Internet-of-things method and system for monitoring and early warning collapsible loess foundation collapse

Abstract

The invention discloses an internet-of-things method and system for collapsible loess foundation collapse monitoring and early warning, and relates to the technical field of foundation monitoring. The invention relates to an internet-of-things system for monitoring and early warning collapsible loess foundation collapse. The internet-of-things system comprises a loess foundation partition division module and a loess foundation monitoring and early warning module. According to the method, the minimum monitoring unit is constructed, the rainfall data, the earth surface data and the rainfall prediction evaluation model are combined, and the time-space depth fusion is performed on the rainfall data by using the CNN and the LSTM model, so that the accuracy of rainfall prediction is improved, and accurate monitoring and early warning of the collapsible loess foundation area are realized; the monitoring area can be divided into a plurality of early warning subareas through the depth clustering technology in combination with the rainfall condition and the earth surface data, the early warning accuracy and response speed are improved, disasters caused by collapsible loess foundation collapse are effectively prevented, and fine management of the loess foundation area is achieved.

Description

Technical Field

[0001] The present invention relates to the field of foundation monitoring technology, and in particular to an Internet of Things method and system for monitoring and early warning of collapsible loess foundation subsidence. Background Art

[0002] Collapsible loess foundation refers to the phenomenon that it is prone to settlement or collapse due to factors such as loose loess layers, humidity changes and human activities. The potential risk of foundation collapse is common in loess areas, especially in areas with dense populations and accelerated urbanization. This problem is more prominent. Traditional foundation monitoring technologies mostly rely on simple geological surveys or analysis based on geological monitoring data. However, these methods have great limitations in real-time, accuracy and monitoring range.

[0003] It is necessary to conduct on-site investigations based on the characteristic properties of collapsible loess foundations, and to carry out reasonable monitoring and early warning in combination with precipitation prediction technology and zoning monitoring. When abnormal collapse is detected, disaster prevention resources should be deployed in a timely manner, such as arranging personnel evacuation and equipment deployment in advance, so as to improve the efficiency of emergency response and reduce the impact of disasters on people’s lives and property. Summary of the invention

[0004] The present invention aims to provide an Internet of Things method and system for monitoring and early warning of collapsible loess foundation subsidence, combining the actual characteristics of the loess foundation to perform zoning monitoring and improve the response speed.

[0005] The IoT method for monitoring and early warning of collapsible loess foundation collapse includes the following steps: Construct the minimum monitoring unit; divide the target early warning loess foundation area based on the minimum monitoring unit, and obtain the target early warning loess foundation monitoring unit J n , n=1, 2, …, N; N is the total number of minimum monitoring units contained in the target early warning loess foundation area; Obtain the precipitation data of the target warning loess foundation area; Based on the precipitation data of the target warning loess foundation area and the precipitation prediction evaluation model, the target warning loess foundation monitoring unit J n Analyze and obtain the precipitation risk value F of the target early warning monitoring unit n ; The target warning loess foundation area precipitation data contains precipitation spatial data and precipitation time data; the precipitation prediction and evaluation model is used to combine the CNN model and the LSTM model to predict and evaluate the precipitation risk situation; Obtain the surface data of the target warning loess foundation area; based on the surface data of the target warning loess foundation area and the target warning monitoring unit precipitation risk value F n and loess foundation zoning monitoring and analysis model for target early warning loess foundation monitoring unit J n Perform deep clustering to obtain the target early warning loess foundation monitoring zone C m, m=1, 2, …, M; M is the total number of target early warning loess foundation monitoring zones in the target early warning loess foundation area; the loess foundation zone monitoring and analysis model is used to perform deep clustering division of the target early warning loess foundation area in combination with precipitation conditions and surface data; In all target early warning loess foundation monitoring zone C m Install monitoring and early warning devices to warn the target loess foundation monitoring zone C at preset time intervals m Carry out monitoring and early warning, and obtain the target early warning signal X for loess foundation collapse m ; When receiving the target warning signal of loess foundation collapse X m If it is abnormal, a collapse warning will be issued; otherwise, no operation will be performed.

[0006] As a preferred technical solution of the present invention, the precipitation prediction and evaluation model includes a data processing layer, a precipitation spatial feature extraction layer, a precipitation time feature extraction layer and a partition precipitation risk prediction layer; The data processing layer is used to warn the loess foundation monitoring unit J according to the target n The precipitation spatial data in the target warning loess foundation area precipitation data is divided into precipitation area spatial data Q n , the precipitation time data in the target warning loess foundation area precipitation data is divided into precipitation area time data S n ; The precipitation spatial feature extraction layer is used to extract the precipitation spatial feature according to the precipitation area spatial data Q n Perform feature extraction to obtain the precipitation area spatial data feature Q n '; The precipitation time feature extraction layer is used to extract the precipitation time data S n Perform feature extraction to obtain the precipitation area time data feature S n '; The regional precipitation risk prediction layer is used to transform the spatial data characteristics Q n ' and precipitation area time data characteristics S n 'Perform feature matching to obtain the precipitation risk value F of the target early warning monitoring unit n .

[0007] As a preferred technical solution of the present invention, the specific steps of constructing a precipitation prediction and evaluation model include: The precipitation spatial feature extraction layer is constructed using the CNN model; the precipitation spatial feature extraction layer includes a convolution layer, a pooling layer, and a feature fusion layer; The precipitation time feature extraction layer is constructed using the LSTM model; the partition precipitation risk prediction layer is constructed using the multi-head attention mechanism, which includes two fully connected layers and an output layer; Specific steps for training the precipitation prediction evaluation model: Collect several groups of precipitation forecast evaluation training samples; each group of precipitation forecast evaluation training samples contains precipitation monitoring data and corresponding marked precipitation risk values; combine several groups of precipitation forecast evaluation training samples to obtain a precipitation forecast evaluation training set; The precipitation prediction and evaluation training set is input into the precipitation prediction and evaluation model for model training to obtain the initial precipitation prediction and evaluation model; the initial precipitation prediction and evaluation model is evaluated. If the initial precipitation prediction and evaluation model passes the model evaluation, the initial precipitation prediction and evaluation model is used as the precipitation prediction and evaluation model to complete the model training; otherwise, the precipitation prediction and evaluation training set is used to continue the model training.

[0008] As a preferred technical solution of the present invention, the loess foundation partition monitoring and analysis model includes a partition data processing layer, a partition matrix construction layer, a partition monitoring clustering layer, an optimized partition adjustment layer and a result output layer; The partition data processing layer is used to early-warn the loess foundation monitoring unit J according to the target n The surface data of the target early warning loess foundation area is divided into regional surface data D n ; Regional surface data D n And the target early warning monitoring unit precipitation risk value F n Perform feature fusion to obtain monitoring partition feature data R n ; The partition matrix construction layer is used to early-warn the loess foundation monitoring unit according to the target n and monitoring partition characteristic data R n Construct a matrix to obtain the loess foundation partition matrix; The partition monitoring clustering layer is used to cluster the loess foundation partition matrix and obtain the initial partition scheme of the loess foundation; The optimized partition adjustment layer is used to optimize the initial partition scheme of the loess foundation and obtain the target early warning loess foundation monitoring partition C. m ; The result output layer is used to output the target early warning loess foundation monitoring partition C m .

[0009] As a preferred technical solution of the present invention, the specific steps of clustering the loess foundation partition matrix in the partition monitoring clustering layer include: The partition monitoring clustering layer is constructed by the autoencoder layer, the graph convolution layer and the self-supervision layer; the autoencoder layer and the graph convolution layer are connected in parallel, and both the autoencoder layer and the graph convolution layer contain B layers, b=1, 2, ..., B; In the bth self-encoding layer, feature extraction is performed on the loess foundation partition matrix to obtain the loess foundation partition matrix feature (b); In the bth graph convolutional layer, the loess foundation partition matrix feature (b) is received and restructured to obtain the loess foundation partition matrix feature structure (b); In the self-supervised layer, clustering feature learning is performed on the loess foundation partition matrix features (b) and the loess foundation partition matrix feature structure (b) to obtain the initial partition scheme of the loess foundation; Among them, the partition monitoring clustering layer is optimized using the loss function Z, Z=Loss(α)+W1*Loss(β)+W2*Loss(γ), Loss(α) represents the loss function of the autoencoder layer, Loss(β) represents the clustering loss function of the autoencoder layer, W1 represents the weight of the clustering loss function of the autoencoder layer, Loss(γ) represents the clustering loss function of the graph convolution layer, and W2 represents the weight of the clustering loss function of the graph convolution layer.

[0010] As a preferred technical solution of the present invention, the specific steps of optimizing and adjusting the initial partitioning scheme of the loess foundation in the optimization partitioning adjustment layer include: Based on the initial partitioning scheme of loess foundation, K optimal loess foundation partitioning H is constructed k , k=1, 2, ..., K; the initial zoning scheme of loess foundation contains several isolated monitoring units; each optimized loess foundation zoning H k A number of isolated monitoring units are randomly divided; K optimized loess foundation partitions are divided into H k Combination, get the optimized loess foundation partition iterative population; set the maximum number of iterations, optimize the loess foundation partition H k The fitness is the fitness G k ; Calculate the fitness G k Specific steps: Optimizing the division of loess foundation k Calculate the silhouette coefficient and get the average silhouette coefficient; use the average silhouette coefficient as the optimal loess foundation partition H k The fitness G k ; The optimized loess foundation partitioning iterative population is updated, and a number of isolated monitoring units are reallocated in each population iteration using the adding point strategy to obtain a new optimized loess foundation partitioning iterative population; When the maximum number of iterations is reached, the optimized loess foundation partition corresponding to the maximum fitness is output, which is the optimal optimized loess foundation partition. According to the optimal optimized loess foundation partition, the target early warning loess foundation monitoring partition C is generated. m , m=1, 2,…, M, where M is the total number of target warning loess foundation monitoring zones in the target warning loess foundation area.

[0011] As a preferred technical solution of the present invention, the target early warning loess foundation monitoring zone C m The specific steps for monitoring and early warning include: Get target warning loess foundation monitoring zone C m Corresponding loess foundation monitoring data P m ; The loess foundation monitoring data P m Compared with the preset monitoring collapse warning threshold range, if the loess foundation monitoring data P m If the preset monitoring collapse warning threshold is exceeded, the target warning loess foundation collapse warning signal X m is abnormal; otherwise, the target warning signal of loess foundation collapse is X m is normal.

[0012] The IoT system for monitoring and early warning of collapsible loess foundation subsidence includes: The loess foundation zoning module includes a minimum unit setting unit, an index analysis unit and a zoning clustering unit; the minimum unit setting unit is used to construct a minimum monitoring unit; based on the minimum monitoring unit, the target early warning loess foundation area is divided to obtain the target early warning loess foundation monitoring unit J n , n=1, 2, …, N; N is the total number of minimum monitoring units in the target early warning loess foundation area; the index analysis unit is used to obtain the precipitation data of the target early warning loess foundation area; based on the precipitation data of the target early warning loess foundation area and the precipitation prediction evaluation model, the target early warning loess foundation monitoring unit J n Analyze and obtain the precipitation risk value F of the target early warning monitoring unit n ; The precipitation data of the target warning loess foundation area contains precipitation spatial data and precipitation time data; the precipitation prediction and evaluation model is used to combine the CNN model and the LSTM model to predict and evaluate the precipitation risk situation; the partition clustering unit is used to obtain the surface data of the target warning loess foundation area; based on the surface data of the target warning loess foundation area and the precipitation risk value F of the target warning monitoring unit n and loess foundation zoning monitoring and analysis model for target early warning loess foundation monitoring unit J n Perform deep clustering to obtain the target early warning loess foundation monitoring zone C m , m=1, 2, …, M; M is the total number of target early warning loess foundation monitoring zones in the target early warning loess foundation area; the loess foundation zone monitoring and analysis model is used to perform deep clustering division of the target early warning loess foundation area in combination with precipitation conditions and surface data; The loess foundation monitoring and early warning module includes a monitoring and early warning unit; the monitoring and early warning unit is used to monitor all target early warning loess foundation monitoring areas C mInstall monitoring and early warning devices to warn the target loess foundation monitoring zone C at preset time intervals m Carry out monitoring and early warning, and obtain the target early warning signal X for loess foundation collapse m ; When receiving the target warning signal of loess foundation collapse X m If it is abnormal, a collapse warning will be issued; otherwise, no operation will be performed.

[0013] The present invention has the following advantages: 1. The present invention constructs a minimum monitoring unit and combines precipitation data, surface data and a precipitation prediction and evaluation model, and uses CNN and LSTM models to perform deep spatiotemporal fusion of precipitation data, thereby improving the accuracy of precipitation prediction and realizing accurate monitoring and early warning of collapsible loess foundation areas. Through deep clustering technology, combined with precipitation conditions and surface data, the monitoring area can be divided into multiple early warning zones, each of which is equipped with a monitoring and early warning device to realize real-time monitoring and early warning. This method improves the accuracy and response speed of early warning, effectively prevents disasters caused by collapse of collapsible loess foundations, and realizes refined management of loess foundation areas.

[0014] 2. The present invention can realize accurate and efficient monitoring zoning through the loess foundation zoning monitoring and analysis model constructed based on precipitation conditions and surface data, combined with deep clustering and optimization adjustment; through multi-source data fusion and feature extraction, the scientificity and accuracy of zoning are significantly improved, and deep clustering combined with optimization adjustment ensures the homogeneity and spatial compactness within the partition; supports real-time adjustment and dynamic monitoring capabilities, adapts to changes in the external environment, improves the utilization efficiency of monitoring resources, and reduces the overall monitoring cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a schematic diagram of the structure of the Internet of Things system for monitoring and early warning of collapsible loess foundation subsidence adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0017] Embodiment 1, the Internet of Things method for monitoring and early warning of collapsible loess foundation collapse, comprises the following steps: Construct the minimum monitoring unit; divide the target early warning loess foundation area based on the minimum monitoring unit, and obtain the target early warning loess foundation monitoring unit J n , n=1, 2, …, N; N is the total number of minimum monitoring units contained in the target early warning loess foundation area; After dividing the area into the smallest monitoring unit, the soil characteristics, collapse risk and dynamic changes of each small area can be identified more accurately. The detailed unit division avoids the problem of attribute averaging caused by too large a monitoring range, improves the reliability of early warning, and provides data support for subsequent zoning monitoring and division. For example, if the target early warning loess foundation area is 500 square meters, the smallest monitoring unit can be set to 5 square meters or other. Based on the zoning division of the smallest unit, localized and zoned early warning can be achieved, and high-risk areas can be quickly located to facilitate emergency response. Obtain the precipitation data of the target warning loess foundation area; Based on the precipitation data of the target warning loess foundation area and the precipitation prediction evaluation model, the target warning loess foundation monitoring unit J n Analyze and obtain the precipitation risk value F of the target early warning monitoring unit n ; The target warning loess foundation area precipitation data contains precipitation spatial data and precipitation time data; the precipitation prediction and evaluation model is used to combine the CNN model and the LSTM model to predict and evaluate the precipitation risk situation; The precipitation prediction and evaluation model includes a data processing layer, a precipitation spatial feature extraction layer, a precipitation temporal feature extraction layer, and a partitioned precipitation risk prediction layer; The data processing layer is used to warn the loess foundation monitoring unit J according to the target n The precipitation spatial data in the target warning loess foundation area precipitation data is divided into precipitation area spatial data Q n , the precipitation time data in the target warning loess foundation area precipitation data is divided into precipitation area time data S n ; The precipitation spatial feature extraction layer is used to extract the precipitation spatial feature according to the precipitation area spatial data Q n Perform feature extraction to obtain the precipitation area spatial data feature Q n '; The precipitation time feature extraction layer is used to extract the precipitation time data S n Perform feature extraction to obtain the precipitation area time data feature S n '; The regional precipitation risk prediction layer is used to transform the spatial data characteristics Q n ' and precipitation area time data characteristics S n 'Perform feature matching to obtain the precipitation risk value F of the target early warning monitoring unit n ; Using the CNN model to extract features from precipitation spatial data can effectively capture spatial distribution patterns, such as differences in local rainfall intensity or changes in precipitation gradients within a region. The LSTM model can extract precipitation time data to identify precipitation trends and mutation points, capture long-term dependencies in time series, and combine spatial and temporal features in the partitioned precipitation risk prediction layer to more accurately assess precipitation risks and provide scientific risk values ​​for loess foundation monitoring units. By calculating independent risk values ​​for each monitoring unit, partitioned precipitation risk assessment can be achieved for the target area. By capturing spatial distribution features with CNN and identifying time series changes with LSTM, risks caused by sudden, regional heavy rainfall or continuous rainfall can be accommodated. The specific steps to build a precipitation prediction and evaluation model include: The precipitation spatial feature extraction layer is constructed using the CNN model; the precipitation spatial feature extraction layer includes a convolution layer, a pooling layer, and a feature fusion layer; using CNN to extract local and global features of precipitation spatial distribution can more accurately characterize regional rainfall patterns; The precipitation time feature extraction layer is constructed using the LSTM model; the partition precipitation risk prediction layer is constructed using the multi-head attention mechanism, which includes two fully connected layers and an output layer; LSTM’s dependency modeling on precipitation time series can effectively predict rainfall intensity changes and future trends; the multi-head attention mechanism combines spatial and temporal features, fully considering the interaction between the two, and enhancing the scientific nature of precipitation risk prediction; Specific steps for training the precipitation prediction evaluation model: Collect several groups of precipitation forecast evaluation training samples; each group of precipitation forecast evaluation training samples contains precipitation monitoring data and corresponding marked precipitation risk values; combine several groups of precipitation forecast evaluation training samples to obtain a precipitation forecast evaluation training set; The precipitation prediction evaluation training set is input into the precipitation prediction evaluation model for model training to obtain the initial precipitation prediction evaluation model; the initial precipitation prediction evaluation model is evaluated, and if the initial precipitation prediction evaluation model passes the model evaluation, the initial precipitation prediction evaluation model is used as the precipitation prediction evaluation model to complete the model training; otherwise, the precipitation prediction evaluation training set is used to continue the model training; The zoning analysis model can provide an independent precipitation risk value for each monitoring unit, realize accurate early warning, implement targeted monitoring and prevention and control strategies for different regions according to the risk value, and improve resource utilization efficiency; Obtain the surface data of the target warning loess foundation area; based on the surface data of the target warning loess foundation area and the target warning monitoring unit precipitation risk value F n and loess foundation zoning monitoring and analysis model for target early warning loess foundation monitoring unit J nPerform deep clustering to obtain the target early warning loess foundation monitoring zone C m , m=1, 2, …, M; M is the total number of target early warning loess foundation monitoring zones in the target early warning loess foundation area; the loess foundation zone monitoring and analysis model is used to perform deep clustering division of the target early warning loess foundation area in combination with precipitation conditions and surface data; The loess foundation zoning monitoring and analysis model includes zoning data processing layer, zoning matrix construction layer, zoning monitoring clustering layer, optimized zoning adjustment layer and result output layer; The partition data processing layer is used to early-warn the loess foundation monitoring unit J according to the target n The surface data of the target early warning loess foundation area is divided into regional surface data D n ; Regional surface data D n And the target early warning monitoring unit precipitation risk value F n Perform feature fusion to obtain monitoring partition feature data R n ; The partition matrix construction layer is used to early-warn the loess foundation monitoring unit according to the target n and monitoring partition characteristic data R n Construct a matrix to obtain the loess foundation partition matrix; The partition monitoring clustering layer is used to cluster the loess foundation partition matrix and obtain the initial partition scheme of the loess foundation; The optimized partition adjustment layer is used to optimize the initial partition scheme of the loess foundation and obtain the target early warning loess foundation monitoring partition C. m ; The result output layer is used to output the target early warning loess foundation monitoring partition C m ; The precipitation risk value and regional surface data are integrated to construct comprehensive monitoring zoning features, which effectively improves the scientific nature of zoning division; the dynamic impact of precipitation on loess foundation and the static properties of surface features are taken into account to make the zoning results more realistic; the constructed loess foundation zoning matrix contains multi-dimensional information such as space, time and precipitation risk, which can more comprehensively describe regional characteristics and provide accurate basic data for clustering; the zoning monitoring clustering layer fully explores the characteristic patterns of loess foundation through a deep clustering algorithm, divides monitoring units with similar attributes into the same area, ensures the homogeneity within the zoning, and can identify complex geological features, precipitation effects and nonlinear relationships between regional attributes; the optimization zoning adjustment layer is based on the initial zoning scheme, further considers factors such as spatial continuity, compactness and monitoring resource allocation of the zoning, optimizes the zoning results, and makes the zoning division more valuable for practical application; The specific steps of clustering the loess foundation partition matrix in the partition monitoring clustering layer include: The partition monitoring clustering layer is constructed by the autoencoding layer, the graph convolution layer and the self-supervision layer; the autoencoding layer and the graph convolution layer are connected in parallel, and both the autoencoding layer and the graph convolution layer contain a B layer, b=1, 2, ..., B; the specific value of B is set by professional technicians according to actual conditions; In the bth self-encoding layer, feature extraction is performed on the loess foundation partition matrix to obtain the loess foundation partition matrix feature (b); In the bth graph convolutional layer, the loess foundation partition matrix feature (b) is received and restructured to obtain the loess foundation partition matrix feature structure (b); In the self-supervised layer, clustering feature learning is performed on the loess foundation partition matrix features (b) and the loess foundation partition matrix feature structure (b) to obtain the initial partition scheme of the loess foundation; Among them, the partition monitoring clustering layer is optimized using the loss function Z, Z=Loss(α)+W1*Loss(β)+W2*Loss(γ), Loss(α) represents the loss function of the autoencoder layer, Loss(β) represents the clustering loss function of the autoencoder layer, W1 represents the weight of the clustering loss function of the autoencoder layer, Loss(γ) represents the clustering loss function of the graph convolution layer, and W2 represents the weight of the clustering loss function of the graph convolution layer; the specific parameters of W1 and W2 are set by professional and technical personnel according to actual conditions; The autoencoder layer captures the global nonlinear characteristics of the loess foundation partition matrix, such as the comprehensive characteristics of precipitation and surface data. The graph convolution layer uses the structural information of the partition matrix to reorganize the features through the adjacency matrix and learn the spatial correlation between regions. The two are connected in parallel to ensure that the global characteristics and structural information of the partition matrix are fully integrated, which significantly improves the accuracy and scientificity of the partition. The deep features of the loess foundation partition are gradually extracted through a multi-layer design to ensure that features from low to high levels are captured, providing a richer representation for clustering. Taking into account the characteristics of the autoencoder layer and the graph convolution layer, the weights of the two clustering losses are adjusted through W1 and W2 to dynamically balance the global characteristics. The importance of features and local structural characteristics; according to different loess foundation partition characteristics (such as significant surface attributes or dominant precipitation influence), the weights can be adjusted to adapt to different scenarios; the self-supervision layer clusters the features of the self-encoding layer and the graph convolution layer to capture the deep correlation between the features, and generates an initial partitioning scheme based on the similarity of precipitation risks and the consistency of surface attributes, so that the monitoring units in the same area have a high degree of consistency; the self-encoding layer performs dimensionality reduction processing on the partition matrix, making clustering feature learning more efficient and reducing computational complexity; using the dynamic optimization mechanism of the loss function, combined with the weight feedback adjustment of different losses, the partitioning performance is gradually improved; The specific steps of optimizing and adjusting the initial zoning scheme of loess foundation in the optimization zoning adjustment layer include: Based on the initial partitioning scheme of loess foundation, K optimal loess foundation partitioning H is constructed k , k=1, 2, ..., K; the initial zoning scheme of loess foundation contains several isolated monitoring units; each optimized loess foundation zoning H k A number of isolated monitoring units are randomly divided; K optimized loess foundation partitions are divided into H k Combination, get the optimized loess foundation partition iterative population; set the maximum number of iterations, optimize the loess foundation partition H k The fitness is the fitness G k ;The maximum number of iterations is set by professional technicians based on actual conditions; When the characteristics of some data points are too different from those of other points and cannot be attributed to any cluster center, isolated monitoring units may appear. Isolated monitoring units are usually located at the edge or abnormal area of ​​the data distribution. They may be caused by data noise, incomplete data collection, real sparse data characteristics, or insufficient robustness of the model to abnormal data. Further partition adjustment is required for isolated monitoring units. Calculate the fitness G k Specific steps: Optimizing the division of loess foundation k Calculate the silhouette coefficient and get the average silhouette coefficient; use the average silhouette coefficient as the optimal loess foundation partition H k The fitness G k ; The optimized loess foundation partitioning iterative population is updated, and a number of isolated monitoring units are reallocated in each population iteration using the adding point strategy to obtain a new optimized loess foundation partitioning iterative population; When the maximum number of iterations is reached, the optimized loess foundation partition corresponding to the maximum fitness is output, which is the optimal optimized loess foundation partition. According to the optimal optimized loess foundation partition, the target early warning loess foundation monitoring partition C is generated. m , m=1, 2, …, M, where M is the total number of target early warning loess foundation monitoring zones in the target early warning loess foundation area; The silhouette coefficient is used as the fitness index to quantify the similarity within the partition and the difference between the partitions, providing a scientific evaluation standard for optimizing the partition. The high fitness partition ensures the internal consistency and external difference of the partition, making the characteristics of the monitoring units in the same area closer and the characteristics between different areas clearly distinguished; the isolated monitoring units in the initial partition are reallocated to avoid the loss or invalidity of the isolated monitoring units. The multiple iterative explorations of random allocation improve the diversity of the partition results and help find a better partition scheme; the point addition strategy is combined with the fitness evaluation to dynamically optimize the partition structure and eliminate unreasonable isolation phenomena; the population update mechanism is used to avoid the local optimal problem of a single solution, which is particularly suitable for dealing with the problem of large-scale loess foundation regional partitioning with complex geological characteristics and diversified data. The efficient dynamic adjustment method of isolated monitoring units is suitable for fine-grained partition optimization in large-scale areas; In all target early warning loess foundation monitoring zone C m Install monitoring and early warning devices to warn the target loess foundation monitoring zone C at preset time intervals m Carry out monitoring and early warning, and obtain the target early warning signal X for loess foundation collapse m ; When receiving the target warning signal of loess foundation collapse X m When it is abnormal, a collapse warning is issued; otherwise, no operation is performed; the preset time interval changes according to the different monitoring areas; if the probability of collapse in the area is high, the monitoring frequency is increased; otherwise, the monitoring frequency is reduced, and the specific time interval is adjusted and set according to the actual situation; Target early warning loess foundation monitoring zone C m The specific steps for monitoring and early warning include: Get target warning loess foundation monitoring zone C m Corresponding loess foundation monitoring data P m ; The loess foundation monitoring data P m Compared with the preset monitoring collapse warning threshold range, if the loess foundation monitoring data P m If the preset monitoring collapse warning threshold is exceeded, the target warning loess foundation collapse warning signal X m is abnormal; otherwise, the target warning signal of loess foundation collapse is X m is normal; Adjust the monitoring frequency according to the risk of collapse in the subarea. Increase the monitoring frequency in high-risk subareas to capture abnormal signals more timely and reduce the response delay when collapse occurs. Reduce the monitoring frequency in low-risk subareas to reduce unnecessary monitoring, save resources and maintain the early warning effect. Compare the real-time monitoring data with the preset threshold range to effectively identify abnormal signals beyond the safety range and ensure the accuracy of the early warning. The preset threshold range is set by professional technicians according to the actual situation. Different threshold ranges can be set according to the different attributes of different partitions. The time interval can be flexibly set according to the actual situation of the partition, supporting dynamic monitoring strategies under different geological conditions, rainfall intensity or external environmental influences, and enhancing early warning adaptability; the monitoring frequency and early warning threshold are independently set according to the characteristics of each partition, providing more targeted management and early warning solutions; High-frequency monitoring and real-time early warning reduce losses when collapse disasters occur and buy time for disaster prevention. Enhanced monitoring of high-risk areas can identify collapse risks in advance and avoid the spread of disasters or the occurrence of secondary disasters. At the same time, it can adapt to the geological and environmental characteristics of different loess foundation areas and support diverse monitoring and early warning needs.

[0018] Example 2, IoT system for monitoring and early warning of collapsible loess foundation subsidence, see Figure 1 As shown, including: The loess foundation zoning module includes a minimum unit setting unit, an index analysis unit and a zoning clustering unit; the minimum unit setting unit is used to construct a minimum monitoring unit; based on the minimum monitoring unit, the target early warning loess foundation area is divided to obtain the target early warning loess foundation monitoring unit J n , n=1, 2, …, N; N is the total number of minimum monitoring units in the target early warning loess foundation area; the index analysis unit is used to obtain the precipitation data of the target early warning loess foundation area; based on the precipitation data of the target early warning loess foundation area and the precipitation prediction evaluation model, the target early warning loess foundation monitoring unit J n Analyze and obtain the precipitation risk value F of the target early warning monitoring unit n ; The precipitation data of the target warning loess foundation area contains precipitation spatial data and precipitation time data; the precipitation prediction and evaluation model is used to combine the CNN model and the LSTM model to predict and evaluate the precipitation risk situation; the partition clustering unit is used to obtain the surface data of the target warning loess foundation area; based on the surface data of the target warning loess foundation area and the precipitation risk value F of the target warning monitoring unit n and loess foundation zoning monitoring and analysis model for target early warning loess foundation monitoring unit J n Perform deep clustering to obtain the target early warning loess foundation monitoring zone C m , m=1, 2, …, M; M is the total number of target early warning loess foundation monitoring zones in the target early warning loess foundation area; the loess foundation zone monitoring and analysis model is used to perform deep clustering division of the target early warning loess foundation area in combination with precipitation conditions and surface data; The loess foundation monitoring and early warning module includes a monitoring and early warning unit; the monitoring and early warning unit is used to monitor all target early warning loess foundation monitoring areas C m Install monitoring and early warning devices to warn the target loess foundation monitoring zone C at preset time intervalsm Carry out monitoring and early warning, and obtain the target early warning signal X for loess foundation collapse m ; When receiving the target warning signal of loess foundation collapse X m If it is abnormal, a collapse warning will be issued; otherwise, no operation will be performed.

[0019] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. An IoT method for monitoring and early warning of collapsible loess foundation subsidence, characterized in that: The following steps are involved: Construct the minimum monitoring unit; divide the target early warning loess foundation area based on the minimum monitoring unit, and obtain the target early warning loess foundation monitoring unit J n , n=1,2,…,N; N is the total number of minimum monitoring units included in the target early warning loess foundation area; Obtain precipitation data for the target warning loess foundation area; Based on the precipitation data of target early warning loess foundation area and precipitation prediction evaluation model, the target early warning loess foundation monitoring unit J n Analyze and obtain the precipitation risk value F of the target early warning monitoring unit n ; The target warning loess foundation area precipitation data contains precipitation spatial data and precipitation time data; the precipitation prediction and evaluation model is used to combine the CNN model and the LSTM model to predict and evaluate the precipitation risk situation; Obtain surface data of the target early warning loess foundation area; Based on the surface data of the loess foundation area for target early warning and the precipitation risk value F of the target early warning monitoring unit n and loess foundation zoning monitoring and analysis model for target early warning loess foundation monitoring unit J n Perform deep clustering to obtain the target early warning loess foundation monitoring zone C m , m=1,2,…,M; M is the total number of target early warning loess foundation monitoring subareas in the target early warning loess foundation area; The loess foundation zoning monitoring and analysis model is used to combine precipitation conditions and surface data to perform in-depth clustering of the target warning loess foundation area; In all target early warning loess foundation monitoring zone C m Install monitoring and early warning devices to warn the target loess foundation monitoring zone C at preset time intervals m Carry out monitoring and early warning, and obtain the target early warning signal X for loess foundation collapse m ; When receiving the target warning signal of loess foundation collapse X m When it is abnormal, a collapse warning is issued; Otherwise, no action.

2. The IoT method for monitoring and early warning of collapsible loess foundation subsidence according to claim 1 is characterized in that: The precipitation prediction and evaluation model includes a data processing layer, a precipitation spatial feature extraction layer, a precipitation temporal feature extraction layer, and a partitioned precipitation risk prediction layer; The data processing layer is used to warn the loess foundation monitoring unit J according to the target n The precipitation spatial data in the target warning loess foundation area precipitation data is divided into precipitation area spatial data Q n , the precipitation time data in the target warning loess foundation area precipitation data is divided into precipitation area time data S n ; The precipitation spatial feature extraction layer is used to extract the precipitation spatial feature according to the precipitation area spatial data Q n Perform feature extraction to obtain the precipitation area spatial data feature Q n '; The precipitation time feature extraction layer is used to extract the precipitation time data S n Perform feature extraction to obtain the precipitation area time data feature S n '; The regional precipitation risk prediction layer is used to transform the spatial data characteristics Q n ' and precipitation area time data characteristics S n 'Perform feature matching to obtain the precipitation risk value F of the target early warning monitoring unit n .

3. The IoT method for monitoring and early warning of collapsible loess foundation subsidence according to claim 2 is characterized in that: The specific steps to build a precipitation prediction and evaluation model include: The precipitation spatial feature extraction layer is constructed using the CNN model; the precipitation spatial feature extraction layer includes a convolution layer, a pooling layer, and a feature fusion layer; The precipitation time feature extraction layer is constructed using the LSTM model; the partition precipitation risk prediction layer is constructed using the multi-head attention mechanism, which includes two fully connected layers and an output layer; Specific steps for training the precipitation prediction evaluation model: Collect several groups of precipitation forecast evaluation training samples; each group of precipitation forecast evaluation training samples contains precipitation monitoring data and corresponding marked precipitation risk values; combine several groups of precipitation forecast evaluation training samples to obtain a precipitation forecast evaluation training set; The precipitation prediction and evaluation training set is input into the precipitation prediction and evaluation model for model training to obtain the initial precipitation prediction and evaluation model; the initial precipitation prediction and evaluation model is evaluated. If the initial precipitation prediction and evaluation model passes the model evaluation, the initial precipitation prediction and evaluation model is used as the precipitation prediction and evaluation model to complete the model training; otherwise, the precipitation prediction and evaluation training set is used to continue the model training.

4. The IoT method for monitoring and early warning of collapsible loess foundation subsidence according to claim 3 is characterized in that: The loess foundation zoning monitoring and analysis model includes zoning data processing layer, zoning matrix construction layer, zoning monitoring clustering layer, optimized zoning adjustment layer and result output layer; The partition data processing layer is used to early-warn the loess foundation monitoring unit J according to the target n The surface data of the target early warning loess foundation area is divided into regional surface data D n ; Regional surface data D n And the target early warning monitoring unit precipitation risk value F n Perform feature fusion to obtain monitoring partition feature data R n ; The partition matrix construction layer is used to early-warn the loess foundation monitoring unit according to the target n and monitoring partition characteristic data R n Construct a matrix to obtain the loess foundation partition matrix; The partition monitoring clustering layer is used to cluster the loess foundation partition matrix and obtain the initial partition scheme of the loess foundation; The optimized partition adjustment layer is used to optimize the initial partition scheme of the loess foundation and obtain the target early warning loess foundation monitoring partition C. m ; The result output layer is used to output the target early warning loess foundation monitoring partition C m .

5. The IoT method for monitoring and early warning of collapsible loess foundation subsidence according to claim 4 is characterized in that: The specific steps of clustering the loess foundation partition matrix in the partition monitoring clustering layer include: The partition monitoring clustering layer is constructed by the autoencoder layer, the graph convolution layer and the self-supervision layer; the autoencoder layer and the graph convolution layer are connected in parallel, and both the autoencoder layer and the graph convolution layer contain B layers, b=1, 2, ..., B; In the bth self-encoding layer, feature extraction is performed on the loess foundation partition matrix to obtain the loess foundation partition matrix feature (b); In the bth graph convolutional layer, the loess foundation partition matrix feature (b) is received and restructured to obtain the loess foundation partition matrix feature structure (b); In the self-supervised layer, clustering feature learning is performed on the loess foundation partition matrix features (b) and the loess foundation partition matrix feature structure (b) to obtain the initial partition scheme of the loess foundation; Among them, the partition monitoring clustering layer is optimized using the loss function Z, Z=Loss(α)+W1*Loss(β)+W2*Loss(γ), Loss(α) represents the loss function of the autoencoder layer, Loss(β) represents the clustering loss function of the autoencoder layer, W1 represents the weight of the clustering loss function of the autoencoder layer, Loss(γ) represents the clustering loss function of the graph convolution layer, and W2 represents the weight of the clustering loss function of the graph convolution layer.

6. The IoT method for monitoring and early warning of collapsible loess foundation subsidence according to claim 5 is characterized in that: The specific steps of optimizing and adjusting the initial zoning scheme of loess foundation in the optimization zoning adjustment layer include: Based on the initial partitioning scheme of loess foundation, K optimal loess foundation partitioning H is constructed k , k=1, 2, ..., K; the initial zoning scheme of loess foundation contains several isolated monitoring units; each optimized loess foundation zoning H k A number of isolated monitoring units are randomly divided; K optimized loess foundation partitions are divided into H k Combination, get the optimized loess foundation partition iterative population; set the maximum number of iterations, optimize the loess foundation partition H k The fitness is the fitness G k ; Calculate the fitness G k Specific steps: Optimizing the division of loess foundation k Calculate the silhouette coefficient and get the average silhouette coefficient; use the average silhouette coefficient as the optimal loess foundation partition H k The fitness G k ; The optimized loess foundation partitioning iterative population is updated, and a number of isolated monitoring units are reallocated in each population iteration using the adding point strategy to obtain a new optimized loess foundation partitioning iterative population; When the maximum number of iterations is reached, the optimized loess foundation partition corresponding to the maximum fitness is output, which is the optimal optimized loess foundation partition. According to the optimal optimized loess foundation partition, the target early warning loess foundation monitoring partition C is generated. m , m=1, 2,…, M, where M is the total number of target warning loess foundation monitoring zones in the target warning loess foundation area.

7. The IoT method for monitoring and early warning of collapsible loess foundation subsidence according to claim 6 is characterized in that: Target early warning loess foundation monitoring zone C m The specific steps for monitoring and early warning include: Get target warning loess foundation monitoring zone C m Corresponding loess foundation monitoring data P m ; The loess foundation monitoring data P m Compared with the preset monitoring collapse warning threshold range, if the loess foundation monitoring data P m If the preset monitoring collapse warning threshold is exceeded, the target warning loess foundation collapse warning signal X m is abnormal; otherwise, the target warning signal of loess foundation collapse is X m is normal.

8. The IoT system for monitoring and early warning of collapsible loess foundation subsidence is characterized by: The system applies the IoT method for monitoring and early warning of collapsible loess foundation subsidence as described in any one of claims 1 to 7, including: The loess foundation zoning module includes a minimum unit setting unit, an index analysis unit and a zoning clustering unit; the minimum unit setting unit is used to construct a minimum monitoring unit; based on the minimum monitoring unit, the target early warning loess foundation area is divided to obtain the target early warning loess foundation monitoring unit J n , n=1, 2, …, N; N is the total number of minimum monitoring units in the target early warning loess foundation area; the index analysis unit is used to obtain the precipitation data of the target early warning loess foundation area; based on the precipitation data of the target early warning loess foundation area and the precipitation prediction evaluation model, the target early warning loess foundation monitoring unit J n Analyze and obtain the precipitation risk value F of the target early warning monitoring unit n ; The precipitation data of the target warning loess foundation area contains precipitation spatial data and precipitation time data; the precipitation prediction and evaluation model is used to combine the CNN model and the LSTM model to predict and evaluate the precipitation risk situation; the partition clustering unit is used to obtain the surface data of the target warning loess foundation area; based on the surface data of the target warning loess foundation area and the precipitation risk value F of the target warning monitoring unit n and loess foundation zoning monitoring and analysis model for target early warning loess foundation monitoring unit J n Perform deep clustering to obtain the target early warning loess foundation monitoring zone C m , m=1, 2, …, M; M is the total number of target early warning loess foundation monitoring zones in the target early warning loess foundation area; the loess foundation zone monitoring and analysis model is used to perform deep clustering division of the target early warning loess foundation area in combination with precipitation conditions and surface data; The loess foundation monitoring and early warning module includes a monitoring and early warning unit; the monitoring and early warning unit is used to monitor all target early warning loess foundation monitoring areas C m Install monitoring and early warning devices to warn the target loess foundation monitoring zone C at preset time intervals m Carry out monitoring and early warning, and obtain the target early warning signal X for loess foundation collapse m ; When receiving the target warning signal of loess foundation collapse X m If it is abnormal, a collapse warning will be issued; otherwise, no operation will be performed.

Images