenvalue decomposition to obtain a new feature matrix;

[0038] The landslide risk model construction module is used to build a landslide risk model based on the deep learning model, input a new feature matrix into the model, obtain the landslide risk index as a label, and train the landslide risk model;

[0039] The landslide risk probability acquisition module is used to analyze the landslide risk index and the meteorological data of the corresponding timestamp to obtain the landslide risk probability;

[0040] The real-time monitoring and risk assessment module is used to collect slope-related data and meteorological data in real time, execute the data preprocessing and feature extraction module to obtain the real-time feature matrix, input the real-time feature matrix into the trained landslide risk model, obtain the landslide risk index output by the model, calculate the landslide risk probability at the current moment, and compare it with the preset risk threshold to judge the current landslide safety status.

[0041] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0042] The present invention adopts a multi-source data fusion method, combines multi-dimensional data such as meteorological monitoring, soil moisture, groundwater level and displacement deviation, and constructs a deep learning model to achieve dynamic monitoring and assessment of landslide risks. This technology not only significantly improves the accuracy of landslide disaster identification and reduces the probability of missed reports and false reports, but also can provide timely warnings before disasters occur, ensuring that relevant departments can quickly take countermeasures, thereby effectively reducing casualties and property losses. Through this systematic solution, the intelligent level of landslide disaster monitoring and early warning has been significantly improved, promoting the modernization of emergency management. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0044] Figure 2 Schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0046] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0047] Example:

[0048] See also Figure 1 The present invention provides a landslide hazard identification method based on multi-source data fusion, and the specific steps include:

[0049] Step 1: Arrange relevant sensors at multiple locations of the landslide to obtain slope-related data, including soil moisture, groundwater level, horizontal displacement and vertical displacement data, and obtain meteorological data at the corresponding time according to the timestamp, wherein the meteorological data includes rainfall, temperature and wind speed data;

[0050] In this embodiment, the specific logic for collecting slope-related data and meteorological data is as follows:

[0051] The landslide area was divided into multiple monitoring grid areas of 5m×5m in size. Humidity sensors and displacement meters were arranged at the center of each grid to obtain soil moisture, horizontal displacement and vertical displacement data of the monitoring point. A water level sensor was installed in the groundwater level monitoring well at the bottom edge of the landslide to obtain groundwater level data. The collected slope-related data were marked with timestamps, and the meteorological data at the same timestamp were recorded.

[0052] Through refined monitoring, multi-dimensional data of the landslide area can be obtained, thus providing a solid data foundation for subsequent risk assessment and early warning. By dividing the landslide area into 5m×5m monitoring grids and arranging humidity sensors and displacement meters in the center of each grid, it is possible to achieve detailed monitoring of the slope state and capture key indicators such as soil moisture changes and displacement trends, which helps to identify potential landslide triggers. In addition, installing a water level sensor at the bottom of the landslide to monitor groundwater level changes can effectively reflect the impact of hydrological conditions on landslides. By correlating the collected slope data with the corresponding meteorological data, the relationship between meteorological factors and landslide risks can be more comprehensively analyzed, improving the accuracy and reliability of the model.

[0053] Changes in soil moisture affect the bearing capacity and friction of the soil. Wet soil is more prone to sliding, so regular monitoring can identify potential landslide risks in a timely manner. Rising groundwater levels can lead to soil saturation and increase the probability of landslides. Monitoring groundwater levels can reveal the impact of moisture on slope stability. Horizontal and vertical displacement data can provide actual deformation information of the slope, help analyze the movement trend of the slope, and detect the precursors of landslides early. Meteorological factors such as rainfall, temperature, and wind speed can directly affect parameters such as soil moisture and groundwater levels. Understanding meteorological conditions can help determine the timeliness of landslide risks. Comprehensive monitoring of these data can help researchers better understand the triggering mechanism of landslides, conduct effective risk assessment and early warning, and thus reduce the losses and hazards caused by landslides.

[0054] Step 2: Preprocess the collected slope-related data and meteorological data, then use principal component analysis to extract features from the preprocessed data to obtain the covariance matrix, and perform eigenvalue decomposition to obtain a new feature matrix;

[0055] In this embodiment, the specific logic for obtaining a new feature matrix is ​​as follows:

[0056] The collected slope-related data and meteorological data were cleaned, and the missing data were processed by the missing value interpolation method. The data with the same timestamp were integrated into a set of observation vectors X. The principal component analysis was used to extract features from the preprocessed data. The formula for obtaining the covariance matrix of the data was as follows:

[0057]

[0058] Where C is the calculated covariance matrix, n is the total number of data samples, X is the observation vector composed of the collected data, and μ is the mean vector of the data;

[0059] Perform eigenvalue analysis on the calculated covariance matrix, and the formula for obtaining the eigenvector is:

[0060] Cv=λv

[0061] Among them, λ is the eigenvalue and v is the corresponding eigenvector;

[0062] The calculated eigenvalues ​​are sorted from large to small, and the eigenvectors corresponding to the first K eigenvalues ​​are selected to form a new matrix. The formula for projecting the new characteristic matrix is:

[0063] Y=XV k

[0064] Among them, Y is the new feature matrix obtained by projection, V k A new matrix consisting of the eigenvectors corresponding to the first K selected eigenvalues.

[0065] The collected data contains noise and missing values. Through data cleaning and interpolation, the accuracy and completeness of the data can be ensured, thus laying the foundation for subsequent analysis. Landslide-related data usually have high dimensions and redundant features. Principal component analysis can extract the most important features from them and reduce the complexity of the data. This helps to improve computational efficiency and reduce model training time. By calculating the covariance matrix, the correlation between the features can be understood, and the factors that have the greatest impact on landslide risk can be identified, thus providing a basis for subsequent risk assessment. In eigenvalue decomposition, sorting eigenvalues ​​can determine the main components and provide the most informative features to describe the data. This helps to remove noise and irrelevant features and enhance the explanatory power of the model. The new feature matrix obtained by projection can effectively simplify the data structure while retaining the main information, making subsequent model training and disaster identification more efficient and accurate.

[0066] Step 3: Build a landslide risk model based on the deep learning model, input the new feature matrix into the model, obtain the landslide risk index as a label, and train the landslide risk model;

[0067] In this embodiment, the specific logic for constructing the landslide risk model is:

[0068] The expert scoring method is used to determine the landslide risk index corresponding to the slope-related data and meteorological data at each timestamp. A landslide risk model is constructed based on the combination of convolutional neural networks and long short-term memory networks. The new feature matrix obtained after principal component analysis is input into the landslide risk model. After the convolution operation is performed through the CNN layer, the state is updated through the LSTM layer. The input layer uses the softmax activation function to output the landslide risk index predicted by the model as the label. The loss function selects the monitoring verification loss. When the verification loss function drops below 0.01 in five consecutive training cycles, the model training is completed.

[0069] By combining convolutional neural networks and long short-term memory networks, the spatial features and time series characteristics in the data can be effectively captured. CNN is good at processing structured data and can identify local associations between features, while LSTM can process time series data and capture the dynamic changes of landslide risk. The new feature matrix processed by principal component analysis is input into the model to ensure that the model training uses high-quality data that has undergone dimensionality reduction and feature extraction. This helps to improve the learning efficiency and prediction accuracy of the model. The landslide risk index is output as a label through the softmax activation function, so that the model can provide intuitive risk assessment results. This probabilistic output form can help decision makers better understand the level of landslide risk and formulate corresponding countermeasures. Using monitoring verification loss as the loss function ensures that the model can evaluate its performance in real time during training. When the verification loss drops below 0.01 in five consecutive training cycles, it indicates that the model has reached a convergence state and has good generalization ability.

[0070] Step 4: Analyze the landslide risk index and the meteorological data of the corresponding timestamp to obtain the landslide risk probability;

[0071] In this embodiment, the specific logic for obtaining the landslide risk probability is:

[0072] The landslide risk index of each grid area is obtained through the model, and the average of the landslide risk index of all grid areas is calibrated as the landslide risk index of the slope. The landslide risk index of the slope and the rainfall, temperature and wind speed data in the same time period are analyzed, and the formula for obtaining the landslide risk probability of the slope is:

[0073]

[0074] Among them, Q is the landslide risk index of the slope, m is the number of grid areas, and q i is the landslide risk index of the i-th grid area output by the model;

[0075]

[0076] Among them, P is the landslide risk probability of the slope, R is the rainfall, T is the temperature, W is the wind speed, e is the nonlinear parameter, k is the adjustment parameter, β1, β2, β3, and β4 are the weight coefficients of each item, β1>β2>β3>β4>0, and β1+β2+β3+β4=1.

[0077] By calculating the landslide risk index and combining it with meteorological data, the probability of landslide risk in a specific area can be quantified, which helps to transform qualitative judgments into quantitative risk assessments, making them easier to understand and use. The occurrence of landslides is often the result of multiple factors. Considering meteorological factors such as rainfall, temperature and wind speed can more comprehensively assess the risk of landslides. This comprehensive analysis can more accurately warn of situations where landslide risks may exist. The use of weighting coefficients and nonlinear parameters allows the model to flexibly adjust the degree of influence of different factors on the probability of landslide risk. This flexibility allows the model to be adjusted accordingly according to actual conditions. Here, e>0, the range is between [0, 10]. When all other factors are zero, the value of e can determine the minimum value of P, thereby preventing the model from estimating a probability value of 0 in extreme cases, thereby improving the accuracy and reliability of the prediction. k>0, the range is between [0, 5]. The value of k is usually used to set the sensitivity of the model. A larger k makes the combination of other parameters (Q, R, T, W) required for P to reach a higher value more demanding, thereby increasing the model's requirements for the "trigger conditions" for landslide occurrence. By adjusting the value of k, the risk probability output by the model can be controlled to a certain extent, making it more consistent with the actual observed frequency of risk events. The obtained landslide risk probability can provide a scientific basis for decision makers to help them formulate corresponding early warning and response strategies. Through real-time monitoring and analysis, early warnings can be issued in time to reduce the threat of landslides to life and property.

[0078] Q represents the landslide risk index of a certain slope, which is the average of the landslide risk indexes of all grid areas. By averaging the landslide risk indexes of multiple small areas (grid areas), the overall risk level of the slope can be obtained. This method takes into account spatial differences, making the risk assessment more comprehensive. The higher the landslide risk index, the greater the landslide risk of the slope, which leads to an increase in the probability of landslide risk of the slope. This is because a high risk index indicates more unstable conditions, which in turn increases the possibility of landslides. When rainfall increases, soil moisture increases, which may cause slope instability, thereby increasing P. High temperatures may cause soil drying and water evaporation, affecting soil stability. In some cases, extreme temperatures may also affect soil structure, thereby affecting landslide risk. Strong winds may cause soil or vegetation above the slope to shift, thereby reducing the stability of the slope and causing P to increase.

[0079] Soil moisture is usually one of the most important factors because it directly affects the shear strength and saturation of the soil. By analyzing soil moisture, groundwater level, horizontal displacement and vertical displacement data, it is very important to monitor whether there is a landslide disaster on the slope. Therefore, the weight of the landslide risk index obtained by analyzing these four sets of data is the highest. Rainfall is an important cause of landslides, especially when the rainfall intensity and duration are large. Therefore, its weight is second only to the landslide risk index. The impact of temperature on landslide risk is relatively small, usually related to the evaporation of soil moisture and the freeze-thaw cycle, so its weight is lower than the weight of rainfall. The direct impact of wind speed on landslides is usually small, and its weight coefficient is the lowest, and can even be ignored in some cases. The constraint condition of β1+β2+β3+β4=1 ensures that the filling hazard index is a normalized indicator, ensuring that the total contribution of the weighted combination of the three factors is 100%.

[0080] Step 5: Collect slope-related data and meteorological data in real time, execute step 2 to obtain the real-time feature matrix, input the real-time feature matrix into the trained landslide risk model, obtain the landslide risk index output by the model, calculate the landslide risk probability at the current moment, and compare it with the preset risk threshold to judge the current landslide safety status.

[0081] In this embodiment, the specific logic for judging the current landslide safety status is as follows:

[0082] Collect soil moisture, groundwater level, horizontal displacement and vertical displacement data of the landslide at the current moment, perform data preprocessing, obtain the real-time feature matrix, input the real-time feature matrix into the model, output the landslide risk index at the current moment, and combine the rainfall, temperature and wind speed data at this moment to obtain the landslide risk probability at the current moment, compare the landslide risk probability at the current moment with the preset risk threshold, if the landslide risk probability at the current moment is higher than the preset risk threshold, issue a landslide risk warning, if the landslide risk probability at the current moment is lower than the preset risk threshold, continue monitoring.

[0083] The value of the risk threshold here is the lowest landslide risk probability when a landslide disaster occurs in historical data. By collecting slope-related data in real time, such as soil moisture, groundwater level, horizontal displacement and vertical displacement, the dynamic factors affecting landslides can be obtained in a timely manner. The real-time collected data is input into the trained landslide risk model, and the landslide risk index and probability at the current moment are output, so that the risk assessment is not only based on historical data, but also reflects the current actual situation. By comparing the landslide risk probability at the current moment with the preset risk threshold, an effective early warning mechanism can be formed. When the risk probability is higher than the threshold, a landslide risk warning is issued in a timely manner, so that corresponding preventive measures can be taken to reduce the risk of disasters.

[0084] See also Figure 2 The present invention further provides a landslide hazard identification system based on multi-source data fusion, wherein the landslide hazard identification system based on multi-source data fusion is used to implement the above-mentioned landslide hazard identification method based on multi-source data fusion, and comprises:

[0085] A data acquisition module is used to arrange relevant sensors at multiple locations of the landslide to obtain slope-related data, including soil moisture, groundwater level, horizontal displacement and vertical displacement data, and obtain meteorological data at the corresponding time according to the timestamp, and the meteorological data includes rainfall, temperature and wind speed data;

[0086] The data preprocessing and feature extraction module is used to preprocess the collected slope-related data and meteorological data, and then use principal component analysis to extract features from the preprocessed data to obtain the covariance matrix, and perform eigenvalue decomposition to obtain a new feature matrix;

[0087] The landslide risk model construction module is used to build a landslide risk model based on the deep learning model, input a new feature matrix into the model, obtain the landslide risk index as a label, and train the landslide risk model;

[0088] The landslide risk probability acquisition module is used to analyze the landslide risk index and the meteorological data of the corresponding timestamp to obtain the landslide risk probability;

[0089] The real-time monitoring and risk assessment module is used to collect slope-related data and meteorological data in real time, execute the data preprocessing and feature extraction module to obtain the real-time feature matrix, input the real-time feature matrix into the trained landslide risk model, obtain the landslide risk index output by the model, calculate the landslide risk probability at the current moment, and compare it with the preset risk threshold to judge the current landslide safety status.

[0090] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0091] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0092] 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, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0093] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. The landslide hazard identification method based on multi-source data fusion is characterized by: The specific steps include: Step 1: Arrange relevant sensors at multiple locations of the landslide to obtain slope-related data, including soil moisture, groundwater level, horizontal displacement and vertical displacement data, and obtain meteorological data at the corresponding time according to the timestamp, wherein the meteorological data includes rainfall, temperature and wind speed data; Step 2: Preprocess the collected slope-related data and meteorological data, then use principal component analysis to extract features from the preprocessed data to obtain the covariance matrix, and perform eigenvalue decomposition to obtain a new feature matrix; Step 3: Build a landslide risk model based on the deep learning model, input the new feature matrix into the model, obtain the landslide risk index as a label, and train the landslide risk model; Step 4: Analyze the landslide risk index and the meteorological data of the corresponding timestamp to obtain the landslide risk probability; Step 5: Collect slope-related data and meteorological data in real time, execute step 2 to obtain the real-time feature matrix, input the real-time feature matrix into the trained landslide risk model, obtain the landslide risk index output by the model, calculate the landslide risk probability at the current moment, and compare it with the preset risk threshold to judge the current landslide safety status.

2. The landslide hazard identification method based on multi-source data fusion according to claim 1 is characterized in that: The specific logic for collecting slope-related data and meteorological data is as follows: The landslide area was divided into multiple monitoring grid areas of 5m×5m in size. Humidity sensors and displacement meters were arranged at the center of each grid to obtain soil moisture, horizontal displacement and vertical displacement data of the monitoring point. A water level sensor was installed in the groundwater level monitoring well at the bottom edge of the landslide to obtain groundwater level data. The collected slope-related data were marked with timestamps, and the meteorological data at the same timestamp were recorded.

3. The landslide hazard identification method based on multi-source data fusion according to claim 2 is characterized in that: The specific logic for obtaining a new feature matrix is ​​as follows: The collected slope-related data and meteorological data were cleaned, and the missing data were processed by the missing value interpolation method. The data with the same timestamp were integrated into a set of observation vectors X. The principal component analysis was used to extract features from the preprocessed data. The formula for obtaining the covariance matrix of the data was as follows: Where C is the calculated covariance matrix, n is the total number of data samples, X is the observation vector composed of the collected data, and μ is the mean vector of the data; Perform eigenvalue analysis on the calculated covariance matrix, and the formula for obtaining the eigenvector is: Cv=λv Among them, λ is the eigenvalue and v is the corresponding eigenvector; The calculated eigenvalues ​​are sorted from large to small, and the eigenvectors corresponding to the first K eigenvalues ​​are selected to form a new matrix. The formula for projecting the new characteristic matrix is: Y=XV k Among them, Y is the new feature matrix obtained by projection, V k A new matrix consisting of the eigenvectors corresponding to the first K eigenvalues ​​selected.

4. The landslide hazard identification method based on multi-source data fusion according to claim 3 is characterized in that: The specific logic for constructing the landslide risk model is as follows: A landslide risk model is constructed based on the combination of convolutional neural network and long short-term memory network. The new feature matrix obtained after principal component analysis is input into the landslide risk model. After convolution operation through the CNN layer, the state is updated through the LSTM layer. The input layer uses the softmax activation function to output the landslide risk index predicted by the model as the label. The loss function selects monitoring verification loss. When the verification loss function drops below 0.01 in five consecutive training cycles, the model training is completed.

5. The landslide hazard identification method based on multi-source data fusion according to claim 4 is characterized in that: The specific logic for obtaining the landslide risk probability is as follows: The landslide risk index of each grid area is obtained through the model, and the average of the landslide risk index of all grid areas is calibrated as the landslide risk index of the slope. The landslide risk index of the slope and the rainfall, temperature and wind speed data at the same time stamp are analyzed, and the formula for obtaining the landslide risk probability of the slope is: Among them, Q is the landslide risk index of the slope, m is the number of grid areas, and q i is the landslide risk index of the i-th grid area output by the model; Among them, P is the landslide risk probability of the slope, R is the rainfall, T is the temperature, W is the wind speed, e is the nonlinear parameter, k is the adjustment parameter, β1, β2, β3, and β4 are the weight coefficients of each item, β1>β2>β3>β4>0, and β1+β2+β3+β4=1.

6. The landslide hazard identification method based on multi-source data fusion according to claim 5 is characterized in that: The specific logic for judging the current landslide safety status is as follows: Collect soil moisture, groundwater level, horizontal displacement and vertical displacement data of the landslide at the current moment, perform data preprocessing, obtain the real-time feature matrix, input the real-time feature matrix into the model, output the landslide risk index at the current moment, and combine the rainfall, temperature and wind speed data at this moment to obtain the landslide risk probability at the current moment, compare the landslide risk probability at the current moment with the preset risk threshold, if the landslide risk probability at the current moment is higher than the preset risk threshold, issue a landslide risk warning, if the landslide risk probability at the current moment is lower than the preset risk threshold, continue monitoring.

7. The landslide hazard identification system based on multi-source data fusion is characterized by: The landslide hazard identification system based on multi-source data fusion is used to implement the landslide hazard identification method based on multi-source data fusion according to any one of claims 1 to 6, comprising: A data acquisition module is used to arrange relevant sensors at multiple locations of the landslide to obtain slope-related data, including soil moisture, groundwater level, horizontal displacement and vertical displacement data, and obtain meteorological data at the corresponding time according to the timestamp, and the meteorological data includes rainfall, temperature and wind speed data; The data preprocessing and feature extraction module is used to preprocess the collected slope-related data and meteorological data, and then use principal component analysis to extract features from the preprocessed data to obtain the covariance matrix, and perform eigenvalue decomposition to obtain a new feature matrix; The landslide risk model construction module is used to build a landslide risk model based on the deep learning model, input a new feature matrix into the model, obtain the landslide risk index as a label, and train the landslide risk model; The landslide risk probability acquisition module is used to analyze the landslide risk index and the meteorological data of the corresponding timestamp to obtain the landslide risk probability; The real-time monitoring and risk assessment module is used to collect slope-related data and meteorological data in real time, execute the data preprocessing and feature extraction module to obtain the real-time feature matrix, input the real-time feature matrix into the trained landslide risk model, obtain the landslide risk index output by the model, calculate the landslide risk probability at the current moment, and compare it with the preset risk threshold to judge the current landslide safety status.

Images