Karst basin coupling hydrological model prediction method

Abstract

The invention relates to the field of hydrology and water resource management and karst hydrodynamics, and discloses a karst basin coupling hydrology model prediction method, which comprises the following steps: S1, collecting and standardizing multi-source heterogeneous monitoring data of a karst basin; s2, constructing a hydrological model for dynamic coupling of surface runoff, fracture seepage and a pipeline network; s3, performing global optimization on the multi-dimensional hydrological parameters by adopting a hybrid intelligent optimization method; s4, performing iterative correction on the model based on a multi-level verification system; and S5, outputting a hydrological prediction result and generating disaster early warning information. By adopting a dynamic adjusting technology of karst area underground river and surface water flow network joint modeling, multi-karst water-containing medium characteristic analysis and integration of underground river and surface water system response characteristics, the problems that a traditional method cannot accurately predict underground water flow and flow velocity, and is difficult to deal with complexity of a karst water system, lack of data and the like are solved.

Description

Technical Field

[0001] The present invention relates to the fields of hydrological water resources management and karst hydrodynamics, and specifically to a prediction method for a coupled hydrological model in a karst basin. Background Art

[0002] In karst areas, due to the special structure of the aquifer system and the extremely inhomogeneous aquifer medium, the laws of hydrological movement are very different from those of general basins. There are many similarities between the underground river system and the surface river in karst areas, especially in the high similarity between the karst pipeline structure of the underground river system and the surface water system network. The response of the underground river system to rainfall is also very rapid, and there are many common characteristics between its flow rate and velocity changes and those of surface rivers. However, due to the complex geological conditions and hydrological characteristics in karst areas, there are still significant deficiencies in the monitoring and prediction of karst water systems in the prior art. Especially in the case of scarce data and difficult data acquisition, it is impossible to provide an effective prediction basis for the water resources in karst areas, which affects the smooth progress of related projects.

[0003] The existing technical solutions mainly focus on the hydrological prediction and water resources management of surface water basins, while the research on underground rivers is relatively weak. As an important water source in karst areas, the changes in the water volume and flow velocity of the underground river system are usually relatively complex, restricted by various factors such as the karst pipeline structure and the characteristics of the aquifer medium. Traditional hydrological models mostly rely on the changes in surface water systems for prediction, ignoring the mutual conversion and linkage effects between groundwater systems and surface water systems. Due to the inhomogeneity of karst water systems, this method often leads to a large deviation between the prediction results and the actual situation, and cannot effectively solve the complex problems faced in karst water resources management.

[0004] In addition, there are also great limitations in the monitoring of groundwater flow in traditional technologies. Existing methods mostly estimate the flow rate by monitoring the changes in groundwater levels. However, due to the particularity of karst underground rivers, this indirect estimation is often difficult to accurately reflect the true flow rate of underground rivers. Especially when the flow rate attenuation curve of karst springs shows a multi-segment curve, the existing methods cannot effectively capture the influence of different karst media on water flow, resulting in inaccurate prediction of water flow attenuation. In the absence of sufficient monitoring data, traditional water flow prediction models cannot provide reliable hydrological data, which poses a great challenge to the construction and water resources utilization in karst areas.

[0005] Due to these deficiencies in the prior art, the development of water resources, construction planning, and risk assessment in karst areas often rely on rough assumptions or simplified models. These methods do not fully consider the dynamic interaction relationship between underground rivers and surface water systems, nor can they solve the problems of accurately predicting underground river flow rates and precipitation process responses in practical applications. Therefore, in view of these problems, the present invention proposes a prediction method for a coupled hydrological model in a karst basin. Summary of the Invention

[0006] In view of the deficiencies of the prior art, the present invention provides a prediction method for a coupled hydrological model in a karst basin, which solves the problems of data scarcity and inaccurate prediction in the prediction of underground river flow and water resource management in karst areas.

[0007] To achieve the above objectives, the present invention is realized through the following technical solutions: A prediction method for a coupled hydrological model in a karst basin, comprising the following steps:

[0008] S1. Collect and standardize multi-source heterogeneous monitoring data in the karst basin;

[0009] S2. Construct a hydrological model with dynamic coupling of surface runoff, fissure seepage and pipeline network;

[0010] S3. Use a hybrid intelligent optimization method to globally optimize multi-dimensional hydrological parameters;

[0011] S4. Iteratively correct the model based on a multi-level verification system;

[0012] S5. Output hydrological prediction results and generate disaster warning information.

[0013] Preferably, the step S1 includes:

[0014] S1.1. Use an unmanned aerial vehicle hyperspectral imaging system to obtain surface runoff characteristic information;

[0015] S1.2. Deploy a fiber optic pressure sensor network to collect groundwater flow dynamic data;

[0016] S1.3. Generate a standardized input data set through a spatio-temporal interpolation algorithm and dimensional normalization processing.

[0017] Preferably, the realization of the dynamic coupling in the step S2 includes:

[0018] When the cumulative rainfall exceeds a preset threshold, trigger the rapid recharge logic from surface runoff to the pipeline network;

[0019] Correct the spatio-temporal resolution difference between surface and underground water flows through a phase difference compensation function;

[0020] The phase difference compensation function calculates the spatio-temporal lag parameter using a cross-correlation algorithm.

[0021] Preferably, the step S3 includes a three-stage optimization framework, which is executed as follows:

[0022] Generate an initial solution set covering the feasible region of parameters based on a low-discrepancy sequence;

[0023] Parallelly execute the particle swarm algorithm and the genetic algorithm to obtain the Pareto front solution;

[0024] Screen the optimal solution set through parameter sensitivity analysis and physical constraint rules;

[0025] When it is detected that the parameter change rate exceeds the statistical confidence interval, automatically restart the optimization process.

[0026] Preferably, the multi-level verification system in step S4 includes:

[0027] Verification of the numerical solution of the unit-level equation;

[0028] Verification of the module-level standard software comparison;

[0029] Verification of the system-level continuous operation stability;

[0030] Verification of the matching of the tracer test at the field level.

[0031] Preferably, in the verification of the matching of the tracer test at the field level, if the tracer recovery rate is lower than 85%, trigger the topological reconstruction of the underground pipeline.

[0032] Preferably, the execution of step S5 includes:

[0033] Real-time evaluate the disaster probability through the water inrush risk index, and the index is calculated as the normalized ratio of the predicted flow to the safety flow threshold;

[0034] When the output result triggers a red warning signal, synchronously generate a three-dimensional visualization map of the hydraulic gradient field.

[0035] Preferably, the flow law of the pipeline network in step S2 is described by an improved Hagen-Poiseuille equation, and its inertial correction term coefficient is calibrated to 0.17 through historical rainstorm event data.

[0036] Preferably, the low-discrepancy sequence in step S3 adopts the Sobol sequence, and the analysis of the parameter sensitivity adopts the SHAP value algorithm, and its screening threshold is set to the top 20% of the key parameters.

[0037] The present invention provides a prediction method for a coupled hydrological model in a karst basin. It has the following beneficial effects:

[0038] 1. By adopting the technology of jointly modeling the underground river and surface water flow network in karst areas, the present invention achieves the technical effect of accurately predicting the groundwater flow rate and velocity. Compared with the existing method that only relies on surface water data for prediction, the present invention can more comprehensively evaluate and predict the hydrological movement law in karst areas by combining the common characteristics of the underground river system and surface water flow network. Especially in the case of scarce data, the technical solution of the present invention effectively makes up for the deficiency that it is difficult to obtain groundwater data by traditional methods, and provides more reliable support for construction and water resource management in karst areas.

[0039] 2. Through the analysis model of multiple karst aquifer medium characteristics, the present invention successfully solves the problem of multi-segmental change of the water volume attenuation curve in the karst water system. Compared with the existing method that usually assumes the water flow change as a single linear process, the multi-segmental curve analysis method of the present invention can accurately simulate the complex underground water flow characteristics in karst areas, avoid the prediction error caused by ignoring the influence of different karst media, improve the accuracy of underground river flow prediction, and provide a more scientific basis for water resource development and management in karst areas.

[0040] 3. By adopting the dynamic adjustment technology that integrates the response characteristics of the underground river and surface water system, the present invention achieves the effect of real-time monitoring and predicting the change of groundwater flow rate during the construction period in karst areas. Different from the situation in the existing technology where it is difficult to make accurate predictions, the present invention can quickly respond to the precipitation process and adjust and predict the water flow change according to the characteristics of the underground river system, solves the problem of being unable to obtain effective data in the case of lack of data or difficult to collect, provides more accurate hydrological support for engineering construction in karst areas, and reduces the potential risks caused by water flow change. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic flow chart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] Please refer to the attached Figure 1 , the embodiment of the present invention provides a prediction method for a coupled hydrological model of a karst basin, including the following steps:

[0044] S1. Collect and standardize the multi-source heterogeneous monitoring data of the karst basin;

[0045] This step aims to construct a standardized data base for basin hydrological elements, break through the linkage barriers between surface and underground observation systems through multi-source data fusion technology, and provide a unified data specification for subsequent dynamic coupling modeling. In the integration of heterogeneous data sources, key technical issues such as spatio-temporal reference differences, dimensional conflicts, and abnormal data interference are mainly solved to form an input data set with complete spatial coverage and self-consistent physical meaning.

[0046] In some embodiments, the layout of the data acquisition system includes the following technical configurations:

[0047] Specifically, an integrated space-air-ground sensing network is adopted to achieve full-element capture. The hyperspectral imager carried by the UAV cluster collects the spectral characteristics of surface runoff with a ground resolution of 0.5 m, and its working band covers the visible light to short-wave infrared range (400 - 2500 nm). For the dynamic monitoring of groundwater flow, an optical fiber pressure sensing array is arranged along the karst pipeline. The single-point measurement accuracy of the array is ±0.1% FS, and the sampling frequency is 10 Hz. Additionally, a fluorescent nanotracer chelated with terbium (Tb 3+ ) is injected at the intersection of key seepage channels, and a laser fluorescence detection device is used to track the downstream concentration.

[0048] The implementation method of data standardization is as follows:

[0049] In a possible implementation, the original data needs to be preprocessed in multiple dimensions to obtain input features adapted to the model:

[0050] Unification of spatial reference: Interpolate the discrete site data into a 1 km × 1 km grid field. The interpolation algorithm uses the Kriging method, and its semi-variogram model is defined as:

[0051]

[0052] In the formula, c 0 is the nugget effect constant, reflecting the microscopic random error; c is the structural variance; a is the range parameter, characterizing the spatial correlation threshold; h is the spatial lag distance (i.e., the spatial distance between sample points). The parameters are preferably determined by cross-validation.

[0053] Dimensional normalization: Construct independent standardization functions for each hydrological parameter. For the parameter set x i , its conversion formula is:

[0054]

[0055] Among them, μ i is the sample mean, and σ i is the standard deviation. The coefficient 3σ is selected to normalize 99.7% of the data to the [-1, 1] interval to avoid unbalanced parameter weights.

[0056] The implementation process of data quality control involves a multi-criterion verification mechanism:

[0057] Generally, physical constraint rules are set to identify and repair abnormal data. For example, the groundwater seepage velocity follows the upper limit constraint of Darcy's law:

[0058]

[0059] where v max is the maximum flow velocity; K sat is the saturated hydraulic conductivity; is the hydraulic gradient; 1.2 is the empirical correction coefficient.

[0060] When the detected flow velocity v > v max , the data correction process is triggered. Specifically, the abnormal value is replaced with the weighted mean of the valid value of the previous time series and the monitoring value of the adjacent site.

[0061] In a typical implementation scenario, tracer data is used to verify the pipeline connectivity through time-frequency joint analysis. The waveform characteristics of the tracer concentration are extracted based on wavelet transform:

[0062]

[0063] where Ψ is the Morlet mother wavelet function, a is the scale factor, b is the translation factor, C(t) is the transformed signal, t is the time variable, and c(t) is the original signal. The topological accuracy of the underground pipeline is judged by the coincidence degree of the arrival time of the main peak.

[0064] It should be noted that the intermediate parameters (such as nugget constant c 0 , range a, etc.) generated in the above processing links are updated online through an adaptive feedback mechanism. For example, the semi-variogram parameters are dynamically optimized using the sliding window method:

[0065]

[0066] where N is the length of the update window, which is dynamically adjusted according to the system storage resources, a t+1 is the optimization parameter, is the optimization operation, γ obs (h k ) is the observed value, γ model (h k , a) is the model output value, ||·|| 2 is the square norm, and h k is the time step.

[0067] Through the above technical means, this step realizes the deep integration and quality improvement of multi-source heterogeneous data. The generated standardized data set can be directly input into the coupling model calculation process of step S2 and provide a reliable training basis for subsequent parameter optimization. The model parameters involved in the processing process (such as nugget constant c 0 , standard deviation σ i , etc.) are stored in the distributed database, supporting multi-user concurrent access and subsequent analysis and traceability.

[0068] S2. Construct a hydrological model for the dynamic coupling of surface runoff, fissure seepage and pipeline network;

[0069] Based on the above standardized data set, this step establishes a mathematical and physical model representing the unique double-layer water cycle structure in the karst area by introducing a multi-scale coupling mechanism and a non-linear transfer function. The innovation of the model lies in solving the technical problems that traditional methods are difficult to reflect the rapid interaction between surface and groundwater and the lag effect of pipeline flow, and achieving the balance between the completeness of physical processes and computational efficiency.

[0070] The construction scheme of the surface runoff module is as follows:

[0071] In some embodiments, an improved SCS-CN model is used to simulate the surface runoff generation process. A compensation term for the storage effect of karst depressions is added to the traditional water balance equation:

[0072]

[0073] In the formula, f c is the storage capacity of the depression per unit area (m 3 / km 2 ), t c is the concentration time (h), Q s is the surface runoff, P is the precipitation, and S is the potential maximum runoff storage. The model parameters are jointly calibrated by the vegetation coverage and land use type retrieved by the UAV hyperspectral in step S1. Among them, the infiltration parameter K d is calculated by the following formula:

[0074]

[0075] Among them, NDVI is the normalized difference vegetation index, K s,soil is the saturated hydraulic conductivity, K d is the infiltration capacity, 0.03 is an empirical coefficient, and exp(-0.5·NDVI) is the vegetation coverage influence factor.

[0076] The technical implementation of the fissure seepage module includes the following contents:

[0077] Specifically, the three-dimensional Richards equation for the unsaturated zone is used to describe the water flow movement in the fractured medium, and its discrete form is as follows:

[0078]

[0079] In the formula, C(h) is the unsaturated water capacity (dθ / dh), K(h) is expressed by an exponential model, α is a fitting parameter, K(h) is the permeability, is the gradient operator, z is the ground elevation, Q ex is the external exchange source term. The boundary conditions are dynamically applied through the measured water head data of the fiber optic pressure sensing array in step S1. The spatial discretization uses unstructured tetrahedral meshes, and the node density is increased to 3 times that of the adjacent area in the karst pipeline intersection area.

[0080] The following improvement measures are adopted for the mathematical characterization of the pipeline network module:

[0081] In a possible implementation, the pipeline flow model introduces an inertial effect correction term based on the Hagen-Poiseuille equation:

[0082]

[0083] In the formula, β is the dynamic correction coefficient (rated as 0.17 through the data of the Puding test site in Guizhou), μ is the water flow viscosity, Q p is the pipeline flow rate, ΔH is the pressure difference, L is the pipeline length, is the time change rate of the flow rate. The pipeline topology is dynamically adjusted according to the inversion result of the tracer migration path in step S1, and the value range of the diameter D is 0.5 - 8.0 m.

[0084] The implementation logic of the dynamic coupling mechanism is as follows:

[0085] Generally, the interaction process is triggered according to the real-time rainfall data in step S1. When the rainfall intensity I r > 50 mm / h, the fast recharge channel from surface runoff to the pipeline network is activated. The recharge term is calculated by the following formula:

[0086] Q ex = min(Q s ·γ(t), Q p,max - Q p )

[0087] In the formula, γ(t) is the time-varying recharge coefficient, the time difference between the surface and underground responses is matched through the cross-correlation algorithm, Q ex is the recharge amount, Q s is the surface runoff, γ(t) is a function of time, Q p,max is the maximum pipeline flow rate, Q pis the current pipeline flow rate. The correlation coefficient is dynamically updated by the following formula:

[0088]

[0089] where Δt is determined by phase difference analysis, the calculation window length is fixed at 6 hours, γ(t) is the cross-correlation function, Q s (τ) is the surface runoff signal, Q p (τ + Δt) is the pipeline flow rate signal, ∥Q s ∥ is the norm of the surface runoff, ∥Q p ∥ is the norm of the pipeline flow rate, Δt is the time delay, and t 0 is the starting time.

[0090] It should be noted that the spatial heterogeneity parameters (such as k s , α, etc.) are dynamically optimized by the hybrid optimization algorithm in step S3. During the model calculation process, the unit-level verification process in step S4 is triggered. If the node mass conservation error exceeds 1e-6 within 3 consecutive time steps, the numerical stability enhancement algorithm is automatically called.

[0091] S3. Adopt a hybrid intelligent optimization method to globally optimize multi-dimensional hydrological parameters;

[0092] Based on the standardized data set generated in step S1 and the dynamic coupling model established in step S2, this step realizes the efficient inversion of karst hydrological parameters by integrating an intelligent search algorithm and a physical constraint mechanism. The innovation of this method lies in overcoming the technical bottleneck that traditional parameter tuning methods are prone to fall into local optima and have high computational costs, and establishing an intelligent reduction strategy and a multi-criterion decision-making mechanism for the parameter feasible region.

[0093] The generation strategy of the initial parameter solution set involves the following key technologies:

[0094] In some embodiments, an improved low-discrepancy sequence is used to cover the multi-dimensional parameter space. For the n-dimensional hydrological parameter Θ = (θ 1 ,..., θ n ), the initial samples are generated by the Sobol sequence:

[0095]

[0096] where m is the number of bits in the binary expansion, ν {i,j}∈{0,1} is the preset number of directions, is the i-th dimensional parameter of the K-th sample, ν i,j is the j-th digit of the i-th dimension, 2 -j is the binary number weight. The sample size is dynamically adjusted according to the results of parameter sensitivity analysis and is initially set to 10n 2 (n is the parameter dimension).

[0097] The execution process of the hybrid optimization framework is as follows:

[0098] Specifically, a phased strategy is adopted to achieve the progressive optimization of the parameter solution set:

[0099] Coarse search stage: Use Latin hypercube sampling to screen out a subset of feasible solutions that satisfy the mass conservation condition. The mass conservation constraint equation is:

[0100]

[0101] In the formula, ΔS is the change in storage variable, ET is the evapotranspiration, and Q g is the groundwater discharge.

[0102] Fine-tuning parameter stage: Design a PSO-GA hybrid algorithm, where the particle swarm algorithm adopts a dynamic inertia weight:

[0103]

[0104] In the formula:

[0105] w(t0: The inertia weight value at the t-th iteration (dimensionless)

[0106] w max : Initial maximum weight (take 0.9)

[0107] w min : Termination minimum weight (take 0.4)

[0108] t: Current iteration number (loop variable, t ∈ [0, T max )

[0109] T max : Total number of iterations (default 200 times, can be extended according to the parameter dimension).

[0110] Adaptive adjustment of the crossover probability of the genetic algorithm:

[0111]

[0112] Among them, p c is the crossover probability, 0.7 is the initial crossover probability, and exp is the exponential function.

[0113] Solution screening stage: Based on the SHAP value, evaluate the contribution degree of the parameters to the objective function, and discard the non-sensitive parameter combinations whose cumulative contribution ratio is less than 5%.

[0114] The re-learning mechanism based on physical constraints includes the following content:

[0115] In a possible implementation, when it is detected that the parameter drift exceeds the statistical confidence interval, physical-guided neural network-assisted optimization is triggered. The network input layer contains 10 hydrological characteristic parameters (such as cumulative rainfall, surface runoff coefficient, etc.), and the hidden layer is activated by the following formula:

[0116]

[0117] where σ(x) is the Sigmoid function value, x is the input value, b is the bias term, and e is the base of the natural logarithm.

[0118] The network training objective function fuses the numerical residual and the physical constraint term of Darcy's law:

[0119]

[0120] where α = 0.6 is the balance factor determined by cross-validation, L is the total loss, α is the weighting coefficient, MSE is the mean square error, μ is the fluid viscosity, K is the permeability, v is the flow velocity, is the L2 norm (squared).

[0121] Implementation details of the key parameter self-update logic:

[0122] For time-varying parameters (such as fracture permeability K), a dual-threshold trigger mechanism is set:

[0123] Primary monitoring: When the daily average change rate ΔK / K > η 1 (η 1 = 20%), local parameter re-optimization is started;

[0124] Secondary monitoring: When the cumulative change Σ|ΔK| > η 2 (η 2 = 3σ k ), the global parameter reconstruction process is triggered.

[0125] It should be noted that the above optimization process and the verification system in step S4 form a closed-loop feedback. For example, when the unit-level verification in step S4 detects that the objective function deteriorates for 5 consecutive iterations, this step is automatically recalled to re-initialize the search space.

[0126] S4. Iteratively correct the model based on a multi-level verification system;

[0127] In this step, by constructing a multi-level closed-loop verification system, cross-validation and adaptive feedback correction are performed on the optimized parameters output in step S3 and the coupled model in step S2. This method breaks through the limitation of the single-dimensional verification of the traditional verification system and realizes the dual guarantee of the physical mechanism correctness and actual prediction accuracy of the karst hydrological model.

[0128] The technical solution for unit-level verification is as follows:

[0129] In some embodiments, Newton iteration residual analysis is used to test the numerical solution stability of partial differential equations. Define the unit mass conservation residual as:

[0130]

[0131] where δ i is the perturbation amount, Δt n is the time step, Ω i is the control region, φ is the scalar field, F is the flux vector field, is the divergence, is the time rate of change.

[0132] If within 5 consecutive time steps then trigger the local grid encryption operation. The encryption standard is:

[0133]

[0134] where h old , h new respectively represent the grid sizes before and after encryption, δ i is the perturbation amount, and σ is the standard deviation or reference value.

[0135] The execution logic of module-level verification includes the following key points:

[0136] In a possible implementation, the process line comparison verification is performed by docking with the standard hydrological software (HEC-HMS). Define the matching degree index:

[0137]

[0138] where γ is the percentage of relative error, Q obs,k is the observed value, Q sin,k is the simulated value, and N is the total number of samples.

[0139] When γ > 20%, activate the parameter sensitivity recheck process. The recheck is based on the SHAP value sorting result in step S3, and the top 30% sensitive parameters are preferentially corrected.

[0140] The key technical features of system-level verification are as follows:

[0141] Specifically, the system stability is evaluated by the Lyapunov exponent:

[0142]

[0143] In the formula, J(k) is the Jacobian matrix of the model in step S2, v kis the perturbation vector, n is the number of samples, and k is the index. If the maximum Lyapunov exponent λ max > 5.0, it is determined that the system is in a chaotic state, and the simulation step size is forced to be shortened to 1 / 5 of the original value.

[0144] The technical deepening content of the external field level verification includes:

[0145] Generally, the accuracy of the underground pipeline topology is verified by combining the tracer recovery test. Design the recovery rate threshold control strategy:

[0146]

[0147] where R is the total concentration ratio, C total,obs is the total concentration to be observed, and C total,inj is the injection concentration.

[0148] When triggered, the pipeline topology model is reconstructed based on the UAV microtopography scanning data in step S1. The key update formula is:

[0149]

[0150] where t peak is the peak time of the tracer concentration, the old pipeline diameter D old is taken from the current model parameters in step S2, D new is the new parameter value, t peak,obs is the observed peak time, and t peak,sim is the simulated peak time.

[0151] The decision logic of the self-diagnosis module covers the following content:

[0152] When the same parameter exceeds the threshold in three consecutive rounds of unit-level verification, trigger the optimization path bifurcation judgment:

[0153] Return to step S1 to review the monitoring data

[0154] where θ S3 is the currently estimated parameter, and θ prior is the prior parameter.

[0155] Otherwise, jump to step S3 to perform the parameter space expansion operation:

[0156]

[0157] where is the maximum value of the i-th dimension, and is the minimum value of the i-th dimension.

[0158] S5. Output the hydrological prediction results and generate disaster warning information.

[0159] Based on the verified model in step S4 and real-time monitoring data, this step generates disaster warning information covering key nodes of karst depressions, sinkholes, and underground rivers by integrating a multi-source forecast scenario library and a mutation recognition algorithm. The method solves the technical pain points of the traditional warning system, such as lagging response and fuzzy spatial positioning, and realizes the dual functions of dynamic risk quantitative assessment and emergency decision-making assistance.

[0160] In some embodiments, a multi-threaded architecture is adopted to extract key risk factors in parallel:

[0161] The surface runoff depth is directly mapped by the output of the coupling model in step S2:

[0162]

[0163] In the formula, Q rain is dynamically input by the radar rainfall inversion data in step S1, Q inf is the fissure seepage volume, Q evap is the evapotranspiration loss, H s (t) is the water volume or water level.

[0164] The advanced prediction of the peak pressure in the underground pipeline is solved based on the characteristic line method:

[0165]

[0166] In the formula, the wave velocity β is called through the optimized parameter library in step S3, H is the water depth, x is the spatial position, and g is the acceleration due to gravity.

[0167] Specifically, a fuzzy evaluation system is constructed to determine the warning level threshold:

[0168] Yellow warning trigger condition (critical warning state):

[0169] When and lasts for 2 hours.

[0170] Among them, R f is the flow velocity ratio, v pipe is the pipeline flow velocity, v crit is the critical flow velocity, is the head acceleration.

[0171] Orange warning trigger condition (high-risk state):

[0172] R f ≥0.92 and more than 3 sudden changes in the backflow volume of sinkholes are detected.

[0173] Red warning trigger condition (emergency state):

[0174] Hs (t) > H hist_max or

[0175] wherein, historical maximum runoff depth H hist_max is extracted from the standardized database in step S1.

[0176] Early warning information generation coding and verification mechanism

[0177] In a possible implementation, a three-stage information processing flow is adopted:

[0178] Information compression: Cluster the early warning data of adjacent nodes based on time series similarity, and set the compression ratio threshold to:

[0179]

[0180] wherein, v i is the velocity vector i, v j is the velocity vector j, ∥·∥ 2 is the Euclidean norm, v i , v j is the velocity scalar, max(v i , v j ) is the maximum velocity.

[0181] When the conditions are met, they are merged into area-level early warnings, and the geographical accuracy is maintained at a 1 km grid.

[0182] Digital signature: Use elliptic curve encryption to generate a check code:

[0183] Sig = ECDSA(Hash(t||Q pcak ||Coord)) The private key is stored in the verification server whitelist in step S4.

[0184] Multicast: Push through the 4G / Beidou dual-channel to a preset terminal group, and trigger the real-time model degradation operation in step S2 to release communication resources when it fails.

[0185] Logical constraints of the feedback optimization interface

[0186] It should be noted that when the false alarm rate of the red early warning exceeds η = 5% continuously for 3 times, a parameter rollback mechanism is triggered:

[0187] θ rollback = Median(θ t-1 , θ t-2 , θ t-3 )

[0188] where the historical parameter set θ t-kSelect from the optimization path library in step S3 and verify the physical rationality through the unit-level verification in step S4.

[0189] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A karst basin coupled hydrological model prediction method, characterized in that: The following steps are involved: S1. Collect and standardize multi-source heterogeneous monitoring data in karst basins; S2. Construct a hydrological model that dynamically couples surface runoff, fracture seepage, and pipeline networks; S3, using hybrid intelligent optimization method to globally optimize multi-dimensional hydrological parameters; S4, iteratively modify the model based on a multi-level verification system; S5. Output hydrological prediction results and generate disaster warning information.

2. A karst basin coupled hydrological model prediction method according to claim 1, characterized in that: The step S1 comprises: S1.

1. Use UAV hyperspectral imaging system to obtain surface runoff characteristics information; S1.2, deploy a fiber optic pressure sensor network to collect groundwater flow dynamic data; S1.

3. Generate a standardized input data set through spatiotemporal interpolation algorithm and dimensional normalization processing.

3. The karst basin coupled hydrological model prediction method according to claim 1 is characterized in that: The implementation of dynamic coupling in step S2 includes: When the accumulated rainfall exceeds the preset threshold, the rapid recharge logic of surface runoff to the pipeline network is triggered; The difference in temporal and spatial resolution between surface and groundwater flow is corrected by a phase difference compensation function; The phase difference compensation function uses a cross-correlation algorithm to calculate the spatiotemporal lag parameters.

4. The method for predicting coupled hydrological models in karst watersheds according to claim 1, characterized in that: The step S3 includes a three-stage optimization framework, which is performed as follows: Generate an initial solution set covering the feasible domain of parameters based on low-discrepancy sequences; Parallel execution of particle swarm optimization and genetic algorithm to obtain Pareto front solutions; Screen the optimal solution set through parameter sensitivity analysis and physical constraint rules; When it is detected that the parameter change rate exceeds the statistical confidence interval, the optimization process is automatically restarted.

5. The method for predicting coupled hydrological models in karst watersheds according to claim 1, characterized in that: The multi-level verification system in step S4 includes: Verification of numerical solutions of unit-level equations; Module-level standard software comparison and verification; System-level continuous operation stability verification; Field level tracer test matching verification.

6. A karst basin coupled hydrological model prediction method according to claim 5, characterized in that: In the field-level tracer test matching verification, if the tracer recovery rate is lower than 85%, the underground pipeline topology reconstruction is triggered.

7. The method for predicting a karst basin coupled hydrological model according to claim 1, characterized in that: The execution of step S5 includes: Real-time assessment of disaster probability through a water inrush risk index, which is calculated as a normalized ratio of predicted flow to a safe flow threshold; When the output result triggers a red warning signal, a three-dimensional hydraulic gradient field visualization map is generated synchronously.

8. A karst basin coupled hydrological model prediction method according to claim 3, characterized in that: The pipeline network flow law in step S2 is described by the improved Hagen-Poiseuille equation, and its inertia correction coefficient is calibrated to 0.17 based on historical rainstorm event data.

9. A karst basin coupled hydrological model prediction method according to claim 4, characterized in that: In the step S3, the low-difference sequence adopts the Sobol sequence, and the parameter sensitivity analysis adopts the SHAP value algorithm, and its screening threshold is set to the top 20% of key parameters.

Images