A system and method for intelligent early warning and prevention of damage to a schist reservoir area slope

Abstract

The application relates to the technical field of geological disaster monitoring and early warning, and discloses a schist reservoir area slope damage intelligent early warning and prevention and control system and method, which comprises a data fusion and knowledge base construction module, an intelligent early warning analysis module, a cloud-edge collaborative deployment platform and a prevention and control decision support module. The data fusion and knowledge base construction module accesses multi-source engineering data and runs a Bayesian updating algorithm, calculates and updates a slope damage mechanism knowledge base. The intelligent early warning analysis module executes logical operation by using a hybrid early warning model coupled by a chain physical criterion and an artificial intelligence time series prediction network, and outputs an early warning signal. The cloud-edge collaborative deployment platform executes on-site model reasoning and cloud iteration training. The prevention and control decision support module matches a disposal plan after receiving the early warning signal, and transmits feedback data of on-site disposal back to the corresponding module. The application realizes early identification driven by mechanism and data, and completes self-adaptive calibration and logic evolution of system parameters through a disaster prevention experience closed-loop feedback mechanism.

Description

Technical Field

[0001] This invention relates to the field of geological disaster monitoring and early warning technology, specifically to an intelligent early warning and prevention system and method for slope damage in schist reservoir areas. Background Technology

[0002] Under the long-term operation of water storage and periodic water level regulation, the slopes of water conservancy project reservoirs experience repeated wet-dry cycles and water-rock stress coupling. For schist slopes, this unique reservoir environment leads to continuous softening and deterioration of the characteristic mineral components within the rock mass, resulting in accumulated local damage and even overall slope instability and failure. To mitigate such engineering geological risks, various sensors are typically deployed at the project site to collect data on displacement, seepage, and microseismic activity, and these are combined with early warning systems for real-time status monitoring and assessment.

[0003] Existing slope early warning systems have significant limitations in long-term engineering applications. Current early warning logic primarily relies on initially set fixed thresholds or purely data-driven prediction algorithms. The rock mechanics parameters and critical damage thresholds used in system early warnings are mostly calibrated once in the early stages of the project through indoor static physical tests. These fixed parameters cannot accurately reflect the dynamic strength decay characteristics of schist during long-term water environment evolution, causing the judgment criteria in the later stages of system operation to gradually deviate from the actual physical state of the slope. Simultaneously, conventional artificial intelligence prediction models only focus on the mathematical time-series changes of apparent monitoring data, lacking constraints from the underlying rock mechanics evolution mechanism. When dealing with complex and unpredictable conditions such as heavy rainfall or sudden water level drops, these purely data-driven models often only issue alarm commands after a significant increase in the absolute displacement of the slope surface reaches the static red line, failing to identify early warning signs of localized fractures within the rock mass from subtle changes in underlying physical parameters and acoustic emission characteristics, resulting in a lag in early warning actions.

[0004] Furthermore, the current early warning system operates on a one-way data flow with on-site disaster prevention and response. After the system triggers an early warning and on-site construction personnel implement contingency plans such as reinforcement or drainage, it cannot quantitatively assess the actual control effectiveness of these specific disaster prevention measures. On-site disaster prevention experience data and control results cannot be fed back into the underlying data architecture of the early warning system, preventing the early warning model from adjusting parameters and evolving its logic based on real feedback from the construction site. Consequently, the system's risk assessment mechanism cannot achieve self-calibration and a closed-loop system. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent early warning and prevention system and method for slope damage in schist reservoir areas. It solves the problems of existing slope early warning systems, which rely on static mechanical parameters and pure data-driven models for a long time, resulting in delayed early warning under complex working conditions, and lack a closed-loop calibration mechanism to feed back the on-site disaster prevention and control effects to the underlying algorithm.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides an intelligent early warning and prevention system for slope damage in schist reservoir areas, comprising: The system includes a data fusion and knowledge base construction module, which integrates indoor experimental data, field monitoring data, and numerical simulation data, and runs a Bayesian update algorithm to calculate and update the slope damage mechanism knowledge base. An intelligent early warning analysis module, connected to the data fusion and knowledge base construction module, incorporates a hybrid early warning model composed of chained physical criteria and an artificial intelligence time-series prediction network. This model receives parameters from the slope damage mechanism knowledge base, performs early warning logic operations, and outputs early warning signals. A cloud-edge collaborative deployment platform, consisting of field edge computing devices and a cloud server, acquires real-time monitoring data and performs model inference, while the cloud server iteratively trains the hybrid early warning model. A prevention and control decision support module, connected to the intelligent early warning analysis module, matches the received early warning signal with a response plan, records feedback data from the field response, and transmits the feedback data to both the data fusion and knowledge base construction module and the intelligent early warning analysis module.

[0007] In the system provided by this invention, to address the problem that static mechanical parameters cannot adapt to the long-term evolution of slope engineering environments, the indoor test data is obtained through a water-rock stress dual-path coupling test. The data fusion and knowledge base construction module is internally configured with a deformation stability criterion as the trigger condition for the parameter update program. When the calculated axial strain rate of the slope rock mass is consistently less than a preset strain rate threshold within a set time period, the system determines that the slope has reached a deformation-stable state and initiates the Bayesian update algorithm. The algorithm extracts the prior probability distribution of the long-term strength attenuation coefficient of the slope, inputs newly received field monitoring data as likelihood evidence into the Bayesian posterior calculation program, calculates and outputs the posterior probability distribution interval of the long-term strength attenuation coefficient, and thus stores the long-term strength attenuation coefficient and the calibrated critical damage thresholds for each stage in the slope damage mechanism knowledge base, achieving dynamic adaptive adjustment of the underlying mechanism parameters over time and with the environment.

[0008] To address the lagging early warning shortcomings of traditional data-driven prediction models, the intelligent early warning analysis module utilizes the principles of continuum mechanics to calculate a damage variable reflecting the degree of rock mass deterioration. This variable is calculated by subtracting the ratio of the rock mass's equivalent elastic modulus to its initial elastic modulus. The chain-like physical criterion compares the calculated damage variable with extracted damage accumulation and local fracturing thresholds. Simultaneously, the artificial intelligence time-series prediction network in the hybrid early warning model employs a long short-term memory network to extract multi-source historical time-series features and predict the displacement acceleration rate within a set time window. When the predicted displacement acceleration rate exceeds an empirical threshold, and the ratio of acoustic emission event rates within a continuous sampling period exceeds the abrupt change critical coefficient, and the chain-like physical criterion determines that the damage variable exceeds the local fracturing threshold, the module triggers and outputs an orange-level early warning signal, thus constructing a mechanism- and data-driven early identification system.

[0009] At the system deployment architecture level, the on-site edge computing device performs filtering and noise reduction preprocessing on the raw displacement signals, pore water pressure signals, and microseismic signals collected by on-site sensors to eliminate outliers. It loads a microservice-encapsulated hybrid early warning model into its local memory and directly performs forward inference operations on the standardized data stream. When extreme environments cause network communication disruptions, the edge computing device maintains independent risk assessment logic using locally fixed mechanistic parameters. The cloud server establishes a distributed database to store historical sequences. When new monitoring feature data reaches a preset data volume threshold, it initiates a retraining program for the artificial intelligence time-series prediction network and distributes the iteratively optimized model and updated physical damage thresholds synchronously.

[0010] To achieve closed-loop control throughout the disaster prevention process, the prevention and control decision support module deploys a prevention and control knowledge base built on a logic rule engine. This knowledge base analyzes the characteristic conditions carried by early warning signals and retrieves them, outputting a response plan containing the coordinates of key inspection areas and construction drilling schemes. After implementation, the module extracts monitoring data sequences within the evaluation time window, calculates the actual deformation convergence rate of the slope rock mass, and compares the deviation with the expected target value in the response plan to generate corresponding confidence weight factors. The data fusion and knowledge base construction module introduces positive confidence weight factors into the Bayesian update algorithm to improve the lower limit of the local rupture threshold. The intelligent early warning analysis module uses confidence weight factors to correct the scores of corresponding early warning rules in the hybrid early warning model, thereby establishing a closed-loop channel for disaster prevention experience data to flow back to the underlying algorithm.

[0011] The second aspect of this invention provides an intelligent early warning and control method for slope damage in schist reservoir areas, applied to the aforementioned intelligent early warning and control system for slope damage in schist reservoir areas, comprising the following steps: The data fusion and knowledge base construction module accesses indoor test data, field monitoring data, and numerical simulation data, and runs a Bayesian update algorithm to calculate and update the slope damage mechanism knowledge base. The intelligent early warning analysis module receives parameters from the slope damage mechanism knowledge base, uses a hybrid early warning model composed of chain-based physical criteria and artificial intelligence time-series prediction network to perform early warning logic operations, and outputs early warning signals. The field edge computing device of the cloud-edge collaborative deployment platform acquires real-time monitoring data and performs model inference, and the cloud server iteratively trains the hybrid early warning model. After receiving the early warning signal, the prevention and control decision support module matches the disposal plan, records the feedback data of the field disposal, and transmits the feedback data to the data fusion and knowledge base construction module and the intelligent early warning analysis module.

[0012] During the execution of this method, the dynamic updating of underlying mechanism parameters is synchronized with the perception of the on-site environment, and physical damage assessment and temporal feature prediction are coupled and calculated in parallel. The response actions generated after the early warning is triggered are transformed into quantified status indicators, and the system parameter boundaries are reshaped through an error evaluation function, forming a complete engineering application link from parameter calibration, risk perception, disaster prevention execution to logical evolution.

[0013] This invention provides an intelligent early warning and prevention system and method for slope damage in schist reservoir areas. It has the following beneficial effects: 1. This invention uses a data fusion and knowledge base construction module to run a Bayesian update algorithm, which integrates indoor test data obtained from the water-rock stress dual-path coupling test with field monitoring data, extracts the prior probability distribution of the long-term strength attenuation coefficient of schist, and calculates the posterior probability distribution interval using field likelihood evidence. This dynamically updates the critical damage threshold in the slope damage mechanism knowledge base, changing the traditional early warning system's long-term reliance on static mechanical parameters. This allows the underlying physical judgment parameters to adaptively adjust with the evolution of the actual water-rock environment in the reservoir area, improving the physical reliability and accuracy of the early warning benchmark under long-term monitoring.

[0014] 2. This invention constructs a hybrid early warning model that couples a chain-like physical criterion with an artificial intelligence time-series prediction network. The intelligent early warning analysis module calculates damage variables based on the decay of the equivalent elastic modulus and displacement acceleration rates predicted by a long short-term memory network in parallel, and performs a comprehensive judgment in conjunction with the monitoring status of acoustic emission event rates. This overcomes the early warning lag defect of a single pure data prediction model when dealing with complex and sporadic geological conditions, enabling the system to identify the precursor features of local rupture inside the rock mass and output early warning signals before the absolute displacement on the slope surface reaches the conventional alarm threshold.

[0015] 3. Based on the cloud-edge collaborative deployment architecture, this invention establishes a closed-loop mechanism for the feedback of disaster prevention experience to the algorithm layer. After the prevention and control decision support module matches the emergency response plan and implements it on-site, it generates a confidence weight factor by calculating the deviation between the actual deformation convergence rate and the expected target. The system feeds back this confidence weight factor to improve the lower limit of the local rupture threshold in the Bayesian update algorithm and to correct the score of the corresponding early warning rule in the hybrid early warning model. The actual disaster prevention reinforcement and drainage effects on the engineering site are transformed into quantitative parameters to participate in the next iteration of the model, realizing the self-calibration of the judgment logic of the entire early warning and prevention system. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram illustrating the multi-source data fusion and dynamic knowledge base construction of the present invention; Figure 3 This is a logical architecture diagram of the hybrid early warning model of the present invention; Figure 4 This is a diagram of the cloud-edge collaborative system deployment and prevention closed-loop system architecture of the present invention; Figure 5 This is a comparison curve of the monitoring time sequence and multi-level early warning triggering of slope displacement in a certain rock reservoir area according to the present invention; Figure 6 This is a convergence verification diagram showing the error between the model-predicted displacement acceleration rate and the actual displacement acceleration rate in the field. Detailed Implementation

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See attached document Figure 1 and Figure 4 This invention provides an intelligent early warning and prevention system for slope damage in schist reservoir areas, comprising: a data fusion and knowledge base construction module, an intelligent early warning analysis module, a cloud-edge collaborative deployment platform, and a prevention and control decision support module.

[0019] The data fusion and knowledge base construction module is configured to access and process indoor experimental data, field monitoring data, and numerical simulation data. Internally, this module runs an adaptive learning algorithm to calculate and update the slope damage mechanism knowledge base.

[0020] The intelligent early warning analysis module incorporates a hybrid early warning model that couples physical criteria with an artificial intelligence time-series prediction algorithm. This module communicates with the data fusion and knowledge base construction module to receive parameters from the slope damage mechanism knowledge base and execute early warning logic operations.

[0021] The cloud-edge collaborative deployment platform consists of on-site edge computing devices and a cloud server. The edge computing devices connect to various on-site monitoring sensors to acquire real-time monitoring data and perform preprocessing and model inference. The cloud server is used to store historical data and iteratively train the hybrid early warning model.

[0022] The prevention and control decision support module has a built-in prevention and control knowledge base. It connects to the intelligent early warning analysis module to match corresponding emergency response plans upon receiving an early warning signal. This module is also configured to record feedback data from on-site responses and transmit the feedback data to the data fusion and knowledge base construction module.

[0023] Based on the above system architecture, the intelligent early warning and prevention method for schist reservoir slope damage of the present invention includes the following working steps.

[0024] The system first acquires indoor physical and mechanical test data. Based on the geological conditions of the target reservoir bank slope, researchers prepare schist joint specimens with pre-set dip angles. The schist joint specimens are placed in a triaxial pressure-flow coupling testing machine, where graded axial stresses and periodically varying reservoir water pressures are applied simultaneously. During this process, the testing equipment acquires macroscopic mechanical response data and acoustic emission signals from the rock samples, and extracts microstructural parameters of the schist joint specimens before and after the test.

[0025] The system executes multi-source data fusion and knowledge base construction steps. The data fusion and knowledge base construction module acquires the mechanism data from the aforementioned indoor physical and mechanical experiments, real-time monitoring data of the target slope, and numerical simulation results. Using a preset deformation stability criterion as a trigger condition, the data fusion and knowledge base construction module applies an adaptive learning algorithm to calibrate the damage threshold and model parameters of the schist, thereby establishing a slope damage mechanism knowledge base.

[0026] The system executes the steps of constructing a hybrid early warning model. The intelligent early warning analysis module establishes a chain-like physical criterion based on damage thresholds in the slope damage mechanism knowledge base, while simultaneously training a time-series prediction model using historical monitoring data. The intelligent early warning analysis module combines the chain-like physical criterion with the time-series prediction model to form a hybrid early warning model.

[0027] The system deploys and issues early warnings through an early warning platform. A hybrid early warning model is deployed on a cloud-edge collaborative deployment platform. On-site edge computing devices receive real-time monitoring data from sensors and input it into the hybrid early warning model for inference and comparison. When the monitored indicators reach preset threshold conditions, the cloud-edge collaborative deployment platform outputs early warning information of the corresponding level.

[0028] The system executes prevention and control decision-making and feedback steps. The prevention and control decision support module receives early warning information, retrieves and outputs a response plan from the prevention and control knowledge base based on the early warning level and disease type. After construction personnel carry out on-site response according to the response plan, the system collects the on-site monitoring results after the response. The prevention and control decision support module converts the monitoring results into weighted data and inputs it into the data fusion and knowledge base construction module for subsequent updates to the damage threshold range and adjustments to the confidence level of the early warning rules.

[0029] See attached document Figure 1 The experimental part of this invention specifically includes performing water-rock-stress dual-path coupling tests and macro- and micro-damage source tracing.

[0030] Based on the engineering geological conditions of the target reservoir bank, researchers drilled schist cores from representative rock strata on-site. The drilled cores were then processed into standard cylindrical specimens using rock cutting equipment. The end faces of the standard cylindrical specimens were ground to ensure that the flatness and perpendicularity of both ends met the standard rock mechanics testing specifications. Joint surfaces of different specifications were fabricated on the processed standard cylindrical specimens using precision cutting equipment. The fabricated specimens included discontinuous joint specimens and single joint specimens, with the angles between the pre-fabricated joint surfaces and the specimen axis set at 30°, 40°, 50°, and 60°. According to the preset penetration requirements, the cutting depth of the joint surfaces was controlled at 1cm, 2cm, 3cm, and 4cm, respectively.

[0031] The processed schist jointed specimens were placed in the main pressure chamber of the triaxial pressure-flow coupling testing machine. This machine integrates an axial servo loading system, a confining pressure control system, a pore water pressure control system, and a constant temperature control system. Before the test, the ambient temperature inside the pressure chamber was adjusted and kept constant at 30°C by the constant temperature control system to simulate the actual soil and rock temperature boundary conditions of the schist reservoir drawdown zone.

[0032] The testing machine applies graded axial stress to the jointed schist specimens. Based on the preliminary test results of standard unjointed schist specimens from the same batch, the estimated uniaxial compressive strength of this type of schist is obtained. 10% of this estimated uniaxial compressive strength is set as a fixed loading interval. The axial servo loading system increases the axial stress applied to the top of the jointed schist specimen in stages according to this loading interval.

[0033] During each stage of constant axial stress, the pore water pressure control system simultaneously performs a cyclic loading procedure on the schist jointed specimen. The set peak pore water pressure levels are divided into three independent levels: 0.10 MPa, 0.20 MPa, and 0.30 MPa. Within a complete test cycle, the pore water pressure control system first uniformly increases the water pressure to one of the set peak levels at a rate of 0.05 MPa / min, and then maintains a stable pressure at that peak. Once the specimen's deformation reaches a preset stability criterion, the pore water pressure control system uniformly reduces the water pressure back to its initial state at the same rate.

[0034] Subsequently, the pore water pressure control system empties the permeable water from the pressure chamber. The constant temperature control system introduces circulating gas at a set temperature and humidity, allowing the schist jointed sample to undergo a dehydration and air-drying process for a specified time. This depressurization and air-drying process corresponds to the periodic drop in reservoir water level and the exposure stage of the slope rock mass. The periodic rise and fall of pore water pressure and the introduction of circulating gas generate dynamic water pressure scouring and capillary tension generated by alternating wet and dry conditions within the micro-fractures of the schist, thereby recreating the softening and exfoliation effect of the actual drawdown zone's water-stress coupled environment on the schist's unique clay mineral components.

[0035] During the water-rock-stress dual-path coupling test, the data acquisition system recorded the physical response of the schist jointed specimens in real time. Linear variable displacement sensors and circumferential extensometers installed on the outside of the specimens measured the axial and radial displacements, respectively, and the computer converted the displacement data into stress-strain curves. Acoustic emission sensors arranged on the specimen base or sidewalls continuously acquired elastic waves released during the development of microcracks within the rock, recording the number of acoustic emission events and energy release rate data.

[0036] After completing the mechanical tests, researchers extracted micro-rock fragments from near the main fracture zone of the fractured schist jointed sample and dehydrated and dried them. X-ray diffraction was used to analyze the mineral composition of the micro-rock fragments before and after the tests, obtaining data on the evolution of clay mineral content within the schist. Scanning electron microscopy was used to image the fracture morphology of the micro-rock fragments. Image analysis software was used to perform grayscale recognition and binarization extraction on the microscopic images acquired by the scanning electron microscope, quantitatively calculating the changes in porosity and microfracture density within the schist. By combining the changes in microscopic parameters with the macroscopic mechanical attenuation curves, the damage source of the schist degradation process was traced.

[0037] See attached document Figure 2 The multi-source data fusion and dynamic knowledge base adaptive update process provided by this invention is executed in the data fusion and knowledge base construction module.

[0038] The data fusion and knowledge base construction module acquires slope engineering parameters from different dimensions. This module receives macroscopic mechanical attenuation curves and microstructural change values ​​from the water-rock stress dual-path coupling test. Simultaneously, the module reads monitoring data transmitted from sensor arrays deployed at the target slope site via a network interface. On-site sensors collect multi-dimensional engineering information, including absolute displacement of the slope surface output by the Global Navigation Satellite System, relative deformation of the rock strata recorded by the crack gauge, pore water head measured by the groundwater level gauge, and elastic wave signals generated by internal rock fractures acquired by the microseismic equipment. The module also imports slope failure simulation data calculated using numerical simulation software.

[0039] The data fusion and knowledge base construction module performs timestamp alignment on the aforementioned multi-source heterogeneous data to eliminate time errors caused by inconsistent sampling frequencies of various sensors, constructing a multi-source information matrix reflecting the current physical state of the slope. The system sets a deformation stability criterion in this module, which serves as a trigger for subsequent calculations. The system continuously receives the deformation of the rock mass and performs time differentiation to obtain the axial strain rate. When the calculated axial strain rate is less than 0.001 mm / h for two consecutive hours, the data fusion and knowledge base construction module determines that the rock mass has reached a deformation stability state. At this point, the module extracts the multi-source information matrix data under this stable state and initiates the parameter update program.

[0040] In the parameter update procedure, the data fusion and knowledge base construction module runs a Bayesian update algorithm to iteratively calculate the damage threshold and constitutive model parameters of the schist. Let the model parameter vector to be updated be... This vector contains the long-term strength attenuation coefficient and mechanical elastic modulus, reflecting the softening characteristics of schist when exposed to water. Let the new set of multi-source information matrix data acquired by the data fusion and knowledge base construction module at the current time point be denoted as . The algorithm extracts the prior distribution of the model parameter vector. And combined with the likelihood function that reflects the actual state of the site. Calculate the posterior distribution of the model parameter vector. The calculation follows the formula. .

[0041] Through multiple rounds of Bayesian formula calculations, the data fusion and knowledge base construction module utilizes the newly added field monitoring data in each round. Continuously adjust the model parameter vector The probability distribution range is determined. This calculation process transforms the static initial mechanical parameters obtained from indoor tests into dynamic parameters that evolve with the reservoir environment and time. The data fusion and knowledge base construction module stores the updated long-term strength attenuation coefficient and the calibrated critical damage thresholds for each stage in a structured manner, generating a slope damage mechanism knowledge base. The slope damage mechanism knowledge base is updated synchronously with the continuous input of field monitoring data, providing basic threshold support for the system's early warning logic.

[0042] See attached document Figure 3 The hybrid early warning model driven by both mechanism and data in this invention is calculated and constructed in the intelligent early warning analysis module.

[0043] The intelligent early warning analysis module reads dynamic mechanical parameters from the slope damage mechanism knowledge base through a data interface. It then uses the principles of continuum mechanics to calculate damage variables reflecting the degree of schist degradation. Damage variables The calculation follows the formula .in, This represents the initial elastic modulus of schist when it is not disturbed by engineering work. Representative schist experience Equivalent elastic modulus after secondary water-rock stress cyclic coupling.

[0044] The intelligent early warning analysis module extracts two key critical values ​​in the schist evolution process based on a slope damage mechanism knowledge base. These two values ​​are the damage accumulation threshold and the damage accumulation threshold, respectively. and local rupture threshold The intelligent early warning analysis module compares the calculated damage variables in real time. Compared to the above threshold. When the damage variable Exceeding the damage accumulation threshold At that time, the intelligent early warning analysis module determines and records that the current slope rock mass is in the damage accumulation stage. When the damage variable Further exceeding the local rupture threshold Furthermore, when the on-site sensor array detects a sudden change in the rock mass shear displacement, the intelligent early warning analysis module determines that the current slope rock mass has entered the local fracturing stage.

[0045] While determining physical thresholds, the intelligent early warning analysis module performs parallel artificial intelligence time series prediction. The module employs a Long Short-Term Memory (LSTM) network or a Transformer network as its underlying architecture. It extracts long-term accumulated displacement, water pressure, and stress data from field sensors as input feature sequences. Through forward propagation and gradient backpropagation of historical multi-source time series features, the module outputs a predicted value for the slope displacement acceleration rate within a set future time window. .

[0046] The intelligent early warning analysis module performs a hybrid early warning logic threshold comparison. It extracts an empirical threshold for displacement acceleration based on historical rupture case statistics. When the predicted value output by the artificial intelligence time series prediction model Greater than this empirical threshold When the intelligent early warning analysis module detects a simultaneous surge in the acoustic emission event rate, it extracts the current physical judgment status. The judgment logic for a surge in the acoustic emission event rate is as follows: calculate the ratio of the average number of acoustic emission events within the current time window to the average number of acoustic emission events within a historical undisturbed background time window. If this ratio exceeds the set acoustic emission mutation threshold for three consecutive sampling periods, it is determined to be a surge in the acoustic emission event rate. If the chain-like physical criteria determine that the slope is in a local fracturing stage at this time, the logic control unit of the intelligent early warning analysis module triggers and outputs an orange warning signal.

[0047] The intelligent early warning analysis module continuously receives the overall macroscopic deformation magnitude and regional stability parameters of the slope. When physical criteria combined with geomechanical analysis confirm that the slope deformation exceeds the set safety tolerance, and the artificial intelligence prediction model indicates that the acceleration rate shows a divergent trend, the intelligent early warning analysis module determines that the slope slip surface has been connected. At this time, the intelligent early warning analysis module triggers and outputs a red warning signal.

[0048] See attached document Figure 4 The early warning platform deployment and early warning release process of this invention is executed in the cloud-edge collaborative deployment platform.

[0049] Researchers have encapsulated the trained and validated hybrid early warning model into lightweight microservice components. The cloud-edge collaborative deployment platform adopts a physical architecture that combines edge computing devices with cloud servers. The edge computing devices are deployed at the on-site monitoring stations on the target slope and connect to various on-site monitoring sensors via data interfaces. The edge computing devices perform filtering and noise reduction preprocessing on the raw displacement signals, pore water pressure signals, and microseismic signals collected in real time by the on-site sensors. This preprocessing eliminates outliers caused by environmental background noise and abnormal fluctuations in equipment levels, outputting a data stream that meets the input standards of the hybrid early warning model.

[0050] Edge computing devices load hybrid early warning model microservices into their local memory. They directly perform forward inference operations on the input standardized data stream. The edge computing devices then transmit the preliminary early warning assessment results generated by the inference, along with compressed on-site feature data, to the cloud server via a communication network. When extreme geological conditions cause on-site network communication disruptions, the edge computing devices utilize locally fixed slope damage mechanism parameters and the hybrid early warning model microservices to maintain independent risk assessment logic on-site and directly trigger local alarm devices to output early warning signals via the regional bus.

[0051] The cloud server handles global resource scheduling and high-concurrency computing tasks for the cloud-edge collaborative deployment platform. It establishes a distributed database to store historical monitoring data from multiple edge computing devices, indoor macro- and micro-physical test data, and the on-site handling feedback information described below. The cloud server utilizes its own computing cluster to run a deep learning environment. When the accumulated new monitoring feature data in the database reaches a preset data volume threshold, the cloud server initiates the retraining process for the time-series prediction model. The cloud server packages the iteratively optimized hybrid early warning model microservice and the updated physical damage thresholds, and distributes them synchronously to the corresponding on-site edge computing devices.

[0052] The cloud-edge collaborative deployment platform is a visualized early warning platform supporting access from both web and mobile devices. The cloud server pushes processed engineering data and analysis results to various display modules of this visualized early warning platform. The geographic information system map display module spatially maps and renders the physical coordinates of on-site sensors with the established 3D geological model of the slope. The real-time data curve module plots the trend lines of various monitoring indicators within a set time window. The early warning information dashboard module outputs the triggered early warning level, trigger time, and specific location of the damage in list format. The visualized early warning platform includes a report export module for business personnel to download historical data and is configured with a manual experience import interface to receive correction parameters for the early warning logic control thresholds from external geological experts.

[0053] See attached document Figure 4 The closed-loop process of prevention and control decision-making and disaster prevention experience feedback in this invention is executed in the prevention and control decision support module.

[0054] Researchers deployed a prevention and control knowledge base within the prevention and control decision support module. This knowledge base, built on a logic rule engine, stores standardized treatment data for the softening and deterioration characteristics of schist when exposed to water and the destructive properties of joint control. When the intelligent early warning analysis module outputs an early warning signal, the prevention and control decision support module simultaneously receives it. The module then analyzes the early warning signal to determine the warning level, the affected area, and the specific type of slope damage (such as rock creep or localized landslide). Finally, the module uses the logic rule engine to perform conditional searches and parameter matching within the prevention and control knowledge base.

[0055] The prevention and control decision support module outputs corresponding emergency response plans based on the search results. These plans include the coordinates of key inspection areas, the drilling layout for groundwater drainage and pressure reduction, evacuation routes for personnel and equipment, and response time limits for the responsible departments. The module then distributes these plans to the receiving terminals of on-site construction and management personnel via the communication network. The on-site disaster prevention team implements specific reinforcement and drainage measures on the target slope according to the plan.

[0056] After the implementation of the mitigation measures, the system continuously collects and uploads engineering monitoring data such as displacement and pore water pressure of the target slope through on-site edge computing devices. The prevention and control decision support module sets a fixed evaluation time window. The module extracts the monitoring data sequence within this evaluation time window and calculates state indicators such as the deformation convergence rate and pore water pressure dissipation rate of the slope rock mass. The module compares the calculated actual state indicators with the expected target values ​​in the original mitigation plan to generate a quantitative assessment value of the actual implementation effect of the mitigation measures. Based on a preset mapping function, the module converts the quantitative assessment value into a numerical confidence weight factor. When the actual deformation convergence rate is higher than the expected target value, the module outputs a positive confidence weight factor.

[0057] The prevention and control decision support module sends the feedback data stream containing the confidence weight factor back to the data fusion and knowledge base construction module and the intelligent early warning analysis module, respectively. The data fusion and knowledge base construction module introduces the positive confidence weight factor into the Bayesian update algorithm, increasing the prior probability of the corresponding slope parameters, thereby raising the lower limit of the corresponding local rupture threshold in the slope damage mechanism knowledge base. The intelligent early warning analysis module uses this confidence weight factor to increase the score of the early warning rule corresponding to this type of disease in the hybrid early warning model. Conversely, when the actual state indicators do not meet expectations, a negative confidence weight factor is output to lower the model threshold. This closed-loop feedback of disaster prevention and control data completes the parameter calibration and logical evolution of the entire system.

[0058] Specific application examples: To aid in understanding the present invention, a schist slope in the drawdown zone of a water conservancy project reservoir is used as a specific application scenario for verification and illustration. Engineers deployed surface displacement gauges, deep inclinometers, groundwater pressure gauges, and a microseismic acoustic emission sensor array in the slope's slip deformation-sensitive area. Under the operational conditions of reservoir impoundment and periodic water level regulation, the cloud-edge collaborative deployment platform continuously receives feature data transmitted from multiple source sensors on-site.

[0059] Based on the results of previous indoor dual-path coupling tests conducted on the schist of this slope, the data fusion and knowledge base construction module extracted the initial elastic modulus of the schist and the long-term strength attenuation coefficient reflecting its softening characteristics upon contact with water. Combined with baseline monitoring data from the initial on-site phase, the module ran a Bayesian update algorithm to determine the damage accumulation threshold and local fracturing threshold of the rock mass in the target area. With repeated periodic rises and falls in the reservoir water level, the damage variables calculated by the intelligent early warning analysis module gradually accumulated over time and with environmental evolution.

[0060] Combined with appendix Figure 5 The monitoring time series and early warning triggering curves are compared to illustrate the actual effects. Under the accidental conditions of continuous heavy rainfall and a sudden drop in water level in the reservoir area, the displacement monitoring data curve of the slope surface shows an accelerating upward trend. (Attached) Figure 5 The diagram illustrates a comparison of the response nodes of the hybrid early warning model of this invention with those of a conventional pure data-driven prediction model. When the time node reaches the first feature assessment moment shown in the diagram, the conventional pure data-driven prediction model does not output any action command because the absolute value of the input surface displacement has not yet reached the statically set alarm threshold. The hybrid early warning model of this invention extracts the physical criterion state at the same time, confirming that the calculated damage variable has exceeded the local fracture threshold. Simultaneously, the intelligent early warning analysis module identifies that the acoustic emission event rate exceeds the abrupt change critical coefficient within three consecutive sampling periods, and the displacement acceleration rate output by the artificial intelligence time-series prediction is greater than the empirical threshold. Based on the above triggering conditions, the logic control unit of the intelligent early warning analysis module outputs an orange early warning signal in advance. This process verifies that by introducing physical mechanisms and dynamic damage thresholds, the system can identify precursory fracture characteristics within the rock mass earlier from changes in underlying physical parameters, overcoming the early warning lag defect of the pure data-driven model under complex working conditions.

[0061] See attached document Figure 6 This graph is used to verify the effectiveness of the data closed-loop feedback mechanism formed between the prevention and control decision support module and the data fusion and knowledge base construction module. The scatter and broken line distribution in the graph reflects the residual evolution between the predicted output value and the actual field observation value of the hybrid early warning model under multiple rounds of field data input and parameter iteration cycles.

[0062] In the initial stage of system deployment and operation, since the knowledge base parameters mainly rely on prior calibration of indoor static physical test data, supplementary... Figure 6The data distribution on the left shows that the model prediction error fluctuates to some extent. When local deformation occurred on site and triggered the emergency response plan issued by the prevention and control decision support module, the on-site construction team implemented disaster prevention measures such as surface drainage and slope toe ballast. The system continuously records the actual deformation convergence rate of the slope after implementing the above measures. The prevention and control decision support module converts this actual deformation convergence rate into a confidence weight factor and feeds it back to the data layer. The adaptive learning algorithm receives this confidence weight factor and the multi-source monitoring matrix before and after deformation, starts the Bayesian posterior calculation program, and re-corrects the long-term strength attenuation coefficient in the slope damage mechanism knowledge base. (See attached...) Figure 6 As shown by the data curve on the right, after multiple feedback loop cycles of parameter self-evolution, the residual between the displacement acceleration rate predicted by the model and the actual displacement acceleration rate observed in the field gradually decreases and converges to a stable confidence interval near zero. This error convergence verification process proves that the system architecture provided by this invention has the ability to perform parameter self-calibration and logical evolution in complex geological environments, ensuring the physical reliability of the early warning model under long-term operation.

Claims

1. A smart early warning and prevention system for slope damage in schist reservoir areas, characterized in that, include: The data fusion and knowledge base construction module is used to access indoor test data, field monitoring data and numerical simulation data, and run the Bayesian update algorithm to calculate and update the slope damage mechanism knowledge base; The intelligent early warning analysis module is connected to the data fusion and knowledge base construction module. It has a built-in hybrid early warning model composed of chain-like physical criteria and artificial intelligence time-series prediction network. It is used to receive parameters from the slope damage mechanism knowledge base, perform early warning logic operations, and output early warning signals. The cloud-edge collaborative deployment platform consists of on-site edge computing devices and a cloud server. The on-site edge computing devices are used to acquire real-time monitoring data and perform model inference, while the cloud server is used to iteratively train the hybrid early warning model. The prevention and control decision support module is connected to the intelligent early warning analysis module. It is used to match the response plan after receiving the early warning signal, record the feedback data of the on-site response, and transmit the feedback data to the data fusion and knowledge base construction module and the intelligent early warning analysis module.

2. The intelligent early warning and prevention system for schist reservoir slope damage according to claim 1, characterized in that, The indoor test data were obtained through a dual-path coupling test of water and rock stress. The Bayesian update algorithm running in the data fusion and knowledge base construction module is used to extract the prior probability distribution of the long-term strength attenuation coefficient of the slope, and input the newly received field monitoring data as likelihood evidence into the Bayesian posterior calculation program to calculate and output the posterior probability distribution interval of the long-term strength attenuation coefficient. Thus, the long-term strength attenuation coefficient and the calibrated critical damage thresholds for each stage are stored in the slope damage mechanism knowledge base.

3. The intelligent early warning and prevention system for schist reservoir slope damage according to claim 2, characterized in that, The data fusion and knowledge base construction module has a deformation stability criterion set internally as the trigger condition for the parameter update program; When the data fusion and knowledge base construction module calculates that the axial strain rate of the slope rock mass is less than the preset strain rate threshold for a continuous set time, it determines that the slope has reached a stable deformation state and starts the Bayesian update algorithm.

4. The intelligent early warning and prevention system for schist reservoir slope damage according to claim 1, characterized in that, The intelligent early warning analysis module uses the principle of continuum mechanics to calculate damage variables that reflect the degree of rock mass deterioration. The calculation model of the damage variables is a value minus the ratio of the equivalent elastic modulus of the rock mass to the initial elastic modulus. The logic of the chain-like physical criterion is to compare the calculated damage variable with the damage accumulation threshold and local rupture threshold extracted from the slope damage mechanism knowledge base.

5. The intelligent early warning and prevention system for schist reservoir slope damage according to claim 4, characterized in that, The artificial intelligence time series prediction network in the hybrid early warning model adopts a long short-term memory network to extract multi-source historical time series features and predict the displacement acceleration rate within a set time window. When the displacement acceleration rate output by the artificial intelligence time-series prediction network is greater than the empirical threshold, and the ratio of acoustic emission event rate within a continuous sampling period exceeds the mutation critical coefficient, and the chain-like physical criterion determines that the damage variable exceeds the local rupture threshold, the intelligent early warning analysis module triggers and outputs the orange-level early warning signal.

6. The intelligent early warning and prevention system for schist reservoir slope damage according to claim 1, characterized in that, The on-site edge computing device is used to perform filtering and noise reduction preprocessing on the raw displacement signals, pore water pressure signals and micro-vibration signals collected by on-site sensors to eliminate outliers; The on-site edge computing device loads the microservice-encapsulated hybrid early warning model in local memory and directly performs forward inference operations on the standardized data stream; When network communication is interrupted, the field edge computing device uses locally fixed mechanism parameters to maintain independent risk assessment logic.

7. The intelligent early warning and prevention system for schist reservoir slope damage according to claim 6, characterized in that, The cloud server has a distributed database for storing historical sequences of the on-site monitoring data. When the newly added monitoring feature data accumulated in the distributed database reaches the preset data volume threshold, the cloud server starts the retraining program of the artificial intelligence time series prediction network and sends the iteratively optimized hybrid early warning model and the updated physical damage threshold to the corresponding on-site edge computing device.

8. The intelligent early warning and prevention system for schist reservoir slope damage according to claim 1, characterized in that, The prevention and control decision support module is equipped with a prevention and control knowledge base built on a logic rule engine. The prevention and control decision support module analyzes the warning level and specific slope damage type carried by the warning signal, uses the logical rule engine to perform conditional retrieval in the prevention and control knowledge base, and outputs the treatment plan containing the coordinates of key inspection areas and the construction hole layout scheme.

9. The intelligent early warning and prevention system for schist reservoir slope damage according to claim 8, characterized in that, The prevention and control decision support module extracts the monitoring data sequence within the evaluation time window, calculates the actual deformation convergence rate of the slope rock mass, and compares the actual deformation convergence rate with the expected target value in the treatment plan to generate the corresponding confidence weight factor as the feedback data. The data fusion and knowledge base construction module introduces the positive confidence weight factor into the Bayesian update algorithm to improve the lower limit of the local rupture threshold. The intelligent early warning analysis module uses the confidence weight factor to correct the score of the corresponding early warning rule in the hybrid early warning model.

10. A method for intelligent early warning and prevention of slope damage in schist reservoir areas, characterized in that, The intelligent early warning and prevention system for schist reservoir slope damage, applied to any one of claims 1-9, comprises the following steps: The data fusion and knowledge base construction module accesses indoor test data, field monitoring data and numerical simulation data, and runs the Bayesian update algorithm to calculate and update the slope damage mechanism knowledge base. The intelligent early warning analysis module receives parameters from the slope damage mechanism knowledge base, performs early warning logic operations using a hybrid early warning model composed of chain-based physical criteria and artificial intelligence time-series prediction network, and outputs early warning signals. The edge computing devices on the cloud-edge collaborative deployment platform acquire real-time monitoring data and perform model inference, while the cloud server iteratively trains the hybrid early warning model. Upon receiving the early warning signal, the prevention and control decision support module matches the response plan, records the feedback data of the on-site response, and transmits the feedback data to the data fusion and knowledge base construction module and the intelligent early warning analysis module.

Images