A dynamic geological modeling method and system based on multi-source heterogeneous data adaptive update

Abstract

The application relates to the field of mine geological modeling, in particular to a dynamic geological modeling method and system based on adaptive updating of multi-source heterogeneous data. The method comprises the following steps: acquiring tailing pond state feature data set, constructing a time-space lag prediction model through a machine learning method according to the tailing pond state feature data set; acquiring a tailing pond operation plan information set, predicting a pore water pressure change information set inside a tailing dam in a future period of time by using the time-space lag prediction model based on the tailing pond operation plan information set and a current ore pulp liquid level state; and performing dynamic evaluation of tailing dam stability according to the pore water pressure change information set and outputting stability early warning information. In the mine geological modeling process, timely allocation of emergency resources and accurate grasp of prevention and control time are realized, and chain instability of dam body and disastrous dam break accidents are reduced.

Description

Technical Field

[0001] This application relates to the field of mine geological modeling, and in particular to a dynamic geological modeling method and system based on adaptive updates of multi-source heterogeneous data. Background Technology

[0002] In the key area of ​​efficient resource utilization, tailings dam risk accurate early warning and stability control technology has become the core support for maintaining safe production in mines and preventing major environmental risks. It determines the safety of people's lives and property downstream and the safety of regional ecological environment, and directly affects the sustainable development of mining and social harmony and stability.

[0003] However, existing tailings dam stability monitoring and early warning methods mostly focus on macroscopic monitoring of physical and mechanical indicators, failing to fully consider the hysteresis effect and nonlinear evolution characteristics of the complex response inside the dam system under multi-field coupling. This results in insufficient sensitivity for early identification of potential risks and limited early warning accuracy, which not only leads to misallocation of emergency resources and delays in prevention and control, but may also trigger a chain reaction of dam instability and catastrophic dam failure accidents. Summary of the Invention

[0004] This application provides a dynamic geological modeling method and system based on adaptive updates of multi-source heterogeneous data to solve the above-mentioned technical problems.

[0005] In a first aspect, this application provides a dynamic geological modeling method based on adaptive updates of multi-source heterogeneous data. The method includes: acquiring a tailings dam state characteristic dataset; constructing a spatiotemporal lag prediction model based on the tailings dam state characteristic dataset using machine learning methods; acquiring a tailings dam operation plan information set; predicting a set of pore water pressure changes inside the tailings dam in future periods based on the tailings dam operation plan information set and the current slurry level state using the spatiotemporal lag prediction model; and performing dynamic stability assessment of the tailings dam based on the pore water pressure change information set, and outputting stability early warning information.

[0006] The above technical solution first acquires a dataset of tailings dam state characteristics and then uses machine learning to construct a spatiotemporal lag prediction model to capture the spatiotemporal lag effect of state changes. Next, it acquires a set of tailings dam operation plan information and the current slurry level, combining this with the model to predict future changes in pore water pressure within the tailings dam. Finally, based on this information set and the dam's structure and material properties, it performs a dynamic stability assessment, outputting stability warning information including warning level, area, and countermeasures, which is then sent to control personnel. By accurately capturing the spatiotemporal lag effect of tailings dam state and combining operation plans with real-time status to achieve accurate prediction of pore water pressure, it enables dynamic assessment and timely warning of tailings dam stability. This significantly improves the real-time performance and accuracy of tailings dam safety monitoring, allowing for early detection of risks and providing management with more time to respond, effectively reducing the probability of accidents such as dam failures, and ensuring safe and sustainable mine production.

[0007] Optionally, the acquisition of the tailings dam state characteristic dataset includes: joint monitoring using radar level gauges deployed at the clarified liquid level in the dam area and ultrasonic flow meters deployed at the ore discharge branch pipes; and using data fusion technology to eliminate monitoring errors caused by floating debris and rainfall disturbances on the dam surface, generating a high-confidence liquid level dynamic change sequence; periodically collecting tailings slurry samples and surrounding soil from different ore discharge points for laboratory particle size and chemical composition analysis; and constructing a slurry spatial distribution model reflecting the tailings particle size distribution, permeability coefficient, and mineral chemical composition of different sedimentary areas based on spatial interpolation algorithms. The tailings slurry physicochemical properties data are generated. Using an array of pore pressure sensors embedded at different elevations and in different depositional areas within the tailings dam, regional permeability data is extracted. Sensor locations sensitive to ore discharge operations and exhibiting significant permeability characteristics are marked as key monitoring points to ensure accurate capture of the dynamic water pressure response along the main seepage paths within the dam, generating pore pressure response time-series data within the dam. Based on the liquid level dynamic change sequence, the tailings slurry physicochemical properties data, and the pore pressure response time-series data within the dam, the tailings dam state characteristic dataset is generated.

[0008] Optionally, the step of constructing a spatiotemporal lag prediction model based on the tailings dam state feature dataset using machine learning methods includes: extracting temporal correlation features between liquid level change events and corresponding pore pressure responses at each key monitoring point of the dam body in different historical periods, based on the time-series data of pore pressure response inside the dam body and the dynamic change sequence of liquid level; the temporal correlation features are used to characterize the time dependency between liquid level changes and pore pressure responses; generating a lag fingerprint coefficient set based on physical and chemical lag fingerprint coefficients; constructing a lag primitive model for each key monitoring point based on the lag fingerprint coefficient set, which can quantitatively describe the magnitude and time of pressure response caused by liquid level changes; and integrating the lag primitive models of all key points to generate the spatiotemporal lag prediction model.

[0009] Optionally, generating a set of hysteresis fingerprint coefficients based on physical and chemical hysteresis fingerprint coefficients includes: determining the unique level-pore pressure hysteresis pattern of each key monitoring point through event alignment and causal correlation analysis based on the liquid level change event and the temporal correlation characteristics, and generating the physical hysteresis fingerprint coefficient containing hysteresis time and response amplitude; analyzing the catalytic or inhibitory effects of different chemical components on the pore water pressure transmission process based on the physicochemical properties data of the tailings slurry, and generating the chemical hysteresis fingerprint coefficient characterizing the effect of chemical reactions on the hysteresis time adjustment; and combining the physical and chemical hysteresis fingerprint coefficients to generate the set of hysteresis fingerprint coefficients.

[0010] Optionally, the step of determining the unique liquid level-pore pressure hysteresis pattern of each key monitoring point based on the liquid level change event and the temporal correlation characteristics, and generating a physical hysteresis fingerprint coefficient containing hysteresis time and response amplitude, includes: extracting the corresponding pore pressure response curve based on the liquid level change event of each key monitoring point; aligning the liquid level change event and the pore pressure response curve using a dynamic time warping method to generate a co-change segment; within the aligned co-change segment, using a time-domain causal analysis method to determine the correlation strength between liquid level change and pore pressure change at different hysteresis times, and screening out multiple candidate hysteresis periods with a correlation strength higher than a preset strength threshold; for each candidate hysteresis period, determining the pore pressure change caused by a unit liquid level change within this period as the response amplitude, and verifying it by combining the statistical significance of this period's recurrence in historical events, and eliminating candidate periods with statistical significance lower than a preset level; integrating the retained hysteresis periods and their corresponding response amplitudes to generate the physical hysteresis fingerprint coefficient containing the hysteresis time and response amplitude.

[0011] Optionally, the step of analyzing the catalytic or inhibitory effects of different chemical components on the pore water pressure transmission process based on the physicochemical properties data of the tailings slurry, and generating a chemical hysteresis fingerprint coefficient that determines the effect of chemical reactions on the hysteresis time adjustment, includes: extracting the mineral chemical composition and ion concentration of tailings slurry samples from the sedimentation area of ​​each key monitoring point based on the physicochemical properties data of the tailings slurry; analyzing the enhancing or weakening effect of specific chemical components on water flow resistance during the pore water pressure transmission process, identifying key chemical factors that adjust the hysteresis time; obtaining the statistical relationship between the concentration changes of key chemical factors in real time and the corresponding pore pressure response to the hysteresis time in historical data, quantifying the correction strength of each chemical factor on the hysteresis time, and generating the chemical hysteresis fingerprint coefficient.

[0012] Optionally, the step of acquiring the tailings dam operation plan information set, based on the tailings dam operation plan information set and the current slurry level, and using the spatiotemporal lag prediction model to predict the pore water pressure change information set inside the tailings dam in the future period, includes: the tailings dam operation plan information set includes the planned total discharge volume, the planned discharge point location sequence, and the planned slurry concentration for the future period; based on the planned total discharge volume and the planned slurry concentration, combined with the current slurry level, and through material balance and rheological characteristic analysis, the spatiotemporal change process of the slurry level in the future period is deduced, and a slurry level prediction sequence is generated; based on the planned discharge point location sequence, its... The corresponding sedimentation area is identified, and the physicochemical properties of the tailings slurry corresponding to this area are extracted based on the slurry spatial distribution model. Based on the physicochemical properties of the tailings slurry, the physical influence factors of future ore discharge behavior on the seepage path of the dam body and the chemical influence factors on the lag response of pore water pressure are determined. The liquid level projection sequence, the physical influence factors, and the chemical influence factors are input into the spatiotemporal lag prediction model. Through the liquid level change events and the lag fingerprint coefficient set, the dynamic response process of pore water pressure in different time periods and different dam areas is simulated, generating a set of pore water pressure change information expressed in the form of a spatiotemporal grid.

[0013] Optionally, determining the chemical influencing factors on the hysteretic response to future pore water pressure based on the tailings slurry physicochemical property data includes: extracting the mineral chemical composition of the tailings slurry from the physicochemical property data for the sedimentary area corresponding to the planned ore discharge point location sequence; identifying chemical components whose catalytic or inhibitory effects on the pore water pressure transmission process exceed a preset chemical threshold as key chemical components and obtaining their concentrations; calculating the rate of change in the pore medium structure due to the concentration change of the key chemical components in the pore water environment based on the real-time reaction kinetic parameters of the key chemical components, and then deriving their effect on pore water pressure. The potential impact of the lag time is assessed. Based on the chemical lag fingerprint coefficient, the potential impact is calibrated and normalized to quantify it into a specific chemical correction adjustment value for the lag time. When the chemical correction adjustment value is positive, it indicates that the chemical component has a catalytic effect on the pore water pressure response; when the chemical correction adjustment value is negative, it indicates that the chemical component has an inhibitory effect on the pore water pressure response. The absolute value of the chemical correction adjustment value indicates the strength of the effect. Based on the positive or negative attribute of the chemical correction adjustment value, the chemical influence factor is generated.

[0014] Optionally, the step of dynamically assessing the stability of the tailings dam based on the pore water pressure change information set and outputting stability early warning information includes: analyzing the adverse effects of future increases in pore water pressure on the structural stability of the dam body based on the pore water pressure change information set and the slurry spatial distribution model, dynamically extrapolating the overall safety status for each future period, and generating a future safety factor change curve; analyzing the future safety factor change curve to identify potentially dangerous periods where the safety factor is below the safety threshold or the rate of decrease exceeds the critical rate; and generating the stability early warning information based on the future safety factor change curve and the potentially dangerous periods.

[0015] Secondly, this application provides a dynamic geological modeling system based on adaptive updates of multi-source heterogeneous data. The system includes: a model building module for acquiring a tailings dam state feature dataset and constructing a spatiotemporal lag prediction model using machine learning methods based on the tailings dam state feature dataset; an information extraction module for acquiring a tailings dam operation plan information set and predicting a set of pore water pressure changes inside the tailings dam in future periods based on the tailings dam operation plan information set and the current slurry level state using the spatiotemporal lag prediction model; and a strategy generation module for dynamically assessing the stability of the tailings dam based on the pore water pressure change information set, generating a tailings dam control strategy, and outputting a stability early warning log. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram illustrating an application scenario provided in one embodiment of this application;

[0018] Figure 2 A flowchart illustrating an assessment method for landscaping maintenance provided in one embodiment of this application;

[0019] Figure 3 This is a schematic diagram of the structure of an assessment system for landscaping maintenance provided in one embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0021] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0022] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0023] Existing tailings dam stability monitoring and early warning methods mostly focus on macroscopic monitoring of physical and mechanical indicators, failing to fully consider the hysteresis effect and nonlinear evolution characteristics of the complex internal response of the dam system under multi-field coupling. This results in insufficient sensitivity for early identification of potential risks and limited early warning accuracy, which not only leads to misallocation of emergency resources and delays in prevention and control, but may also trigger a chain reaction of dam instability and catastrophic dam failure accidents.

[0024] Based on this, this application provides a dynamic geological modeling method and system based on adaptive updates of multi-source heterogeneous data. First, a tailings dam state characteristic dataset is acquired, and a spatiotemporal lag prediction model is constructed using machine learning to capture the spatiotemporal lag effect of state changes. Next, a tailings dam operation plan information set and the current slurry level are acquired, and combined with the aforementioned model, a future set of information on changes in pore water pressure inside the tailings dam is predicted. Finally, based on this information set, and combined with the dam structure and material properties, a dynamic stability assessment is performed, outputting stability warning information including warning level, area, and countermeasures, which is then sent to control personnel. By accurately capturing the spatiotemporal lag effect of the tailings dam state and combining the operation plan with real-time status to achieve accurate prediction of pore water pressure, dynamic assessment and timely warning of tailings dam stability are realized. This significantly improves the real-time performance and accuracy of tailings dam safety monitoring, enabling early detection of risks and providing management with time to respond, effectively reducing the probability of accidents such as dam failures, and ensuring mine production safety and sustainable development.

[0025] Figure 1 This is a schematic diagram illustrating an application scenario provided by this application. In mine geological modeling, the method provided in this application can improve the real-time performance and accuracy of tailings dam safety monitoring, enabling early detection of risks and providing management with more time to respond, effectively reducing the probability of accidents such as dam failures, and ensuring mine production safety and sustainable development.

[0026] Specifically, the method of this application is applied to any server that communicates with a multimodal sensor array and a production operation scheduling database. The server acquires a tailings dam state feature dataset provided by the multimodal sensor array and a tailings dam operation plan information set provided by the production operation scheduling database. First, the tailings dam state feature dataset is acquired, and a spatiotemporal lag prediction model is constructed using machine learning to capture the spatiotemporal lag effect of state changes. Next, the tailings dam operation plan information set and the current slurry level are acquired, and the model is combined to predict future changes in pore water pressure inside the tailings dam. Finally, based on this information set and the dam structure and material properties, a dynamic stability assessment is performed, outputting stability warning information including warning level, area, and countermeasures, and sending it to control personnel. Specific implementation details can be found in the following embodiments.

[0027] Figure 2 This is a flowchart illustrating a dynamic geological modeling method based on adaptive updates of multi-source heterogeneous data, provided as an embodiment of this application. The method of this embodiment can be applied to servers in the above-described scenarios. Figure 2 As shown, the method includes:

[0028] S201. Obtain the tailings dam state feature dataset. Based on the tailings dam state feature dataset, construct a spatiotemporal lag prediction model using machine learning methods.

[0029] A tailings dam state characteristic dataset can be a multi-source data set used to characterize the physical state, chemical state, and historical evolution of the tailings dam in the spatiotemporal dimensions. This includes dynamic liquid level change sequences, physicochemical property data of tailings slurry, and time-series data of pore pressure response within the dam body. The data originates from a multimodal sensor array deployed in the tailings dam area. A spatiotemporal lag prediction model can be a machine learning model capable of integrating spatiotemporal dependencies and predicting changes in physical parameters within the tailings dam at a future timeframe. Its key feature is its ability to capture the lag effects of data in both time and space.

[0030] Specifically, the safe operation of tailings dams is dynamically affected by a variety of factors, and these effects exhibit significant temporal and spatial lags. From a temporal perspective, the continuous discharge of slurry gradually alters the dam's load. This load change's impact on the internal stress distribution of the dam is not immediate but accumulates over time, taking hours or even days to fully penetrate the deeper layers of the dam. The impact of rainfall on the dam's seepage flow also lags; rainwater infiltration into the dam and subsequent changes in seepage flow require time, and this lag varies depending on factors such as rainfall intensity and the permeability of the dam materials. From a spatial perspective, displacement in a certain area of ​​the dam may trigger stress redistribution in adjacent areas, leading to a delayed displacement response in distant areas. Changes in water level upstream of the dam will have a delayed impact on pore water pressure downstream of the dam section through water pressure transmission, and the degree of impact varies with distance. Traditional geological modeling methods often employ static analysis or simple time-series predictions, which cannot effectively quantify these complex spatiotemporal lag relationships, resulting in low accuracy in predicting tailings dam status changes and failing to meet safety monitoring requirements. This step involves integrating multi-source heterogeneous monitoring data (i.e., tailings dam state characteristic datasets) and employing machine learning methods (such as spatiotemporal graphical neural networks, temporal convolutional networks, or recurrent neural networks considering lag effects) to construct an intelligent model with spatiotemporal lag prediction capabilities. By acquiring a comprehensive tailings dam state characteristic dataset and utilizing machine learning methods to construct a spatiotemporal lag prediction model, the aforementioned temporal and spatial lag effects can be accurately characterized. This provides a reliable model foundation for subsequent pore water pressure prediction, ensuring that the prediction results truly reflect the dynamic changes of the tailings dam.

[0031] S202. Obtain the tailings dam operation plan information set. Based on the tailings dam operation plan information set and the current slurry level, use a time-space lag prediction model to predict the pore water pressure change information set inside the tailings dam in the future period.

[0032] The tailings dam operation plan information set can be a detailed collection of information on the daily production operation arrangements of the tailings dam, including the planned total amount of tailings to be discharged in the future, the planned sequence of discharge points, and the planned slurry concentration. The data comes from the production operation scheduling database. The current slurry level status can be information such as the actual slurry level height and the rate of change of the slurry in the tailings dam at the current moment. The pore water pressure change information set can be a sequence of data describing the changes in pore water pressure values ​​at different depths and locations of the tailings dam over a future period (such as 24 hours or 72 hours), reflecting key characteristics such as the increasing and decreasing trends of pore water pressure and the time of peak occurrence.

[0033] Specifically, the tailings dam's operational plan directly determines the input volume and distribution of the slurry. For example, a large discharge during a concentrated period can rapidly increase the dam's load, thereby altering the stress balance within the dam. The impact of this load change on pore water pressure has a time lag, requiring a spatiotemporal lag prediction model to accurately predict its future changes. The current slurry level is a real-time feedback parameter during operation. If the rate of level rise exceeds expectations, it may prematurely trigger abnormal changes in pore water pressure, forming a core factor influencing pore water pressure along with the operational plan. Pore water pressure is a key indicator for measuring tailings dam stability; excessively high pressure can reduce the dam's shear strength, increasing the risk of dam failure. Traditional methods often rely solely on historical data for simple predictions, ignoring the dynamic impact of the operational plan and real-time slurry level, and failing to consider spatiotemporal lag effects, leading to significant deviations between predictions and actual conditions, making early risk warning difficult. This step introduces the tailings dam operational plan information set and real-time slurry level as external driving factors and boundary conditions, inputting them into a pre-trained spatiotemporal lag prediction model. By acquiring the tailings dam operation plan information set and combining it with the current slurry level, the spatiotemporal lag prediction model can be used to predict the pore water pressure change information set for future periods. This can fully incorporate the lag effect of the operation schedule and the real-time feedback of the status, accurately predict the trend of pore water pressure change, and provide a timely and reliable basis for the stability assessment of the tailings dam.

[0034] S203. Based on the pore water pressure change information set, conduct dynamic assessment of tailings dam stability and output stability early warning information.

[0035] Stability warning information can be a prompt generated based on the results of dynamic stability assessment, including future safety factor change curves, potential dangerous periods, and recommended countermeasures.

[0036] Specifically, the stability of tailings dams is the core of the safe operation of tailings ponds, and changes in pore water pressure are the most direct factor causing stability fluctuations. When pore water pressure continues to rise, the effective stress between the dam particles decreases, and the shear strength declines. If it exceeds the safety threshold, it may trigger a landslide or dam failure in a short period of time. Traditional stability assessments are mostly conducted manually on a periodic basis, with long assessment cycles. They cannot track the dynamic changes in pore water pressure in real time, and the assessment process does not fully consider the lagging trend of pore water pressure changes, resulting in serious lag in early warnings. It is difficult to take timely preventive measures before an accident occurs. For example, if the pore water pressure has shown a rapid upward trend, but the periodic assessment has not been carried out, the best time for handling may be missed, causing irreparable losses. This step conducts a dynamic stability assessment of tailings dams based on the pore water pressure change information set. It can monitor every change node of pore water pressure in real time, and judge the stability status in advance based on its future change trend. Once any signs of exceeding the safe range are detected, the corresponding level of stability early warning information is immediately output and sent to the control personnel. By integrating multi-source heterogeneous tailings dam status monitoring data and operation plan information, a machine learning prediction model capable of capturing spatiotemporal lag effects was constructed, enabling accurate prediction of future pore water pressure changes. Based on this, a dynamic stability assessment was then conducted.

[0037] The method provided in this embodiment first acquires a tailings dam state characteristic dataset, and then uses machine learning to construct a spatiotemporal lag prediction model to capture the spatiotemporal lag effect of state changes. Next, it acquires a tailings dam operation plan information set and the current slurry level, and combines this with the aforementioned model to predict future changes in pore water pressure inside the tailings dam. Finally, based on this information set, and considering the dam structure and material properties, it performs a dynamic stability assessment, outputting stability warning information including warning level, area, and countermeasures, and sends it to control personnel. By accurately capturing the spatiotemporal lag effect of the tailings dam state and combining the operation plan with real-time status to achieve accurate prediction of pore water pressure, it enables dynamic assessment and timely warning of tailings dam stability. This significantly improves the real-time performance and accuracy of tailings dam safety monitoring, allowing for early detection of risks and providing management with more time to respond, effectively reducing the probability of accidents such as dam failures, and ensuring mine production safety and sustainable development.

[0038] In some embodiments, radar level gauges deployed at the clarified liquid level in the reservoir area and ultrasonic flow meters deployed at the ore discharge branch pipes are used for joint monitoring. Data fusion technology is used to eliminate monitoring errors caused by floating objects on the reservoir surface and rainfall disturbances, generating a high-confidence liquid level dynamic change sequence. Laboratory particle size and chemical composition analysis is performed by periodically collecting tailings slurry samples and surrounding soil from different ore discharge points. Based on spatial interpolation algorithms, a slurry spatial distribution model reflecting the tailings particle size distribution, permeability coefficient, and mineral chemical composition of different sedimentary areas is constructed, generating physicochemical property data of the tailings slurry. A pore pressure sensor array buried at different elevations and in different sedimentary areas of the tailings dam is used. Based on the pore pressure sensor array, regional permeability data is extracted. Sensor points that are sensitive to ore discharge operations and have significant permeability characteristics are marked as key monitoring points to ensure accurate capture of the dynamic response of water pressure along the main seepage paths within the dam, generating time-series data of pore pressure response inside the dam. Based on the liquid level dynamic change sequence, tailings slurry physicochemical property data, and pore pressure response time-series data inside the dam, a tailings dam state characteristic dataset is generated.

[0039] Radar level gauges can be level monitoring devices deployed at the clarified liquid level in the reservoir area, used to collect the liquid level height in real time; the data comes directly from the equipment monitoring results. Ultrasonic flow meters can be flow monitoring devices deployed at the ore discharge branch pipe, used to measure the slurry discharge flow rate; the data comes from the equipment's real-time monitoring. The liquid level dynamic change sequence can be a continuous data sequence reflecting the change of liquid level in the reservoir area over time, generated by data fusion from radar level gauge and ultrasonic flow meter data. Spatial interpolation algorithms can be mathematical methods that extrapolate the spatial distribution characteristics of the entire domain based on data from finite sampling points, used to construct a slurry spatial distribution model. The slurry spatial distribution model can be a spatial distribution model reflecting the tailings particle size distribution, permeability coefficient, and mineral chemical composition of different sedimentary areas, generated after processing and analyzing data using spatial interpolation algorithms. Tailings slurry physicochemical property data can be a dataset containing physicochemical properties such as tailings particle size distribution, permeability coefficient, and mineral chemical composition, derived from the output results of the slurry spatial distribution model. The pore pressure sensor array can be a monitoring network composed of multiple pore water pressure sensors buried at different elevations and in different sedimentary areas of the tailings dam; the data comes from the water pressure values ​​collected by the sensors in real time. Regional permeability data can be quantitative data reflecting the permeability of different areas of the dam body, extracted from monitoring data of the pore pressure sensor array. Key monitoring points can be sensor locations that are sensitive to ore discharge operations and have significant permeability characteristics, identified manually by combining sensor data feature markings. Pore pressure response time-series data inside the dam body can be a data sequence recording the dynamic changes of water pressure over time along the main seepage paths inside the dam body, derived from continuous monitoring of key monitoring points.

[0040] Specifically, the tailings dam state characteristic dataset is the foundation for constructing a spatiotemporal lag prediction model. Its accuracy and comprehensiveness directly determine the reliability of subsequent pore water pressure prediction and stability assessment. However, the tailings dam environment is complex, and single monitoring methods are easily affected by interference, leading to large data errors and incomplete information: In tailings dam level monitoring, surface floating objects (such as tailings particles and debris) can obscure radar signals, and rainfall can cause abnormal fluctuations in the liquid level, making it difficult to obtain the true liquid level using only radar level gauges; changes in the flow rate of the discharge branch pipe can also indirectly affect the liquid level, requiring cross-validation with flow rate data; the physicochemical properties of tailings slurry have significant spatial variation. Differences: The slurry concentration and particle size vary at different ore discharge points, and the resulting dam structure (such as permeability) also exhibits regional differences. Analyzing only a small number of sampling points cannot reflect the overall distribution and may overlook the characteristics of key depositional areas, leading to biased judgments about dam stability. In monitoring pore water pressure within the dam, although sensors are widely distributed, some points are not sensitive to ore discharge operations or are located in areas with poor permeability. Their data cannot effectively reflect pressure changes along the main seepage paths. Blindly using all sensor data will introduce redundant information, reduce model efficiency, and even mask dangerous signals. To address these issues, this step first uses radar level gauges deployed around the reservoir area to monitor the liquid level in real time, and simultaneously uses ultrasonic flow meters installed on each ore discharge branch pipe to monitor the slurry injection flow rate. After receiving this raw data, the central data acquisition system calls data fusion algorithms (such as Kalman filters) to jointly process the data, eliminating errors introduced by factors such as surface debris obstruction or rainfall disturbance, ultimately generating a high-confidence sequence of dynamic liquid level changes. Meanwhile, tailings slurry samples and surrounding soil samples are collected periodically (e.g., monthly) from different tailings discharge points and sent to the laboratory for particle size and chemical composition analysis using equipment such as laser particle size analyzers and X-ray fluorescence spectrometers to obtain accurate physicochemical parameters. Subsequently, based on the analysis results of these samples and their corresponding spatial coordinates, spatial interpolation algorithms (such as Kriging) are used to construct a slurry spatial distribution model that continuously reflects the spatial distribution of tailings particle size distribution, permeability coefficient, and chemical composition throughout the entire tailings reservoir area. Structured physicochemical property data of the tailings slurry are then extracted from this model. In addition, pore water pressure readings are continuously collected by an array of pore pressure sensors pre-embedded at different elevations and in different areas within the tailings dam. The system analyzes this data to calculate the response sensitivity (e.g., pressure change rate) of each sensor point to the tailings discharge operation, thereby assessing the permeability of the area. The points with the most sensitive response and the most significant permeability characteristics are marked as key monitoring points. Finally, the system extracts pore pressure data from these key monitoring points to form time-series data of pore pressure response within the dam.Finally, the data management module integrates and formats the above-mentioned liquid level dynamic change sequence, tailings slurry physicochemical property data, and dam internal pore pressure response time series data to generate a tailings dam state characteristic dataset for subsequent modeling and analysis.

[0041] The method provided in this embodiment effectively eliminates interference factors in liquid level monitoring by combining multi-device monitoring and data fusion, ensuring the authenticity and reliability of liquid level data. By screening key monitoring points and focusing on pore pressure changes along the core seepage path, redundant data is reduced and the effectiveness of information is improved. The final tailings dam state characteristic dataset is comprehensive and accurate, laying a high-quality data foundation for the construction of subsequent spatiotemporal lag prediction models. This helps improve the accuracy of pore water pressure prediction and the reliability of tailings dam stability assessment, providing strong support for tailings dam safety control.

[0042] In some embodiments, based on the time-series data of pore pressure response and the dynamic change sequence of liquid level inside the dam, the temporal correlation features between liquid level change events and corresponding pore pressure responses in different historical periods are extracted for each key monitoring point of the dam. The temporal correlation features are used to characterize the time dependence between liquid level changes and pore pressure responses. A hysteresis fingerprint coefficient set is generated based on the physical hysteresis fingerprint coefficient and the chemical hysteresis fingerprint coefficient. Based on the hysteresis fingerprint coefficient set, a hysteresis primitive model that can quantitatively describe the pressure response amplitude and response time caused by liquid level changes is constructed for each key monitoring point. The hysteresis primitive models of all key points are integrated to generate a spatiotemporal hysteresis prediction model.

[0043] The liquid level dynamic change sequence can be a high-confidence data sequence reflecting the change of liquid level in the reservoir over time, generated by joint monitoring of radar level gauges and ultrasonic flow meters and data fusion. The physical hysteresis fingerprint coefficient can be a quantitative coefficient describing the influence of the dam's physical properties (such as permeability coefficient and particle size distribution) on the hysteresis effect of pore pressure response, calculated based on the physicochemical properties data of tailings slurry. The chemical hysteresis fingerprint coefficient can be a quantitative coefficient describing the influence of the chemical composition of tailings slurry (such as mineral composition) on the hysteresis effect of pore pressure response, derived from the chemical composition analysis results in the physicochemical properties data of tailings slurry. The hysteresis fingerprint coefficient set can be a dataset integrating the physical and chemical hysteresis fingerprint coefficients, used to quantify the multi-dimensional impact of hysteresis effects. The hysteresis primitive model can be a model that quantitatively describes the amplitude and response time of the pore pressure response caused by liquid level changes for a single key monitoring point, constructed based on the hysteresis fingerprint coefficient set.

[0044] Specifically, the spatiotemporal lag prediction model is the core tool for achieving accurate prediction of pore water pressure in tailings dams. The scientific nature of its construction directly determines the reliability of subsequent stability assessments. The pore water pressure response of tailings dams exhibits significant spatiotemporal lag characteristics. This lag is caused by multiple factors: From a temporal perspective, the impact of liquid level changes on pore pressure is not immediate but is constrained by the permeability of the dam body—in areas with low permeability, water infiltration is slow, resulting in a long lag time for the pore pressure response; conversely, areas with high permeability respond more rapidly. For example, when the liquid level in the reservoir suddenly rises, pore pressure changes may occur on the surface of the dam body within a few hours, while the response in deeper areas may be delayed for several days. From a spatial perspective, the tailings particle size distribution and mineral composition differ in different sedimentary areas, leading to spatial heterogeneity in the amplitude and lag time of the pore pressure response. For example, the pore pressure response amplitude is smaller but the lag time is longer in fine-grained sedimentary areas, while the opposite is true in coarse-grained areas. To address the above issues, this step first involves the system reading the tailings dam state characteristic dataset to obtain the time-series data of the pore pressure response within the dam body and the dynamic change sequence of the liquid level. For each marked key monitoring point, the system employs time-series analysis methods (such as cross-correlation analysis) to analyze the correspondence between a series of historical liquid level change events (such as liquid level rise) and their pore water pressure response at that point, extracting time-series correlation features (such as maximum correlation coefficient and lag time) to quantitatively describe the time dependence between the two. Simultaneously, based on the physicochemical properties data of the tailings slurry, the system calculates the physical lag fingerprint coefficient (based on physical parameters such as average particle size and permeability coefficient) for the area where the key monitoring point is located. Chemical hysteresis fingerprint coefficients (based on chemical parameters such as clay content and specific mineral content) are generated and combined to form a set of hysteresis fingerprint coefficients corresponding to that point. Subsequently, using the temporal correlation features and the set of hysteresis fingerprint coefficients of that point as input features, and the historical pore pressure response data of that point as training labels, machine learning algorithms (such as gradient boosting decision trees) are used to train the model, resulting in a hysteresis primitive model specific to that point that can predict its future pressure response amplitude and time. Finally, the system uses ensemble techniques such as spatial interpolation or graph neural networks to integrate and couple the hysteresis primitive models of all key monitoring points to form a global spatiotemporal hysteresis prediction model that can predict the spatiotemporal evolution of pore water pressure inside the entire dam body.

[0045] The method provided in this embodiment accurately captures the time dependence between liquid level change and pore pressure response based on extracted time-series correlation features, laying the foundation for the quantification of hysteresis effects. By introducing physical and chemical hysteresis fingerprint coefficients, the influence of the dam's physical properties and chemical composition on hysteresis effects is fully considered, making the model more consistent with actual working conditions. The generated model can accurately predict pore water pressure changes in different regions and time periods, providing a reliable basis for subsequent dynamic assessment of tailings dam stability, helping to identify potential safety risks in advance and improve the safety management level of tailings dams.

[0046] In some embodiments, based on the liquid level change events and temporal correlation characteristics, the unique liquid level-pore pressure hysteresis law of each key monitoring point is determined through event alignment and causal correlation analysis, generating a physical hysteresis fingerprint coefficient that includes hysteresis time and response amplitude; based on the physicochemical properties data of tailings slurry, the catalytic or inhibitory effects of different chemical components on the pore water pressure transmission process are analyzed, generating a chemical hysteresis fingerprint coefficient that characterizes the effect of chemical reactions on hysteresis time adjustment; the physical hysteresis fingerprint coefficient and the chemical hysteresis fingerprint coefficient are combined to generate a hysteresis fingerprint coefficient set.

[0047] The level-pore pressure hysteresis law can be a quantitative conclusion describing the relationship between level changes and pore pressure responses at each key monitoring point, determined through event alignment and causal correlation analysis (e.g., "The pressure response at this point lags behind the level rise event by an average of 4.5 hours, and the pressure rise is approximately 0.8 times the level rise"). The hysteresis time can be a component of the physical hysteresis fingerprint coefficient, referring to the delay in pore water pressure response relative to the level change event. The response amplitude can be the pore pressure change caused by a unit level change, calculated within the candidate hysteresis period.

[0048] Specifically, the hysteresis fingerprint coefficient set is the core input for constructing the hysteresis primitive model. Its accuracy directly determines the ability of the spatiotemporal hysteresis prediction model to characterize the pore pressure response. The hysteresis effect of the pore pressure response in tailings dams is the result of the combined effects of physical and chemical factors. Ignoring the influence of either dimension will lead to model prediction bias. The basic law of hysteresis dominated by physical factors is that physical properties such as the dam's permeability coefficient and particle size distribution determine the water infiltration rate, which in turn affects the hysteresis time and amplitude of the pore pressure response. For example, the permeability coefficient is high in coarse-grained areas, resulting in a fast and large pore pressure response after liquid level changes; the opposite is true in fine-grained areas, where the hysteresis time is shorter. While the lag is long and small, physical properties alone cannot explain certain abnormal lag phenomena—for example, within the same physical property region, there are significant differences in pore pressure response lag time; chemical factors exacerbate the complexity of the lag: chemical components in tailings slurry (such as clay minerals swelling upon contact with water and sulfides oxidizing to produce acid) can alter pore structure and indirectly affect water conduction. For instance, in areas with high clay mineral content, pore blockage after water absorption will prolong the pore pressure response lag time; while acidic components may dissolve some minerals, shortening the lag time. These chemical effects interfere with the physical lag law, and if not quantified separately, they will mask the true lag characteristics. To address the above issues, this step first involves the system reading the dynamic change sequence of the liquid level and the time-series data of the pore pressure response within the dam body. For each key monitoring point, the system scans the liquid level data to identify significant liquid level change events (such as continuous rises exceeding a threshold). For each such event, the system searches for the corresponding response event within the pore pressure data sequence. Next, event alignment and causal correlation analysis are performed to precisely align the liquid level event and the pressure response event on the time axis. Statistical methods are then used to verify whether the former significantly causes the latter. Through this analysis, the unique liquid level-pore pressure hysteresis pattern at that point is determined (such as average hysteresis time, response time, etc.). The system calculates the physical hysteresis fingerprint coefficient for the point by comparing the amplitude ratios. Simultaneously, it queries the physicochemical properties of the tailings slurry corresponding to the key point to obtain its chemical composition information (such as montmorillonite content). Then, based on a pre-set expert knowledge base or chemical-mechanical coupling model, it analyzes the catalytic or inhibitory effects of these chemical components on the pore water pressure transmission process (e.g., for every 1% increase in montmorillonite content, the hysteresis time increases by 0.2 hours), and quantifies and generates the chemical hysteresis fingerprint coefficient. Finally, the system combines the physical and chemical hysteresis fingerprint coefficients for this point to generate a complete set of hysteresis fingerprint coefficients for that point, including the baseline value and adjustment amounts.

[0049] The physical hysteresis fingerprint coefficients generated by event alignment and causal analysis, as provided in this embodiment, accurately capture the fundamental influence of the dam's physical properties on pore pressure response, providing a quantitative benchmark for hysteresis effects. Combined with chemical hysteresis fingerprint coefficients obtained from chemical composition analysis, this supplements the correction effect of chemical reactions on hysteresis patterns, solving the problem of hysteresis anomalies under complex geological conditions. The hysteresis fingerprint coefficient set comprehensively covers the key factors affecting pore pressure response, enabling the subsequently constructed hysteresis primitive model to more realistically reflect spatiotemporal hysteresis characteristics, thereby improving the accuracy of pore water pressure prediction and providing a more reliable basis for tailings dam stability assessment.

[0050] In some embodiments, based on the liquid level change event at each key monitoring point, the corresponding pore pressure response curve is extracted. The liquid level change event and the pore pressure response curve are aligned using a dynamic time warping method to generate a co-change segment. Within the aligned co-change segment, a time-domain causal analysis method is used to determine the correlation strength between liquid level change and pore pressure change at different lag times, and multiple candidate lag periods with correlation strength higher than a preset strength threshold are selected. For each candidate lag period, the pore pressure change caused by a unit liquid level change within this period is determined as the response amplitude, and the statistical significance of this period's recurrence in historical events is used for verification. Candidate periods with statistical significance lower than a preset level are eliminated. The retained lag periods and their corresponding response amplitudes are integrated to generate a physical lag fingerprint coefficient containing lag time and response amplitude.

[0051] The pore pressure response curve can be a curve describing the change of pore water pressure at key monitoring points over time, generated from time-series data of pore pressure response within the dam body. The coordinated change segment can be a time period, after dynamic time warping, where the liquid level change event and the pore pressure response curve show a synchronous trend. The time-domain causal analysis method can be a statistical method that analyzes the causal relationship between variables in the time dimension, used to quantify the correlation strength between liquid level changes and pore pressure changes. The candidate lag period can be a lag time interval where the correlation strength is higher than a preset strength threshold.

[0052] Specifically, the physical lag fingerprint coefficient is a core indicator for quantifying the impact of dam physical properties on pore pressure response. Its accurate extraction directly determines the reliability of the lag primitive model. The time correlation between liquid level change and pore pressure response has complex characteristics that traditional methods struggle to capture accurately: Time axis misalignment problem: The time scales of liquid level change events and pore pressure response curves often do not match. For example, the liquid level may rise rapidly within 1 hour, while the pore pressure response may be scattered over 3-6 hours. Direct comparison can lead to distortion in the correlation analysis. Correlation interference problem: Pore pressure changes may be affected by other factors (such as seepage from adjacent areas), resulting in "pseudo-correlation" periods unrelated to liquid level changes. If these are mistakenly included in the analysis, it can lead to misjudgment of lag patterns. Statistical reliability problem: Some lag periods may only appear in individual historical events and do not have universal patterns. If directly used as fingerprint coefficients, it will reduce the model's generalization ability. To address the above issues, this step first involves the system reading historical data from a key monitoring point. For each historical liquid level change event at that point, the system extracts the corresponding time window data from its pore pressure response time series data, forming the pore pressure response curve for that event. Next, using dynamic time warping, the system performs nonlinear matching between the data sequence of the liquid level change event and the pore pressure response curve to achieve precise event alignment and identify the most synchronized co-change segment. Within this aligned co-change segment, the system employs time-domain causal analysis to calculate the correlation strength between liquid level changes and pore pressure changes under different assumed lag times. Then, based on… A preset intensity threshold is used to filter out all candidate lag periods with correlation strength higher than the threshold. Then, the system backtracks historical data to examine the frequency of each candidate lag period in multiple independent events, evaluates its statistical significance, and removes candidate periods with low statistical significance based on preset levels. For each remaining candidate lag period, the system calculates the average pore pressure change corresponding to the unit liquid level change within that period and determines it as the response amplitude. Finally, the system packages all these verified lag periods and their corresponding response amplitude parameters to generate a unique physical lag fingerprint coefficient for the key monitoring point, containing both lag time and response amplitude.

[0053] The method provided in this embodiment achieves precise alignment between liquid level changes and pore pressure responses, solving the correlation distortion problem caused by time axis misalignment; time-domain causal analysis effectively filters out highly correlated lag periods, eliminating the influence of interference factors; the finally generated physical lag fingerprint coefficient can realistically and stably reflect the liquid level-pore pressure lag law of key monitoring points, providing a reliable physical characteristic quantification basis for the subsequent construction of lag element models, which helps to improve the accuracy of pore water pressure prediction.

[0054] In some embodiments, based on the physicochemical properties of the tailings slurry, the mineral chemical composition and ion concentration of the tailings slurry sample in the sedimentation area where each key monitoring point is located are extracted; the enhancing or weakening effect of specific chemical components on water flow resistance during pore water pressure transmission is analyzed, and key chemical factors that have an adjusting effect on lag time are identified; the statistical relationship between the concentration changes of key chemical factors in real time and the corresponding pore pressure response lag time in historical data is obtained, the correction strength of each chemical factor on lag time is quantified, and a chemical lag fingerprint coefficient is generated.

[0055] Key chemical factors can be chemical components that significantly adjust the lag time of pore water pressure transmission, and can be determined by analyzing the correlation between chemical components and lag time.

[0056] Specifically, the chemical hysteresis fingerprint coefficient is a core indicator for quantifying the influence of chemical action on the hysteresis effect of pore pressure response. Its accurate extraction is crucial for perfecting the hysteresis law. The physical hysteresis fingerprint coefficient only reflects the influence of the physical properties of the dam body, while chemical components directly interfere with water flow conduction by changing the pore structure. If this factor is ignored, it will lead to a distortion in the characterization of the hysteresis law. The influence of chemical components on water flow resistance is complex: clay minerals expand when they come into contact with water and fill the pores, increasing water flow resistance and prolonging the pore pressure response hysteresis time; while acidic ions may dissolve mineral particles, expand pore channels, and shorten the hysteresis time. For example, in a certain sedimentary area containing a high proportion of montmorillonite (clay mineral), the pore pressure response hysteresis time is 30% longer than in areas with the same physical properties after a change in liquid level. This difference cannot be explained by physical hysteresis fingerprint alone. The influence of key chemical factors is specific: different chemical components have different directions and intensities of action. For example, calcium ions may promote particle aggregation and reduce permeability, while sulfate ions may crystallize and block pores, exacerbating the hysteresis. If all chemical components are analyzed together, the independent effects of key factors will be masked, leading to correction bias. To address the above issues, this step first involves the system reading the physicochemical properties of the tailings slurry. For tailings slurry samples from the deposition area of ​​each key monitoring point, the system extracts the mineral chemical composition (e.g., through XRD data analysis) and ion concentration (e.g., by reading IC analysis results). Next, the system analyzes how these chemical components enhance or weaken water flow resistance during pore water pressure transmission by altering fluid viscosity, particle surface charge, and double-layer thickness. Based on this mechanism analysis and statistical correlation analysis of historical data, the system identifies a few key factors that have the most significant effect on hysteresis time adjustment. The system identifies key chemical factors (such as "montmorillonite content" and "pore water sodium adsorption ratio SAR"). Then, it retrieves records from a historical database, which contain the concentration changes of these key chemical factors at different historical time points and the actual observed pore pressure response lag time at that time. Through statistical methods such as regression analysis, the system quantifies the correction strength of each chemical factor's unit concentration change on the lag time (e.g., "for every 1 unit increase in SAR, the lag time decreases by 0.15 hours"). Finally, the system combines all key chemical factors and their corresponding correction strengths to generate the chemical lag fingerprint coefficient corresponding to that monitoring point.

[0057] The method provided in this embodiment extracts chemical composition information of the area where key monitoring points are located, accurately identifying the chemical factors affecting the lag time of pore pressure response. By identifying key chemical factors, interference from irrelevant chemical components is eliminated, focusing on the core influencing factors. By quantifying the correction intensity, the adjustment effect of chemical action on lag time is scientifically characterized. The generated chemical lag fingerprint coefficient compensates for the limitation of physical lag fingerprints that only consider physical characteristics, making the characterization of lag laws more comprehensive, thereby improving the accuracy of the spatiotemporal lag prediction model and providing a more reliable basis for tailings dam pore water pressure prediction and stability assessment.

[0058] In some embodiments, the tailings dam operation plan information set includes the planned total discharge volume, planned discharge point location sequence, and planned slurry concentration for future periods. Based on the planned discharge volume and planned slurry concentration, combined with the current slurry level, the spatiotemporal change process of the dam level in the future period is deduced through material balance and rheological characteristic analysis, generating a level prediction sequence. Based on the planned discharge point location sequence, the corresponding sedimentation area is identified, and the physicochemical properties data of the tailings slurry corresponding to this area are extracted according to the slurry spatial distribution model. Based on the physicochemical properties data of the tailings slurry, the physical influence factors of future discharge behavior on the seepage path of the dam body and the chemical influence factors on the lag response of pore water pressure are determined. The level prediction sequence, physical influence factors, and chemical influence factors are input into the spatiotemporal lag prediction model, and the dynamic response process of pore water pressure in different periods and different dam areas is simulated through the liquid level change events and lag fingerprint coefficient set, generating a pore water pressure change information set expressed in the form of a spatiotemporal grid.

[0059] The planned total tailings discharge can be the total weight or volume of tailings slurry planned for discharge in the future, derived from the tailings dam operation plan information set. The planned discharge point sequence can be the spatial order of tailings slurry discharge points in the future, such as the coordinates of discharge outlets in different areas, determined by the tailings dam operation plan information set. The planned slurry concentration can be the ratio of solid particles to liquid in the future discharged tailings slurry, derived from the tailings dam operation plan information set. Material balance analysis can be a method of extrapolating liquid level changes by calculating the relationship between the planned discharge volume and the existing capacity of the dam area, based on the planned total discharge volume and the current slurry liquid level state. Rheological property analysis can be an analytical method for studying the impact of tailings slurry flow characteristics (such as viscosity and fluidity) on liquid level distribution, combined with the planned slurry concentration. The liquid level extrapolation sequence can be a sequence of data predicting the changes in the dam area liquid level over time and space in the future, generated by material balance and rheological property analysis. The physical influence factor can be the amount by which future discharge behavior changes the physical properties (such as permeability coefficient and pore structure) of the seepage path in the dam body, determined based on the physicochemical property data of the tailings slurry. The chemical influence factor can be the adjustment amount of the hysteretic response of pore water pressure caused by changes in chemical composition due to future ore dumping behavior, which is derived from the chemical composition in the physicochemical property data of tailings slurry.

[0060] Specifically, predicting future changes in pore water pressure is a prerequisite for tailings dam stability assessment, and its accuracy directly affects the effectiveness of safety warnings. Traditional prediction methods rely solely on historical data or static models, making it difficult to cope with the dynamic changes brought about by ore discharge operations, and thus have significant limitations: The impact of the operation plan on the liquid level has spatiotemporal complexity: the planned total amount of ore discharged determines the overall magnitude of liquid level changes, the planned location of the ore discharge point determines the spatial distribution of liquid level changes (e.g., concentrated ore discharge in a certain area will lead to a sudden rise in local liquid level), and the planned slurry concentration affects the tailings accumulation rate (high-concentration slurry deposits faster, and the liquid level rises more rapidly). If these factors are ignored, the liquid level prediction will deviate significantly from the actual level, leading to distorted pore pressure predictions; Ore discharge behavior changes the seepage characteristics of the dam body: newly discharged tailings will change the physical structure (e.g., particle size distribution, permeability coefficient) and chemical composition (e.g., the introduction of new mineral components) of the original deposition area, thereby changing the seepage path and pore pressure lag law. For example, continuous ore discharge into fine-particle areas may reduce their permeability and prolong the pore pressure response lag time; while slurry containing acidic components may dissolve the original minerals and shorten the lag time. To address the above issues, this step first involves the system reading the tailings dam operation plan information set to obtain the planned total discharge volume, planned discharge point location sequence, and planned slurry concentration. Simultaneously, the system acquires the current slurry level status (real-time monitoring value). Next, based on the planned discharge volume and planned slurry concentration, combined with the current slurry level status, the system calculates the total slurry increment using the material balance principle and analyzes the diffusion and deposition process of the slurry after it enters the dam, considering rheological characteristics. This allows for the prediction of the spatiotemporal changes in the dam's slurry level over future periods, generating a slurry level prediction sequence. Simultaneously, the system analyzes the planned discharge point location sequence and identifies the corresponding deposition areas in the geographic information system. Finally, the system queries the slurry spatial distribution model to extract these target deposition areas. The system uses data on the physicochemical properties of the tailings slurry corresponding to the accumulation area (including physical properties such as expected permeability coefficient and chemical properties such as mineral composition). Based on this data, the system determines the physical impact factors (such as estimated loading amount and permeability change) and chemical impact factors (i.e., the expected change in the chemical lag fingerprint coefficient corresponding to the new area) that this ore discharge operation will have on the dam body. Finally, the system inputs the liquid level projection sequence, physical impact factors, and chemical impact factors into the pre-constructed spatiotemporal lag prediction model. This model uses its built-in lag fingerprint coefficient set to simulate the response of each dam unit to future liquid level change events and changes in the physicochemical environment, and finally generates a set of pore water pressure change information covering the entire dam body in the form of a spatiotemporal grid.

[0061] By integrating tailings dam operation plan information through the method provided in this embodiment, the spatiotemporal changes of future liquid levels are accurately predicted, avoiding the disconnect between static prediction and actual ore discharge behavior. By identifying physical and chemical influencing factors, the changes in seepage characteristics and hysteresis patterns of the dam body caused by new ore discharge are fully considered, making the pore pressure prediction more consistent with dynamic operating conditions. With the help of the spatiotemporal hysteresis prediction model, a refined simulation of the pore pressure response in different time periods and regions is achieved. The final pore water pressure change information set can truly reflect the future pressure dynamics inside the dam body, providing a reliable basis for subsequent stability assessment and helping to identify potential risks in advance.

[0062] In some embodiments, for the sedimentary area corresponding to the planned ore dumping location sequence, the mineral chemical composition is extracted from the physicochemical property data of the tailings slurry. Chemical components whose catalytic or inhibitory effects on the pore water pressure transmission process exceed a preset chemical threshold are identified as key chemical components, and their concentrations are obtained. Based on the real-time reaction kinetic parameters of the key chemical components in the pore water environment, the rate of change of their concentration on the pore medium structure is calculated, thereby deriving the potential impact on the pore water pressure lag time. Based on the chemical lag fingerprint coefficient, the potential impact is calibrated and normalized, quantifying the potential impact into specific chemical correction adjustment values ​​for the lag time. When the chemical correction adjustment value is positive, it indicates that the chemical component has a catalytic effect on the pore water pressure response; when the chemical correction adjustment value is negative, it indicates that the chemical component has an inhibitory effect on the pore water pressure response. The absolute value of the chemical correction adjustment value indicates the strength of the effect. Based on the positive and negative attributes of the chemical correction adjustment value, a chemical influence factor is generated.

[0063] Real-time reaction kinetic parameters can be parameters describing the reaction rate of key chemical components in a pore water environment, such as the reaction rate constant, obtained through laboratory measurements or fitting historical data. Chemical correction values ​​are specific adjustments to the pore water pressure lag time of key chemical components after calibration and normalization; positive values ​​indicate catalysis, negative values ​​indicate inhibition, and absolute values ​​indicate the intensity of the effect. Catalysis can be represented by a negative chemical correction value, meaning that the chemical component will accelerate the conduction of pore water pressure and shorten the lag time. Inhibition can be represented by a positive chemical correction value, meaning that the chemical component will delay the conduction of pore water pressure and prolong the lag time.

[0064] Specifically, chemical influencing factors are key parameters for accurately predicting the lag response of pore water pressure. Their scientific determination is directly related to the predictive model's ability to capture chemical effects. Chemical components in tailings slurry significantly affect the lag behavior of pore water pressure by altering pore structure and water flow resistance. Ignoring their dynamic changes can lead to prediction bias. The influence of key chemical components is concentration-dependent: the effect of the same chemical component on lag time varies significantly at different concentrations. For example, low concentrations of clay minerals may slightly prolong the lag time, while high concentrations may double the lag time due to pore blockage. If calculated using only fixed values, the lag effect at high concentrations will be underestimated. Ore discharge operations cause changes in the spatiotemporal distribution of chemical components: different planned ore discharge points will cause newly discharged tailings slurry to accumulate in specific areas, changing the concentration of chemical components in those areas (e.g., high-sulfur ore discharge at a certain ore discharge point will increase the local sulfate ion concentration). This change will dynamically adjust the lag response of pore water pressure. If historical chemical data is used, it will be impossible to reflect future new influences. To address the above issues, this step first involves the system extracting the complete mineral chemical composition of the tailings slurry from the physicochemical properties data of the future sedimentary areas corresponding to the planned ore dumping location sequence. Then, the system compares the known effect intensity of each chemical component with preset chemical thresholds to identify all key chemical components with significant effects and obtains their predicted concentrations in the new slurry. Next, the system retrieves the real-time reaction kinetic parameters of these key chemical components from its built-in knowledge base. Based on these parameters and concentration values, the system uses a built-in chemical-hydraulic coupling model to calculate the rate of change of the component concentration on the pore medium structure (manifested as the rate of change of parameters such as permeability), and then derives its effect on the pore water pressure in the area. The system first calculates the potential impact of the force lag time (a preliminary, uncalibrated estimate). Then, it calls a pre-trained chemical lag fingerprint coefficient for the monitoring points corresponding to the prediction area, uses the calibration information contained within it to calibrate the calculated potential impact, and then eliminates dimensional differences through normalization. At this point, the potential impact is precisely quantified into a specific chemical correction value. The system determines the positive or negative attribute of this value: negative indicates a catalytic effect; positive indicates an inhibitory effect, and the absolute value represents the strength of the effect. Finally, based on the positive or negative attribute and numerical value of this chemical correction value, the system packages and generates the final chemical impact factor that can be directly used by the prediction model.

[0065] By introducing reaction kinetic parameters through the method provided in this embodiment, the dynamic timeliness of the effects of chemical components is captured, overcoming the limitations of static parameters; the action mechanism of key chemical components is distinguished, making the quantification of chemical influence more accurate; the generated chemical influence factor can truly reflect the dynamic adjustment of the lag response of chemical action to pore water pressure during future ore discharge, providing reliable input for the spatiotemporal lag prediction model, further improving the accuracy of pore water pressure change prediction, and laying a more scientific foundation for tailings dam stability assessment.

[0066] In some embodiments, based on the pore water pressure change information set and the slurry spatial distribution model, the adverse effects of future increases in pore water pressure on the stability of the dam structure are analyzed, the overall safety status of each future period is dynamically extrapolated, and a future safety factor change curve is generated; the future safety factor change curve is analyzed to identify potential dangerous periods when the safety factor is below the safety threshold or the rate of decrease exceeds the critical rate; and stability early warning information is generated based on the future safety factor change curve and the potential dangerous periods.

[0067] The future safety factor change curve can be a curve describing the overall safety factor of the tailings dam changing over time in various future periods, generated through dynamic extrapolation of the dam's safety status. Potentially hazardous periods can be future periods where the safety factor is below the safety threshold or the rate of decline exceeds the critical rate, identified from the future safety factor change curve.

[0068] Specifically, dynamic assessment of tailings dam stability is a core element in preventing dam failure accidents. Its scientific validity is directly related to the safe operation of tailings ponds. Traditional assessment methods have defects such as staticity and lag, making it difficult to cope with complex dynamic changes. Manual inspections or periodic monitoring have long cycles (such as once a month), making it difficult to track rapid changes in pore water pressure in real time. For example, after a certain ore discharge operation, the pore water pressure in a local area may rise sharply within 24 hours, but the next assessment may not be conducted until one month later, which may miss the best time for disposal. To address the above issues, this step first involves the system reading a set of pore water pressure variation information and a slurry spatial distribution model. The pore water pressure variation information set provides predicted pore water pressure values ​​for each future time point and for each three-dimensional grid cell within the dam body. The slurry spatial distribution model provides soil shear strength parameters (such as cohesion and internal friction angle) for each grid cell. Next, based on the principle of limit equilibrium methods (such as the Bishop method and Janbu method), the system analyzes how these future pore water pressures (as forces) affect the stability of the dam body. The system dynamically extrapolates the overall safety state of the dam body at each future calculation point, i.e., calculates the overall safety factor at each time point, thereby generating a comprehensive safety curve. The system first generates a future safety factor change curve for a predicted time range. Then, it analyzes this curve, comparing the safety factor value at each point with a preset safety threshold. Simultaneously, it calculates the instantaneous rate of decline of the curve at each time point and compares it with a preset critical rate. Through this scanning analysis, the system identifies all periods where the safety factor is below the threshold and all periods where the rate of decline exceeds the critical rate. These periods are marked as potentially dangerous periods. Finally, based on the complete future safety factor change curve and the identified potentially dangerous periods, the system generates a stability warning message containing detailed information such as specific risk descriptions, risk occurrence times, risk durations, and risk levels, and outputs it to the monitoring platform.

[0069] By combining the pore water pressure change information set with the slurry spatial distribution model provided in this embodiment, dynamic assessment of dam stability is achieved, overcoming the limitations of traditional static assessment. The future safety factor change curve intuitively reflects the evolution trend of stability, making potential risks predictable. The accurately identified potential danger periods provide management departments with sufficient time to respond. The graded early warning and targeted measures suggestions improve the effectiveness of emergency response.

[0070] Figure 3 This application provides a schematic diagram of the structure of a dynamic geological modeling system based on adaptive updates of multi-source heterogeneous data, as shown in one embodiment. Figure 3As shown in this embodiment, an adaptive updating dynamic geological modeling system 300 based on multi-source heterogeneous data includes: a model building module 301, an information extraction module 302, and a strategy generation module 303. The model building module 301 is used to acquire a tailings dam state feature dataset and, based on the tailings dam state feature dataset, construct a spatiotemporal lag prediction model using machine learning methods. The information extraction module 302 is used to acquire a tailings dam operation plan information set and, based on the tailings dam operation plan information set and the current slurry level, predict a future period of pore water pressure change information set inside the tailings dam using the spatiotemporal lag prediction model. The strategy generation module 303 is used to perform dynamic stability assessment of the tailings dam based on the pore water pressure change information set, generate a tailings dam control strategy, and output a stability early warning log.

[0071] Optionally, when the model building module 301 acquires the tailings dam state characteristic dataset, it is specifically used for: joint monitoring using radar level gauges deployed at the clarified liquid level in the dam area and ultrasonic flow meters deployed at the ore discharge branch pipes; and using data fusion technology to eliminate monitoring errors caused by floating debris and rainfall disturbances on the dam surface, generating a high-confidence liquid level dynamic change sequence; and periodically collecting tailings slurry samples and surrounding soil from different ore discharge points for laboratory particle size and chemical composition analysis, and constructing a model reflecting the tailings particle size distribution, permeability coefficient, and mineral chemical composition of different sedimentary areas based on spatial interpolation algorithms. A spatial distribution model of the tailings slurry is generated to produce physicochemical property data of the tailings slurry. A pore pressure sensor array is installed at different elevations and in different depositional areas within the tailings dam. Based on this array, regional permeability data is extracted. Sensor locations sensitive to ore discharge operations and exhibiting significant permeability characteristics are marked as key monitoring points to ensure accurate capture of the dynamic response of water pressure along the main seepage paths within the dam, generating time-series data of the pore pressure response inside the dam. Based on the dynamic change sequence of the liquid level, the physicochemical property data of the tailings slurry, and the time-series data of the pore pressure response inside the dam, a tailings dam state characteristic dataset is generated. Based on the time-series data of pore pressure response inside the dam body and the dynamic change sequence of liquid level, for each key monitoring point of the dam body, the time-series correlation features between liquid level change events and corresponding pore pressure responses in different historical periods are extracted; the time-series correlation features are used to characterize the time dependency between liquid level changes and pore pressure responses; a hysteresis fingerprint coefficient set is generated based on the physical hysteresis fingerprint coefficient and the chemical hysteresis fingerprint coefficient; based on the hysteresis fingerprint coefficient set, a hysteresis primitive model is constructed for each key monitoring point that can quantitatively describe the magnitude and time of the pressure response caused by liquid level changes; the hysteresis primitive models of all key points are integrated to generate the spatiotemporal hysteresis prediction model.

[0072] Optionally, when the model building module 301 constructs a spatiotemporal lag prediction model based on the tailings dam state feature dataset using machine learning methods, it specifically performs the following: Based on the time-series data of the pore pressure response inside the dam and the dynamic change sequence of the liquid level, it extracts the temporal correlation features between the liquid level change events and the corresponding pore pressure response at each key monitoring point of the dam in different historical periods; the temporal correlation features are used to characterize the temporal dependency between liquid level change and pore pressure response; it generates a lag fingerprint coefficient set based on the physical and chemical lag fingerprint coefficients; based on the lag fingerprint coefficient set, it constructs a lag primitive model for each key monitoring point that can quantitatively describe the pressure response amplitude and response time caused by the liquid level change; and it integrates the lag primitive models of all key points to generate the spatiotemporal lag prediction model.

[0073] Optionally, when generating a hysteresis fingerprint coefficient set based on the physical hysteresis fingerprint coefficient and the chemical hysteresis fingerprint coefficient, the model building module 301 is specifically used for: determining the unique liquid level-pore pressure hysteresis law of each key monitoring point through event alignment and causal correlation analysis based on the liquid level change event and the time-series correlation characteristics, and generating the physical hysteresis fingerprint coefficient containing hysteresis time and response amplitude; analyzing the catalytic or inhibitory effects of different chemical components on the pore water pressure transmission process based on the physicochemical property data of the tailings slurry, and generating the chemical hysteresis fingerprint coefficient characterizing the effect of chemical reaction on the hysteresis time adjustment; and combining the physical hysteresis fingerprint coefficient and the chemical hysteresis fingerprint coefficient to generate the hysteresis fingerprint coefficient set.

[0074] Optionally, when the model building module 301 determines the unique liquid level-pore pressure hysteresis pattern of each key monitoring point based on the liquid level change event and the temporal correlation characteristics, and generates a physical hysteresis fingerprint coefficient containing hysteresis time and response amplitude, it is specifically used to: extract the corresponding pore pressure response curve based on the liquid level change event of each key monitoring point; align the liquid level change event and the pore pressure response curve using a dynamic time warping method to generate a coordinated change segment; and within the aligned coordinated change segment, use... A time-domain causal analysis method is used to determine the correlation strength between liquid level changes and pore pressure changes under different lag times, and to screen out multiple candidate lag periods with a correlation strength higher than a preset strength threshold. For each candidate lag period, the amount of pore pressure change caused by a unit liquid level change within this period is determined as the response amplitude, and the statistical significance of this period's recurrence in historical events is used for verification. Candidate periods with statistical significance lower than a preset level are eliminated. The retained lag periods and their corresponding response amplitudes are integrated to generate the physical lag fingerprint coefficient, which includes the lag time and response amplitude.

[0075] Optionally, when the model building module 301 analyzes the catalytic or inhibitory effects of different chemical components on the pore water pressure transmission process based on the physicochemical properties data of the tailings slurry and generates a chemical hysteresis fingerprint coefficient that adjusts the hysteresis time, it is specifically used to: extract the mineral chemical composition and ion concentration of the tailings slurry sample from the sedimentation area where each key monitoring point is located, based on the physicochemical properties data of the tailings slurry; analyze the enhancing or weakening effect of specific chemical components on water flow resistance during the pore water pressure transmission process, identify key chemical factors that adjust the hysteresis time; obtain the statistical relationship between the concentration changes of key chemical factors in real time and the corresponding pore pressure response to the hysteresis time in historical data, quantify the correction strength of each chemical factor on the hysteresis time, and generate the chemical hysteresis fingerprint coefficient.

[0076] Optionally, when the information extraction module 302 predicts the pore water pressure change information set inside the tailings dam in the future period based on the acquired tailings dam operation plan information set and the current slurry level, using the spatiotemporal lag prediction model, the following is specifically used: the tailings dam operation plan information set includes the planned total ore discharge, the planned ore discharge point location sequence, and the planned slurry concentration for the future period; based on the planned total ore discharge and the planned slurry concentration, combined with the current slurry level, the spatiotemporal change process of the slurry level in the future period is deduced through material balance and rheological characteristic analysis, and a slurry level prediction sequence is generated; based on the planned ore discharge... The point location sequence is used to identify the corresponding sedimentation area, and the physicochemical properties of the tailings slurry corresponding to this area are extracted according to the slurry spatial distribution model. Based on the physicochemical properties of the tailings slurry, the physical influence factors of future ore discharge behavior on the seepage path of the dam body and the chemical influence factors on the lag response of pore water pressure are determined. The liquid level projection sequence, the physical influence factors and the chemical influence factors are input into the spatiotemporal lag prediction model. Through the liquid level change events and the lag fingerprint coefficient set, the dynamic response process of pore water pressure in different time periods and different dam areas is simulated, and the pore water pressure change information set expressed in the form of a spatiotemporal grid is generated.

[0077] Optionally, when determining the chemical influencing factors on the hysteretic response to future pore water pressure based on the physicochemical properties data of the tailings slurry, the information extraction module 302 is specifically used for: extracting the mineral chemical composition of the tailings slurry from the physicochemical properties data of the sedimentary area corresponding to the planned ore dumping point location sequence; identifying chemical components whose catalytic or inhibitory effects on the pore water pressure transmission process exceed a preset chemical threshold as key chemical components and obtaining their concentrations; calculating the rate of change of the concentration of the key chemical components on the pore medium structure based on the real-time reaction kinetic parameters of the key chemical components in the pore water environment, and then deriving... The potential impact on pore water pressure lag time is determined; based on the chemical lag fingerprint coefficient, the potential impact is calibrated and normalized to quantify it into a specific chemical correction adjustment value for the lag time; when the chemical correction adjustment value is positive, it indicates that the chemical component has a catalytic effect on the pore water pressure response; when the chemical correction adjustment value is negative, it indicates that the chemical component has an inhibitory effect on the pore water pressure response; the absolute value of the chemical correction adjustment value indicates the strength of the effect; based on the positive or negative attribute of the chemical correction adjustment value, the chemical influence factor is generated.

[0078] Optionally, the strategy generation module 303 is specifically used to: analyze the adverse effects of future increases in pore water pressure on the stability of the dam structure based on the pore water pressure change information set and the slurry spatial distribution model, dynamically extrapolate the overall safety status for each future period, and generate a future safety factor change curve; analyze the future safety factor change curve to identify potential dangerous periods where the safety factor is below the safety threshold or the rate of decrease exceeds the critical rate; and generate the stability warning information based on the future safety factor change curve and the potential dangerous periods.

[0079] The system in this embodiment can be used to execute the methods of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

Claims

1. A dynamic geological modeling method based on adaptive updating of multi-source heterogeneous data, characterized in that, include: Obtain a tailings dam state feature dataset, and construct a spatiotemporal lag prediction model based on the tailings dam state feature dataset using machine learning methods; Obtain a set of tailings dam operation plan information; based on the set of tailings dam operation plan information and the current slurry level, use the spatiotemporal lag prediction model to predict the set of pore water pressure change information inside the tailings dam in the future period. Based on the pore water pressure change information set, a dynamic assessment of tailings dam stability is performed, and stability early warning information is output. The step of constructing a spatiotemporal lag prediction model based on the tailings dam state feature dataset using machine learning methods includes: Based on the time series data of pore pressure response and liquid level dynamic change sequence inside the dam, the time series correlation features between liquid level change events and corresponding pore pressure response in different historical periods are extracted for each key monitoring point of the dam. The time-series correlation feature is used to characterize the time dependency between liquid level change and pore pressure response; Generate a set of hysteresis fingerprint coefficients based on the physical and chemical hysteresis fingerprint coefficients; Based on the set of hysteresis fingerprint coefficients, a hysteresis primitive model is constructed for each key monitoring point to quantitatively describe the pressure response amplitude and response time caused by liquid level changes. The spatiotemporal lag prediction model is generated by integrating the lag primitive models of all key points.

2. The method according to claim 1, characterized in that, The process of obtaining the tailings dam status feature dataset includes: By jointly monitoring the liquid level using radar level gauges deployed at the clarification level in the reservoir area and ultrasonic flow meters deployed at the ore discharge branch pipes, and by using data fusion technology to eliminate monitoring errors caused by floating objects on the reservoir surface and rainfall disturbances, a high-confidence liquid level dynamic change sequence is generated. Laboratory particle size and chemical composition analysis were conducted by periodically collecting tailings slurry samples and surrounding soil from different ore discharge points. Based on spatial interpolation algorithms, a spatial distribution model of tailings slurry reflecting the particle size distribution, permeability coefficient, and mineral chemical composition of tailings in different sedimentary areas was constructed, generating physicochemical property data of tailings slurry. By burying pore pressure sensor arrays at different elevations and in different sedimentation areas of the tailings dam, regional permeability data is extracted based on the pore pressure sensor arrays. Sensor points that are sensitive to ore discharge operations and have significant permeability characteristics are marked as key monitoring points to ensure accurate capture of the dynamic response of water pressure on the main seepage path in the dam body and generate time series data of pore pressure response inside the dam body. The tailings dam state feature dataset is generated based on the liquid level dynamic change sequence, the tailings slurry physicochemical property data, and the dam body internal pore pressure response time series data.

3. The method according to claim 2, characterized in that, The step of generating a set of hysteresis fingerprint coefficients based on physical and chemical hysteresis fingerprint coefficients includes: Based on the liquid level change events and the time-series correlation characteristics, the unique liquid level-orifice pressure hysteresis pattern of each key monitoring point is determined through event alignment and causal correlation analysis, and the physical hysteresis fingerprint coefficient containing hysteresis time and response amplitude is generated. Based on the physicochemical properties data of the tailings slurry, the catalytic or inhibitory effects of different chemical components on the pore water pressure transmission process are analyzed, and the chemical hysteresis fingerprint coefficient, which characterizes the effect of chemical reaction on the hysteresis time adjustment, is generated. The physical hysteresis fingerprint coefficients are combined with the chemical hysteresis fingerprint coefficients to generate the hysteresis fingerprint coefficient set.

4. The method according to claim 3, characterized in that, Based on the liquid level change events and the temporal correlation characteristics, the unique liquid level-orifice pressure hysteresis pattern of each key monitoring point is determined through event alignment and causal correlation analysis, generating a physical hysteresis fingerprint coefficient containing hysteresis time and response amplitude, including: Based on the liquid level change event at each key monitoring point, the corresponding pore pressure response curve is extracted. The liquid level change event and the pore pressure response curve are aligned using a dynamic time warping method to generate a coordinated change segment. Within the aligned coordinated change segment, a time-domain causal analysis method is used to determine the correlation strength between liquid level change and pore pressure change under different lag times, and to screen out multiple candidate lag periods with a correlation strength higher than a preset strength threshold. For each candidate lag period, the change in pore pressure caused by a unit change in liquid level within this period is determined as the response amplitude, and the statistical significance of this period recurring in historical events is used for verification. Candidate periods with statistical significance lower than a preset level are eliminated. The retained lag time periods and their corresponding response amplitudes are integrated to generate the physical lag fingerprint coefficients, which include the lag time and response amplitude.

5. The method according to claim 4, characterized in that, The step of analyzing the catalytic or inhibitory effects of different chemical components on the pore water pressure transmission process based on the physicochemical properties data of the tailings slurry, and generating the chemical hysteresis fingerprint coefficient characterizing the effect of chemical reactions on the hysteresis time adjustment, includes: Based on the physicochemical properties data of the tailings slurry, the mineral chemical composition and ion concentration of the tailings slurry sample in the sedimentation area where each key monitoring point is located are extracted. The effects of specific chemical components on the flow resistance during pore water pressure transmission were analyzed to identify key chemical factors that adjust the lag time. The statistical relationship between the concentration changes of key chemical factors in real time and the corresponding lag time of pore pressure response is obtained from historical data. The correction strength of each chemical factor on the lag time is quantified, and the chemical lag fingerprint coefficient is generated.

6. The method according to claim 5, characterized in that, The process of acquiring the tailings dam operation plan information set, based on the tailings dam operation plan information set and the current slurry level, utilizes the spatiotemporal lag prediction model to predict the pore water pressure change information set inside the tailings dam in future periods, including: The tailings dam operation plan information set includes the planned total amount of tailings to be discharged in the future time period, the planned sequence of ore discharge points, and the planned slurry concentration. Based on the planned total ore discharge and the planned slurry concentration, combined with the current slurry level, the spatiotemporal change process of the reservoir level in the future period is deduced through material balance and rheological characteristic analysis, and a level prediction sequence is generated. Based on the planned ore dumping point location sequence, the corresponding sedimentation area is identified, and the physicochemical properties data of the tailings slurry corresponding to this area are extracted according to the slurry spatial distribution model. Based on the physicochemical properties data of the tailings slurry, the physical influence factors of future ore discharge behavior on the seepage path of the dam body and the chemical influence factors on the hysteretic response to pore water pressure are determined. The liquid level projection sequence, the physical influence factor, and the chemical influence factor are input into the spatiotemporal lag prediction model. Through the liquid level change events and the lag fingerprint coefficient set, the dynamic response process of pore water pressure in different time periods and different dam areas is simulated, and the pore water pressure change information set expressed in spatiotemporal grid form is generated.

7. The method according to claim 6, characterized in that, The determination of chemical influencing factors on the future pore water pressure hysteresis response based on the physicochemical properties data of the tailings slurry includes: For the sedimentary areas corresponding to the planned ore dumping point location sequence, the mineral chemical composition is extracted from the physicochemical property data of the tailings slurry. Chemical components whose catalytic or inhibitory effects on the pore water pressure transmission process exceed a preset chemical threshold are identified as key chemical components and their concentrations are obtained. Based on the real-time reaction kinetic parameters of the key chemical components in the pore water environment, the rate of change of their concentration on the structure of the pore medium is calculated, and then the potential influence on the pore water pressure hysteresis time is deduced. Based on the chemical hysteresis fingerprint coefficient, the potential impact level is calibrated and normalized, and the potential impact level is quantified into a specific chemical correction adjustment value for the hysteresis time. When the chemical correction adjustment value is positive, it indicates that the chemical component has a catalytic effect on the pore water pressure response; When the chemical correction adjustment value is negative, it indicates that the chemical component has an inhibitory effect on the pore water pressure response; The absolute value of the chemical correction adjustment value indicates the strength of its effect; The chemical influence factor is generated based on the positive or negative attributes of the chemical correction adjustment value.

8. The method according to claim 7, characterized in that, The process of dynamically assessing the stability of the tailings dam based on the pore water pressure change information set and outputting stability early warning information includes: Based on the pore water pressure change information set and the slurry spatial distribution model, the adverse effects of future pore water pressure increases on the stability of the dam structure are analyzed, the overall safety status of each future period is dynamically predicted, and the future safety factor change curve is generated. Analyze the future safety factor change curve to identify potentially dangerous periods when the safety factor is below the safety threshold or the rate of decline exceeds the critical rate; The stability warning information is generated based on the future safety factor change curve and the potential danger period.

9. A dynamic geological modeling system based on adaptive updates of multi-source heterogeneous data, characterized in that, The method applied to any one of claims 1-8 includes: The model building module is used to acquire a tailings dam state feature dataset and, based on the tailings dam state feature dataset, to build a spatiotemporal lag prediction model using machine learning methods. The information extraction module is used to obtain the tailings dam operation plan information set, and based on the tailings dam operation plan information set and the current slurry level, it uses the spatiotemporal lag prediction model to predict the pore water pressure change information set inside the tailings dam in the future period. The strategy generation module is used to perform dynamic evaluation of tailings dam stability based on the pore water pressure change information set, generate tailings dam control strategies, and output stability early warning logs.

Images