Characteristic derivation method based on clinker components

Abstract

The invention provides a characteristic derivation method based on clinker components, and belongs to the technical field of process data processing. The method comprises the following steps of: 1, acquiring time sequence data of process parameters on a cement production line through a quality control system, a quality management system and a sintering system DSC, and acquiring a clinker strength measured value of a corresponding batch through a clinker strength application system; meanwhile, relevant data of the clinker are aligned and integrated according to company names, sampling positions on a production line and date information. 2, abnormal values in the data are filtered through a z-score method; and 3, constructing a mineral phase dynamic statistical feature data set based on daily details and a derivative feature data set based on a time sequence for the data. And 4, combining the mineral phase dynamic statistical feature data set and the derivative feature data set based on the time sequence with unprocessed original data according to a company name, a sampling position on a production line and a date, and outputting an enhanced feature data set. According to the method, the data expression effect is improved, so that the data availability is enhanced.

Description

Technical Field

[0001] This invention belongs to the field of process data processing technology. Specifically, this invention relates to a method for characterization based on clinker components. Background Technology

[0002] In the cement production process, clinker, as the main active component of cement, is the decisive factor in the final strength performance of cement. Accurate prediction of clinker strength helps optimize production processes, reduce energy consumption, and improve production efficiency.

[0003] Chinese Patent 113033923A discloses a method, apparatus, and system for predicting, evaluating, and optimizing the performance of cement clinker. It employs machine learning to construct a performance prediction model for cement clinker based on historical production data. Production data of the cement clinker to be tested is acquired and input into the performance prediction model to obtain predicted performance data. The performance of the cement clinker to be tested is evaluated and analyzed. If the predicted performance data meets the set requirements, it is directly output. If the predicted performance data does not meet the set requirements, the production data is optimized until the predicted performance data meets the set requirements.

[0004] Current clinker strength prediction models rely on clinker chemical composition data as a single static indicator, which suffers from insufficient information dimensions and limited expressive power. They cannot fully reflect the interaction between components, the formation law of mineral phases, and the dynamic fluctuations of the production process, making it difficult to meet the needs of refined modeling and in-depth analysis. Summary of the Invention

[0005] The present invention aims to provide a feature derivation method based on clinker components, so as to improve the expression effect of data by performing feature derivation on clinker-related data, thereby enhancing the usability of data.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] This invention provides a method for characterizing clinker components, comprising the following steps:

[0008] Step 1: Obtain relevant data on clinker;

[0009] Step 2: Perform data preprocessing on relevant clinker data;

[0010] Step 3: Construct a derived feature dataset based on the preprocessed data;

[0011] Step 4: Merge the derived feature dataset with the original data to output the enhanced feature dataset.

[0012] In step one, the time-series data of process parameters on the cement production line are obtained through the quality control system, quality management system, and calcination system DSC, and the measured values ​​of clinker strength for the corresponding batch are obtained through the clinker strength application system.

[0013] In step one, the relevant data of clinker are aligned and integrated based on the company name, sampling location on the production line, and date information.

[0014] In step two, outliers in the data are filtered out using the z-score method.

[0015] In step three, a dynamic statistical feature dataset of mineral phases based on daily details and a derived feature dataset based on time series are constructed for the data.

[0016] The method for constructing dynamic statistical features of mineral phases based on daily details is as follows: the data are grouped according to the company, sampling location on the production line, and date dimension. Various statistical calculations are performed on the daily detailed data of mineral phases, rate values, clinker free calcium, and liter weight. Each feature is derived into seven types of derived features: maximum value, minimum value, mean, standard deviation, median, difference ratio, and summation.

[0017] One method for constructing time-series-based derived features is to quantify the timestamps corresponding to the data, introduce the day_diff feature, and record the difference in the number of months, weeks, and days from the base date.

[0018] In step four, the daily detailed mineral phase dynamic statistical feature dataset and the time-series derived feature dataset are merged with the unprocessed raw data by company name, sampling location on the production line, and date, and an enhanced feature dataset is output.

[0019] The effectiveness of feature derivation is determined by comparing the impact of data before and after feature derivation on the performance of the prediction model.

[0020] The technical effects of this invention are as follows:

[0021] (1) This invention achieves efficient expression and analysis of clinker chemical composition data through systematic and multi-dimensional feature construction, thereby enhancing the usability of data in intensity prediction, process optimization and intelligent control.

[0022] (2) The present invention enhances the interpretability of the model, breaks through the limitations of traditional single components, and deeply mines production process information through multi-dimensional feature derivation such as dynamic statistics of mineral phases. This not only significantly improves the interpretability and effectiveness of the features, but also solves the problems of traditional single features and lack of process correlation.

[0023] (3) This invention improves the predictive performance of the model by innovatively adopting a time-series numerical coding strategy to replace the traditional date processing method, effectively capturing the environmental cycle pattern. At the same time, it avoids dimensional explosion through high-level feature aggregation, accurately quantifies the kiln condition stability and the influence of the external environment, and greatly improves the model accuracy.

[0024] (4) This invention is generalizable and provides a systematic and generalizable process for deriving the characteristics of clinker components. The method is flexible, scalable, and adaptable to different plants and process routes. Attached Figure Description

[0025] This manual includes the following figures, which illustrate the following:

[0026] Figure 1 This is a flowchart of a method for characterizing clinker components according to the present invention. Detailed Implementation

[0027] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, in order to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention, and to facilitate its implementation.

[0028] This invention provides a method for characterizing clinker components, comprising the following steps:

[0029] Step 1: Obtain relevant data on clinker;

[0030] Step 2: Perform data preprocessing on relevant clinker data;

[0031] Step 3: Construct a derived feature dataset based on the preprocessed data;

[0032] Step 4: Merge the derived feature dataset with the original data to output the enhanced feature dataset.

[0033] In step one, the time-series data of process parameters on the cement production line are obtained through the quality control system, quality management system, and calcination system DSC, and the measured values ​​of clinker strength for the corresponding batch are obtained through the clinker strength application system.

[0034] In step one, the relevant data of clinker are aligned and integrated based on the company name, sampling location on the production line, and date information.

[0035] In step two, outliers in the data are filtered out using the z-score method.

[0036] In step three, a dynamic statistical feature dataset of mineral phases based on daily details and a derived feature dataset based on time series are constructed for the data.

[0037] The method for constructing dynamic statistical features of mineral phases based on daily details is as follows: the data are grouped according to the company, sampling location on the production line, and date dimension. Various statistical calculations are performed on the daily detailed data of mineral phases, rate values, clinker free calcium, and liter weight. Each feature is derived into seven types of derived features: maximum value, minimum value, mean, standard deviation, median, difference ratio, and summation.

[0038] One method for constructing time-series-based derived features is to quantify the timestamps corresponding to the data, introduce the day_diff feature, and record the difference in the number of months, weeks, and days from the base date.

[0039] In step four, the daily detailed mineral phase dynamic statistical feature dataset and the time-series derived feature dataset are merged with the unprocessed raw data by company name, sampling location on the production line, and date, and an enhanced feature dataset is output.

[0040] The effectiveness of feature derivation is determined by comparing the impact of data before and after feature derivation on the performance of the prediction model.

[0041] The following describes in detail a feature derivation method based on clinker components according to the present invention.

[0042] Time-series data of process parameters on the cement production line are acquired through the quality control system, quality management system, and calcination system (DSC). Measured clinker strength values ​​for corresponding batches are obtained through the clinker strength application system. These process parameters include, but are not limited to, clinker composition, mineral composition, dicalcium silicate (C2S), tetracalcium aluminoferrite (C4AF), tricalcium aluminate (C3A), calcination process, and mineralizer parameters. Due to the numerous processes on the cement production line, the inclusion of time-series data of process parameters better reflects the accuracy of the data. Furthermore, the collected data is aligned and integrated based on company name, sampling location on the production line, and date information.

[0043] In the cement production process, data thresholds are determined based on process rules and the macroscopic value ranges of various characteristics. Data preprocessing is performed on clinker-related data. This invention uses the z-score method to filter outliers. The principle of the z-score method is to calculate the standard deviation distance (z-value) between a data point and the mean to determine its degree of deviation from the overall distribution, and to remove data points with larger absolute values ​​of the standard deviation distance as outliers. This invention describes the data preprocessing process through examples, using the statistical z-score method to find and filter outliers, setting the z-score to a large value (z-score=5) to filter only extreme outliers.

[0044] This invention can also filter chemical elements according to process rules, such as: C3S + C2S + C3A + C4AF >= 90%; C3S > 40%; C2S > 15%; 5 <= C3A <= 11%; 0.1 <= f-CaO <= 2.5%. Alternatively, filtering can be based on the macroscopic value range of each characteristic, such as 0.5%. <f-CaO<1.0%;5.0%<Al2O3<6.0%;50%<C3S<80%;5%<C2S<30%;3%<C3A<30%;2%<C4AF<20%;50%<CaO<70%;2%<Fe2O3<6%。

[0045] Based on the preprocessed data, this invention constructs a derived feature dataset by creating a daily detailed mineral phase dynamic statistical feature dataset and a time-series-based derived feature dataset. The method for constructing the daily detailed mineral phase dynamic statistical features involves grouping the data according to company, sampling location on the production line, and date. Various statistical calculations are performed on the daily detailed data of mineral phases and rate values. Each feature yields seven types of derived features: maximum value, minimum value, mean, standard deviation, median, difference ratio, and summation, used to characterize the long-term and short-term stability of the calcination process. Similarly, for the daily detailed data of core kiln operating indicators (clinker free calcium and liter weight), these seven statistical features are derived. The daily standard deviation of clinker free calcium is used to characterize the stability of kiln temperature control on that day, and the daily standard deviation of liter weight is used to reflect the uniformity of clinker quality on that day. This method retains more daily detailed information to achieve hourly data-assisted daily prediction while avoiding feature dimension explosion. The mineral phases in this invention include SiO2, Al2O3, Fe2O3, CaO, MgO, K2O, Na2O, R2O, SO3, C3S, C2S, C3A, and C4AF; the ratio values ​​include KH, KH-, SM, and IM, where KH and KH- are lime saturation coefficients, SM is the silicon ratio, and IM is the aluminum ratio. In the embodiments of this invention, we recalculated these elements using the following formula:

[0046] KH=(CaO-1.65*Al2O3-0.35*Fe2O3-0.7*SO3) / (2.8*SiO2)

[0047] KH-=((CaO-f-CaO)-1.65*Al2O3-0.35*Fe2O3-0.7*SO3) / (2.8*SiO2)

[0048] SM = SiO2 / (Al2O3+Fe2O3)

[0049] IM = Al2O3 / Fe2O3

[0050] One method for constructing time-series-based derived features is to quantify the timestamps corresponding to the data and introduce the day_diff feature, i.e., the difference in the number of days, recording the difference in the number of months, weeks, and days compared to the base date. This fully considers the relative time difference between different data points, reflecting the impact of the external environment on clinker cooling. Extracting abstract date features helps to capture long-term trends or cyclical change patterns.

[0051] The derived feature dataset is merged with the original data to output an enhanced feature dataset. Specifically, the daily detailed mineral facies dynamic statistical feature dataset and the time-series derived feature dataset are merged with the unprocessed original data by company name, sampling location, and date to output an enhanced feature dataset.

[0052] The effectiveness of feature derivation was assessed by comparing the impact of data before and after feature derivation on the performance of the prediction model. Specifically, the performance of the prediction models before and after feature derivation was compared using Random Forest, SVM, and XGBoost algorithms. For all models, both the prediction error (MAE) and the explanatory power (R²) were improved after using derivation features. This demonstrates that constructing features with process and statistical significance from the original process and composition data is key to improving the success of model prediction tasks. Table 1 compares the model performance before and after using derivation features.

[0053]

[0054] Table 1

[0055] The beneficial effects of the present invention are described in detail below.

[0056] This invention achieves efficient expression and analysis of clinker chemical composition data through systematic and multi-dimensional feature construction, thereby enhancing the usability of data in intensity prediction, process optimization and intelligent control.

[0057] This invention enhances the interpretability of the model, breaks through the limitations of traditional single-component models, and deeply mines production process information through multi-dimensional feature derivation such as dynamic statistics of mineral phases. This not only significantly improves the interpretability and effectiveness of features, but also solves the problems of traditional single features and lack of process correlation.

[0058] This invention improves the predictive performance of the model by innovatively adopting a time-series numerical coding strategy to replace the traditional date processing method, effectively capturing the environmental cycle pattern. At the same time, it avoids dimensionality explosion through high-level feature aggregation, accurately quantifies the kiln condition stability and the influence of the external environment, and significantly improves the model accuracy.

[0059] This invention is versatile, providing a systematic and generalizable process for deriving the characteristics of clinker components. The method is flexible, scalable, and adaptable to different plants and process routes.

[0060] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution; or the direct application of the inventive concept and technical solution to other situations without modification, are all within the protection scope of the present invention.

Claims

1. A method for characterizing clinker components, characterized in that: Includes the following steps: Step 1: Obtain relevant data on clinker; Step 2: Perform data preprocessing on relevant clinker data; Step 3: Construct a derived feature dataset based on the preprocessed data; Step 4: Merge the derived feature dataset with the original data to output the enhanced feature dataset.

2. The method for characterizing clinker components as described in claim 1, characterized in that: In step one, the time-series data of process parameters on the cement production line are obtained through the quality control system, quality management system, and calcination system, and the measured values ​​of clinker strength for the corresponding batch are obtained through the clinker strength application system.

3. The method for characterizing clinker components as described in claim 1, characterized in that: In step one, the relevant data of clinker are aligned and integrated based on the company name, sampling location on the production line, and date information.

4. The method for characterizing clinker components as described in claim 1, characterized in that: In step two, outliers in the data are filtered out using the z-score method.

5. The method for characterizing clinker components as described in claim 1, characterized in that: In step three, a dynamic statistical feature dataset of mineral phases based on daily details and a derived feature dataset based on time series are constructed for the data.

6. The method for characterizing clinker components as described in claim 5, characterized in that: The method for constructing dynamic statistical features of mineral phases based on daily details is as follows: the data are grouped according to the company, sampling location on the production line, and date dimension. Various statistical calculations are performed on the daily detailed data of mineral phases, rate values, clinker free calcium, and liter weight. Each feature is derived into seven types of derived features: maximum value, minimum value, mean, standard deviation, median, difference ratio, and summation.

7. The method for characterizing clinker components as described in claim 5, characterized in that: One method for constructing time-series-based derived features is to quantify the timestamps corresponding to the data, introduce the day_diff feature, and record the difference in the number of months, weeks, and days from the base date.

8. The method for characterizing clinker components as described in claim 1, characterized in that: In step four, the daily detailed mineral phase dynamic statistical feature dataset and the time-series derived feature dataset are merged with the unprocessed raw data by company name, sampling location on the production line, and date, and an enhanced feature dataset is output.

9. The method for characterizing clinker components as described in claim 1, characterized in that: The effectiveness of feature derivation is determined by comparing the impact of data before and after feature derivation on the performance of the prediction model.

Images