An adaptive local dynamic coke quality prediction method

Abstract

The present invention provides an adaptive local dynamic coke quality prediction method, including the steps of: obtaining a complete set of historical coal blends; converting a coal blending plan to be predicted and the historical coal blending plans into a matrix to perform a distance measurement; extracting some historical coal blending plans from the complete set of historical coal blends as coal blending plan neighbors subS based on the distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ; training a predetermined model using the historical coal blending plans in the complete set of historical coal blends or the coal blending plan neighbors subS based on the coal blending plan neighbors subS and preset conditions; and inputting the coal blending plan to be predicted into the target prediction model for prediction. The present invention also provides an adaptive local dynamic coke quality prediction system. The present invention is particularly suitable for coal blend prediction and scientific coal blending in coke companies that produce a large number of different coal types, and can achieve substantial cost reductions and efficiency improvements.

Description

[Technical Field]

[0001] This application claims priority to a Chinese patent application filed with the China Patent Office on October 12, 2022, bearing application number 202211247136.3 and entitled "Adaptive local dynamic coke quality prediction method," the entire contents of which are incorporated herein by reference.

[0002] The present invention relates to the field of industrial production coal blending planning, and more particularly to an adaptive local dynamic coke quality prediction method and system. [Background technology]

[0003] As coal blending and coking technology continues to improve, companies' requirements for coke quality and the stability of coke quality are also increasing. In addition, the shortage of high-quality coking coal and fluctuations in the domestic and international coal and coke markets pose certain challenges to coking companies' coal raw material procurement, which further affects the control of final coke quality. Therefore, accurately predicting coke quality based on coal blending planning has become a major problem that coking companies must urgently solve.

[0004] Although there are many conventional coke quality prediction models, most of them cannot adapt to the work concepts of coal blending operators, and dynamic adjustment is even more difficult. Therefore, when there is a lack of data, the model's predictive performance is relatively low and cannot capture more data information, resulting in significant discrepancies with expectations. However, when there is sufficient data, because coal blending and coking are closely related to the production environment, raw coal origin, inventory structure, etc., large amounts of data will introduce noise into the model, leading to large deviations in the model results and further affecting the final coke quality prediction.

[0005] Therefore, for coke enterprises, a stable and low-error coke quality prediction model can, on the one hand, avoid fluctuations in coke quality caused by changes in coal raw materials, and, on the other hand, achieve effective cost reduction under the premise that coke quality is controlled. Summary of the Invention [Problem to be solved by the invention]

[0006] The present invention aims to provide an adaptive local dynamic coke quality prediction method and system with improved accuracy. [Means for solving the problem]

[0007] In order to solve the above technical problems, the present invention provides an adaptive local dynamic coke quality prediction method, obtaining a complete set of historical coal blends including a plurality of historical coal blend plans; a step of converting the coal blending plan to be predicted and the historical coal blending plans into a matrix, performing distance measurement, and calculating a distance d between the coal blending plan to be predicted and each historical coal blending plan; extracting some historical coal blending plans from the historical coal blending complete set as coal blending plan neighbors subS based on a distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ; Based on the coal blending plan neighborhood subS and a preset condition, training a predetermined model using the historical coal blend complete set or historical coal blending plans in the coal blending plan neighborhood subS to obtain a target prediction model; and inputting the coal blending plan to be predicted into a target prediction model to predict and obtain the local dynamic coke quality.

[0008] Preferably, the step of training a predetermined model using a historical coal blend complete set or a historical coal blend plan in the coal blend plan neighborhood subS based on the coal blend plan neighborhood subS and the preset condition to obtain a target prediction model specifically includes: A step of determining whether the coal blending plan neighborhood subS satisfies a preset condition; If the coal blending plan neighborhood subS satisfies a predetermined condition, training a first predetermined model based on the historical coal blending plan in the coal blending plan neighborhood subS to obtain a target prediction model; If the coal blending plan neighborhood subS does not satisfy the preset condition, training a second predetermined model based on the historical coal blending plans in the historical coal blend complete set to obtain a target prediction model.

[0009] Preferably, the step of determining whether the coal blending plan neighborhood subS satisfies a predetermined condition specifically includes: If the coal blending plan neighborhood subS is not empty, determining that a preset condition is satisfied; and a step of determining that the predetermined condition is not satisfied if the coal blending plan neighborhood subS is empty.

[0010] Preferably, the first predetermined model is a general linear model; The second predetermined model is a regression tree model.

[0011] Preferably, the step of extracting some historical coal blending plans from the historical coal blending complete set as coal blending plan neighbors subS based on the distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ specifically includes the following steps: a step of determining whether a distance d between the coal blending plan to be predicted and each historical coal blending plan is equal to or less than a preset threshold δ; and adding the corresponding historical coal blending plan to the coal blending plan neighborhood subS when the distance d between the coal blending plan to be predicted and each historical coal blending plan is equal to or less than a preset threshold δ.

[0012] Preferably, the threshold value δ is a minimum coal blending number threshold or a shortest distance threshold.

[0013] Preferably, the step of training a predetermined model using a historical coal blend complete set or a historical coal blend plan in the coal blend plan neighborhood subS to obtain a target prediction model specifically includes: The method includes a step of training a predetermined model using the coal type ratio and coal quality index of the historical coal blending plan as input parameters and the coke quality of the historical coal blending plan as an output parameter to obtain a target prediction model.

[0014] Preferably, both the historical coal blending plan and the coal blending plan to be predicted include a coal type ratio and a coal quality index, The matrix is ​​as follows: JPEG2025533265000002.jpg2074 In the formula, JPEG2025533265000003.jpg1111 is the coal type ratio of each single coal type in the coal blending plan, JPEG2025533265000004.jpg1634 is the coal quality index of each single coal type in the coal blending plan, m is the number of single coal types in the coal blending plan, and n is the number of coal quality indexes of each single coal type in the coal blending plan, Next, calculate the distance d between the coal blending plan to be predicted and each historical coal blending plan, The formula for measuring the distance between the coal blending plan to be predicted and the historical coal blending plan is as follows: JPEG2025533265000005.jpg1044 In the formula, Q and Q' are the transformed matrices of the coal blending plan to be predicted and the historical coal blending plan, respectively, and f dist is the distance measurement method and d is the distance.

[0015] Preferably, the distance measurement method is Manhattan distance method, Euclidean distance method, Chebyshev distance method, Minkowski distance method, cosine angle distance method or Mahalanobis distance method.

[0016] The present invention further provides an adaptive local dynamic coke quality prediction system for realizing the adaptive local dynamic coke quality prediction method, an acquisition module for acquiring a complete set of historical coal blends, the complete set including a plurality of historical coal blend plans; a conversion module that converts the coal blending plan to be predicted and the historical coal blending plan into matrices; a calculation module that measures the coal blending plan to be predicted and the historical coal blending plans converted into matrices, and calculates a distance d between the coal blending plan to be predicted and each historical coal blending plan; an extraction module that extracts some historical coal blending plans as coal blending plan neighbors subS from the historical coal blending complete set based on the distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ; a training module for training a predetermined model using a historical coal blend complete set or a historical coal blend plan in the coal blend plan neighborhood subS based on the coal blend plan neighborhood subS and a preset condition to obtain a target prediction model; and a prediction module for inputting a coal blending plan to be predicted into a target prediction model to predict and obtain local dynamic coke quality. [Effects of the Invention]

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: In the present invention, a coal blending plan is queried and searched based on historical coal blending plans to construct a similar coal blending plan neighborhood, the criterion for the similar coal blending plan neighborhood is based on measured distance, and the prediction model constructed based on the similar coal blending plan neighborhood may be linear or nonlinear, and this prediction method may be called "local prediction". Considering that constructing a coal blending plan neighborhood also requires a certain amount of calculation time, the selection of a prediction model can be selected based on a balance between accuracy and calculation speed. Compared with traditional single-step prediction models, the accuracy of the coke quality calculated through the four steps of "query," "search," "construction of coal blending plan neighborhood," and "local prediction" is improved to a certain extent, and is particularly suitable for coal blending prediction and scientific coal blending in coke companies with wide coal supply routes and a wide variety of coal types used in production, which can achieve substantial cost reductions and efficiency improvements.

[0018] In the present invention, a dataset similar to the queried coal blend list is searched and constructed from the historical dataset using the coal blend list to be queried as a neighborhood of the coal blend plan, and a prediction model is constructed using the neighborhood of the coal blend plan as a training dataset, thereby dynamically modeling each time a query is made, and finally, a coke quality index prediction for the queried coal blend list, including coke CRI, CSR, M10, M40, etc., is obtained. [Brief explanation of the drawings]

[0019] [Figure 1] 1 is a flow chart of the adaptive local dynamic coke quality prediction method of the present invention. [Figure 2] FIG. 1 is a schematic diagram of the predictive performance of the model. DETAILED DESCRIPTION OF THE INVENTION

[0020] In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, the present invention may be embodied in many other forms different from those described herein, and those skilled in the art of coal blending may make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] The terms used in one or more embodiments herein do not limit one or more embodiments herein, but are merely used to describe a particular embodiment. Unless otherwise clearly indicated in the specification, the singular forms "a," "the," and "the" used in one or more embodiments and claims herein are intended to include the plural forms. Furthermore, the term "and / or" used in one or more embodiments herein means to include any and all possible combinations of one or more associated items.

[0022] Herein, in one or more embodiments herein, terms such as first, second, etc. are used to describe various types of information, but such information is not limited to these terms. These terms are merely used to distinguish between information of the same type. For example, a first may be referred to as a second, and similarly, a second may be referred to as a first, without departing from the scope of one or more embodiments herein. Depending on the context, the term "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0023] The present invention will be described in more detail below in conjunction with FIG. The adaptive local dynamic coke quality prediction method includes the following steps. 1),obtain a complete set of historical coal blends, which includes multiple historical coal blend plans. 2) The coal blending plan to be predicted and the historical coal blending plans are converted into a matrix and measured, and the distance d between the coal blending plan to be predicted and each historical coal blending plan is calculated. The historical coal blending plan mainly includes coal type ratios and coal quality indexes, and therefore the coal blending plan matrix Q is expressed by the matrix as follows: JPEG2025533265000006.jpg2057 In the formula, JPEG2025533265000007.jpg1111 is the coal type ratio of each single coal type in the coal blending plan, JPEG2025533265000008.jpg1833 is the coal quality index of each single coal type in the coal blending plan, m is the number of single coal types in the coal blending plan, and n is the number of coal quality indexes of each single coal type in the coal blending plan.

[0024] Then, it is necessary to calculate the distance between the coal blending plan to be predicted and all historical coal blending plans. When two coal blending plans are measured and expressed as distance, they are defined as follows: JPEG2025533265000009.jpg1068 In the formula, Q and Q' are the transformed matrices of the coal blending plan to be predicted and the historical coal blending plan, respectively, and f dist is the distance measurement method and d is the distance.

[0025] Considering that the coal types used in the coal blending plan are variable, i.e., m is not fixed, numerical processing is performed on the coal blending plan to facilitate the calculation of the distance between the two inequality matrices (the coal blending plan to be predicted and the historical coal blending plan). Although the types of coal types used in specific plants vary, considering generality, most coal use scenarios are combined based on the China National Standard for Coal Classification (GB / T 5751-2009) to classify the coal types in each coal blending plan into eight coal types, including gas coal, fatty coal, coking coal, lean coal, 1 / 3 coking coal, very little lean coal, gas fat coal, and poor coal. Therefore, the coal type is combined and based on the coal blending ratio and the corresponding coal quality index of a single coal type, Q∈R m*n A∈R 8*nA is the transformed matrix of the coal blending plan, which can be transformed through the following process. Regarding coal blending plan Q, The numbers of coal types belonging to the gas coal type (QM) are {0,…,m1}, The coal types belonging to the fatty coal (FM) type are numbered {m1+1,…,m2}, The number of coal types belonging to the coking coal (JM) type is {m2+1,…,m3}, The numbers of coal types belonging to lean coal (SM) type are {m3+1,...,m4}, The numbers of coal types belonging to the 1 / 3 coking coal (1 / 3JM) type are {m4+1,…,m5}, The coal types belonging to the lean coal (PS) type are numbered {m5+1,...,m6}, The numbers of coal types belonging to the gas-fatty coal (QF) type are {m6+1,...,m7}, The coal types belonging to the poor coal (PM) type are numbered {m7+1,...,m8}.

[0026] Therefore, the result is as follows: JPEG2025533265000010.jpg56147 JPEG2025533265000011.jpg19147 Transformed matrix of coal blending plan to be forecast For JPEG2025533265000012.jpg718, the result is as follows: JPEG2025533265000013.jpg52142 Therefore, the formula (1) becomes as follows: JPEG2025533265000014.jpg12121 Distance measurement method dist can be selected from Table 1. JPEG2025533265000015.jpg91143

[0027] In the formula, X 1kis the data in the Kth row of the transformed matrix of the coal blending plan to be predicted, and X 2k is the data in the Kth row of the transformed matrix of the historical coal blending plan, X1 is the transformed matrix of the coal blending plan to be forecasted, X2 is the transformed matrix of the historical coal blending plan, n indicates the n-dimensional space, and d 12 indicates distance, just like d.

[0028] 3) Based on the distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ, some historical coal blending plans are extracted from the complete historical coal blending set as coal blending plan neighbors subS. The coal blending plan neighborhood range can be limited by a threshold value δ, and the threshold value δ may be the number of coal blending plans. For example, δ historical coal blending plans with the smallest distance d are obtained as the coal blending plan neighborhood subS, and may be a distance threshold (for example, in the case of a method of measuring similarity by distance). For the coal blending plan set S' obtained by the similarity measurement method, the coal blending plan neighborhood of the coal blending plan that is the query target can be obtained by threshold screening. This coal blending plan neighborhood is a subset subS of the historical coal blending plan complete set S, and by combining the distance d, it becomes as follows: JPEG2025533265000016.jpg2576 4), based on the coal blending plan neighborhood subS and the preset conditions, a predetermined model is trained using the historical coal blending complete set or the historical coal blending plan in the coal blending plan neighborhood subS to obtain a target prediction model. Among them, coal type ratio and coal quality index are input parameters, and coke quality is output. The historical coal blending plan includes coal type ratio, coal quality index, and coke quality. The coal blending plan neighborhood subS is used as data input, and the prediction model can be selected from a general linear model, a support vector machine, a regression tree, and a multilayer perceptron. However, in actual production, considering that the number of coal blending plans is limited, the regression tree model and the general linear model are selected for prediction, and the predetermined models include a first predetermined model and a second predetermined model. The model selection can be determined according to Table 2 below. JPEG2025533265000017.jpg20143

[0029] If the coal blending plan neighborhood subS is empty, all the historical coal blending plans in the historical coal blending complete set S in step 1 are jointly used as input.

[0030] Regarding the prediction model (first predetermined model) selected when the coal blending plan neighborhood subS is not empty, the following points are mainly considered. (1) Generally, the amount of data on the coal blending plan neighborhood subS is relatively small, and if a complex model is used, overfitting is likely to occur. (2) Considering the calculation speed, it is necessary to construct a coal blending planning neighborhood before prediction, and if the coal blending planning neighborhood subS is not empty, it is necessary to train the model each time, which will result in a certain amount of calculation consumption. Therefore, selecting the general linear model can improve the calculation speed. Regarding the prediction model (second predetermined model) selected when the coal blending plan neighborhood subS is empty, the following points are mainly considered. (1) The amount of data in the complete historical coal blend set S is sufficient to build a model with a relatively high degree of complexity. (2) The model trained on the complete set of historical coal blends S is reusable, and there is no need to build the model every time a query is made. 5), the coal blending plan to be predicted is input into the target prediction model and predicted, and the coke quality of the coal blending plan to be predicted is obtained.

[0031] The present invention further provides an adaptive local dynamic coke quality prediction system for realizing the adaptive local dynamic coke quality prediction method, an acquisition module for acquiring a complete set of historical coal blends, the complete set including a plurality of historical coal blend plans; a conversion module that converts the coal blending plan to be predicted and the historical coal blending plan into matrices; a calculation module that measures the coal blending plan to be predicted and the historical coal blending plans converted into matrices, and calculates a distance d between the coal blending plan to be predicted and each historical coal blending plan; an extraction module that extracts some historical coal blending plans as coal blending plan neighbors subS from the historical coal blending complete set based on the distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ; a training module for training a predetermined model using a historical coal blend complete set or a historical coal blend plan in the coal blend plan neighborhood subS based on the coal blend plan neighborhood subS and a preset condition to obtain a target prediction model; and a prediction module for inputting a coal blending plan to be predicted into a target prediction model to predict and obtain local dynamic coke quality.

[0032] In order to better illustrate the technical effects of the present invention, the present invention provides the following specific example to illustrate the above technical flow. Example 1 Step 1: Build a complete set of historical coal blending plans. A total of 212 coal blend plans from the past year are selected from the historical coal blend plans, and the first 200 are selected as historical coal blend plans for constructing a complete set of historical coal blends. The remaining 12 historical coal blend plans are used as the test data set. Each historical coal blend plan has a total of 13 coal quality indicators (including three proximate analysis indicators, two caking indicators, and eight coal lithology indicators) for each single coal type, which are converted as follows for each historical coal blend plan: For each historical coal blending plan, the numbers of coal types belonging to the gas coal type (QM) are {0,…,m1}, the numbers of coal types belonging to the fatty coal (FM) type are {m1+1,…,m2}, the numbers of coal types belonging to the jumbo coal (JM) type are {m2+1,…,m3}, the numbers of coal types belonging to the lean coal (SM) type are {m3+1,…,m4}, and the numbers of coal types belonging to the 1 / 3 jumbo coal (1 / 3JM) type are {m4+1,…,m5}; The numbers of coal types belonging to the lean coal (PS) type are {m5+1,...,m6}, the numbers of coal types belonging to the gas-fatty coal (QF) type are {m6+1,...,m7}, and the numbers of coal types belonging to the lean coal (PM) type are {m7+1,...,m8}, where m1, m2, m3, m4, m5, m6, m7, and m8 all belong to integer fields, and can be as follows according to the type to which a single coal type belongs and the proportion of a single coal type in the coal blending plan: JPEG2025533265000018.jpg63138 Each coal blending plan can be expressed as a matrix of size 8*13. That is, the complete set of coal blending plans has a total of 200 data, each of which is an 8*13 matrix, and there are 12 test data sets, each of which is an 8*13 matrix.

[0033] Step 2, construct a forecasting model based on the complete set of historical coal blends. For the historical coal blend complete set, its specification is 200*8*13, and the historical coal blend complete set is processed, that is, the column-wise sum is performed for each data, so that the specification of the historical coal blend complete set after conversion is 200*1*13, and the conversion of each data is as follows: JPEG2025533265000019.jpg2390 Among them, JPEG2025533265000020.jpg1155. Based on the transformed complete set of historical coal blends, a prediction model is constructed using a regression tree algorithm. Because it is a tree model, in the regression problem, each leaf outputs a predicted value, and the predicted value is generally the average value of the outputs of the elements in the training set included in that leaf. JPEG2025533265000021.jpg1271 For historical coal blending plans, the label y i are the CSR, CRI, M10, M40, ash content, and sulfur content of the coke, and in this embodiment, the CSR of the coke is selected as the label. Therefore, by combining the above data transformations and setting the complete set of historical coal blends after transformation, predictive modeling is performed as follows: JPEG2025533265000022.jpg1178 x i is a 13-dimensional vector, JPEG2025533265000023.jpg1139. The regression problem is to find a function that minimizes the mean squared error (mse) by fitting the elements in a dataset D. The goal is to build JPEG2025533265000024.jpg711. JPEG2025533265000025.jpg1882 Assuming that the constructed regression tree has M leaves, we divide the input space x into M units R1, R2, …, R M This means that there are at most M distinct predictors in the regression tree, and the formula for minimizing the MSE is: JPEG2025533265000026.jpg2399 c m denotes the predicted value of the mth leaf.

[0034] To minimize the MSE of the entire CART, we simply minimize the MSE of each leaf. To minimize the MSE of a single leaf, we simply set the predicted value to the average of the training set elements contained in the leaf, i.e., JPEG2025533265000027.jpg1383 Therefore, at each split, when selecting the splitting variable and splitting point (i.e., selecting the feature and the split that divides the feature space into two), the MSE on the training set of the tree model is minimized, i.e., the sum of the MSE of each leaf is minimized.

[0035] Therefore, at each split, the splitting variable and splitting point are selected (i.e., the feature and the split that divides the feature space into two) to minimize the MSE on the training set of the tree model, i.e., to minimize the sum of the MSEs of each leaf.

[0036] Here, a heuristic method is used to traverse all splitting variables and splitting points, and the case with the smallest sum of MSE of the leaf node is selected as the splitting point. If the jth coal quality index and its value s are selected as the splitting variable and splitting point, the splitting variable and splitting point will divide the input space of the parent node into two. JPEG2025533265000028.jpg22117 The formula for selecting the split variable j and split point s for the regression tree is as follows: JPEG2025533265000029.jpg11147 By adopting the traverse method, the division variable j and the division point s can be found. First, the coal quality index j is fixed, and then the optimal division s for that coal quality index is selected. By performing this for each coal quality index, the optimal division corresponding to the 13 coal quality indexes can be obtained. By selecting the minimum value from these 13 values, the global optimal value (j, s) can be obtained.

[0037] In formula (5), the first term The c1 value obtained from JPEG2025533265000030.jpg957 is calculated according to formula (3): JPEG2025533265000031.jpg755, and similarly, in the second section, JPEG2025533265000032.jpg865, with the following corresponding output values: JPEG2025533265000033.jpg13137 We continue to split the two obtained subregions until the condition is met, where we set the tree depth to 10. Finally, we divide the input space into M regions R1, R2, ..., R M Here, the optimal value for M is obtained by hyperparameter search, and finally, a regression tree prediction model is generated. JPEG2025533265000034.jpg1189 In the formula, M is the spatial domain, m is a positive integer greater than 1, and c m represents the predicted value of the mth leaf, and R m represents the mth input spatial region, x represents the input variables, and I is the identity matrix.

[0038] Step 3: Build a query coal blending plan neighborhood For each test coal blending plan that is a prediction target in the 12 test data sets and that has been converted using the conversion method described in step 1, a column-wise sum is performed as follows. JPEG2025533265000035.jpg2397 Among them, JPEG2025533265000036.jpg1068. The Euclidean distance is used to measure the similarity between two coal blend plans, and is calculated as follows: JPEG2025533265000037.jpg16147 For each test coal blending plan, calculate the distance between the test coal blending plan and each data in the complete data set, set a threshold δ, and i,j is less than or equal to the threshold δ, it is an element in the coal blending plan neighborhood of the coal blending plan to be predicted, and if all distances in the complete set data are greater than the threshold δ, the coal blending plan neighborhood of the test coal blending plan is empty.

[0039] Step 4: Build a predictive model based on the query coal blending plan neighborhood Considering that a coal blending plan neighborhood needs to be constructed each time a coal blending plan is queried, and the coal blending plan neighborhood is not necessarily the same each time, this process takes a long time to calculate, and the amount of data in the obtained coal blending plan neighborhood is generally less than the complete set of historical coal blends, if the coal blending plan neighborhood is not empty, a linear model is used as the prediction model, as follows: JPEG2025533265000038.jpg11137 JPEG2025533265000039.jpg821, JPEG2025533265000040.jpg1043, JPEG2025533265000041.jpg89 is a neighborhood. The solution of the general linear model is found using the least squares method, and α0 ~ α 13 are model parameters.

[0040] Step 5, Local Dynamic Prediction For the coal blending plan neighborhood of the query coal blending plan, if the coal blending plan neighborhood is not empty, the model obtained in step 3 is selected as the prediction model, and if the coal blending plan neighborhood is empty, the model obtained in step 2 is selected as the prediction model. Of the 12 query coal blending plans used for testing in this example, the coal blending plan neighborhood of 10 query coal blending plans is not empty, and the coal blending plan neighborhood of the other two query coal blending plans is empty.

[0041] Step 6: Evaluate the results In this example, the model obtained in step 2 is selected as the reference model, and adaptive local dynamic prediction is not added, i.e., steps 3 and 4 are not performed, and coke CSR is used as the coke quality prediction target. The comparison results achieved are as shown in the following table. JPEG2025533265000042.jpg20143 As can be seen from the results, the error of the adaptive local dynamic prediction model of the present invention is only 2.23, which is better than the 2.84 of the reference model, and therefore the following conclusions can be drawn: (1) The adaptive local dynamic prediction model performs a single search of the coal blending neighborhood for each input test data in most cases (i.e., when the coal blending neighborhood is not empty), which can effectively remove noise data and improve data utilization efficiency. (2) The adaptive local dynamic prediction model remodels each time the coal blending planning neighborhood is not empty, allowing the model to capture the information contained in the data more flexibly and reduce the model error.

[0042] In some embodiments provided by the present invention, the disclosed devices and methods may be realized in other forms. For example, the device embodiments described above are merely illustrative, and the division of the modules, components, or units is merely a division of logic functions. In actual implementation, other division forms may exist, for example, multiple units, components, or assemblies may be combined or integrated with other devices, or some features may be omitted or not implemented.

[0043] The units may or may not be physically separated, and the components represented as units may be one physical unit or multiple physical units, i.e., they may be located in one place or distributed across multiple different places, and some or all of the units may be selected according to actual needs to achieve the objective of the solution of this embodiment.

[0044] Furthermore, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The integrated units may be realized in the form of hardware or in the form of software functional units.

[0045] In particular, based on the disclosed embodiments of the present invention, the processes described above with reference to the flowcharts may be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product, comprising a computer program stored on a computer-readable medium, the computer program including program code for executing the methods illustrated in the flowcharts. In such embodiments, the computer program may be downloaded and installed from a network via a communication interface and / or installed from a removable medium. When executed by a central processing unit (CPU), the computer program performs the functions defined in the methods of the present invention. Here, the computer-readable medium of the present invention may be a computer-readable signal medium, a computer-readable storage medium, or any combination of the above. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof.

[0046] The flowcharts and block diagrams in the figures illustrate possible system architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram represents a module, program segment, or portion of code, which includes one or more executable instructions for implementing a specified logical function. Also, in some alternative implementations, the functions depicted in the blocks may occur in a different order than depicted in the figures. For example, two blocks shown in succession may actually be executed in parallel or in the reverse order, as determined by the functionality involved. Also, each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware system or a combination of dedicated hardware and computer instructions to perform the specified functions or operations.

[0047] The above are only specific embodiments of the present invention, and the scope of protection of the present invention is not limited thereto, and any changes or substitutions within the technical scope disclosed in the present invention should fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined based on the scope of protection of the claims.

Claims

1. 1. A method for predicting adaptive local dynamic coke quality, comprising: obtaining a complete set of historical coal blends including a plurality of historical coal blend plans; a step of converting the coal blending plan to be predicted and the historical coal blending plans into a matrix, performing distance measurement, and calculating a distance d between the coal blending plan to be predicted and each historical coal blending plan; extracting some historical coal blending plans from the historical coal blending complete set as coal blending plan neighborhood subS based on a distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ; According to the coal blending plan neighborhood subS and the preset conditions, training a predetermined model using the historical coal blending complete set or the historical coal blending plans in the coal blending plan neighborhood subS to obtain a target prediction model; inputting a coal blending plan to be predicted into a target prediction model to predict and obtain local dynamic coke quality; 1. A method for predicting adaptive local dynamic coke quality, comprising:

2. The step of training a predetermined model using the historical coal blending complete set or the historical coal blending plans in the coal blending plan neighborhood subS according to the coal blending plan neighborhood subS and the preset conditions to obtain a target prediction model is specifically as follows: A step of determining whether the coal blending plan neighborhood subS satisfies a preset condition; If the coal blending plan neighborhood subS satisfies a predetermined condition, training a first predetermined model based on the historical coal blending plan in the coal blending plan neighborhood subS to obtain a target prediction model; If the coal blending plan neighborhood subS does not satisfy the preset condition, training a second predetermined model based on the historical coal blending plan in the historical coal blend complete set to obtain a target prediction model; 2. The method of claim 1, further comprising:

3. Specifically, the step of determining whether the coal blending plan neighborhood subS satisfies the predetermined condition includes the following steps: If the coal blending plan neighborhood subS is not empty, determining that a preset condition is satisfied; If the coal blending plan neighborhood subS is empty, determining that the predetermined condition is not satisfied; 3. The adaptive local dynamic coke quality prediction method of claim 2, comprising:

4. the first predetermined model is a general linear model; the second predetermined model is a regression tree model; The adaptive local dynamic coke quality prediction method of claim 3.

5. Specifically, the step of extracting some historical coal blending plans as coal blending plan neighborhood subS from the historical coal blending complete set based on the distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ includes the following steps: a step of determining whether a distance d between the coal blending plan to be predicted and each historical coal blending plan is equal to or less than a predetermined threshold δ; When a distance d between the coal blending plan to be predicted and each historical coal blending plan is equal to or less than a predetermined threshold δ, the corresponding historical coal blending plan is added to the coal blending plan neighborhood subS; 2. The method of claim 1, further comprising:

6. The threshold δ is a minimum coal blending number threshold or a shortest distance threshold; The adaptive local dynamic coke quality prediction method of claim 5.

7. The step of training a predetermined model using a historical coal blending complete set or a historical coal blending plan in the coal blending plan neighborhood subS to obtain a target prediction model is specifically: training a predetermined model using the coal type ratio and coal quality index of the historical coal blending plan as input parameters and the coke quality of the historical coal blending plan as an output parameter to obtain a target prediction model; The adaptive local dynamic coke quality prediction method of claim 1.

8. Both the historical coal blending plan and the coal blending plan to be predicted include a coal type ratio and a coal quality index, The matrix is ​​as follows: In the formula, is the coal type ratio of each single coal type in the coal blending plan, is the coal quality index of each single coal type in the coal blending plan, m is the number of single coal types in the coal blending plan, n is the number of coal quality indexes of each single coal type in the coal blending plan, Next, calculate the distance d between the coal blending plan to be predicted and each historical coal blending plan, The formula for measuring the distance between the coal blending plan to be predicted and the historical coal blending plan is as follows: In the formula, Q and Q' are the transformed matrices of the coal blending plan to be predicted and the historical coal blending plan, respectively, and f ｄｉｓｔ is the distance measurement method and d is the distance, The adaptive local dynamic coke quality prediction method of claim 1.

9. The distance measurement method is Manhattan distance method, Euclidean distance method, Chebyshev distance method, Minkowski distance method, cosine angle distance method, or Mahalanobis distance method. The adaptive local dynamic coke quality prediction method of claim 8.

10. An adaptive local dynamic coke quality prediction system for implementing the adaptive local dynamic coke quality prediction method according to any one of claims 1 to 9, comprising: an acquisition module for acquiring a complete set of historical coal blends, the complete set including a plurality of historical coal blend plans; a conversion module that converts the coal blending plan to be predicted and the historical coal blending plan into matrices; a calculation module that measures the coal blending plan to be predicted and the historical coal blending plans converted into matrices, and calculates a distance d between the coal blending plan to be predicted and each historical coal blending plan; an extraction module that extracts some historical coal blending plans as coal blending plan neighborhood subS from the historical coal blending complete set based on the distance d between the coal blending plan to be predicted and each historical coal blending plan and a preset threshold δ; a training module for training a predetermined model using a historical coal blend complete set or a historical coal blend plan in the coal blend plan neighborhood subS based on the coal blend plan neighborhood subS and a preset condition to obtain a target prediction model; a prediction module for inputting a coal blending plan to be predicted into a target prediction model to predict and obtain local dynamic coke quality; An adaptive local dynamic coke quality prediction system comprising:

Images