CN-122022491-A - Tailings dam risk assessment method and system based on LSTM safety coefficient prediction

CN122022491ACN 122022491 ACN122022491 ACN 122022491ACN-122022491-A

Abstract

The application discloses a tailing dam risk assessment method and system based on LSTM safety coefficient prediction, and relates to the field of rock-soil safety monitoring; and extracting derivative features through feature engineering, and constructing a composite feature sequence comprising static features and dynamic time features by combining normalized time and attenuation coefficients. And then constructing an LSTM model, taking the composite characteristic sequence of the history dam environment as input data, and taking the history safety coefficient sequence as output data for model training. And inputting the current dam environment composite characteristic sequence into a trained model, predicting a safety coefficient sequence, calculating a rainfall accumulation effect according to rainfall intensity and rainfall duration, and carrying out comprehensive risk assessment on the tailing dam. The application enhances the representativeness and accuracy of the prediction of the safety coefficient of the tailing dam, and utilizes the rainfall accumulation effect value to carry out risk judgment, thereby further improving the accuracy of the risk assessment of the tailing dam.

Inventors

ZHANG XIAOBO
WANG GUODONG
MA YONGLI
YAO CHI
CHENG JUN
YANG JIANHUA
YE ZHIWEI

Assignees

南昌大学

Dates

Publication Date: 20260512
Application Date: 20260323

Claims (10)

1. A tailings dam risk assessment method based on LSTM security coefficient prediction, the method comprising: Acquiring historical dam environment data and corresponding historical safety coefficient data of a continuous time point of a target tailing dam, and respectively obtaining a historical dam environment sequence and a historical safety coefficient sequence by constructing a time sequence, wherein the historical dam environment data comprises dam height, dam slope ratio, cohesive force, internal friction angle, permeability coefficient, reservoir water level, rainfall intensity and rainfall duration; extracting derivative features of samples in a historical dam environment sequence one by utilizing feature engineering to obtain historical derivative features, normalization time and attenuation coefficients of the samples, combining the historical derivative features with features in the historical dam environment sequence to serve as static features, and taking the normalization time and the attenuation coefficients as dynamic time features to construct a historical dam environment composite feature sequence, wherein the historical derivative features comprise dam high slope ratio features, shear strength parameter features, infiltration logarithmic features, rainfall dam height ratio features, reservoir water level dam height ratio features and dam morphology features; Constructing an LSTM model, taking a composite characteristic sequence of the history dam environment as input data, taking a history safety coefficient sequence as output data, taking the minimum loss function as an optimization target, and training the LSTM model to obtain a trained LSTM model; Acquiring a current dam environment composite characteristic sequence of a target tailing dam, and inputting the current dam environment composite characteristic sequence into a trained LSTM model to obtain a predicted safety coefficient sequence; Calculating a rainfall accumulation effect value by using rainfall intensity and rainfall duration of a preset time point, performing risk assessment on the target tailing dam based on the rainfall accumulation effect value and a predicted safety coefficient sequence, obtaining a risk assessment result of the target tailing dam, and generating a corresponding risk assessment report.
2. The tailing dam risk assessment method based on LSTM security coefficient prediction according to claim 1, wherein after obtaining the historical dam environment data and the corresponding historical security coefficient data of the continuous time points of the target tailing dam, and respectively obtaining the historical dam environment sequence and the historical security coefficient sequence by constructing the time sequence, further comprises: And respectively preprocessing the historical dam environment data and the historical safety coefficient data to update the historical dam environment data and the historical safety coefficient data, wherein the preprocessing at least comprises cleaning, outlier rejection and normalization processing.
3. The tailings dam risk assessment method based on LSTM security coefficient prediction of claim 1, wherein the dam high slope ratio feature is calculated according to the following formula: ; the calculation formula of the shear strength parameter characteristics is as follows: ; The calculation formula of the infiltration logarithmic characteristic is as follows: ; the calculation formula of the rainfall dam height ratio characteristic is as follows: ; the calculation formula of the height ratio characteristic of the reservoir water level dam is as follows: ; the calculation formula of the morphological characteristics of the dam body is as follows: ; In the formula, Is a dam high slope ratio characteristic; Is a shear strength parameter characteristic; is characterized by the infiltration logarithm; the characteristics of the rainfall dam height ratio are that; The dam height ratio characteristics of the reservoir water level are; Is a morphological feature of a dam body; Is a dam height; Is a dam slope ratio; Is cohesive force; Is an internal friction angle; is the permeability coefficient; is the reservoir water level; Is the intensity of rainfall.
4. The tailing dam risk assessment method based on LSTM safety coefficient prediction according to claim 1, wherein the method is characterized by constructing an LSTM model, taking a history dam environment composite characteristic sequence as input data, taking a history safety coefficient sequence as output data, taking a minimum loss function as an optimization target, training the LSTM model, and obtaining a trained LSTM model, and specifically comprises the following steps: Constructing an LSTM model; Taking the historical dam environment composite characteristic sequence as input data, taking the historical safety coefficient sequence as output data, and dividing a training set and a verification set according to a preset proportion; setting an initial learning rate, a learning rate attenuation coefficient, a maximum training round and an early-stop endurance value of an Adam optimizer in an LSTM model, and training the LSTM model; and selecting optimal parameters of the LSTM model by using the verification set, and taking the model with the minimum loss function value corresponding to the safety coefficient prediction error as the trained LSTM model.
5. The tailing dam risk assessment method based on LSTM safety coefficient prediction according to claim 1 is characterized in that a network architecture of an LSTM model comprises a sequence input layer, a first LSTM layer, a first Dropout layer, a second LSTM layer, a second Dropout layer, a full connection layer and a regression output layer which are sequentially connected, wherein the output mode of the first LSTM layer is set to be a sequence mode and is used for outputting sequence characteristics to the second LSTM layer, the Dropout probability of the first Dropout layer is set to be 0.2-0.4, the Dropout probability of the second Dropout layer is set to be 0.1-0.3, and the Dropout probability is used for inhibiting model overfitting.
6. The method for evaluating the risk of a tailings dam based on LSTM safety coefficient prediction according to claim 1, wherein the method for evaluating the risk of the target tailings dam based on the rainfall intensity and the rainfall duration at the preset time point calculates a rainfall accumulation effect value, and performs risk evaluation on the target tailings dam based on the rainfall accumulation effect value and the predicted safety coefficient sequence, so as to obtain a risk evaluation result of the target tailings dam, and generates a corresponding risk evaluation report, and specifically comprises the steps of: Calculating a rainfall accumulation effect value by using rainfall intensity and rainfall duration of a preset time point; predicting safety coefficients corresponding to preset time points in the safety coefficient sequence based on the rainfall accumulation effect value index; Calculating according to the rainfall accumulation effect value to obtain prediction uncertainty, and carrying out weighted calculation on the safety coefficient corresponding to the preset time point according to the prediction uncertainty to obtain a weighted safety coefficient; Judging a risk assessment result based on the weighted safety coefficient, wherein the risk assessment result comprises high risk, medium risk and low risk; And generating a risk assessment report corresponding to the risk assessment result.
7. The tailing dam risk assessment method based on LSTM safety coefficient prediction according to claim 6, wherein the risk assessment result is determined based on the weighted safety coefficient, specifically comprising: When the weighted safety coefficient is smaller than the first safety coefficient threshold value, judging that the risk assessment result is high risk; When the pre-weighted safety coefficient is larger than or equal to a first safety coefficient threshold value and smaller than a second safety coefficient threshold value, judging that the risk assessment result is a risk, wherein the first safety coefficient threshold value is smaller than the second safety coefficient threshold value; And when the weighted safety coefficient is greater than or equal to the second safety coefficient threshold value, judging that the risk assessment result is low risk.
8. The tailings dam risk assessment method based on LSTM security coefficient prediction of claim 6 wherein the calculation formula of the prediction uncertainty is as follows: ; In the formula, To predict uncertainty values; base E [0.02,0.05] for a preset base uncertainty value; Accumulating effect values for rainfall; the calculation formula of the weighted safety coefficient is as follows: ; In the formula, Is a weighted safety coefficient; is a safety coefficient; the effect value is accumulated for rainfall.
9. The tailing dam risk assessment method based on LSTM safety factor prediction according to claim 1, wherein the rainfall accumulation effect value is calculated as follows: ; In the formula, Accumulating effect values for rainfall; is rainfall intensity; Is the duration of rainfall.
10. A computer system comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the LSTM security coefficient prediction-based tailing dam risk assessment method of any one of claims 1-9.

Description

Tailings dam risk assessment method and system based on LSTM safety coefficient prediction Technical Field The application relates to the field of rock and soil safety monitoring, in particular to a tailing dam risk assessment method and system based on LSTM safety coefficient prediction. Background The tailings dam is a core structure for storing tailings in the mining production process, the stability of the tailings dam is directly related to the life and property safety of mine personnel, the ecological environment safety and the continuity of mining production, the safety coefficient is a core quantitative index for evaluating the stability of the tailings dam, and the accurate and real-time prediction of the safety coefficient is a key for the safety management of the tailings dam. At present, the tailing dam safety coefficient prediction mainly depends on two traditional methods, namely, a numerical simulation method (a finite element method and a limit balance method) based on a physical mechanism is required to depend on an accurate rock-soil constitutive model and parameter setting, the calculation flow is complex and time-consuming and lengthy, real-time prediction requirements are difficult to meet, and more importantly, static assumption is adopted in the method for processing variable loads such as rainfall and the like, the influence of rainfall accumulation time-varying effect and pore water pressure dynamic change on the safety coefficient cannot be quantized, and the prediction precision is limited. Secondly, although the statistical empirical formula method is simple and convenient to calculate, the accuracy is low, the generalization capability is poor, the formula is mainly derived from data fitting under specific working conditions, complex nonlinear interaction among multi-source factors such as dam parameters, hydrologic conditions and the like cannot be captured, and the safety coefficient at a single time point can only be output, so that the sequential prediction of the safety coefficient along with the time evolution track cannot be realized. Under some conditions, the machine learning method is gradually applied to the field, but the existing application mostly adopts a static neural network or a support vector machine, so that single-point prediction of the safety coefficient under a single working condition can be realized, and the time sequence evolution characteristic of the safety coefficient can not be fully utilized. In summary, the existing tailing dam risk assessment has the problems that the safety coefficient prediction representativeness is poor, the accuracy is low, and the dynamic influence of the rainfall accumulation effect on the dam stability is not fully considered, so that the assessment result is inaccurate. Disclosure of Invention The application aims to provide a tailing dam risk assessment method and system based on LSTM safety coefficient prediction, which can enhance the representativeness and accuracy of the tailing dam safety coefficient prediction and improve the accuracy of the tailing dam risk assessment. In order to achieve the above object, the present application provides the following solutions: In a first aspect, the application provides a tailing dam risk assessment method based on LSTM safety coefficient prediction, which comprises the steps of obtaining historical dam environment data and corresponding historical safety coefficient data of a target tailing dam at continuous time points, and respectively obtaining a historical dam environment sequence and a historical safety coefficient sequence by constructing a time sequence; the historical dam environment data comprises dam height, dam slope ratio, cohesive force, internal friction angle, permeability coefficient, reservoir water level, rainfall intensity and rainfall duration, wherein derivative characteristics are extracted from samples in a historical dam environment sequence one by utilizing characteristic engineering to obtain historical derivative characteristics, normalized time and attenuation coefficient of the samples, the historical derivative characteristics and the characteristics in the historical dam environment sequence are combined to be used as static characteristics, the normalized time and the attenuation coefficient are used as dynamic time characteristics to construct a historical dam environment composite characteristic sequence, the historical derivative characteristics comprise dam height slope ratio characteristics, shear strength parameter characteristics, permeability logarithmic characteristics, rainfall dam height ratio characteristics, reservoir water level dam height ratio characteristics and dam morphology characteristics, an LSTM model is constructed, the historical dam environment composite characteristic sequence is used as input data, the historical safety coefficient sequence is used as output data, the loss function is minimum to be used as an optimi