Method for identifying abnormal tailings dam seepage monitoring data based on grey model and neural network

Abstract

This invention discloses a method for identifying anomalies in tailings dam seepage monitoring data based on a grey model and neural network. First, monitoring data is acquired, preprocessed using the arithmetic square root, and generated by a single accumulation. Then, a GM(1,1) grey model is established, and a dynamic adaptive threshold interval is constructed to coarsely screen anomaly data. Data is updated using a metabolic process; data exceeding the threshold is considered anomaly. Next, the grey correlation degree between the anomaly data and related measuring point data is calculated, and the input variable for the radial basis function neural network is selected accordingly. The network is trained to obtain the neural network's seepage prediction value, and the relative error is used to determine whether the anomaly is caused by environmental factors. Unidentified anomalies are retested and verified, and instruments are checked. If the retest is normal, the measurement error data is removed or corrected; if the retest is abnormal, it is determined to be a structural change and an early warning is issued. This invention combines a grey model and a neural network to achieve efficient and accurate identification of anomaly data and automatic identification of environmental factor responses.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy engineering safety management technology, and in particular relates to a method for identifying anomalies in tailings dam seepage monitoring data based on grey models and neural networks. Background Technology

[0002] Tailings dams are crucial facilities for discharging tailings during mineral processing. They are typically formed by damming valleys or enclosing land. As a key structure, the safety and stability of tailings dams directly impact the safety of life and property downstream and the ecological environment. Tailings dams are generally designed with permeable bodies, and the phreatic line of the dam body is considered the "lifeline" of the tailings dam; its depth changes directly reflect the state of the seepage field. In recent years, with the increasing national requirements for safe production and environmental protection, the accuracy and real-time nature of tailings dam seepage monitoring data have received increasing attention. Health diagnosis of the seepage field based on phreatic line depth monitoring data has become an important means of ensuring the safe operation of tailings dam projects.

[0003] In the identification of anomalies in existing tailings dam seepage monitoring data, scholars have introduced various mathematical theories and methods, among which regression models are the most widely used. These models identify anomalous measurements in the data by establishing statistical relationships between environmental quantities (such as reservoir water level and rainfall) and effect quantities (such as the depth of the seepage line). For example, robust regression methods based on M-estimation are used to determine outliers in the data sequence; and BC(a) confidence intervals are introduced based on statistical models to identify cracks in concrete dams. These methods typically require the construction of high-precision fitting models and rely on a relatively complete monitoring data system to achieve preliminary identification of anomalous data.

[0004] However, existing methods have significant shortcomings in practical applications. On the one hand, the investment in safety supervision of tailings dams is generally lower than that of reservoir dams, monitoring projects are incomplete, and the level of automation is low. In addition, the complex and variable seepage field leads to low fitting accuracy of the constructed statistical regression models, making it difficult to accurately identify abnormal data. On the other hand, for the identified abnormal data, it is necessary to further determine whether the cause is monitoring error, environmental change, or structural deterioration. Currently, this relies heavily on manual investigation and experience-based correction, resulting in low identification efficiency and poor timeliness. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide a method for identifying anomalies in tailings dam seepage monitoring data based on grey models and neural networks, which solves the problems of insufficient fitting accuracy and difficulty in efficiently distinguishing the causes of numerical anomalies in existing statistical regression models, and realizes efficient and accurate identification of abnormal data and automatic identification of environmental response.

[0006] Technical solution: The anomaly identification method for tailings dam seepage monitoring data based on grey model and neural network as described in this invention includes the following steps:

[0007] S1. Acquire tailings dam seepage monitoring data, perform arithmetic square root preprocessing on the seepage monitoring data to obtain a preprocessed seepage monitoring data sequence, and perform a first-time accumulation generation on the preprocessed seepage monitoring data sequence to obtain a first-time accumulation generation sequence; establish a GM(1,1) grey prediction model based on the first-time accumulation generation sequence, solve the grey differential equation parameters, and obtain the grey seepage prediction value at the corresponding monitoring time; construct an anomaly identification threshold interval based on the deviation between the grey seepage prediction value and the acquired seepage monitoring data, and update the preprocessed seepage monitoring data sequence using a metabolic update method; identify seepage monitoring data whose deviation exceeds the anomaly identification threshold interval as abnormal monitoring data, and identify the monitoring points corresponding to the abnormal monitoring data as abnormal monitoring points;

[0008] S2. Obtain monitoring data of relevant measuring points corresponding to the abnormal monitoring points, and perform mean transformation on the abnormal monitoring data and the monitoring data of relevant measuring points. After transformation, calculate the gray correlation degree between the abnormal monitoring points and the relevant measuring points, and select target measuring point data as the input variable of the neural network based on the gray correlation degree. Construct a radial basis function neural network model, and input the abnormal monitoring data and the target measuring point data into the radial basis function neural network model for training to obtain the neural network seepage prediction value corresponding to the abnormal monitoring points. Based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, identify the environmental quantity response of the abnormal monitoring data.

[0009] S3. Retest and verify the abnormal monitoring data that were not identified as environmental quantity responses in step S2, and check the monitoring instruments. When the retest results are normal, the abnormal monitoring data is identified as measurement error data, and the original monitoring data corresponding to the abnormal monitoring data is removed or corrected. When the retest results are abnormal, the abnormal monitoring data is identified as tailings dam structural change data, and the corresponding abnormal warning information is output.

[0010] The technical solution proposed in this invention, by introducing a grey model to construct a dynamic prediction interval, can alleviate the problem of insufficient fitting accuracy of traditional statistical regression models for nonlinear and small-sample monitoring data, and achieve preliminary and efficient identification of abnormal monitoring data. Furthermore, by combining grey relational analysis of strongly correlated measurement point data with a radial basis function neural network model, it helps to automatically distinguish whether abnormal data is caused by changes in environmental quantities, improving the ability to identify the causes of numerical anomalies. Finally, through retesting and verification and instrument checks, the interference of measurement error data can be reduced, and data confirmed as structural changes can be used for anomaly early warning. Therefore, for tailings dam seepage monitoring with incomplete monitoring projects and a lack of environmental monitoring values, the overall accuracy and practicality of anomaly identification of measurement data are enhanced.

[0011] Preferably, step S1, which involves performing arithmetic square root preprocessing on the seepage monitoring data to obtain a preprocessed seepage monitoring data sequence, and then performing a single accumulation on the preprocessed seepage monitoring data sequence to generate the following:

[0012] Set the original seepage monitoring data sequence ,in Here, n is the sequence number of the monitoring time, and n is the sequence length. This represents the raw seepage monitoring data at time k; the raw seepage monitoring data is preprocessed using the arithmetic square root to obtain the preprocessed seepage monitoring data sequence. ,in ; then Perform one accumulation generation to obtain an accumulation generation sequence. Where j is the index variable in the summation, This represents the cumulative sum of the preprocessed data over the first k time steps.

[0013] By performing arithmetic square root preprocessing and single-step accumulation on the original seepage monitoring data, this step helps to weaken the fluctuations and noise interference that may exist in the original data, making the generated new sequence more suitable for grey model modeling. The introduction of accumulation generation can enhance the intrinsic trend information of the sequence, thereby providing a more favorable data foundation for the stable solution of the subsequent GM(1,1) model parameters, and to a certain extent improving the ability of the prediction model to characterize the seepage change law.

[0014] Preferably, step S1, which involves establishing a GM(1,1) grey prediction model, solving the grey differential equation parameters, obtaining the grey seepage prediction value at the corresponding monitoring time, and updating the preprocessed seepage monitoring data sequence using a metabolic update method, includes:

[0015] Generate a sequence by accumulating once. Treating it as a function of time, we establish a first-order differential equation:

[0016]

[0017] Where a is the development coefficient, b is the gray action quantity, and t is the time variable; let the vector to be identified be... The parameter estimates of a and b are obtained by solving using the least squares method. and ;

[0018] Therefore, the time response formula of the GM(1,1) grey prediction model is established:

[0019]

[0020] in, This is the first value of the preprocessed data. It is a natural constant. The current time number. This is the cumulative predicted value at time k+1; restoring the above predicted value yields the predicted value within the preprocessing domain:

[0021] (k=2,3,…,n)

[0022] in, This is the actual accumulated value at time k. The predicted value of the preprocessed data at time k+1 is given; the predicted value of the preprocessed data is then subjected to inverse preprocessing to obtain the predicted value of gray seepage under the original dimensions:

[0023]

[0024] in, This is the predicted gray seepage value at time k+1; using a metabolic update method, the above calculated values ​​are... The latest data is added to the original sequence, while the oldest data in the original sequence is removed. This forms a new, equal-length original data sequence, which is used for dynamic prediction at the next time step.

[0025] Solving the parameters of the grey differential equation using the least squares method and establishing the time response formula enables dynamic prediction of seepage monitoring data, while restoring it to the original dimensions for practical engineering applications. Employing a metabolic update method to incorporate the latest predicted values ​​into the sequence and remove the oldest data helps the model continuously reflect the changing trend of the current seepage state, thereby suppressing the accumulation of prediction errors to a certain extent and enhancing the model's adaptability to abnormal fluctuations in monitoring data.

[0026] Preferably, step S1, which involves constructing an anomaly identification threshold range based on the deviation between the gray seepage prediction value and the acquired seepage monitoring data, includes:

[0027] For each monitoring time point, calculate the predicted value of gray seepage. The relative deviation between the data and the seepage monitoring data at that moment was calculated, and the deviation distribution of all historical normal moments was statistically analyzed to obtain the mean μ and standard deviation σ of the deviation. Based on the prediction accuracy requirements of the grey model, a threshold range for anomaly identification was set. λ is a preset confidence coefficient; the current seepage monitoring data is compared with the gray seepage prediction value: if the deviation falls within the threshold interval T, the monitoring data is determined to be normal; if the deviation exceeds the threshold interval, the monitoring data is determined to be abnormal monitoring data; the threshold interval T is recalculated with the addition of new data using the metabolically updated sequence to achieve dynamic adaptive adjustment.

[0028] By constructing an anomaly identification threshold range based on the deviation distribution between predicted and actual values, and introducing a confidence coefficient for adjustment, this step can provide a dynamic basis for anomaly judgment in seepage monitoring data. At the same time, the threshold range is recalculated with the updated sequence after metabolism, which helps to adapt the discrimination criteria to the changing characteristics of the data sequence, thereby improving the adaptability and reliability of anomaly monitoring data identification to a certain extent.

[0029] Preferably, step S2, which involves performing mean transformation on the abnormal monitoring data and related measuring point monitoring data, and then calculating the grey correlation degree between the abnormal monitoring points and related measuring points after the transformation, includes:

[0030] Let the monitoring data sequence of the measuring point where the abnormal monitoring data is located be the reference sequence. Where k is the time number and n is the sequence length. This represents the raw seepage monitoring data at time k; there are m related measurement point data sequences as comparison sequences. ,in , Let represent the observed value of the i-th relevant measuring point at time k; perform mean transformation on the reference sequence and each comparison sequence respectively, using the following transformation formula:

[0031]

[0032] in The denominator is the k-th value in the current sequence being transformed, i.e., the reference sequence or a comparison sequence. The arithmetic mean of the entire sequence. The transformed dimensionless values ​​are used; after transformation, the correlation coefficient between each comparison series and the reference series at each time step is calculated using the following formula. :

[0033]

[0034] in: This represents the absolute difference between the reference sequence and the i-th comparison sequence at time k; It represents the minimum absolute difference between all compared sequences and the reference sequence at all times; The maximum absolute difference between all compared sequences and the reference sequence at all times; The resolution coefficient is used to calculate the correlation between each comparison sequence and the reference sequence using the following formula. :

[0035]

[0036] in, It reflects the overall correlation between the data of the i-th measurement point and the data of the abnormal monitoring point. According to the size of the gray correlation degree, the measurement point data ranked in the top N by gray correlation degree are selected as the target measurement point data input to the radial basis function neural network.

[0037] By performing mean transformation on the abnormal monitoring data and related measurement point data, this step can convert data with different dimensions into a dimensionless form, which is convenient for subsequent correlation analysis. On this basis, the grey correlation degree between each related measurement point and the monitoring data is calculated, which helps to select variables with a high degree of correlation from multiple related measurement points as input to the neural network. This reduces the redundancy of input variables to a certain extent and improves the ability of the subsequent radial basis function neural network model to represent the response of environmental quantities.

[0038] Preferably, step S2, which involves constructing a radial basis function neural network model and inputting the anomaly monitoring data and target measurement point data into the radial basis function neural network model for training, includes:

[0039] The radial basis function neural network adopts a three-layer feedforward network structure, including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined according to the total number of monitoring indicators selected by grey relational analysis in step S2, and the seepage monitoring data of the measurement point where the abnormal monitoring data is located and the surrounding target measurement points are used as the input vector. The number of nodes in the hidden layer is determined by trial and error or cross-validation. Each hidden layer node corresponds to a radial basis function, and its center vector and expansion constant are the parameters to be trained. The number of nodes in the output layer is set to 1, and the output value is the neural network seepage prediction value of the measurement point where the abnormal monitoring data is located.

[0040] The network parameters are initialized as follows: a portion of the input samples are randomly selected as the center vectors of the hidden layer nodes, or the input data is clustered using a clustering algorithm, and the cluster centers are used as the initial values ​​of the center vectors; the expansion constant is set to a certain proportion of the average distance between the input data samples, according to the distribution of the input data.

[0041] Model training and dataset partitioning are as follows: the preprocessed dataset is divided into training set, validation set and test set according to a preset ratio; the training set is used for parameter learning of the neural network; the validation set is used to evaluate the model performance during training to prevent overfitting; the test set is used to perform final performance testing on the trained model; after training is completed, the neural network seepage prediction value of the measurement point where the abnormal monitoring data is located is obtained using the test set.

[0042] By constructing a three-layer radial basis function neural network and determining the input layer nodes based on the index selected according to grey relational analysis, this step can organically combine anomaly monitoring data with relevant measurement point data to form a targeted seepage prediction model. Using trial-and-error or cross-validation to determine the number of hidden layer nodes, and combining clustering or sample selection methods for center vector initialization, helps improve the model's adaptability and training efficiency. Simultaneously, dividing the dataset into training, validation, and test sets allows for monitoring of model performance during training, helping to alleviate overfitting and thus enhancing the radial basis function neural network's ability to generalize to neural network seepage prediction values.

[0043] Preferably, step S2, which involves identifying the environmental response of the abnormal monitoring data based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, includes:

[0044] The neural network seepage prediction value output by the radial basis function neural network is compared with the corresponding actual value of the abnormal monitoring data, and the relative error between the two is calculated. A relative error threshold is set, which is predetermined based on engineering experience or statistical methods. If the relative error is less than or equal to the relative error threshold, the abnormal monitoring data is determined to be a normal response caused by changes in environmental quantities, i.e., it is identified as an environmental quantity response. If the relative error is greater than the relative error threshold, the abnormal monitoring data is determined not to be an environmental quantity response, and step S3 is required for retesting and verification and monitoring instrument inspection.

[0045] By comparing the neural network seepage predictions output by the radial basis function neural network with the actual values ​​of abnormal monitoring data and calculating the relative error, this step can quantitatively distinguish the response of environmental quantities based on a set relative error threshold. If the relative error is within the threshold range, the abnormal monitoring data is attributed to a normal response caused by changes in environmental quantities; otherwise, it is transferred to the subsequent retesting and verification stage. This mechanism helps to further distinguish between the influence of environmental quantities and non-environmental quantities in abnormal data, thereby improving the precision of identifying the causes of abnormalities in seepage monitoring data.

[0046] Secondly, the tailings dam seepage monitoring data anomaly identification system based on grey model and neural network described in this invention includes:

[0047] The data acquisition module is used to acquire tailings dam seepage monitoring data;

[0048] The gray model anomaly identification module is used to preprocess the seepage monitoring data using the arithmetic square root to obtain a preprocessed seepage monitoring data sequence, and to perform a single accumulation generation on the preprocessed seepage monitoring data sequence to obtain a single accumulation generation sequence; based on the single accumulation generation sequence, a GM(1,1) gray prediction model is established, and the parameters of the gray differential equation are solved to obtain the gray seepage prediction value at the corresponding monitoring time; anomaly identification threshold interval is constructed according to the deviation between the gray seepage prediction value and the obtained seepage monitoring data, and the preprocessed seepage monitoring data sequence is updated using a metabolic update method; seepage monitoring data whose deviation exceeds the anomaly identification threshold interval is identified as abnormal monitoring data, and the monitoring point corresponding to the abnormal monitoring data is identified as an abnormal monitoring point;

[0049] An environmental quantity response identification module is used to acquire monitoring data of relevant measuring points corresponding to the abnormal monitoring points, and to perform mean transformation on the abnormal monitoring data and the monitoring data of relevant measuring points. After transformation, the gray correlation degree between the abnormal monitoring points and the relevant measuring points is calculated. Target measuring point data is selected as the input variable of the neural network based on the gray correlation degree. A radial basis function neural network model is constructed, and the abnormal monitoring data and the target measuring point data are input into the radial basis function neural network model for training to obtain the neural network seepage prediction value corresponding to the abnormal monitoring points. Based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, environmental quantity response identification is performed on the abnormal monitoring data.

[0050] The result determination and processing module is used to retest and verify abnormal monitoring data that are not identified as environmental quantity responses in the environmental quantity response identification module and to check the monitoring instruments. When the retest result is normal, the abnormal monitoring data is identified as measurement error data, and the original monitoring data corresponding to the abnormal monitoring data is removed or corrected. When the retest result is abnormal, the abnormal monitoring data is identified as tailings dam structural change data, and the corresponding abnormal warning information is output.

[0051] Thirdly, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed by the processor to identify anomalies in tailings dam seepage monitoring data based on gray models and neural networks.

[0052] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for identifying anomalies in tailings dam seepage monitoring data based on gray models and neural networks.

[0053] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: 1. This invention establishes a dynamic adaptive threshold range using the GM(1,1) grey model to achieve coarse screening of abnormal data, and then combines grey relational degree and radial basis function neural network to perform fine judgment on environmental response, thereby achieving efficient and accurate identification of abnormal data and automatic identification of environmental response; 2. This invention forms a complete closed-loop data processing flow from "abnormal identification (coarse screening)" to "cause identification (fine judgment)" and then to "final decision-making". Through a two-step method, a large amount of "false anomaly" data that may be environmental response is retained as normal data, effectively reducing the false alarm rate; 3. This invention can automatically distinguish between measurement error data, environmental response data and tailings dam structural change data, freeing technicians from the tedious data screening and allowing them to accurately focus on truly noteworthy risk signals, providing reliable data support for tailings dam operation management and risk prevention and control. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0055] Figure 2 This is a topographical overview of the LS tailings dam according to the present invention;

[0056] Figure 3 This is a schematic diagram illustrating the identification results of environmental quantity responses of some abnormal data points at the K2 measuring point in this invention. Detailed Implementation

[0057] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0058] This invention provides a method for identifying anomalies in tailings dam seepage monitoring data based on a grey model and neural network, such as... Figure 1 As shown, it includes the following steps:

[0059] S1. Acquire tailings dam seepage monitoring data, perform arithmetic square root preprocessing on the seepage monitoring data to obtain a preprocessed seepage monitoring data sequence, and perform a first-time accumulation generation on the preprocessed seepage monitoring data sequence to obtain a first-time accumulation generation sequence; establish a GM(1,1) grey prediction model based on the first-time accumulation generation sequence, solve the grey differential equation parameters, and obtain the grey seepage prediction value at the corresponding monitoring time; construct an anomaly identification threshold interval based on the deviation between the grey seepage prediction value and the acquired seepage monitoring data, and update the preprocessed seepage monitoring data sequence using a metabolic update method; identify seepage monitoring data whose deviation exceeds the anomaly identification threshold interval as abnormal monitoring data, and identify the monitoring points corresponding to the abnormal monitoring data as abnormal monitoring points;

[0060] S2. Obtain monitoring data of relevant measuring points corresponding to the abnormal monitoring points, and perform mean transformation on the abnormal monitoring data and the monitoring data of relevant measuring points. After transformation, calculate the gray correlation degree between the abnormal monitoring points and the relevant measuring points, and select target measuring point data as the input variable of the neural network based on the gray correlation degree. Construct a radial basis function neural network model, and input the abnormal monitoring data and the target measuring point data into the radial basis function neural network model for training to obtain the neural network seepage prediction value corresponding to the abnormal monitoring points. Based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, identify the environmental quantity response of the abnormal monitoring data.

[0061] S3. Retest and verify the abnormal monitoring data that were not identified as environmental quantity responses in step S2, and check the monitoring instruments. When the retest results are normal, the abnormal monitoring data is identified as measurement error data, and the original monitoring data corresponding to the abnormal monitoring data is removed or corrected. When the retest results are abnormal, the abnormal monitoring data is identified as tailings dam structural change data, and the corresponding abnormal warning information is output.

[0062] Furthermore, step S1, which involves performing arithmetic square root preprocessing on the seepage monitoring data to obtain a preprocessed seepage monitoring data sequence, and then performing an accumulation operation on the preprocessed seepage monitoring data sequence to generate the following:

[0063] Set the original seepage monitoring data sequence ,in Here, n is the sequence number of the monitoring time, and n is the sequence length. This represents the raw seepage monitoring data at time k; the raw seepage monitoring data is preprocessed using the arithmetic square root to obtain the preprocessed seepage monitoring data sequence. ,in ; then Perform one accumulation generation to obtain an accumulation generation sequence. Where j is the index variable in the summation, This represents the cumulative sum of the preprocessed data over the first k time steps.

[0064] Furthermore, step S1, which involves establishing a GM(1,1) grey prediction model, solving the grey differential equation parameters, obtaining the grey seepage prediction value at the corresponding monitoring time, and updating the preprocessed seepage monitoring data sequence using a metabolic update method, includes:

[0065] Generate a sequence by accumulating once. Treating it as a function of time, we establish a first-order differential equation:

[0066]

[0067] Where a is the development coefficient, b is the gray action quantity, and t is the time variable; let the vector to be identified be... The parameter estimates of a and b are obtained by solving using the least squares method. and ;

[0068] Therefore, the time response formula of the GM(1,1) grey prediction model is established:

[0069]

[0070] in, This is the first value of the preprocessed data. It is a natural constant. The current time number. This is the cumulative predicted value at time k+1; restoring the above predicted value yields the predicted value within the preprocessing domain:

[0071] (k=2,3,…,n)

[0072] in, This is the actual accumulated value at time k. The predicted value of the preprocessed data at time k+1 is given; the predicted value of the preprocessed data is then subjected to inverse preprocessing to obtain the predicted value of gray seepage under the original dimensions:

[0073]

[0074] in, The gray seepage prediction value at time k+1 is used; the above calculated value is updated using a metabolic update method. The latest data is added to the original sequence, while the oldest data in the original sequence is removed. This forms a new, equal-length original data sequence, which is used for dynamic prediction at the next time step.

[0075] Furthermore, step S1, which involves constructing an anomaly identification threshold range based on the deviation between the gray seepage prediction value and the acquired seepage monitoring data, includes:

[0076] For each monitoring time point, calculate the predicted value of gray seepage. The relative deviation between the data and the seepage monitoring data at that moment was calculated, and the deviation distribution of all historical normal moments was statistically analyzed to obtain the mean μ and standard deviation σ of the deviation. Based on the prediction accuracy requirements of the grey model, a threshold range for anomaly identification was set. λ is a preset confidence coefficient; the current seepage monitoring data is compared with the gray seepage prediction value: if the deviation falls within the threshold interval T, the monitoring data is determined to be normal; if the deviation exceeds the threshold interval, the monitoring data is determined to be abnormal monitoring data; the threshold interval T is recalculated with the addition of new data using the metabolically updated sequence to achieve dynamic adaptive adjustment.

[0077] Furthermore, step S2, which involves performing mean transformation on the abnormal monitoring data and related measuring point monitoring data, and then calculating the grey correlation degree between the abnormal monitoring points and related measuring points after the transformation, includes:

[0078] Let the monitoring data sequence of the measuring point where the abnormal monitoring data is located be the reference sequence. Where k is the time number and n is the sequence length. This represents the raw seepage monitoring data at time k; there are m related measurement point data sequences as comparison sequences. ,in , Let represent the observed value of the i-th relevant measuring point at time k; perform mean transformation on the reference sequence and each comparison sequence respectively, using the following transformation formula:

[0079]

[0080] in The denominator is the k-th value in the current sequence being transformed, i.e., the reference sequence or a comparison sequence. The arithmetic mean of the entire sequence. The transformed dimensionless values ​​are used; after transformation, the correlation coefficient between each comparison series and the reference series at each time step is calculated using the following formula. :

[0081]

[0082] in: This represents the absolute difference between the reference sequence and the i-th comparison sequence at time k; It represents the minimum absolute difference between all compared sequences and the reference sequence at all times; The maximum absolute difference between all compared sequences and the reference sequence at all times; The resolution coefficient is used to calculate the correlation between each comparison sequence and the reference sequence using the following formula. :

[0083]

[0084] in It reflects the overall correlation between the i-th related measurement point and the anomaly monitoring data. Based on the size of the grey correlation degree, the measurement point data with the top N grey correlation degrees are selected as the target measurement point data input to the radial basis function neural network.

[0085] Furthermore, step S2, which involves constructing a radial basis function neural network model and inputting the anomaly monitoring data and target measurement point data into the radial basis function neural network model for training, includes:

[0086] The radial basis function neural network adopts a three-layer feedforward network structure, including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined according to the total number of monitoring indicators selected by grey relational analysis in step S2, and the seepage monitoring data of the measurement point where the abnormal monitoring data is located and the surrounding target measurement points are used as the input vector. The number of nodes in the hidden layer is determined by trial and error or cross-validation. Each hidden layer node corresponds to a radial basis function, and its center vector and expansion constant are the parameters to be trained. The number of nodes in the output layer is set to 1, and the output value is the neural network seepage prediction value of the measurement point where the abnormal monitoring data is located.

[0087] The network parameters are initialized as follows: a portion of the input samples are randomly selected as the center vectors of the hidden layer nodes, or the input data is clustered using a clustering algorithm, and the cluster centers are used as the initial values ​​of the center vectors; the expansion constant is set to a certain proportion of the average distance between the input data samples, according to the distribution of the input data.

[0088] Model training and dataset partitioning are as follows: the preprocessed dataset is divided into training set, validation set and test set according to a preset ratio; the training set is used for parameter learning of the neural network; the validation set is used to evaluate the model performance during training to prevent overfitting; the test set is used to perform final performance testing on the trained model; after training is completed, the neural network seepage prediction value of the measurement point where the abnormal monitoring data is located is obtained using the test set.

[0089] Furthermore, step S2, which involves identifying the environmental response of the abnormal monitoring data based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, includes:

[0090] The neural network seepage prediction value output by the radial basis function neural network is compared with the corresponding actual value of the abnormal monitoring data, and the relative error between the two is calculated. A relative error threshold is set, which is predetermined based on engineering experience or statistical methods. If the relative error is less than or equal to the relative error threshold, the abnormal monitoring data is determined to be a normal response caused by changes in environmental quantities, i.e., it is identified as an environmental quantity response. If the relative error is greater than the relative error threshold, the abnormal monitoring data is determined not to be an environmental quantity response, and step S3 is required for retesting and verification and monitoring instrument inspection.

[0091] Based on a similar inventive concept, this invention also provides a tailings dam seepage monitoring data anomaly identification system based on gray model and neural network, corresponding to the aforementioned tailings dam seepage monitoring data anomaly identification method based on gray model and neural network, comprising:

[0092] The data acquisition module is used to acquire tailings dam seepage monitoring data;

[0093] The gray model anomaly identification module is used to preprocess the seepage monitoring data using the arithmetic square root to obtain a preprocessed seepage monitoring data sequence, and to perform a single accumulation generation on the preprocessed seepage monitoring data sequence to obtain a single accumulation generation sequence; based on the single accumulation generation sequence, a GM(1,1) gray prediction model is established, and the parameters of the gray differential equation are solved to obtain the gray seepage prediction value at the corresponding monitoring time; anomaly identification threshold interval is constructed according to the deviation between the gray seepage prediction value and the obtained seepage monitoring data, and the preprocessed seepage monitoring data sequence is updated using a metabolic update method; seepage monitoring data whose deviation exceeds the anomaly identification threshold interval is identified as abnormal monitoring data, and the monitoring point corresponding to the abnormal monitoring data is identified as an abnormal monitoring point;

[0094] An environmental quantity response identification module is used to acquire monitoring data of relevant measuring points corresponding to the abnormal monitoring points, and to perform mean transformation on the abnormal monitoring data and the monitoring data of relevant measuring points. After transformation, the gray correlation degree between the abnormal monitoring points and the relevant measuring points is calculated. Target measuring point data is selected as the input variable of the neural network based on the gray correlation degree. A radial basis function neural network model is constructed, and the abnormal monitoring data and the target measuring point data are input into the radial basis function neural network model for training to obtain the neural network seepage prediction value corresponding to the abnormal monitoring points. Based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, environmental quantity response identification is performed on the abnormal monitoring data.

[0095] The result determination and processing module is used to retest and verify abnormal monitoring data that are not identified as environmental quantity responses in the environmental quantity response identification module and to check the monitoring instruments. When the retest result is normal, the abnormal monitoring data is identified as measurement error data, and the original monitoring data corresponding to the abnormal monitoring data is removed or corrected. When the retest result is abnormal, the abnormal monitoring data is identified as tailings dam structural change data, and the corresponding abnormal warning information is output.

[0096] The invention will now be illustrated with a specific example.

[0097] Step S1: Identify anomaly monitoring data based on gray model

[0098] 1.1 Project Overview

[0099] The LS phosphogypsum tailings dam employs a wet stacking method. The project was put into operation in January 2006, with a final designed total dam height of approximately 130m. By September 2010, the tailings dam had been upgraded to a 10-stage sub-dam, with a crest elevation of 2020m and a stacking height of 80m. The initial dam was 30m high, constructed with earth filling. Rockfill drainage systems were installed on the upstream slope and at the dam base, while drainage prisms were installed on the downstream dam. The tailings dam uses an upstream dam construction method. The sub-dam crest is 5-8m wide and 5m high, with an overall stacking slope ratio of approximately 1:4.0. The LS tailings dam is located in a wide, gently sloping valley running east-west, with a terrain that slopes from north to south and from west to east. The geological structure of the dam site consists of Lower Cambrian Qiongzhusi Formation silty mudstone and carbonaceous shale.

[0100] The tailings dam seepage monitoring uses piezometers. Nine piezometers were installed at the 2nd, 4th, and 5th sub-dam sections during the initial safety assessment in March 2008, designated as observation points K1-9. The tailings dam topographic overview and monitoring point layout are shown in the following figure. Figure 2 As shown, the green dashed line is the boundary marker between the initial dam and the accumulation dam.

[0101] 1.2 Constructing the Grey Model

[0102] The grey prediction model GM(1,1) is suitable for short- to medium-term predictions and requires relatively few raw data points, typically 6-15. The grey prediction model has been applied in tailings dam phreatic line prediction, and its calculation accuracy meets engineering requirements. However, it has not yet been applied to the identification of abnormal monitoring points. This study validated the model using raw data from the K4 observation point, located in the middle of the secondary sub-dam. Data sample segments with abrupt change points were selected from the monitoring data, and a GM(1,1) model with 7 samples was established. The predicted value for the next time point was calculated using the prediction model, and compared with the measured results. The residual results are shown in Table 1.

[0103] Table 1. Prediction results of the GM(1,1) model

[0104] Time series 8 9 10 11 12 13 14 15 Monitoring value / m 17.45 17.20 17.39 17.15 17.31 17.00 16.90 16.07 Predicted value / m 17.32 17.26 17.36 17.21 17.19 16.84 16.98 16.91 relative error 0.74% 0.35% 0.17% 0.35% 0.69% 0.94% 0.47% 5.23%

[0105] 1.3 Anomaly Data Identification

[0106] The calculation results above show that when the infiltration line monitoring data changes steadily without abrupt changes, the grey prediction model can effectively predict the next sample data, with the relative error between the model and the measured value controlled within 3%. When abnormal data points appear in the observed values, such as time series point 15 in this example, the relative error between the monitored value and the predicted value will increase significantly. Therefore, this model can be used to identify abnormal data points.

[0107] Anomaly identification was performed on a total of 2436 monitoring data points across 9 measurement points, and the results were compared with those obtained manually. Manual verification identified 64 anomalous data points, resulting in a false negative rate of 40.63%. The grey model method, however, only missed one anomalous data point, with a false negative rate of 1.56%. These results demonstrate that the proposed grey model-based anomaly identification method overcomes the problem of low fitting accuracy in statistical models, which are unsuitable for anomaly identification, and exhibits good applicability and accuracy.

[0108] Step S2: Environmental Quantity Response Identification

[0109] 2.1 Correlation Analysis

[0110] Correlation analysis was used to analyze the observation points K1-K9 of the seepage line. K1-K4 are located in the secondary sub-dam, K5-K8 are located in the quaternary sub-dam, and K9 is located in the quinary sub-dam. The correlation matrix between the observation points was calculated using monitoring data from June 2009 to February 2010, as shown in Table 2.

[0111] Table 2. Correlation Matrix of Observation Points K1-K9 R

[0112] K1 K2 K3 K4 K5 K6 K7 K8 K9 K1 1 0.958 0.949 0.932 0.925 0.737 0.927 0.711 0.913 K2 0.958 1 0.969 0.952 0.935 0.745 0.942 0.73 0.932 K3 0.949 0.969 1 0.966 0.929 0.758 0.958 0.743 0.935 K4 0.932 0.952 0.966 1 0.923 0.754 0.968 0.757 0.936 K5 0.925 0.935 0.929 0.923 1 0.747 0.917 0.737 0.912 K6 0.737 0.745 0.758 0.754 0.747 1 0.756 0.599 0.74 K7 0.927 0.942 0.958 0.968 0.917 0.756 1 0.75 0.931 K8 0.711 0.73 0.743 0.757 0.737 0.599 0.75 1 0.728 K9 0.913 0.932 0.935 0.936 0.912 0.74 0.931 0.728 1

[0113] The correlation matrix R of each monitoring point was obtained by calculation, where R 12 =0.958, R 13 =0.949, R 14 =0.932 represents the correlation between K2, K3, K4 and K1, respectively. The input values ​​of the target measurement point neural network are determined by comparing the correlation between monitoring points. Measurement points K2, K3, and K4 are selected as the input vector for K1. Similarly, the three measurement points with the highest correlation are selected as the input vectors for each monitoring point. (K1, K3, K4), (K4, K2, K1), and (K3, K2, K7) are selected sequentially as the input vectors for K2, K3, and K4.

[0114] 2.2 Identifying Abnormal Data Based on Neural Networks

[0115] The monitoring value sequence of the target measurement point is used as the output vector, and the monitoring value sequences of the remaining measurement points are used as the input vector. Fifty sets of measurements from the first segment of the monitoring data are selected for analysis, with 30 sets used as training samples and the remaining 20 sets as test samples. After inputting the training samples into the network, the network adjusts the weights and neuron centers according to the learning rules to achieve the specified minimum error index. After training, the trained neural network is used to test the test samples. Except for a few points with large fluctuations, the other measurement points show good agreement, indicating that the constructed neural network model has good accuracy.

[0116] The prediction results show that the relative errors between the model's predicted values ​​and the measured values ​​for normal data points are all within 5%, while the relative errors for outliers (non-environmental response) are all above 8%. Therefore, setting the relative error threshold δ to 5% is sufficient to effectively identify outliers. This model is used to identify environmental response quantities for outlier data points in historical monitoring data. If the relative error is less than δ, the abrupt change is considered an environmental response, and the data point is retained as normal data; if the relative error is greater than δ, the cause of the abrupt change is excluded as an environmental response.

[0117] Figure 3 The environmental response identification results for observation point K2 are shown. Red stars indicate anomalous data points that are not environmental responses, while green stars indicate anomalous data points that are environmental responses. The relative errors of anomalous data points 2, 3, and 5 are all less than 3%, while the relative errors of data points 1 and 4 are 8.34% and 9.45%, respectively. Therefore, anomalous data points 2, 3, and 5 are determined to be environmental responses and entered into the database as normal data. Anomalous data points 1 and 4 are determined to be non-environmental responses and retained as anomalous data points. Using this identification method, environmental responses were identified for all anomalous data points. Out of a total of 64 anomalous data points, 37 were identified as environmental responses. Compared with the results of manual verification, 2 were missed, resulting in an error rate of 3.13%. This indicates that the method is reasonable, effective, and highly accurate, meeting the requirements for engineering applications.

[0118] Step S3: Since the data used in this paper is historical data, only historical data is used for anomaly identification and environmental impact assessment. No structural changes have occurred during the operation of the tailings dam; therefore, all abnormal measurement points identified in the environmental response are treated as outliers caused by measurement errors and are removed and corrected.

[0119] The present invention also discloses an electronic device.

[0120] Specifically, the electronic device can be a desktop computer, laptop computer, handheld computer, or cloud server, etc. This computer device may include, but is not limited to, a processor and memory. The processor and memory can be connected via a bus or other means. The processor can be a Central Processing Unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, graphics processing units (GPUs), embedded neural network processing units (NPUs) or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0121] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules. The processor executes various functional applications and data processing by running non-transitory software programs, instructions, and modules stored in memory. Memory may include a program storage area and a data storage area. The program storage area may store the control unit and the application program required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, memory may include high-speed random access memory and non-transitory memory. In some embodiments, memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0122] The present invention also discloses a computer-readable storage medium.

[0123] Specifically, the computer-readable storage medium is used to store a computer program, which, when executed by a processor, implements the methods described in the above method implementation.

[0124] Those skilled in the art will understand that all or part of the processes in the methods described above can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

Claims

1. A method for identifying anomalies in tailings dam seepage monitoring data based on grey model and neural network, characterized in that, Includes the following steps: S1. Acquire tailings dam seepage monitoring data, perform arithmetic square root preprocessing on the seepage monitoring data to obtain a preprocessed seepage monitoring data sequence, and perform a first-time accumulation generation on the preprocessed seepage monitoring data sequence to obtain a first-time accumulation generation sequence; establish a GM(1,1) grey prediction model based on the first-time accumulation generation sequence, solve the grey differential equation parameters, and obtain the grey seepage prediction value at the corresponding monitoring time; construct an anomaly identification threshold interval based on the deviation between the grey seepage prediction value and the acquired seepage monitoring data, and update the preprocessed seepage monitoring data sequence using a metabolic update method; identify seepage monitoring data whose deviation exceeds the anomaly identification threshold interval as abnormal monitoring data, and identify the monitoring points corresponding to the abnormal monitoring data as abnormal monitoring points; S2. Obtain monitoring data of relevant measuring points corresponding to the abnormal monitoring points, and perform mean transformation on the abnormal monitoring data and the monitoring data of relevant measuring points. After transformation, calculate the gray correlation degree between the abnormal monitoring points and the relevant measuring points, and select target measuring point data as the input variable of the neural network based on the gray correlation degree. Construct a radial basis function neural network model, and input the abnormal monitoring data and the target measuring point data into the radial basis function neural network model for training to obtain the neural network seepage prediction value corresponding to the abnormal monitoring points. Based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, identify the environmental quantity response of the abnormal monitoring data. S3. Retest and verify the abnormal monitoring data that were not identified as environmental quantity responses in step S2, and check the monitoring instruments. When the retest results are normal, the abnormal monitoring data is identified as measurement error data, and the original monitoring data corresponding to the abnormal monitoring data is removed or corrected. When the retest results are abnormal, the abnormal monitoring data is identified as tailings dam structural change data, and the corresponding abnormal warning information is output.

2. The method according to claim 1, characterized in that, Step S1 involves performing arithmetic square root preprocessing on the seepage monitoring data to obtain a preprocessed seepage monitoring data sequence, and then performing a single accumulation on the preprocessed seepage monitoring data sequence to generate the following: Set the original seepage monitoring data sequence ,in Here, n is the sequence number of the monitoring time, and n is the sequence length. This represents the raw seepage monitoring data at time k; the raw seepage monitoring data is preprocessed using the arithmetic square root to obtain the preprocessed seepage monitoring data sequence. ,in ; then Perform one accumulation generation to obtain an accumulation generation sequence. Where j is the index variable in the summation, This represents the cumulative sum of the preprocessed data over the first k time steps.

3. The method according to claim 2, characterized in that, Step S1, which involves establishing a GM(1,1) grey prediction model, solving the grey differential equation parameters, obtaining the grey seepage prediction value at the corresponding monitoring time, and updating the preprocessed seepage monitoring data sequence using a metabolic update method, includes: Generate a sequence by accumulating once. Treating it as a function of time, we establish a first-order differential equation: Where a is the development coefficient, b is the gray action quantity, and t is the time variable; let the vector to be identified be... The parameter estimates of a and b are obtained by solving using the least squares method. and ; Therefore, the time response formula of the GM(1,1) grey prediction model is established: in, This is the first value of the preprocessed data. It is a natural constant. The current time number. This is the cumulative predicted value at time k+1; restoring the above predicted value yields the predicted value within the preprocessing domain: (k=2,3,…,n) in, This is the actual accumulated value at time k. The predicted value of the preprocessed data at time k+1 is given; the predicted value of the preprocessed data is then subjected to inverse preprocessing to obtain the predicted value of gray seepage under the original dimensions: ;in, This is the predicted gray seepage value at time k+1; using a metabolic update method, the above calculated values ​​are... The latest data is added to the original sequence, while the oldest data in the original sequence is removed. This forms a new, equal-length original data sequence, which is used for dynamic prediction at the next time step.

4. The method according to claim 1, characterized in that, Step S1, which involves constructing an anomaly identification threshold range based on the deviation between the predicted gray seepage value and the acquired seepage monitoring data, includes: For each monitoring time point, calculate the predicted value of gray seepage. The relative deviation between the data and the seepage monitoring data at that moment was calculated, and the deviation distribution of all historical normal moments was statistically analyzed to obtain the mean μ and standard deviation σ of the deviation. Based on the prediction accuracy requirements of the grey model, a threshold range for anomaly identification was set. λ is a preset confidence coefficient; the current seepage monitoring data is compared with the gray seepage prediction value: if the deviation falls within the threshold interval T, the monitoring data is determined to be normal; if the deviation exceeds the threshold interval, the monitoring data is determined to be abnormal monitoring data; the threshold interval T is recalculated with the addition of new data using the metabolically updated sequence to achieve dynamic adaptive adjustment.

5. The method according to claim 1, characterized in that, Step S2 involves performing mean transformation on the abnormal monitoring data and related measuring point monitoring data, and then calculating the grey correlation degree between the abnormal monitoring points and related measuring points after the transformation process. Let the monitoring data sequence of the measuring point where the abnormal monitoring data is located be the reference sequence. Where k is the time number and n is the sequence length. This represents the raw seepage monitoring data at time k; there are m related measurement point data sequences as comparison sequences. ,in , Let represent the observed value of the i-th measuring point at time k; perform mean transformation on the reference sequence and each comparison sequence respectively, using the following transformation formula: in The denominator is the k-th value in the current sequence being transformed, i.e., the reference sequence or a comparison sequence. The arithmetic mean of the entire sequence. The transformed dimensionless values ​​are used; after transformation, the correlation coefficient between each comparison series and the reference series at each time step is calculated using the following formula. : in: This represents the absolute difference between the reference sequence and the i-th comparison sequence at time k; It represents the minimum absolute difference between all compared sequences and the reference sequence at all times; The maximum absolute difference between all compared sequences and the reference sequence at all times; The resolution coefficient is used to calculate the correlation between each comparison sequence and the reference sequence using the following formula. : ;in It reflects the overall correlation between the data of the i-th measurement point and the data of the abnormal monitoring point. According to the size of the gray correlation degree, the measurement point data ranked in the top N by gray correlation degree are selected as the target measurement point data input to the radial basis function neural network.

6. The method according to claim 1, characterized in that, Step S2, which involves constructing a radial basis function neural network model and training it by inputting the anomaly monitoring data and target measurement point data into the radial basis function neural network model, includes: The radial basis function neural network adopts a three-layer feedforward network structure, including an input layer, a hidden layer, and an output layer. The number of nodes in the input layer is determined according to the total number of monitoring indicators selected by grey relational analysis in step S2. The seepage monitoring data of the measurement point where the abnormal monitoring data is located and its surrounding strongly correlated measurement points, i.e., the target measurement point data, are used as the input vector. The number of nodes in the hidden layer is determined by trial and error or cross-validation. Each hidden layer node corresponds to a radial basis function, and its center vector and expansion constant are the parameters to be trained. The number of nodes in the output layer is set to 1, and the output value is the neural network seepage prediction value of the measurement point where the abnormal monitoring data is located. The network parameters are initialized as follows: a portion of the input samples are randomly selected as the center vectors of the hidden layer nodes, or the input data is clustered using a clustering algorithm, and the cluster centers are used as the initial values ​​of the center vectors; the expansion constant is set to a certain proportion of the average distance between the input data samples, according to the distribution of the input data. Model training and dataset partitioning are as follows: the preprocessed dataset is divided into training set, validation set and test set according to a preset ratio; the training set is used for parameter learning of the neural network; the validation set is used to evaluate the model performance during training to prevent overfitting; the test set is used to perform final performance testing on the trained model; after training is completed, the neural network seepage prediction value of the measurement point where the abnormal monitoring data is located is obtained using the test set.

7. The method according to claim 1, characterized in that, Step S2, which involves identifying the environmental response of the abnormal monitoring data based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, includes: The neural network seepage prediction value output by the radial basis function neural network is compared with the corresponding actual value of the abnormal monitoring data, and the relative error between the two is calculated. A relative error threshold is set, which is predetermined based on engineering experience or statistical methods. If the relative error is less than or equal to the relative error threshold, the abnormal monitoring data is determined to be a normal response caused by changes in environmental quantities, i.e., it is identified as an environmental quantity response. If the relative error is greater than the relative error threshold, the abnormal monitoring data is determined not to be an environmental quantity response, and step S3 is required for retesting and verification and monitoring instrument inspection.

8. A tailings dam seepage monitoring data anomaly identification system based on grey model and neural network, characterized in that, include: The data acquisition module is used to acquire tailings dam seepage monitoring data; The gray model anomaly identification module is used to preprocess the seepage monitoring data using the arithmetic square root to obtain a preprocessed seepage monitoring data sequence, and to perform a single accumulation generation on the preprocessed seepage monitoring data sequence to obtain a single accumulation generation sequence; based on the single accumulation generation sequence, a GM(1,1) gray prediction model is established, and the parameters of the gray differential equation are solved to obtain the gray seepage prediction value at the corresponding monitoring time; anomaly identification threshold interval is constructed according to the deviation between the gray seepage prediction value and the obtained seepage monitoring data, and the preprocessed seepage monitoring data sequence is updated using a metabolic update method; seepage monitoring data whose deviation exceeds the anomaly identification threshold interval is identified as abnormal monitoring data, and the monitoring point corresponding to the abnormal monitoring data is identified as an abnormal monitoring point; An environmental quantity response identification module is used to acquire monitoring data of relevant measuring points corresponding to the abnormal monitoring points, and to perform mean transformation on the abnormal monitoring data and the monitoring data of relevant measuring points. After transformation, the gray correlation degree between the abnormal monitoring points and the relevant measuring points is calculated. Target measuring point data is selected as the input variable of the neural network based on the gray correlation degree. A radial basis function neural network model is constructed, and the abnormal monitoring data and the target measuring point data are input into the radial basis function neural network model for training to obtain the neural network seepage prediction value corresponding to the abnormal monitoring points. Based on the relative error between the neural network seepage prediction value and the abnormal monitoring data, environmental quantity response identification is performed on the abnormal monitoring data. The result determination and processing module is used to retest and verify abnormal monitoring data that are not identified as environmental quantity responses in the environmental quantity response identification module and to check the monitoring instruments. When the retest result is normal, the abnormal monitoring data is identified as measurement error data, and the original monitoring data corresponding to the abnormal monitoring data is removed or corrected. When the retest result is abnormal, the abnormal monitoring data is identified as tailings dam structural change data, and the corresponding abnormal warning information is output.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the tailings dam seepage monitoring data anomaly identification method based on gray model and neural network as described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the tailings dam seepage monitoring data anomaly identification method based on gray model and neural network according to any one of claims 1 to 7.

Images