Multi-element grade prediction method based on multi-task cooperation and covariance modeling

Abstract

The invention relates to the technical field of intelligent ore processing, and discloses a multi-element grade prediction method based on multi-task collaboration and covariance modeling. The method comprises the steps that an ore grade prediction model is constructed, the ore grade prediction model comprises a feature coding layer, a fusion processing layer and an output layer, the feature coding layer is provided with a plurality of encoders for multi-modal data of ore, the fusion processing layer is provided with fusion blocks based on an attention mechanism, and the output layer is constructed based on a multi-layer perception mechanism; the method comprises the following steps: acquiring multi-modal data of to-be-detected ore, inputting the multi-modal data into an ore grade prediction model, respectively acquiring corresponding single-modal features through a plurality of encoders, acquiring fusion features by a fusion processing layer based on the single-modal features, acquiring a mean vector and a covariance matrix by an output layer based on the fusion features, and performing prediction distribution. And analyzing the ore grade data according to the prediction distribution result and regulating and controlling the process flow. The problem that the ore grade cannot be effectively and accurately detected by the existing method for predicting the ore grade by adopting machine learning is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent ore processing, and in particular to a multi-element grade prediction method based on multi-task collaboration and covariance modeling. BACKGROUND

[0002] Fast and accurate prediction of multi-element grade of ore is a core link for realizing efficient utilization of resources and optimization control of the beneficiation process in mines. Traditionally, the determination of element grade is heavily dependent on laboratory chemical analysis. Although this method has high accuracy, it has disadvantages such as long analysis period, high cost, and serious lag, and cannot meet the needs of real-time decision-making in the production site. In order to overcome this bottleneck, in recent years, spectral analysis technology (such as X-ray fluorescence spectroscopy, XRF) based on machine learning has been applied to grade prediction, realizing the progress from offline analysis to online estimation. However, these existing technical solutions still have significant deficiencies, mainly in the following aspects: Single model architecture, ignoring element correlation; existing methods usually adopt the strategy of “single-element independent modeling”, and establish a prediction model for each target element. This approach cannot take advantage of the inherent symbiotic or associated relationship between elements in ore, resulting in model redundancy, high computational cost, and the predicted results may deviate from the actual situation due to the neglect of the synergistic or antagonistic effects between elements.

[0003] Single data utilization, weak feature fusion capability; most technologies only focus on using spectral data (such as XRF or NIR) for prediction, and fail to effectively integrate a large amount of time-series process data (such as grinding and flotation parameters) and sample metadata (such as sampling information and environmental conditions) generated in industrial production processes. This single data perspective limits the model's ability to capture key factors affecting grade changes from multiple dimensions, restricting further improvement of prediction accuracy.

[0004] Single prediction output, lack of uncertainty evaluation; traditional models usually only output point estimates of grade, and cannot provide uncertainty of predicted values and correlation information between different element grade predictions. This reduces the reliability of the prediction results, making it difficult to provide more in-depth quantitative basis for subsequent process optimization and risk decision-making.

[0005] Lack of robustness, difficult to handle missing labels; in actual production environment, due to detection cost or process limitations, some element assay values (labels) of part of the samples may be missing. Existing methods lack effective mechanisms to handle such incomplete label data, often leading to a reduction in available training samples or the need for complex preprocessing, affecting the practicality and generalization ability of the model in real industrial scenarios.

[0006] The above shortcomings show that the prior art is difficult to meet the needs of modern mines for high-precision, high-reliability, strong interpretability and intelligent grade prediction that can adapt to complex industrial data. Therefore, a new prediction method that can comprehensively utilize multi-source data, cooperatively model multi-element relationships, and output uncertainty information with strong robustness is urgently needed. SUMMARY

[0007] The present application provides a multi-element grade prediction method based on multi-task cooperation and covariance modeling to solve the problem that existing methods of predicting ore grade using machine learning cannot effectively and accurately detect ore grade.

[0008] To achieve the above purpose, the present application realizes the technical scheme as follows: The present application provides a multi-element grade prediction method based on multi-task cooperation and covariance modeling, comprising the following steps: Step 1: Constructing an ore grade prediction model, the ore grade prediction model comprising a feature encoding layer, a fusion processing layer and an output layer, the feature encoding layer comprising a plurality of encoders set for multi-modal data of the ore, the fusion processing layer comprising a fusion block set based on an attention mechanism, and the output layer being built based on a multi-layer perception mechanism; Step 2: Obtaining multi-modal data of the ore to be detected, inputting the multi-modal data into the ore grade prediction model, first acquiring corresponding single-modal features through the plurality of encoders of the feature encoding layer, obtaining fusion features based on the single-modal features and the fusion block of the fusion processing layer, obtaining a mean vector and a covariance matrix based on the fusion features of the output layer, and performing a prediction distribution based on the mean vector and the covariance matrix to obtain a prediction distribution result, analyzing the ore grade data and regulating the process flow according to the prediction distribution result.

[0009] Through the above design, a covariance modeling mechanism under label missing is designed for the case of missing part of the element assay results in the ore sample. The loss component corresponding to the missing label is dynamically shielded during the training process, while its contribution to the covariance matrix is retained. This method ensures effective training of the model on incomplete data sets and improves the usability in actual industrial scenarios.

[0010] Further, the multi-modal data comprises ore spectral data, process time series data and environmental metadata; the multi-modal data is input into the ore grade prediction model after preprocessing, and the preprocessing comprises spectral resampling and normalization processing.

[0011] Further, the ore spectral data comprises X-ray fluorescence spectral data and near-infrared spectral data; the plurality of encoders comprises a spectral data encoder set for the ore spectral data, and the spectral data encoder is built based on a pre-trained deep convolutional neural network.

[0012] Further, the process time series data includes grinding data, flotation data and classification data, the grinding data includes current, voltage, power consumption, feed amount, discharge amount and grinding product particle size of the grinding device in operation, the flotation data includes flotation tank gas amount and stirring speed of the flotation device, and the classification data includes overflow speed, underflow speed and cyclone pressure of the classification device. The plurality of encoders includes time series data encoders configured for grinding data, flotation data and classification process data, and the time series data encoders are constructed based on a pre-trained time series model.

[0013] Further, the environmental metadata includes sampling depth, sampling time, environmental humidity and environmental temperature, and the plurality of encoders includes a metadata encoder configured for environmental metadata. The metadata encoder is constructed based on a pre-trained embedding network.

[0014] Through the above design, the ore spectrum, process time series data and environmental metadata are fused, the multi-modal attention mechanism and covariance modeling are used, the collaborative and accurate prediction of multiple element grades is realized, the prediction accuracy and efficiency are significantly improved, the correlation between elements is explicitly described and the prediction uncertainty is provided, and the adaptive capacity for missing label data is good, effectively meeting the needs of intelligent and high-reliability grade prediction technology in complex industrial sites.

[0015] Further, the fusion processing layer unifies the feature dimensions of all single-modal features, calculates the attention scores of each single-modal feature through a fusion block, and constructs the fusion features after fusion weighting based on the attention scores.

[0016] Further, the step of obtaining the mean vector and the covariance matrix based on the fusion features by the output layer specifically includes: obtaining the mean vector based on the linear conversion of the fusion features by the output layer, obtaining the diagonal elements and the non-diagonal elements based on the fusion features combined with a multi-layer perceptron, obtaining a lower triangular matrix according to the diagonal elements and the non-diagonal elements, and obtaining the covariance matrix based on the lower triangular matrix combined with a transpose operation.

[0017] Further, the step of obtaining the diagonal elements and the non-diagonal elements based on the fusion features combined with a multi-layer perceptron specifically includes: obtaining a plurality of elements based on the fusion features combined with a multi-layer perceptron, obtaining the diagonal elements by screening a predetermined number of elements and combining a softplus activation, and obtaining the non-diagonal elements by filling the non-diagonal part of the lower triangular matrix with the remaining elements after screening.

[0018] Further, the loss function of the ore grade prediction model is constructed based on the multivariate Gaussian negative log-likelihood loss of the prediction distribution and the true label.

[0019] Through the above design, the multivariate Gaussian negative log-likelihood loss is introduced into the loss function, and the prediction mean and the covariance matrix between elements are learned simultaneously. Through the constraint of the covariance term, the model can explicitly capture the correlation between different metal element grades, improve the prediction stability and adaptability under complex geological conditions, and better depict the statistical dependence between elements.

[0020] Beneficial effects: The multi-task cooperative and covariance modeling multi-element grade prediction method provided by the application proposes a multi-target cooperative prediction architecture based on deep learning, which can face various complex ore systems and realize joint modeling and distributed prediction of multi-element grades.

[0021] Firstly, the application introduces multi-modal information such as ore spectral data, process time series data and environmental metadata into the model, extracts key features in different modalities using pre-trained spectral encoders, time series encoders and metadata encoders, and realizes dynamic weighted fusion of multi-source features through a soft fusion module of attention mechanism, so as to fully mine the multi-dimensional information of ore and improve the integrity and effectiveness of feature expression.

[0022] Secondly, the application introduces a covariance modeling mechanism in the output layer, learns the prediction mean vector and the covariance matrix simultaneously, constructs the prediction distribution of multi-element grade, and realizes the expansion from traditional "point estimation" to "distribution estimation". The prediction distribution can explicitly reflect the correlation and constraint relationship between different elements, and quantitatively describe the uncertainty of the prediction result, so that the model can not only provide the most possible prediction value of multi-element, but also provide the prediction confidence interval and related structure information.

[0023] Thirdly, through the constraint of the covariance matrix, the model can capture the symbiotic, associated or antagonistic characteristics between multi-elements, and improve the stability and robustness in the scene of complex ore composition, strong volatility and incomplete labels. Especially in the case of partial sample label loss, the label loss covariance modeling mechanism designed by the application can dynamically shield the loss component corresponding to the missing label, while retaining its contribution to the covariance matrix, ensuring effective training of the model on incomplete data sets.

[0024] In summary, the multi-task cooperative and covariance modeling method of the application can realize cooperative high-precision prediction of multi-element grade, which can not only improve the accuracy and interpretability of the prediction result, but also provide quantitative support for beneficiation process control, ore blending optimization and intelligent decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The network structure diagram of the ore grade prediction model of the embodiment of the application is shown in the figure. Figure 2A flowchart of an implementation process of the ore grade prediction model of the embodiment of the present application. DETAILED DESCRIPTION

[0026] The technical solutions of the present application will be described below clearly and completely. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0027] Unless otherwise defined, the technical terms or scientific terms used in the present application shall have the usual meanings understood by those skilled in the art. The terms "first", "second", and similar words used in the present application do not represent any order, number, or importance, but are only used to distinguish different components. Similarly, the terms "one" or "a" and similar words do not represent a quantity limitation, but represent the existence of at least one. The terms "connected" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The terms "up", "down", "left", "right", and the like are only used to represent relative positional relationships, which change accordingly when the absolute position of the described object changes.

[0028] Please refer to Figure 1 The embodiment of the present application provides a multi-element grade prediction method for multi-task cooperation and covariance modeling, which comprises the following steps: Step 1: Constructing an ore grade prediction model, the ore grade prediction model comprising a feature encoding layer, a fusion processing layer, and an output layer, the feature encoding layer comprising a plurality of encoders configured for multi-modal data of the ore, the fusion processing layer comprising a fusion block configured based on an attention mechanism, and the output layer being configured based on a multi-layer perception mechanism; Among them, the plurality of encoders of the feature encoding layer comprises a spectral data encoder configured for spectral data of the ore, the spectral data encoder being configured based on a pre-trained deep convolutional neural network, a time series data encoder configured for grinding, flotation, and classification process, the time series data encoder being configured based on a pre-trained time series model, a metadata encoder configured for mining environment, the metadata encoder being configured based on a pre-trained embedding network, and the specific encoder configuration effectively realizes feature extraction and more effectively expresses the characteristics of the ore in different aspects, so that the subsequent ore grade prediction is more accurate.

[0029] The multi-modal data comprises ore spectral data, process time series data, and environmental metadata. The ore spectral data include X-ray fluorescence spectral data and near-infrared spectral data, wherein the X-ray fluorescence spectral data are obtained by detecting the element composition and element content of the ore by using an X-ray fluorescence spectrometer, and the near-infrared spectral data are obtained by detecting the mineral composition, structural water and hydroxyl of the ore by using a near-infrared tube. The process timing data include grinding data, flotation data and classification data, the grinding data include the current, voltage, power consumption, feed amount, discharge amount and grinding product particle size of the grinding device during work, the flotation data include the flotation tank gas amount and stirring speed of the flotation device, and the classification data include the overflow speed, underflow speed and cyclone pressure of the classification device, and the timing data are obtained by window aggregation when the ore spectral data are measured; The environmental metadata include sampling depth, sampling time, environmental humidity and environmental temperature, and the metadata are obtained by manual recording; In the embodiment, the multi-modal data of the tungsten ore are collected, 4096 energy points are obtained by detection of the X-ray fluorescence spectrometer, 2151 wavelength points are obtained by detection of the near-infrared spectrometer, and the grinding data, flotation data and classification data are collected by process sensors, and 60 seconds of data are taken.

[0030] After the collection, the multi-modal data are preprocessed, and the preprocessing specifically includes spectral resampling of the spectral data and normalization of the multi-modal data, and in this case, a uniform sampling grid with a size of 1024 is defined by spectral resampling, and the 4096 energy points detected by the X-ray fluorescence spectrometer and the 2151 wavelength points detected by the near-infrared spectrometer are unified to the same number of data points to realize data alignment.

[0031] A loss function of the ore grade prediction model is constructed based on a multivariate Gaussian negative log-likelihood loss of a prediction distribution and a true label, and is specifically represented by the following formula: ; Wherein, represents the loss function, represents the true label of the ore sample, and ; represents the mean in the prediction distribution; represents the transposition operation; represents the covariance matrix in the prediction distribution.

[0032] ​Step 2: Obtain the multi-modal data of the ore to be detected, input the multi-modal data into the ore grade prediction model, first obtain the corresponding single-modal features through the multiple encoders of the feature encoding layer, obtain the fusion features based on the single-modal features and the fusion block in the fusion processing layer, obtain the mean vector and the covariance matrix based on the fusion features in the output layer, and perform prediction distribution based on the mean vector and the covariance matrix to obtain the prediction distribution result, analyze the ore grade data according to the prediction distribution result, and control the process flow.

[0033] Please refer to Figure 2 , after the multi-modal data is preprocessed and input into the ore grade prediction model, the preprocessing includes spectral resampling of spectral data and normalization of multi-modal data, then the multiple encoders of the ore grade prediction model are used for feature extraction to obtain multiple single-modal features, in this embodiment, the feature dimensions of the multiple single-modal features extracted by the fusion processing layer are all unified to 128, the fusion block calculates the attention score of each single-modal feature by using soft attention fusion, and the fusion feature after fusion weighting is constructed based on the attention score.

[0034] After obtaining the fusion features, the output layer obtains the mean vector based on the fusion features and linear conversion. ; Wherein, represents a learnable weight matrix, and ; represents the fusion features of the i-th ore sample; represents a learnable bias, and ; represents the number of elements to be predicted, which is 3 here.

[0035] The output layer obtains a plurality of elements according to the set multi-layer perceptron, which is represented by the following formula: ; Wherein, represents the element grade of the j-th ore sample; represents a multi-layer perceptron; here .

[0036] The first three elements are selected from the obtained plurality of elements, and the diagonal elements are obtained by combining the softplus activation, which is represented by the following formula: ; Wherein, represents the j-th component, , represents the obtained diagonal elements; ​​This represents the numerical stability constant, which is a very small positive number to avoid the diagonal elements being calculated close to 0, leading to a singular covariance matrix. The off-diagonal elements are obtained by filling the off-diagonal portion of the lower triangle of the matrix with the remaining selected elements, as expressed by the following formula: ; in, This indicates the non-diagonal element being retrieved; This represents the remaining elements after selection. ; and These represent the row and column indices of the matrix, respectively; in this embodiment... , ,and .

[0037] Here, to clarify the filling order of the off-diagonal elements in the lower triangle, let the total number of elements to be predicted be . The model outputs from the multilayer perceptron. scalars as vectors Store, before One element is activated by softplus to become a diagonal element, and the remaining elements are... The elements are filled into the non-diagonal positions of the lower triangle in row-major order.

[0038] Based on diagonal elements and off-diagonal elements Obtain the lower triangular matrix Based on the lower triangular matrix The covariance matrix is ​​obtained by combining the transpose operation; here, we first obtain the lower triangular matrix. The reason is that the covariance matrix must satisfy symmetry and positive semidefiniteness. To avoid the problem that directly predicting the covariance matrix is ​​difficult to satisfy symmetry and positive semidefiniteness, Cholesky decomposition is used here, expressed by the following formula: ; in, This represents the obtained covariance matrix; Further from the mean vector With covariance matrix The obtained predicted distribution results are represented as follows: ; in, This represents the elemental grade predicted by the ore grade prediction model. In this embodiment, six elements are acquired from the multilayer perceptron. The first three elements are selected and combined with softplus activation to obtain the diagonal elements. The remaining three elements are used as off-diagonal elements. The lower triangular matrix is ​​then obtained from the diagonal and off-diagonal elements. The covariance matrix is ​​further obtained through transpose operation. From the obtained covariance matrix, the uncertainty and correlation of each element prediction can be obtained. The diagonal elements reflect the single-element prediction variance and can be used to evaluate the prediction confidence interval. The off-diagonal elements reflect the synergistic or restrictive relationship between different elements and can be used to analyze the symbiotic characteristics of multiple elements.

[0039] Specifically, the predicted distribution can provide information not only about the tungsten content in tungsten ore (… ),copper( ) and sulfur ( The most likely grade estimates for the three elements can also reflect the statistical correlation and prediction uncertainty among the predicted values ​​of different elements. This output has a direct guiding role in the actual production and operation of tungsten ore beneficiation plants.

[0040] The mean vector represents the expected value of each element's grade and can be used to monitor the trend of changes in raw ore grade in real time. When predicting... When the mean value decreases, it indicates that the raw ore has become dilute. The concentrator can accordingly increase the grinding fineness, appropriately increase the amount of flotation reagents, or extend the flotation time to maintain concentrate quality and recovery rate. When the predicted mean increases, it indicates that the raw ore is enriched. The amount of reagent added or the flotation intensity can be appropriately reduced to reduce reagent consumption and energy consumption.

[0041] at the same time, and The predicted average can also serve as an important reference for process adjustments: if An increase in the predicted mean indicates enhanced activity of copper minerals, which can be addressed by optimizing the collector ratio and adjusting the stirring intensity; if An increase in the predicted mean indicates an increase in the proportion of pyrite or sulfide ores. The impact of sulfides on scheelite flotation should be reduced by increasing the pH or adding inhibitors such as lime or sodium sulfide.

[0042] Furthermore, the mean prediction results can also provide a quantitative basis for ore blending in different mining areas, such as when a certain mining area... When the content is high but the sulfur content is also high, it can be blended with low-sulfur ore sources to achieve stabilization of the grade of the ore feed and stable production control.

[0043] On the other hand, the covariance matrix provides information on the correlation and prediction uncertainty among its elements. The diagonal elements reflect the variance of the model's predictions for each element, which can be used to assess the confidence level of the prediction results. When the prediction variance is large, it indicates that the model's prediction reliability for the current operating condition is reduced. The operating system can automatically trigger a "steady-state hold" strategy, temporarily refraining from adjusting process parameters and increasing the frequency of online monitoring. When the variance is small and stable, an adaptive control mode can be enabled to dynamically optimize flotation aeration rate, stirring speed, or reagent dosage. Off-diagonal elements reflect the synergistic or restrictive relationships between different elements. For example, when the prediction result... When the covariance is positive, it indicates a synergistic enrichment relationship between tungsten and copper, making a combined flotation reagent system suitable to improve the overall recovery rate; when... When the covariance is negative, it indicates that sulfides have an inhibitory effect on scheelite flotation, and the pH value should be adjusted appropriately or sulfide ore inhibition measures should be strengthened. A strong positive correlation in covariance indicates strong copper-sulfur symbiosis, which allows for optimization of the type and ratio of collectors to improve sorting selectivity.

[0044] Furthermore, by continuously monitoring changes in the structure of the covariance matrix, process status monitoring and anomaly diagnosis can also be achieved. When the structure of the covariance matrix changes significantly within a short period of time, such as... The correlation suddenly weakened or When the correlation reverses, it may indicate a sudden change in ore properties or abnormal equipment operation. The system can issue an alarm signal based on this, prompting the user to check the classification pressure, flotation air volume, or raw ore particle size distribution.

[0045] In summary, the predicted distribution output in this embodiment provides the following three aspects of decision support for the beneficiation plant's production: 1. Real-time grade prediction and process control based on mean vector to achieve adaptive optimization of feed flow and reagent regime; 2. Uncertainty analysis and element correlation characterization based on covariance matrix provide guidance for risk assessment and coupling relationships in the flotation process; 3. An anomaly detection mechanism based on predictive distribution dynamic changes provides data support for production stability maintenance and equipment condition diagnosis.

[0046] Therefore, the mean vector and covariance matrix output by the ore grade prediction model can enable the beneficiation plant to transform from "single grade prediction" to "multi-element collaborative modeling and intelligent decision support", which can significantly improve the automation, stability and economic benefits of the tungsten ore beneficiation process.

[0047] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A multi-element grade prediction method based on multi-task collaboration and covariance modeling, characterized in that, Includes the following steps: Step 1: Construct an ore grade prediction model. The ore grade prediction model includes a feature encoding layer, a fusion processing layer, and an output layer. The feature encoding layer includes multiple encoders set for multimodal data of the ore. The fusion processing layer includes a fusion block set based on an attention mechanism. The output layer is constructed based on a multilayer perceptron. Step 2: Obtain multimodal data of the ore to be tested. Input the multimodal data into the ore grade prediction model. First, obtain the corresponding single-modal features through multiple encoders in the feature encoding layer. The fusion processing layer obtains fusion features based on the single-modal features and the fusion block. The output layer obtains the mean vector and covariance matrix based on the fusion features, and performs prediction distribution based on the mean vector and covariance matrix to obtain the prediction distribution results. Analyze the ore grade data and adjust the process flow according to the prediction distribution results.

2. The multi-element grade prediction method based on multi-task collaboration and covariance modeling according to claim 1, characterized in that, The multimodal data includes ore spectral data, process time series data, and environmental metadata; the multimodal data is preprocessed and then input into the ore grade prediction model, the preprocessing including spectral resampling and normalization.

3. The multi-element grade prediction method based on multi-task collaboration and covariance modeling according to claim 2, characterized in that, The ore spectral data includes X-ray fluorescence spectral data and near-infrared spectral data; the multiple encoders include a spectral data encoder designed for the ore spectral data, which is constructed based on a pre-trained deep convolutional neural network.

4. The multi-element grade prediction method based on multi-task collaboration and covariance modeling according to claim 2, characterized in that, The process timing data includes grinding data, flotation data, and classification data. The grinding data includes the current, voltage, power consumption, feed rate, discharge rate, and grinding product particle size of the grinding equipment during operation. The flotation data includes the air volume and stirring speed of the flotation cell of the flotation equipment. The classification data includes the overflow velocity, underflow velocity, and hydrocyclone pressure of the classification equipment. The plurality of encoders includes a time-series data encoder configured for grinding data, flotation data, and classification process data, the time-series data encoder being constructed based on a pre-trained time-series model.

5. The multi-element grade prediction method based on multi-task collaboration and covariance modeling according to claim 2, characterized in that, The environmental metadata includes sampling depth, sampling time, ambient humidity, and ambient temperature, and the multiple encoders include a metadata encoder set for the environmental metadata; The metadata encoder is built based on a pre-trained embedding network.

6. The multi-element grade prediction method based on multi-task collaboration and covariance modeling according to any one of claims 1-5, characterized in that, The fusion processing layer unifies the feature dimensions of all single-modal features and calculates the attention score of each single-modal feature through the fusion block, and constructs the fused weighted feature based on the attention score.

7. The multi-element grade prediction method based on multi-task collaboration and covariance modeling according to any one of claims 1-5, characterized in that, The steps of obtaining the mean vector and covariance matrix based on the fusion features in the output layer specifically include: the output layer obtains the mean vector after linear transformation based on the fusion features, then obtains the diagonal and off-diagonal elements based on the fusion features and a multilayer perceptron, obtains the lower triangular matrix based on the diagonal and off-diagonal elements, and obtains the covariance matrix based on the lower triangular matrix and a transpose operation.

8. The multi-element grade prediction method based on multi-task collaboration and covariance modeling according to claim 7, characterized in that, The steps for obtaining diagonal and off-diagonal elements based on fusion features and multilayer perceptron specifically include: obtaining several elements based on fusion features and multilayer perceptron, selecting a predetermined number of elements from them and then using softplus activation to obtain diagonal elements, and filling the remaining elements into the off-diagonal part of the lower triangle of the matrix to obtain off-diagonal elements.

9. The multi-element grade prediction method based on multi-task collaboration and covariance modeling according to claim 8, characterized in that, The loss function of the ore grade prediction model is constructed based on the multivariate Gaussian negative log-likelihood loss of the predicted distribution and the true label.

Images