Wind power coal dressing intelligent control system

Abstract

The invention discloses an intelligent control system for pneumatic coal dressing, which belongs to the technical field of intelligent coal dressing and comprises a sensing acquisition module, an intelligent coal type identification module, a winnowing strategy optimization module and an intelligent winnowing decision and control module. According to the coal type identification method based on multi-mode embedded driving, intelligent coal type identification is carried out, multi-aspect information of a coal sample is comprehensively extracted, the coal type identification accuracy and stability are improved, the system can more accurately grasp coal quality changes and quickly respond to coal type fluctuation, and therefore more reliable data support is provided for the wind power coal dressing process; a state-driven multi-target self-adaptive wind strategy optimization method is adopted for wind separation strategy optimization, under the condition of state change of coal types, granularity, coal seam thickness and the like, wind separation parameters are dynamically adjusted, indexes such as ash content, yield, energy consumption and gangue mixing rate are collaboratively optimized, and the wind separation coal product quality and the system operation efficiency are effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent coal separation, and particularly relates to a wind power coal separation intelligent control system. BACKGROUND

[0002] The wind power coal separation refers to a physical coal separation method that uses wind power as a separation medium to separate coal and coal gangue in raw coal according to physical differences such as density, particle size and suspension characteristics, so as to obtain clean coal. The wind power coal separation intelligent control system uses artificial intelligence technology, combines multi-modal information generated in the wind separation process, automatically identifies the coal type, and dynamically optimizes the key parameters such as wind speed, air valve and coal supply amount according to the change of the wind separation working condition, so as to improve the accuracy and stability of coal separation, increase the yield of clean coal, and adapt to the demand of complex coal quality and variable working conditions.

[0003] However, in the existing wind power coal separation intelligent control process, the coal type changes frequently but is difficult to be accurately identified in time, and the identification method is single and easy to be disturbed, which leads to the technical problems of lagging adjustment of wind separation parameters and unstable wind separation effect. In addition, the wind separation parameters are set depending on artificial experience, and the system lacks sensitive response ability to the change of coal quality, which leads to the technical problems of unstable coal separation efficiency, inaccurate ash control, high coal gangue mixing rate, and difficulty in meeting the modern coal separation demand of high efficiency and low consumption. SUMMARY

[0004] In view of the above problems, the wind power coal separation intelligent control system is provided to overcome the defects of the prior art. In the existing wind power coal separation intelligent control process, the coal type changes frequently but is difficult to be accurately identified in time, and the identification method is single and easy to be disturbed, which leads to the technical problems of lagging adjustment of wind separation parameters and unstable wind separation effect. The coal type identification method driven by multi-modal embedding is creatively used for intelligent coal type identification, the multi-aspect information of the coal sample is comprehensively extracted, the accuracy and stability of coal type identification are improved, the system can more accurately grasp the change of coal quality and quickly respond to the fluctuation of coal type, so as to provide more reliable data support for the wind power coal separation process. In the existing wind power coal separation intelligent control process, the wind separation parameters are set depending on artificial experience, and the system lacks sensitive response ability to the change of coal quality, which leads to the technical problems of unstable coal separation efficiency, inaccurate ash control, high coal gangue mixing rate, and difficulty in meeting the modern coal separation demand of high efficiency and low consumption. The multi-target adaptive wind strategy optimization method driven by state is creatively used for wind separation strategy optimization. Under the condition of the change of the coal type, particle size and coal seam thickness, the wind separation parameters are dynamically adjusted, the indexes such as ash content, yield, energy consumption and coal gangue mixing rate are cooperatively optimized, and the product quality of the wind power coal separation and the system operation efficiency are effectively improved.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent control system for wind-powered coal separation, comprising: a sensing and acquisition module, an intelligent coal type identification module, an air separation strategy optimization module, and an air separation intelligent decision-making and control module;

[0006] The sensing and collecting module is used for sensing and collecting, and obtains wind selection status information and coal sample collection information through sensing and collecting, and sends the wind selection status information to the wind selection strategy optimization module, and sends the coal sample collection information to the intelligent coal type identification module;

[0007] The intelligent coal type identification module is used for intelligent coal type identification. Based on the coal sample collection information, a multimodal embedding-driven coal type identification method is used to perform intelligent coal type identification, obtain coal type reference information, and send the coal type reference information to the wind selection strategy optimization module;

[0008] The wind separation strategy optimization module is used to optimize the wind separation strategy. Based on the coal type reference information and wind separation state information, a state-driven multi-objective adaptive wind strategy optimization method is used to optimize the wind separation strategy to obtain the optimal wind coal separation strategy, and the optimal wind coal separation strategy is sent to the wind separation intelligent decision and control module;

[0009] The wind separation intelligent decision and control module is used for wind separation intelligent decision and control. Through wind separation intelligent decision and control, decision instructions are implemented, the optimal wind coal separation strategy is achieved, and an intelligent wind separation control report is generated.

[0010] Furthermore, the perception and collection is specifically to collect wind separation parameters and coal seam thickness in real time during the wind coal preparation process to obtain wind separation status information, and at the same time collect RGB images of coal samples through industrial cameras, collect laser particle size measurement data of coal samples through laser particle size analyzers, and collect near-infrared spectrum data of coal samples through infrared sensors to obtain coal sample collection information.

[0011] Furthermore, the intelligent coal type identification is specifically based on the coal sample collection information, and adopts a multimodal embedding driven coal type identification method to perform intelligent coal type identification to obtain coal type reference information, including the following steps: coal sample feature extraction, modal alignment modeling, self-supervised coal type identification and coal type reference information generation;

[0012] The coal sample feature extraction specifically involves extracting coal particle appearance features from the coal sample RGB image by constructing a lightweight convolutional neural network, extracting particle size features from the coal sample laser particle size measurement data by using a statistical feature method, and obtaining spectral features by performing Fourier transform on the coal sample near-infrared spectrum data;

[0013] The modal alignment modeling is used to align and map different modal features into a unified low-dimensional semantic embedding space. Specifically, by designing an image encoder, feature mapping is performed on the coal particle appearance features to obtain coal particle appearance embedding features; by designing a particle size encoder, feature mapping is performed on the particle size features to obtain particle size embedding features; by designing a spectral encoder, feature mapping is performed on the spectral features to obtain spectral embedding features; in the feature mapping process, a modal consistency loss function is introduced to align the semantics between different modalities; finally, the coal particle appearance embedding features, particle size embedding features and spectral embedding features are averaged and fused to generate multimodal embedding features;

[0014] The image encoder is designed, specifically, by setting three convolutional layers in the image encoder, wherein the number of channels of the three convolutional layers is set to 64, 128, and 256, respectively, the convolution kernel size of each convolutional layer is 3×3, the stride is 1, and the ReLU activation function, batch normalization, and random dropout mechanism are all adopted; a global average pooling layer and a fully connected layer are set after the last convolutional layer, and the output dimension of the image encoder is set to 128;

[0015] The particle size encoder is designed by using a multilayer perceptron with two fully connected layers as the particle size encoder, setting 64 nodes in the first fully connected layer and 128 nodes in the second fully connected layer, and the first fully connected layer adopts a ReLU activation function, and the output dimension of the particle size encoder is set to 128;

[0016] The spectral encoder is designed, specifically, by setting three one-dimensional convolutional layers in the spectral encoder, wherein the number of channels of the three one-dimensional convolutional layers is set to 32, 64, and 128 respectively, and the convolution kernel size of each one-dimensional convolutional layer is 5, the stride is 1, and the ReLU activation function is used; a fully connected layer is set after the last one-dimensional convolutional layer, and the output dimension of the spectral encoder is set to 128;

[0017] The self-supervised coal type identification specifically involves performing unsupervised clustering of multimodal embedding features using the KMeans clustering algorithm to generate coal type cluster labels; constructing a Transformer classifier, taking the multimodal embedding features as input and the coal type cluster labels as supervision signals, and introducing a state consistency loss function to perform model training to obtain a coal type classifier; then, using the coal type classifier to identify coal types and obtain coal type categories;

[0018] The coal type reference information is generated by performing the coal sample feature extraction, the modal alignment modeling and the self-supervised coal type identification to obtain the coal type reference information, and the coal type reference information includes coal particle appearance characteristics, particle size characteristics, spectral characteristics and coal type category.

[0019] Furthermore, the wind separation strategy optimization is used to optimize the wind coal separation strategy. Specifically, based on the coal type reference information and wind separation state information, a state-driven multi-objective adaptive wind strategy optimization method is used to optimize the wind separation strategy to obtain the optimal wind coal separation strategy, including the following steps: wind separation state encoding, construction of improved wind separation multi-objective reward, initial design of strategy network, fine-tuning loss training and wind coal separation strategy optimization;

[0020] The wind selection state encoding is specifically based on the coal type reference information and the wind selection state information, by constructing a lightweight state embedding network, constructing the original state vector, obtaining the wind selection state encoding vector, and using the wind selection state encoding vector as the state parameter of the strategy network;

[0021] The improved wind separation multi-objective reward is constructed, specifically, an improved wind separation multi-objective reward function is constructed including an ash control penalty item, a yield reward item, an energy consumption penalty item, and a gangue rate penalty item, and the improved wind separation multi-objective reward function is used as an immediate reward function of the strategy network;

[0022] The strategy network is initially set up by defining action parameters, and using a standard reinforcement learning algorithm to perform strategy network training initialization settings based on the winnowing state encoding vector and the improved winnowing multi-objective reward function, and obtaining the optimal strategy network through strategy network training to perform winnowing parameter control;

[0023] The action parameters specifically include wind speed adjustment value, air valve opening angle adjustment value and coal feeding rate adjustment value;

[0024] The fine-tuning loss training specifically calculates the deviation index in real time during the policy network training process. When the deviation index is greater than the fine-tuning threshold, the policy fine-tuning mechanism is executed. The policy fine-tuning mechanism specifically calculates the policy offset based on the difference between the predicted action and the actual executed action, as well as the deviation between the expected reward and the actual reward, and constructs a fine-tuning policy loss function, updates the policy network parameters, and performs policy network training optimization to obtain the optimized optimal policy network.

[0025] The wind coal separation strategy optimization specifically uses the optimal strategy network to optimize the wind coal separation strategy based on the coal type reference information and wind separation status information to obtain the optimal wind coal separation strategy.

[0026] Furthermore, the intelligent decision-making and control of wind separation is specifically to use the optimal wind coal separation strategy as a control instruction, control the wind coal separation system in real time, and generate an intelligent wind separation control report based on coal type reference information and the optimal wind coal separation strategy.

[0027] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0028] (1) In view of the technical problems in the existing intelligent control process of wind coal preparation, there are frequent changes in coal types but it is difficult to identify them in a timely and accurate manner, and the identification method is single and easily interfered with, resulting in delayed adjustment of wind separation parameters and unstable wind separation effects. This solution creatively adopts a multimodal embedded-driven coal type identification method for intelligent coal type identification, comprehensively extracts various aspects of coal sample information, improves the accuracy and stability of coal type identification, enables the system to more accurately grasp the changes in coal quality, and quickly respond to coal type fluctuations, thereby providing more reliable data support for the wind coal preparation process.

[0029] (2) In view of the technical problems in the existing intelligent control process of wind coal preparation, there is a reliance on manual experience to set wind selection parameters, and a lack of sensitive response capabilities to changes in coal quality, which leads to unstable coal preparation efficiency, inaccurate ash content control, and a high coal gangue mixing rate, making it difficult to meet the high-efficiency and low-consumption modern coal preparation needs. This solution creatively adopts a state-driven multi-objective adaptive wind strategy optimization method to optimize the wind selection strategy. Under the conditions of changes in coal type, particle size, coal seam thickness, etc., the wind selection parameters are dynamically adjusted to coordinately optimize indicators such as ash content, yield, energy consumption, and coal gangue mixing rate, effectively improving the quality of wind coal preparation products and the system operation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A schematic diagram of a module of a wind-powered coal preparation intelligent control system provided by the present invention;

[0031] Figure 2 This is a flow chart of the intelligent coal type identification module;

[0032] Figure 3 Flowchart of the winnowing strategy optimization module.

[0033] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0035] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0036] Example 1, see Figure 1 The present invention provides an intelligent control system for wind coal separation, comprising: a sensing and acquisition module, an intelligent coal type identification module, a wind separation strategy optimization module, and a wind separation intelligent decision-making and control module;

[0037] The sensing and collecting module is used for sensing and collecting, and obtains wind selection status information and coal sample collection information through sensing and collecting, and sends the wind selection status information to the wind selection strategy optimization module, and sends the coal sample collection information to the intelligent coal type identification module;

[0038] The intelligent coal type identification module is used for intelligent coal type identification. Based on the coal sample collection information, a multimodal embedding-driven coal type identification method is used to perform intelligent coal type identification, obtain coal type reference information, and send the coal type reference information to the wind selection strategy optimization module;

[0039] The wind separation strategy optimization module is used to optimize the wind separation strategy. Based on the coal type reference information and wind separation state information, a state-driven multi-objective adaptive wind strategy optimization method is used to optimize the wind separation strategy to obtain the optimal wind coal separation strategy, and the optimal wind coal separation strategy is sent to the wind separation intelligent decision and control module;

[0040] The wind separation intelligent decision and control module is used for wind separation intelligent decision and control. Through wind separation intelligent decision and control, decision instructions are implemented, the optimal wind coal separation strategy is achieved, and an intelligent wind separation control report is generated.

[0041] Example 2, see Figure 1 This embodiment is based on the above embodiment. Specifically, the sensing and acquisition is to collect wind separation parameters and coal seam thickness in real time during the wind coal separation process to obtain wind separation status information. At the same time, an industrial camera is used to collect RGB images of coal samples, a laser particle size analyzer is used to collect laser particle size measurement data of coal samples, and an infrared sensor is used to collect near-infrared spectrum data of coal samples to obtain coal sample collection information.

[0042] The wind separation status information includes wind separation parameters and coal seam thickness;

[0043] The wind selection parameters include wind speed, air valve opening angle and coal feeding rate;

[0044] The coal sample collection information includes coal sample RGB images, coal sample laser particle size measurement data and coal sample near-infrared spectrum data;

[0045] The acquisition frequency of the industrial camera is set to 30 frames per second.

[0046] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The intelligent coal type identification is specifically based on the coal sample collection information. A multimodal embedding-driven coal type identification method is used to perform intelligent coal type identification to obtain coal type reference information. The method includes the following steps: coal sample feature extraction, modal alignment modeling, self-supervised coal type identification, and coal type reference information generation.

[0047] The coal sample feature extraction specifically involves extracting coal particle appearance features from the coal sample RGB image by constructing a lightweight convolutional neural network, extracting particle size features from the coal sample laser particle size measurement data by using a statistical feature method, and obtaining spectral features by performing Fourier transform on the coal sample near-infrared spectrum data;

[0048] Preferably, the lightweight convolutional neural network is specifically a ResNet18 network;

[0049] The modal alignment modeling is used to align and map different modal features into a unified low-dimensional semantic embedding space. Specifically, by designing an image encoder, feature mapping is performed on the coal particle appearance features to obtain coal particle appearance embedding features; by designing a particle size encoder, feature mapping is performed on the particle size features to obtain particle size embedding features; by designing a spectral encoder, feature mapping is performed on the spectral features to obtain spectral embedding features; in the feature mapping process, a modal consistency loss function is introduced to align the semantics between different modalities; finally, the coal particle appearance embedding features, particle size embedding features and spectral embedding features are averaged and fused to generate multimodal embedding features;

[0050] The image encoder is designed, specifically, by setting three convolutional layers in the image encoder, wherein the number of channels of the three convolutional layers is set to 64, 128, and 256, respectively, the convolution kernel size of each convolutional layer is 3×3, the stride is 1, and the ReLU activation function, batch normalization, and random dropout mechanism are all adopted; a global average pooling layer and a fully connected layer are set after the last convolutional layer, and the output dimension of the image encoder is set to 128;

[0051] The particle size encoder is designed by using a multilayer perceptron with two fully connected layers as the particle size encoder, setting 64 nodes in the first fully connected layer and 128 nodes in the second fully connected layer, and the first fully connected layer adopts a ReLU activation function, and the output dimension of the particle size encoder is set to 128;

[0052] The spectral encoder is designed, specifically, by setting three one-dimensional convolutional layers in the spectral encoder, wherein the number of channels of the three one-dimensional convolutional layers is set to 32, 64, and 128 respectively, and the convolution kernel size of each one-dimensional convolutional layer is 5, the stride is 1, and the ReLU activation function is used; a fully connected layer is set after the last one-dimensional convolutional layer, and the output dimension of the spectral encoder is set to 128;

[0053] The calculation formula of the modal consistency loss function is:

[0054] ;

[0055] Where, L mc is the modal consistency loss function, i is the coal sample index, N is the total number of coal samples, p is the first modal index, q is the second modal index, M is the total number of modalities, is the embedded feature of the i-th coal sample in the p-th mode, is the embedded feature of the i-th coal sample in the q-th mode, ||·||2 is the L2 norm symbol;

[0056] The calculation formula for generating multimodal embedding features is:

[0057] ;

[0058] Where Z i is the multimodal embedding feature of the i-th coal sample, is the embedding feature of the i-th coal sample in the first mode, specifically the embedding feature of the coal particle appearance of the i-th coal sample, is the embedding feature of the i-th coal sample in the second mode, specifically the particle size embedding feature of the i-th coal sample, is the embedded feature of the i-th coal sample in the third mode, specifically refers to the spectral embedded feature of the i-th coal sample;

[0059] The self-supervised coal type identification specifically involves performing unsupervised clustering of multimodal embedding features using the KMeans clustering algorithm to generate coal type cluster labels; constructing a Transformer classifier, taking the multimodal embedding features as input and the coal type cluster labels as supervision signals, and introducing a state consistency loss function to perform model training to obtain a coal type classifier; then, using the coal type classifier to identify coal types and obtain coal type categories;

[0060] The calculation formula of the state consistency loss function is:

[0061] ;

[0062] Where, L labelis the state consistency loss function, C(·) is the Transformer classifier, onehot(·) is the one-hot encoding operation, is the coal type cluster label of the i-th coal sample;

[0063] The coal type reference information is generated by performing the coal sample feature extraction, the modal alignment modeling, and the self-supervised coal type identification to obtain the coal type reference information, wherein the coal type reference information includes coal particle appearance features, particle size features, spectral features, and coal type category;

[0064] By performing the above operations, in order to address the technical problems in the existing intelligent control process of wind coal preparation, such as frequent changes in coal types but difficulty in timely and accurate identification, and the single identification method being easily interfered with, resulting in delayed adjustment of wind separation parameters and unstable wind separation effects, this solution creatively adopts a multimodal embedding-driven coal type identification method for intelligent coal type identification, comprehensively extracts multi-faceted information of coal samples, improves the accuracy and stability of coal type identification, enables the system to more accurately grasp changes in coal quality, and quickly respond to fluctuations in coal types, thereby providing more reliable data support for the wind coal preparation process.

[0065] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. The wind separation strategy optimization is used to optimize the wind coal separation strategy. Specifically, based on the coal type reference information and wind separation state information, a state-driven multi-objective adaptive wind strategy optimization method is used to optimize the wind separation strategy to obtain the optimal wind coal separation strategy. The method includes the following steps: wind separation state encoding, constructing an improved wind separation multi-objective reward, initializing the strategy network, fine-tuning the loss training, and optimizing the wind coal separation strategy.

[0066] The wind selection state encoding is specifically based on the coal type reference information and the wind selection state information, by constructing a lightweight state embedding network, constructing the original state vector, obtaining the wind selection state encoding vector, and using the wind selection state encoding vector as the state parameter of the strategy network;

[0067] The calculation formula of the winnowing state coding vector is:

[0068] ;

[0069] Where s t is the winnowing state encoding vector, y t is the one-hot encoding of coal type, d t is the particle size characteristic, l t is the coal seam thickness, v t is the wind speed, is the air valve opening angle, t is the time step index;

[0070] Preferably, the lightweight state embedding network is specifically a lightweight Transformer structure;

[0071] The improved wind separation multi-objective reward is constructed, specifically, an improved wind separation multi-objective reward function is constructed including an ash control penalty item, a yield reward item, an energy consumption penalty item, and a gangue rate penalty item, and the improved wind separation multi-objective reward function is used as an immediate reward function of the strategy network;

[0072] The calculation formula of the improved winnowing multi-objective reward function is:

[0073] ;

[0074] Where r t is an improved winnowing multi-objective reward function, is the ash content control weight, R ash (t) is the ash content control penalty term, is the yield weight, R yield (t) is the productivity bonus item, is the energy consumption weight, R energy (t) is the energy consumption penalty term, is the weight of coal gangue rate, R reject (t) is the coal gangue rate penalty item;

[0075] Preferably, the calculation formula of the ash content control penalty term is:

[0076] ;

[0077] Where a t is the actual ash content of current clean coal, a target It is the standard ash content set by the industry;

[0078] Preferably, the calculation formula of the productivity bonus item is:

[0079] ;

[0080] Where m clean (t) is the clean coal mass per unit time, m total (t) is the total mass of coal fed per unit time;

[0081] Preferably, the calculation formula of the energy consumption penalty term is:

[0082] ;

[0083] Where, is the energy consumption weight factor, v t is the current wind speed;

[0084] Preferably, the calculation formula of the gangue rate penalty term is:

[0085] ;

[0086] Where, is the gangue weight factor, m waste (t) is the mass of gangue mixed into clean coal per unit time;

[0087] The strategy network is initially set up by defining action parameters, and using a standard reinforcement learning algorithm to perform strategy network training initialization settings based on the winnowing state encoding vector and the improved winnowing multi-objective reward function, and obtaining the optimal strategy network through strategy network training to perform winnowing parameter control;

[0088] The action parameters specifically include wind speed adjustment value, air valve opening angle adjustment value and coal feeding rate adjustment value, and the calculation formula is:

[0089] ;

[0090] Where a t is the action parameter, is the wind speed adjustment value, is the air valve opening angle adjustment value, is the coal feed rate adjustment value;

[0091] The fine-tuning loss training specifically calculates the deviation index in real time during the policy network training process. When the deviation index is greater than the fine-tuning threshold, the policy fine-tuning mechanism is executed. The policy fine-tuning mechanism specifically calculates the policy offset based on the difference between the predicted action and the actual executed action, as well as the deviation between the expected reward and the actual reward, and constructs a fine-tuning policy loss function, updates the policy network parameters, and performs policy network training optimization to obtain the optimized optimal policy network.

[0092] The calculation formula of the deviation index is:

[0093] ;

[0094] Where, is the deviation indicator, is the actual execution of the action, is the predicted action, is the advantage estimation action;

[0095] The calculation formula of the fine-tuning threshold is:

[0096] ;

[0097] Where, is the deviation threshold, is the mean of the historical deviation index, k is the tolerance coefficient with a range of [1.5, 3], is the standard deviation of the historical deviation indicator;

[0098] The calculation formula of the fine-tuning strategy loss function is:

[0099] ;

[0100] Where, is the fine-tuning strategy loss function, t max is the maximum time step, is the adjustment coefficient, is the expected reward, r t It is an actual reward;

[0101] The calculation formula for the update strategy network parameters is:

[0102] ;

[0103] Where, are policy network parameters, is the assignment operator, is the policy network learning rate, is the gradient operator with respect to the policy network parameters;

[0104] The wind coal separation strategy optimization is specifically to use the optimal strategy network to optimize the wind coal separation strategy according to the coal type reference information and wind separation state information to obtain the optimal wind coal separation strategy;

[0105] By performing the above operations, in order to address the technical problems in the existing intelligent control process of wind coal preparation, there is a reliance on manual experience to set wind separation parameters, a lack of sensitive response capabilities to changes in coal quality, resulting in unstable coal preparation efficiency, inaccurate ash control, and a high coal-gangue mixing rate, making it difficult to meet the high-efficiency and low-consumption modern coal preparation needs. This solution creatively adopts a state-driven multi-objective adaptive wind strategy optimization method to optimize the wind separation strategy. Under the conditions of changes in coal type, particle size, coal seam thickness, etc., the wind separation parameters are dynamically adjusted to coordinately optimize indicators such as ash content, yield, energy consumption, and coal-gangue mixing rate, effectively improving the quality of wind coal preparation products and system operation efficiency.

[0106] Example 5, see Figure 1 This embodiment is based on the above embodiment. The intelligent decision-making and control of wind separation is specifically to use the optimal wind coal separation strategy as a control instruction, control the wind coal separation system in real time, and generate an intelligent wind separation control report based on the coal type reference information and the optimal wind coal separation strategy.

[0107] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0108] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0109] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. An intelligent control system for wind-powered coal preparation, characterized by: include: Perception and collection module, intelligent coal type identification module, wind selection strategy optimization module and wind selection intelligent decision and control module; The sensing and collecting module is used for sensing and collecting, and obtains wind selection status information and coal sample collection information through sensing and collecting, and sends the wind selection status information to the wind selection strategy optimization module, and sends the coal sample collection information to the intelligent coal type identification module; The intelligent coal type identification module is used for intelligent coal type identification. Based on the coal sample collection information, a multimodal embedding-driven coal type identification method is used to perform intelligent coal type identification, obtain coal type reference information, and send the coal type reference information to the wind selection strategy optimization module; The wind separation strategy optimization module is used to optimize the wind separation strategy. Based on the coal type reference information and wind separation state information, a state-driven multi-objective adaptive wind strategy optimization method is used to optimize the wind separation strategy to obtain the optimal wind coal separation strategy, and the optimal wind coal separation strategy is sent to the wind separation intelligent decision and control module; The wind separation intelligent decision and control module is used for wind separation intelligent decision and control. Through wind separation intelligent decision and control, decision instructions are implemented, the optimal wind coal separation strategy is achieved, and an intelligent wind separation control report is generated.

2. The intelligent control system for wind-powered coal preparation according to claim 1, characterized in that: The intelligent coal type identification method is specifically based on the coal sample collection information and uses a multimodal embedding-driven coal type identification method to perform intelligent coal type identification to obtain coal type reference information, including the following steps: coal sample feature extraction, modal alignment modeling, self-supervised coal type identification and coal type reference information generation; The coal sample feature extraction specifically involves extracting coal particle appearance features from the coal sample RGB image by constructing a lightweight convolutional neural network, extracting particle size features from the coal sample laser particle size measurement data by using a statistical feature method, and obtaining spectral features by performing Fourier transform on the coal sample near-infrared spectrum data; The modal alignment modeling is used to align and map different modal features into a unified low-dimensional semantic embedding space. Specifically, by designing an image encoder, feature mapping is performed on the coal particle appearance features to obtain coal particle appearance embedding features; by designing a particle size encoder, feature mapping is performed on the particle size features to obtain particle size embedding features; by designing a spectral encoder, feature mapping is performed on the spectral features to obtain spectral embedding features; in the feature mapping process, a modal consistency loss function is introduced to align the semantics between different modalities; finally, the coal particle appearance embedding features, particle size embedding features and spectral embedding features are averaged and fused to generate multimodal embedding features; The self-supervised coal type identification specifically involves performing unsupervised clustering of multimodal embedding features using the KMeans clustering algorithm to generate coal type cluster labels; constructing a Transformer classifier, taking the multimodal embedding features as input and the coal type cluster labels as supervision signals, and introducing a state consistency loss function to perform model training to obtain a coal type classifier; then, using the coal type classifier to identify coal types and obtain coal type categories; The coal type reference information is generated by performing the coal sample feature extraction, the modal alignment modeling and the self-supervised coal type identification to obtain the coal type reference information, and the coal type reference information includes coal particle appearance characteristics, particle size characteristics, spectral characteristics and coal type category.

3. The intelligent control system for wind-powered coal preparation according to claim 2, characterized in that: In the modality alignment modeling, the image encoder is designed, specifically, three convolutional layers are set in the image encoder, the number of channels of the three convolutional layers are set to 64, 128 and 256 respectively, the convolution kernel size of each convolutional layer is 3×3, the step size is 1, and the ReLU activation function, batch normalization and random dropout mechanism are all used; a global average pooling layer and a fully connected layer are set after the last convolutional layer, and the output dimension of the image encoder is set to 128; The particle size encoder is designed by using a multilayer perceptron with two fully connected layers as the particle size encoder, setting 64 nodes in the first fully connected layer and 128 nodes in the second fully connected layer, and the first fully connected layer adopts a ReLU activation function, and the output dimension of the particle size encoder is set to 128; The spectral encoder is designed by setting three one-dimensional convolutional layers in the spectral encoder, wherein the number of channels of the three one-dimensional convolutional layers is set to 32, 64 and 128 respectively, the convolution kernel size of each one-dimensional convolutional layer is 5, the step size is 1, and the ReLU activation function is adopted; a fully connected layer is set after the last one-dimensional convolutional layer, and the output dimension of the spectral encoder is set to 128.

4. The intelligent control system for wind-powered coal preparation according to claim 3 is characterized in that: The wind separation strategy optimization is used to optimize the wind coal separation strategy. Specifically, based on the coal type reference information and wind separation state information, a state-driven multi-objective adaptive wind separation strategy optimization method is used to optimize the wind separation strategy to obtain the optimal wind coal separation strategy, including the following steps: wind separation state encoding, construction of improved wind separation multi-objective reward, initial design of strategy network, fine-tuning loss training and wind coal separation strategy optimization; The wind selection state encoding is specifically based on the coal type reference information and the wind selection state information, by constructing a lightweight state embedding network, constructing the original state vector, obtaining the wind selection state encoding vector, and using the wind selection state encoding vector as the state parameter of the strategy network; The improved wind separation multi-objective reward is constructed, specifically, an improved wind separation multi-objective reward function is constructed including an ash control penalty item, a yield reward item, an energy consumption penalty item, and a gangue rate penalty item, and the improved wind separation multi-objective reward function is used as an immediate reward function of the strategy network; The strategy network is initially set up by defining action parameters, and using a standard reinforcement learning algorithm to perform strategy network training initialization settings based on the winnowing state encoding vector and the improved winnowing multi-objective reward function, and obtaining the optimal strategy network through strategy network training to perform winnowing parameter control; The action parameters specifically include wind speed adjustment value, air valve opening angle adjustment value and coal feeding rate adjustment value; The fine-tuning loss training specifically calculates the deviation index in real time during the policy network training process, and executes the policy fine-tuning mechanism when the deviation index is greater than the fine-tuning threshold; The wind coal separation strategy optimization specifically uses the optimal strategy network to optimize the wind coal separation strategy based on the coal type reference information and wind separation status information to obtain the optimal wind coal separation strategy.

5. The intelligent control system for wind-powered coal preparation according to claim 4 is characterized in that: In the fine-tuning loss training, the policy fine-tuning mechanism is specifically to calculate the policy offset based on the difference between the predicted action and the actual executed action, as well as the deviation between the expected reward and the actual reward, and construct a fine-tuning policy loss function, update the policy network parameters, and perform policy network training optimization to obtain the optimized optimal policy network.

6. The intelligent control system for wind-powered coal preparation according to claim 5, characterized in that: The intelligent decision-making and control of wind separation specifically uses the optimal wind coal separation strategy as a control instruction to control the wind coal separation system in real time, and generates an intelligent wind separation control report based on coal type reference information and the optimal wind coal separation strategy.

7. The intelligent control system for wind-powered coal preparation according to claim 6, characterized in that: The sensing and collection is specifically to collect wind separation parameters and coal seam thickness in real time during the wind coal separation process to obtain wind separation status information, and at the same time collect RGB images of coal samples through industrial cameras, collect laser particle size measurement data of coal samples through laser particle size analyzers, and collect near-infrared spectrum data of coal samples through infrared sensors to obtain coal sample collection information.

Images