Intelligent dosing method and system for lead zinc ore flotation

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an intelligent dosing method and system for lead zinc ore flotation to improve the efficiency, stability and economic benefits of a flotation process. The method comprises the following steps: calculating a first cross-modal feature and a second cross-modal feature, fusing the first cross-modal feature and the second cross-modal feature with a bubble image feature and an element grade feature through residual connection, and then processing to obtain a multi-modal feature sequence; inputting the multi-modal feature sequence of each time step into an LSTM network to obtain a corresponding LSTM hidden state sequence; calculating the time attention weight of the hidden state sequence of each time step in the time window, and performing weighted summation on the corresponding LSTM hidden state sequence according to the time attention weight to obtain a context vector in the time window; and finally, sharing the weighted and summed context vector serving as an input to a full connection layer corresponding to the two detection heads so as to respectively obtain probability distribution of predicted medicament types and a dose value of each medicament.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and automatic control, and particularly relates to an intelligent dosing method and system for lead-zinc ore flotation. Background Art

[0002] In the mining industry, lead-zinc ore flotation is a core process aimed at efficiently separating valuable metals from waste rock in ores. During the flotation process, reasonable dosing of reagents is crucial for improving the recovery rate of target minerals and the process efficiency. However, existing flotation dosing methods have significant deficiencies in accuracy, real-time performance, and intelligence, severely restricting the stability and economic benefits of the flotation process.

[0003] Firstly, traditional dosing methods mainly rely on the experience and subjective judgment of operators. This manual operation method is easily affected by the skill level, working state, and experience deviation of operators, resulting in instability and inaccuracy of reagent dosing, thereby reducing the flotation effect. Insufficient reagent dosage may lead to a decrease in the recovery rate of target minerals, while excessive reagent dosage may cause problems such as resource waste, increased production costs, and environmental pollution.

[0004] Secondly, although some semi-automatic dosing devices (such as self-service dosing machines) have emerged in the industry, they still highly rely on manual experience - operators need to preset the dosing time interval and reagent dosage by observing bubble images and element grade data. This mode relying on manual setting lacks the ability to perceive the dynamic changes in the flotation process in real time, cannot make timely adjustments according to the changes in the properties of the pulp in the flotation process, and is difficult to achieve true closed-loop control and efficient operation.

[0005] In addition, the existing dosing process has a low level of intelligence and cannot meet the dynamic regulation requirements of the flotation process. During the flotation process, the properties of the pulp and flotation conditions may fluctuate continuously with changes in raw material quality or environmental factors. Due to strong hysteresis, traditional dosing methods are difficult to intelligently adjust the reagent type and dosage, resulting in increased volatility of the flotation effect, especially with low efficiency in large-scale continuous production.

[0006] In summary, there is still much room for improvement in the accuracy, real-time performance, and intelligence of existing flotation dosing methods. Therefore, there is an urgent need for a new dosing method and device that combines advanced technologies and intelligent control algorithms to achieve real-time monitoring and dynamic regulation of the flotation process, thereby improving the efficiency, stability, and economic benefits of the flotation process. Summary of the Invention

[0007] The object of the present invention is to disclose an intelligent dosing method and system for lead-zinc ore flotation to improve the efficiency, stability, and economic benefits of the flotation process.

[0008] To achieve the above object, the intelligent dosing method for lead-zinc ore flotation disclosed by the present invention includes: Step S1: During the lead-zinc ore flotation process, extract the bubble image features based on the bubble images, and at the same time, collect the grade data of at least two metal elements in the pulp sample in the process in real time through an on-line analyzer to construct the element grade features; Step S2: Calculate the query vector, key vector, and value vector of the bubble image features and the element grade features based on the attention mechanism respectively; Step S3: Based on the key vector and value vector of the element grade features, and the first cross-modal feature calculated by the cross-attention mechanism using the query vector of the bubble image features, and at the same time, based on the key vector and value vector of the bubble image features, and the second cross-modal feature calculated by the cross-attention mechanism using the query vector of the element grade features, after fusing the first cross-modal feature, the second cross-modal feature, the bubble image features, and the element grade features through residual connection, perform layer normalization processing to obtain the final multi-modal feature sequence; Step S4: In any time window, input the multi-modal feature sequence of each time step into the LSTM network to obtain the LSTM hidden state sequence corresponding to each time step; Step S5: Calculate the temporal attention weights of the hidden state sequences corresponding to each time step within the time window, and perform weighted summation on the corresponding LSTM hidden state sequences according to the temporal attention weights to obtain the context vector within the time window; Step S6: Use the context vector after weighted summation as the input to be shared by the fully connected layers corresponding to the two detection heads to respectively obtain the probability distribution of the predicted reagent types and the dosage values of each reagent.

[0009] Preferably, in step S1, based on the original image of 224×224×3 input, extract 64-dimensional image features through the feature extraction network of the pre-trained ResNet-18.

[0010] Preferably, in step S2, calculate the query vector, key vector, and value vector of the bubble image features and the element grade features based on the attention mechanism through independent linear transformation. The specific calculation process is as follows:

[0011] Among them, is the 64-dimensional image feature; are trainable different weight matrices, and ; are respectively the query vector, key vector, and value vector of the bubble image features based on the attention mechanism;

[0012] Among them, is the 22-dimensional element grade feature, is the different weight matrices that can be trained, and ; are respectively the query vector, key vector, and value vector of the element grade feature based on the attention mechanism.

[0013] Preferably, the calculation formula of step S3 specifically includes: ; ; Among them, is the first cross-modal feature, is the second cross-modal feature, and are hyperparameter temperature coefficients, and the superscript is the matrix transpose symbol.

[0014] Preferably, the calculation formula of step S5 specifically includes: ; ; ; Among them, is the time window length, is the hidden state sequence output by the LSTM network corresponding to the time step ; is the weight matrix that can be trained; is the bias term that can be trained; is the weight of the time step in the time window; is the normalized weight after conversion based on the Softmax function; is the context vector within the time window obtained after weighted summation.

[0015] Preferably, the specific calculation formula of step S6 includes: ; Among them, is the classification task weight matrix; is the classification task bias term; is the linear output of the classification task; is the probability of the reagent type; is the natural constant, represents the standard deviation.

[0016] ; Among them, is the regression task weight matrix; is the regression task bias term; is the predicted value of the drug dosage.

[0017] Preferably, in the process of training the model for implementing steps S2 to S6, the loss function is composed of the weighted sum of the multi-label binary classification cross-entropy loss of the drug type information and the mean square loss of the drug dosage, and the loss gradient of the model backpropagation is calculated by taking the derivative of the loss.

[0018] To achieve the above object, the present invention also discloses an intelligent drug addition system for lead-zinc ore flotation, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the above method is implemented.

[0019] The present invention can improve the efficiency, stability, and economic benefits of the flotation process, and has the following beneficial effects: 1. First, calculate the query vector, key vector, and value vector based on the attention mechanism for the multi-dimensional bubble image features and element grade features, and then perform a unique cross-modal attention calculation on the two types of features, specifically including: the first cross-modal feature calculated through the cross-attention mechanism according to the key vector and value vector based on the element grade features and the query vector based on the bubble image features. At the same time, the second cross-modal feature calculated through the cross-attention mechanism according to the key vector and value vector based on the bubble image features and the query vector based on the element grade features. After fusing the first cross-modal feature, the second cross-modal feature, the bubble image features, and the element grade features through residual connection, and then performing layer normalization processing to obtain the final multi-modal feature sequence. Based on this multi-modal feature sequence, compared with omitting the unique cross-modal attention calculation for the two types of features during the fusion process of the bubble image features and the element grade features, the interaction method between features is optimized, and it can better capture more key features based on overall collaboration at each time step for the real-time state of the flotation process.

[0020] 2. By calculating the temporal attention weights of the corresponding hidden state sequences at each time step within the time window, and performing weighted summation on the corresponding LSTM hidden state sequences according to the temporal attention weights to obtain the context vector within the time window, the key operation time points (i.e., the time step with the largest temporal attention weight in the hidden state sequence obtained by LSTM) can be accurately identified, thereby improving the prediction accuracy of the subsequent probability distribution of the drug types and the dosage values of each drug.

[0021] 3. Predict the probability distribution of the drug types and the dosage values of each drug based on the same context vector within the time window, which solves the problem of insufficient collaboration caused by the separation of the two types of tasks in the traditional method.

[0022] Next, the present invention will be further described in detail with reference to the accompanying drawings. Brief Description of the Drawings

[0023] The drawings forming a part of this application are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 It is a schematic flow diagram of an intelligent dosing method for lead-zinc ore flotation disclosed in an embodiment of the present invention. Detailed Embodiments

[0024] The embodiments of the present invention will be described in detail below with reference to the drawings. However, the present invention can be implemented in many different ways defined and covered by the claims.

[0025] Embodiment 1 This embodiment discloses an intelligent dosing method for lead-zinc ore flotation, as Figure 1 shown, including: Step S1: During the lead-zinc ore flotation process, extract the bubble image features based on the bubble image, and at the same time, collect the grade data of at least two metal elements in the pulp sample in the process in real time through an on-line analyzer to construct the element grade features.

[0026] Preferably, in this step, based on the original image of 224×224×3 input, 64-dimensional image features are extracted through the feature extraction network of the pre-trained ResNet-18.

[0027] For the original training data set of the bubble image, several data augmentation methods can be randomly selected to expand the training set. The relevant augmentation methods include but are not limited to operations such as rotation, flipping, and cropping. By performing data augmentation on the bubble image, not only can diverse transformations and noises be introduced, thereby improving the generalization ability of the model and enabling it to better adapt to the bubble image in different scenarios and conditions, but also more abundant training samples can be generated, effectively expanding the scale of the training data set and alleviating the problem of scarce data distribution.

[0028] In specific deployment, on the flotation process production line, the grade data of multiple metal elements (which can be referred to the subsequent specific examples) in the pulp sample in key processes such as roughing, cleaning, and scavenging can be collected in real time through an on-line grade analyzer to construct multi-dimensional element grade features.

[0029] Step S2: Calculate the query vector, key vector, and value vector based on the attention mechanism for the bubble image features and the element grade features respectively.

[0030] In this step, the query vector, key vector, and value vector based on the attention mechanism for the bubble image features and the element grade features can be calculated through independent linear transformation. The specific calculation process is as follows:

[0031] Among them, is a 64-dimensional image feature; are different trainable weight matrices, and ; are the query vector, key vector, and value vector of the bubble image feature based on the attention mechanism, respectively.

[0032]

[0033] Among them, is a 22-dimensional element grade feature, are different trainable weight matrices, and ; are the query vector, key vector, and value vector of the element grade feature based on the attention mechanism, respectively.

[0034] Step S3: The first cross-modal feature calculated through the cross-attention mechanism based on the key vector and value vector of the element grade feature, and the query vector of the bubble image feature, and at the same time, the second cross-modal feature calculated through the cross-attention mechanism based on the key vector and value vector of the bubble image feature, and the query vector of the element grade feature. After fusing the first cross-modal feature, the second cross-modal feature with the bubble image feature and the element grade feature through residual connection, and then performing layer normalization processing, the final multi-modal feature sequence is obtained.

[0035] Referring to the subsequent steps, the features in the multi-modal feature sequence fused in this step are aligned based on time steps.

[0036] The specific calculation formula of this step includes: ; ; Among them, is the first cross-modal feature, is the second cross-modal feature, and are hyperparameter temperature coefficients. Optionally, , , the superscript is the matrix transpose symbol.

[0037] The fused multi-modal feature sequence can be expressed as: ; Among them, is the weight coefficient, and LayerNorm is the normalization function. Optionally, , which means we assign greater weight to the element grade characteristics and let them play a dominant role in feature calculation.

[0038] Step S4: In any time window, input the multi-modal feature sequence of each time step into the LSTM network to obtain the LSTM hidden state sequence corresponding to each time step.

[0039] For example, input the multi-modal feature sequence , then after LSTM processing, the output is , where . Optionally, the time window T = 5.

[0040] In this step, since the hidden state dimension of the LSTM can be independent of the input dimension, that is, the hidden state dimension can be customized. Therefore, in this embodiment, the 86 (i.e., the sum of 64 + 22) -dimensional features are mapped to 128 dimensions, which can enhance the expression ability of the features and facilitate the subsequent processing of the time attention mechanism and the task head (classification and regression).

[0041] Step S5: Calculate the time attention weights of the hidden state sequences corresponding to each time step within the time window, and perform weighted summation on the corresponding LSTM hidden state sequences according to the time attention weights to obtain the context vector within the time window.

[0042] The specific calculation formula for this step includes: ; ; ; where is the time window length, is the hidden state sequence output by the LSTM network corresponding to the time step ; is the trainable weight matrix; is the trainable bias term; is the weight of the time step in the time window; is the normalized weight after conversion based on the Softmax function; is the context vector obtained after weighted summation within the time window.

[0043] In the dimension processing of this embodiment, although the bubble image and the grade data are used as the basis for adding medicine, the weight of the grade data is greater. Therefore, the dimension of the image features is intentionally reduced to prevent the element grade features from being submerged when cross-attention calculation is performed between the bubble image features and the element grade features. In addition, during the cross-modal feature attention calculation, the key vector of the element grade features is divided by 0.5 to strengthen its attention weight, making the element grade features play a dominant role. On the contrary, the key vector of the bubble image features is divided by 2 to weaken its attention weight, making the bubble image features play an auxiliary role.

[0044] Step S6: Use the context vector after weighted summation as the input and share it with the fully connected layers corresponding to the two detection heads to obtain the probability distribution of the predicted medicine types and the dosage values of each medicine respectively.

[0045] Preferably, the specific calculation formula for this step includes: ; where, is the weight matrix for the classification task; is the bias term for the classification task; is the linear output of the classification task; is the probability of the medicine type; is the natural constant, represents the standard deviation.

[0046] ; where, is the weight matrix for the regression task; is the bias term for the regression task; is the predicted value of the medicine dosage.

[0047] In this embodiment, during the process of training the model for implementing steps S2 to S6, the loss function can be composed of the weighted sum of the multi-label binary classification cross-entropy loss of the medicine type information and the mean square loss of the medicine dosage, and the loss gradient of the model's backpropagation is calculated by taking the derivative of the loss.

[0048] Optionally, the multi-label binary classification cross-entropy loss of the medicine type information , and its formal representation can be: ; where, represents the number of batch samples, is the index of the medicine type (illustrated with 10 medicines as an example), is the th sample th true label of the th The predicted probability of the th medicament in the

[0049] samples, with values in the range [0, 1]. Optionally, calculate the mean squared loss of the medicament dosage The formal expression can be: ; where represents the true dosage value of the th medicament in the th sample, and the th predicted dosage value of the th medicament in the

[0050] The total loss function can be: ; where λ is the loss weight coefficient, and its value can be 0.5. Further, take the derivative of the final loss to calculate the loss gradient, and backpropagate this gradient in the network model to optimize the model parameters. The calculation formula can be: ; where is the loss gradient parameter to be learned by the network model in the th iteration process; represents the learning rate.

[0051] Among them, when annotating the bubble image (including the enhanced image) and the corresponding chemical addition information based on this embodiment, the association data between the bubble image (including the original data and the enhanced data), the raw ore Zn grade, the raw ore Pb grade, the total lead concentrate Zn grade, the total lead concentrate Pb grade, the zinc concentrate Zn grade, the zinc concentrate Pb grade, the zinc scavenging three tailings Zn grade, the zinc scavenging three tailings Pb grade, the high lead concentrate Zn grade, the high lead concentrate Pb grade, the low lead concentrate Zn grade, the low lead concentrate Pb grade, the lead quick rough tailings Zn grade, the lead quick rough tailings Pb grade, the zinc scavenging four bubbles Zn grade, the zinc scavenging four bubbles Pb grade, the zinc scavenging four tailings Zn grade, the zinc scavenging four tailings Pb grade, the lead tailings Zn grade, the lead tailings Pb grade, the zinc quick rough tailings Zn grade, the zinc quick rough tailings Pb grade and the chemical agent type sequence and the chemical agent dosage sequence is established. During the annotation process of a real chemical agent and dosage sequence, the first 10 elements represent the chemical agent type sequence, and the elements from the 11th to the 20th represent the chemical agent dosage sequence. For example: the first element in the chemical agent sequence represents copper sulfate, the second element represents SD-11, the third element represents ethyl thionocarbamate, the fourth element represents cassiterite activator, the fifth element represents lead nitrate, the sixth element represents No. 2 oil, the seventh element represents GZT, the eighth element represents butyl xanthate, the ninth element represents SND-21, and the tenth element represents cassiterite collector. In addition, they are all binary classification data, where 0 represents none and 1 represents presence. Correspondingly, the first element in the dosage sequence represents the weight of copper sulfate, the second element represents the weight of SD-11, the third element represents the weight of ethyl thionocarbamate, the fourth element represents the weight of cassiterite activator, the fifth element represents the weight of lead nitrate, the sixth element represents the weight of No. 2 oil, the seventh element represents the weight of GZT, the eighth element represents the weight of butyl xanthate, the ninth element represents the weight of SND-21, and the tenth element represents the weight of cassiterite collector; among them, the dosage data are usually floating-point numbers, and the weight unit is kg.

[0052] Embodiment 2 This embodiment discloses an intelligent chemical addition system for lead-zinc ore flotation, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method disclosed in the above embodiment is implemented.

[0053] In summary, the methods and systems respectively disclosed in the above two embodiments of the present invention can improve the efficiency, stability and economic benefits of the flotation process, and have the following beneficial effects: 1. First, calculate the query vector, key vector, and value vector for the multi-dimensional bubble image features and element grade features based on the attention mechanism. Then, perform a unique cross-modal attention calculation on the two types of features, specifically including: the first cross-modal feature calculated through the cross-attention mechanism based on the key vector and value vector of the element grade features and the query vector of the bubble image features. At the same time, the second cross-modal feature calculated through the cross-attention mechanism based on the key vector and value vector of the bubble image features and the query vector of the element grade features. After fusing the first cross-modal feature, the second cross-modal feature, the bubble image features, and the element grade features through residual connection, perform layer normalization to obtain the final multi-modal feature sequence. Based on this multi-modal feature sequence, compared with omitting the unique cross-modal attention calculation for the two types of features during the fusion process of the bubble image features and the element grade features, the interaction method between features is optimized, and it is more capable of capturing more key features based on overall collaboration at each time step for the real-time state of the flotation process.

[0054] 2. By calculating the temporal attention weights of the hidden state sequences corresponding to each time step within the time window and performing weighted summation on the corresponding LSTM hidden state sequences according to the temporal attention weights to obtain the context vector within the time window, the key operation time points (i.e., the time step with the largest temporal attention weight in the hidden state sequence obtained by LSTM) can be accurately identified, thereby improving the prediction accuracy of the probability distribution of the reagent types and the dosage values of each reagent in the subsequent process.

[0055] 3. Predict the probability distribution of the reagent types and the dosage values of each reagent based on the same context vector within the time window, which solves the problem of insufficient collaboration caused by the separation of the two types of tasks in the traditional method.

[0056] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent dosing method for lead-zinc ore flotation, characterized in that: include: Step S1, in the flotation process of lead-zinc ore, extracting bubble image features based on bubble images, and collecting grade data of at least two metal elements of the slurry sample in the process in real time through an online analyzer to construct element grade features; Step S2, respectively calculating the query vector, key vector and value vector of the bubble image feature and the element grade feature based on the attention mechanism; Step S3: a first cross-modal feature is calculated through a cross-attention mechanism based on a key vector and a value vector based on an element quality feature and a query vector based on a bubble image feature, and a second cross-modal feature is calculated through a cross-attention mechanism based on a key vector and a value vector based on an element quality feature and a query vector based on an element quality feature, the first cross-modal feature, the second cross-modal feature, the bubble image feature and the element quality feature are fused through a residual connection, and a final multi-modal feature sequence is obtained through layer normalization. Step S4: In any time window, the multimodal feature sequence of each time step is input into the LSTM network to obtain the LSTM hidden state sequence corresponding to each time step; Step S5, calculating the time attention weight of the hidden state sequence corresponding to each time step in the time window, and performing weighted summation on the corresponding LSTM hidden state sequence according to the time attention weight to obtain the context vector in the time window; Step S6: Share the weighted summed context vector as input to the fully connected layers corresponding to the two detection heads to obtain the probability distribution of the predicted drug type and the dosage value of each drug respectively.

2. The intelligent dosing method for lead-zinc ore flotation according to claim 1, characterized in that: In step S1, based on the input 224×224×3 original image, a 64-dimensional image feature is extracted through a pre-trained ResNet-18 feature extraction network.

3. The intelligent dosing method for lead-zinc ore flotation according to claim 2 is characterized in that: In step S2, the query vector, key vector and value vector of the bubble image features and element grade features based on the attention mechanism are calculated through independent linear transformation. The specific calculation process is: in, It is a 64-dimensional image feature; are different weight matrices that can be trained, and ; are respectively the query vector, key vector and value vector of the bubble image feature based on the attention mechanism; The number of dimensions of the corresponding parameter is the corresponding superscript value; in, It is the element grade characteristic of 22 dimensions. are different weight matrices that can be trained, and ; They are respectively the query vector, key vector and value vector of the element quality features based on the attention mechanism.

4. The intelligent dosing method for lead-zinc ore flotation according to claim 3 is characterized in that: The calculation formula of step S3 specifically includes: ； ； in, is the first cross-modal feature, is the second cross-modal feature, and is the temperature coefficient of the hyperparameter, with the superscript is the matrix transpose symbol.

5. The intelligent dosing method for lead-zinc ore flotation according to claim 4, characterized in that: The calculation formula of step S5 specifically includes: ； ； ； in, is the time window length, Corresponding time step for LSTM network The output hidden state sequence; is the trainable weight matrix; is a trainable bias term; is the time step Weights in time windows; for Normalized weights after conversion based on the Softmax function; The context vector within the time window is obtained by weighted summation.

6. The intelligent dosing method for lead-zinc ore flotation according to claim 5, characterized in that: The specific calculation formula of step S6 includes: ； in, is the classification task weight matrix; is the bias term for the classification task; Linear output for classification tasks; is the probability of drug type; is a natural constant, represents standard deviation; ； in, is the regression task weight matrix; is the regression task bias term; The predicted value of drug dosage.

7. The intelligent dosing method for lead-zinc ore flotation according to any one of claims 1 to 6, characterized in that: In the process of training the model for implementing steps S2 to S6, the loss function is composed of the weighted sum of the multi-label binary cross entropy loss of the drug type information and the mean square loss of the drug dosage, and the loss gradient of the model back propagation is obtained by deriving the loss.

8. An intelligent dosing system for lead-zinc ore flotation, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images