Tailing pond risk assessment method

Abstract

The application relates to a tailing pond risk assessment method. The method comprises the following steps: determining a tailing pond risk assessment system and preprocessing sample data of each index; determining sub-basin information and water ecology sensitivity; calculating the comprehensive weight of each index; constructing a dynamic Bayesian network model according to the tailing pond risk assessment system, the preprocessed sample data of each index and the comprehensive weight of each index; based on a dynamic Bayesian network model, according to the rainfall capacity and the sample data, predicting the pollutant permeation capacity; constructing a pollutant diffusion model according to the information of the sub-basin where the tailing pond is located; simulating a pollutant diffusion result according to the pollutant permeation amount based on a pollutant diffusion model; and evaluating the risk state of each sub-basin according to the pollutant diffusion result and the water ecology sensitivity. According to the method, the risk of the tailings pond is dynamically evaluated, the accuracy, practicability and perspectiveness of risk evaluation are improved, risk evaluation is carried out on the harm caused by the downstream drainage basin, and therefore the more scientific and more comprehensive risk evaluation method is provided.

Description

Technical Field

[0001] This application relates to the technical field of tailings pond risk assessment, and particularly to a tailings pond risk assessment method. Background Art

[0002] A tailings pond is one of the important engineering construction contents of a metal ore mining and beneficiation project, and is a debris flow hazard source with high potential energy.

[0003] The Chinese patent application with the application number CN202111192961.3 and the publication date of March 1, 2022 discloses a risk monitoring and early warning system for tailings ponds based on the Internet of Things, including a risk comprehensive monitoring module, a risk intelligent assessment module, a risk precise early warning module, a risk trend prediction module, and a disaster simulation and simulation module; the risk comprehensive monitoring module is used to realize the dynamic comprehensive display of tailings pond data, obtain the monitoring data and monitoring video data of tailings pond sensors, and the monitoring data includes rainfall, reservoir water level, dry beach length, phreatic line, surface displacement of the dam body, internal displacement of the dam body, surface displacement of the geological landslide body in the reservoir area, etc.; the monitoring video data includes the monitoring data of parts such as the overflow well, the ore discharge place at the beach top, the tail discharge pipeline, the downstream slope of the dam body, the reservoir water level gauge, and the dry beach benchmark; establish a GIS map of the tailings pond, and input the obtained monitoring data and monitoring video data into the GIS map to facilitate the comprehensive control of the Internet of Things perception point positions of each tailings pond through the GIS map.

[0004] In the process of implementing the above solution, the applicant found the following problems:

[0005] When assessing the risk of a tailings pond, it mainly relies on monitoring the real-time data of the tailings pond and comparing the real-time data with a threshold value to conduct a static risk assessment. The static assessment of comparing real-time data with a threshold value makes the above solution lack dynamics and foresight, thereby resulting in low accuracy of risk assessment and lagging risk early warning. In addition, the above solution can only assess the risk of the tailings pond itself and cannot consider in real time the risk hazards that the tailings pond may cause to the downstream basin, thus making the assessment risk not comprehensive enough and the comprehensiveness low. Summary of the Invention

[0006] Based on this, it is necessary to provide a tailings pond risk assessment method that can dynamically assess the risks existing in the tailings pond and can assess the risk hazards caused by the tailings pond to the downstream basin in view of the above technical problems.

[0007] A tailings pond risk assessment method provided by this application includes: determining a tailings pond risk assessment system and preprocessing the sample data of each index in the tailings pond risk assessment system; determining the information of the sub - watershed where the tailings pond is located and the water ecological sensitivity; calculating the comprehensive weight of each index; constructing a dynamic Bayesian network model based on the tailings pond risk assessment system, the preprocessed sample data of each index, and the comprehensive weight of each index; predicting the pollutant infiltration amount based on the dynamic Bayesian network model according to the rainfall and the sample data; constructing a pollutant diffusion model according to the information of the sub - watershed where the tailings pond is located; simulating the pollutant diffusion result based on the pollutant diffusion model according to the pollutant infiltration amount; and assessing the risk status of each sub - watershed according to the pollutant diffusion result and the water ecological sensitivity.

[0008] In one embodiment, determining the information of the sub - watershed where the tailings pond is located includes: calculating the water flow direction of each pixel according to the preset DEM data to generate a flow accumulation grid; extracting the information of the sub - watershed where the tailings pond is located according to the flow accumulation result of the flow accumulation grid; wherein the sub - watershed information includes the sub - watershed boundary, the respective flow and flow velocity of each sub - watershed.

[0009] In one embodiment, determining the water ecological sensitivity includes: weighting the flow and flow velocity of the sub - watershed and the preset water environment functional area data, re - classifying the weighted result to generate the water ecological sensitivity, and normalizing the water ecological sensitivity value.

[0010] In one embodiment, calculating the comprehensive weight of each index includes: determining the subjective weight of each index based on the analytic hierarchy process; calculating the objective weight of each index based on the entropy method; and calculating the comprehensive weight of each index according to the subjective weight and the objective weight.

[0011] In one embodiment, constructing a dynamic Bayesian network model based on the tailings pond risk assessment system, the preprocessed sample data of each index, and the comprehensive weight of each index includes: constructing a static Bayesian network model according to the tailings pond risk assessment system; assigning the prior probability of each node index in the static Bayesian network model according to the comprehensive weight of each index; determining the time transfer relationship of each index itself and merging the time - slice data into a format suitable for learning to generate an initial dynamic Bayesian network model; and training the initial dynamic Bayesian network model according to the sample data of each index to generate a target dynamic Bayesian network model.

[0012] In one embodiment, predicting the pollutant infiltration amount based on the dynamic Bayesian network model according to the rainfall and the sample data includes: predicting the probability of the infiltration amount being in different risk levels based on the dynamic Bayesian network model according to the rainfall; determining the infiltration amount values corresponding to different risk levels according to the sample data; and generating the pollutant infiltration amount according to the probability in different risk levels and the corresponding infiltration amount values.

[0013] In one embodiment, a pollutant diffusion model is constructed according to the sub-basin demarcation information where the tailings pond is located, including: establishing a node set according to the sub-basin demarcation; generating a pollutant diffusion model based on the node set according to the water flow direction and the direction and weight of the edges between the flow velocity nodes.

[0014] In one embodiment, after generating the pollutant diffusion model, it further includes: calculating the eigenvector centrality, local clustering coefficient, in-degree, and out-degree of each node according to the pollutant diffusion model.

[0015] In one embodiment, the risk status of each sub-basin is evaluated according to the pollutant diffusion result and the water ecological sensitivity, including: calculating the difference between the in-degree and out-degree of each node; calculating the comprehensive risk of each node according to the pollutant diffusion result, water ecological sensitivity, eigenvector centrality of each node, local clustering coefficient of each node, and the difference between the in-degree and out-degree of each node; normalizing the numerical values of the comprehensive risk of each node, and dividing the basins corresponding to each node into corresponding risk levels according to the numerical value ranges of different risks.

[0016] The present application adopts the above-mentioned tailings pond risk assessment method, which has the following beneficial effects:

[0017] 1. By introducing the deep learning method, a time series prediction of rainfall is carried out by constructing an LSTM model, and the predicted rainfall result is input into the constructed Bayesian network model to obtain the real-time probability of different risk levels of the tailings pond under different rainfall conditions, thus realizing the dynamic risk assessment of the tailings pond, which helps to improve the accuracy, practicability, and forward-looking of the risk assessment.

[0018] 2. By combining the output probability of the Bayesian network with the historical monitoring data, a risk assessment of the harm caused by the leachate of the tailings pond to the downstream basin is realized, and combined with the ecological sensitivity assessment result, the acute exposure standard of heavy metal pollutants, and the complex network model, thus providing a more scientific and comprehensive risk assessment method, making the risk assessment method of the present application have strong comprehensiveness, dynamics, and the advantage of refined management in the current field of tailings pond risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of a tailings pond risk assessment method in one embodiment;

[0020] Figure 2 It is a schematic diagram of the classification of tailings pond risk indicators in one embodiment;

[0021] Figure 3 It is a schematic diagram of the numerical values of indicators in each risk level after pretreatment in one embodiment;

[0022] Figure 4 It is a schematic diagram of the weights of each indicator in one embodiment;

[0023] Figure 5 Schematic diagram of the Bayesian network model in an embodiment;

[0024] Figure 6 Sub - watershed pollutant diffusion model in an embodiment;

[0025] Figure 7 Schematic diagram of the adjacency matrix in an embodiment;

[0026] Figure 8 Schematic diagram of the risk change trend of each sub - watershed at different time slices in an embodiment;

[0027] Figure 9 Schematic diagram of the risk levels of each sub - watershed at time slice 3 in an embodiment. Detailed implementation manners

[0028] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0029] Refer to Figure 1 , a tailings pond risk assessment method provided by the present application includes:

[0030] S101, determining a tailings pond risk assessment system and pre - processing the sample data of each index in the tailings pond risk assessment system; determining the information of the sub - watershed where the tailings pond is located and the water ecological sensitivity.

[0031] In the present application, the tailings pond risk assessment system includes primary indexes and secondary indexes, and the ranges of each index at different risk levels are determined. Among them, the primary indexes include dam body instability, flood overtopping, and seepage failure; the secondary indexes include rainfall, outer slope ratio, dry beach length, reservoir water level, beach top elevation, position of the phreatic line, surface displacement, and dry beach slope.

[0032] Pre - processing the sample data of each index includes handling missing values, outliers, and standardizing the data format. Among them, the state values of other nodes except rainfall are determined according to relevant specifications; the state values of the rainfall node are determined according to the local actual situation and relevant specifications. For quantifiable risk indexes, fuzzy intervals are used to normalize the original values, quantitative index values and qualitative indexes, and the normalized index interval values are calculated through the membership function.

[0033] Specifically, the calculation formula for positive indexes is:

[0034]

[0035]

[0036] The calculation formula for the negative index is:

[0037]

[0038] where a ij is the value of the j-th index at the i-th risk level; L is the minimum value or the left-side value of the index a ij for calculating the lower limit of normalization; U is the maximum value or the right-side value of the index a ij for calculating the upper limit of normalization. For a positive index, is the lower limit of normalization of the index a ij which is the index divided by the maximum upper bound of the index j among all risk levels i is the upper limit of normalization of the index a ij which is the index divided by the maximum upper bound of the index j among all risk levels i For a negative index, is the lower limit of normalization of the index a ij which is the minimum lower bound of the index j among all risk levels i divided by is the upper limit of normalization of the index a ij which is the minimum lower divided by where is the lower bound or the left-side value of the index a ij and is used to calculate the lower limit of the normalization interval in both the case of positive and negative indices; is the upper bound or the right-side value of the index a ij which is used to calculate the upper limit of the normalization interval in positive indices and the lower limit of the normalization interval in negative indices.

[0039] Referring to Figure 2 and Figure 3 , taking the position of the phreatic line as an example, the positive index of the position of the phreatic line is where is the maximum value among all index data, which is 8. So the interval range of risk level A is

[0040] Taking the dry beach slope as an example, the negative index of the dry beach slope is the minimum value among all index values, which is 0.1. So the interval range of risk level D is:

[0041] S201, calculate the comprehensive weight of each index.

[0042] S301, construct a dynamic Bayesian network model based on the tailings pond risk assessment system, the sample data of each index after pretreatment, and the comprehensive weight of each index.

[0043] S401, predict the pollutant infiltration amount based on the dynamic Bayesian network model according to the rainfall and the sample data;

[0044] S501, construct a pollutant diffusion model according to the information of the sub-watershed where the tailings pond is located.

[0045] S601, simulate the pollutant diffusion result based on the pollutant diffusion model according to the pollutant infiltration amount.

[0046] S701, evaluate the risk status of each sub-watershed according to the pollutant diffusion result and the water ecological sensitivity.

[0047] Introduce the deep learning method, perform time series prediction on the rainfall by constructing an LSTM model, and input the predicted rainfall result into the constructed Bayesian network model to obtain the real-time probability of different risk levels of the tailings pond under different rainfall conditions, thus realizing the dynamic risk assessment of the tailings pond, which helps to improve the accuracy, practicability and forward-looking of the risk assessment. In addition, combining the output probability of the Bayesian network with the historical monitoring data realizes the risk assessment of the harm caused by the leachate of the tailings pond to the downstream watershed, and combines the ecological sensitivity assessment result, the acute exposure standard of heavy metal pollutants and the complex network model, thus providing a more scientific and comprehensive risk assessment method, making the risk assessment method of this application have strong comprehensiveness, dynamics and the advantages of refined management in the current field of tailings pond risk assessment.

[0048] In one embodiment, determining the information of the sub-watershed where the tailings pond is located includes: calculating the water flow direction of each pixel according to the preset DEM data to generate a flow accumulation raster; extracting the information of the sub-watershed where the tailings pond is located according to the flow accumulation result of the flow accumulation raster; wherein, the sub-watershed information includes the sub-watershed boundary, the respective flow rates and flow velocities of the sub-watersheds.

[0049] In one embodiment, determining the water ecological sensitivity includes: weighting the flow rate and flow velocity of the sub-watershed and the preset water environment function area data, reclassifying the weighted result to generate the water ecological sensitivity, and normalizing the water ecological sensitivity value.

[0050] Specifically, the flow rate and velocity data of the sub-watersheds that have been extracted are used, combined with the pre-collected water environment functional area data, to conduct a watershed water ecological sensitivity assessment. The flow rate, velocity, and the preset water environment functional area data of the sub-watersheds are weighted according to a flow velocity weight of 0.3, a flow rate weight of 0.3, and a water environment functional area weight of 0.6. The weighted calculation results are reclassified to obtain the water ecological sensitivity, and finally, the numerical values of the water ecological sensitivity are normalized.

[0051] In one embodiment, referring to Figure 4 , the comprehensive weights of each index are calculated, including: determining the subjective weights of each index based on the analytic hierarchy process; calculating the objective weights of each index based on the entropy method; and calculating the comprehensive weights of each index according to the subjective weights and objective weights.

[0052] Specifically, the formula for calculating the objective weights of each index using the entropy method is:

[0053]

[0054] where s ij is the proportion of the i-th evaluated object under the j-th index; e j is the entropy value of the j-th index; n is the number of evaluation indicators; r j (j = 1, 2... n) is the weight of the j-th index.

[0055] Specifically, the comprehensive weights of each index are obtained using the matrix method, and the formula is as follows:

[0056] a i = v i / (v i + w i )

[0057] b i = w i / (v i + w i )

[0058]

[0059] where a i is the relative importance degree of index i obtained by the analytic hierarchy process; b i is the relative importance degree of index i obtained by the entropy method; v i is the subjective weight obtained by the analytic hierarchy process; w i is the objective weight obtained by the entropy method; C i is the comprehensive weight.

[0060] In one embodiment, a dynamic Bayesian network model is constructed based on the tailings pond risk assessment system, sample data of each indicator after preprocessing, and the comprehensive weight of each indicator, including: constructing a static Bayesian network model based on the tailings pond risk assessment system; assigning a priori probability of each node indicator in the static Bayesian network model according to the comprehensive weight of each indicator; determining the time transfer relationship of each indicator itself and merging the time slice data into a format suitable for learning to generate an initial dynamic Bayesian network model; training the initial dynamic Bayesian network model based on the sample data of each indicator to generate a target dynamic Bayesian network model.

[0061] Specifically, refer to Figure 5 First, a static Bayesian network structure is constructed based on the tailings pond risk assessment system and the actual scenario. Then, the hill climbing algorithm is used to learn the parameters of the sample data of each indicator after preprocessing to correct and improve the static Bayesian network structure.

[0062] In this embodiment, a hill climbing algorithm optimizes the structure of a static Bayesian network model through local search. First, possible indicator nodes and the connections between them are defined. Then, based on data fit or the information criterion, the algorithm gradually optimizes the local structure to find the optimal static Bayesian network model. The hill climbing algorithm generates, evaluates, and selects a locally optimized network structure, iterating until a stopping condition is reached, ultimately outputting the optimal Bayesian network topology.

[0063] After the static Bayesian network model is constructed, the prior probabilities of each node indicator in the static Bayesian network model are assigned based on the sample data and natural laws of each indicator. The conditional probabilities of the nodes are calculated based on the sample data and prior probabilities of each indicator. These conditional probabilities are then weighted using the calculated indicator weights to generate a dynamically weighted Bayesian network model. The edges between nodes represent conditional dependencies between variables, which are modeled using conditional probability distributions. Weighting can involve assigning different weights to certain conditional probabilities to reflect the strength or importance of certain dependencies. For example, weights can be used to adjust the degree to which one variable influences another.

[0064] After the prior probability assignment is completed, the time transfer relationship of each indicator itself is determined in the already constructed dynamic weighted Bayesian network model, and the time slice data is merged into a format suitable for learning to complete the construction of the dynamic Bayesian network model.

[0065] The adjustment of prior probabilities is usually done at the beginning of training as part of model initialization. During particle filtering or other training processes, the prior probabilities are gradually updated to posterior probabilities through the likelihood function of the observed data.

[0066] Due to the conditional independence assumption of the Bayesian network, the calculation formula for the posterior probability is as follows:

[0067]

[0068] Among them, P(v|A1, A2, … A j ) is the conditional probability of event v occurring under the known multiple conditions A1, A2, … A j , calculated based on the conditional probabilities of these conditions and the prior probability of event v. P(v) represents the prior probability of event v, that is, the probability of event v occurring when there is no other information. P(A i |v) is the conditional probability of the i-th child node, indicating the probability of condition A i occurring when event v is known to occur. C i is the weight of each index.

[0069] This application can use the particle filter algorithm, the expectation maximization algorithm (EM algorithm), Kalman filtering, variational Bayes, and Markov chain Monte Carlo (MCMC) to train the dynamic Bayesian network model. Preferably, it is the particle filter algorithm.

[0070] By combining the particle filter algorithm, the state distribution is estimated by sampling particles (possible combinations of node states), updating weights (based on evidence), and resampling until the change rate of the posterior probability distribution is lower than the threshold after the prior probabilities of each node have been updated multiple times, and then the training stops. Finally, by inputting the rainfall results predicted by the LSTM model as evidence, the probabilities of the tailings pond being in different risk levels under different future rainfall conditions are obtained.

[0071] The risk assessment of the tailings pond involves complex non-linear relationships (such as geological changes, rainfall impacts) and non-Gaussian distributions (such as extreme event probabilities). The particle filter can effectively handle such problems through Monte Carlo sampling. Moreover, the particle filter is suitable for online learning, can update the state estimate in real time according to new observation data, and is applicable to the dynamic monitoring scenario of the tailings pond. Finally, the particle filter has less dependence on model assumptions and can adapt to different types of dynamic Bayesian network structures and probability distributions.

[0072] The particle filter algorithm uses a set of randomly sampled particles and their weights to approximate and recursively infer the posterior distribution of the hidden state. At each moment, the particles predict new states according to the state transition distribution, adjust their weights based on the current observation, and finally select representative particles through resampling, so as to achieve efficient estimation of the hidden variables in a complex or non-linear dynamic model.

[0073] The sample data in this application is historical data. Taking the node of "seepage amount" as an example: for the "seepage amount" node, since it only depends on the root node of "rainfall amount", the predicted future 24-hour rainfall amount using the LSTM model is used as evidence. Taking the risk level of rainfall amount in one hour as an example, given "rainfall amount == D", based on the conditional probability distribution learned by the model from historical data, the probabilities of each risk level of seepage amount are obtained by the following formula:

[0074]

[0075] where P(Q i = i|R = D) is the conditional probability, which is the probability that the seepage amount is at level i under the condition that the rainfall amount is at level D, and P(Q i = i, R = D) is the joint probability, indicating the probability that the seepage amount Q i is i and the rainfall amount is D at the same time; P(R = D) is the probability of the rainfall amount at risk level D. Through this formula, the probability that the seepage amount is at level i under the condition that the rainfall amount is D can be calculated. In this way, the conditional probability distribution of the seepage amount under four different risk levels can be obtained. Therefore, the change of the risk level probability of the seepage amount in the future 24 hours can be inferred using the predicted future 24-hour rainfall amount by the LSTM model.

[0076] In this application, the LSTM model is used to predict the time series of rainfall amount. The construction of the LSTM model includes the following steps: divide the rainfall amount data into a training set and a test set, and slice it into a sequence format by sliding the window in chronological order; design the LSTM model, including an input layer, an LSTM layer, a fully connected layer, and an output layer, set the hyperparameters and use the training set to train the model, use the mean square error as the loss function, and stop training when the change rate of the loss function is continuously lower than a certain threshold for multiple times. After training, evaluate the model effect on the test set and use evaluation indicators such as the root mean square error to measure the performance. Finally, use the model to predict the future rainfall amount.

[0077] The Long Short-Term Memory (LSTM) network is a recurrent neural network suitable for rainfall forecasting. It uses forget gates, input gates, and output gates to capture long-term and short-term dependencies in time series, overcoming the vanishing gradient problem of traditional RNNs. For rainfall forecasting, the LSTM takes historical rainfall as input, constructs time series samples using a sliding window, and outputs rainfall at a specific moment in the future. The training process includes data normalization, model initialization, forward propagation to calculate predicted values, mean squared error loss calculation, backpropagation to update parameters over time, and multiple rounds of iterative optimization. The LSTM retains long-term information through cell states and conveys short-term memory through hidden states. It combines tanh and sigmoid functions to model nonlinear relationships, ultimately generating accurate forecasts. Evaluation uses metrics such as the mean squared error (MSE) and mean average error (MAE) on the test set, and optimizes performance by adjusting hyperparameters or using early stopping strategies to adapt to the seasonality and complexity of rainfall data.

[0078] In one embodiment, the pollutant infiltration amount is predicted based on rainfall and sample data based on a dynamic Bayesian network model, including: predicting the probability of the infiltration amount being at different risk levels based on rainfall based on the dynamic Bayesian network model; determining the infiltration amount values corresponding to the different risk levels based on the sample data; and generating the pollutant infiltration amount, i.e., the infiltration amount of the tailings pond leachate, based on the probabilities and corresponding infiltration amount values at different risk levels.

[0079] Using an existing Bayesian network model, we reasoned based on predicted rainfall data to determine the probability of infiltration at different risk levels. We then determined the magnitude of infiltration at each risk level based on historical monitoring data. The risk probabilities and the resulting infiltration amounts were substituted into a weighted average formula, where the probabilities of the different risk levels were treated as weights to calculate the predicted infiltration amount. This application uses four risk levels as an example, and the specific formula is as follows:

[0080] Q=R1*q1+R2*q2+R3*q3+R4*q4

[0081] Among them, Q represents the infiltration rate of tailings pond leachate, R1, R2, R3, and R4 are the probabilities of the Bayesian network infiltration rate indicators at different risk levels; q1, q2, q3, and q4 are the average values of the indicator ranges at different risk levels.

[0082] In one embodiment, a pollutant diffusion model is constructed based on the sub-basin boundary information where the tailings pond is located, including: establishing a node set based on the sub-basin boundary; based on the node set, generating a pollutant diffusion model according to the direction and weight of the edges between the water flow direction and flow rate nodes.

[0083] Reference Figure 6 and Figure 7, A node set is established according to the sub - basin demarcation, and an adjacency matrix is generated based on the water flow direction and velocity information. The mathematical properties of the matrix can be used for complex network analysis, where the flow direction is transformed into the direction of the edges of the complex network, and the velocity is the weight of the edges, clearly representing the connection relationships between sub - basins and generating a pollutant diffusion model.

[0084] In one embodiment, after generating the pollutant diffusion model, it further includes: calculating the eigenvector centrality, local clustering coefficient, in - degree, and out - degree of each node according to the pollutant diffusion model.

[0085] Specifically, let the adjacency matrix of the network be A = A ij , for a directed network as follows:

[0086]

[0087] Among them, eigenvector centrality is an index to measure the importance of nodes in the network. It is based on the connectivity between nodes and their neighbor nodes and the importance of neighbor nodes. The calculation formula is as follows:

[0088]

[0089] Among them, EC i is the eigenvector centrality of node i; λ is the largest eigenvalue of the adjacency matrix A; A ij is the element in the i - th row and j - th column of the adjacency matrix A, indicating whether node i is connected to node j; x j is the eigenvector centrality of node j.

[0090] The local clustering coefficient measures the degree of clustering of nodes in the network, that is, the connection ratio between its neighbors. The calculation formula is as follows:

[0091]

[0092] Among them, CC i is the local clustering coefficient of node i; A ij is the element in the i - th row and j - th column of the adjacency matrix A, indicating whether node i is connected to node j; is the total degree of node i, including in - degree and out - degree; A jk is the element in the j - th row and k - th column of the adjacency matrix A, indicating whether node j is connected to node k; A ki is the element in the k - th row and i - th column of the adjacency matrix A, indicating whether node k is connected to node i;

[0093] The in-degree is the number of directed edges pointing to a node; the out-degree is the number of directed edges pointing out from a node.

[0094] In one embodiment, based on the pollutant diffusion model, simulating the pollutant diffusion result according to the pollutant penetration amount includes: for each node, calculating the convective transport amount along each edge, and at the same time, calculating the diffusion transport amount between the node and its adjacent nodes. For each node, accumulate all the inflow amounts and subtract the outflow amount to update the current pollutant concentration. After the update, perform attenuation processing on the concentrations of all nodes. At the end of each time step, multiply the concentration by 0.8 to simulate the attenuation of pollutants due to natural processes such as sedimentation and degradation.

[0095] Specifically, the calculation formula for the initial mixing process is:

[0096]

[0097] where C mix is the pollutant concentration after the leachate and river water are completely mixed at the input point; Q is the penetration amount of the tailings pond leachate; C0 is the initial concentration of pollutants in the leachate; Q river is the flow rate of the river in the basin where the tailings pond is located; C river is the river background concentration (the pollutant concentration before the leachate input).

[0098] The calculation formula for the convective transport process is:

[0099] Q a，ij = k a * u ij * C mix

[0100] where Q a，ij is the convective transport amount from node i to adjacent node j; k a is the convective transfer coefficient; u ij is the water flow velocity from node i to node j.

[0101] The calculation formula for the diffusion transport process is:

[0102] Q d,ij = k d * (C i - C j )

[0103] where Q d,ij is the diffusion transport amount from node i to adjacent node j; k d is the diffusion transfer coefficient; C i - C j is the concentration difference between node i and node j.

[0104] The calculation formula for the concentration update process of nodes based on convective transport and diffusive transport is as follows:

[0105] Q net,i =∑ j∈Nin(i) (Q a,ji +Q d,ji )-∑ j∈Nout(i) (Q a,ij +Q d,ij )

[0106]

[0107] C i (t + Δt)=C i (t)+ΔC i

[0108] Where, Q net,i is the net transport volume (inflow minus outflow) of node i; Nin(i) is the set of adjacent nodes flowing into node i; Nout(i) is the set of adjacent nodes flowing out from node i; Q a,ji is the convective transport volume from node j to node i; Q d,ji is the diffusive transport volume from node j to node i; Q a,ij is the convective transport volume from node i to node j; Q d,ij is the diffusive transport volume from node i to node j; ΔC i is the concentration change of node i within the time step Δt; Δt is the time step of the simulation; V i is the water volume of node i; C i (t) is the pollutant concentration of node i at time t; C i (t + Δt) is the pollutant concentration at the next time step.

[0109] Finally, the formula for the attenuation of the obtained pollutant concentration is:

[0110] C=C i (t + Δt)*0.8

[0111] Where, C is the pollutant concentration of the simulation result; 0.8 is the attenuation factor.

[0112] In one embodiment, the risk status of each sub - basin is evaluated according to the pollutant diffusion result and water ecological sensitivity, including: calculating the difference between the in - degree and out - degree of each node; calculating the comprehensive risk of each node according to the pollutant diffusion result, water ecological sensitivity, eigenvector centrality of each node, local clustering coefficient of each node, and the difference between the in - degree and out - degree of each node; normalizing the numerical value of the comprehensive risk of each node, and dividing the basin corresponding to each node into corresponding risk levels according to the numerical value range of different risks.

[0113] Specifically, referring toFigure 8 and Figure 9 , based on the simulation results of pollutant migration and diffusion in leachate, the comprehensive risk is calculated using the following formula.

[0114] Risk=(C×S)+(w1×ln(1+DC)+w2×EC+w3×CC)

[0115] Wherein, C is the pollutant concentration of the simulation results, S is the water ecological sensitivity, w1, w2, and w3 are their respective weight coefficients (obtained according to natural laws), and DC, EC, and CC are the difference between in-degree and out-degree, eigenvector centrality, and local clustering coefficient in the network structure index, respectively.

[0116] Finally, the calculation results of the comprehensive risk are normalized, and each sub-basin is divided into high-risk, relatively high-risk, medium-low-risk, and low-risk areas according to different numerical ranges. The risk status of each sub-basin in the next 24h is calculated according to the future infiltration prediction results.

[0117] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0118] The above-described embodiments only represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A tailings pond risk assessment method, characterized in that, Including: Determine the tailings pond risk assessment system and preprocess the sample data of each index in the tailings pond risk assessment system; Determine the information of the sub-watershed where the tailings pond is located and the water ecological sensitivity; Calculate the comprehensive weight of each index; Construct a dynamic Bayesian network model based on the tailings pond risk assessment system, the preprocessed sample data of each index, and the comprehensive weight of each index; Predict the pollutant infiltration amount based on the rainfall and sample data using the dynamic Bayesian network model; Construct a pollutant diffusion model according to the information of the sub-watershed where the tailings pond is located; Simulate the pollutant diffusion result based on the pollutant infiltration amount using the pollutant diffusion model; Evaluate the risk status of each sub-watershed based on the pollutant diffusion result and the water ecological sensitivity.

2. The method according to claim 1, wherein Determine the information of the sub-watershed where the tailings pond is located, including: Calculate the water flow direction of each pixel according to the preset DEM data and generate a flow accumulation raster; Extract the information of the sub-watershed where the tailings pond is located according to the flow accumulation result of the flow accumulation raster; among them, the sub-watershed information includes the sub-watershed boundary, the respective flow and flow velocity of each sub-watershed.

3. The method according to claim 2, wherein Determine the water ecological sensitivity, including: weighting the flow and flow velocity of the sub-watershed and the preset water environment functional area data, reclassifying the weighted result to generate the water ecological sensitivity, and normalizing the water ecological sensitivity value.

4. The method according to claim 1, wherein Calculate the comprehensive weight of each index, including: Determine the subjective weight of each index based on the analytic hierarchy process; Calculate the objective weight of each index based on the entropy method; Calculate the comprehensive weight of each index according to the subjective weight and the objective weight.

5. The method according to claim 1, characterized in that, Construct a dynamic Bayesian network model based on the tailings pond risk assessment system, the preprocessed sample data of each index, and the comprehensive weight of each index, including: Construct a static Bayesian network model according to the tailings pond risk assessment system; Assign the prior probability of each node index in the static Bayesian network model according to the comprehensive weight of each index; Determine the time transfer relationship of each index itself and merge the time slice data into a format suitable for learning to generate an initial dynamic Bayesian network model; Train the initial dynamic Bayesian network model according to the sample data of each index to generate a target dynamic Bayesian network model.

6. The method according to claim 1, characterized in that, Predict the pollutant infiltration amount based on the rainfall and sample data using the dynamic Bayesian network model, including: Predict the probability of the infiltration amount being in different risk levels based on the rainfall using the dynamic Bayesian network model; Determine the infiltration amount value corresponding to different risk levels according to the sample data; Generate the pollutant infiltration amount according to the probability and the corresponding infiltration amount value under different risk levels.

7. The method according to claim 3, wherein Construct a pollutant diffusion model according to the sub-watershed boundary information of the tailings pond, including: Establish a node set according to the sub-watershed boundary; Generate a pollutant diffusion model based on the node set according to the direction and weight of the edges between the water flow direction and flow velocity nodes.

8. The method according to claim 7, wherein After generating the pollutant diffusion model, it also includes: Calculate the eigenvector centrality, local clustering coefficient, in-degree, and out-degree of each node according to the pollutant diffusion model.

9. The method according to claim 8, wherein Evaluate the risk status of each sub-watershed based on the pollutant diffusion result and the water ecological sensitivity, including: Calculate the difference between the in-degree and out-degree of each node; Calculate the comprehensive risk of each node according to the pollutant diffusion result, water ecological sensitivity, eigenvector centrality of each node, local clustering coefficient of each node, and the difference between the in-degree and out-degree of each node; Normalize the numerical values of the comprehensive risk of each node, and divide the corresponding basins of each node into corresponding risk levels according to the numerical value ranges of different risks.

Images