Intelligent landslide decision-making method and system based on Internet of Things perception technology

Abstract

The invention relates to an intelligent landslide decision-making method and system based on the Internet of Things sensing technology, and the method comprises the steps: receiving and processing multi-source sensing data, and obtaining a slope space model through the combination of slope point cloud data; according to the multi-source sensing data and the adaptive learning model, the landslide boundary, the current situation and the prediction result are intelligently analyzed with the help of a landslide disaster knowledge map, and are fed back to the virtual model in real time; generating a plurality of disaster prevention schemes based on an intelligent disaster decision result in combination with multi-source sensing data and a slope disaster associated knowledge graph, and screening out an optimal scheme with the lowest cost; finally, through an automatic design program, reinforcing design parameters are generated for the optimal scheme, a final design scheme is formed after optimization, and the final design scheme is visually displayed in the virtual model. According to the invention, through integrating multi-source sensing data, constructing a virtual slope space model, introducing intelligent decision and automatic design and other means, whole-course intelligent management of slope disaster prevention and control is realized, and the efficiency and level of disaster prevention and reduction are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of landslide monitoring, and in particular to a landslide intelligent decision-making method and system based on Internet of Things sensing technology. Background Art

[0002] Landslide is a common geological disaster that poses a serious threat to human life and property. In order to detect signs of landslide as early as possible and take effective early warning and preventive measures, the use of IoT sensing technology for landslide monitoring and intelligent decision-making has become a promising direction. Through real-time sensing and intelligent decision-making, the possibility of landslide can be analyzed early, effective preventive measures can be taken in advance, and disaster losses can be reduced. This is of great significance for protecting infrastructure, houses and other properties.

[0003] Landslide monitoring requires the comprehensive use of multi-source data, including surface displacement, deformation, stress, water level, etc. How to effectively integrate and process these heterogeneous data to improve the overall understanding of the landslide status is a complex issue. Establishing an accurate landslide prediction model requires comprehensive consideration of multiple factors, including geological conditions, meteorological conditions, hydrological conditions, etc. The complex relationship between these factors increases the complexity of the model, and in-depth research is needed to establish a more accurate and reliable prediction model, and ultimately make reliable decisions to guide the optimization of reinforcement plans. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a landslide intelligent decision-making method and system based on Internet of Things sensing technology, which solves the technical problem that the prior art is difficult to efficiently integrate and analyze data from multiple sources and with different structures, and to build a prediction model that can accurately predict landslide risks and consider the complex relationships between multiple factors.

[0006] (II) Technical solution

[0007] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0008] In a first aspect, an embodiment of the present invention provides a landslide intelligent decision-making method based on Internet of Things sensing technology, comprising:

[0009] Receive and process multi-source perception data collected by pre-deployed slope data perception equipment, combine the slope surface point cloud data obtained by photogrammetry and 3D laser scanning, and construct a virtual slope space model;

[0010] According to multi-source sensing data and at least one adaptive learning model, and by introducing a landslide disaster knowledge graph for analysis, a disaster intelligent decision-making result including at least one of a landslide boundary identification result, a landslide status assessment result and a landslide disaster prediction result is obtained, and the result is fed back into a virtual slope space model;

[0011] Based on the results of intelligent disaster decision-making, multiple disaster prevention plans are generated by combining the acquired multi-source perception data and the knowledge graph of slope disaster association, and the best disaster prevention plan is selected according to the lowest cost;

[0012] Using an automated design program, the slope reinforcement design parameters are generated for the best disaster prevention solution. After optimization calculation, the optimal design solution is generated and fed back into the virtual slope space model.

[0013] Optionally, based on multi-source sensing data and at least one adaptive learning model, a landslide disaster knowledge graph is introduced for analysis to obtain a disaster intelligent decision result including at least one of a landslide boundary identification result, a landslide status assessment result, and a landslide disaster prediction result, and the result is fed back to the virtual slope space model, including:

[0014] Add the MobileNet module as the backbone network based on the pre-trained DeepLabV3+ network, and adjust the number of categories in the output layer according to task requirements to obtain a lightweight recognition model. Use the lightweight recognition model to identify landslide boundaries in the acquired target area images.

[0015] Establish a landslide disaster evaluation index system, combine the AHP-GMR correlation fusion mechanism to perform significance and correlation analysis and judgment of multi-source perception data, and output the landslide disaster status evaluation results through data fusion;

[0016] The prediction model is constructed using the EMD-Elman learning algorithm, and the prediction model is trained based on multi-source perception data and the introduced previous slope disaster related knowledge graph to output the landslide disaster prediction results;

[0017] The landslide boundary identification, landslide hazard status evaluation results and landslide hazard prediction results are fed back into the virtual slope space model as the results of intelligent disaster decision-making for visualization.

[0018] Optionally, a MobileNet module is added as a backbone network based on the pre-trained DeepLabV3+ network, and the number of categories in the output layer is adjusted according to task requirements to obtain a lightweight recognition model. The lightweight recognition model is used to perform landslide boundary recognition on the acquired target area image, including:

[0019] The remote sensing image dataset of the landslide area is collected by drone equipment, and the slope area is annotated and labeled based on expert experience;

[0020] Create data storage objects for remote sensing image datasets and corresponding labels respectively;

[0021] In response to the task requirements, the pre-trained DeepLabV3+ network was selected as the basic model, the MobileNet module was added to the basic model as the backbone network, and the number of output layer categories of the model was adjusted to be consistent with the task requirements, thus obtaining a lightweight landslide impact area recognition model;

[0022] Pass the data storage object, lightweight landslide impact area identification model and pre-configured training option parameters to the TrainNetwork function for model training;

[0023] The preprocessed remote sensing image of the target area is obtained and loaded, and the loaded remote sensing image of the target area is input into the trained lightweight landslide impact area recognition model to output the recognition result of the landslide boundary.

[0024] Optionally, a landslide hazard evaluation index system is established, and the significance and correlation analysis and judgment of multi-source perception data are carried out in combination with the AHP-GMR correlation fusion mechanism, and the landslide hazard status evaluation results are output through data fusion, including:

[0025] According to the obtained landslide hazard evaluation index data, a landslide hazard evaluation index system including multiple factors affecting slope stability is constructed; wherein each factor is configured with multiple indicators to form a factor indicator sequence;

[0026] Collect corresponding quantitative or qualitative evaluation data according to the factors in the system, convert the qualitative evaluation indicators into quantitative ones, and generate the original factor matrix X;

[0027] Select a factor from the original factor matrix X as the reference evaluation index sequence to form the evaluation index sequence X e ;

[0028] The ratio of the factors in the original factor matrix X to the average value of all factors is used as the standardized factor value of the factor to form the standardized factor matrix Z, and the X in the evaluation index sequence is used as the standardized factor value of the factor. e The ratio of the evaluation index to the average value of all evaluation indexes is used as the standardized evaluation index value to form a standardized evaluation index sequence Z e ;

[0029] Using the AHP-GMR association fusion mechanism, each element in the standardized factor matrix Z and the standardized evaluation index sequence Z are calculated one by one. e The absolute difference of the corresponding elements, and determine the maximum difference and the minimum difference;

[0030] Based on the maximum difference and the minimum difference, calculate the standardized factor matrix Z and the standardized evaluation index sequence Z e The correlation coefficient at each moment, and the average value of the correlation coefficient at each moment is taken as the correlation between the jth factor and the corresponding indicator;

[0031] Sort by the degree of association to form an association sequence;

[0032] Select the factors in the original factor matrix X as the evaluation index sequence to calculate the correlation degree, and combine the correlation sequences obtained by each calculation to obtain the correlation matrix;

[0033] For multi-source sensing data, the correlation degree in the correlation matrix is ​​used as the weight for data fusion, and the intelligent evaluation index FoS of landslide disaster is obtained through normalization processing;

[0034] The Kriging difference algorithm is used to interpolate the global spatial data of the smart evaluation index FoS to achieve a continuous expression of the smart evaluation index FoS and transmit it back to the virtual slope spatial model;

[0035] in,

[0036] The original factor matrix X is:

[0037]

[0038] In the formula, n is the number of columns and m is the number of rows;

[0039] Evaluation index sequence X e for:

[0040] X e =(x e1 ,x e2 ,x e3 ,…,x em ) T ;

[0041] In the formula, e is the selected factor;

[0042] The normalized factor matrix Z is:

[0043] Z={z ij};

[0044]

[0045] In the formula, x ij represents the i-th index value of the j-th factor;

[0046] Standardized evaluation index sequence Z e for:

[0047] Z e=(z e1 ,z e2 ,z e3 ,…,z em ) T ;

[0048]

[0049] The minimum difference is:

[0050] Δmin j =min|Z ij -Z ei |;

[0051] The maximum difference is:

[0052] Δmax j =max|Z ij -Z ei |, i = 1, 2, 3......m;

[0053] The correlation coefficients at each moment are:

[0054]

[0055] In the formula, ρ is the resolution coefficient, taking [0, 1]. The smaller ρ is, the greater the sensitivity, the smaller the correlation coefficient value, and Δ is the difference between x ij and △min;

[0056] The correlation degree between the jth factor and the corresponding index is:

[0057]

[0058] The wisdom evaluation index FoS of each point i is:

[0059]

[0060] Optionally, use the EMD-Elman learning algorithm to construct a prediction model, and train the prediction model according to the multi-source perception data and the introduced knowledge graph of previous slope disasters to output the landslide disaster prediction results including:

[0061] Obtain multi-source perception data, and perform preprocessing operations on the collected multi-source perception data, including denoising processing, normalization processing, and extracting the incremental and acceleration feature curves related to landslide prediction from the normalized data;

[0062] Decompose the incremental and acceleration feature curves related to landslide prediction to establish a learning data set including trend terms, periodic terms, and random terms;

[0063] The empirical mode decomposition method is used to decompose the preprocessed multi-source data to obtain several IMF components;

[0064] For each IMF component, multiple Elman neural network prediction sub-models are constructed;

[0065] Using the learning data set and the introduced knowledge graph of slope hazards, each prediction sub-model is trained. At the same time, the prediction results are compared with the actual data to determine whether the prediction results meet the set RMSE threshold.

[0066] If the set RMSE threshold is met, the prediction results of all trained prediction sub-models are integrated using a weighted average scheme to obtain a comprehensive prediction result curve. According to the characteristics of multi-source perception data, the parameters of each prediction sub-model are combined into a prediction weight matrix for the landslide and then stored.

[0067] In a second aspect, an embodiment of the present invention provides a landslide intelligent decision-making system based on Internet of Things sensing technology, comprising:

[0068] The data layer is configured with slope data sensing devices for collecting multi-source sensing data, a data processing module for data input, storage and retrieval, and a digital twin module for receiving and processing multi-source sensing data collected by pre-deployed slope data sensing devices, combining the slope surface point cloud data obtained by photogrammetry and three-dimensional laser scanning, and constructing a virtual slope space model;

[0069] The decision-making layer is configured with an expert system for establishing an expert knowledge graph by integrating professional knowledge, experience and rules, a disaster intelligent decision-making module for obtaining a disaster intelligent decision-making result including at least one of a landslide boundary identification result, a landslide status assessment result and a landslide disaster prediction result based on multi-source perception data and at least one adaptive learning model, and feeding back the disaster intelligent decision-making result to the virtual slope space model, and an optimal disaster prevention scheme generation module for generating multiple disaster prevention schemes based on the disaster intelligent decision-making result, combining the obtained multi-source perception data and the slope disaster related knowledge graph, and selecting the optimal disaster prevention scheme according to the lowest cost, and an automated reinforcement module for generating slope reinforcement design parameters for the optimal disaster prevention scheme using an automated design program, generating an optimal design scheme after optimization calculation, and feeding back the optimal design scheme to the virtual slope space model;

[0070] The display layer is configured with a perception data display unit, a three-dimensional model display unit, a decision result display unit, a project information display unit, an automatic design display unit, and an automatic report display unit;

[0071] The management layer is configured with a user authority management unit, a project authority management unit, a log management unit, a security management unit, and a data management unit.

[0072] Optionally, the disaster intelligent decision-making module includes:

[0073] The landslide boundary recognition unit is used to add the MobileNet module as the backbone network based on the pre-trained DeepLabV3+ network, and adjust the number of categories in the output layer according to the task requirements to obtain a lightweight recognition model. The lightweight recognition model is used to identify the landslide boundary of the acquired target area image;

[0074] The landslide disaster status evaluation unit is used to establish a landslide disaster evaluation index system, combine the AHP-GMR correlation fusion mechanism to perform significance and correlation analysis and judgment of multi-source perception data, and output the landslide disaster status evaluation results through data fusion;

[0075] The landslide disaster prediction unit is used to build a prediction model using the EMD-Elman learning algorithm, train the prediction model based on multi-source perception data and the introduced previous slope disaster related knowledge graph to output the landslide disaster prediction results;

[0076] The feedback and display unit is used to feed back the landslide boundary identification, landslide disaster status evaluation results and landslide disaster prediction results as disaster intelligent decision-making results into the virtual slope space model for visual display.

[0077] Optionally, the landslide boundary identification unit includes:

[0078] The label generation subunit is used to collect remote sensing image datasets of the landslide area through UAV equipment, and to annotate the slope area and generate labels based on expert experience;

[0079] A memory generation subunit is used to create data memory objects for remote sensing image datasets and corresponding labels respectively;

[0080] The model lightweight subunit is used to select the pre-trained DeepLabV3+ network as the basic model according to the acquired task requirements, add the MobileNet module as the backbone network on the basic model, and adjust the number of categories in the output layer of the model to be consistent with the task requirements, so as to obtain a lightweight landslide impact area identification model;

[0081] The lightweight module training subunit is used to pass the data storage object, the lightweight landslide impact area identification model and the pre-configured training option parameters to the TrainNetwork function for model training;

[0082] The landslide boundary recognition result subunit is used to obtain and load the pre-processed remote sensing image of the target area, and input the loaded remote sensing image of the target area into the trained lightweight landslide impact area recognition model to output the recognition result of the landslide boundary.

[0083] Optionally, the landslide hazard status assessment unit includes:

[0084] The system construction subunit is used to construct a landslide hazard evaluation index system including multiple factors affecting slope stability according to the obtained landslide hazard evaluation index data; wherein each factor is configured with multiple indicators to form a factor indicator sequence;

[0085] The original factor matrix generation subunit is used to collect corresponding quantitative or qualitative evaluation data according to the factors in the system, convert the qualitative evaluation indicators into quantitative ones, and generate the original factor matrix X;

[0086] Select a factor from the original factor matrix X as the evaluation index sequence to form the evaluation index sequence X e ;

[0087] The standardization subunit is used to take the ratio of the factors in the original factor matrix X to the average value of all factors as the standardized factor value of the factor to form the standardized factor matrix Z, and to convert the X in the evaluation index sequence into e The ratio of the evaluation index to the average value of all evaluation indexes is used as the standardized evaluation index value to form a standardized evaluation index sequence Z e ;

[0088] The difference determination subunit is used to calculate the difference between each element in the standardized factor matrix Z and the standardized evaluation index sequence Z one by one by using the AHP-GMR association fusion mechanism. e The absolute difference of the corresponding elements, and determine the maximum difference and the minimum difference;

[0089] The correlation degree obtaining subunit is used to calculate the standardized factor matrix Z and the standardized evaluation index sequence Z based on the maximum difference and the minimum difference. e The correlation coefficient at each moment, and the average value of the correlation coefficient at each moment is taken as the correlation between the jth factor and the corresponding indicator;

[0090] The association relationship determination subunit is used to sort the associations according to the size of the association degree to form an association sequence, select the factors in the original factor matrix X in turn as the evaluation index sequence to calculate the association degree, and combine the association sequences obtained by each calculation to obtain the association matrix;

[0091] The smart evaluation index output subunit is used to fuse the multi-source perception data using the correlation degree in the correlation matrix as the weight, and obtain the smart evaluation index FoS of landslide disaster through normalization processing;

[0092] The interpolation subunit is used to interpolate the global spatial data of the smart evaluation index FoS using the Kriging interpolation algorithm, realize the continuous expression of the smart evaluation index FoS, and transmit it back to the virtual slope space model;

[0093] in,

[0094] The original factor matrix X is:

[0095]

[0096] In the formula, n is the number of columns and m is the number of rows;

[0097] Evaluation index sequence X e for:

[0098] X e =(x e1 ,x e2 ,x e3 ,…,x em ) T ;

[0099] In the formula, e is the selected factor;

[0100] The normalized factor matrix Z is:

[0101]

[0102] In the formula, x ij represents the i-th index value of the j-th factor;

[0103] Standardized evaluation index sequence Z e for:

[0104] Z e =(z e1 ,z e2 ,z e3 ,…,z em ) T ;

[0105]

[0106] The minimum difference is:

[0107] Δmin j =min|Z ij -Z ei |;

[0108] The maximum difference is:

[0109] Δmax j =max|Z ij -Z ei |, i = 1, 2, 3...m;

[0110] The correlation coefficient at each moment is:

[0111]

[0112] In the formula, ρ is the resolution coefficient, which is [0,1]. The smaller ρ is, the greater the sensitivity is, the smaller the correlation coefficient is, and Δ is X. ij The difference with △min;

[0113] The correlation between the jth factor and the corresponding indicator is:

[0114]

[0115] FoS, the intelligence evaluation index of each point i for:

[0116]

[0117] Optionally, the landslide hazard prediction unit includes:

[0118] A preprocessing subunit is used to obtain multi-source sensing data, and perform preprocessing operations including denoising and normalization on the collected multi-source sensing data, and extracting increment and acceleration characteristic curves related to landslide prediction from the normalized data;

[0119] A learning data set output subunit is used to decompose the increment and acceleration characteristic curves related to landslide prediction to establish a learning data set containing trend items, period items and random items;

[0120] The prediction sub-model construction sub-unit is used to decompose the pre-processed multi-source data using the empirical mode decomposition method to obtain several IMF components, and for each IMF component, construct multiple Elman neural network prediction sub-models respectively;

[0121] The training and judgment subunit is used to train each prediction submodel using the learning data set and the introduced knowledge graph of previous slope hazards, and compare the prediction results with the actual data to determine whether the prediction results meet the set RMSE threshold.

[0122] The fusion and storage subunit is used to fuse the prediction results of all trained prediction sub-models by using a weighted average scheme if the set RMSE threshold is met, to obtain a comprehensive prediction result curve, and to combine the parameters of each prediction sub-model into a prediction weight matrix for the landslide according to the characteristics of multi-source perception data, and then store it.

[0123] (III) Beneficial effects

[0124] The beneficial effects of the present invention are:

[0125] First, by receiving and processing multi-source sensing data collected by pre-deployed slope data sensing equipment, it is possible to obtain real-time and accurate key information on the slope, including physical, mechanical, and environmental aspects. Combining the slope surface point cloud data obtained by photogrammetry and 3D laser scanning technology, a detailed and virtual slope spatial model can be constructed, which not only helps to fully understand the real-time status of the slope, but also provides an important data basis for subsequent disaster prediction and prevention work.

[0126] Secondly, by using multi-source sensing data and adaptive learning models, and introducing landslide disaster knowledge graphs for analysis, landslide boundaries can be identified more intelligently, the current status of landslides can be assessed, and landslide disasters can be predicted. These results not only reflect the current stability of the slope, but also predict possible disaster risks in the future, providing a scientific decision-making basis for timely disaster prevention measures. At the same time, these decision results are fed back to the virtual slope space model, realizing the dynamic synchronization of the model and the actual slope status, further improving the practicality and accuracy of the model.

[0127] Furthermore, based on the results of intelligent disaster decision-making, combined with multi-source perception data and slope disaster-related knowledge graphs, a variety of disaster prevention plans can be generated. The best disaster prevention plan is selected based on the principle of minimum cost, which not only ensures the optimization of disaster prevention effects, but also achieves efficient use of resources.

[0128] Finally, the computer automated design program is used to quickly generate the slope reinforcement design parameters for the best disaster prevention solution, and the optimal design solution is generated after optimization calculation. This process not only improves the design efficiency, but also ensures the scientificity and rationality of the design solution. At the same time, the optimal design solution is fed back to the virtual slope space model, which can further verify the feasibility of the design solution, intuitively display the reinforcement effect, and provide strong guidance for actual construction.

[0129] Therefore, the present invention realizes the full intelligent management of slope disaster prevention and control by integrating multi-source perception data, constructing a virtual slope space model, introducing intelligent decision-making and automated design and other technical means, and significantly improves the efficiency and level of disaster prevention and mitigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0130] Figure 1 A schematic diagram of a flow chart of a method provided by an embodiment of the present invention;

[0131] Figure 2 A schematic diagram of a process of intelligent boundary identification of a method provided in an embodiment of the present invention;

[0132] Figure 3 A schematic diagram of a process of intelligent disaster prediction according to a method provided in an embodiment of the present invention;

[0133] Figure 4 A schematic diagram of the composition of a system provided by an embodiment of the present invention;

[0134] Figure 5 This is a display interface diagram of the system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0135] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0136] like Figure 1 As shown, an intelligent landslide decision-making method based on the Internet of Things sensing technology proposed in an embodiment of the present invention includes: receiving and processing multi-source sensing data collected by pre-deployed slope data sensing equipment, combining the slope surface point cloud data obtained by photogrammetry and three-dimensional laser scanning, and constructing a virtual slope space model; according to the multi-source sensing data and at least one adaptive learning model, and introducing the landslide disaster knowledge map for analysis, obtaining a disaster intelligent decision-making result including at least one of the landslide boundary recognition result, the landslide status assessment result and the landslide disaster prediction result, and feeding it back to the virtual slope space model; based on the disaster intelligent decision-making result, combining the obtained multi-source sensing data and the slope disaster related knowledge map, generating multiple disaster prevention plans, and screening out the best disaster prevention plan according to the lowest cost; using an automated design program to generate slope reinforcement design parameters for the best disaster prevention plan, generating the optimal design plan after optimization calculation, and feeding it back to the virtual slope space model.

[0137] First, by receiving and processing multi-source sensing data collected by pre-deployed slope data sensing equipment, it is possible to obtain real-time and accurate key information on the slope, including physical, mechanical, and environmental aspects. Combining the slope surface point cloud data obtained by photogrammetry and 3D laser scanning technology, a detailed and virtual slope spatial model can be constructed, which not only helps to fully understand the real-time status of the slope, but also provides an important data basis for subsequent disaster prediction and prevention work.

[0138] Secondly, by using multi-source sensing data and adaptive learning models, and introducing landslide disaster knowledge graphs for analysis, landslide boundaries can be identified more intelligently, the current status of landslides can be assessed, and landslide disasters can be predicted. These results not only reflect the current stability of the slope, but also predict possible disaster risks in the future, providing a scientific decision-making basis for timely disaster prevention measures. At the same time, these decision results are fed back to the virtual slope space model, realizing the dynamic synchronization of the model and the actual slope status, further improving the practicality and accuracy of the model.

[0139] Furthermore, based on the results of intelligent disaster decision-making, combined with multi-source perception data and slope disaster-related knowledge graphs, a variety of disaster prevention plans can be generated. The best disaster prevention plan is selected based on the principle of minimum cost, which not only ensures the optimization of disaster prevention effects, but also achieves efficient use of resources.

[0140] Finally, the computer automated design program is used to quickly generate the slope reinforcement design parameters for the best disaster prevention solution, and the optimal design solution is generated after optimization calculation. This process not only improves the design efficiency, but also ensures the scientificity and rationality of the design solution. At the same time, the optimal design solution is fed back to the virtual slope space model, which can further verify the feasibility of the design solution, intuitively display the reinforcement effect, and provide strong guidance for actual construction.

[0141] Therefore, the present invention realizes the full intelligent management of slope disaster prevention and control by integrating multi-source perception data, constructing a virtual slope space model, introducing intelligent decision-making and automated design and other technical means, and significantly improves the efficiency and level of disaster prevention and mitigation.

[0142] In order to better understand the above technical solution, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0143] Specifically, an embodiment of the present invention provides a landslide intelligent decision-making method based on Internet of Things sensing technology, which includes:

[0144] S1. Receive and process multi-source perception data collected by pre-deployed slope data perception equipment, combine the slope surface point cloud data obtained by photogrammetry and 3D laser scanning, and construct a virtual slope space model.

[0145] Specifically, slope data perception mainly relies on burying sensors on the surface and inside of the slope to achieve real-time collection of slope physical, mechanical, environmental and other data. Slope data perception equipment refers to a combination of equipment used to collect slope physical and mechanical data, including intelligent perception sensors, digital collection equipment and wireless transmission components.

[0146] As the terminal of the perception system, the intelligent perception sensor can continuously collect various monitoring data of the slope, including displacement, deformation, stress, surface cracks, groundwater level, pore water pressure, etc. Commonly used perception devices include optical fiber displacement sensors, laser displacement sensors, wireless strain gauges, wireless crack sensors, fully automatic water level gauges, sonic detectors, fully automatic pore water pressure gauges, etc.

[0147] Digital acquisition equipment is used to analyze and process the data transmitted by intelligent sensing sensors. Digital acquisition equipment has high reliability and accuracy, ensuring long-term stable operation, and performs data verification and format conversion on the collected data to achieve preliminary denoising of the data.

[0148] The wireless transmission component is used to transmit the data collected by the digital acquisition equipment to the central cloud server. The transmission system is based on the wireless transmission technology of the 5G cellular network, which can achieve a large coverage range and ensure a high transmission speed.

[0149] Next, the digital twin modeling of the slope is carried out by applying the data collected by the slope data perception device and combining the slope surface point cloud data obtained by photogrammetry and 3D laser scanning. The slope digital twin model combines the physical model of the slope with the multidimensional data model to build a virtual slope space model. The digital twin model can show the evolution process of the slope and realize multi-source data collaboration. At the same time, it can also be updated in real time based on the data collected by the slope data perception system to reflect the real state of the slope in real time.

[0150] S2. Based on multi-source perception data and at least one adaptive learning model, and introducing a landslide disaster knowledge graph for analysis, obtain a disaster intelligent decision-making result including at least one of a landslide boundary identification result, a landslide status assessment result and a landslide disaster prediction result, and feed it back to the virtual slope space model.

[0151] Based on the previous knowledge graph of landslide disasters, the present invention combines intelligent algorithms to perform intelligent identification and analysis of monitoring data, realize intelligent selection of the optimal disaster prevention strategy, and return the decision results to the digital twin model for display. Specifically, it can be subdivided into three main contents: intelligent identification of landslide boundaries, intelligent assessment of landslide status, and intelligent prediction of landslide development trends.

[0152] Further, step S2 includes:

[0153] S21. Add the MobileNet module as the backbone network based on the pre-trained DeepLabV3+ network, and adjust the number of categories in the output layer according to task requirements to obtain a lightweight recognition model. Use the lightweight recognition model to identify the landslide boundary in the acquired target area image.

[0154] Furthermore, step S21 includes: first, collecting remote sensing image data sets of the landslide area through unmanned aerial vehicle equipment, and annotating the slope area and generating labels in combination with expert experience; second, creating data storage objects for the remote sensing image data sets and corresponding labels respectively; in response to task requirements, selecting the pre-trained DeepLabV3+ network as the basic model, adding the MobileNet module as the backbone network on the basic model, and adjusting the number of output layer categories of the model to be consistent with the task requirements, so as to obtain a lightweight landslide impact area identification model; then, passing the data storage object, the lightweight landslide impact area identification model and the pre-configured training option parameters to the TrainNetwork function for model training; further, obtaining and loading the pre-processed remote sensing image of the target area, and inputting the loaded remote sensing image of the target area into the trained lightweight landslide impact area identification model, and outputting the recognition result of the landslide boundary.

[0155] In a specific embodiment, reference Figure 2 , using MobileNet as the backbone network and combining it with the DeepLabV3+ network architecture, a lightweight landslide impact area recognition model was constructed. The specific recognition decision steps are as follows

[0156] (1) Data preparation: A remote sensing image dataset of the landslide area is collected by drone equipment. Experienced geological experts annotate the slope area and generate labels.

[0157] (2) Create data storage objects: Create data storage objects for landslide images and corresponding labels respectively to facilitate subsequent data loading and processing.

[0158] (3) Build a deep learning model: Select an appropriate deep learning model based on the task requirements. Use the pre-trained model DeepLabV3+ network as the basis, and then add the MobileNet module as the backbone network. When building the model, ensure that the number of categories in the output layer is consistent with the task.

[0159] (4) Configure training options: Define training parameters, including loss function, optimizer, learning rate, and training batch size. Common loss functions include cross entropy. Optimizers can be SGD, Adam, etc.

[0160] (5) Model training: Call the TrainNetwork function to train the model. Pass the data storage object, deep learning model, and training option parameters to the TrainNetwork function, and save the model after training.

[0161] (6) Identify landslide areas: Use the trained model to identify landslide-affected areas in new remote sensing images. Load the pre-processed remote sensing image of the target area collected by the sensor and obtain the identification results.

[0162] (7) Evaluate performance: Use the test data set to evaluate the trained model and calculate parameter indicators such as accuracy and recall to evaluate the performance of the model. At the same time, return the evaluation results to step (5) to iterate the model.

[0163] S22. Establish a landslide hazard evaluation index system, combine the AHP-GMR correlation fusion mechanism to conduct significance and correlation analysis and judgment of multi-source perception data, and output the landslide hazard status evaluation results through data fusion.

[0164] Furthermore, step S22 includes: first, constructing a landslide hazard evaluation index system including multiple factors affecting slope stability according to the acquired landslide hazard evaluation index data; wherein each factor is configured with multiple indicators to form a factor indicator sequence; second, collecting corresponding quantitative or qualitative evaluation data according to the factors in the system, converting the qualitative evaluation indicators into quantitative ones, and generating an original factor matrix X; then, selecting a factor from the original factor matrix X as a reference evaluation index sequence to form an evaluation index sequence X e ; Then, the ratio of the factors in the original factor matrix X to the average value of all factors is used as the standardized factor value of the factor to form the standardized factor matrix Z, and the X in the evaluation index sequence is e The ratio of the evaluation index to the average value of all evaluation indexes is used as the standardized evaluation index value to form a standardized evaluation index sequence Z e ; Then, using the AHP-GMR association fusion mechanism, each element in the standardized factor matrix Z and the standardized evaluation index sequence Z are calculated one by one e The absolute difference of the corresponding elements is calculated, and the maximum difference and the minimum difference are determined; furthermore, based on the maximum difference and the minimum difference, the standardized factor matrix Z and the standardized evaluation index sequence Z are calculated. e The correlation coefficient at each moment is calculated, and the average value of the correlation coefficient at each moment is taken as the correlation between the jth factor and the corresponding indicator; then, the correlation is sorted according to the size of the correlation to form a correlation sequence; at the same time, the factors in the original factor matrix X are selected in turn as the evaluation indicator sequence X eThe correlation degree is calculated, and the correlation sequences obtained by each calculation are combined to obtain the correlation matrix; then, for multi-source perception data, the correlation degree in the correlation matrix is ​​used as the weight for data fusion, and the smart evaluation index FoS of landslide disaster is obtained through normalization processing; finally, the Kriging interpolation algorithm is used to interpolate the global spatial data of the smart evaluation index FoS to realize the continuous expression of the smart evaluation index FoS, and then it is transmitted back to the virtual slope space model.

[0165] In a specific embodiment, an intelligent disaster expert assessment system is established, and the significance and correlation analysis and judgment of monitoring data are carried out by means of the AHP-GMR (hierarchical analysis-grey correlation) data fusion mechanism, and the correlation and significance evaluation index between various factors is constructed. The specific steps are as follows:

[0166] (1) Based on the landslide hazard evaluation index, an evaluation index system including various factors affecting slope stability is established. According to the factors in the system, quantitative or qualitative evaluation data are collected and evaluated, and the qualitative evaluation index is quantitatively converted to generate the original factor matrix X:

[0167]

[0168] The factor matrix contains n factors, and each factor has m indicators to form a factor indicator sequence. ij Represents the i-th index value of the j-th factor.

[0169] Next, select a factor from the factor matrix as the evaluation index sequence, and set the evaluation index sequence to X e for:

[0170] X e =(x e1 ,x e2 ,x e3 ,…,x em ) T ; e is the selected factor.

[0171] (2) Standardize the factor matrix. Calculate the average value of each factor index, and use the ratio of the factor to the average value as the standardized evaluation value of the factor. At the same time, standardize the evaluation index sequence to obtain the standardized factor matrix Z and the standardized evaluation index sequence Z. e :

[0172] The normalized factor matrix Z is:

[0173]

[0174] In the formula, x ij represents the i-th index value of the j-th factor;

[0175] Standardized evaluation index sequence Z e for:

[0176] Z e =(z e1 ,z e2 ,z e3 ,…,z em ) T ;

[0177]

[0178] (3) For the standardized factor matrix Z, calculate each factor sequence and the standardized evaluation index sequence Z one by one e The absolute difference of the corresponding elements, and determine the maximum and minimum differences. The maximum value Δmax of the absolute interpolation of the jth factor j and minimum value Δmin j .

[0179] The minimum difference is:

[0180] Δmin j =min|Z ij -Z ei |;

[0181] The maximum difference is:

[0182] Δmax j =max|Z ij -Z ei |, i = 1, 2, 3...m;

[0183] (4) Calculate the correlation coefficient between each factor sequence and the corresponding element of the evaluation index sequence. The degree of correlation involved here is essentially the correlation coefficient between each factor sequence Z j Curve and evaluation index sequence Z e The degree of geometric similarity. For each factor sequence Z j , i.e. the jth column of Z, and compare it with the standardized evaluation index sequence Z e , and calculate the correlation coefficient ξ(x ij ):

[0184]

[0185] In the formula, ρ is the resolution coefficient, which is [0,1]. The smaller ρ is, the greater the sensitivity is, the smaller the correlation coefficient is, and Δ is x. ij The difference with △min.

[0186] (5) Calculation of correlation coefficient. The correlation coefficient is a comparison of the factor sequences X j With the standardized evaluation index sequence X eThe correlation value over the entire sequence interval. In order to obtain an overall judgment value, it is necessary to calculate the average value of the correlation coefficient at each moment in the sequence interval as a quantitative representation of the correlation between the factor sequence and the standardized evaluation index sequence. The correlation between the jth factor and the evaluation index is r j for:

[0187]

[0188] (6) Relevance ranking. The degree of correlation between each factor and the evaluation index is ranked according to the correlation r j In order of size, the n factor sequence X j For the evaluation index sequence X e The correlation degrees of the factors are arranged in order of size to form a correlation sequence, which reflects the correlation of each factor sequence to the evaluation index sequence. 1 >r 2 , it means that the correlation between factor 1 and the evaluation index is better than factor 2, that is, the factor 1 sequence X 1 The change pattern and evaluation index sequence X e The more similar the change patterns are.

[0189] (7) Establish a correlation matrix. According to the above steps of calculating the correlation, select the factors in the factor matrix as the evaluation index sequence for correlation calculation. Combine the correlation sequence r obtained in each calculation to obtain the correlation matrix. According to the order of the indicators and the size of the correlation, the degree of correlation between each factor sequence and the evaluation index sequence can be significantly distinguished. In the actual data fusion process, the factor sequence with a correlation less than 0.5 can be eliminated according to the size of the correlation to reduce data redundancy.

[0190] (8) Data fusion and intelligent evaluation. The correlation degree in the correlation matrix is ​​used as the weight for data fusion and normalization to obtain the intelligent evaluation index FoS for each point. i for:

[0191]

[0192] (9) After obtaining the FoS index of the key points, the Kriging interpolation algorithm is used to interpolate the spatial data in the entire domain, and the interpolation results are sent back to the virtual slope spatial model for data display.

[0193] S23. Use the EMD-Elman learning algorithm to build a prediction model, and train the prediction model based on multi-source perception data and the introduced previous slope disaster related knowledge graph to output landslide disaster prediction results.

[0194] Furthermore, step S23 includes: first, obtaining multi-source sensing data, performing preprocessing operations including denoising, normalization, and extracting incremental and acceleration characteristic curves related to landslide prediction from the normalized data; second, decomposing the incremental and acceleration characteristic curves related to landslide prediction to establish a learning data set including trend terms, periodic terms, and random terms; then, using the empirical mode decomposition method to decompose the preprocessed multi-source data to obtain several IMF components; then, for each IMF component, respectively constructing multiple Elman neural network prediction sub-models; then, using the learning data set and the introduced previous slope disaster-related knowledge graph, each prediction sub-model is trained, and the prediction results are compared with the actual data to determine whether the prediction results meet the set RMSE threshold. If they meet the set RMSE threshold, the prediction results of all trained prediction sub-models are merged using a weighted average scheme to obtain a comprehensive prediction result curve, and according to the characteristics of the multi-source sensing data, the parameters of each prediction sub-model are combined into a prediction weight matrix of the landslide, and then stored.

[0195] In a specific embodiment, the AHP-GMR data association fusion system is applied, and the EMD (empirical mode decomposition) data decomposition method is used to process the displacement, GNSS and other data curve data in the multi-source perception data collected by the sensor, respectively, and a learning data set consisting of trend items, period items, and random items is established for these data sequences, and the Elman machine learning analysis algorithm is applied for data fusion and prediction. Figure 3 , the specific steps are as follows:

[0196] (1) Multi-source data collection: Data are collected from different sensors and monitoring equipment, including surface displacement, soil moisture, rainfall, etc., which are key parameters in landslide monitoring.

[0197] (2) Data preprocessing: The collected data is denoised to eliminate the interference caused by environmental noise or equipment errors. Then, the data is normalized to ensure that data from different sources are comparable. The normalized data is differentiated to extract the characteristic curves of increment and acceleration that are useful for predicting landslides.

[0198] (3) Data decomposition and data set augmentation: The EMD method is used to decompose the preprocessed data (displacement, velocity, acceleration, environmental factors, physical parameters, etc.) into several IMFs, and the monitoring data series curve is decomposed into trend term curves, periodic term curves, random term curves, etc.

[0199] (4) Elman neural network learning and prediction: For each IMF component, we construct an Elman neural network prediction model. The prediction results are compared with the actual results, and the results of different components are analyzed by RMSE threshold judgment. According to the specific characteristics of the monitoring data, the model parameters are combined into the prediction weight matrix of the landslide and stored.

[0200] (5) Data fusion: In the fusion layer, the prediction results of all Elman neural networks are fused and the weighted average scheme is used to fuse the results to obtain a comprehensive prediction result curve. The relevant parameters are stored in the database and displayed on the virtual slope space model by reading the database information.

[0201] S24. The landslide boundary identification, landslide disaster status evaluation results and landslide disaster prediction results are fed back into the virtual slope space model as disaster intelligent decision-making results for visual display.

[0202] S3. Based on the results of intelligent disaster decision-making, combined with the acquired multi-source perception data and slope disaster-related knowledge graph, a variety of disaster prevention plans are generated, and the best disaster prevention plan is selected according to the lowest cost.

[0203] S4. Utilize the automated design program to generate slope reinforcement design parameters for the best disaster prevention solution. After optimization calculation, the optimal design solution is generated and fed back into the virtual slope space model.

[0204] In the automated reinforcement scheme, with the help of the results of intelligent disaster decision-making, the best disaster prevention strategy is recommended according to the intelligent decision, and the automated slope reinforcement design is carried out. The intelligent disaster decision-making system can analyze the geometric shape, geological conditions, seismic parameters, rainfall parameters and other factors of the slope according to the actual situation of the slope, and analyze it according to the knowledge map associated with slope disasters, generate a variety of disaster prevention schemes, and analyze and compare each scheme. On the premise of meeting the disaster prevention requirements, the optimal disaster prevention scheme with the lowest cost is selected. Using the computer automated design program, according to the optimal disaster prevention strategy, the type, size, location and other parameters of the reinforcement structure are automatically generated, and the optimization calculation is performed to generate the optimal design scheme.

[0205] Additionally, refer to Figure 4 The embodiment of the present invention provides a landslide intelligent decision-making system based on Internet of Things sensing technology, including:

[0206] The data layer is equipped with slope data perception equipment for collecting multi-source perception data, a data processing module for data input, storage and retrieval, and a digital twin module for receiving and processing multi-source perception data collected by pre-deployed slope data perception equipment, combining the slope surface point cloud data obtained by photogrammetry and three-dimensional laser scanning to construct a virtual slope space model.

[0207] The decision-making layer is configured with an expert system for establishing an expert knowledge graph by integrating professional knowledge, experience and rules, a disaster intelligent decision-making module for analyzing based on multi-source perception data and at least one adaptive learning model, and introducing a landslide disaster knowledge graph to obtain a disaster intelligent decision-making result including at least one of a landslide boundary identification result, a landslide current status assessment result and a landslide disaster prediction result, and feeding it back to the virtual slope space model, and a disaster intelligent decision-making module for generating multiple disaster prevention plans based on the disaster intelligent decision-making result, combining the obtained multi-source perception data and the slope disaster related knowledge graph, and selecting the best disaster prevention plan according to the lowest cost. The module also generates a module for generating a best disaster prevention plan according to the best disaster prevention plan by using an automated design program, generates slope reinforcement design parameters for the best disaster prevention plan, generates an optimal design plan after optimization calculation, and feeds it back to the virtual slope space model.

[0208] The display layer is configured with a perception data display unit, a three-dimensional model display unit, a decision result display unit, a project information display unit, an automatic design display unit and an automatic report display unit.

[0209] And, the management layer is configured with a user authority management unit, a project authority management unit, a log management unit, a security management unit and a data management unit.

[0210] Specifically, the disaster intelligent decision-making module includes: a landslide boundary recognition unit, which is used to add a MobileNet module as the backbone network based on the pre-trained DeepLabV3+ network, and adjust the number of categories in the output layer according to task requirements to obtain a lightweight recognition model, and use the lightweight recognition model to identify the landslide boundary of the acquired target area image; a landslide disaster status evaluation unit, which is used to establish a landslide disaster evaluation index system, combine the AHP-GMR correlation fusion mechanism to perform significance and correlation analysis and judgment of multi-source perception data, and output the landslide disaster status evaluation results after data fusion; and a landslide disaster prediction unit, which is used to construct a prediction model using the EMD-Elman learning algorithm, train the prediction model based on multi-source perception data and the introduced previous slope disaster association knowledge graph, and output the landslide disaster prediction results.

[0211] Therefore, the disaster intelligent decision-making module uses the deep learning DeepLab+MobileNet network to intelligently identify the landslide-affected area; establishes an intelligent evaluation system for disaster assessment experts, uses the AHP-GMR theory to measure the correlation and significance between various factors, and performs data fusion to obtain the final evaluation results. Using the EMD-Elman learning algorithm, based on the previous slope disaster-related knowledge graph, the decision model is trained to conduct comprehensive prediction of multi-source monitoring data. The deep learning algorithm can use historical data to train the decision model and improve the accuracy of intelligent identification, evaluation, and prediction.

[0212] In one embodiment, the landslide boundary identification unit includes:

[0213] The label generation subunit is used to collect remote sensing image datasets of the landslide area through UAV equipment, and to annotate the slope area and generate labels based on expert experience;

[0214] A memory generation subunit is used to create data memory objects for remote sensing image datasets and corresponding labels respectively;

[0215] The model lightweight subunit is used to select the pre-trained DeepLabV3+ network as the basic model according to the acquired task requirements, add the MobileNet module as the backbone network on the basic model, and adjust the number of categories in the output layer of the model to be consistent with the task requirements, so as to obtain a lightweight landslide impact area identification model;

[0216] The lightweight module training subunit is used to pass the data storage object, the lightweight landslide impact area identification model and the pre-configured training option parameters to the TrainNetwork function for model training;

[0217] The landslide boundary recognition result subunit is used to obtain and load the pre-processed remote sensing image of the target area, and input the loaded remote sensing image of the target area into the trained lightweight landslide impact area recognition model to output the recognition result of the landslide boundary.

[0218] In another embodiment, the landslide hazard status assessment unit includes:

[0219] The system construction subunit is used to construct a landslide hazard evaluation index system including multiple factors affecting slope stability according to the obtained landslide hazard evaluation index data; wherein each factor is configured with multiple indicators to form a factor indicator sequence;

[0220] The original factor matrix generation subunit is used to collect corresponding quantitative or qualitative evaluation data according to the factors in the system, convert the qualitative evaluation indicators into quantitative ones, and generate the original factor matrix X;

[0221] The evaluation index sequence generation subunit is used to select a factor from the original factor matrix X as a reference evaluation index sequence to form an evaluation index sequence X e ;

[0222] The standardization subunit is used to take the ratio of the factors in the original factor matrix X to the average value of all factors as the standardized factor value of the factor to form the standardized factor matrix Z, and to convert the X in the evaluation index sequence into e The ratio of the evaluation index to the average value of all evaluation indexes is used as the standardized evaluation index value to form a standardized evaluation index sequence Z e ;

[0223] The difference determination subunit is used to calculate the difference between each element in the standardized factor matrix Z and the standardized evaluation index sequence Z one by one by using the AHP-GMR association fusion mechanism. e The absolute difference of the corresponding elements, and determine the maximum difference and the minimum difference;

[0224] The correlation degree obtaining subunit is used to calculate the correlation coefficient between the standardized factor matrix Z and the standardized evaluation index sequence Ze at each moment based on the maximum difference and the minimum difference, and obtain the average value of the correlation coefficient at each moment as the correlation degree between the jth factor and the corresponding index;

[0225] The association relationship determination subunit is used to sort the associations according to the size of the association degree to form an association sequence, and select the factors in the original factor matrix X in turn as the evaluation index sequence X e Perform correlation calculation, combine the correlation sequences obtained in each calculation, and obtain a correlation matrix;

[0226] The smart evaluation index output subunit is used to fuse the multi-source perception data using the correlation degree in the correlation matrix as the weight, and obtain the smart evaluation index FoS of landslide disaster through normalization processing;

[0227] The interpolation subunit is used to use the Kriging interpolation algorithm to interpolate the global spatial data of the smart evaluation index FoS, realize the continuous expression of the smart evaluation index FoS, and transmit it back to the virtual slope space model.

[0228] In yet another embodiment, the landslide hazard prediction unit comprises:

[0229] A preprocessing subunit is used to obtain multi-source sensing data, and perform preprocessing operations including denoising and normalization on the collected multi-source sensing data, and extracting increment and acceleration characteristic curves related to landslide prediction from the normalized data;

[0230] A learning data set output subunit is used to decompose the increment and acceleration characteristic curves related to landslide prediction to establish a learning data set containing trend items, period items and random items;

[0231] The prediction sub-model construction sub-unit is used to decompose the pre-processed multi-source data using the empirical mode decomposition method to obtain several IMF components, and for each IMF component, construct multiple Elman neural network prediction sub-models respectively;

[0232] The training and judgment subunit is used to train each prediction submodel using the learning data set and the introduced knowledge graph of previous slope hazards, and compare the prediction results with the actual data to determine whether the prediction results meet the set RMSE threshold.

[0233] The fusion and storage subunit is used to fuse the prediction results of all trained prediction sub-models by using a weighted average scheme if the set RMSE threshold is met, to obtain a comprehensive prediction result curve, and to combine the parameters of each prediction sub-model into a prediction weight matrix for the landslide according to the characteristics of multi-source perception data, and then store it.

[0234] The feedback and display unit is used to feed back the landslide boundary identification, landslide disaster status evaluation results and landslide disaster prediction results as disaster intelligent decision-making results into the virtual slope space model for visual display.

[0235] In addition, the decision-making layer also has: an expert evaluation system that integrates professional knowledge in fields such as geology, meteorology, and civil engineering, allowing experts in the field to share experience and knowledge, establish an expert knowledge map, integrate professional knowledge, experience, and rules in a structured format, form a comprehensive professional knowledge network, and integrate expert experience into the landslide risk assessment and decision-making process. Build a decision rule library, integrate experts' decision-making principles in different situations, provide an expert feedback interface, and enable practitioners in professional fields to update the knowledge base in real time. Use a fuzzy logic reasoning engine to fuzzify and quantify expert experience, and establish a quantifiable and explainable intelligent expert evaluation model.

[0236] Therefore, the decision-making layer can use data mining technology to conduct in-depth mining of historical data and identify the inherent laws and changing trends of historical data. Use machine learning algorithms to learn real-time perception data, enhance the module's prediction generalization ability, and improve the accuracy of landslide risk prediction under different working conditions. Enhance the interpretability of the prediction model, establish a prediction model driven by physical laws, and enhance the trust of engineering personnel in the model output. Introduce deep learning technology, build a deep neural network, learn complex geological patterns from landslide geometry data, and enhance the system's analytical ability to identify potential landslide risks. Use perception of real-time data streams, combined with anomaly detection algorithms, to identify and mark unusual geological events in real time, and improve the module's sensitivity to emergencies. Develop visual analysis tools to display data analysis results through charts, graphics, etc., so that decision makers can intuitively understand the analysis process and results.

[0237] At the same time, the decision-making layer applies machine learning algorithms to perform trend analysis based on historical data and real-time perception data to identify the evolution trend of potential landslide risks in advance. A neural network prediction model is established to verify and test its prediction ability for different working conditions and improve the robustness of the module. When a potential landslide risk is predicted, alarm information is generated in real time and fed back to the digital twin model. An adaptive learning algorithm is introduced to enable the system to automatically iterate and optimize the prediction model based on new perception data to adapt to real-time changes in landslide geology, hydrology, and environment. The intelligent prediction module is used to switch between multiple scenes to evaluate the landslide risk under different working conditions. Multiple factors are considered in the intelligent decision-making process, including perception data, expert evaluation results, and intelligent prediction results. A fuzzy comprehensive evaluation algorithm is used to set a risk threshold, and a decision warning is triggered when the landslide risk coefficient exceeds the threshold. Risk management and pattern management tools are provided to help decision makers formulate appropriate response strategies under different risk levels and damage modes. A collaborative decision-making platform is designed to achieve information sharing and multi-party collaborative decision-making among decision makers in different professional fields.

[0238] Secondly, it is important to understand that the management layer includes modules such as user authority management, project authority management, log management, security management, and data management. The following is a detailed introduction to each module in the management layer:

[0239] User rights management refers to the definition of user roles and the allocation of rights. Appropriate system functions and data access rights are assigned according to different roles such as administrators, engineers, and decision makers to ensure that users can only access information within their rights.

[0240] Project permission management includes project team management and project access rights, determining team members involved in the project and assigning corresponding permissions. Controlling the directional access of project data by members with different project permissions to ensure information security and consistency.

[0241] Log management records user operations, including login, data modification, system configuration changes, etc., in order to track system activities. At the same time, it records system abnormalities in real time to provide support for problem troubleshooting.

[0242] Security management, using authentication mechanisms such as two-factor authentication to ensure that only authorized users can log into the system. At the same time, multiple encryptions are performed on sensitive data in transmission and storage to prevent data leakage and tampering. Firewalls and intrusion detection systems are deployed to protect the system from network attacks.

[0243] Data management, regular data backup, and establishment of effective recovery mechanisms to ensure secure data sharing at different authority levels.

[0244] Next, the data layer includes modules such as data perception, data processing, and data twins.

[0245] Data perception includes physical perception devices, mechanical perception devices, and environmental perception devices.

[0246] Physical sensing equipment. Automatic inclinometers are deployed in potential landslide areas to monitor the tilt angle of the surface in real time, provide physical sensing data about surface changes, and accurately judge the stability of the surface. Fully automatic strain gauges are installed at key deformation locations to monitor the displacement of soil and rock and provide soil deformation information.

[0247] Mechanical sensing equipment, automatic stress gauges are deployed at key locations such as the slope foot and slope shoulder to sense the changes in slope tensile stress in real time. Pressure sensors are deployed near the sliding surface inside the slope to monitor the pressure changes inside the landslide soil.

[0248] Environmental sensing equipment, using automated weather stations to collect environmental parameters such as temperature, humidity, wind speed, etc., to assess the potential impact of meteorological conditions on landslides. Automatic groundwater level meters are used to measure groundwater levels, provide soil saturation and hydrological information, and sense the water level inside the slope. Automatic rain gauges are used to measure rainfall and mark the amount of rainfall in a hydrological year.

[0249] And, data processing includes data input, data storage, and data retrieval.

[0250] Data input, through the Internet of Things sensing technology, real-time collection of data generated by various sensing devices. Standardized cleaning, screening and processing of different types of sensor data, compiled into a unified data format. Establish a standardized digital interface to achieve unified mapping of multi-modal data.

[0251] Data storage: Stores structured data from sensing devices and provides a high-speed, reliable data access interface. Archives historical data to support long-term trend analysis, model training, and system performance evaluation.

[0252] Data retrieval, establish a multi-dimensional semantic index system, and realize multi-modal data retrieval. Provide a real-time query interface to retrieve the latest data in the current landslide perception system in real time. Allow users to retrieve and analyze historical data according to time range and key parameters, support database knowledge iteration, and adapt to the long-term risk assessment needs of the system.

[0253] In addition, data twins include three-dimensional geological models and geographic information models. Three-dimensional geological model. Use slope geometry data to build a three-dimensional geological model to show the physical geometric characteristics and relationships of underground structures and different strata of the slope. Combined with sensing equipment to collect data, calculate the mechanical parameters of the geological body, simulate the deformation of the geological body, predict the change range and movement trend of the slope rock and soil body, and analyze the dynamic evolution process of the landslide. Geographic information model integrates real-time sensing data with the geographic information system (GIS) to provide accurate spatial information of landslides and their monitoring points in geodetic coordinates, and support spatial analysis and real-life three-dimensional visualization.

[0254] Furthermore, reference Figure 5 ,The display layer includes modules such as perception data display, ,three-dimensional model display, decision result display, project information display, ,automatic design display, and automatic report display.

[0255] The perception data display module sets up a real-time monitoring dashboard on the digital twin model interface, provides intuitive and dynamic display of key perception data, and presents it in color, charts and numbers. It provides interactive historical trend query charts, allowing users to select specific time periods and perception parameters as needed, supports overlapping time period data display, and multiple content display in the same time period.

[0256] The twin model display module has a built-in highly interactive 3D digital twin model. Users can translate, zoom and rotate to gain in-depth insights into the structural information and stratum distribution of different rock and soil bodies, and display the topography of potential landslide areas, including ridges, rivers and other geographical elements. It also supports point-and-click information query of components.

[0257] The decision-making result display module provides a detailed risk assessment report, combining charts, graphs and text descriptions to highlight the results of expert evaluation and intelligent analysis and prediction, and present a panoramic view of the landslide risk status to decision makers. The alarm information and notifications generated by the system are displayed in real time in the digital twin model, and timely risk warnings are provided in the form of pop-up windows, color markings, etc.

[0258] In summary, the present invention discloses a landslide intelligent decision-making method and system based on the Internet of Things sensing technology. The Internet of Things sensing technology provided by the present invention can sense various parameters of the slope in real time, such as inclination, displacement, and underground Internet of Things sensing technology can collect landslide monitoring data in real time, providing timely reference basis for landslide intelligent decision-making; it can simultaneously and continuously collect a variety of landslide sensing data, including surface displacement, stress, deformation, surface cracks, groundwater level, etc., to provide a comprehensive analysis basis for landslide intelligent decision-making. Automatically collecting, transmitting, and processing landslide sensing data can effectively reduce manual intervention, improve the efficiency of landslide intelligent decision-making, and provide an economical data solution for landslide intelligent decision-making.

[0259] At the same time, an intelligent decision-making system based on the expert evaluation system was constructed. The system can effectively integrate the massive data collected by the IoT sensing technology to provide scientific decision-making support for landslide early warning and emergency response. The system can directly collect landslide monitoring data from IoT sensing devices, pre-process the data, use data mining, machine learning and other technologies to analyze and evaluate the landslide monitoring data, integrate expert knowledge graphs into the decision-making model, and combine the landslide risk assessment results to provide scientific decision-making support for landslide early warning and emergency response, effectively avoid decision-making bias, and improve the scientificity, rationality and timeliness of decision-making.

[0260] Furthermore, the present invention realizes the interaction between the sensing data and the digital twin model. This interaction can effectively improve the accuracy and efficiency of intelligent decision-making for landslides. The landslide digital twin model is a virtual landslide constructed using data collected by the Internet of Things sensing technology. The model can simulate the evolution of the landslide, assess the landslide risk, and can be updated according to real-time sensing data to reflect the latest status of the landslide, providing a more accurate reference for landslide warning and emergency response. At the same time, the technical threshold of decision makers can be greatly reduced, and the scope of application of the system can be expanded.

[0261] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0262] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions.

[0263] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments after knowing the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0264] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention should also include these modifications and variations.

Claims

1. A landslide intelligent decision-making method based on Internet of Things sensing technology, characterized in that: include: Receive and process multi-source perception data collected by pre-deployed slope data perception equipment, combine the slope surface point cloud data obtained by photogrammetry and 3D laser scanning, and construct a virtual slope space model; According to multi-source sensing data and at least one adaptive learning model, and by introducing a landslide disaster knowledge graph for analysis, a disaster intelligent decision-making result including at least one of a landslide boundary identification result, a landslide status assessment result and a landslide disaster prediction result is obtained, and the result is fed back into a virtual slope space model; Based on the results of intelligent disaster decision-making, multiple disaster prevention plans are generated by combining the acquired multi-source perception data and the knowledge graph associated with slope disasters, and the best disaster prevention plan is selected according to the lowest cost; Using an automated design program, the slope reinforcement design parameters are generated for the best disaster prevention solution. After optimization calculation, the optimal design solution is generated and fed back into the virtual slope space model.

2. The landslide intelligent decision-making method based on Internet of Things sensing technology as claimed in claim 1 is characterized in that: According to multi-source sensing data and at least one adaptive learning model, and by introducing a landslide disaster knowledge graph for analysis, a disaster intelligent decision result including at least one of a landslide boundary identification result, a landslide status assessment result and a landslide disaster prediction result is obtained, and fed back to the virtual slope space model, including: Add the MobileNet module as the backbone network based on the pre-trained DeepLabV3+ network, and adjust the number of categories in the output layer according to task requirements to obtain a lightweight recognition model. Use the lightweight recognition model to identify landslide boundaries in the acquired target area images. Establish a landslide disaster evaluation index system, combine the AHP-GMR correlation fusion mechanism to perform significance and correlation analysis and judgment of multi-source perception data, and output the landslide disaster status evaluation results through data fusion; The prediction model is constructed using the EMD-Elman learning algorithm, and the prediction model is trained based on multi-source perception data and the introduced previous slope disaster related knowledge graph to output the landslide disaster prediction results; The landslide boundary identification, landslide hazard status evaluation results and landslide hazard prediction results are fed back into the virtual slope space model as the results of intelligent disaster decision-making for visualization.

3. The landslide intelligent decision-making method based on Internet of Things sensing technology as claimed in claim 2 is characterized in that: Add the MobileNet module as the backbone network based on the pre-trained DeepLabV3+ network, and adjust the number of categories in the output layer according to task requirements to obtain a lightweight recognition model. Use the lightweight recognition model to perform landslide boundary recognition on the acquired target area image, including: The remote sensing image dataset of the landslide area is collected by drone equipment, and the slope area is annotated and labeled based on expert experience; Create data storage objects for remote sensing image datasets and corresponding labels respectively; In response to the task requirements, the pre-trained DeepLabV3+ network was selected as the basic model, the MobileNet module was added to the basic model as the backbone network, and the number of output layer categories of the model was adjusted to be consistent with the task requirements, thus obtaining a lightweight landslide impact area recognition model; Pass the data storage object, lightweight landslide impact area identification model and pre-configured training option parameters to the TrainNetwork function for model training; The preprocessed remote sensing image of the target area is obtained and loaded, and the loaded remote sensing image of the target area is input into the trained lightweight landslide impact area recognition model to output the recognition result of the landslide boundary.

4. The landslide intelligent decision-making method based on Internet of Things sensing technology as claimed in claim 2 is characterized in that: A landslide disaster evaluation index system is established, and the significance and correlation analysis and judgment of multi-source perception data are carried out in combination with the AHP-GMR correlation fusion mechanism. The landslide disaster status evaluation results are output through data fusion, including: According to the obtained landslide hazard evaluation index data, a landslide hazard evaluation index system including multiple factors affecting slope stability is constructed; wherein each factor is configured with multiple indicators to form a factor indicator sequence; Collect corresponding quantitative or qualitative evaluation data according to the factors in the system, convert the qualitative evaluation indicators into quantitative ones, and generate the original factor matrix X; Select a factor from the original factor matrix X as the reference evaluation index sequence to form the evaluation index sequence X e ; The ratio of the factors in the original factor matrix X to the average value of all factors is used as the standardized factor value of the factor to form the standardized factor matrix Z, and the X in the evaluation index sequence is used as the standardized factor value of the factor. e The ratio of the evaluation index to the average value of all evaluation indexes is used as the standardized evaluation index value to form a standardized evaluation index sequence Z e ; Using the AHP-GMR association fusion mechanism, each element in the standardized factor matrix Z and the standardized evaluation index sequence Z are calculated one by one. e The absolute difference of the corresponding elements, and determine the maximum difference and the minimum difference; Based on the maximum difference and the minimum difference, calculate the standardized factor matrix Z and the standardized evaluation index sequence Z e The correlation coefficient at each moment, and the average value of the correlation coefficient at each moment is taken as the correlation between the jth factor and the corresponding indicator; Sort by the degree of association to form an association sequence; Select the factors in the original factor matrix X as the evaluation index sequence to calculate the correlation degree, and combine the correlation sequences obtained by each calculation to obtain the correlation matrix; For multi-source sensing data, the correlation degree in the correlation matrix is ​​used as the weight for data fusion, and the intelligent evaluation index FoS of landslide disaster is obtained through normalization processing; The Kriging difference algorithm is used to interpolate the global spatial data of the smart evaluation index FoS to achieve a continuous expression of the smart evaluation index FoS and transmit it back to the virtual slope spatial model; in, The original factor matrix X is: In the formula, n is the number of columns and m is the number of rows; Evaluation index sequence X e for: X e =(x e1 ,x e2 ,x e3 ,…,x em ) T ; In the formula, e is the selected factor; The normalized factor matrix Z is: From={from ij }; In the formula, x ij represents the i-th index value of the j-th factor; Standardized evaluation index sequence Z e for: WITH e =(from e1 ,With e2 ,With e3 ,…,With em ) T ; The minimum difference is: Δmin j =min|Z ij -WITH ei |; The maximum difference is: Δmax j =max|Z ij -Z ei |,i=1,2,3......m; The correlation coefficient at each moment is: In the formula, ρ is the resolution coefficient, which is [0,1]. The smaller ρ is, the greater the sensitivity is, the smaller the correlation coefficient is, and Δ is x. ij The difference with △min; The correlation between the jth factor and the corresponding indicator is: FoS, the intelligence evaluation index of each point i for:

5. The landslide intelligent decision-making method based on Internet of Things sensing technology as claimed in claim 2 is characterized in that: The prediction model is constructed using the EMD-Elman learning algorithm. The prediction model is trained based on multi-source perception data and the introduced previous slope disaster related knowledge graph to output landslide disaster prediction results including: Acquire multi-source sensing data, and perform preprocessing operations including denoising, normalization, and extracting increment and acceleration characteristic curves related to landslide prediction from the normalized data. Decomposing the increment and acceleration characteristic curves related to landslide prediction to establish a learning data set containing trend terms, periodic terms and random terms; The empirical mode decomposition method is used to decompose the preprocessed multi-source data to obtain several IMF components; For each IMF component, multiple Elman neural network prediction sub-models are constructed; Using the learning data set and the introduced knowledge graph of slope hazards, each prediction sub-model is trained. At the same time, the prediction results are compared with the actual data to determine whether the prediction results meet the set RMSE threshold. If the set RMSE threshold is met, the prediction results of all trained prediction sub-models are integrated using a weighted average scheme to obtain a comprehensive prediction result curve. According to the characteristics of multi-source perception data, the parameters of each prediction sub-model are combined into a prediction weight matrix for the landslide and then stored.

6. A landslide intelligent decision-making system based on Internet of Things sensing technology, characterized in that: include: The data layer is configured with slope data sensing devices for collecting multi-source sensing data, a data processing module for data input, storage and retrieval, and a digital twin module for receiving and processing multi-source sensing data collected by pre-deployed slope data sensing devices, combining the slope surface point cloud data obtained by photogrammetry and three-dimensional laser scanning, and constructing a virtual slope space model; The decision-making layer is configured with an expert system for establishing an expert knowledge graph by integrating professional knowledge, experience and rules, a disaster intelligent decision-making module for obtaining a disaster intelligent decision-making result including at least one of a landslide boundary identification result, a landslide status assessment result and a landslide disaster prediction result based on multi-source perception data and at least one adaptive learning model, and feeding back the disaster intelligent decision-making result to the virtual slope space model, and an optimal disaster prevention scheme generation module for generating multiple disaster prevention schemes based on the disaster intelligent decision-making result, combining the obtained multi-source perception data and the slope disaster related knowledge graph, and selecting the optimal disaster prevention scheme according to the lowest cost, and an automated reinforcement module for generating slope reinforcement design parameters for the optimal disaster prevention scheme using an automated design program, generating an optimal design scheme after optimization calculation, and feeding back the optimal design scheme to the virtual slope space model; The display layer is configured with a perception data display unit, a 3D model display unit, a decision result display unit, a project information display unit, an automatic design display unit, and an automatic report display unit; The management layer is configured with a user authority management unit, a project authority management unit, a log management unit, a security management unit, and a data management unit.

7. The landslide intelligent decision-making system based on Internet of Things sensing technology as claimed in claim 6 is characterized in that: Disaster intelligent decision-making modules include: The landslide boundary recognition unit is used to add the MobileNet module as the backbone network based on the pre-trained DeepLabV3+ network, and adjust the number of categories in the output layer according to the task requirements to obtain a lightweight recognition model. The lightweight recognition model is used to identify the landslide boundary of the acquired target area image; The landslide disaster status evaluation unit is used to establish a landslide disaster evaluation index system, combine the AHP-GMR correlation fusion mechanism to perform significance and correlation analysis and judgment of multi-source perception data, and output the landslide disaster status evaluation results through data fusion; The landslide disaster prediction unit is used to build a prediction model using the EMD-Elman learning algorithm, train the prediction model based on multi-source perception data and the introduced previous slope disaster related knowledge graph to output the landslide disaster prediction results; The feedback and display unit is used to feed back the landslide boundary identification, landslide disaster status evaluation results and landslide disaster prediction results as disaster intelligent decision-making results into the virtual slope space model for visual display.

8. The landslide intelligent decision-making method based on Internet of Things sensing technology as claimed in claim 7 is characterized in that: The landslide boundary identification unit includes: The label generation subunit is used to collect remote sensing image datasets of the landslide area through UAV equipment, and to annotate the slope area and generate labels based on expert experience; A memory generation subunit is used to create data memory objects for remote sensing image datasets and corresponding labels respectively; The model lightweight subunit is used to select the pre-trained DeepLabV3+ network as the basic model according to the acquired task requirements, add the MobileNet module as the backbone network on the basic model, and adjust the number of categories in the output layer of the model to be consistent with the task requirements, so as to obtain a lightweight landslide impact area identification model; The lightweight module training subunit is used to pass the data storage object, the lightweight landslide impact area identification model and the pre-configured training option parameters to the TrainNetwork function for model training; The landslide boundary recognition result subunit is used to obtain and load the pre-processed remote sensing image of the target area, and input the loaded remote sensing image of the target area into the trained lightweight landslide impact area recognition model to output the recognition result of the landslide boundary.

9. The landslide intelligent decision-making method based on Internet of Things sensing technology as claimed in claim 7 is characterized in that: The landslide hazard status assessment units include: The system construction subunit is used to construct a landslide hazard evaluation index system including multiple factors affecting slope stability according to the obtained landslide hazard evaluation index data; wherein each factor is configured with multiple indicators to form a factor indicator sequence; The original factor matrix generation subunit is used to collect corresponding quantitative or qualitative evaluation data according to the factors in the system, convert the qualitative evaluation indicators into quantitative ones, and generate the original factor matrix X; Select a factor from the original factor matrix X as the evaluation index sequence to form the evaluation index sequence X e ; The standardization subunit is used to take the ratio of the factors in the original factor matrix X to the average value of all factors as the standardized factor value of the factor to form the standardized factor matrix Z, and to convert the X in the evaluation index sequence into e The ratio of the evaluation index to the average value of all evaluation indexes is used as the standardized evaluation index value to form a standardized evaluation index sequence Z e ; The difference determination subunit is used to calculate the difference between each element in the standardized factor matrix Z and the standardized evaluation index sequence Z one by one by using the AHP-GMR association fusion mechanism. e The absolute difference of the corresponding elements, and determine the maximum difference and the minimum difference; The correlation degree obtaining subunit is used to calculate the standardized factor matrix Z and the standardized evaluation index sequence Z based on the maximum difference and the minimum difference. e The correlation coefficient at each moment, and the average value of the correlation coefficient at each moment is taken as the correlation between the jth factor and the corresponding indicator; The association relationship determination subunit is used to sort the associations according to the size of the association degree to form an association sequence, select the factors in the original factor matrix X in turn as the evaluation index sequence to calculate the association degree, and combine the association sequences obtained by each calculation to obtain the association matrix; The smart evaluation index output subunit is used to fuse the multi-source perception data using the correlation degree in the correlation matrix as the weight, and obtain the smart evaluation index FoS of landslide disaster through normalization processing; The interpolation subunit is used to interpolate the global spatial data of the smart evaluation index FoS using the Kriging interpolation algorithm, realize the continuous expression of the smart evaluation index FoS, and transmit it back to the virtual slope space model; in, The original factor matrix X is: In the formula, n is the number of columns and m is the number of rows; Evaluation index sequence X e for: X e =(x e1 ,x e2 ,x e3 ,…,x em ) T ; In the formula, e is the selected factor; The normalized factor matrix Z is: From={from ij }; In the formula, x ij represents the i-th index value of the j-th factor; Standardized evaluation index sequence Z e for: WITH e =(from e1 ,With e2 ,With e3 ,…,With em ) T ; The minimum difference is: Δmin j =min|Z ij -WITH ei |; The maximum difference is: Δmax j =max|Z ij -Z ei |,i=1,2,3......m; The correlation coefficient at each moment is: In the formula, ρ is the resolution coefficient, which is [0,1]. The smaller ρ is, the greater the sensitivity is, the smaller the correlation coefficient is, and Δ is X. ij The difference with △min; The correlation between the jth factor and the corresponding indicator is: FoS, the intelligence evaluation index of each point i for:

10. The landslide intelligent decision-making method based on Internet of Things sensing technology according to claim 7 is characterized in that: The landslide hazard prediction unit includes: A preprocessing subunit is used to obtain multi-source sensing data, and perform preprocessing operations including denoising and normalization on the collected multi-source sensing data, and extracting increment and acceleration characteristic curves related to landslide prediction from the normalized data; A learning data set output subunit is used to decompose the increment and acceleration characteristic curves related to landslide prediction to establish a learning data set containing trend items, period items and random items; The prediction sub-model construction sub-unit is used to decompose the pre-processed multi-source data using the empirical mode decomposition method to obtain several IMF components, and for each IMF component, construct multiple Elman neural network prediction sub-models respectively; The training and judgment subunit is used to train each prediction submodel using the learning data set and the introduced knowledge graph of previous slope hazards, and compare the prediction results with the actual data to determine whether the prediction results meet the set RMSE threshold. The fusion and storage subunit is used to fuse the prediction results of all trained prediction sub-models by using a weighted average scheme if the set RMSE threshold is met, to obtain a comprehensive prediction result curve, and to combine the parameters of each prediction sub-model into a prediction weight matrix for the landslide according to the characteristics of multi-source perception data, and then store it.

Images