Reservoir bank collapse real-time early warning system and method based on multi-source data collaboration

Abstract

The invention belongs to the technical field of reservoir bank collapse monitoring and early warning, and particularly relates to a reservoir bank collapse real-time early warning system and method based on multi-source data collaboration. The system comprises a distributed optical fiber sensing network, a micro borehole clinometer array and an edge computing terminal, the edge computing terminal comprises a data acquisition module, a data processing module, a model computing module and an early warning module, and the model computing module adopts a dynamic prediction model cooperatively driven by a physical mechanism model and a deep reinforcement learning algorithm. And the pre-warning module analyzes and predicts the data preprocessed by the data acquisition module, and outputs a visual risk cloud picture and a graded pre-warning signal according to a prediction result and a self-adaptive pre-warning threshold algorithm. According to the invention, the strain and temperature information of the bank slope is monitored through the distributed optical fiber network, the change of inclination angles at different depths in the bank slope is monitored through the micro borehole inclinometer array, data fusion and analysis processing are carried out through the edge computing terminal, and finally real-time monitoring and early warning of the bank collapse risk are realized.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of reservoir bank collapse monitoring and early warning, and particularly relates to a reservoir bank collapse real-time early warning system and method based on multi-source data collaboration. BACKGROUND

[0002] The problem of reservoir bank collapse has always been an important issue in the field of water conservancy engineering. In the prior art, the monitoring and prediction of reservoir bank collapse mainly rely on traditional methods. Traditional bank collapse prediction usually relies on manual profile measurement and empirical parameters. Manual profile measurement requires staff to operate on site regularly, which not only consumes a lot of manpower, material resources and time, but also has limited measurement frequency and cannot obtain real-time bank collapse related data. Empirical parameters are based on past engineering experience, which cannot accurately reflect the actual situation of the current reservoir bank slope. The existing models for reservoir bank collapse analysis are mostly two-dimensional or simplified three-dimensional models. In the construction process of these models, in order to simplify the calculation, the micro-macro cross-scale effects of internal crack development, seepage field change and freeze-thaw cycle of the bank slope are ignored. The internal crack development of the bank slope changes the mechanical properties and seepage path of the soil, the change of the seepage field affects the pore water pressure and effective stress of the soil, and the freeze-thaw cycle destroys the structure of the soil and reduces the strength of the soil. These micro-macro cross-scale factors play a key role in the progressive failure of loess / soft rock bank slope. Since the existing models ignore these factors, it is difficult to accurately quantify the critical threshold of the progressive failure of loess / soft rock bank slope. This makes it impossible to determine whether the bank slope is in a dangerous state in time in actual engineering, increasing the risk of reservoir bank collapse. In addition, in terms of data processing, the existing technology lacks real-time fusion analysis means for multi-source heterogeneous data such as GNSS settlement, InSAR deformation, pore water pressure and soil moisture content. These multi-source heterogeneous data contain different information in the process of reservoir bank collapse, and separate analysis of these data cannot fully understand the development of bank collapse.

[0003] In summary, the prior art has many deficiencies in the monitoring and prediction of reservoir bank collapse. The traditional bank collapse prediction method cannot dynamically capture the whole process of bank collapse evolution, resulting in lagging prediction results and large errors; the existing model ignores the micro-macro cross-scale effects inside the bank slope, making it difficult to quantify the critical threshold of bank collapse; and there is a lack of real-time fusion analysis means for multi-source heterogeneous data, which cannot realize high-precision bank collapse risk warning. SUMMARY

[0004] In view of the above problems, the purpose of the present application is to provide a reservoir bank collapse real-time early warning system and method based on multi-source data collaboration, which monitors the strain and temperature information of the bank slope by laying a distributed optical fiber network, installs a micro borehole tiltmeter array to monitor the inclination angle changes at different depths inside the bank slope, and finally realizes real-time monitoring and early warning of bank collapse risk through data fusion and analysis processing by an edge computing terminal.

[0005] The technical scheme of the present application is as follows: a reservoir bank collapse real-time early warning system based on multi-source data collaboration, comprising a distributed optical fiber sensing network, a micro borehole tiltmeter array and an edge computing terminal, wherein:

[0006] The distributed optical fiber sensing network comprises a surface optical fiber subnet and an internal optical fiber subnet, and is layered along the surface and inside of the reservoir bank slope, for real-time monitoring of the strain and temperature information of the bank slope;

[0007] The micro borehole tiltmeter array comprises a plurality of micro borehole tiltmeters, and a mounting bracket is arranged below the micro borehole tiltmeters, and the micro borehole tiltmeter array is vertically arranged in the key area of the reservoir bank slope, for measuring the inclination angle changes at different depths inside the bank slope;

[0008] The edge computing terminal is connected with the distributed optical fiber sensing network and the micro borehole tiltmeter array through wired or wireless mode, and the edge computing terminal comprises a data acquisition module, a data processing module, a model calculation module and an early warning module, the data acquisition module is used for collecting multi-source heterogeneous data from the distributed optical fiber sensing network, the micro borehole tiltmeter array and other sensors, the other sensors are specifically GNSS settlement meters, InSAR deformation meters, pore water pressure sensors and soil moisture content sensors, the data processing module is used for preprocessing the collected data, the model calculation module is used for analyzing and predicting the preprocessed data by using a dynamic prediction model driven by a physical mechanism model and a deep reinforcement learning algorithm, and the early warning module is used for outputting a visual risk cloud map and a graded early warning signal according to the prediction result and an adaptive early warning threshold algorithm.

[0009] The surface optical fiber subnet is arranged in a grid shape along the slope surface of the reservoir bank, the horizontal and vertical spacings of adjacent optical fibers are both 1m, and the optical fibers are closely attached to the slope surface through anchor rods or concrete bases, the optical fibers of the internal optical fiber subnet are buried in the inside of the bank slope through PVC protection pipes and are layered in the depth range of 0.5-3m; the micro borehole tiltmeter array is arranged in a plurality of vertical boreholes, the borehole diameter is 0m-30m, the center distance of adjacent boreholes is 5m, and each tiltmeter is installed in each borehole at an interval of 2m, forming a three-dimensional monitoring network.

[0010] The data processing module pre-processes the collected data, including coordinate alignment, abnormal value elimination and normalization, and the formula for normalization is:

[0011]

[0012] In the formula, x is the original observation value, μ is the arithmetic mean of the data set, σ is the standard deviation of the data set, x norm is the normalized value.

[0013] The model calculation module adopts an improved Green-Ampt infiltration model coupled with Mohr-Coulomb strength criterion and ST-GNN spatiotemporal graph convolution network to drive a dynamic prediction model, and analyzes and predicts the pre-processed data. The improved Green-Ampt infiltration model considers the thickness of the unsaturated infiltration layer and the cumulative infiltration amount, and the specific formula is:

[0014]

[0015] In the formula, Z f is the wetting front depth, cm; Ks is the saturated permeability coefficient of the soil, cm / min; t is the infiltration front migration time, min; θs is the saturated volumetric water content of the soil, %; θi is the initial volumetric water content of the soil, %; b is the thickness of the infiltration layer, cm; the improved Green-Ampt infiltration model is used to describe the rainfall infiltration process of the bank slope soil, and the unsaturated characteristics and pore structure of the bank slope soil are considered;

[0016] Mohr-Coulomb strength criterion combined with pore water pressure and Biot coefficient, the specific formula is:

[0017]

[0018] In the formula, F s is the strength coefficient; c0 is the cohesion of the soil, Mpa; σ n is the normal stress, Mpa; is the internal friction angle of the soil, °; α is the Biot coefficient; the Mohr-Coulomb strength criterion is used to judge the failure condition of the bank slope soil;

[0019] The structural equation of the ST-GNN spatiotemporal graph convolution network structure includes:

[0020] Spatial graph convolution layer:

[0021] Temporal gating unit (GRU): h t = GRU(h t-1 , x t )

[0022] In the formula, H (l) representing the characteristics of the first layer, representing a dynamic adjacency matrix, representing a degree matrix, W (l) representing a learnable parameter, h t representing the hidden state at the t-th moment, x t representing the input data at the t-th moment; the spatio-temporal graph neural network ST-GNN is used to process the spatio-temporal features of multi-source heterogeneous data and mine the spatio-temporal correlation in the data.

[0023] The early warning module is based on a joint early warning mode of a cloud model and Bayesian optimization, and specifically includes:

[0024] The cloud model membership function is:

[0025] The Bayesian optimization dynamically adjusts:

[0026] The adaptive early warning threshold is T 预警 =E+k*en*Bayesian weight

[0027] In the formula, μ(x) is the cloud model membership, E is the cloud expectation value, en is the cloud entropy, β is the learning rate, γ is the decay coefficient, k is the safety coefficient, P(D / ω) is the Bayesian posterior probability, and the Bayesian weight reflects the confidence of historical data and real-time data;

[0028] The cloud model represents the uncertainty and fuzziness of the bank collapse risk through the cloud expectation value E and the cloud entropy en, and can effectively process complex data; the Bayesian optimization continuously updates the prior probability through the posterior probability P(D|w) to optimize the early warning threshold; and the early warning threshold is dynamically adjusted through the formula T 预警 =E+k*en*Bayesian weight, which can reflect the risk changes under different conditions.

[0029] The reward function of multi-objective optimization reinforcement learning is used to balance the early warning accuracy and protection cost of the early warning module, specifically:

[0030] The reward function is:

[0031] In the formula: R t is the reward value at the t-th moment, α is the early warning accuracy weight coefficient, β is the protection cost weight coefficient, the prediction error is the error between the model prediction and the actual value, the protection cost t is the protection cost at the t-th moment, and the total cost is the preset total protection cost.

[0032] The early warning module further comprises a display unit and a communication unit; the display unit is used for displaying a visual risk cloud chart and a graded early warning signal; the communication unit is used for sending the graded early warning signal to a terminal device of a relevant person; the display unit adopts a high-resolution display screen to clearly display the visual risk cloud chart and the graded early warning signal; the communication unit supports multiple communication modes, specifically, short message, email, APP push, to ensure that the graded early warning signal can be sent to the relevant person in time; the display unit and the communication unit of the early warning module cooperate with each other to provide intuitive information and timely early warning for the user.

[0033] A reservoir bank collapse real-time early warning method based on multi-source data collaboration, using a reservoir bank collapse real-time early warning system based on multi-source data collaboration as described above, comprising the following steps:

[0034] S1: Obtain strain and temperature information on the surface and inside of the reservoir bank slope through a distributed optical fiber sensing network, and obtain inclination angle change information at different depths inside the reservoir bank slope through a micro borehole tiltmeter array;

[0035] S2: The edge computing terminal collects multi-source heterogeneous data transmitted by the distributed optical fiber sensing network, the micro borehole tiltmeter array and other sensors, and performs early warning after real-time fusion analysis of these data, specifically including:

[0036] S31: An improved Green-Ampt infiltration model coupled with Mohr-Coulomb strength criterion is used to describe the infiltration process and strength characteristics of the bank slope soil body;

[0037] S32: A deep reinforcement learning algorithm is used to fuse a spatiotemporal graph neural network ST-GNN for processing multi-source heterogeneous data and mining spatiotemporal features in the data to predict the development trend of bank collapse;

[0038] S33: Based on a cloud model and Bayesian optimization, the early warning threshold can be dynamically adjusted according to the actual situation of the bank slope;

[0039] S34: Different early warning signals are graded according to the risk degree of bank collapse, and emergency early warning is issued according to different graded early warning signals.

[0040] The technical effect of the present application is that: 1. The present application can obtain multi-source heterogeneous data such as strain, inclination, settlement, deformation, pore water pressure and soil moisture content of the bank slope through the comprehensive use of the distributed optical fiber sensing network, the array of micro borehole inclinometers and other sensors. Real-time collection and analysis of these data enable dynamic capture of the whole process of bank collapse evolution under the coupling effect of reservoir water level fluctuation, rainfall erosion and rock-soil mass creep, and solve the problem of lagging prediction results and large errors of traditional methods. 2. The present application uses an improved reen-Ampt infiltration model coupled with Mohr-Coulomb strength criterion to describe the infiltration process and strength characteristics of the bank slope soil. A depth reinforcement learning algorithm is used to fuse a spatiotemporal graph neural network ST-GNN for processing multi-source heterogeneous data and mining spatiotemporal features in the data to predict the development trend of bank collapse. Based on the cloud model and Bayesian optimization, the warning threshold can be dynamically adjusted according to the actual situation of the bank slope, fully considering the microscopic-macroscopic cross-scale influence of crack development, seepage field change and freeze-thaw cycle of the bank slope, and accurately quantifying the critical threshold of progressive failure of loess / soft rock bank slope. Through real-time fusion analysis of multi-source heterogeneous data, millimeter-level deformation capture and hour-level warning of bank collapse risk are realized, greatly improving the spatiotemporal accuracy of bank collapse prediction. 3. Compared with the traditional bank collapse monitoring and prediction method which requires a large amount of manual input and may lead to failure to take effective measures in time when bank collapse occurs due to inaccurate prediction, causing damage to reservoir facilities and loss of surrounding properties, the present application realizes intelligent monitoring and warning based on the multi-source data collaborative reservoir bank collapse real-time warning system, reduces the labor cost, and accurate prediction and timely warning can take preventive measures in advance to reduce the damage of bank collapse to reservoir facilities and the surrounding environment, avoiding huge economic losses caused by bank collapse.

[0041] The following will be further described with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 It is a composition schematic diagram of a multi-source data collaborative reservoir bank collapse real-time warning system according to an embodiment of the present application.

[0043] Figure 2 It is a composition schematic diagram of an edge computing terminal according to the present application.

[0044] Figure 3 It is a data processing flowchart of a multi-source data collaborative reservoir bank collapse real-time warning system according to the present application. DETAILED DESCRIPTION

[0045] Embodiment 1

[0046] As Figure 1 , Figure 2As shown, a reservoir bank collapse real-time early warning system based on multi-source data collaboration, a reservoir bank collapse real-time early warning system based on multi-source data collaboration, including distributed optical fiber sensing network, micro borehole tiltmeter array and edge computing terminal, wherein:

[0047] The distributed optical fiber sensing network includes a surface optical fiber subnetwork and an internal optical fiber subnetwork, and is arranged along the surface and internal layers of the reservoir bank slope for real-time monitoring of the strain and temperature information of the bank slope.

[0048] The micro borehole tiltmeter array includes a plurality of micro borehole tiltmeters, and the micro borehole tiltmeters are provided with mounting brackets below the micro borehole tiltmeters, and the micro borehole tiltmeter array is vertically arranged in the key area of the reservoir bank slope for measuring the change of the inclination angle at different depths inside the bank slope.

[0049] The edge computing terminal is connected with the distributed optical fiber sensing network and the micro borehole tiltmeter array through wired or wireless mode, and the edge computing terminal includes a data acquisition module, a data processing module, a model calculation module and an early warning module, the data acquisition module is used for collecting multi-source heterogeneous data transmitted by the distributed optical fiber sensing network, the micro borehole tiltmeter array and other sensors, the other sensors are GNSS settlement instrument, InSAR deformation instrument, pore water pressure sensor and soil moisture content sensor, the data processing module pre-processes the collected data, the model calculation module adopts a dynamic prediction model driven by a physical mechanism model and a deep reinforcement learning algorithm to analyze and predict the pre-processed data, and the early warning module outputs a visual risk cloud map and a graded early warning signal according to the prediction result and an adaptive early warning threshold algorithm.

[0050] The edge computing terminal is the data processing and analysis core of the present application. It is responsible for collecting multi-source heterogeneous data transmitted by the distributed optical fiber sensing network, the micro borehole tiltmeter array and other sensors (such as GNSS settlement instrument, InSAR deformation instrument, pore water pressure sensor, soil moisture content sensor, etc.), and performing real-time fusion analysis on these data. The edge computing terminal adopts advanced artificial intelligence algorithms and data processing technology, and can quickly and accurately process a large amount of data.

[0051] The edge computing terminal comprises a processor, a storage and an interface module; the processor adopts a RuiCore RK3588 chip, integrates four-core ARM Cortex-A76 and four-core Cortex-A55, is on-board with 16GB memory, and is internally provided with an NPU, supports INT4 / INT8 / INT16 / FP16 mixed operation, and has a floating point operation capacity of 6TOPS; the storage is an extendable SSD hard disk; the interface module comprises two-way gigabit Ethernet interface, five-way RS-485 interface and fiber interface, and is respectively used for communication with a cloud platform, various sensors and a distributed fiber sensing network; the terminal is packaged in a closed container of a bank slope monitoring station, the power supply adopts a 1+1 redundant configuration of a solar panel and a storage battery, and is provided with a three-fan cooling system, and can continuously operate in an environment of-20℃ to 60℃.

[0052] The surface fiber sub-network is arranged in a grid shape along the bank slope surface of the reservoir, the transverse and longitudinal spacings of adjacent fibers are both 1m, and the surface fiber sub-network is closely attached to the slope surface through anchor rods or concrete bases, the fibers of the internal fiber sub-network are buried in the interior of the bank slope through PVC protection pipes and are arranged in layers in a depth range of 0.5-3m.

[0053] The distributed fiber sensing network is an important part of the present application. It lays optical fibers on the surface and inside of the reservoir bank slope, and uses the sensing characteristics of the optical fibers to obtain strain and temperature information of the bank slope. Optical fiber sensing technology has the advantages of high sensitivity, wide distribution range, strong anti-interference ability, etc. When the bank slope deforms, the optical fiber will be stretched or compressed, causing changes in the internal light signal. By detecting the changes in the light signal, the strain of the bank slope can be accurately measured. At the same time, the optical fiber can also measure the temperature, because the change of temperature will also affect the propagation of light signal. The distributed fiber sensing network can monitor the deformation of the bank slope in real time and continuously, providing important data support for bank collapse prediction.

[0054] The array of micro borehole inclinometers is arranged in a plurality of vertical boreholes, the borehole diameter is 50-100mm, the depth is 0m-30m, and the center distance between adjacent boreholes is 5m. Each inclinometer is installed in each borehole at an interval of 2m, forming a three-dimensional monitoring network.

[0055] The array of micro borehole inclinometers is a device for measuring the inclination changes inside the bank slope. It is composed of multiple micro borehole inclinometers, which are installed in the boreholes inside the bank slope. The micro borehole inclinometer can measure the inclination angle changes at different depths in the borehole, so as to understand the deformation of the bank slope inside. Compared with traditional inclinometers, the micro borehole inclinometer has the advantages of small size, high precision, easy installation, etc. By arranging micro borehole inclinometers at different positions and depths of the bank slope, the deformation distribution inside the bank slope can be comprehensively mastered, providing more accurate data for bank collapse prediction.

[0056] The data processing module pre-processes the collected data, including coordinate alignment, outlier rejection and normalization, and the normalization formula is:

[0057]

[0058] In the formula, x is the original observation value, μ is the arithmetic mean of the data set, σ is the standard deviation of the data set, x norm is the normalized value.

[0059] The model calculation module adopts an improved Green-Ampt infiltration model coupled with Mohr-Coulomb strength criterion and ST-GNN space-time graph convolution network collaborative driving dynamic prediction model to analyze and predict the pre-processed data. The improved Green-Ampt infiltration model considers the unsaturated infiltration layer thickness and cumulative infiltration, and the specific formula is:

[0060]

[0061] In the formula, Z f is the wetting front depth, cm; Ks is the saturated permeability coefficient of the soil, cm / min; t is the infiltration front migration time, min; θs is the saturated volumetric water content of the soil, %; θi is the initial volumetric water content of the soil, %; b is the infiltration layer thickness, cm; the improved Green-Ampt infiltration model is used to describe the rainfall infiltration process of the bank slope soil, considering the unsaturated characteristics and pore structure of the bank slope soil; the model uses the infiltration layer thickness parameter to more accurately simulate the infiltration process of unsaturated soil, defines the cumulative infiltration and infiltration depth relationship as a piecewise function to improve the calculation accuracy, considers the influence of matric suction head on the infiltration process, and enhances the applicability of the model;

[0062] Mohr-Coulomb strength criterion combined with pore water pressure and Biot coefficient, the specific formula is:

[0063]

[0064] In the formula, F s is the strength coefficient; c0 is the cohesion of the soil, Mpa; σ n is the normal stress, Mpa; is the internal friction angle of the soil, °; α is the Biot coefficient; the Mohr-Coulomb strength criterion is used to judge the failure condition of the bank slope soil;

[0065] The application innovatively adopts a dynamic prediction model driven by a physical mechanism model and a deep reinforcement learning algorithm, the physical mechanism model is a improved Green-Ampt infiltration model coupled with Mohr-Coulomb strength criterion, which is used for describing the infiltration process and strength characteristics of the bank slope soil body, the improved Green-Ampt infiltration model considers the unsaturated characteristics and pore structure of the bank slope soil body, and can more accurately simulate the rainfall infiltration process, the Mohr-Coulomb strength criterion is used to judge the failure condition of the bank slope soil body, by coupling the two models, the mechanical behavior of the bank slope under the action of rainfall, reservoir water level fluctuation and other factors can be better described;

[0066] The structure equation of the ST-GNN spatio-temporal graph convolution network structure comprises:

[0067] The spatial graph convolution layer comprises:

[0068] The time gate unit (GRU) is h t = GRU (h t-1 , x t )

[0069] In the formula, H (l) represents the feature representation of the lth layer, represents a dynamic adjacency matrix, represents a degree matrix, W (l) represents a learnable parameter, h t represents the hidden state at the tth moment, x t represents the input data at the tth moment; the spatio-temporal graph neural network ST-GNN is used for processing the spatio-temporal features of multi-source heterogeneous data and mining the spatio-temporal correlation in the data.

[0070] The early warning module is based on a joint early warning mode of a cloud model and Bayesian optimization, and specifically comprises:

[0071] The cloud model membership function is:

[0072] The Bayesian optimization dynamically adjusts:

[0073] The adaptive early warning threshold is T 预警 =E+k*en*Bayes weight

[0074] In the formula, μ (x) is the cloud membership degree, E is the cloud expectation value, en is the cloud entropy, β is the learning rate, γ is the attenuation coefficient, k is the safety coefficient, P (D / ω) is the Bayesian posterior probability, and the Bayesian weight reflects the confidence degree of historical data and real-time data;

[0075] The cloud model represents the uncertainty and fuzziness of the bank collapse risk through cloud expectation E and cloud entropy en, and can effectively process complex data; the Bayesian optimization continuously updates the prior probability through the posterior probability P(D|w), and optimizes the early warning threshold; the early warning threshold is dynamically adjusted through the formula T 预警 = E + k * en * Bayesian weight, which can reflect the risk changes under different conditions, and the algorithm based on the cloud model and Bayesian optimization can dynamically adjust the early warning threshold according to the actual situation of the bank slope. The cloud model can process the uncertainty and fuzziness of the data, and the Bayesian optimization can continuously optimize the early warning threshold according to the historical data and real-time data. Through the early warning threshold algorithm, it can more accurately judge whether the bank slope is in a dangerous state, and avoid false positives and false negatives.

[0076] The reinforcement learning reward function of multi-objective optimization is used to balance the early warning accuracy and protection cost of the early warning module, specifically:

[0077] Reward function:

[0078] In the formula: R t is the reward value at time t, alpha is the early warning accuracy weight coefficient, beta is the protection cost weight coefficient, the prediction error is the error between the model prediction and the actual value, the protection cost t is the protection cost at time t, and the total cost is the preset total protection cost.

[0079] The present application balances the early warning accuracy and protection cost through the weight coefficient, realizes multi-objective optimization, and at the same time, the reward value changes with the prediction error and protection cost, so as to encourage the model to develop in the optimal direction. Finally, the weight coefficient can be adjusted according to the actual demand to adapt to the early warning demand in different scenes.

[0080] The early warning module further comprises a display unit and a communication unit; the display unit is used to display the visual risk cloud chart and the graded early warning signal; the communication unit is used to send the graded early warning signal to the terminal device of the relevant personnel; the display unit adopts a high-resolution display screen to clearly display the visual risk cloud chart and the graded early warning signal; the communication unit supports multiple communication modes, specifically SMS, email, APP push, to ensure that the graded early warning signal can be sent to the relevant personnel in time; the display unit and the communication unit of the early warning module cooperate with each other to provide intuitive information and timely early warning for the user.

[0081] The graded early warning signal of the present application is graded according to the risk degree of bank collapse, and provides clear emergency response guidance for reservoir management personnel. Through visual output, the reservoir management personnel can quickly understand the bank collapse situation and take corresponding measures to improve the emergency response efficiency.

[0082] As Figure 3As shown, a reservoir bank collapse real-time early warning method based on multi-source data collaboration uses a reservoir bank collapse real-time early warning system based on multi-source data collaboration as described above, and includes the following steps:

[0083] S1: Obtain strain and temperature information on the surface and inside of the reservoir bank slope through a distributed optical fiber sensing network, and obtain inclination angle change information at different depths inside the reservoir bank slope through a micro borehole tiltmeter array;

[0084] S2: The edge computing terminal collects multi-source heterogeneous data transmitted by the distributed optical fiber sensing network, the micro borehole tiltmeter array and other sensors, and performs real-time fusion analysis on these data and performs early warning, specifically including:

[0085] S31: An improved Green-Ampt infiltration model coupled with Mohr-Coulomb strength criterion is used to describe the infiltration process and strength characteristics of the bank slope soil;

[0086] S32: A deep reinforcement learning algorithm is used to fuse a spatiotemporal graph neural network ST-GNN for processing multi-source heterogeneous data and mining spatiotemporal features in the data to predict the development trend of bank collapse;

[0087] S33: Based on cloud model and Bayesian optimization, the early warning threshold can be dynamically adjusted according to the actual situation of the bank slope;

[0088] S34: Different early warning signals are classified according to the risk degree of bank collapse, and emergency early warning is issued according to different classified early warning signals.

[0089] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A reservoir bank collapse real-time early warning system based on multi-source data collaboration, characterized in that: The application relates to a reservoir slope monitoring system, which comprises a distributed optical fiber sensing network, a micro borehole tiltmeter array and an edge computing terminal, wherein: The distributed optical fiber sensing network comprises a surface optical fiber subnet and an internal optical fiber subnet, is arranged along the surface and internal layers of a reservoir slope and is used for monitoring the strain and temperature information of the slope in real time; The micro borehole tiltmeter array comprises a plurality of micro borehole tiltmeters, the micro borehole tiltmeters are provided with mounting supports below the micro borehole tiltmeters, the micro borehole tiltmeter array is vertically arranged in a key area of the reservoir slope and is used for measuring the inclination angle changes at different depths in the slope; The edge computing terminal is connected with the distributed optical fiber sensing network and the micro borehole tiltmeter array through wired or wireless modes, the edge computing terminal comprises a data acquisition module, a data processing module, a model calculation module and a warning module, the data acquisition module is used for collecting multi-source heterogeneous data transmitted by the distributed optical fiber sensing network, the micro borehole tiltmeter array and other sensors, the other sensors are specifically GNSS settlement meters, InSAR deformation meters, pore water pressure sensors and soil moisture content sensors, the data processing module is used for pre-processing the collected data, the model calculation module is used for analyzing and predicting the pre-processed data by adopting a dynamic prediction model which is driven by a physical mechanism model and a deep reinforcement learning algorithm, and the warning module is used for outputting a visual risk cloud map and a graded warning signal according to a prediction result and an adaptive warning threshold algorithm. 2.The reservoir bank collapse real-time early warning system based on multi-source data collaboration according to claim 1, characterized in that: The surface optical fiber sub-network is arranged in a grid shape along the slope surface of the reservoir bank slope, the horizontal and vertical distances between adjacent optical fibers are both 1 m, and the surface optical fiber sub-network is closely attached to the slope surface through anchor rods or concrete bases, the optical fibers of the internal optical fiber sub-network are buried in the internal bank slope through PVC protection pipes and are arranged in layers in the depth range of 0.5-3 m; the array of the micro borehole tiltmeters is arranged in a plurality of vertical boreholes The depth of the boreholes is 0-30 m, the center distance between adjacent boreholes is 5 m, and each tiltmeter is installed in each borehole at a distance of 2 m to form a three-dimensional monitoring network. 3.The reservoir bank collapse real-time early warning system based on multi-source data collaboration according to claim 1, characterized in that: The data processing module is used for pre-processing the collected data, including coordinate alignment, abnormal value elimination and normalization, and the normalization formula is: where x is the original observation, μ is the arithmetic mean of the set of data, σ is the standard deviation of the set of data, and x norm is the normalized value.

4. The reservoir bank collapse real-time early warning system based on multi-source data collaboration of claim 1, characterized in that: The model calculation module adopts a dynamic prediction model which is driven by an improved Green-Ampt infiltration model, a Mohr-Coulomb strength criterion and an ST-GNN spatiotemporal graph convolution network, is used for analyzing and predicting the pre-processed data, the improved Green-Ampt infiltration model considers the unsaturated infiltration layer thickness and the cumulative infiltration amount, and the specific formula is: wherein Z f is the wetting front depth, cm; Ks is the saturated hydraulic conductivity of the soil, cm / min; t is the wetting front travel time, min; θs is the saturated volumetric water content of the soil, %; θi is the initial volumetric water content of the soil, %; b is the thickness of the wetting layer, cm; the improved Green-Ampt infiltration model is used to describe the rainfall infiltration process of the bank slope soil, and the unsaturated characteristics and pore structure of the bank slope soil are considered; The Mohr-Coulomb strength criterion is combined with pore water pressure and a Biot coefficient, and the specific formula is: In the formula, F s is the strength coefficient; c0 is the soil cohesion, Mpa; σ n is the normal stress, Mpa; is the internal friction angle of soil, °; α is the Biot coefficient; the Mohr-Coulomb strength criterion is used to judge the failure condition of the bank slope soil The structure equation of the ST-GNN spatiotemporal graph convolution network structure comprises: spatial graph convolutional layer: Time-gated unit (GRU): h t = GRU(h t-1 , x t ) In the formula, H (l) representing a feature of the first layer, representing a dynamic adjacency matrix, representing a degree matrix, W (l) representing a learnable parameter, h t representing a hidden state at the t-th moment, x t representing input data at the t-th moment; the spatio-temporal graph neural network ST-GNN is used for processing spatio-temporal features of multi-source heterogeneous data and mining spatio-temporal correlations in the data.

5. The reservoir bank collapse real-time early warning system based on multi-source data collaboration according to claim 1, characterized in that: The warning module is based on a joint warning mode of a cloud model and Bayesian optimization, and specifically comprises: Cloud model membership function: Bayesian optimization dynamically adjusts: Adaptive warning threshold: T 预警 = E + k * en * Bayes weight In the formula, mu (x) is the cloud model membership degree, E is a cloud expectation value, en is a cloud entropy, beta is a learning rate, gamma is a decay coefficient, k is a safety coefficient, and P (D / omega) is a Bayesian posterior probability; the Bayesian weight reflects the confidence degree of historical data and real-time data; The cloud model represents the uncertainty and fuzziness of the bank collapse risk by cloud expectation E and cloud entropy en, and can effectively process complex data; the Bayesian optimization continuously updates the prior probability by posterior probability P(D|w), and optimizes the early warning threshold; and the early warning threshold is dynamically adjusted by the formula T 预警 =E+k*en*Bayesian weight, which can reflect the risk changes under different conditions. 6.The reservoir bank collapse real-time early warning system based on multi-source data collaboration according to claim 1, characterized in that: The warning accuracy and protection cost of the warning module are balanced through a reinforcement learning reward function of multi-objective optimization, and the specific formula is: Reward function: In the formula: R t is the reward value at time t, a is the warning accuracy weight coefficient, β is the protection cost weight coefficient, the prediction error is the error between the model prediction and the actual value, the protection cost t is the protection cost at time t, and the total cost is the preset total protection cost.

7. The reservoir bank collapse real-time early warning system based on multi-source data collaboration according to claim 1, characterized in that: The warning module further comprises a display unit and a communication unit; the display unit is used for displaying the visual risk cloud map and the graded warning signal; and the communication unit is used for sending the graded warning signal to terminal equipment of relevant personnel; The display unit adopts a high-resolution display screen to clearly display the visual risk cloud map and the graded warning signal. The communication unit supports multiple communication modes, specifically short message, email, APP push, to ensure that the graded early warning signals can be sent to relevant personnel in time; the display unit and the communication unit of the early warning module cooperate with each other to provide intuitive information and timely early warning for users.

8. A reservoir bank collapse real-time early warning method based on multi-source data collaboration, using the reservoir bank collapse real-time early warning system based on multi-source data collaboration according to claim 1, characterized in that: The method comprises the following steps: S1: Obtain strain and temperature information on the surface and inside of the reservoir slope through a distributed optical fiber sensing network, and obtain inclination angle change information at different depths inside the reservoir slope through a micro borehole tiltmeter array; S2: An edge computing terminal collects multi-source heterogeneous data transmitted by the distributed optical fiber sensing network, the micro borehole tiltmeter array and other sensors, and performs early warning after real-time fusion analysis of the data, specifically including: S31: An improved Green-Ampt infiltration model is coupled with Mohr-Coulomb strength criterion to describe the infiltration process and strength characteristics of the slope soil; S32: A deep reinforcement learning algorithm is fused with a spatiotemporal graph neural network ST-GNN to process multi-source heterogeneous data and mine spatiotemporal features in the data, and to predict the development trend of bank collapse; S33: Based on a cloud model and Bayesian optimization, the early warning threshold can be dynamically adjusted according to the actual situation of the slope; S34: Different early warning signals are graded according to the risk degree of bank collapse, and emergency early warning is given according to different graded early warning signals.

Images