Clean coal yield prediction learning tree method
By proposing a learning tree method in the prediction of refined coal yield, training a multi-level cluster and combining a fully connected neural network for prediction, the existing methods are solved in terms of causal characteristics and explanatory characteristics, and efficient prediction of complex coal quality characteristics is achieved.
Patent Information
- Application Number
- CN202510070489.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-16
- Publication Date
- 2025-06-03
AI Technical Summary
The existing method for prediction of refined coal yields is difficult to comprehensively extract causal characteristics, timing modeling requires strict alignment of data time, and the prediction results lack interpretation ability, making it difficult to meet the needs of different washing and selection processes.
A learning tree method for the yield prediction of fine coal is proposed. By obtaining historical washing sample data, a data set is made, and the coal mass cluster, moisture cluster, ash cluster, ash cluster, inlet washing volume cluster, and washing working parameter cluster are trained in turn. Finally, the output results of these clusters are used as the input of the yield predictor of fine coal, for training and real-time prediction.
This method can effectively extract interactive information of complex characteristics such as coal quality, moisture, ash, and wash-in, improve the adaptability and prediction accuracy of complex and diverse coal quality characteristics, and provides a high-mobility precision coal yield prediction solution.
Smart Images
Figure CN120086738A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of clean coal yield prediction, in particular to a learning tree method for predicting clean coal yield. Background Art
[0002] Clean coal washing is a key link in coal mine production. Its core goal is to maximize the clean coal yield through an efficient separation process, while reducing the ash content and impurity content of raw coal. In recent years, with the development of data mining and machine learning technologies, data-driven washing result prediction models have gradually become a research hotspot.
[0003] Existing clean coal prediction methods mainly use time series algorithms to predict the yield and ash content, and are modeled based on long short-term memory neural networks (LSTM), dilated causal convolutions, LightGBM, etc. However, these methods are difficult to comprehensively extract causal features, such as raw coal composition, washing volume, and washing conditions. In addition, time series modeling requires strict alignment of the data time generated by each process, and the on-site implementation is difficult. Another part of the methods rely on experts to construct multivariate features and combine them with BP neural networks for prediction. However, both the time series regression model and the BP regression model have the problem that the prediction results lack interpretability and are difficult to meet the requirements of different washing processes. Summary of the Invention
[0004] In view of the above problems, the present invention proposes a learning tree method for predicting clean coal yield.
[0005] An embodiment of the present invention provides a learning tree method for predicting clean coal yield, and the learning tree method for predicting clean coal yield includes:
[0006] Obtain historical washing sample data;
[0007] Use the historical washing sample data to produce a data set;
[0008] Based on the data set, sequentially train a coal quality clusterer, a moisture clusterer, an ash clusterer, a washing volume clusterer, and a washing working parameter clusterer;
[0009] Use the output results of the coal quality clusterer, the moisture clusterer, the ash clusterer, the washing volume clusterer, and the washing working parameter clusterer, combined with the data set as the input of the clean coal yield predictor, and train to obtain a clean coal yield predictor;
[0010] Use the clean coal yield predictor to predict the clean coal yield in real time.
[0011] Optionally, using the historical washing sample data to produce a data set includes:
[0012] Mark multiple target data corresponding to the coal quality characteristics, raw coal moisture, raw coal ash, raw coal washing volume, washing operation parameters, and clean coal yield respectively from the historical washing sample data;
[0013] Use the target data corresponding to the coal quality characteristics, the raw coal moisture, the raw coal ash, the raw coal washing volume, the washing operation parameters, and the clean coal yield respectively as the features of the data set, and use the target data corresponding to the clean coal yield as the label of the data set;
[0014] Among them, the coal quality characteristics include: density characteristics, particle size characteristics, friability;
[0015] The washing operation parameters include: the pressure and flow rate at the combined medium inlet, and the separation density.
[0016] Optionally, based on the data set, train a coal quality clusterer, a moisture clusterer, an ash clusterer, a washing volume clusterer, and a washing operation parameter clusterer in sequence, including:
[0017] Based on the data set, first train a coal quality clusterer;
[0018] Use the output result of the coal quality clusterer, combined with the data set as the input of the moisture clusterer, and train the moisture clusterer;
[0019] Use the output results of the coal quality clusterer and the moisture clusterer, combined with the data set as the input of the ash clusterer, and train the ash clusterer;
[0020] Use the output results of the coal quality clusterer, the moisture clusterer, and the ash clusterer, combined with the data set as the input of the washing volume clusterer, and train the washing volume clusterer;
[0021] Finally, use the output results of the coal quality clusterer, the moisture clusterer, the ash clusterer, and the washing volume clusterer, combined with the data set as the input of the washing operation parameter clusterer, and train the washing operation parameter clusterer.
[0022] Optionally, based on the data set, first train a coal quality clusterer, including:
[0023] Use the comprehensive similarity metric formula to calculate the similarity between samples in the data set;
[0024] Based on the sample similarity calculation results, use the Gaussian mixture model to cluster the density, particle size, and friability characteristics of the data set, and automatically identify the coal quality categories;
[0025] By fitting a multi-Gaussian distribution, capture the non-linear relationship and potential class distribution between features, and improve the recognition accuracy of the coal quality categories.
[0026] Optionally, the comprehensive similarity metric formula is:
[0027]
[0028] In the above formula, S(i, j) represents the similarity between sample i and sample j, S[i, j] ∈ [0, 1], and the closer the value is to 1, the more similar sample i and sample j are; M represents the number of feature dimensions; w m represents the weight of feature m, indicating its contribution to the similarity calculation; x i，m and x j，m respectively represent the values of sample i and sample j on feature m; σ m represents the scale factor of feature m, which is used to normalize the difference of features.
[0029] Optionally, before using the Gaussian mixture model to cluster the density, granularity, and friability features of the dataset, it further includes:
[0030] According to the characteristics of the similarity matrix, initialize the key parameters of the Gaussian mixture model; specifically including:
[0031] Determine the number of clusters K through a predefined value or based on spectral analysis;
[0032] Initialize using the principal component direction of the similarity matrix, extract the first K eigenvectors, and use them as the initial positions of the cluster centers;
[0033] Construct a covariance matrix, which is initialized as a diagonal matrix, and its value is determined by the feature distribution of the samples;
[0034] Set the mixing weights, which are set to a uniform distribution, that is, mixing weight = 1 / K, indicating that the initial weights of each clustering component are equal.
[0035] Optionally, using the output result of the coal quality clusterer, combined with the dataset as the input of the moisture clusterer, train the moisture clusterer, including:
[0036] Using the output result of the coal quality clusterer: each coal quality category, combined with the raw coal moisture in the dataset as the input of the moisture clusterer, use a density-based clustering algorithm to cluster the moisture features, and identify each moisture content category corresponding to each coal quality category.
[0037] Optionally, using the output results of both the coal quality clusterer and the moisture clusterer, combined with the dataset as the input of the ash clusterer, train the ash clusterer, including:
[0038] Using each of the coal quality categories and each of the moisture content categories, combined with the raw coal ash content in the dataset as the input to the ash classifier, hierarchical clustering algorithm is used to cluster the ash characteristics, and each ash content category corresponding to each moisture content is identified.
[0039] Optionally, using the output results of the coal quality classifier, the moisture classifier, and the ash classifier, combined with the dataset as the input to the washing quantity classifier, the washing quantity classifier is trained, including:
[0040] Using each of the coal quality categories, each of the moisture content categories, each of the ash content categories, combined with the raw coal washing quantity in the dataset as the input to the washing quantity classifier, KMeans clustering algorithm is used to cluster the washing quantity characteristics, and each washing quantity category corresponding to each ash content is identified.
[0041] Optionally, using the output results of the coal quality classifier, the moisture classifier, the ash classifier, and the washing quantity classifier, combined with the dataset as the input to the coal preparation working parameter classifier, the coal preparation working parameter classifier is trained, including:
[0042] Using each of the coal quality categories, each of the moisture content categories, each of the ash content categories, each of the washing quantity categories, combined with the coal preparation working parameters in the dataset as the input to the coal preparation working parameter classifier, any clustering algorithm is used to cluster the coal preparation working parameter characteristics, and each coal preparation parameter category corresponding to each washing quantity category is identified.
[0043] Optionally, using the output results of the coal quality classifier, the moisture classifier, the ash classifier, the washing quantity classifier, and the coal preparation working parameter classifier, combined with the dataset as the input to the clean coal yield predictor, the clean coal yield predictor is trained, including:
[0044] Using the fully connected neural network algorithm to perform supervised learning and training on each of the coal quality categories, each of the moisture content categories, each of the ash content categories, each of the washing quantity categories, and each of the coal preparation parameter categories, to obtain the clean coal yield predictor.
[0045] Optionally, using the clean coal yield predictor to predict the clean coal yield in real time, including:
[0046] Obtain real-time data;
[0047] Sequentially load the trained coal quality classifier, moisture classifier, ash classifier, washing quantity classifier, and coal preparation working parameter classifier;
[0048] Identify the real-time data sequentially through the coal quality clustering device, the moisture content clustering device, the ash content clustering device, the washing quantity clustering device, and the washing operation parameter clustering device, and obtain the results corresponding to the real-time data in sequence: each coal quality category, each moisture content category corresponding to each coal quality category, each ash content category corresponding to each moisture content category, each washing quantity category corresponding to each ash content category, and each washing parameter category corresponding to each washing quantity category;
[0049] Load the trained clean coal yield predictor, and input the results corresponding to the real-time data into it to predict the clean coal yield in real time.
[0050] An embodiment of the present invention provides a clean coal yield prediction learning tree system, and the clean coal yield prediction learning tree system includes:
[0051] A data acquisition module, configured to acquire historical washing sample data;
[0052] A production module, configured to produce a data set by using the historical washing sample data;
[0053] A training clustering device module, configured to sequentially train a coal quality clustering device, a moisture content clustering device, an ash content clustering device, a washing quantity clustering device, and a washing operation parameter clustering device based on the data set;
[0054] A training predictor module, configured to use the output results of the coal quality clustering device, the moisture content clustering device, the ash content clustering device, the washing quantity clustering device, and the washing operation parameter clustering device, and combine the data set as the input of the clean coal yield predictor to train the clean coal yield predictor;
[0055] A real-time prediction module, configured to use the clean coal yield predictor to predict the clean coal yield in real time.
[0056] Optionally, the production module is specifically configured to:
[0057] Mark multiple target data corresponding to coal quality characteristics, raw coal moisture, raw coal ash content, raw coal washing quantity, washing operation parameters, and clean coal yield from the historical washing sample data;
[0058] Use the target data corresponding to the coal quality characteristics, the raw coal moisture, the raw coal ash content, the raw coal washing quantity, the washing operation parameters, and the clean coal yield as the features of the data set, and use the target data corresponding to the clean coal yield as the label of the data set;
[0059] Wherein, the coal quality characteristics include: density characteristics, particle size characteristics, and friability;
[0060] The coal preparation working parameters include: the pressure and flow rate at the combined medium inlet, and the separation density.
[0061] Optionally, the training clusterer module includes:
[0062] A coal quality clusterer training sub-module, which is used to first train a coal quality clusterer based on the data set;
[0063] A moisture clusterer training sub-module, which is used to use the output result of the coal quality clusterer and combine the data set as the input of the moisture clusterer to train the moisture clusterer;
[0064] An ash content clusterer training sub-module, which is used to use the output results of the coal quality clusterer and the moisture clusterer and combine the data set as the input of the ash content clusterer to train the ash content clusterer;
[0065] A washed quantity clusterer training sub-module, which is used to use the output results of the coal quality clusterer, the moisture clusterer, and the ash content clusterer and combine the data set as the input of the washed quantity clusterer to train the washed quantity clusterer;
[0066] A coal preparation working parameter clusterer training sub-module, which is used to use the output results of the coal quality clusterer, the moisture clusterer, the ash content clusterer, and the washed quantity clusterer and combine the data set as the input of the coal preparation working parameter clusterer to train the coal preparation working parameter clusterer.
[0067] Optionally, the coal quality clusterer training sub-module is specifically used for:
[0068] Calculating the similarity between samples in the data set using the comprehensive similarity metric formula;
[0069] Based on the sample similarity calculation result, using the Gaussian mixture model to cluster the density, particle size, and friability characteristics of the data set, and automatically identifying the coal quality categories;
[0070] By fitting multiple Gaussian distributions, capturing the non-linear relationship and potential class distribution between features, and improving the recognition accuracy of the coal quality categories.
[0071] Among them, the comprehensive similarity metric formula is:
[0072]
[0073] In the above formula, S(i,j) represents the similarity between sample i and sample j, S[i,j] ∈ [0,1], and the value closer to 1 indicates that sample i and sample j are more similar; M represents the number of feature dimensions; w m represents the weight of feature m, indicating its contribution to the similarity calculation; xi，m and x j，m represent the values of sample i and sample j on feature m respectively; σ m represents the scale factor of feature m, which is used to normalize the difference of features.
[0074] Optionally, the training coal quality clustering sub-module is further configured to:
[0075] According to the characteristics of the similarity matrix, initialize the key parameters of the Gaussian mixture model; specifically including:
[0076] Determine the number of clusters K through predefined values or based on spectral analysis;
[0077] Use the principal component direction of the similarity matrix for initialization, extract the first K eigenvectors as the initial positions of the cluster centers;
[0078] Construct a covariance matrix, which is initialized as a diagonal matrix, and its value is determined by the feature distribution of the samples;
[0079] Set the mixing weights, which are set to a uniform distribution, that is, mixing weight = 1 / K, indicating that the initial weights of each clustering component are equal.
[0080] Optionally, the training moisture clustering sub-module is specifically configured to:
[0081] Using the output result of the coal quality clustering: each coal quality category, combined with the raw coal moisture in the dataset as the input of the moisture clustering, and using a density-based clustering algorithm to cluster the moisture features to identify each moisture content category corresponding to each coal quality category.
[0082] Optionally, the training ash content clustering sub-module is specifically configured to:
[0083] Using each coal quality category, each moisture content category, combined with the raw coal ash content in the dataset as the input of the ash content clustering, and using a hierarchical clustering algorithm to cluster the ash content features to identify each ash content category corresponding to each moisture content.
[0084] Optionally, the training washing quantity clustering sub-module is specifically configured to:
[0085] Using each coal quality category, each moisture content category, each ash content category, combined with the raw coal washing quantity in the dataset as the input of the washing quantity clustering, and using the KMeans clustering algorithm to cluster the washing quantity features to identify each washing quantity category corresponding to each ash content.
[0086] Optionally, the training washing operation parameter clustering sub-module is specifically configured to:
[0087] Using each type of coal quality category, each type of moisture content category, each type of ash content category, and each type of washing quantity category, combined with the washing operation parameters in the dataset as the input of the washing operation parameter clusterer, any clustering algorithm is used to cluster the characteristics of the washing operation parameters, and each type of washing parameter category corresponding to each type of washing quantity category is identified.
[0088] Optionally, the training predictor module is specifically configured to:
[0089] Use the fully connected neural network algorithm to perform supervised learning and training on each type of coal quality category, each type of moisture content category, each type of ash content category, each type of washing quantity category, and each type of washing parameter category to obtain the clean coal yield predictor.
[0090] Optionally, the real-time prediction module is specifically configured to:
[0091] Obtain real-time data;
[0092] Load the trained coal quality clusterer, moisture clusterer, ash clusterer, washing quantity clusterer, and washing operation parameter clusterer in sequence;
[0093] Sequentially identify the real-time data through the coal quality clusterer, the moisture clusterer, the ash clusterer, the washing quantity clusterer, and the washing operation parameter clusterer in sequence, and obtain the results corresponding to the real-time data in sequence: each type of coal quality category, each type of moisture content category corresponding to each type of coal quality category, each type of ash content category corresponding to each type of moisture content category, each type of washing quantity category corresponding to each type of ash content category, and each type of washing parameter category corresponding to each type of washing quantity category;
[0094] Load the trained clean coal yield predictor, and input the results corresponding to the real-time data into it to predict the clean coal yield in real time.
[0095] The clean coal yield prediction learning tree method provided by the present invention includes: first, obtaining historical washing sample data; then using the historical washing sample data to produce a dataset; then, based on this dataset, training a coal quality clusterer, a moisture clusterer, an ash clusterer, a washing quantity clusterer, and a washing operation parameter clusterer in sequence.
[0096] After that, using the output results of the coal quality clusterer, the moisture clusterer, the ash clusterer, the washing quantity clusterer, and the washing operation parameter clusterer, combined with the dataset as the input of the clean coal yield predictor, training to obtain the clean coal yield predictor; finally, using the clean coal yield predictor to predict the clean coal yield in real time.
[0097] In view of the problems existing in the prediction of traditional clean coal yield, the present invention proposes a non-temporal clean coal yield prediction method, which comprehensively considers the entire process from raw coal to be washed, washing conditions to the output of clean coal. Through a multi-level classifier, fine-grained statistical recognition is carried out for each link, and the recognition result is used as the intermediate process of clean coal yield prediction. This method has both high prediction performance and result interpretability, and provides a highly migratory paradigm framework for the prediction of clean coal in different washing processes. The present invention proposes a clean coal yield prediction learning tree method that combines rule matching and machine learning. By combining multi-stage clustering and supervised learning, that is, the way of hierarchical clustering and prediction, it solves the problem of difficult handling of complex feature interactions such as coal quality, moisture, ash content, and washing volume in traditional models, improves the adaptability and prediction accuracy for complex and diverse coal quality characteristics, brings important technical support for coal washing, and has high practicability. BRIEF DESCRIPTION OF THE DRAWINGS
[0098] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered as limiting the present invention. Moreover, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:
[0099] Figure 1 is a flowchart of a clean coal yield prediction learning tree method according to an embodiment of the present invention;
[0100] Figure 2 is a modular schematic diagram of a clean coal yield prediction learning tree system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0101] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, which are only a part of the embodiments of the present invention, rather than all the embodiments, and are not used to limit the present invention.
[0102] Referring to Figure 1 , a flowchart of a clean coal yield prediction learning tree method according to an embodiment of the present invention is shown. The clean coal yield prediction learning tree method includes:
[0103] Step 101: Obtain historical washing sample data.
[0104] For the clean coal yield prediction learning tree method of the present invention, the collection and cleaning of historical data need to be carried out first, that is: obtain historical washing sample data. There are various ways to obtain historical washing sample data. For example: historical washing sample data can be obtained from a database.
[0105] After obtaining the historical washing sample data, preprocessing methods such as deduplication, complementation, and normalization can be used to clean the historical washing sample data to ensure the quality of the obtained data, lay a good foundation for the training of subsequent clusterers and clean coal yield predictors, and further lay a good foundation for the real-time prediction of clean coal yield.
[0106] Step 102: Use the historical washing sample data to produce a data set.
[0107] After obtaining the historical washing sample data, a data set can be produced using the historical washing sample data. This data set serves as the input and label for subsequent clusterers and clean coal yield predictors. A better method for producing a data set using the historical washing sample data includes:
[0108] Mark out the multi-dimensional feature data corresponding to the coal quality characteristics, raw coal moisture, raw coal washing volume, and washing operation parameters from the historical washing sample data, and use them as multiple target data, as well as another target data corresponding to the clean coal yield; all of these are data related to the clean coal yield.
[0109] Use the target data corresponding to the coal quality characteristics, raw coal moisture, raw coal washing volume, washing operation parameters, and clean coal yield as the multi-dimensional features of the data set, and use the target data corresponding to the clean coal yield as the label of the data set to guide subsequent learning and prediction.
[0110] Among them, the coal quality characteristics include: density characteristics, particle size characteristics, friability and other characteristics; the washing operation parameters include: the pressure and flow rate at the combined medium inlet, separation density and other parameters.
[0111] Step 103: Based on the data set, sequentially train a coal quality clusterer, a moisture clusterer, an ash clusterer, a washing volume clusterer, and a washing operation parameter clusterer.
[0112] After the data set is made, a coal quality clusterer, a moisture clusterer, an ash clusterer, a washing volume clusterer, and a washing operation parameter clusterer can be sequentially trained based on the data set. A better method for training each clusterer includes:
[0113] Based on the data set, first train a coal quality clusterer; then use the output result of the coal quality clusterer and combine it with the data set as the input of the moisture clusterer to train the moisture clusterer; then use the output results of the coal quality clusterer and the moisture clusterer, and combine them with the data set as the input of the ash clusterer to train the ash clusterer.
[0114] Subsequently, using the output results of the coal quality clusterer, moisture clusterer, and ash content clusterer, combined with the dataset as the input of the washed quantity clusterer, the washed quantity clusterer is trained; finally, using the output results of the coal quality clusterer, moisture clusterer, ash content clusterer, and washed quantity clusterer, combined with the dataset as the input of the coal preparation working parameter clusterer, the coal preparation working parameter clusterer is trained.
[0115] The present invention creatively proposes using the coal quality clusterer as the first clusterer to be trained, and successively using the output results of the previous clusterers as the input of the subsequent clusterers to train the subsequent clusterers. Specifically:
[0116] The method for first training the coal quality clusterer based on the dataset includes:
[0117] Use the comprehensive similarity metric formula to calculate the similarity between samples in the dataset; based on the sample similarity calculation results, use the Gaussian mixture model to cluster the density, particle size, and friability characteristics of the dataset to automatically identify the coal quality categories. At the same time, by fitting multiple Gaussian distributions, the non-linear relationships and potential class distributions between features can be captured, improving the recognition accuracy of coal quality categories.
[0118] Among them, the comprehensive similarity metric formula is:
[0119]
[0120] In the above formula, S(i, j) represents the similarity between sample i and sample j, S[i, j] ∈ [0, 1], and the closer the value is to 1, the more similar sample i and sample j are; M represents the number of feature dimensions; w m represents the weight of feature m, indicating its contribution to the similarity calculation; x i，m and x j，m respectively represent the values of sample i and sample j on feature m; σ m represents the scale factor of feature m, which is used to normalize the difference of the feature.
[0121] In addition, before using the Gaussian mixture model to cluster the density, particle size, and friability characteristics of the dataset, the Gaussian mixture model also needs to be initialized, which includes:
[0122] According to the characteristics of the similarity matrix, initialize the key parameters of the Gaussian mixture model; specifically including:
[0123] Determine the number of clusters K through predefined values or based on spectral analysis (such as spectral clustering, etc.); use the principal component direction of the similarity matrix for initialization (such as singular value decomposition, etc.), extract the first K eigenvectors as the initial positions of the cluster centers, that is, set the mean vector μ k .
[0124] Reconstruct the covariance matrix, which is initialized as a diagonal matrix, and its values are determined by the characteristic distribution of the samples (e.g., the variance of each feature), that is, set the covariance matrix Σ k Finally, set the mixing weight πk, which is set to a uniform distribution, that is, the mixing weight πk = 1 / K, indicating that the initial weights of each clustering component are equal.
[0125] The training of the coal quality clusterer can be achieved through the above method.
[0126] A better method for training a moisture clusterer by using the output result of the coal quality clusterer and combining the dataset as the input of the moisture clusterer includes:
[0127] Using the output result of the coal quality clusterer: each coal quality category, and combining the raw coal moisture in the dataset as the input of the moisture clusterer, and using the density-based clustering algorithm to cluster the moisture characteristics to identify each moisture content category corresponding to each coal quality category.
[0128] A better method for training an ash content clusterer by using the output results of the coal quality clusterer and the moisture clusterer and combining the dataset as the input of the ash content clusterer includes:
[0129] Using each coal quality category and each moisture content category (that is, each moisture content category corresponding to each coal quality category), and combining the raw coal ash content in the dataset as the input of the ash content clusterer, and using the hierarchical clustering algorithm to cluster the ash content characteristics to identify each ash content category corresponding to each moisture content.
[0130] A better method for training a washing amount clusterer by using the output results of the coal quality clusterer, the moisture clusterer, and the ash content clusterer and combining the dataset as the input of the washing amount clusterer includes:
[0131] Using each coal quality category, each moisture content category, and each ash content category (that is, under the node of each moisture content category corresponding to each coal quality category, each ash content category corresponding to each moisture content category), and combining the raw coal washing amount in the dataset as the input of the washing amount clusterer, and using the KMeans clustering algorithm to cluster the washing amount characteristics to identify each washing amount category corresponding to each ash content.
[0132] A better method for training a coal preparation working parameter clusterer by using the output results of the coal quality clusterer, the moisture clusterer, the ash content clusterer, and the washing amount clusterer and combining the dataset as the input of the coal preparation working parameter clusterer includes:
[0133] For each coal quality category, each moisture content category, each ash content category, and each washing amount category (that is, under each moisture content category corresponding to each coal quality category, under each ash content category corresponding to each moisture content category at a lower level, and under each washing amount category corresponding to each ash content category), combined with the washing operation parameters in the dataset as the input to the washing operation parameter clustering algorithm, any clustering algorithm is used to cluster the characteristics of the washing operation parameters, and each washing parameter category corresponding to each washing amount category is identified.
[0134] By adaptively selecting algorithms suitable for different feature distributions, the above clustering algorithms achieve efficient classification and dynamic optimization of various characteristics such as coal quality, moisture, ash, washing amount, and washing operation parameters, providing multi-dimensional feature input support for subsequent prediction of clean coal yield.
[0135] Step 104: Using the output results of the coal quality clustering algorithm, moisture clustering algorithm, ash clustering algorithm, washing amount clustering algorithm, and washing operation parameter clustering algorithm, combined with the dataset as the input to the clean coal yield predictor, the clean coal yield predictor is trained.
[0136] After training each clustering algorithm, using the output results of each clustering algorithm, combined with the dataset as the input to the clean coal yield predictor, the clean coal yield predictor is trained. A preferred method for training the clean coal yield predictor includes:
[0137] Using the fully connected neural network algorithm for supervised learning and training of each coal quality category, each moisture content category (that is, each moisture content category corresponding to each coal quality category), each ash content category (that is, under each moisture content category corresponding to each coal quality category, each ash content category corresponding to each moisture content category), each washing amount category (that is, under each moisture content category corresponding to each coal quality category, under each ash content category corresponding to each moisture content category at a lower level, and under each washing amount category corresponding to each ash content category), and each washing parameter category (that is, under each moisture content category corresponding to each coal quality category, under each ash content category corresponding to each moisture content category at a lower level, under each washing amount category corresponding to each ash content category at the lowest level, and each washing amount category corresponding to each washing amount category), to obtain the clean coal yield predictor.
[0138] Step 105: Use the clean coal yield predictor to predict the clean coal yield in real time.
[0139] After the clean coal yield predictor is trained, it can be used to predict the clean coal yield in real time. A better method for using the clean coal yield predictor to predict the clean coal yield in real time includes:
[0140] Obtain real-time data; for example: obtain real-time raw coal data in the form of a database or a detection device, etc. Then, load the trained coal quality clustering device, moisture clustering device, ash content clustering device, washing volume clustering device, and washing operation parameter clustering device in sequence.
[0141] Identify the real-time raw coal data through the coal quality clustering device, moisture clustering device, ash content clustering device, washing volume clustering device, and washing operation parameter clustering device in sequence, and obtain the corresponding results of the real-time raw coal data in sequence: each coal quality category, each moisture content category corresponding to each coal quality category, each ash content category corresponding to each moisture content category, each washing volume category corresponding to each ash content category, and each washing parameter category corresponding to each washing volume category.
[0142] Finally, load the trained clean coal yield predictor, and input the corresponding results of the above real-time raw coal data into it, then the clean coal yield can be predicted in real time.
[0143] Through the above method, the clean coal yield can be predicted in real time quickly, effectively, and accurately.
[0144] In the embodiment of the present invention, based on the above clean coal yield prediction learning tree method, a clean coal yield prediction learning tree system is also proposed. This system can execute the clean coal yield prediction learning tree method described in any one of steps 101 to 105 above, and realize fast, effective, and accurate real-time prediction of the clean coal yield. Refer to Figure 2 As shown in the modular schematic diagram of a clean coal yield prediction learning tree system, this clean coal yield prediction learning tree system includes:
[0145] A data acquisition module 210, which is used to acquire historical washing sample data;
[0146] A production module 220, which is used to produce a data set by using the historical washing sample data;
[0147] A clustering device training module 230, which is used to sequentially train a coal quality clustering device, a moisture clustering device, an ash content clustering device, a washing volume clustering device, and a washing operation parameter clustering device based on the data set;
[0148] A predictor training module 240, which is used to use the output results of the coal quality clustering device, the moisture clustering device, the ash content clustering device, the washing volume clustering device, and the washing operation parameter clustering device, and combine the data set as the input of the clean coal yield predictor to train the clean coal yield predictor;
[0149] A real-time prediction module 250, which is used to facilitate the real-time prediction of the clean coal yield by the clean coal yield predictor.
[0150] Optionally, the production module 220 is specifically configured to:
[0151] Mark multiple target data corresponding to the coal quality characteristics, raw coal moisture, raw coal ash content, raw coal washing volume, washing operation parameters, and clean coal yield respectively from the historical washing sample data;
[0152] Use the target data corresponding to the coal quality characteristics, the raw coal moisture, the raw coal ash content, the raw coal washing volume, the washing operation parameters, and the clean coal yield respectively as the features of the data set, and use the target data corresponding to the clean coal yield as the label of the data set;
[0153] Wherein, the coal quality characteristics include: density characteristics, particle size characteristics, friability;
[0154] The washing operation parameters include: the pressure and flow rate at the combined medium inlet, and the separation density.
[0155] Optionally, the training clusterer module 230 includes:
[0156] A coal quality clusterer training sub-module, which is used to first train a coal quality clusterer based on the data set;
[0157] A moisture clusterer training sub-module, which is used to use the output result of the coal quality clusterer and combine the data set as the input of the moisture clusterer to train the moisture clusterer;
[0158] An ash content clusterer training sub-module, which is used to use the output results of both the coal quality clusterer and the moisture clusterer and combine the data set as the input of the ash content clusterer to train the ash content clusterer;
[0159] A washing volume clusterer training sub-module, which is used to use the output results of the coal quality clusterer, the moisture clusterer, and the ash content clusterer and combine the data set as the input of the washing volume clusterer to train the washing volume clusterer;
[0160] A washing operation parameter clusterer training sub-module, which is used to use the output results of the coal quality clusterer, the moisture clusterer, the ash content clusterer, and the washing volume clusterer and combine the data set as the input of the washing operation parameter clusterer to train the washing operation parameter clusterer.
[0161] Optionally, the coal quality clusterer training sub-module is specifically configured to:
[0162] Calculate the similarity between samples in the data set using the comprehensive similarity measurement formula;
[0163] Based on the calculation results of sample similarity, use the Gaussian mixture model to cluster the density, granularity, and friability characteristics of the dataset, and automatically identify the coal quality categories;
[0164] By fitting a multi-Gaussian distribution, capture the non-linear relationship and potential class distribution between features, and improve the recognition accuracy of the coal quality categories.
[0165] Among them, the comprehensive similarity metric formula is:
[0166]
[0167] In the above formula, S(i, j) represents the similarity between sample i and sample j, S[i, j] ∈ [0, 1], and the closer the value is to 1, the more similar sample i and sample j are; M represents the number of feature dimensions; w m represents the weight of feature m, indicating its contribution to the similarity calculation; x i，m and x j，m represent the values of sample i and sample j on feature m respectively; σ m represents the scale factor of feature m, which is used to normalize the difference of features.
[0168] Optionally, the training coal quality clusterer sub-module is also used for:
[0169] According to the characteristics of the similarity matrix, initialize the key parameters of the Gaussian mixture model; specifically including:
[0170] Determine the number of clusters K through predefined values or based on spectral analysis;
[0171] Use the principal component direction of the similarity matrix for initialization, extract the first K eigenvectors as the initial positions of the cluster centers;
[0172] Construct a covariance matrix, which is initialized as a diagonal matrix, and its value is determined by the feature distribution of the samples;
[0173] Set the mixing weights, which are set to a uniform distribution, that is, the mixing weight = 1 / K, indicating that the initial weights of each clustering component are equal.
[0174] Optionally, the training moisture clusterer sub-module is specifically used for:
[0175] Using the output results of the coal quality clusterer: each coal quality category, combined with the raw coal moisture in the dataset as the input of the moisture clusterer, and using a density-based clustering algorithm to cluster the moisture characteristics, and identify each moisture content category corresponding to each coal quality category.
[0176] Optionally, the training ash clusterer sub-module is specifically used for:
[0177] Using each of the coal quality categories and each of the moisture content categories, and combining the raw coal ash in the dataset as the input to the ash classifier, hierarchical clustering algorithm is used to cluster the ash characteristics, and each ash content category corresponding to each moisture content is identified.
[0178] Optionally, the training washery input volume classifier sub-module is specifically configured to:
[0179] Using each of the coal quality categories, each of the moisture content categories, each of the ash content categories, and combining the raw coal washery input volume in the dataset as the input to the washery input volume classifier, KMeans clustering algorithm is used to cluster the washery input volume characteristics, and each washery input volume category corresponding to each ash content is identified.
[0180] Optionally, the training washery working parameter classifier sub-module is specifically configured to:
[0181] Using each of the coal quality categories, each of the moisture content categories, each of the ash content categories, each of the washery input volume categories, and combining the washery working parameters in the dataset as the input to the washery working parameter classifier, any clustering algorithm is used to cluster the washery working parameter characteristics, and each washery parameter category corresponding to each washery input volume category is identified.
[0182] Optionally, the training predictor module 240 is specifically configured to:
[0183] Using the fully connected neural network algorithm to perform supervised learning and training on each of the coal quality categories, each of the moisture content categories, each of the ash content categories, each of the washery input volume categories, and each of the washery parameter categories, and obtaining the clean coal yield predictor.
[0184] Optionally, the real-time prediction module 250 is specifically configured to:
[0185] Obtain real-time data;
[0186] Load the trained coal quality classifier, moisture classifier, ash classifier, washery volume classifier, and washery working parameter classifier in sequence;
[0187] Sequentially identify the real-time data through the coal quality classifier, the moisture classifier, the ash classifier, the washery volume classifier, and the washery working parameter classifier, and sequentially obtain the results corresponding to the real-time data: each coal quality category, each moisture content category corresponding to each coal quality category, each ash content category corresponding to each moisture content category, each washery input volume category corresponding to each ash content category, and each washery parameter category corresponding to each washery input volume category;
[0188] Load the trained clean coal yield predictor, input the results corresponding to the real-time data into it, and predict the clean coal yield in real time.
[0189] In summary, the clean coal yield prediction learning tree method provided by the present invention includes: first, obtaining historical washing sample data; then using the historical washing sample data to produce a data set; and then, based on this data set, training a coal quality clusterer, a moisture clusterer, an ash clusterer, a washing quantity clusterer, and a washing operation parameter clusterer in sequence.
[0190] After that, taking the output results of the coal quality clusterer, the moisture clusterer, the ash clusterer, the washing quantity clusterer, and the washing operation parameter clusterer, and combining the data set as the input of the clean coal yield predictor, training the clean coal yield predictor; finally, using the clean coal yield predictor to predict the clean coal yield in real time.
[0191] In view of the problems existing in the traditional clean coal yield prediction, the present invention proposes a clean coal yield prediction learning tree method that combines rule matching and machine learning. By combining multi-stage clustering and supervised learning, that is, the way of hierarchical clustering and prediction, it solves the problem of difficult handling of complex feature interactions such as coal quality, moisture, ash, and washing quantity in the traditional model, improves the adaptability and prediction accuracy for complex and diverse coal quality characteristics, provides important technical support for coal washing, and has high practicality.
[0192] Although the preferred embodiments of the embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.
[0193] Finally, it should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or terminal device including the said element.
[0194] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit of the present invention and the scope protected by the claims. All of these are within the protection scope of the present invention.
Claims
1. A clean coal yield prediction learning tree method, characterized in that: The clean coal yield prediction learning tree method comprises: Obtain historical washing sample data; Using the historical cleansed sample data to create a data set; Based on the data set, a coal quality clusterer, a moisture clusterer, an ash clusterer, an inflow washing amount clusterer, and a washing working parameter clusterer are trained in sequence; The output results of the coal quality clusterer, the moisture clusterer, the ash clusterer, the washing amount clusterer, and the washing working parameter clusterer are combined with the data set as the input of the clean coal yield predictor to train the clean coal yield predictor; This helps the clean coal yield predictor to predict the clean coal yield in real time.
2. The clean coal yield prediction learning tree method according to claim 1, characterized in that: The data set is produced by using the historical cleansing sample data, including: Marking multiple target data corresponding to coal quality characteristics, raw coal moisture, raw coal ash content, raw coal washing amount, washing working parameters and clean coal yield from the historical washing sample data; The target data corresponding to the coal quality characteristics, the raw coal moisture content, the raw coal ash content, the raw coal washing amount, the washing working parameters and the clean coal yield are used as the features of the data set, and the target data corresponding to the clean coal yield is used as the label of the data set; The coal quality characteristics include: density characteristics, particle size characteristics, and fragility; The washing and sorting working parameters include: the inlet pressure and flow rate, and the sorting density.
3. The clean coal yield prediction learning tree method according to claim 2, characterized in that: Based on the data set, a coal quality clusterer, a moisture clusterer, an ash clusterer, an inflow clusterer, and a washing working parameter clusterer are trained in sequence, including: Based on the data set, a coal quality clusterer is firstly trained; The output result of the coal quality clusterer is combined with the data set as the input of the moisture clusterer to train the moisture clusterer; The output results of the coal quality clusterer and the moisture clusterer are combined with the data set as the input of the ash clusterer to train the ash clusterer; The output results of the coal quality clusterer, the moisture clusterer and the ash clusterer are combined with the data set as the input of the washing amount clusterer to train the washing amount clusterer; Finally, the output results of the coal quality clusterer, the moisture clusterer, the ash clusterer and the washing amount clusterer are combined with the data set as the input of the washing work parameter clusterer to train the washing work parameter clusterer.
4. The clean coal yield prediction learning tree method according to claim 3, characterized in that: Based on the data set, a coal quality clusterer is first trained, including: Using a comprehensive similarity measurement formula to calculate the similarity between samples in the data set; Based on the sample similarity calculation results, the density, particle size and fragility characteristics of the data set are clustered using a Gaussian mixture model to automatically identify the coal quality category; By fitting multiple Gaussian distributions, the nonlinear relationship between features and the potential category distribution are captured, thereby improving the recognition accuracy of the coal quality category.
5. The clean coal yield prediction learning tree method according to claim 4, characterized in that: The comprehensive similarity measurement formula is: In the above formula, S(i, j)] represents the similarity between sample i and sample j, S[i, j]∈[0,1], the closer the value is to 1, the more similar sample i and sample j are; M represents the number of feature dimensions; w m represents the weight of feature m, indicating its contribution to the similarity calculation; x i，m and x j，m Respectively represent the values of sample i and sample j on feature m; σ m Represents the scaling factor of feature m, used to normalize the feature differences.
6. The clean coal yield prediction learning tree method according to claim 4, characterized in that: Prior to clustering the density, granularity, and fragility features of the dataset using a Gaussian mixture model, the following steps are also included: According to the characteristics of the similarity matrix, the key parameters of the Gaussian mixture model are initialized; specifically: Determine the number of clusters K by predefined value or based on spectral analysis; Initialize the principal component direction of the similarity matrix and extract the first K eigenvectors as the initial position of the cluster center; Construct a covariance matrix, which is initialized as a diagonal matrix and the value is determined by the characteristic distribution of the sample; The mixed weight is set to be uniformly distributed, that is, mixed weight = 1 / K, indicating that the initial weights of the cluster components are equal.
7. The clean coal yield prediction learning tree method according to claim 3, characterized in that: The output result of the coal quality clusterer is combined with the data set as the input of the moisture clusterer to train the moisture clusterer, including: The output result of the coal quality clusterer is used as the input of the moisture clusterer: each coal quality category is combined with the raw coal moisture in the data set, and the moisture features are clustered using a density-based clustering algorithm to identify each moisture content category corresponding to each coal quality category.
8. The clean coal yield prediction learning tree method according to claim 7, characterized in that: The output results of the coal quality clusterer and the moisture clusterer are combined with the data set as the input of the ash clusterer to train the ash clusterer, including: Each coal quality category and each moisture content category are combined with the raw coal ash in the data set as the input of the ash clusterer, and the ash features are clustered using a hierarchical clustering algorithm to identify each ash content category corresponding to each moisture content.
9. The clean coal yield prediction learning tree method according to claim 8, characterized in that: The output results of the coal quality clusterer, the moisture clusterer, and the ash clusterer are combined with the data set as the input of the washing amount clusterer, and the washing amount clusterer is trained to obtain the washing amount clusterer, including: Taking each coal quality category, each moisture content category, and each ash content category as the input of the washing amount clusterer, the KMeans clustering algorithm is used to cluster the washing amount features to identify each washing amount category corresponding to each ash content.
10. The clean coal yield prediction learning tree method according to claim 9, characterized in that: The output results of the coal quality clusterer, the moisture clusterer, the ash clusterer, and the washing amount clusterer are combined with the data set as the input of the washing work parameter clusterer, and the washing work parameter clusterer is trained to obtain the washing work parameter clusterer, including: Taking each coal quality category, each moisture content category, each ash content category, and each washing amount category, combined with the washing work parameters in the data set as the input of the washing work parameter clusterer, any clustering algorithm is used to cluster the washing work parameter characteristics to identify each washing parameter category corresponding to each washing amount category.
11. The clean coal yield prediction learning tree method according to claim 10, characterized in that: The output results of the coal quality clusterer, the moisture clusterer, the ash clusterer, the washing amount clusterer, and the washing working parameter clusterer are combined with the data set as the input of the clean coal yield predictor to train the clean coal yield predictor, including: A fully connected neural network algorithm is used to perform supervised learning and training on each coal quality category, each moisture content category, each ash content category, each washing amount category, and each washing parameter category to obtain the clean coal yield predictor.
12. The clean coal yield prediction learning tree method according to claim 10, characterized in that: The clean coal yield predictor is useful for real-time prediction of the clean coal yield, including: Get real-time data; Load the trained coal quality clusterer, moisture clusterer, ash clusterer, washing amount clusterer, and washing working parameter clusterer in sequence; The real-time data is identified in sequence by the coal quality clusterer, the moisture clusterer, the ash clusterer, the washing amount clusterer, and the washing operation parameter clusterer, and the results corresponding to the real-time data are obtained in sequence: each coal quality category, each moisture content category corresponding to each coal quality category, each ash content category corresponding to each moisture content category, each washing amount category corresponding to each ash content category, and each washing parameter category corresponding to each washing amount category; The trained clean coal yield predictor is loaded, and the result corresponding to the real-time data is input into it, and the clean coal yield is obtained by real-time prediction.