Landslide prediction method and system based on multi-source data fusion

Abstract

The invention belongs to the technical field of landslide prediction, and discloses a landslide prediction method and system based on multi-source data fusion, and the method comprises the steps: collecting multi-source heterogeneous landslide related data, carrying out the preprocessing of the landslide related data through a data fusion technology, and obtaining a landslide data set; performing noise reduction on the landslide data set; mining nonlinear correlation characteristics in the landslide data set after noise reduction, and constructing high-order statistics capable of depicting a landslide evolution process; constructing a plurality of prediction sub-models according to the high-order statistics, and fusing the prediction sub-models to form a landslide prediction model; and inputting the to-be-measured high-order statistical magnitude into the landslide prediction model, and if the to-be-measured high-order statistical magnitude exceeds a preset danger threshold, determining that a landslide risk exists. The method has high universality and practicability, and can be widely applied to the field of monitoring and early warning of geological disasters such as landslide.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of landslide prediction, in particular to a landslide prediction method and system based on multi-source data fusion. BACKGROUND

[0002] Landslide prediction is a complex process that needs to consider multiple influencing factors. In actual landslide prediction, there are often problems such as multiple data sources, inconsistent data formats, and large data volumes. How to effectively integrate and utilize these multi-source heterogeneous data is a technical problem that needs to be solved. In addition, there may be correlations and redundancies between different data. Directly using raw data for modeling and prediction not only increases the computational complexity, but also may affect the prediction accuracy. Therefore, how to extract the most effective features from the original multi-source data and remove the interference of redundancy and noise is a big challenge in the data preprocessing stage. At the same time, landslide prediction itself is a nonlinear and complex process. It is difficult to well fit the internal laws of landslide occurrence by using only traditional statistical methods. How to build a landslide prediction model with superior performance, while ensuring the prediction accuracy and the generalization ability of the model, is a technical problem that has been explored in this field. In addition, since the occurrence of landslide disasters is often sudden and destructive, higher requirements are put forward for the real-time performance and robustness of the prediction model. SUMMARY

[0003] In order to solve the above technical problems, the present application provides a landslide prediction method and system based on multi-source data fusion, which realizes intelligent, accurate and interpretable early warning of landslide risk through the organic combination of data fusion, feature extraction, ensemble learning, incremental updating and knowledge reasoning, and provides strong technical support for landslide disaster prevention. The present application has strong universality and practicality, and can be widely applied in the field of monitoring and early warning of landslide and other geological disasters.

[0004] The present application discloses a landslide prediction method based on multi-source data fusion, which comprises:

[0005] Collecting multi-source heterogeneous landslide-related data, using data fusion technology to preprocess the landslide-related data, and obtaining a landslide data set;

[0006] De-noising the landslide data set;

[0007] Mining the inherent nonlinear correlation characteristics of the de-noised landslide data set, and constructing high-order statistics capable of describing the evolution process of landslide;

[0008] According to the high-order statistics, a plurality of prediction sub-models are constructed and fused to form a landslide prediction model;

[0009] The to-be-tested high-order statistics is input into the landslide prediction model, and if the to-be-tested high-order statistics exceeds a preset danger threshold, it is determined that there is a landslide risk.

[0010] Preferably, the data fusion technology is used to preprocess the landslide-related data to obtain a landslide data set, which includes:

[0011] According to the sources and formats of the multi-source heterogeneous landslide-related data, different categories of landslide data subsets are obtained;

[0012] According to each landslide data subset, a corresponding data preprocessing method is used to clean, convert and standardize the data, and eliminate data redundancy and inconsistency;

[0013] The preprocessed landslide data subsets are fused, and a unified format of landslide data is generated according to a preset data model and fusion rule;

[0014] The fused landslide data is quality evaluated to obtain a landslide data set.

[0015] Preferably, the noise reduction of the landslide data set includes:

[0016] The wavelet transform is performed on the landslide data set to map the data to wavelet basis functions of different scales to realize multi-resolution analysis in time-frequency domain;

[0017] According to the amplitude of the wavelet transform coefficient, a suitable threshold is set to perform threshold processing on the wavelet coefficient to filter out high-frequency noise components with small amplitudes;

[0018] The wavelet coefficient after threshold processing is inverse wavelet transformed to reconstruct the noise-reduced landslide data;

[0019] The singular value decomposition technology is used to decompose the noise-reduced landslide data matrix into a product form of a left singular matrix, a singular value matrix and a right singular matrix;

[0020] According to the size of the singular value, the rank of the data matrix is determined, and the singular vectors corresponding to the singular values below the preset requirement are discarded to remove the nonlinear interference components in the data;

[0021] The data matrix is reconstructed using the retained singular values and singular vectors to obtain a landslide data set after removing random disturbance and nonlinear interference.

[0022] Preferably, the inherent nonlinear correlation characteristics of the noise-reduced landslide data set are mined to construct high-order statistics capable of describing the evolution process of the landslide, including:

[0023] According to the filtered monitoring data, a nonlinear time series analysis method is used to mine the inherent nonlinear correlation characteristics of the data;

[0024] According to the nonlinear correlation characteristics inherent in the data, the surface displacement data are analyzed, and a high-order statistical quantity capable of describing the landslide evolution process is constructed through a nonlinear dynamic model.

[0025] Preferably, the plurality of prediction sub-models are constructed and fused to form the landslide prediction model, which comprises:

[0026] By analyzing the displacement evolution characteristics of landslides under the coupling action of rainfall and reservoir water, the dynamic response law of deformation in different stages of landslides to reservoir water level changes and rainfall is obtained, so as to form the understanding of the physical meaning of each displacement component, the importance of displacement influencing factors and the hysteresis;

[0027] The importance and hysteresis of the influencing factors are quantitatively analyzed and expressed by combining the displacement addition model, the wavelet denoising, the VMD decomposition and the grey correlation analysis method, so as to prepare displacement components and influencing factors meeting the preset requirements for the prediction model;

[0028] After model training, effect evaluation and optimization, the step-type landslide displacement is efficiently predicted, wherein the prediction of the trend term and the periodic term displacement component is realized by using the double exponential smoothing method and the Informer model respectively, and the prediction result of the cumulative displacement is obtained by adding the predicted values of each component.

[0029] Preferably, the method further comprises: using an incremental learning mechanism to update and adjust the landslide prediction model online, when a new landslide event data is monitored, the new landslide event data is added to the training set, and the landslide prediction model weight is fine-tuned through an optimization algorithm, so that the landslide prediction model can adapt to the changing data distribution and capture new patterns of landslide occurrence.

[0030] Preferably, the method further comprises: introducing the causal reasoning and knowledge graph technology, constructing a landslide risk assessment framework containing multi-field knowledge of geology, meteorology and hydrology, determining the logical dependency relationship between different factors through causal relationship mining, combining expert knowledge to constrain and correct the prediction result, and forming a comprehensive landslide early warning decision scheme.

[0031] The application also discloses a landslide prediction system based on multi-source data fusion, which is used to realize any one of the methods, and comprises a fusion module, a denoising module, a mining module, a construction module and a prediction module.

[0032] The fusion module is used for collecting multi-source heterogeneous landslide related data, and pre-processing the landslide related data by using a data fusion technology to obtain a landslide data set.

[0033] The denoising module is used for denoising the landslide data set.

[0034] The mining module is used for mining the nonlinear correlation characteristics inherent in the landslide data set after noise reduction, and constructing high-order statistics capable of depicting the landslide evolution process;

[0035] The constructing module is used for constructing a plurality of prediction sub-models and fusing to form a landslide prediction model according to the high-order statistics;

[0036] The prediction module is used for inputting the to-be-tested high-order statistics into the landslide prediction model, and if the to-be-tested high-order statistics exceeds a preset danger threshold, it is determined that there is a landslide risk.

[0037] Compared with the prior art, the beneficial effects of the present application are:

[0038] The application discloses a landslide prediction method and system based on multi-source data fusion. First, multi-source landslide related data is fused and preprocessed, noise is reduced by using wavelet analysis and other technologies, and high-order statistical features are extracted. Then, an ensemble learning model is constructed for landslide prediction, and an incremental learning mechanism is introduced to realize online updating of the model. To improve prediction accuracy and interpretability, the application also fuses causal reasoning and knowledge graph technology to construct a landslide risk assessment framework of multi-domain knowledge. Through the organic combination of data fusion, feature extraction, ensemble learning, incremental updating and knowledge reasoning technologies, the application realizes intelligent, accurate and interpretable early warning of landslide risk, providing strong technical support for landslide disaster prevention. The method has strong universality and practicality, and can be widely applied to the monitoring and early warning field of landslides and other geological disasters. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed to be used in the embodiments. Obviously, the drawings described in the following embodiments are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0040] Figure 1 A landslide prediction method based on multi-source data fusion according to an embodiment of the present application;

[0041] Figure 2 The whole structure of the Informer model according to an embodiment of the present application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0043] It should be noted that, unless otherwise defined, technical terms or scientific terms used in the embodiments of the present disclosure shall have the common meaning understood by one of ordinary skill in the art to which the present disclosure belongs. The terms "first", "second", and similar terms used in the embodiments of the present disclosure do not denote any order, quantity, or importance, but are used to distinguish different components. The terms "include", "contain", and similar terms mean that the elements or objects before the terms encompass the elements or objects listed after the terms and their equivalents, and do not exclude other elements or objects. The terms "connect" or "connected" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "upper", "lower", "left", "right", and the like are used only to represent relative positional relationships, and when the absolute positions of the described objects change, the relative positional relationships can also change accordingly.

[0044] In order to make the above objectives, characteristics and advantages of the present application more apparent, comprehensible and easier to understand, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Embodiment one

[0046] As shown in Figure 1 , the embodiment of the present application provides a landslide prediction method based on multi-source data fusion, the method comprising:

[0047] Collecting multi-source heterogeneous landslide related data, using data fusion technology to preprocess the landslide related data to obtain a landslide data set;

[0048] De-noising the landslide data set;

[0049] Mining the inherent nonlinear correlation characteristics of the de-noised landslide data set, and constructing high-order statistics capable of depicting the evolution process of the landslide;

[0050] According to the high-order statistics, a plurality of prediction sub-models are constructed and fused to form a landslide prediction model;

[0051] Inputting the to-be-tested high-order statistics into the landslide prediction model, if the to-be-tested high-order statistics exceed a preset danger threshold, it is determined that there is a landslide risk.

[0052] In the present embodiment, the data fusion technology is used to preprocess the landslide related data to obtain a landslide data set, which includes:

[0053] According to the source and format of the multi-source heterogeneous landslide related data, different categories of landslide data subsets are obtained;

[0054] According to each landslide data subset, a corresponding data preprocessing method is adopted to clean, convert and standardize the data, and eliminate data redundancy and inconsistency;

[0055] The preprocessed landslide data subsets are fused, and a unified format of landslide data is generated according to a preset data model and fusion rule;

[0056] The fused landslide data is quality evaluated to obtain a landslide data set.

[0057] Specifically, landslide data sources are diverse, including remote sensing images, geological surveys, rainfall records, etc. Taking remote sensing images as an example, they can be divided into optical images and radar images. Optical images can obtain surface shape changes, and radar images can detect surface deformation. Geological survey data includes rock and soil parameters, underground water level, etc., and rainfall records provide rainfall, duration, etc. For remote sensing images, geometric correction and atmospheric correction are needed to eliminate image distortion and atmospheric effects. For geological survey data, outliers such as data that deviate significantly from the normal range of rock and soil strength parameters need to be removed. Rainfall data needs to be time series interpolated to fill in missing values. During data fusion, different source data can be unified into the same spatio-temporal reference system. For example, remote sensing images, geological maps, and rainfall distribution maps can be superimposed on the same coordinate system to achieve spatial correspondence of multi-source data. In the time dimension, monitor data at different times are aligned to a unified time sequence. Quality evaluation uses multiple indicators, such as data completeness evaluation of missing value proportion and accuracy evaluation of deviation from measured values. For example, if the coverage rate of a monitoring station in a certain area is less than sixty percent, temporary stations need to be added or interpolation methods need to be used to supplement.

[0058] In this embodiment, the noise reduction of the landslide data set includes:

[0059] By wavelet transform of the landslide data set, the data is mapped to the wavelet basis function of different scales to realize multi-resolution analysis in time-frequency domain;

[0060] According to the amplitude of the wavelet transform coefficient, a suitable threshold is set to process the wavelet coefficient by thresholding to filter out high-frequency noise components with small amplitude;

[0061] Inverse wavelet transform is performed on the wavelet coefficients after thresholding to reconstruct the noise-reduced landslide data;

[0062] The singular value decomposition technique is used to decompose the noise-reduced landslide data matrix into the product form of the left singular matrix, the singular value matrix and the right singular matrix;

[0063] According to the size of the singular value, the rank of the data matrix is determined, and the singular vectors corresponding to the singular values below the preset requirement are discarded to remove the nonlinear interference components in the data;

[0064] The data matrix is reconstructed by using the reserved singular values and singular vectors to obtain a landslide data set after removing random disturbances and nonlinear interference.

[0065] In this embodiment, the nonlinear correlation characteristics inherent in the denoised landslide data set are mined, and high-order statistics capable of depicting the evolution process of the landslide are constructed, including:

[0066] According to the filtered monitoring data, a nonlinear time series analysis method is used to mine the inherent nonlinear correlation characteristics of the data.

[0067] According to the inherent nonlinear correlation characteristics of the data, the surface displacement data is analyzed, and a nonlinear dynamic model is used to construct high-order statistics capable of depicting the evolution process of the landslide.

[0068] Specifically, for the filtered monitoring data, a nonlinear time series analysis method is used to mine the inherent nonlinear correlation characteristics of the data. Assuming that the surface displacement data is analyzed, it is found that the displacement rate shows exponential growth in a certain time period through a nonlinear dynamic model, which may be a landslide precursor. High-order statistics such as Lyapunov exponent are constructed to quantify this nonlinear feature. According to expert knowledge and historical landslide cases, the dangerous threshold of high-order statistics is preset. For example, if the Lyapunov exponent exceeds 0.5, it indicates that the system stability is very poor and the landslide risk is high. Real-time calculation of the high-order statistics corresponding to the current monitoring data, if it exceeds the threshold, a landslide warning signal is issued. When the landslide warning signal is monitored, the emergency plan is immediately started. According to the warning level, corresponding measures are taken, such as evacuating residents, blocking roads, etc. Assuming that the warning level is three, first, the nearby residents are notified to evacuate, then the relevant departments are notified to conduct on-site investigation, and finally the emergency supplies are started. The changes of the monitoring data and the high-order statistics of the landslide area are continuously tracked, and the dangerous threshold is dynamically adjusted. For example, during continuous rainfall, the threshold may need to be lowered because the soil moisture content increases and the landslide risk rises. Comprehensive analysis of regional geological environmental conditions, rainfall conditions and other influencing factors, combined with the trend of high-order statistics, the development and evolution process of the landslide is predicted.

[0069] In this embodiment, a plurality of prediction sub-models are constructed and fused to form a landslide prediction model, including:

[0070] Through the analysis of the displacement evolution characteristics of landslides under the coupling action of rainfall and reservoir water, the dynamic response law of deformation at different stages of landslides and changes in reservoir water level and rainfall is obtained, so as to form the understanding of the physical significance of each displacement component, the importance and hysteresis of displacement influencing factors;

[0071] Combining the displacement addition model, wavelet denoising, VMD decomposition and grey relational analysis methods, the importance and hysteresis of influencing factors are quantitatively analyzed and expressed, and displacement components and influencing factors that meet the preset requirements are prepared for the prediction model.

[0072] After model training, effect evaluation and optimization, efficient prediction of step-type landslide displacement is achieved. Among them, the prediction of trend term and periodic term displacement components is realized by using double exponential smoothing method and Informer model respectively. The prediction result of cumulative displacement is obtained by adding up the predicted values ​​of each component.

[0073] Specifically, the temporal deformation characteristics of a landslide are the result of the combined effects of internal geological conditions (stress field, rock and soil structure, etc.), external triggering factors (rainfall, water storage and drainage, etc.), and random disturbances (seismic waves, wind loads, etc.). The internal force field and rock and soil structure of the landslide primarily influence the long-term deformation trend of the landslide. Seasonal rainfall and water storage and drainage cause periodic deformation of the landslide, while random factors cause slight random deformation of the landslide. Therefore, the landslide displacement can be expressed using an additive model as:

[0074] f t =T t +P t +R t (1)

[0075] Where, f t represents the cumulative displacement at time t; T t represents the trend term displacement at time t; P t represents the periodic displacement at time t; R t represents the random deformation at time t. Since the random disturbance factors of the landslide body are not easy to capture and have little effect on the landslide displacement, most studies tend to ignore the random deformation. At the same time, considering the physical meaning of each component of the landslide cumulative displacement, this paper removes the random deformation R from Equation (1) by wavelet denoising. t The denoised displacement sequence is decomposed into trend term displacement and period term displacement by variational mode decomposition algorithm.

[0076] Double exponential smoothing (DES) is a time series analysis method for predicting future trends, and is suitable for time series without periodic or seasonal fluctuations. Considering that the trend component of landslide displacement is mainly driven by internal factors such as geological conditions, rather than external factors, the trend component of landslide mainly shows a continuous growth over time, without obvious periodic terms or seasonal fluctuations. Therefore, DES, as a single variable prediction model, is particularly suitable for predicting the trend displacement of landslide development through historical time series. Assume that a landslide trend displacement series T is given. tThe key formula of the DES algorithm is:

[0077] S t = aT t + (1-a) (S t-1 +b t-1 ) 0≤a≤1 (2)

[0078] b t = g (S t -S t-1 ) + (1-g) b t-1 0≤g≤1 (3)

[0079] Y t+1 = S t +b t (4)

[0080] In the formula, Y t represents the exponential smoothing value at time t; b t represents the gradient optimal estimate at time t; a and g are smoothing factors, Y t+1 is the predicted value at time t+1; and the initial values of S t and b t are generally:

[0081]

[0082] The overall structure of the Informer model used in this paper is shown in Figure 2 , wherein the "input vector_enc" and the "input vector_dec" represent the inputs of the Encoder and the Decoder respectively, the "Key" and the "Value" participating in the multi-head confusion attention mechanism calculation in the Decoder are projected from the "concatenated feature map" generated by the Encoder, the "Query" is projected from the result of the normalization layer in the Decoder, and multiple layers of the Encoder and the Decoder can be set in the model.

[0083] In this embodiment, the method further comprises: using an incremental learning mechanism to perform online updating and adjusting on the landslide prediction model, when a new landslide event data is monitored, adding the new landslide event data into the training set, and fine-tuning the landslide prediction model weight through an optimization algorithm, so that the landslide prediction model can adapt to the changing data distribution and capture new patterns of landslide occurrence.

[0084] Specifically, the landslide monitoring system acquires various sensor data and environmental parameters in real time, pre-processes the data to eliminate outliers and noise interference, and extracts key feature indicators. According to the pre-established landslide prediction model, the real-time pre-processed feature data is input into the model for prediction and reasoning to determine whether there is a landslide risk and the risk level. If the prediction result shows that there is a landslide risk and exceeds the preset warning threshold, the early warning mechanism is triggered, and warning information is sent to relevant departments, and the emergency plan is started. When a new landslide event data is monitored, it is labeled as a new training sample, the original training dataset is updated, and gradient descent and other optimization algorithms are used for incremental learning and parameter fine-tuning of the model. The performance indicators of the updated model, such as accuracy and recall rate, are evaluated by cross-validation and other methods. If the performance improvement meets the preset conditions, the updated model is deployed to the online environment. The online prediction effect of the model is continuously tracked, user feedback and actual landslide situations are collected, and samples with prediction errors or missed reports are analyzed for errors and included in the next round of incremental learning training set. The model is regularly evaluated and optimized, and feature selection, parameter tuning and other techniques are used to improve the generalization ability and robustness of the model to adapt to landslide prediction requirements in different regions and scenarios.

[0085] In this embodiment, the method further includes introducing causal reasoning and knowledge graph technology, building a landslide risk assessment framework containing geological, meteorological, and hydrological multi-disciplinary knowledge, determining the logical dependency relationship between different factors through causal relationship mining, combining expert knowledge to constrain and correct the prediction results, and forming a comprehensive landslide early warning decision scheme.

[0086] Specifically, it includes the following steps: Step 1: Construct a landslide risk assessment knowledge graph, and ontologize and semantically associate multi-disciplinary knowledge such as geology, meteorology, and hydrology to form a structured knowledge base;

[0087] Step 2: Use causal reasoning algorithms such as Bayesian networks or causal forests to mine the causal relationships between different factors from the knowledge graph and determine their logical dependencies;

[0088] Step 3: According to the mined causal relationships, build a landslide risk assessment model that considers the influence of each factor on landslide occurrence;

[0089] Step 4: Acquire real-time geological, meteorological, and hydrological monitoring data and input it into the landslide risk assessment model to predict the probability of landslide occurrence;

[0090] Step 5: Compare the prediction results with the expert knowledge in the knowledge graph to determine whether there is deviation or abnormality, and if necessary, constrain and correct it;

[0091] Step 6: The landslide early warning decision scheme is obtained by combining the prediction results of the integrated model and the correction results of the expert knowledge, including early warning level, influence range, and emergency measures, etc.

[0092] Step 7: The landslide early warning decision scheme is presented to the decision makers and the public in an intuitive and easy-to-understand way through visualization technology, improving the interpretability and operability of the early warning information.

[0093] It should be noted that the method of the embodiments of the present disclosure can be executed by a single device, such as a computer or a server, etc. The method of the embodiments can also be applied to a distributed scenario, and completed by multiple devices cooperating with each other. In this distributed scenario, one of the multiple devices can only execute one or more steps in the method of the embodiments of the present disclosure, and the multiple devices interact with each other to complete the method.

[0094] It should be noted that some embodiments of the present disclosure have been described above. Other embodiments are within the scope of the appended claims. In some cases, it should be understood that the size of the serial number of each step in the above embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure. The actions or steps recited in the claims can be executed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0095] Embodiment Two

[0096] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also discloses a landslide prediction system based on multi-source data fusion, which is used to implement any of the methods, and the system comprises a fusion module, a noise reduction module, a mining module, a construction module, and a prediction module.

[0097] The fusion module is used to collect multi-source heterogeneous landslide-related data, and uses data fusion technology to pre-process the landslide-related data to obtain a landslide data set;

[0098] The noise reduction module is used to reduce the noise of the landslide data set;

[0099] The mining module is used to mine the inherent nonlinear correlation characteristics of the landslide data set after noise reduction, and construct high-order statistics capable of describing the landslide evolution process;

[0100] The construction module is used to construct a plurality of prediction sub-models according to the high-order statistics and fuse them to form a landslide prediction model;

[0101] The prediction module is configured to input the to-be-tested high-order statistic into a landslide prediction model, and determine that there is a landslide risk if the to-be-tested high-order statistic exceeds a preset danger threshold.

[0102] The system of the above embodiments is configured to implement the landslide prediction method based on multi-source data fusion of any one of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be described herein again.

[0103] It should be noted that the above landslide prediction system based on multi-source data fusion is embodied in the form of a functional unit. The term "module" herein can be implemented in the form of software and / or hardware, and is not specifically limited.

[0104] For example, the "module" can be a software program, a hardware circuit, or a combination of both, which implements the above functions. The hardware circuit can include an application specific integrated circuit (ASIC), an electronic circuit, a processor (such as a shared processor, a dedicated processor, or a group processor, etc.) and a memory for executing one or more software or firmware programs, a combination logic circuit, and / or other suitable components supporting the described functions.

[0105] The embodiments of the present disclosure are intended to cover all such alternatives, modifications, and variations of the present disclosure falling within the broadest scope of the appended claims. Therefore, any omission, modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A landslide prediction method based on multi-source data fusion, characterized in that: The method comprises: Collect multi-source heterogeneous landslide-related data, use data fusion technology to pre-process the landslide-related data, and obtain the landslide dataset; Denoising of landslide datasets; Mining the inherent nonlinear correlation characteristics of the denoised landslide dataset and constructing high-order statistics that can characterize the landslide evolution process; Based on high-order statistics, multiple prediction sub-models are constructed and integrated to form a landslide prediction model; The high-order statistics to be measured are input into the landslide prediction model. If the high-order statistics to be measured exceed the preset danger threshold, it is determined that there is a landslide risk.

2. The method according to claim 1, characterized in that Data fusion technology is used to pre-process the landslide related data, and the landslide data set obtained includes: According to the sources and formats of multi-source heterogeneous landslide-related data, different types of landslide data subsets are obtained; According to each landslide data subset, the corresponding data preprocessing method is used to clean, transform and standardize the data to eliminate data redundancy and inconsistency; The pre-processed landslide data subsets are fused to generate landslide data in a unified format according to the preset data model and fusion rules; The quality of the fused landslide data was assessed to obtain the landslide dataset.

3. The method according to claim 1, characterized in that Denoising the landslide dataset involves: By performing wavelet transform on the landslide dataset, the data is mapped to wavelet basis functions of different scales to achieve multi-resolution analysis in the time-frequency domain. According to the amplitude of the wavelet transform coefficient, set a suitable threshold, perform threshold processing on the wavelet coefficient, and filter out high-frequency noise components with smaller amplitude; Perform inverse wavelet transform on the wavelet coefficients after threshold processing to reconstruct the landslide data after noise reduction; The singular value decomposition technique is used to decompose the landslide data matrix after noise reduction into the product of the left singular matrix, the singular value matrix and the right singular matrix. According to the size of the singular value, the rank of the data matrix is ​​determined, and the singular vectors corresponding to the singular values ​​below the preset requirements are discarded to remove the nonlinear interference components in the data; The data matrix is ​​reconstructed using the retained singular values ​​and singular vectors to obtain the landslide dataset after removing random disturbances and nonlinear interference.

4. The method according to claim 1, wherein By mining the inherent nonlinear correlation characteristics of the denoised landslide dataset and constructing high-order statistics that can characterize the landslide evolution process, the following are included: Based on the filtered monitoring data, the nonlinear time series analysis method is used to mine the inherent nonlinear correlation characteristics of the data; Based on the inherent nonlinear correlation characteristics of the data, the surface displacement data are analyzed, and through the nonlinear dynamic model, high-order statistics that can characterize the landslide evolution process are constructed.

5. The method according to claim 1, wherein Constructing multiple prediction sub-models and integrating them to form a landslide prediction model includes: By analyzing the evolution characteristics of landslide displacement under the coupling of rainfall and reservoir water, the dynamic response patterns of landslide deformation at different stages to reservoir water level fluctuations and rainfall were obtained, thereby forming an understanding of the physical significance of each displacement component, the importance of displacement influencing factors, and the hysteresis. Combining the displacement addition model, wavelet denoising, VMD decomposition and grey relational analysis methods, the importance and hysteresis of influencing factors are quantitatively analyzed and expressed, and displacement components and influencing factors that meet the preset requirements are prepared for the prediction model. After model training, effect evaluation and optimization, efficient prediction of step-type landslide displacement is achieved. Among them, the prediction of trend term and periodic term displacement components is realized by using double exponential smoothing method and Informer model respectively. The prediction result of cumulative displacement is obtained by adding up the predicted values ​​of each component.

6. The method according to claim 1, characterized in that This method also includes: using an incremental learning mechanism to update and adjust the landslide prediction model online. When new landslide event data is monitored, the new landslide event data is added to the training set, and the landslide prediction model weights are fine-tuned through an optimization algorithm to enable the landslide prediction model to adapt to the ever-changing data distribution and capture new patterns of landslide occurrence.

7. The method according to claim 1, characterized in that This method also includes: introducing causal reasoning and knowledge graph technology to construct a landslide risk assessment framework that includes knowledge from multiple fields such as geology, meteorology, and hydrology. Through causal relationship mining, the logical dependencies between different factors are determined. In combination with expert knowledge, the prediction results are constrained and corrected to form a comprehensive landslide early warning decision-making plan.

8. A landslide prediction system based on multi-source data fusion, the system being used to implement the method according to any one of claims 1 to 7, characterized in that: The system includes: a fusion module, a noise reduction module, a mining module, a construction module and a prediction module; The fusion module is used to collect multi-source heterogeneous landslide-related data, and pre-process the landslide-related data using data fusion technology to obtain a landslide data set; The noise reduction module is used to reduce noise on the landslide dataset; The mining module is used to mine the nonlinear correlation characteristics inherent in the denoised landslide data set and construct high-order statistics that can characterize the landslide evolution process; The construction module is used to construct multiple prediction sub-models based on high-order statistics and fuse them to form a landslide prediction model; The prediction module is used to input the high-order statistic to be measured into the landslide prediction model. If the high-order statistic to be measured exceeds a preset danger threshold, it is determined that there is a landslide risk.

Images