Teacher data generating method, coke grade prediction model generating method, coke grade prediction method, teacher data generating system, coke grade prediction model generating system, coke grade prediction system, and program

Abstract

To provide a technique for improving accuracy in prediction of a grade value of coke.SOLUTION: The present invention is directed to a method of generating teacher data D2 for use in machine learning of a prediction model 20 for predicting coke grade. The method has steps of acquiring a plurality of actual generation results including compound data relating to a compound ratio of a plurality of kinds of raw coal and grade values of coke actually generated based on the compound ratio together with a generation order, and generating the teacher data D2 based on the plurality of actual generation results. When the teacher data D2 is generated, the compound data is used as input and the grade value is used as output. Further, a first feature amount based on the grade value corresponding to one actual generation result among the plurality of actual generation results generated prior to the actual generation result relating to the compound data used as input and a second feature amount based on the grade values corresponding to at least two respective actual generation results successively generated in order among the plurality of actual generation results generated prior to the actual generation result relating to the compound data used as input are used as input to generate the teacher data D2.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to machine learning techniques for predicting coke grade. [Background technology]

[0002] Coke is obtained by carbonizing a coal blend obtained by blending multiple brands of raw coal. In order to obtain coke that satisfies a desired quality (e.g., physical property values ​​such as coke strength DI), various methods for predicting the quality of coke have been proposed.

[0003] For example, Patent Document 1 refers to learning the correspondence between data on the blending ratio of each of a plurality of types of raw coal and the quality value of coke (coke strength). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2022-41892 A Summary of the Invention [Problem to be solved by the invention]

[0005] The present disclosure provides a technique for improving the accuracy of predicting coke quality values. [Means for solving the problem]

[0006] The teacher data generating method of the present disclosure includes a step of acquiring multiple production results, including blending data regarding the blending ratio of multiple types of raw coal and the quality value of coke actually produced based on the blending ratio, along with the production order, and a step of generating teacher data based on the multiple production results, in which in the step of generating teacher data, the blending data is used as input and the quality value is used as output, and the teacher data is generated by inputting a first feature based on the quality value corresponding to one of the multiple production results generated before the production result related to the blending data used as input, and a second feature based on the quality values ​​corresponding to at least two of the multiple production results that are consecutive in production order and generated before the production result related to the blending data used as input. [Brief description of the drawings]

[0007] [Figure 1] FIG. 2 is a block diagram showing a teacher data generation system, a prediction model generation system, and a coke quality prediction system according to the present embodiment. [Diagram 2] 4 is a flowchart showing processing executed by the teacher data generation system of the present embodiment. [Diagram 3] 4 is a flowchart showing a process executed by the predictive model generation system of the present embodiment. [Figure 4] 3 is a flowchart showing a process executed by the coke quality prediction system of the present embodiment. [Diagram 5] FIG. 13 is a diagram showing an example of teacher data. [Figure 6] FIG. 6 is a diagram showing prediction results and actual values ​​in Example 1 and Comparative Example 1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] FIG. 1 is a block diagram showing a teacher data generation system 1, a prediction model generation system 2, and a coke quality prediction system 3 of this embodiment. As shown in FIG. 1, the coke quality prediction system 3 predicts the quality value of a coke obtained from a coal blend blended at the blending ratio of each raw coal (brand) represented by prediction target blending data D6. The coke quality prediction system 3 uses a prediction model 20 generated in advance by machine learning by the prediction model generation system 2. In this embodiment, the coke strength DI is used as the quality value. The prediction target blending data D6 is data regarding the blending ratio of each of multiple types of raw coal, and in FIG. 1, the blending ratio of each of the multiple types of raw coal is directly shown. The prediction model generation system 2 builds the prediction model 20 by learning it through machine learning using the teacher data D2. The teacher data generation system 1 generates teacher data D2 for machine learning the prediction model 20 based on the actual measurement data D1. The actual measurement data D1 includes the quality value of a coke produced from a certain coal blend and blending data regarding the blending ratio of each of the multiple types of raw coal constituting the coal blend. The values ​​shown in the measured data D1 are based on actual measurements. Each system will be described below.

[0009] Each component of the teacher data generation system 1, the prediction model generation system 2, and the coke quality prediction system 3 is realized in a computer including a CPU, RAM, ROM, non-volatile memory, an input / output interface, etc., by the CPU executing information processing according to a program loaded from the ROM or non-volatile memory to the RAM.

[0010] <Teacher data generation system 1> As shown in FIG. 1, the teacher data generation system 1 includes an actual measurement data acquisition unit 10 and a teacher data generation unit 11.

[0011] The actual measurement data acquisition unit 10 acquires the actual measurement data D1 by input from a user or by reading data from an external storage or an internal storage. The actual measurement data D1 includes a plurality of production results and a production order. The production results may be in any form as long as the data is data that associates blending data regarding the blending ratio of each of a plurality of raw coals with the quality value of the coke actually obtained by the blending. The blending ratio may be the ratio itself or an amount that indirectly indicates the ratio. In this embodiment, the actual measurement data D1 is based on the operation result of the coke oven, and the blending ratio of each raw coal is associated with the quality value of the coke obtained by the blending. The number of types (brands) of raw coal is L, but L can be changed as appropriate depending on the convenience of implementation as long as it is multiple. In this embodiment, data for L=129 is used, but the number of L is one example. The actual measurement data D1 in the example shown in FIG. 1 includes m production results such that the production order is 1, 2, ..., m. It shows that the blending data has a generation order of "1" and the quality value of the coke generated by the blending data is α1. Here, the diagram shows data such as a number indicating the generation order, but this is for the purpose of explanation, and data indicating the generation order is not essential. If there is no data indicating the generation order, the generation order can be expressed by the order of the data, and in this case, it can be said that the actual measurement data D1 includes the generation order.

[0012] The teacher data generating unit 11 generates teacher data D2 based on the actual measurement data D1 acquired by the actual measurement data acquiring unit 10. As an example, when the actual measurement data D1 is directly used as the teacher data, the teacher data indicates that the blending ratios (L numerical values) of each of the multiple types of raw coal are input to the prediction model 20, and one output value that is the correct answer from the prediction model 20 is the grade value. That is, there are L (129) parameter inputs. If the number of parameter inputs is large, there is a limit to improving the accuracy of the prediction model 20. Therefore, in this embodiment, in order to reduce the number of parameter inputs, the teacher data generating system 1 is configured as follows.

[0013] As shown in FIG. 1, the teacher data generation system 1 further includes a raw coal property data acquisition unit 12, a dimension reduction processing unit 13, a cluster analysis unit 14, and a relationship identification unit 15.

[0014] The raw coal physical property data acquisition unit 12 shown in FIG. 1 acquires raw coal physical property data D3. In the raw coal physical property data D3, multiple (i types) physical property values ​​of each raw coal are associated with multiple types of raw coal (brands). There are various types of physical property values ​​of raw coal, but in this embodiment, the following 17 types of physical property values ​​are used for i=17 shown in FIG. 1. Of course, this is an example, and any physical property value can be adopted. In this embodiment, each of the multiple types (129 types) of raw coal has 17 types of physical property values, so the raw coal physical property data D3 can be expressed as a matrix of L×i, that is, 129×17. ASH: Ash content in coal FT: coal softening temperature MFT: Maximum fluidity temperature of coal logMF: Maximum fluidity of coal DT: coal resolidification temperature RF: Softening and melting temperature range (DT-FT) Ro: Average maximum reflectance of the vitrinite structure of the coal tissue TI: Amount of inactive components in coal (does not soften or melt) C(dry-base): The proportion of carbon in coal H(dry base): The proportion of hydrogen in coal N (dry base): The proportion of nitrogen in coal O(dry-base): Oxygen content in coal ORI: (Reference) Evaluation of Coal Coking Properties by Pyrolysis Products (Journal of the Japan Institute of Energy, Vol. 73, No. 4, 1994, p. 279-289) HRI: (Reference) Evaluation of coal coking properties using pyrolysis products (Journal of the Japan Institute of Energy, Vol. 73, No. 4, 1994, p. 279-289) dHI: (Reference) Evaluation of coal coking properties using pyrolysis products (Journal of the Japan Institute of Energy, Vol. 73, No. 4, 1994, p. 279-289) VM (dry-ash-free): The amount of volatile matter in coal

[0015] The dimension reduction processing unit 13 shown in FIG. 1 performs a dimension reduction process on a plurality of physical property values ​​of raw coal, and calculates dimension-reduced feature data D4 representing the plurality of physical property values. In this embodiment, one type of raw coal is expressed by 17 types of physical property values. That is, the number of dimensions indicating the characteristics of raw coal is 17. The dimension reduction processing unit 13 reduces the number of dimensions indicating the characteristics of raw coal to a number smaller than 17 by the dimension reduction process. By the dimension reduction, the number of dimensions can be reduced while reproducing the original characteristics, and it is possible to reduce the calculation cost and avoid the curse of dimensionality (overlearning). As an algorithm used for the dimension reduction process, various algorithms such as principal component analysis, factor analysis, multifactor analysis, multiple correspondence analysis, t-SNE, and Autoencoder can be used. In this embodiment, principal component analysis (PCA) is used, and the first principal component, the second principal component, and the third principal component are adopted as the feature data. Of course, this is not limited to this, and various changes are possible. For example, in the principal component analysis, the feature data may be only the first principal component, or the first and second principal components. That is, a predetermined number (any number equal to or greater than 1) of principal components from the first order onwards are selected as feature data. The predetermined number can be set appropriately depending on the required accuracy. The dimension reduction processing unit 13 executes the above-mentioned dimension reduction processing for all raw coals for which there is physical property data in the raw coal physical property data D3.

[0016] In this embodiment, when a principal component analysis was performed on the above 17 types of physical property values, it was found that the first principal component alone could reproduce approximately 64.9% of the features, the first and second principal components alone could reproduce approximately 80.8% of the features, the first to third principal components alone could reproduce approximately 86.8% of the features, and the first to fourth principal components alone could reproduce approximately 92.1% of the features. These are just examples, and since they change depending on the combination of physical property values ​​used, they are merely examples for understanding the dimensionality reduction in this disclosure and do not limit the scope of rights.

[0017] The cluster analysis unit 14 shown in FIG. 1 classifies each of the multiple types (L=129 types in this embodiment) of raw coal into one of multiple clusters by performing cluster analysis on the feature data D4 calculated by the dimension reduction processing unit 13. In this embodiment, the number of initial clusters is set to 35, and the cluster analysis is performed using the x-means algorithm to obtain a calculation result of the number of clusters k=39, but this is not limited as long as it is a cluster analysis. For example, non-hierarchical cluster analysis such as k-means++, g-means, x-means, and xg-means may be used, or hierarchical cluster analysis, DBSCAN, agglomerative clustering, and partition clustering may be used. In this embodiment, the number of clusters k is smaller than the number L of raw coal types, but is not limited to this. The number of clusters may be the same as the number of raw coal types, or the number of clusters may be larger than the number of raw coal types. Note that raw coals that have defects in one or more of the 17 types of physical properties (five exception brands) are not grouped together with other raw coals and are placed in a single cluster. Therefore, the total number of clusters k is 34+5=39.

[0018] 1 identifies the correspondence between raw coal (brand) and clusters, as shown in Fig. 1, based on the classification result of raw coal into clusters by the cluster analysis unit 14. Data D5 indicating the correspondence is stored in the memory unit 16.

[0019] FIG. 5 is a diagram showing an example of teacher data D2. In FIG. 5, teacher data D2 based on the generation results from 89 to 100 in the generation order is shown, which is a part of teacher data D2. Here, one generation result is an aggregate value of the generation results using the same blending data. For example, even if coke is generated for only one day using the first blending data, or generated for multiple consecutive days using the first blending data, both are treated as one generation result as long as the blending data is not changed. If coke is generated using the first blending data, then generated using the second blending data, and then generated using the first blending data, the generation results using the first blending data before the second blending data and the generation results using the first blending data after the second blending data are not integrated and become separate generation results because of the change in blending data.

[0020] The teacher data generating unit 11 shown in FIG. 1 generates teacher data D2 based on a plurality of production results included in the actual measurement data D1 acquired by the actual measurement data acquiring unit 10 and their production order. As shown in FIG. 1, the teacher data D2 inputs the blending data of the production results to the prediction model 20, and outputs the quality value of the production results from the prediction model 20. The blending data may include the same blending as long as it is different from the previous and subsequent production results (for example, the generation order is 95 and 100, 97 and 99). Furthermore, the teacher data generating unit 11 generates teacher data D2 having a first feature amount and a second feature amount as input to the prediction model 20. The first feature amount and the second feature amount will be described with reference to FIG. 5. The first feature amount and the second feature amount are feature amounts related to the time series change of the past performance of the quality value, and are provided not only for quality value prediction based on the blending data, but also for predicting the factor of the variation of the quality value when repeatedly producing coke in a coke oven from the data related to the most recent time series change.

[0021] As shown in FIG. 5, the first feature is called a lag feature, and is a feature based on one of a plurality of past production results. Although one type of first feature may be used, in this embodiment, four types of first feature (l3, l5, l7, l10) are used. The first feature is a value based on a quality value corresponding to one of a plurality of production results generated before the production result related to the input blending data. Specifically, one production result that is the basis of the first feature is a production result generated a first predetermined number of times before the production result related to the input blending data. In this embodiment, there are multiple types (four types) of first feature, and the first predetermined number of each first feature is 3, 5, 7, and 10, respectively. As shown in FIG. 5, the lag feature (l3) input together with the blending data with a generation order of 100 is the quality value (α97) of the generation result three generations ago, the lag feature (l5) is the quality value (α95) of the generation result five generations ago, the lag feature (l7) is the quality value (α93) of the generation result seven generations ago, and the lag feature (l10) is the quality value (α90) of the generation result ten generations ago. Thus, in this embodiment, there are multiple types of first feature, and the first predetermined number of each first feature is different. Note that the number of first feature can be changed arbitrarily, and the first predetermined number can also be changed arbitrarily.

[0022] As shown in FIG. 5, the second feature is called a rolling feature, and is a feature based on at least two generation results that are consecutive in the generation order among a plurality of past generation results. The second feature may be of the first type, but in this embodiment, two types of second feature (r5, r7) are used. The second feature is a value based on the quality value corresponding to at least two generation results that are consecutive in the generation order among a plurality of generation results that were generated before the generation result related to the input blending data. Specifically, the at least two generation results that are the basis of the second feature are a plurality of generation results from a second predetermined number before the generation result related to the input blending data to a third predetermined number before. In this embodiment, there are a plurality of types (two types) of second feature, and the second predetermined number of each second feature is 1, and the third predetermined number is 5 and 7, respectively. As shown in FIG. 5, the rolling feature (r5) input together with the blending data whose generation order is 100 is a statistical value (average value) of the quality values ​​(α99 to 95) of the generation results from one to five before. The rolling feature (r7) is a statistical value (average value) of the quality values ​​(α99-93) of the production results from the previous to the seventh previous. In this manner, in this embodiment, there are a plurality of types of second feature, and at least one of the second predetermined number and the third predetermined number of each second feature is different.

[0023] In addition, the third predetermined number may be the same and the second predetermined number may be different, or both the second predetermined number and the third predetermined number may be different. Furthermore, the number of second feature amounts can be changed arbitrarily, and the second predetermined number and the third predetermined number can be set arbitrarily. In this embodiment, the second feature amount is the average value of the quality values ​​of a plurality of production results, but is not limited to this and may be any statistical value. Examples of the statistical value include the average value, the mode, and the median. In FIG. 5 , the first feature amounts and second feature amounts whose generation orders are 89 to 97 are not shown, but are generated according to the same rules as the first feature amounts and second feature amounts whose generation orders are 98 to 100.

[0024] The teacher data generating unit 11 shown in FIG. 1 further generates teacher data D2 based on the correspondence between raw coal and clusters stored in the storage unit 16. Specifically, the blending ratio of each raw coal in the measured data D1 is converted into the blending ratio of each cluster. For example, as shown in FIG. 1, when cluster 2 corresponds to raw coal 1 (brand 1) and raw coal 3 (brand 3), the blending ratios of all raw coals corresponding to cluster 2 in the measured data D1 are aggregated and converted into blending ratios on a cluster basis. Then, the teacher data generating unit 11 generates teacher data D2 in which the blending ratios of each cluster are converted and input to the prediction model 20, and the quality value of coke is output. In this embodiment, the number L of raw coal types is 129, but the number of clusters is 39, so the number of input parameters to the prediction model 20 can be reduced. In this embodiment, since the physical property values ​​differ depending on the lot even for the same raw coal, the physical property values ​​are measured for each lot. Therefore, approximately 1,500 lots were accurately classified (feature extracted) into clusters with a number of k = 39, which we believe has improved accuracy.

[0025] As shown in FIG. 1, the actual measurement data D1 is based on the actual measurement value of the coke produced in the coke oven, so that even if the blending ratio is the same, the quality value may include variation. For example, even if the blending ratio is the same on April 1 and April 2, the quality value may vary. If the variation is small, the adverse effect on the prediction accuracy of the prediction model 20 is small, but if the variation is large, the prediction accuracy may be adversely affected. Therefore, in order to reduce this adverse effect, it is preferable to configure the actual measurement data acquisition unit 10 as follows. That is, it is preferable that the actual measurement data acquisition unit 10 has an identical blending extraction unit (not shown), a representative value determination unit (not shown), and a data aggregation unit (not shown). The identical blending extraction unit (not shown) extracts multiple quality values ​​of coke having the same combination of blending ratios of multiple types of raw coal from the acquired actual measurement data. The representative value determination unit (not shown) calculates a characteristic value of the quality that is the representative value based on the multiple extracted quality values. Examples of characteristic values ​​include the average value, the mode value, and the median value, but in this embodiment, the average value is used. It has been found that the average value is closer to the true value of the population than the mode value and the median value. The data aggregation unit (not shown) converts one calculated representative value into new actual measurement data aggregated as a new quality value. This makes it possible to reduce or eliminate the variation in data in which the same composition has different quality values. In addition, in order to further improve accuracy, after the representative value (average value) is calculated by the representative value determination unit (not shown), only quality values ​​whose absolute value difference from the representative value (average value) is within a predetermined threshold value may be left, and a representative value may be calculated again from the remaining quality values, and the representative value may be converted into new actual measurement data aggregated as a new quality value. In addition, a data deletion unit may be provided that limits the actual measurement data used for machine learning to only data obtained when the same composition is continuously manufactured a predetermined number of times or more (for example, five or more measurements), and deletes data that has not been continuously manufactured for a predetermined period of time. Note that each of these units may be omitted. The data deletion unit may delete data that has only one identical composition. This is because data that has only one identical composition cannot obtain a representative value (average value), and the quality value may not be a true value. Also, the acquired measured data D1 may include data on operation, and outliers may be removed based on the operation data. Examples of data on operation include coke oven temperature data and measurement data on coal moisture. Data that are outliers in a box-and-whisker plot of oven temperature and moisture over the entire operation period may be removed. Outliers are in the range from the minimum value to the first quartile in the box-and-whisker plot (bottom 25%) and the range from the third quartile to the maximum value in the box-and-whisker plot (top 25%). As described above, the actual data acquisition unit 10 has an identical blend extraction unit (not shown), a representative value determination unit (not shown), and a data aggregation unit (not shown), but the actual data acquisition unit 10 may acquire actual data D1 on which these processes have been performed in advance outside the system.

[0026] <Prediction model generation system 2> 1, the prediction model generation system 2 includes a learning unit 21 that constructs a prediction model 20 by learning through machine learning using teacher data D2. As the prediction model 20, various models such as linear regression, regression tree, random forest, support vector machine, neural network, ensemble, etc. can be used as long as they are supervised machine learning models. In this embodiment, a random forest is used, but the present invention is not limited to this.

[0027] <Coke quality prediction system 3> The coke quality prediction system 3 shown in FIG. 1 includes an actual measurement data acquisition unit 10, a prediction target data acquisition unit 30, a conversion unit 31, and a prediction unit 32.

[0028] The actual measurement data acquiring unit 10 in the coke quality prediction system 3 is the same as the actual measurement data acquiring unit 10 in the teacher data generation system 1, and the acquired actual measurement data D1 includes blending data, but is not limited to this. The performance data acquired by the actual measurement data acquiring unit 10 in the coke quality prediction system 3 only needs to include a plurality of production results including the quality values ​​of the actually generated coke and the production order, and does not need to include blending data.

[0029] The prediction target data acquisition unit 30 acquires prediction target blending data D6 indicating the blending ratios of each of a plurality of raw coals constituting a blended coal, which is a raw material for coke whose quality value is to be predicted. The prediction target blending data D6 includes blending ratios V1, V2, V3, ..., V L-1 ,V L Represents.

[0030] The conversion unit 31 adds the first feature amount and the second feature amount to the prediction target combination data D6 based on the multiple generation results and their generation order included in the actual measurement data D1 acquired by the actual measurement data acquisition unit 10, and converts it into converted input data D7. The first feature amount and the second feature amount can be calculated by the same process as the feature amount added to the teacher data D2 described using FIG.

[0031] The conversion unit 31 converts the blending ratio of each raw coal unit acquired into a blending ratio of each cluster unit using a correspondence relationship (data D5) that previously associates raw coal with a cluster. The conversion process by the conversion unit 31 is the same as the conversion process by the teacher data generation unit 11. The conversion unit 31 converts the blending ratios (X1, X2, X3, ..., X k-1 ,X k ) is converted into post-conversion input data D7 representing the

[0032] The prediction unit 32 uses the prediction model 20 to calculate the mixture ratios (X1, X2, X3, . . . , X k-1 ,X k ), the first feature amount, and the second feature amount, the coke quality value is predicted from the blending data (blend ratio of each cluster unit), the first feature amount, and the second feature amount. As described above, the prediction model 20 is constructed in advance by machine learning by the learning unit 21 of the prediction model generation system 2 so as to output the coke quality value from the blending data (blend ratio of each cluster unit), the first feature amount, and the second feature amount. By inputting the converted input data D7 to the prediction model 20, the prediction model 20 outputs the coke quality value.

[0033] [How to generate training data] The teacher data generation method will be described with reference to Fig. 2. The teacher data generation method is executed by one or more processors constituting the teacher data generation system 1. As shown in Fig. 2, steps ST100 to ST103 and steps ST104 to ST107 can be executed in parallel, and may be executed in parallel, or one may be executed first. The order is not limited. In step ST100, the teacher data generation system 1 acquires raw coal physical property data D3 in which multiple physical property values ​​of each raw coal are associated with each of multiple types (L types) of raw coal. In the next step ST101, the teacher data generation system 1 performs a dimension reduction process on the multiple physical property values ​​of each raw coal in the raw coal physical property data D3 for each of the multiple types (L types) of raw coal, and calculates dimension-reduced feature data D4 representing the multiple physical property values. In the next step ST102, the teacher data generation system 1 classifies the calculated feature data D4 into one of multiple clusters by performing cluster analysis. Next, in step ST103, the teacher data generation system 1 identifies the correspondence between raw coal and clusters based on the results of the cluster analysis. In the next step ST104, the teacher data generation system 1 acquires actual measurement data D1 in which the blending ratio of each of the multiple types (L types) of raw coal is associated with the grade value of the actually obtained coke.

[0034] In step ST105, the teacher data generation system 1 extracts multiple coke quality values ​​in which the combination of blending ratios of multiple types (L type) of raw coal is consecutively the same from the acquired actual measurement data D1. In the next step ST106, the teacher data generation system 1 calculates a representative quality characteristic value (average value in this embodiment) based on the multiple extracted quality values. In the next step ST107, the teacher data generation system 1 converts the calculated representative value into new aggregated actual measurement data as a new quality value. Note that steps ST105 to ST107 can be omitted.

[0035] In the next step ST108, the teacher data generation system 1 uses the correspondence between raw coal and clusters (data D5) to convert the blending ratios of each raw coal in the actual measurement data D1 into blending ratios of each cluster, and generates teacher data that takes the blending ratios of each converted cluster as input and outputs the coke quality value.

[0036] In the next step ST109, the teacher data generation system 1 generates teacher data including the first feature amount and the second feature amount as input, based on the multiple generation results and the generation order included in the result data.

[0037] [How to generate a coke quality prediction model] The coke quality prediction model generation method will be described with reference to Fig. 3. The coke quality prediction model generation method is executed by one or more processors constituting the prediction model generation system 2. As shown in Fig. 3, in step ST200, the above-mentioned teacher data generation process is executed to generate teacher data D2. In the next step ST201, the prediction model generation system 2 generates a prediction model 20 that predicts the quality value of coke by performing machine learning using the teacher data.

[0038] [Coke quality prediction method] The coke quality prediction method will be described with reference to Fig. 4. The coke quality prediction method is executed by one or more processors constituting the coke quality prediction system 3. As shown in Fig. 4, in step ST300, the coke quality prediction system 3 calculates the blending ratios (V1, V2, V3, ..., V L-1 ,V L In the next step ST301, the coke quality prediction system 3 obtains the blending ratios (V1, V2, V3, ..., V L-1 ,V L ) to the cluster-wise blending ratio (X1, X2, X3, …, X k-1 ,X k) into the converted blending data [blending ratios of each cluster unit (X1, X2, X3, ..., X4, ...]], using a prediction model 20 constructed by machine learning so as to output a coke quality value from the blending data (blending ratios of each cluster unit), the first feature amount, and the second feature amount. In the next step ST304, the coke quality prediction system 3 converts the blending data [blending ratios of each cluster unit (X1, X2, X3, ..., X4, ...]], into the converted blending data [blending ratios of each cluster unit (X1, X2, X3, ..., X4, ...]]], using a prediction model 20 constructed by machine learning so as to output a coke quality value from the blending data (blending ratios of each cluster unit), the first feature amount, and the second feature amount. k-1 ,X k )], and predict the quality value of the coke from the first feature amount and the second feature amount.

[0039] [Effects of this disclosure] In order to demonstrate the effects of the present disclosure, Comparative Example 1 and Example 1 are shown below. In comparing Comparative Example 1 and Example 1, the same measured data D1 is used.

[0040] Example 1 [First feature amount and second feature amount are present] The system configuration is shown in FIG. 1. In the training data generation process shown in FIG. 2, all of ST100 to 109 are executed. The number of brands L (number of types of raw coal) used is 129. The number of physical property values ​​j=17. The number k of clusters constituting the blending data is 39. The prediction model 20 is a random forest. There are four types of first feature values. There are two types of second feature values. The number of input parameters to the prediction model 20 is 39+4+2=45.

[0041] Comparative Example 1 [No first feature amount or second feature amount] This is a configuration in which the first feature amount and the second feature amount are excluded from the input parameters to the prediction model 20 in the first embodiment. In the teacher data generation process shown in FIG. 2, ST100 to ST103 are not executed, and step ST109 is not executed. There are 39 input parameters to the prediction model 20. The rest is the same as in the first embodiment.

[0042] FIG. 6 is a diagram showing prediction results and actual values ​​in Example 1 and Comparative Example 1. The vertical axis indicates the coke quality value (coke strength), and the horizontal axis corresponds to the generation order of the actual measurement data D1. Time point 2 is later than time point 1. The correlation coefficient for the actual values ​​of Comparative Example 1 is 0.020. The correlation coefficient for the actual values ​​of Example 1 is 0.343. It can be seen that Example 1 has an improved correlation coefficient and prediction accuracy compared to Comparative Example 1. This makes it clear that the first feature amount and the second feature amount used in ST109 in FIG. 2 are very effective in improving prediction accuracy.

[0043] [1] As described above, in this embodiment, the teacher data generation method is a method for generating teacher data D2 used for machine learning of a prediction model 20 that predicts coke grade, and includes a step of acquiring multiple production results including blending data related to the blending ratio of multiple types of raw coal and the grade value of coke actually generated based on the blending ratio, together with the production order, and a step of generating teacher data D2 based on the multiple production results. In the step of generating the teacher data D2, the blending data is used as input and the grade value is used as output. Furthermore, the teacher data D2 may be generated by inputting a first feature based on the grade value corresponding to one of the multiple production results generated before the production result related to the input blending data, and a second feature based on the grade values ​​corresponding to at least two production results that are consecutive in the production order among the multiple production results generated before the production result related to the input blending data. In this way, since the first feature value and the second feature value are included as inputs to the prediction model 20, it is possible to improve the prediction accuracy of the quality value by taking into account factors that cause the quality value to vary when coke is repeatedly produced in a coke oven.

[0044] [2] In the teacher data generation method described in [1] above, the one generation record that serves as the basis for the first feature may be a generation record that was generated a first predetermined number of times before the generation record related to the input blending data, and the at least two generation records that serve as the basis for the second feature may be a plurality of generation records that are generated from a second predetermined number of times before the generation record related to the input blending data to a third predetermined number of times before. In the coke manufacturing process, operating conditions such as furnace temperature and moisture content and coal blending contents that affect coke quality do not change suddenly but change gradually from past measured values, so it is considered that the coke quality depends on the past coke quality value. As described above, by determining the first predetermined number, the second predetermined number, and the third predetermined number, the feature amount is calculated according to a uniform rule compared to the case where a randomly selected production record is used. Since calculating the feature amount according to a uniform rule matches the idea that the coke quality depends on the past coke quality value, it is possible to provide appropriate processing for machine learning and improve the estimation accuracy of the coke quality value.

[0045] [3] In the teacher data generation method described in [2] above, the teacher data D2 may include multiple types of first features having different first predetermined numbers, and multiple types of second features having different second predetermined numbers and / or third predetermined numbers. In this way, since a plurality of types of feature values ​​with different ranges of production records on which the feature values ​​are based are included, it is possible to take into consideration factors that affect the quality value due to time-series changes, and thus it is possible to improve the prediction accuracy of the quality value.

[0046] [4] In the teaching data generation method described in any one of [1] to [3] above, the blending data may be data representing the blending ratio of each of multiple types of raw coal in terms of a blending ratio of a cluster unit that is pre-associated with the raw coal. If the number of clusters is smaller than the number of raw coal types, the number of input parameters to the prediction model 20 can be reduced, making it possible to improve the prediction accuracy of the grade value.

[0047] [5] The method for generating a prediction model for coke quality in this embodiment may include a step of executing a teacher data generation method described in any one of [1] to [4] above to generate teacher data D2, and a step of generating a prediction model 20 that predicts the quality value of coke by performing machine learning using the generated teacher data D2. It becomes possible to provide a predictive model 20.

[0048] [6] The coke quality prediction method of this embodiment includes the steps of acquiring prediction target blending data regarding the blending ratio of multiple types of raw material coal representing the blend of raw materials for the coke to be predicted, acquiring multiple production results including the quality values ​​of the actually generated coke together with the production order, and predicting the quality value of the coke from an input including the prediction target blending data using a prediction model 20 constructed by machine learning to output a coke quality value from an input including the blending data regarding the blending ratio of multiple types of raw material coal, wherein the input including the prediction target blending data may include a first feature based on a quality value corresponding to one of the multiple production results, and a second feature based on quality values ​​corresponding to at least two of the multiple production results that are consecutive in the production order. In this way, since the first feature value and the second feature value are included as inputs to the prediction model 20, it is possible to improve the prediction accuracy of the quality value by taking into account factors that cause the quality value to vary when coke is repeatedly produced in a coke oven.

[0049] [7] The teacher data generation system 1 of this embodiment is a system that generates teacher data to be used in machine learning of a predictive model that predicts coke quality, and may be equipped with one or more processors that execute the methods described in any of [1] to [4] above.

[0050] [8] The coke quality prediction model generation system 2 of this embodiment may include a learning unit 21 that generates a prediction model 20 that predicts the quality value of coke by machine learning using training data D2 generated by any of the methods described in [1] to [4] above.

[0051] [9] The coke quality prediction system 3 of this embodiment includes a prediction target data acquisition unit 30 that acquires prediction target blending data regarding the blending ratio of multiple types of raw material coal that represents the blend of the raw materials of the coke to be predicted, an actual measurement data acquisition unit 10 that acquires multiple production results including the quality values ​​of the actually generated coke together with the production order, and a prediction unit 32 that predicts the quality value of the coke from an input including the prediction target blending data using a prediction model 20 constructed by machine learning to output a coke quality value from an input including blending data regarding the blending ratio of multiple types of raw material coal, and the input including the prediction target blending data may include a first feature based on a quality value corresponding to one of the multiple production results, and a second feature based on quality values ​​corresponding to at least two of the multiple production results that are consecutive in the production order.

[0052]

[10] The program for generating the training data may be configured to cause one or more processors to execute any one of the above methods [1] to [4].

[0053]

[11] The program for generating a prediction model of coke quality may cause one or more processors to execute the steps of: executing any of the methods described in [1] to [4] above to generate training data D2; and performing machine learning using the generated training data D2 to generate a prediction model 20 that predicts the quality value of coke.

[0054]

[12] A program for predicting coke quality may include a step of acquiring prediction target blending data regarding the blending ratio of multiple types of raw coal representing the blend of raw materials for the coke to be predicted, a step of acquiring multiple production results including the quality values ​​of actually generated coke together with the production order, and a step of predicting the quality value of the coke from an input including the prediction target blending data using a prediction model constructed by machine learning to output a coke quality value from an input including the blending data regarding the blending ratio of multiple types of raw coal, wherein the input including the prediction target blending data may include a first feature based on a quality value corresponding to one of the multiple production results, and a second feature based on quality values ​​corresponding to at least two of the multiple production results that are consecutive in the production order.

[0055] A computer-readable temporary recording medium according to the present embodiment stores the above program.

[0056] Although the embodiments of the present disclosure have been described above based on the drawings, the specific configurations should not be considered to be limited to these embodiments. The scope of the present disclosure is indicated not only by the description of the above-mentioned embodiments but also by the claims, and further includes all modifications within the meaning and scope equivalent to the claims.

[0057] The structures employed in the above-described embodiments may be employed in any other embodiment. The specific configurations of the components are not limited to the above-described embodiments, and may be modified in various ways without departing from the spirit of the present disclosure.

[0058] (A) For example, in the present embodiment, the grade value is described as the coke strength DI, but the present invention is not limited to this, and any grade (physical property value) of coke can be used.

[0059] (B) In this embodiment, the blending data is a conversion of the blending ratio of each of the multiple raw coal types into the blending ratio of each cluster unit, but is not limited to this. For example, the blending data may be the blending ratio of each of the multiple raw coal types themselves. [Explanation of symbols]

[0060] 1: Teacher data generation system 2: Predictive model generation system 3: Coke quality prediction system 10: Actual data acquisition section 20: Predictive model 21: Learning Department 30: Prediction target data acquisition section 32: Prediction section D1: Actual data D2: Training data D6: Prediction target mix data

Claims

1. Acquiring blending data on blending ratios of a plurality of types of raw coal and a plurality of production results including a quality value of coke actually produced based on the blending ratios together with a production order; generating training data based on the plurality of generation results; Including, A method for generating teacher data, in which, in the step of generating the teacher data, the blending data is used as input, the quality value is used as output, and the teacher data is generated by inputting a first feature based on the quality value corresponding to one of the multiple generation results generated before the generation result related to the blending data used as input, and a second feature based on the quality value corresponding to each of at least two generation results that are consecutive in generation order among the multiple generation results generated before the generation result related to the blending data used as input.

2. The one production record serving as a basis for the first characteristic amount is a production record generated a first predetermined number of times before the production record related to the input blending data, The method for generating teacher data as described in claim 1, wherein the at least two generation records that serve as the basis for the second feature are a plurality of generation records ranging from a second predetermined number before to a third predetermined number before the generation record related to the input blending data.

3. 3. The method for generating teacher data according to claim 2, wherein the teacher data includes a plurality of types of the first features, each of which has a different first predetermined number, and a plurality of types of the second features, each of which has a different second predetermined number and / or a different third predetermined number.

4. The method for generating training data according to claim 1 , wherein the blending data represents the blending ratio of each of the plurality of types of raw coal in terms of a blending ratio of a cluster unit that is pre-associated with the raw coal.

5. A step of generating teacher data by executing a method according to any one of claims 1 to 4; generating a prediction model for predicting a grade value of the coke by performing machine learning using the generated teacher data; A method for generating a prediction model for coke grade, comprising:

6. A step of acquiring prediction target blending data regarding a blending ratio of a plurality of types of raw coal representing a blending ratio of raw materials of a coke to be predicted; acquiring a plurality of production records including grade values ​​of actually produced coke together with a production order; A step of predicting a quality value of coke from an input including blending data to be predicted, using a prediction model constructed by machine learning so as to output a quality value of coke from an input including blending data on a blending ratio of a plurality of types of raw coal; Including, A coke quality prediction method, wherein the input including the prediction target blend data includes a first feature based on the quality value corresponding to one of the multiple production results, and a second feature based on the quality values ​​corresponding to at least two production results among the multiple production results that are consecutive in production order.

7. A system for generating training data used in machine learning of a prediction model for predicting coke quality, comprising: A teacher data generation system comprising one or more processors for executing the method according to any one of claims 1 to 4.

8. A coke quality prediction model generation system comprising: a learning unit that generates a prediction model for predicting a quality value of the coke by performing machine learning using teacher data generated by the method according to any one of claims 1 to 4.

9. A prediction target data acquisition unit that acquires prediction target blend data regarding a blend ratio of multiple types of raw coal representing the blend of raw materials of coke to be predicted; an actual measurement data acquisition unit that acquires a plurality of production records including grade values ​​of actually produced coke together with the order of production; A prediction unit that predicts a coke quality value from an input including blending data to be predicted, using a prediction model constructed by machine learning so as to output a coke quality value from an input including blending data on a blending ratio of a plurality of raw coals; Equipped with A coke quality prediction system, wherein the input including the prediction target blend data includes a first feature based on the quality value corresponding to one of the multiple production results, and a second feature based on the quality values ​​corresponding to at least two production results among the multiple production results that have consecutive production orders.

10. A program for causing one or more processors to execute the method according to any one of claims 1 to 4.

11. A step of generating teacher data by executing a method according to any one of claims 1 to 4; generating a prediction model for predicting a grade value of the coke by performing machine learning using the generated teacher data; A program for causing one or more processors to execute the above.

12. A step of acquiring prediction target blending data regarding a blending ratio of a plurality of types of raw coal representing a blending ratio of raw materials of a coke to be predicted; acquiring a plurality of production records including grade values ​​of actually produced coke together with a production order; A step of predicting a quality value of coke from an input including blending data to be predicted, using a prediction model constructed by machine learning so as to output a quality value of coke from an input including blending data on a blending ratio of a plurality of types of raw coal; on one or more processors; The input including the prediction target combination data includes a first feature based on the quality value corresponding to one of the plurality of production results, and a second feature based on the quality value corresponding to at least two of the plurality of production results that are consecutive in production order, program.

Images