A geological disaster comprehensive monitoring and forecasting method based on dual-satellite fusion

Abstract

The present application belongs to the technical field of geological disaster monitoring, and specifically relates to a comprehensive monitoring and forecasting method for geological disasters based on dual-satellite fusion, comprising the following steps: developing Beidou reference stations, monitoring stations and high-precision board cards to obtain coordinate change data of monitoring points; building a "point-line-surface" three-dimensional monitoring network through a BDS / GNSS multiple monitoring point and reference station networking, fusing Beidou high-precision positioning data and multi-source data and obtaining monitoring point displacement data; quantitatively and finely forecasting short-term and nearby heavy rain through an intelligent grid product, and using a reverse modeling sample to train and optimize a model of a heavy rain radar evolution process; constructing a multi-dimensional feature vector matrix, extracting important and key features in a multi-factor influence factor dataset, and on this basis, establishing a comprehensive monitoring and forecasting model of a long short-term memory artificial neural network. The present application can improve the accuracy and timeliness of geological disaster early warning under the complex geological environment of Guangxi-ASEAN, and has a good market application prospect.

Description

Technical Field

[0001] This invention belongs to the field of geological disaster monitoring technology, specifically relating to a comprehensive geological disaster monitoring and forecasting method based on dual-satellite fusion. Background Technology

[0002] Guangxi is located in the mid-to-southern subtropical monsoon climate zone, experiencing frequent heavy rainfall during the flood season. This often triggers flash floods, rising river levels, and inundation of crops, roads, houses, and other economic and human losses. Currently, research on cloud detection using geostationary meteorological satellites is relatively limited both domestically and internationally. Furthermore, Guangxi has a complex geological structure with widespread karst landforms. Coupled with the impact of heavy rainfall, geological disasters occur frequently, with hundreds of incidents annually, such as landslides, collapses, mudslides, ground subsidence, and ground fissures, resulting in direct economic losses ranging from tens of millions to hundreds of millions of yuan each year. Therefore, the monitoring, early warning, and geological exploration of geological disasters require the development of high-precision, highly reliable satellite navigation and positioning technologies.

[0003] To improve the accuracy of geological disaster monitoring and early warning models, this application combines the advantages of BeiDou's high monitoring accuracy and short early warning response time with the rainfall monitoring capabilities of meteorological satellites in specific areas, and provides a comprehensive geological disaster monitoring and forecasting method based on dual-satellite fusion. This method greatly extends the early warning time of geological disasters and is of great significance for geological disaster monitoring, safety management, and emergency response to sudden events in the complex geological environment of Guangxi-ASEAN.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a comprehensive geological disaster monitoring and forecasting method based on dual-satellite fusion, so as to improve the accuracy of geological disaster monitoring and early warning models, extend the timeliness of geological disaster early warning, and buy more valuable time for emergency rescue.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A comprehensive monitoring and forecasting method for geological hazards based on binary satellite fusion includes the following steps:

[0008] S1. Develop integrated Beidou monitoring station and integrated Beidou reference station, as well as high-precision boards that support multi-system satellite positioning to obtain accurate measurement data of surface displacement;

[0009] S2. Based on BDS / GNSS multi-monitoring point and base station networking technology, construct "master-slave" nodes and build a three-dimensional monitoring network of "point-line-surface";

[0010] S3. Data noise reduction and fusion processing are performed based on the adaptive attenuation memory Kalman filter algorithm AFMKF and multi-source sensor information fusion technology.

[0011] S4. Using three-dimensional data from "sky-air-ground" as the input factor of the CNN model, and introducing the time dimension, a nonlinear precipitation estimation model is established from low-to-high-level feature extraction to actual precipitation.

[0012] S5. Construct a modeling sample: S_t+S+S_+t based on precipitation or heavy precipitation observed in real-time, and use the reconstructed modeling sample to train the model of the heavy precipitation radar evolution process.

[0013] S6. Construct a dataset of landslide displacement prediction factors and multi-factor influencing factors in complex environments. Based on this, construct a multi-dimensional feature vector matrix and extract important and key features from environmental factors.

[0014] S7. Using the feature vectors constructed in S6 as early warning factors, and combining them with deformation monitoring curves, medium- and long-term forecast models, short-term forecast models, and imminent slip forecast models are formed based on the characteristics of the time series of data.

[0015] As a preferred option, the specific features of the Beidou monitoring station integrated machine, the Beidou reference station integrated machine, and the high-precision board in step S1 are as follows:

[0016] The high-precision board contains two BDS / GNSS multi-system chips, based on a carrier phase measurement algorithm, and employs a radio interferometer scheme to retrieve the position coordinates of certain nodes and calculate the angular change from these coordinates. The signal processing module of the high-precision board is designed using FPGA (5CEA7U19) + ARM (AM3225). The Beidou monitoring station integrated machine and the Beidou reference station integrated machine are developed based on BDS / GNSS multi-mode chips and are used to monitor deformation quantities such as displacement and angular changes.

[0017] As a preferred approach, S2 constructs a "master-slave" node structure to build a three-dimensional monitoring network of "points, lines, and surfaces," specifically including the following steps:

[0018] A 1+N monitoring network is deployed in each monitoring area. Data transmission between each monitoring station and the reference station is carried out via a local area network and then transmitted back to the server through the reference station. Communication links are designed to receive and transmit satellite observation data. Power is analyzed during data transmission and reception. Data is fused using a sleep mode. Redundant data is processed within the network to minimize the amount of data required while meeting application needs.

[0019] Preferably, data denoising and fusion processing is performed in S3, including the following steps:

[0020] An adaptive attenuation memory Kalman filter (AFMKF) algorithm is adopted, and an attenuation factor is introduced to dynamically track and monitor deformation data, thereby reducing the impact of noise errors on the system. A data fusion algorithm based on AFMKF is designed for the CKF algorithm, and multiple auxiliary sensors are used to assist in the monitoring of disaster points. Data fusion processing is performed on BeiDou high-precision positioning data and multi-source sensor data.

[0021] As a preferred option, the nonlinear precipitation estimation model in S4 has the following characteristics:

[0022] Using three-dimensional data from the sky, air, and ground as input factors, and incorporating the time dimension, a nonlinear precipitation estimation model is established by extracting features from the low, middle, and upper atmospheres to obtain actual precipitation, with the time window being the data from the past 7-8 frames.

[0023] As a preferred option, in the modeling sample S_t+S+S_+t constructed in S5, S_t is the radar echo sequence 1-2 hours before the precipitation process; S is the radar echo time sequence of the entire precipitation process; and S_+t is the radar echo sequence 1-2 hours after the rain.

[0024] As a preferred option, S6 constructs a dataset of landslide displacement prediction factors and multi-factor influencing factors in complex environments, specifically including the following steps:

[0025] Data fusion of multi-dimensional features is performed on surface displacement, gridded rainfall data, and data on ground fissures, unstable slopes, and hydrogeological conditions acquired by BeiDou satellites and Fengyun meteorological satellites. The specific process is as follows: First, the multi-sensor monitoring data of the initial state are centrally stored and managed to generate an initial set; a nonlinear function is applied to each member of the initial set to generate a prediction set; the predicted state is used to estimate the average value and error covariance matrix of the state; the perturbation value is used to generate the observation set; based on the prediction set and the observation set, the Kalman filter gain is calculated to obtain the analysis set, and the overall deformation is extracted.

[0026] As a preferred approach, S7 performs a comprehensive qualitative and quantitative analysis of the deformation stages of the disaster body. The specific steps for dividing the forecasts into medium- and long-term forecasts, short-term forecasts, and imminent landslide forecasts according to the time scales of the deformation and failure stages are as follows:

[0027] Using the feature vectors constructed in S6 as early warning factors, combined with deformation monitoring curves, and considering the characteristics of the time series of data, a complex model based on long short-term memory artificial neural networks of various types of data is established. In the medium and long-term forecast stage, the neural network method is used to predict the cumulative displacement; in the short-term forecast stage, the kinematic method is used to predict the time when the deformation enters the pre-slip stage; in the pre-slip stage, the Verhulst model is used to make a more accurate prediction of the time of disaster occurrence.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) The geological disaster comprehensive monitoring and forecasting method based on dual-satellite fusion of the present invention integrates the advantages of Beidou and Fengyun satellites (hereinafter referred to as "dual satellites") in geological disaster monitoring and early warning applications, solves the problem of insufficient information caused by single data, and improves the accuracy and timeliness of geological disaster monitoring and early warning models through dual-satellite data fusion.

[0030] (2) The geological disaster comprehensive monitoring and forecasting method based on dual-satellite fusion of the present invention transmits ground data based on a three-dimensional monitoring network of "point-line-surface" based on monitoring points and reference stations, which ensures the normal transmission of data in weak signal environments.

[0031] (3) The geological disaster comprehensive monitoring and forecasting method based on dual-star fusion of the present invention performs data fusion on Beidou high-precision positioning data and multi-source sensor data, completes the network processing of redundant data, and solves the problem of isolated deformation data in traditional disaster monitoring. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0033] Figure 2 This is a power consumption diagram for data transmission and reception.

[0034] Figure 3 This is a schematic diagram illustrating data denoising and fusion processing using the AFMKF algorithm;

[0035] Figure 4 It is a schematic diagram showing a series of characteristics of heavy precipitation in radar and satellite images;

[0036] Figure 5 This is a schematic diagram of establishing a graded numerical model for precipitation correction forecasting;

[0037] Figure 6 This is a schematic diagram of multi-sensor data fusion;

[0038] Figure 7 This is a schematic diagram illustrating the process of creating, training, and predicting a BP neural network. Detailed Implementation

[0039] The technical solution of this invention patent will be clearly and completely described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0040] In the description of this invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention.

[0041] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0042] See attached document Figure 1 A comprehensive monitoring and forecasting method for geological hazards based on binary star fusion includes the following steps:

[0043] S1. Develop integrated BeiDou monitoring stations and BeiDou reference stations, as well as high-precision boards and other terminal equipment that support satellite positioning of multiple systems such as BDS and GPS, to obtain accurate measurement data of surface displacement.

[0044] A sustainable landslide monitoring method based on BDS / GNSS was developed. This study analyzed the changes between the real-time three-dimensional coordinates and the initial coordinates of each monitoring point to obtain the changes in the monitoring points. A multi-dimensional deformation monitoring technology combining short and long baselines was adopted to perform error correction and real-time high-precision positioning calculation to obtain accurate surface displacement measurement data. A local calculation scheme based on a local area network was designed to transmit information data.

[0045] Based on a carrier phase measurement algorithm, a high-precision BDS / GNSS board was developed to achieve static millimeter-level accuracy measurement. Specifically, a pair of BDS / GNSS multi-system chips is used to acquire the pseudorange and carrier phase values ​​of the BDS / GNSS multi-system satellites. A radio interferometer scheme is employed, using the distance between the phase centers of the two receiving antennas as the measurement baseline, which is then used as the length measurement value. After the two antennas receive the RF signal, it undergoes low-noise amplification, frequency conversion, and filtering. Coherent measurement is then performed within the receiver. The coherent carrier phase difference data is used to deduce the baseline length through an algorithm. The intersection of multiple lengths allows for the deduction of the position coordinates of certain nodes, from which the angular change can be calculated. The high-precision board's receiver signal processing module is designed using an FPGA (5CEA7U19) + ARM (AM3225) architecture. It receives the intermediate frequency digital signal down-converted by the RF module, processes the digital signal, and achieves positioning calculation and integrated navigation.

[0046] Based on BDS / GNSS multi-mode chips, a Beidou monitoring station integrated unit and a Beidou reference station integrated unit were developed and deployed on deformation monitoring targets such as geological disaster sites and buildings to monitor deformation quantities such as displacement and angle changes. The deformation monitoring station connects to a precision positioning reference station to obtain high-precision reference position data. Simultaneously, a local calculation scheme based on a local area network was designed for information data transmission. The Beidou reference station integrated unit consists of an AC-220V surge protector, a 220V to 12V switching power supply, an industrial control computer, a 4G router, a reference station receiver, and an antenna feeder surge protector. The Beidou monitoring station integrated unit mainly consists of a power module, a receiver module, a data transmission module, and a heat dissipation module. Depending on the power supply method, it can be divided into two versions: a 220V AC mains power supply version and a solar power supply version.

[0047] S2. Based on BDS / GNSS multi-monitoring point and base station networking technology, construct "master-slave" nodes and build a "point-line-surface" three-dimensional monitoring network.

[0048] First, a 1+N monitoring network is deployed in each monitoring area. Data is transmitted between each monitoring station and the base station via a local area network, and finally transmitted back to the server through the base station.

[0049] Secondly, the communication link was designed to receive and transmit satellite observation data. (See attached document.) Figure 2As shown, the power consumption during data transmission and reception is analyzed. A sleep mode is used for data fusion, and redundant data is processed within the network to minimize the required data volume while meeting application needs. An Aopre wireless bridge is used for point-to-point wireless transmission, with a wireless frequency band of 5.8 GHz and a transmission speed of 300 Mbps. When in mountainous or remote areas, a 4G DTU communication scheme is used to achieve long-distance wireless transmission via the 4G network. This module uses a hardware watchdog circuit to monitor the module's operating status in real time, and a short-circuit protection design is added to the voltage input terminal. This scheme has a serial data rate of 300–115200 bps and supports RS232 and RS485 interfaces, compatible with the RS232 and RS485 circuits extended on the STM32F767ZFT6 core processor. When the transmission distance is hundreds of meters or even kilometers, a data radio is used for data transmission.

[0050] S3. Based on the adaptive attenuation memory Kalman filter algorithm AFMKF and multi-source sensor information fusion technology, data noise reduction and fusion processing are performed.

[0051] As attached Figure 3 As shown, the Adaptive Decaying Memory Kalman Filter (AFMKF) algorithm is adopted, and a decay factor is introduced to dynamically track and monitor deformation data, reduce system noise errors, and accelerate the convergence of data monitoring. To address the uncertainty of the decaying memory factor, periodic high-precision observations are used to update the innovation sequence, and the calculation of the decaying memory factor is constrained.

[0052] A data fusion algorithm based on AFMKF was designed for the CKF algorithm, employing multiple auxiliary sensors for disaster point monitoring, including intelligent rainfall monitors, tilt / acceleration monitors, intelligent crack displacement monitors, mud and water level depth monitors, and remote alarm devices—a variety of universal geological disaster monitoring and early warning equipment. Data fusion was performed on BeiDou high-precision positioning data and multi-source sensor data, completing the intra-network processing of redundant data while addressing the problem of isolated deformation data in traditional disaster monitoring.

[0053] S4. The model uses three-dimensional data from the sky, air, and ground as input factors for the CNN model, introduces the time dimension, and establishes a nonlinear precipitation estimation model that extracts features from the low, middle, and upper levels to the actual precipitation.

[0054] The model uses integrated three-dimensional data from the sky, air, and ground as input factors for its CNN model. The model's input considers real-time changes in areal features, as well as upper-level radar echoes and satellite cloud patterns. Furthermore, a time dimension is introduced, using data from the past 7-8 frames as a time window, to establish a nonlinear precipitation estimation model that extracts features from low, middle, and upper atmospheres to determine actual precipitation. This nonlinear precipitation estimation model can be used for:

[0055] (1) Mesoscale characteristics of heavy precipitation and automatic identification;

[0056] By identifying typical cases of short-duration heavy precipitation and combining data from next-generation Doppler radar, wind profiler radar, meteorological geostationary satellites, mesoscale ground stations, single-point radiosonde detection, and numerical model data, a popular learning method was used to extract features of the circulation physical field, obtaining characteristic information of the precipitation formation, maturity, and dissipation stages. Through analysis of these mesoscale characteristics, image pattern recognition and other techniques were employed to automatically identify a series of features of heavy precipitation as seen in radar and satellite images, as shown in the attached figure. Figure 4 As shown.

[0057] (2) Short-term and now-near intelligent forecasting of areas of heavy precipitation;

[0058] A time-series forecasting model for satellite precipitation cloud clusters based on deep convolutional neural networks (CNNs) for heavy precipitation is constructed. This model effectively integrates numerical model data with atmospheric background field information to create multi-dimensional raw inputs. Through learning, it effectively tracks satellite precipitation cloud cluster signals, enabling short-term, near-term candidate region forecasting for heavy precipitation. A time-series extrapolation forecasting model for heavy precipitation areas based on radar video information using deep learning techniques (including RNN, 2D CNN, and 3D CNN models) is also developed. Furthermore, by incorporating the indicative role of mesoscale pressure fields, the model is further compared, corrected, and calibrated to predict future heavy precipitation areas.

[0059] (3) Quantitative and refined forecasting of rainfall levels for heavy precipitation;

[0060] Quantitative evaluation is performed on the radar and satellite imagery describing heavy precipitation obtained from the aforementioned forecasts. A cross-channel 3D cylindrical convolution is employed, where multiple channels, multi-layer data, and various feature quantities are fused to form the network input. Through convolutional learning and evolution, the heavy precipitation magnitudes for six hourly intervals within the next 0-6 hours are obtained. Using the heavy precipitation location networks and parameters from various data sources obtained in the study as the pre-level network and starting point, deconvolution is performed to build the post-level network. Then, the pre- and post-level networks are fused and trained to achieve accurate forecasts of component-level heavy precipitation within a short period. If the training set lacks data on heavy precipitation events of this type, a Deep Adaptive Network (DAN) technique from transfer learning is used to adaptively adjust the heavy precipitation forecast results.

[0061] (4) Nonlinear precipitation forecasting based on multi-model super ensemble forecasting;

[0062] For different levels of precipitation, European fine-grid precipitation data were used as forecasting factors, combined with a set of physical quantity characteristic factors, and fuzzy neural networks were employed to forecast precipitation at various stations. Based on rainfall forecasts from various stations in Guangxi and precipitation field forecasts from all numerical model members, a study was conducted on using the random forest algorithm to achieve station-to-grid forecasting. A hierarchical numerical model precipitation correction forecast model was established, as shown in the appendix. Figure 5 As shown.

[0063] S5 constructs modeling samples based on precipitation or heavy precipitation observed in real-time: Sample: S_t+S+S_+t, and uses the reconstructed modeling samples to train the model of the heavy precipitation radar evolution process.

[0064] Short-duration heavy precipitation events are low-probability weather phenomena, and their sample distribution is highly uneven throughout the radar echo time series. Therefore, a modeling sample based on observed precipitation or heavy precipitation is constructed: Sample: S_t+S+S_+t, where S_t is the radar echo sequence 1-2 hours before the precipitation event; S is the radar echo time series of the entire precipitation event; and S_+t is the radar echo sequence 1-2 hours after the rain. The model is trained using the reconstructed modeling sample when simulating the radar evolution of heavy precipitation.

[0065] S6. Construct a dataset of landslide displacement prediction factors and multi-factor influencing factors in complex environments. Based on this, construct a multi-dimensional feature vector matrix and extract important and key features from the environmental factors.

[0066] Data on surface displacement and gridded rainfall acquired via BeiDou and Fengyun meteorological satellites in S1 and S2, as well as data on ground fissures, unstable slopes, and hydrogeological conditions acquired through a combination of SPOT7 and Landsat8 imagery and field measurements, are centrally stored and managed. First, a RabbitMQ message queue is used to receive data. A RabbitMQ cluster is built on the server, with HAProxy load balancing, creating a mirrored cluster. A publish / subscribe model is adopted; after data is stored in RabbitMQ, the early warning model receives real-time data through a subscription model. A database is constructed to classify and collect multi-source data from various monitoring systems. The data tables mainly include: basic information tables for monitoring stations, basic information tables for monitoring points, monitoring data tables, information tables for early warning objects, early warning log tables, file upload information tables, file upload record tables, reference point port information tables, sensor basic information tables, sensor data tables, system log tables, and scheduled task tables. The hardware and software selected for database construction should be organically integrated according to the physical design scheme of the database.

[0067] The system investigates and assesses the influencing factors of disaster bodies, establishes real-time monitoring for key factors, and transmits data back. In the centralized fusion architecture, the multi-sensor information fusion system uses multiple sensors to acquire measurement information containing various target characteristics and sends it directly to the fusion center. There, it performs processing steps such as spatial registration, data correlation, tracking filtering, and prediction, and appropriately combines and synthesizes this information according to optimization criteria. Spatial registration refers to coordinating synchronous and asynchronous data through registration algorithms to estimate and compensate for sensor biases. The process is shown in the attached diagram. Figure 6 As shown, the Integrated Kalman Filter (ENKF) target tracking algorithm is used for multi-dimensional feature data fusion. The specific process is as follows: First, a series of disturbances are applied to the initial state of the multi-sensor monitoring data to generate an initial set; a nonlinear function is applied to each member of the initial set to generate a prediction set; the predicted state is used to estimate the average state and the error covariance matrix; the disturbance values ​​are used to generate the observation set. Based on the prediction set and the observation set, the Kalman filter gain is calculated to obtain the analysis set, and the overall deformation is extracted.

[0068] S7. Using the characteristic vectors such as surface displacement and rainfall in S6 as early warning factors, and combining them with deformation monitoring curves, medium- and long-term forecast models, short-term forecast models, and landslide forecast models are formed based on the characteristics of the time series of data.

[0069] A comprehensive qualitative and quantitative analysis of the deformation stages of the disaster body is conducted. At the same time, according to the time scale of the deformation and failure stages, the forecasts can be divided into medium- and long-term forecasts, short-term forecasts, and imminent landslide forecasts.

[0070] (1) In the medium to long term, a neural network method is used to predict cumulative displacement. The learning process of a BP neural network is an error propagation and correction process, which is divided into two stages: The first stage proceeds in the forward propagation direction, from the input layer to the hidden layer to the output layer, to obtain the output value of each neuron until the output value of the final output layer is obtained; The second stage proceeds in the backward propagation direction, and adjusts the connection weights between nodes based on the error between the actual output value and the expected output value of the output layer, so that the error is minimized. See Appendix for details. Figure 7 Let there be a learning sample (x). 1p ,x 2p ,…,x 3p (tp)(p is the number of samples), after giving the weights of the network vector W, it can be calculated using the formula:

[0071]

[0072] The output value Y of the network is calculated using the basic formula for the output of hidden layer neurons in a neural network. k The network's output error is defined as:

[0073] dp = t p -Y k

[0074] Define the error function as:

[0075]

[0076] In the backpropagation algorithm, it is along the error function e p The negative gradient direction that varies with W corrects W. Let the correction value of W be Δω, and take...

[0077]

[0078] Where η is the learning rate, taking a value between 0 and 1. Substituting dp and ep into the formula Δω, we get:

[0079]

[0080] This formula can be used as the basic formula for the next step of iterative calculation of the sorted sample.

[0081] To meet actual forecasting needs, and leveraging the nonlinear mapping capabilities of BP neural networks, modeling is performed. Before training the network, the weights and thresholds must be initialized. The principles are as follows: if the network is small, the learning rate η and the number of learning steps can be smaller; if the network is large, larger values ​​should be chosen accordingly, until an appropriate adjustment is reached.

[0082] (2) In the short-term forecasting stage, the time when deformation enters the pre-slip stage is predicted using kinematic methods. Using 45° as a reference, the initial acceleration rate of the landslide's Tt curve after removing the isotropic portion is used to determine the rate value at point M entering the afforestation stage; the ∑at and ∑at values ​​of the landslide are plotted. Find the time interval of constant acceleration in the curve, let it be [n1, n2]. Remove the uniform velocity part n, then the corresponding time interval of constant acceleration in the Tt curve is t = [(n1-n), (n2-n)] days; select point P with t = m in ∑at (that is, when the landslide deforms to this point, the selection of this point is not fixed) as the prediction starting point, then the coordinates of point P in the Tt curve are (mn, T), where T is the actual T in Tt.

[0083] To simplify the quantitative prediction method for parabolic curves, the parabola T = a·t with coefficient a ranging from 0.001 to 0.4 is used. 2 To create the graphic, follow these steps:

[0084] ①Construct ∑at and When the deformation enters the initial acceleration stage, the St curve is converted into the Tt line, and the 45° line is used to determine the initial acceleration rate value after removing the uniform velocity part.

[0085] ②According to ∑at and The curve determines the isostatic acceleration phase, and the prediction starting point is selected. The Tt curve, which determines the isostatic acceleration phase, is compared with the plotted graph to determine the fitted parabolic equation;

[0086] ③ Establish multiple parabolic equations to intercept the Tt curve, and calculate the solution at point M. The coefficients of the established parabolic equations are slightly larger than those of the fitted parabolic equations.

[0087] ④ Using (∑a)-t, Curve and macroscopic deformation signs rule out erroneous results and determine the correct M point.

[0088] (3) During the pre-landslide stage, the Verhulst model is used to make a more accurate prediction of the disaster's occurrence time. Let the original equally spaced monitoring data sequence X be... (0) (t);

[0089] X (0) (t)={X (0) (1), X (0) (2), ...X (0) (n)}

[0090] For X (0) Perform an AGO transformation on (t) to obtain

[0091] X (1) (t)={X (1) (1), X (1) (2), ...X (1) (n)}

[0092] With X (1) (t) Fitting the Verhulst first-order whitening nonlinear differential equation:

[0093]

[0094] In the formula, a and b are undetermined coefficients, which can be obtained using the least squares method, as follows:

[0095]

[0096] Y N =[X (0) (2),X (0) (3),……,X (0) (n)] T

[0097] Substituting the obtained undetermined coefficients into the first-order whitening nonlinear differential equation, the solution to the nonlinear differential equation is obtained:

[0098]

[0099] The above equation represents the established time-based Verhulst nonlinear differential dynamic prediction model. Here, t0 represents the initial time.

[0100] If X (0) If (t) represents non-equal-interval time series data, it is first converted into an equivalent equal-interval sequence before modeling. Since the evolution of a disaster is highly similar to the process of a organism from reproduction to extinction, the critical value (inflection point) a / 2b, representing the transition from maturity (rapid growth) to extinction (slow growth), can be used as the critical displacement value. Thus, a / 2b is replaced... The time t of the landslide can then be solved:

[0101]

[0102] Since the initial time is generally 0, the above equation becomes

[0103]

[0104] t is actually a time series number; the actual slope failure time t′ should be...

[0105] t′=t×Δt

[0106] In the formula, △t is the average interval time of the monitoring data.

[0107] In addition, this embodiment issues geological disaster early warning information according to the development stage, urgency, unstable development trend, and potential harm of the geological disaster. The early warning levels are divided into Level 1, Level 2, Level 3, and Level 4, corresponding to different degrees of geological disaster risk, such as extremely high risk, high risk, relatively high risk, and moderate risk, respectively, and are indicated by red, orange, yellow, and blue. Level 1 is the highest level.

[0108] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A comprehensive monitoring and forecasting method for geological hazards based on binary star fusion, characterized in that, Includes the following steps: S1. Develop integrated Beidou monitoring station and integrated Beidou reference station, as well as high-precision boards that support multi-system satellite positioning to obtain accurate measurement data of surface displacement; S2. Based on BDS / GNSS multi-monitoring point and base station networking technology, construct a "master-slave" node network and build a "point-line-area" three-dimensional monitoring network, that is, deploy a 1+N mode monitoring network, specifically including the following steps: A 1+N monitoring network is deployed in each monitoring area. Data is transmitted between each monitoring station and the reference station via a local area network and then back to the server through the reference station. A communication link is designed to receive and transmit satellite observation data. The power during data transmission and reception is analyzed. Data is fused using a sleep mode. Redundant data is processed within the network to minimize the amount of data required while meeting application needs. S3. Data noise reduction and fusion processing are performed based on the adaptive attenuation memory Kalman filter algorithm AFMKF and multi-source sensor information fusion technology. S4. Using three-dimensional data from "sky-air-ground" as input factors for the CNN model, and introducing the time dimension, a nonlinear precipitation estimation model is established from low-to-high-level feature extraction to actual precipitation. S5. Construct a modeling sample based on precipitation or heavy precipitation observed in real-time: S_t+S+S_+t, where S_t is the radar echo sequence 1-2 hours before the precipitation process; S is the radar echo time sequence of the entire precipitation process; S_+t is the radar echo sequence 1-2 hours after the rain; and use the reconstructed modeling sample to train the model of the radar evolution process of heavy precipitation. S6. Construct a dataset of landslide displacement prediction factors and multi-factor influencing factors in complex environments. Perform multi-dimensional feature data fusion on surface displacement, grid rainfall data, ground fissure, unstable slope, and hydrogeological condition data obtained from Beidou satellite and Fengyun meteorological satellite. On this basis, construct a multi-dimensional feature vector matrix to extract important and key features from environmental factors. S7. Using the feature vectors constructed in S6 as early warning factors, and combining them with deformation monitoring curves, medium- and long-term forecast models, short-term forecast models, and imminent slip forecast models are formed based on the characteristics of the time series of data.

2. The method for comprehensive monitoring and forecasting of geological disasters based on binary star fusion as described in claim 1, characterized in that, In step S1, the specific features of the Beidou monitoring station integrated machine, the Beidou reference station integrated machine, and the high-precision board are as follows: The high-precision board contains two BDS / GNSS multi-system chips, based on a carrier phase measurement algorithm, and uses a radio interferometer scheme to retrieve the position coordinates of certain nodes and calculate the angle change from the coordinates. The high-precision board's signal processing module adopts an FPGA+ARM design; the Beidou monitoring station integrated machine and the Beidou reference station integrated machine are developed based on BDS / GNSS multi-mode chips and are used to monitor deformation detection quantities, including displacement and angle changes.

3. The method for comprehensive monitoring and forecasting of geological disasters based on binary star fusion as described in claim 1, characterized in that, Data denoising and fusion processing is performed in S3, including the following steps: An adaptive attenuation memory Kalman filter (AFMKF) algorithm is adopted, and an attenuation factor is introduced to dynamically track and monitor deformation data, thereby reducing the impact of noise errors on the system. A data fusion algorithm based on AFMKF is designed for the CKF algorithm, and multiple auxiliary sensors are used to assist in the monitoring of disaster points. Data fusion processing is performed on BeiDou high-precision positioning data and multi-source sensor data.

4. The method for comprehensive monitoring and forecasting of geological disasters based on binary star fusion as described in claim 1, characterized in that, The nonlinear precipitation estimation model in S4 has the following characteristics: Using three-dimensional data from the sky, air, and ground as input factors, and incorporating the time dimension, a nonlinear precipitation estimation model is established by extracting features from the low, middle, and upper atmospheres to obtain actual precipitation, with the time window being the data from the past 7-8 frames.

5. The method for comprehensive monitoring and forecasting of geological disasters based on binary star fusion according to claim 1, characterized in that, The data fusion of multi-dimensional features is performed on surface displacement, gridded rainfall data, and data on ground fissures, unstable slopes, and hydrogeological conditions acquired by BeiDou satellites and Fengyun meteorological satellites. The specific process is as follows: First, the multi-sensor monitoring data of the initial state are centrally stored and managed to generate an initial set; a nonlinear function is applied to each member of the initial set to generate a prediction set; the predicted state is used to estimate the average value and error covariance matrix of the state; the perturbation value is used to generate an observation set; based on the prediction set and the observation set, the Kalman filter gain is calculated to obtain the analysis set and extract the overall deformation situation.

6. The method for comprehensive monitoring and forecasting of geological disasters based on binary star fusion according to claim 1, characterized in that, S7 conducts a comprehensive qualitative and quantitative analysis of the deformation stages of the disaster body. Based on the time scales defined by the deformation and failure stages, it categorizes forecasts into medium- and long-term forecasts, short-term forecasts, and imminent landslide forecasts. The specific steps are as follows: Using the feature vector constructed in S6 as an early warning factor, combined with the deformation monitoring curve, and considering the characteristics of the time series of data, a complex model based on long short-term memory artificial neural networks of various types of data is established. In the medium- and long-term forecast stage, the neural network method is used to predict the cumulative displacement; in the short-term forecast stage, the kinematic method is used to predict the time when the deformation enters the pre-slip stage; in the pre-slip stage, the Verhulst model is used to make a more accurate prediction of the disaster occurrence time; the Verhulst model is a time nonlinear differential dynamic forecasting model.

Images