On-line grade monitoring method for reselection and enrichment of low-grade waste rocks

Abstract

The invention relates to the field of low-grade waste rock reselection and enrichment on-line grade monitoring, in particular to a low-grade waste rock reselection and enrichment on-line grade monitoring method. The method comprises the following steps: collecting physical characteristics of a barren rock sample, obtaining physical characteristics of a historical barren rock sample and an actual grade of the corresponding barren rock sample, respectively carrying out standardization processing, and outputting a standardized physical characteristic set and a standardized historical physical characteristic set; based on the standardized physical features and the standardized historical physical features, introducing a nonlinear weighted fusion method based on physical consistency constraint to carry out feature fusion, and constructing a fusion feature set; and based on the fusion feature set and the standardized physical feature set, constructing an optimized support vector regression grade prediction model, obtaining a grade prediction result of the waste rock, and restoring the grade prediction result to an actual grade. The problems that a traditional grade monitoring method is delayed in response, unscientific in multi-source data fusion, poor in prediction accuracy and incapable of adapting to process dynamic changes are solved.

Description

Technical Field

[0001] This invention relates to the field of online grade monitoring of low-grade waste rock gravity separation enrichment, and particularly to a method for online grade monitoring of low-grade waste rock gravity separation enrichment. Background Technology

[0002] In the industrial production of waste rock resorption and enrichment, waste rock grade monitoring is a key link in guiding process adjustments and improving resource utilization. Existing grade monitoring methods are mainly divided into two categories: one is the periodic laboratory testing method that has been used in the industry for a long time. This method obtains waste rock grade data through manual sampling, sample preparation, and laboratory testing, and is the traditional benchmark method for grade determination. The other is the grade prediction method that relies on physical characteristic data collected by sensors on the production line and combines it with traditional regression models. This method attempts to replace some laboratory testing work by constructing a prediction model after simple processing of a small number of features.

[0003] However, both types of quality monitoring methods have significant technical shortcomings: periodic laboratory testing procedures are cumbersome and time-consuming, with obvious response delays; traditional regression models lack scientific rigor in feature fusion, making it difficult to adapt to feature correlation patterns under complex working conditions, resulting in poor prediction accuracy, and neither type of method can dynamically match the needs of changing process parameters. Summary of the Invention

[0004] This invention provides an online grade monitoring method for the gravity separation and enrichment of low-grade waste rock, which solves the technical problems of traditional grade monitoring methods, such as response delay, unscientific multi-source data fusion, poor prediction accuracy, inability to adapt to dynamic changes in the process, and difficulty in achieving real-time and accurate prediction of waste rock grade.

[0005] The present invention provides an online grade monitoring method for gravity separation enrichment of low-grade waste rock, comprising the following steps:

[0006] S1. Collect the physical characteristics of waste rock samples and obtain the physical characteristics of historical waste rock samples and the actual grade of the corresponding waste rock samples; standardize the physical characteristics of the waste rock samples, the physical characteristics of historical waste rock samples, and the actual grade of the corresponding waste rock samples to obtain standardized physical characteristics, standardized historical physical characteristics, and standardized actual grade of the corresponding waste rock samples, and output the standardized physical characteristic set and the standardized historical physical characteristic set; based on the standardized physical characteristics and the standardized historical physical characteristics, introduce a nonlinear weighted fusion method based on physical consistency constraints to perform feature fusion, obtain fused features, and construct the fused feature set;

[0007] S2. Based on the fused feature set and the standardized physical feature set, an optimized support vector regression grade prediction model is constructed to obtain the grade prediction results of waste rock and restore them to the actual grade.

[0008] Preferably, S1 specifically includes:

[0009] In the implementation of the nonlinear weighted fusion method based on physical consistency constraints, the interaction strength between physical features is quantified by calculating the ratio between standardized physical features, and the mutual influence coefficient between physical features is introduced. An exponential function is then applied to fuse the standardized physical features after nonlinear mapping to obtain fused features.

[0010] Preferably, S1 specifically includes:

[0011] The above-mentioned nonlinear weighted fusion method based on physical consistency constraints is applied to the standardized historical physical characteristics to obtain the predicted fusion characteristics; the actual correlation coefficient between the predicted fusion characteristics and the actual grade of the corresponding waste rock sample after standardization is calculated, and the correlation error is obtained by combining the ideal perfect correlation coefficient; the mutual influence coefficient between physical characteristics is obtained by iterating with the goal of minimizing the correlation error.

[0012] Preferably, S2 specifically includes:

[0013] The optimized support vector regression grade prediction model includes a kernel mapping term, a physical consistency correction term, and a second-order difference regularization term.

[0014] Preferably, S2 specifically includes:

[0015] By introducing a Gaussian kernel function, the fused features at each time step are mapped to a high-dimensional space, thus constructing a kernel mapping term.

[0016] Preferably, S2 specifically includes:

[0017] The standardized physical characteristics at a single moment are nonlinearly mapped and weighted summed, and combined with the physical consistency correction coefficient, a physical consistency correction term is constructed to correct the fusion bias.

[0018] Preferably, S2 specifically includes:

[0019] By calculating the second-order difference approximation of the fused features and combining it with the regularization coefficient, a second-order difference regularization term is constructed.

[0020] Preferably, S2 specifically includes:

[0021] The grade prediction results of waste rock output by the optimized support vector regression grade prediction model are normalized to obtain normalized waste rock grade prediction results; the normalized waste rock grade prediction results are then denormalized to restore the actual grade.

[0022] The beneficial effects of the technical solution of the present invention are:

[0023] 1. Simultaneously collect multi-dimensional physical features to obtain the physical characteristics of historical waste rock samples and the actual grade of the corresponding historical waste rock samples, and standardize all data to ensure comparability; adopt a nonlinear weighted fusion method based on physical consistency constraints to strengthen the nonlinear correlation of features and physical consistency, output high-quality fusion features, effectively avoid the information loss problem of linear fusion, and lay a reliable foundation for grade prediction.

[0024] 2. An optimized support vector regression grade prediction model is constructed, which integrates physical consistency correction terms and second-order difference regularization terms. This can suppress feature mutation interference, correct fusion bias, and, combined with normalization constraints, ensure that the waste rock grade prediction results are reasonable and fully meet the requirements of industrial-grade monitoring.

[0025] 3. It adapts to the dynamic fluctuations in waste rock characteristics in industrial scenarios. The waste rock grade prediction results can provide accurate data support for production line adjustments, help optimize production processes and improve enrichment efficiency, and have outstanding engineering application value and promotion potential. Attached Figure Description

[0026] Figure 1 This is a flowchart of an online grade monitoring method for the gravity separation and enrichment of low-grade waste rock according to the present invention. Detailed Implementation

[0027] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0029] The following description, in conjunction with the accompanying drawings, details the specific scheme of the online grade monitoring method for the gravity separation and enrichment of low-grade waste rock provided by this invention.

[0030] See attached document Figure 1 The diagram illustrates a flowchart of an online grade monitoring method for gravity separation and enrichment of low-grade waste rock according to an embodiment of the present invention. The method includes the following steps:

[0031] S1. Collect the physical characteristics of waste rock samples, and obtain the physical characteristics of historical waste rock samples and the actual grades of the corresponding waste rock samples; perform standardization processing on the physical characteristics of waste rock samples, the physical characteristics of historical waste rock samples, and the actual grades of the corresponding waste rock samples respectively, to obtain the standardized physical characteristics, the standardized historical physical characteristics, and the standardized actual grades of the corresponding waste rock samples, and output the standardized physical characteristic set and the standardized historical physical characteristic set; based on the standardized physical characteristics and the standardized historical physical characteristics, introduce a non-linear weighted fusion method based on physical consistency constraints for feature fusion, obtain the fusion features, and construct a fusion feature set.

[0032] First, to ensure the comparability of different physical characteristics, based on the existing sensors on the waste rock gravity concentration production line, such as mineral composition sensors, laser particle size sensors, density sensors, and moisture content sensors, synchronously and continuously collect the physical characteristics of a single group of waste rock samples, such as mineral composition, particle size, density, moisture content ratio, etc., and store them in the industrial real-time archive database supporting the production line, and establish an index according to the sample group and sampling time; at the same time, to establish a benchmark data set for subsequent analysis, the corresponding waste rock samples will be sent to the laboratory for grade analysis after collection, and the obtained actual grade results will be filled back into the corresponding records in the database through the unique identifier of the same sample, ensuring the one-to-one association between historical physical characteristics and actual grades; obtain the physical characteristics of historical waste rock samples and the actual grades of the corresponding historical waste rock samples from the industrial real-time archive database, and then use the min-max normalization method to perform standardization processing on the physical characteristics of the currently collected waste rock samples, the physical characteristics of historical waste rock samples, and the actual grades of the corresponding historical waste rock samples respectively, to obtain the standardized physical characteristics, the standardized historical physical characteristics, and the standardized actual grades of the corresponding waste rock samples, and output the standardized physical characteristic set at all sampling times of synchronous continuous collection and the standardized historical physical characteristic set at all historical sampling times.

[0033] After the standardization processing of physical characteristics is completed, introduce a non-linear weighted fusion method based on physical consistency constraints to fuse all the standardized physical characteristics corresponding to a single sampling time in the standardized physical characteristic set, and obtain the fusion features. The non-linear weighted fusion method based on physical consistency constraints is optimized and designed based on the D-S evidence fusion theory. First, calculate the ratio between the standardized physical characteristics to quantify the interaction intensity between physical characteristics, and introduce an exponential function to amplify the non-linear correlation to avoid the loss of correlation information caused by linear fusion; at the same time, construct a physical consistency constraint mechanism to ensure that the fusion features output in the feature fusion stage conform to the actual ore characteristics. The feature fusion formula is as follows:

[0034]

[0035] Among them, X fused The fused feature is the result obtained by weighting and nonlinearly transforming all standardized physical features within a single sampling time; n represents the total number of standardized physical features within a single sampling time; i and j are the indices of all standardized physical features within a single sampling time; λ i The weighting coefficient λ represents the weighted physical feature at the i-th standardized physical feature at a single sampling time, reflecting the importance of the standardized physical feature at that single sampling time in waste rock grade prediction. The calculation method is as follows: First, the standardized historical physical features at all historical sampling times are analyzed using the entropy weight method to obtain the baseline weight for each physical feature. Then, combined with the historical distribution of each physical feature, the deviation of each standardized physical feature at the current time is quantified by calculating the normal distribution probability density, and the self-information of each standardized physical feature within a single sampling time is calculated. The self-information is normalized to obtain the instantaneous importance score of each standardized physical feature. Finally, the baseline weight and the instantaneous importance score are combined using a method such as weighted geometric mean to obtain the weighting coefficient λ. i , λ i The range of values ​​for is (0,1), and the sum is 1. Let exp(⋅) be the real-time combined effect of all standardized physical features except itself on the i-th standardized physical feature, and let exp(⋅) be the exponential function. This indicates the degree to which the i-th standardized physical feature is affected by the real-time interaction of all physical features except the i-th standardized feature. This allows the fused features to better reflect the actual correlation between physical features of waste rock, avoids ignoring the interaction between standardized physical features, and prevents distortion of the fusion results, thereby improving the accuracy of waste rock grade prediction. These represent the standardized physical features at the i-th and j-th sampling times, respectively; This is the processing function for the i-th standardized physical feature at a single sampling time. It is used to perform a nonlinear mapping on the standardized physical feature at a single sampling time. For example, for standardized physical features with a long-tailed distribution, such as mineral composition, a base-10 logarithmic shift transformation is used to effectively reduce the influence of extreme values; α ij Represents the standardized physical features at the i-th and j-th sampling times. and The mutual influence coefficient is used to quantify the interdependence between standardized physical features at a single sampling time. Its calculation method is as follows: Substitute the standardized historical physical features into the feature fusion formula to obtain the predicted fused features. Calculate the actual correlation coefficient between the predicted fused features and the corresponding waste rock sample's standardized actual grade using Pearson correlation analysis. Then, calculate the difference between the ideal perfect correlation coefficient and the actual correlation coefficient to obtain the correlation error. Finally, with the goal of minimizing the correlation error, iteratively correct the correlation using the existing Gauss-Newton method to obtain the mutual influence coefficient α. ij The value range is [-1, 1]. If it exceeds the range, it will be adjusted by the max-min normalization algorithm. The ideal perfect correlation coefficient represents the optimal correlation between the predicted fusion features and the actual grade. The default value is 1, which means that the physical features and the grade are perfectly positively correlated.

[0036] Finally, the fusion features of a single waste rock sample collected from all sampling moments of the production line are output and constituted as a fusion feature set. This serves as the core input for grade prediction and can significantly improve the scientificity and effectiveness of feature fusion.

[0037] S2. Based on the fused feature set and the standardized physical feature set, an optimized support vector regression grade prediction model is constructed to obtain the grade prediction results of waste rock and restore them to the actual grade.

[0038] After feature fusion, the next step is waste rock grade prediction. Based on the fused feature set and the standardized physical feature set, an optimized support vector regression grade prediction model is constructed. In the construction of this optimized model, a second-order difference regularization term is introduced. The second-order difference effectively controls the impact of abrupt changes in the standardized physical features on the prediction results, improving the stability and accuracy of the optimized model. Regularization effectively suppresses overfitting. The specific formula for the optimized support vector regression grade prediction model is as follows:

[0039]

[0040] Where P is the predicted grade of waste rock; T is the total sampling time collected by a single group of waste rock samples through the production line; t is the time index; β0, β t These are the regression constant and regression coefficients, respectively. β0 is used to correct the overall baseline bias of the optimized support vector regression grade prediction model. t The weights β0 and β1 are used to reflect the contribution of the fusion features at time t to the prediction of waste rock grade. tAll features are obtained through training using a 5-fold cross-validation method, based on standardized historical physical characteristics and the actual grade of corresponding waste rock samples; ϕ(⋅) is a Gaussian kernel function used to map the fused features to a high-dimensional space; X fused,t The fusion characteristics of a single group of waste rock samples at time t; The kernel mapping term is the weighted sum of the fused features at each time step after high-dimensional mapping, which integrates the time-series nonlinear correlation information; γ is the physical consistency correction coefficient, which is used to adjust the strength of the physical consistency correction term. It is obtained by training with 5-fold cross-validation based on the standardized historical physical features and the actual grade of the corresponding waste rock samples after standardization. This represents the i-th standardized physical feature of a single waste rock sample at time t. The physical consistency correction term strengthens the contribution of each standardized physical feature and corrects the fusion bias to ensure that the grade prediction results of waste rock conform to mineralogical laws; η is the regularization coefficient, which controls the penalty intensity on the acceleration of changes in standardized physical features. Based on the standardized historical physical features and the actual grade of the corresponding standardized waste rock samples, the value is determined by grid search optimization, and the value range is

[10] . -5 10 -1 ]; Δx represents the discrete time interval (i.e., the time difference between two adjacent sampling times) corresponding to time t. t The square of Δ2; Δ2 is a second-order difference operator used to capture the dynamic curvature of the fusion features at each time step, in order to adapt to the dynamic fluctuation characteristics of waste rock in industrial production. is the second-order difference approximation of the fused features at time t, used to characterize the second-order variation trend of the fused features; The second-order difference regularization term is the weighted sum of squares of the second-order change trends of the fused features at each time step, used to quantify the degree of abrupt change in the fused features.

[0041] In the prediction process, the fused features at each time step are first mapped to a high-dimensional space. Then, the standardized physical features at a single time step are nonlinearly mapped and weighted and summed. Simultaneously, the approximate values ​​at each time step are calculated using the second-order difference regularization term. Finally, the kernel mapping term, physical consistency correction term, and second-order difference regularization term are summed to obtain the grade prediction result of the waste rock. Since the superposition of the kernel mapping term and the second-order difference regularization term will cause the grade prediction result of the waste rock to exceed the (0,1) interval, the sigmoid function needs to be introduced to normalize the grade prediction result of the waste rock. This ensures that the final normalized grade prediction result of the waste rock strictly falls within the (0,1) interval. After the normalized grade prediction result of the waste rock is denormalized, it is restored to the actual grade, providing data support for production adjustments.

[0042] In summary, an online grade monitoring method for the gravity separation and enrichment of low-grade waste rock has been developed.

Claims

1. A method for online grade monitoring of low-grade waste rock enrichment by gravity separation, characterized in that, Includes the following steps: S1. Collect the physical characteristics of waste rock samples and obtain the physical characteristics of historical waste rock samples and the actual grade of the corresponding waste rock samples; standardize the physical characteristics of the waste rock samples, the physical characteristics of historical waste rock samples, and the actual grade of the corresponding waste rock samples to obtain standardized physical characteristics, standardized historical physical characteristics, and standardized actual grade of the corresponding waste rock samples, and output the standardized physical characteristic set and the standardized historical physical characteristic set; based on the standardized physical characteristics and the standardized historical physical characteristics, introduce a nonlinear weighted fusion method based on physical consistency constraints to perform feature fusion, obtain fused features, and construct the fused feature set; S2. Based on the fused feature set and the standardized physical feature set, an optimized support vector regression grade prediction model is constructed to obtain the grade prediction results of waste rock and restore them to the actual grade.

2. The method for online grade monitoring of low-grade waste rock enrichment by gravity separation according to claim 1, characterized in that, S1 specifically includes: In the implementation of the nonlinear weighted fusion method based on physical consistency constraints, the interaction strength between physical features is quantified by calculating the ratio between standardized physical features, and the mutual influence coefficient between physical features is introduced. An exponential function is then applied to fuse the standardized physical features after nonlinear mapping to obtain fused features.

3. The method for online grade monitoring of low-grade waste rock enrichment by gravity separation according to claim 2, characterized in that, S1 specifically includes: The above-mentioned nonlinear weighted fusion method based on physical consistency constraints is applied to the standardized historical physical characteristics to obtain the predicted fusion characteristics; the actual correlation coefficient between the predicted fusion characteristics and the actual grade of the corresponding waste rock sample after standardization is calculated, and the correlation error is obtained by combining the ideal perfect correlation coefficient; the mutual influence coefficient between physical characteristics is obtained by iterating with the goal of minimizing the correlation error.

4. The method for online grade monitoring of low-grade waste rock enrichment by gravity separation according to claim 1, characterized in that, S2 specifically includes: The optimized support vector regression grade prediction model includes a kernel mapping term, a physical consistency correction term, and a second-order difference regularization term.

5. The method for online grade monitoring of low-grade waste rock enrichment by gravity separation according to claim 4, characterized in that, S2 specifically includes: By introducing a Gaussian kernel function, the fused features at each time step are mapped to a high-dimensional space, thus constructing a kernel mapping term.

6. The method for online grade monitoring of low-grade waste rock enrichment by gravity separation according to claim 4, characterized in that, S2 specifically includes: The standardized physical characteristics at a single moment are nonlinearly mapped and weighted summed, and combined with the physical consistency correction coefficient, a physical consistency correction term is constructed to correct the fusion bias.

7. The method for online grade monitoring of low-grade waste rock enrichment by gravity separation according to claim 4, characterized in that, S2 specifically includes: By calculating the second-order difference approximation of the fused features and combining it with the regularization coefficient, a second-order difference regularization term is constructed.

8. The method for online grade monitoring of low-grade waste rock enrichment by gravity separation according to claim 1, characterized in that, S2 specifically includes: The grade prediction results of waste rock output by the optimized support vector regression grade prediction model are normalized to obtain normalized waste rock grade prediction results; the normalized waste rock grade prediction results are then denormalized to restore the actual grade.

Images