METHOD AND APPARATUS FOR PREDICTING HEAT CONSUMPTION OF CLINKER SINTERING

Abstract

The present disclosure relates to a method and apparatus for predicting the heat consumption of clinker sintering based on a LightGBM model, and relates to the technical field of predicting the heat consumption of clinker sintering. The main objectives of the present disclosure are to improve the prediction accuracy and detection efficiency of the heat consumption of clinker sintering, and to reduce memory usage, so that the method and apparatus can be better practically applied in factories.The main technical solutions of the present disclosure include: obtaining corresponding target production data based on preset process parameters, wherein the preset process parameters are process parameters meeting preset requirements, selected from process parameters related to the heat consumption of clinker sintering; and outputting a prediction result of the heat consumption of clinker sintering by means of the preset LightGBM model, based on the target production data. The method and apparatus of the present disclosure are used for predicting the heat consumption of clinker sintering. Figure 1.

Description

Title of the invention: METHOD AND APPARATUS FOR PREDICTING HEAT CONSUMPTION OF CLINKER SINTERING Technical field

[0001] The present disclosure relates to the technical field of predicting the heat consumption of clinker sintering, and more particularly to a method and apparatus for predicting the heat consumption of clinker sintering based on a LightGBM model. Background

[0002] Cement clinker is a semi-finished cement product obtained by sintering raw materials at high temperature and then cooling the sintered raw materials, the raw materials being configured in appropriate proportions, using limestone, sandstone, aluminum and iron raw materials, etc. as the main raw materials. Being a complex thermodynamic and chemical system with high energy intensity, clinker sintering is the most important technological link in the cement production process, and its energy consumption has a direct impact on the energy utilization rate and economic efficiency of a whole cement production line.As an important indicator for measuring energy consumption in cement production, clinker sintering heat consumption refers to the actual calcination heat consumption per 1 kg of clinker in a calciner. Effective prediction of clinker sintering heat consumption can optimize the production process, improve production efficiency and reduce energy consumption, which is of great significance for ensuring national energy security.

[0003] Currently, in a method for predicting the heat consumption of clinker sintering, a coal consumption detection method and system for cement sintering based on a random forest model are established by using process parameters such as the feed rate, the secondary air temperature, and the temperature in the tail of the calcining kiln in combination with a random forest algorithm. Alternatively, the index prediction of the coal consumption during the cement sintering process is established by using process parameters such as the raw material feed rate, the coal feed rate of the decomposition kiln, and the secondary air temperature in combination with a time series deep belief network. However, the method for predicting the heat consumption of clinker sintering above still has shortcomings in prediction accuracy, computational efficiency and model memory usage, which is not conducive to practical application in factories. Summary

[0004] In view of the above-mentioned problems, the present disclosure provides a method and apparatus for predicting the heat consumption of clinker sintering based on a LightGBM model. The main objectives of the present disclosure are to improve the prediction accuracy and detection efficiency of the heat consumption of clinker sintering, and to reduce memory usage, so that the method and apparatus can be better practically applied in factories.

[0005] To solve the above-mentioned technical problems, the present disclosure provides the following solutions.

[0006] In a first aspect, the present disclosure provides a method for predicting the heat consumption of clinker sintering based on a LightGBM model, which comprises:

[0007] obtaining corresponding target production data based on predefined process parameters, the predefined process parameters being process parameters meeting predefined requirements, selected from process parameters related to the heat consumption of clinker sintering; and

[0008] outputting a prediction result of the heat consumption of clinker sintering by means of the predefined LightGBM model, based on the target production data.

[0009] Optionally, the method further comprises:

[0010] obtaining corresponding historical production data based on predefined process parameters, wherein a marking time of the historical production data is a corresponding initial production time;

[0011] dividing the historical production data into a training set and a test set according to a predefined ratio; and

[0012] obtaining the LightGBM model meeting prediction performance requirements based on the training set and the test set, the LightGBM model being configured to predict the heat consumption of clinker sintering.

[0013] Optionally, obtaining the corresponding historical production data based on predefined process parameters comprises:

[0014] defining a time series data input window based on a time difference between the initial production time and the generation time of the historical production data; and

[0015] obtaining historical production data corresponding to the predefined process parameters by means of the time series data input window.

[0016] Optionally, obtaining the LightGBM model meeting prediction performance requirements based on the training set and the test set comprises:

[0017] optimizing hyperparameters of the LightGBM model using a Bayesian optimization algorithm based on the training set, to obtain a trained LightGBM model;

[0018] evaluating whether the prediction performance of the trained LightGBM model meets the prediction performance requirements based on the test set; and

[0019] determining that the LightGBM model is configured to predict the heat consumption of clinker sintering when the prediction performance of the trained LightGBM model meets the prediction performance requirements.

[0020] Optionally, obtaining the corresponding target production data based on predefined process parameters comprises:

[0021] collecting original production data corresponding to predefined process parameters;

[0022] performing outlier processing on the original production data based on the Hampel filter; and

[0023] transforming the original data subjected to outlier processing into non-dimensional data with a mean of 0 and a variance of 1, based on z-score normalization, in order to obtain the target production data corresponding to the predefined process parameters.

[0024] Optionally, the method further comprises:

[0025] obtaining process parameters related to the heat consumption of clinker sintering, in order to form a set of process parameters; and

[0026] selecting the predefined process parameters by means of a maximum information coefficient correlation analysis method based on the set of process parameters.

[0027] Optionally, selecting the predefined process parameters using a maximum information coefficient correlation analysis method based on the set of process parameters comprises:

[0028] calculating a maximum information coefficient between each of the process parameters of the process parameter set and the heat consumption of clinker sintering;

[0029] ranking the process parameters of the process parameter set from largest to smallest according to the maximum information coefficient, in order to obtain a ranking result;

[0030] calculating a coefficient difference between the maximum information coefficients corresponding to two adjacent parameters among the process parameters in the ranking result; and

[0031] selecting the predefined process parameters from the process parameter set based on the coefficient difference.

[0032] In a second aspect, the present disclosure provides an apparatus for predicting the heat consumption of clinker sintering based on a LightGBM model, comprising:

[0033] a first obtaining unit configured to obtain corresponding target production data on the basis of predefined process parameters, the predefined process parameters being process parameters meeting predefined requirements, selected from process parameters related to the heat consumption of clinker sintering; and

[0034] a prediction unit configured to output a clinker sintering heat consumption prediction result using the predefined LightGBM model, based on the target production data.

[0035] To achieve the above objectives, according to a third aspect of the present disclosure, there is provided a storage medium having a program stored therein. When the program is being executed, a device on which the storage medium is installed is controlled to execute the method for predicting the heat consumption of clinker sintering based on the LightGBM model as mentioned in the first aspect.

[0036] To achieve the above objectives, according to a fourth aspect of the present disclosure, there is provided an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor. When the processor executes the program, all or part of the steps of predicting by the apparatus the heat consumption of clinker sintering based on the LightGBM model as mentioned in the second aspect are implemented.

[0037] By virtue of the above technical solutions, the present disclosure provides the method and apparatus for predicting the heat consumption of clinker sintering based on the LightGBM model. Indeed, the existing methods for predicting the heat consumption of clinker sintering have shortcomings in prediction accuracy, computational efficiency and model memory utilization. Therefore, according to the present disclosure, corresponding target production data is obtained based on predefined process parameters, the predefined process parameters being process parameters meeting predefined requirements, selected from process parameters related to the heat consumption of clinker sintering. A prediction result of the heat consumption of clinker sintering is output by means of the predefined LightGBM model, based on the target production data.The method and apparatus for predicting the heat consumption of clinker sintering provided by the present disclosure improve the prediction accuracy and detection efficiency of the heat consumption of clinker sintering, while using little memory, which is beneficial for practical application in factories.

[0038] The above description is only an overview of the technical solutions of the present disclosure. In order that the technical means of the present disclosure may be more clearly understood and implemented in accordance with the contents of the specification, and in order that the above and other objectives, features, and advantages of the present disclosure may be more easily understood, specific embodiments of the present disclosure are provided below. Brief description of the drawings

[0039] Upon reading the detailed description of the following preferred embodiments, various other advantages and benefits will become apparent to a person of ordinary skill in the art. The accompanying drawings are for the sole purpose of illustrating the preferred embodiments and are not to be construed as limiting the present disclosure. Further, in the drawings, like elements are indicated by like reference numerals. In the drawings:

[0040] [Fig. 1] shows a flowchart of a method for predicting the heat consumption of clinker sintering based on a LightGBM model according to one embodiment of the present disclosure;

[0041] [Fig.2] shows a flowchart of another method for predicting the heat consumption of clinker sintering based on a LightGBM model according to an embodiment of the present disclosure;

[0042] [Fig.3] shows a flowchart of a third method for predicting the heat consumption of clinker sintering based on a LightGBM model according to an embodiment of the present disclosure;

[0043] [Fig.4] shows a block diagram of an apparatus for predicting the heat consumption of clinker sintering based on a LightGBM model according to an embodiment of the present disclosure; and

[0044] [Fig.5] shows a block diagram of another apparatus for predicting the heat consumption of clinker sintering based on a LightGBM model according to an embodiment of the present disclosure. Detailed description

[0045] Exemplary embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are illustrated in the accompanying drawings, it should be understood that the present disclosure may be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure may be understood in a complete and comprehensive manner and may fully convey the scope of the present disclosure to those skilled in the art.

[0046] Existing methods for predicting the heat consumption of clinker sintering still have shortcomings in prediction accuracy, computational efficiency and model memory utilization. In response to this problem, the inventors came up with the idea of ​​making predictions using optimized process parameters in combination with a LightGBM model.

[0047] Therefore, an embodiment of the present disclosure provides a method for predicting the heat consumption of clinker sintering based on the LightGBM model, which improves the prediction accuracy and detection efficiency of the heat consumption of clinker sintering. As shown in [Fig.l], the method comprises the following specific execution steps.

[0048] In step 101, corresponding target production data is obtained based on predefined process parameters.

[0049] The predefined process parameters are process parameters meeting predefined requirements, selected from process parameters related to the heat consumption of clinker sintering. The process parameters are process parameters related to the heat consumption of clinker sintering, which are selected from four links in a clinker sintering system comprising a preheater, a decomposition kiln, a rotary calciner and a grate cooler. For example, the process parameters may comprise 35 process parameters in total: process parameters of the preheater (comprising the outlet temperature of a cylinder C1, the pressure at cone of cylinder C1, the outlet temperature of a cylinder C2, the cone pressure of cylinder C2, the outlet temperature of a cylinder C3, the cone pressure of cylinder C3, the outlet temperature of a cylinder C4, the cone pressure of cylinder C4, the discharge pipe temperature of cylinder C4, the electric current of a high-temperature fan, the rotation speed of the high-temperature fan and the inlet negative pressure of the high-temperature fan), process parameters of the decomposition furnace (including the outlet temperature of the decomposition furnace, the outlet negative pressure of the decomposition furnace, the coal feed at the tail of the calcining furnace, the temperature of a flue chamber, the pressure of the flue chamber and the temperature of the tertiary air entering a furnace),rotary calciner process parameters (including coal feed at the calciner head, feed rate, decomposition rate of input materials, calciner main motor current, calciner rotation speed, calciner head negative pressure, calciner tail negative pressure, calciner head fan current, calciner head fan speed, electric current of a primary fan, air pressure of the primary fan, and secondary air temperature) and grate cooler process parameters (including average pressure under a grate, average speed of a grate plate, grate cooler oil pressure, tertiary air temperature, and tertiary air pressure). However, this embodiment is not limited thereto.

[0050] More specifically, in this embodiment, in analyzing the production processes of cement clinker, after selecting several process parameters related to the heat consumption of cement clinker sintering from the process parameters of devices such as the preheater, the decomposition kiln, the rotary calciner, and the grate cooler, a maximum information coefficient (MIC) between the selected process parameters and the heat consumption of cement clinker sintering is calculated by means of MIC correlation analysis. The higher the MIC, the stronger the correlation between the process parameters and the heat consumption of cement clinker sintering.Therefore, in this embodiment, a plurality of target process parameters having a higher MIC can be selected according to the MIC from among the plurality of process parameters as preset process parameters. Then, corresponding historical production data is obtained from the historical data according to the preset process parameters.

[0051] In step 102, a clinker sintering heat consumption prediction result is output by means of the predefined LightGBM model, based on the target production data.

[0052] After obtaining the target production data in step 101, the target production data is input into a pre-trained LightGBM model to obtain the prediction result of the heat consumption of clinker sintering.

[0053] Based on a gradient boosting decision tree (GBDT) algorithm and using two new technologies, namely gradient-based one-sided sampling (GOSS) and exclusive feature pooling (EFB), LightGBM reduces memory consumption and improves computational speed while ensuring high accuracy, making it more suitable for machine learning tasks with high-dimensional features or big data.

[0054] More specifically, with GOSS, data are first classified according to absolute values ​​of gradients, and ax first 100% instances are selected as large gradient data instances, and then bx (1-a) x 100% data instances are randomly extracted from the remaining data as small gradient data. With GOSS, when calculating information gain, the sampled small gradient data are amplified by (la) / b times, which ensures that more attention is paid to undertrained instances when the original data distribution is hardly changed.

[0055] It is assumed that there are n sample instances which are denoted by {xb ... xn}, where x; represents an s-dimensional vector. In each gradient boosting iteration, a negative gradient of a loss function output by the model is denoted by {gi, ... gn}. The formula for calculating the information gain of a feature j separated at a point d is: where Ai = {xi t A : xÿ < d}, Ar = { x; s A : xÿ > d}, Bi = { xi s B : xÿ < d}, { Xi GB: Xÿ > d}. Therefore, with GOSS, a separation point is determined at using the estimate of Vj(d) Sooner than precise data y which allows to significantly reduce computing costs.

[0056] With EFB, a goal of dimensionality reduction is achieved by grouping exclusive features into a single feature. In practical applications, many high-dimensional features are poorly numerous and mutually exclusive. Exclusive features are grouped into a single feature, and then a feature histogram as for the single feature is constructed from a grouping of features. In this way, the construction complexity can be reduced from O(#datax#feature) to O(#datax#bundle), and #bundle « #feature. In this way, the training speed of the algorithm can be significantly accelerated without compromising accuracy.

[0057] Thanks to GOSS and EFB, the LightGBM model increases the computational speed and reduces the memory consumption while ensuring high accuracy, and is therefore suitable for handling prediction problems with a large volume of data and high-dimensional features.

[0058] As can be seen in the implementation mode based on the embodiment in [Fig.l], the present disclosure provides a method for predicting the heat consumption of clinker sintering based on a LightGBM model. According to the method of the present disclosure, corresponding target production data are obtained based on preset process parameters, and then the preset process parameters are input into the LightGBM model to obtain the prediction result of the heat consumption of clinker cement sintering, which is output by the LightGBM model. The method of the present disclosure can improve the prediction accuracy and detection efficiency of the heat consumption of clinker sintering while using less memory, making it advantageous for practical application in factories.

[0059] For better selection of predefined process parameters and for improving prediction accuracy and efficiency, and further for improving and extending the embodiment shown in [Fig.l], an embodiment of the present disclosure also provides another method for predicting the heat consumption of clinker sintering based on the LightGBM model. As shown in [Fig.2], this method comprises the following steps.

[0060] In step 201, process parameters related to the heat consumption of clinker sintering are obtained to form a process parameter set.

[0061] For example, the process parameters related to the heat consumption of clinker sintering in step 101 are not described here. It should be emphasized that the set of process parameters may also include the heat consumption of clinker sintering (QrR), which is obtained using the following formula: QrR — ^-rQar,net

[0062] where QrR represents the sintering heat consumption per unit of clinker expressed in units of KJ / Kg; mr represents the quantity of coal feed per unit of clinker expressed in units of KJ / Kg; and Q.„ >net represents the lower calorific value of the coal expressed in units of KJ / Kg; Mfr + Myr mr = ~7r,—— (2)

[0063] where Mfr represents the quantity of coal feed to the decomposition kiln expressed in units of t / h; Myr represents the quantity of coal feed at the head of the calcining kiln expressed in units of t / h; Ms represents a feed rate expressed in units of t / h; and ms represents the dose of raw materials required for production per unit of clinker expressed in units of Kg / Kg.

[0064] In step 202, the predefined process parameters are selected using a maximum information coefficient correlation analysis method based on the set of process parameters.

[0065] After obtaining the process parameter set in step 201, a maximum information coefficient between each of the process parameters in the process parameter set and the heat consumption of clinker sintering is calculated. Then, the process parameters in the process parameter set are ranked from largest to smallest according to the maximum information coefficient, to obtain a ranking result. Then, a coefficient difference between the maximum information coefficients corresponding to two adjacent parameters among the process parameters in the ranking result is calculated. Finally, the predefined process parameters are selected from the process parameter set based on the coefficient difference. In this way, it is advantageous to improve the accuracy of the prediction result.

[0066] The maximum information coefficient is calculated according to the following steps.

[0067] Based on the following equation, mutual information I[X;Y] between process parameter X and heat consumption Y of clinker cement sintering is first calculated: / [x;r] = yp(x,y)log2 pfcy) pWp(y) (3)

[0068] where p(x,y) represents a joint probability between process parameter X and heat consumption Y of clinker cement sintering, p(x) represents an edge probability of process parameter X, and p(y) represents an edge probability of process parameter Y.

[0069] Then, according to the mutual information I[X;Y] between the process parameter X and the heat consumption Y of clinker cement sintering, the maximum information coefficient M1C[X;Y] between the process parameter X and the heat consumption Y of clinker cement sintering is calculated based on the following formula: / (X; Y) MIC[X;Y\ = max ---z / --— |x||r| <R log2(mln(x, / )) 0)

[0070] It should be noted that a feature can be obtained by means of other features. Therefore, in general, the larger the information amount of the input features of a prediction model, the higher the accuracy of the prediction result. However, the accuracy of the prediction result may be compromised when too many features weakly correlated with the heat consumption of clinker sintering are input into the prediction model. Therefore, in this step, after ranking the process parameters of the process parameter set from largest to smallest according to the maximum information coefficient, a fluctuation amplitude between several coefficient differences is determined based on the coefficient differences between the maximum information coefficients corresponding to the two adjacent process parameters.

[0071] When the fluctuation amplitude is constant, the coefficient differences between the maximum information coefficients corresponding to the two adjacent process parameters are both less than a predefined coefficient difference, i.e., the maximum information coefficients of the process parameters show a constant decreasing trend, the maximum information coefficients corresponding to the many process parameters show little difference. In this case, the first N target process parameters are selected from the process parameter set according to the ranking result as predefined process parameters.

[0072] When the fluctuation amplitude is not constant, the coefficient differences between the maximum information coefficients from the first-ranked process coefficient to the nth-ranked process coefficient are smaller than the predefined coefficient difference, but the coefficient difference between the nth-ranked process coefficient and the (n+1)th-ranked process coefficient is larger than the predefined coefficient difference, i.e., the differences between the maximum information coefficients corresponding to the process coefficients ranked after the nth and the maximum information coefficients corresponding to the process coefficients ranked before the nth are larger.Therefore, to avoid too many features weakly correlated with clinker sintering heat consumption being entered into the prediction model, in this step, it is determined whether n is greater than N. If n is . greater than N, the first N target process parameters are selected from the process parameter set according to the ranking result as predefined process parameters. If n is not greater than N, it is determined whether the first N process parameters include the process parameters in the process mechanism parameters. For example, in the four links, namely the preheater, the decomposition furnace, the rotary calciner and the grate cooler, if the first N process parameters include the process parameters in the process mechanism parameters, the first N target process parameters are selected from the process parameter set according to the ranking result as predefined process parameters.If the first N process parameters do not include the process parameters in the process mechanism parameters, based on a principle that the selected target process parameters must include at least one process parameter of the process mechanism parameters of each category (such as preheater, decomposition furnace, rotary calciner, and grate cooler), the first N target process parameters are first selected as preset process parameters, and then the (Nn) target process parameters are selected from the process parameters ranked after the nth according to the ranking result, as preset process parameters.

[0073] In step 203, corresponding target production data is obtained based on the predefined process parameters.

[0074] This step is described with reference to the description of step 101 in the above method, and identical content elements are not repeated here.

[0075] After the predefined process parameters are obtained in step 202, original production data corresponding to the predefined process parameters are collected. Then, outlier processing is performed on the original production data, based on the Hampel filter. Then, the original data subjected to the outlier processing are transformed into non-dimensional data with a mean of 0 and a variance of 1, based on z-score normalization, to obtain the target production data corresponding to the predefined process parameters. In this way, it is advantageous to improve the accuracy of the clinker sintering heat consumption prediction result outputted by the prediction model.For example, for the maximum information coefficients, the first 15 process parameters are selected from the 35 process parameters in step 101 as predefined process parameters: the coal feed at the head of the calciner, the coal feed at the tail of the calciner, . the rotational speed of the high-temperature fan, the feed rate, the average speed of the grate cooler, the outlet temperature of the C1 cylinder, the negative cone pressure of the C1 cylinder, the tertiary air temperature, the average pressure under the grate, the secondary air temperature, the electric current of the high-temperature fan, the outlet temperature of the decomposition furnace, the outlet temperature of the C4 cylinder, the air pressure of the primary fan, and the negative cone pressure of the C4 cylinder.

[0076] Outlier processing is performed on the original production data based on the Hampel filter. The specific methods are as follows.

[0077] For a data sequence xb x2, x3, ... xn, assuming that a sample to be tested is xb, that a length of a sliding window is set to 2s+l, and that a number of samples around the sample to be tested is s, a detection process of the sample to be tested is expressed by the following equation: (xit |x; - m;| < 3St xi = 1 W |Xj-mj >3St (5) mi = median(x^s, x-,_s+1, - , xt, ■■■, xi+s_u xi+s) (6) where m / represents a median in the window, and S ; represents a scale estimate of an absolute deviation from the median in the window.

[0078] When an absolute value between the sample to be tested xt and a window median is greater than three times S;, and sample is deleted, and the vacant place is occupied by a local median of the sliding window.

[0079] Furthermore, the original production data collected by each sensor have different engineering units whose values ​​may even differ by several orders of magnitude. The direct use of these data may weaken the roles of variables with smaller values ​​in the model and highlight the roles of variables with larger values. Therefore, it is necessary to normalize the original production data in order to eliminate the negative effects of different dimensions on subsequent stages. The original production data subjected to outlier processing are normalized in order to eliminate the negative effects of the dimensions.The original data are transformed into non-dimensional data with mean 0 and variance 1, based on z-score normalization, to obtain the target production data corresponding to the predefined process parameters, where a transformation formula is presented in Equation (8): . Xt - fl zi =------ a (8)

[0080] where fi represents the mean of the data sequence, and o represents the standard deviation of the data sequence.

[0081] In step 204, a clinker sintering heat consumption prediction result is output by means of the predefined LightGBM model, based on the target production data.

[0082] This step is described with reference to the description of step 102 in the above method, and identical content elements are not repeated here.

[0083] As can be seen in the implementation mode based on the embodiment in [Fig.l], the present disclosure provides a method for predicting the heat consumption of clinker sintering based on a LightGBM model that solves a problem of strong coupling and hysteresis relationship between multivariables. More specifically, in this method, a maximum information coefficient between each of the process parameters in the process parameter set and the heat consumption of clinker sintering is determined. The process parameters in the process parameter set are ranked from largest to smallest according to the maximum information coefficient, to obtain a ranking result. Then, based on the ranking result, the target process parameters are selected from the process parameter set as predefined process parameters to obtain the target production data.Based on maximum information coefficient correlation analysis, target process parameters are selected from several process parameters that affect the heat consumption of clinker cement sintering. In this way, the complex correspondence between clinker sintering heat consumption and factors influencing clinker sintering heat consumption is simplified, the problem of multivariable coupling in the production process is solved, and the generalization of using clinker sintering heat consumption prediction model to predict clinker sintering heat consumption is improved.

[0084] In order to obtain a prediction model with a better prediction effect, a training process is designed. To improve and extend the embodiment shown in [Fig.l], an embodiment of the present disclosure also provides another method for predicting the heat consumption of clinker sintering based on the LightGBM model. As shown in [Fig.3], this method comprises the following steps.

[0085] In step 301, corresponding historical production data is obtained based on the predefined process parameters.

[0086] The marking time of historical production data is a corresponding initial production time.

[0087] A time series data input window is defined based on a time difference between the initial production time and the generation time of the historical production data; and the historical production data corresponding to the predefined process parameters are obtained by means of the time series data input window.

[0088] More specifically, the time series data input window is a sliding time window designed according to process mechanisms for characterizing time series information of clinker production. Varying time series data of a production process under continuous working condition in the clinker sintering process are integrated into a matrix as an input of the prediction model which comprises time-varying delay information in the clinker production process.

[0089] Illustration:

[0090] In cement clinker production, it takes about 50 to 60 minutes for the raw materials to be fully sintered and turned into clinker. Therefore, there is a time delay between the feed rate and the coal feed data and the corresponding heat consumption data. All other variables related to heat consumption also have time delay characteristics, and the delay time of each variable changes dynamically.

[0091] Therefore, the sliding time window, also known as the time series data input window, is constructed. The magnitude of the time window can be set to 60 minutes, and the original data can be divided into a time series with a magnitude of one hour. Finally, time alignment is performed on the data of the heat consumption of clinker cement sintering, so as to obtain the time series of each process parameter corresponding to the heat consumption of clinker cement sintering. The time series is used to extract a time series in the production data of process parameters as an input of the model which can make full use of the collected data.In addition, the time series information of a production stage can be fully input into the model without the need to match the time series of the data, saving the work of calculating the lead time.

[0092] In step 302, the historical production data is divided into a training set and a test set according to a predefined ratio.

[0093] After the historical production data corresponding to the predefined process parameters is obtained in step 301, the historical production data is divided into a training set and a test set according to the predefined ratio. The specific predefined ratio is set according to the activity, and this embodiment does not impose any specific limit.

[0094] In step 303, the LightGBM model meeting the prediction performance requirements is obtained based on the training set and the test set.

[0095] The LightGBM model is configured to predict the heat consumption of clinker sintering. After obtaining the training set and the test set in step 302, the hyperparameters of the LightGBM model are optimized using a Bayesian optimization algorithm based on the training set to obtain a trained LightGBM model. It is evaluated whether the prediction performance of the trained LightGBM model meets the prediction performance requirements based on the test set. It is determined that the LightGBM model is configured to predict the heat consumption of clinker sintering when the prediction performance of the trained LightGBM model meets the prediction performance requirements.

[0096] In one example, the established LightGBM model contains many hyperparameters, and the definition of the hyperparameters has a direct impact on the model performance. Hyperparameter optimization in machine learning aims to find the hyperparameters that enable a machine learning algorithm to achieve the best performance on a validation dataset. As an optimization algorithm based on the Bayesian theorem, the Bayesian optimization algorithm can find a global optimal solution in fewer iterations. The main idea of ​​this algorithm is to find the optimal solution by continuously updating a prior probability distribution.The main reference is made to the performance indicators for the model effects, and the evaluation indicators of regression algorithms mainly include mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and R squared (R2), etc.

[0097] More specifically, in this embodiment, the established LightGBM model contains many hyperparameters. Mainly, eight hyperparameters having a greater impact on the model performance are selected, including: a learning rate (learning_rate), a number of iterations (n_estimators), a maximum tree depth (max_depth), a number of leaf nodes (num_leaves), a minimum number of child node samples (min_child_samples), an L1 regularization coefficient (reg_alpha), an L2 regularization coefficient (reg_lambda), and a column sampling ratio (colsample_bytree). Then, based on the Bayesian optimization algorithm, the hyperparameters of a LightGBM learner are optimized to select an optimal hyperparameter set, where learning_rate is 0.1, n_estimators is 550, max_depth is 12, num_leaves is 12, reg_alpha is 0.21, reg_lambda is 0.35, colsample_bytree is 0.98, and the evaluation metrics used in the model are RMSE, MAE, and R2.

[0098] The Bayesian optimization algorithm is implemented through the following steps.

[0099] Based on the established LightGBM model, the combination of hyperparameters is initialized. In other words, default values ​​are adopted for each hyperparameter of the model. The combination of hyperparameters mainly includes: a learning rate (learning_rate), a number of iterations (n_estimators), a maximum tree depth (max_depth), a number of leaf nodes (num_leaves), a minimum number of child node samples (min_child_samples), an L1 regularization coefficient (reg_alpha), an L2 regularization coefficient (reg_lambda), and a column sampling ratio (colsample_bytree).

[0100] A probabilistic substitution model is established for tree probability estimation (TPE), and a probability distribution definition of the TPE algorithm is shown in equation (9): . . , _ [ / « y < y* pxy W*), y > y* (Q)

[0101] where Z(Gre) represents a density formed by an observed value (xw), its corresponding risk loss value y = f(xw), and y < y*, and g(x) represents the density formed using the remaining observed values ​​except (x(1)).

[0102] With the TPE algorithm, a threshold y* is selected as a certain quantile y of the current observed risk value y, satisfying p(y < y*) = y. By means of l(x) and g(x) of the TPE algorithm, the hyperparameter set is divided into two parts, namely a part with lower risk and a part with higher risk. With the TPE algorithm, a next hyperparameter is selected according to an expected improvement (El) acquisition function, and further optimization is performed by the maximum expected improvement (El), the definition of El being shown in Equation 10: ESy^X) = yyT(x)-ZWfœp(yW / g(x) \ 1 yl(x) + (1 - y)g(x) V + l(x) 7 ) (10)

[0103] To obtain the maximum expected improvement, by evaluating each hyperparameter x at each iteration, the hyperparameter value with maximum El is fed back into the algorithm until the optimal solution x* is obtained:

[0104] . / \(H) x = argmaxEl^xj

[0105] Therefore, in the present disclosure, the Bayesian optimization algorithm is employed to optimize the hyperparameters of the LightGBM learner, which enables the LightGBM model to achieve optimal performance.

[0106] Illustration:

[0107] A typical 5000 t / d clinker production line in a cement plant is used as the research object, the production line being a four-stage preheater. According to the described method, 15 target process parameters are selected, and out of 5000 sample data groups, 3750 data groups are selected as the training set, while the remaining 1250 data groups are used as the test set. These data constitute a data set of the prediction model.A time series data input window is designed according to a cement processing mechanism, the training set of target production data corresponding to the target process parameters is input into the LightGBM model for training, the hyperparameters of the LightGBM learner are optimized by means of the Bayesian optimization algorithm, and then the test set of target production data corresponding to the target process parameters is input into the LightGBM model. Thus, a prediction result of clinker cement sintering heat consumption is obtained, the prediction result is output by a prediction model to predict the heat consumption of clinker cement sintering, to evaluate the prediction performance of the model. It is found that the errors between the predicted values ​​and the actual values ​​are smaller.The comparison results of the prediction performance between the LightGBM model and a BO-LightGBM model subjected to Bayesian optimization are shown in Table 1. Table 1 Comparison of prediction model performance Model RMSE MAE R2 LightGBM 29.5720 18.8352 0.9911 BO-LightGBM 26.2002 16.2728 0.9930

[0108] Further, in another preferred embodiment of the present disclosure, the LightGBM may also use a histogram algorithm to process feature data. Continuous feature data is discretized into k integers as abscissas to establish a diagram. Then, when The data is input into the model, a histogram is established based on the k integers to count the number of occurrences, and an optimal feature segmentation point is determined based on the feature statistics. Thus, the model features only need to be counted once, avoiding the repetitive feature calculation in traditional machine learning algorithms. While ensuring high accuracy, memory consumption is reduced and computing speed is improved, making them more suitable for machine learning tasks with high-dimensional features or big data.

[0109] The data set of the selected input feature is divided into a training set and a test set, the training set is input into the LightGBM model for training, and an input / output layer is constructed for the prediction model for predicting the heat consumption of clinker sintering based on the LightGBM algorithm. The LightGBM uses a histogram algorithm to extract features from a time series data matrix selected in an input part, to construct a time-varying delay feature histogram for the cement sintering process. Finally, the prediction model for predicting the heat consumption of clinker sintering based on the LightGBM algorithm is trained.Therefore, in this embodiment, by means of the LightGBM algorithm, a method for predicting the heat consumption of clinker sintering based on the LightGBM model is provided which can better predict the heat consumption of clinker cement sintering and can be better applied in practice in factories.

[0110] As can be seen from the embodiment in [Fig. 1], the present disclosure provides a method for predicting the heat consumption of clinker sintering based on the LightGBM model. An input feature of this model is selected from pre-processed data related to process parameters by means of the maximum information coefficient (MIC) analysis method, and then a data set corresponding to the input feature is obtained and divided into a training set and a test set.Then, a time series data input window is designed according to a cement processing mechanism, the training set of target production data corresponding to the target process parameters is input into the LightGBM model for training, the hyperparameters of the LightGBM learner are optimized by means of the Bayesian optimization algorithm, and then the test set of target production data corresponding to the predefined process parameters is input into the LightGBM model. In this way, a consumption prediction result . heat consumption of clinker cement sintering is obtained, the prediction result is output by the prediction model to predict the heat consumption of clinker cement sintering. The prediction accuracy and detection efficiency of clinker sintering heat consumption can thus be improved.

[0111] Furthermore, as an implementation of the method shown in [Fig. 1], an embodiment of the present disclosure also provides a clinker sintering heat consumption prediction apparatus based on the LightGBM model to implement the method shown in [Fig. 1]. The embodiment of the apparatus corresponds to the above embodiment, and for ease of reading, the detailed content items of the above embodiment are not unnecessarily described one by one in the embodiment of the apparatus. However, it is clear that the apparatus in this embodiment can implement all the content items correspondingly in the above embodiment. As shown in [Fig.4], the apparatus comprises:

[0112] A first obtaining unit 31 configured to obtain corresponding target production data on the basis of predefined process parameters, the predefined process parameters being process parameters meeting predefined requirements, selected from process parameters related to the heat consumption of clinker sintering; and

[0113] A prediction unit 32 configured to output, on the basis of the target production data obtained from the first obtaining unit 31, a prediction result of the heat consumption of clinker sintering by means of the predefined LightGBM model.

[0114] Furthermore, as an implementation of the method shown in [Fig.2] and 3, an embodiment of the present disclosure also provides another apparatus for predicting the heat consumption of clinker sintering based on the LightGBM model to implement the method shown in [Fig.2] and 3. The embodiment of the apparatus corresponds to the above embodiment, and for ease of reading, the detailed content items of the above embodiment are not unnecessarily described one by one in the embodiment of the apparatus. However, it is clear that the apparatus in this embodiment can implement all the content items correspondingly in the above embodiment. As shown in [Fig.4], the apparatus comprises:

[0115] A first obtaining unit 31 configured to obtain corresponding target production data on the basis of predefined process parameters obtained by a selection unit 34, the predefined process parameters being process parameters meeting predefined requirements, selected from process parameters related to the heat consumption of clinker sintering;

[0116] A prediction unit 32 configured to output, on the basis of the target production data obtained from the first obtaining unit 31, a prediction result of the heat consumption of clinker sintering by means of the predefined LightGBM model obtained from the driving unit 37;

[0117] A second obtaining unit 33 configured to obtain process parameters related to the heat consumption of clinker sintering, in order to form a set of process parameters;

[0118] A selection unit 34 configured to select the predefined process parameters by means of a maximum information coefficient correlation analysis method on the basis of the process parameter set obtained from the second obtaining unit 33;

[0119] A third obtaining unit 35 configured to obtain corresponding historical production data based on predefined process parameters obtained from the selection unit 34, the marking time of the historical production data being a corresponding initial production time;

[0120] A division unit 36 ​​configured to divide the historical production data obtained from the third obtaining unit 35 into a training set and a test set according to a predefined ratio; and

[0121] A training unit 37 configured to obtain the LightGBM model meeting prediction performance requirements based on the training set and the test set obtained from the dividing unit 36, the LightGBM model being configured to predict the heat consumption of clinker sintering.

[0122] Furthermore, the third obtaining unit 35 comprises:

[0123] An adjustment module 351 configured to define a time series data input window based on a time difference between the initial production time and the generation time of the historical production data; and

[0124] An obtaining module 352 configured to obtain the historical production data corresponding to the predefined process parameters by means of the time series data input window obtained from the adjustment module 351.

[0125] Furthermore, the drive unit 37 comprises:

[0126] A training module 371 configured to optimize the hyperparameters of the LightGBM model using a Bayesian optimization algorithm based on the training set, to obtain a trained LightGBM model;

[0127] An evaluation module 372 configured to evaluate, based on the test set, whether the prediction performance of the trained LightGBM model obtained from the training module 371 meets the prediction performance requirements; and

[0128] A determination module 373 configured to determine, when the prediction performance of the trained LightGBM model obtained from the evaluation module 372 meets the prediction performance requirements, that the LightGBM model is configured to predict the heat consumption of clinker sintering.

[0129] Furthermore, the first obtaining unit 31 comprises:

[0130] A collection module 311 configured to collect original production data corresponding to the predefined process parameters;

[0131] A processing module 312 configured to perform outlier processing based on the Hampel filter on the original production data obtained from the obtaining module 311; and

[0132] An obtaining module 313 configured to transform, based on z-score normalization, the original data subjected to outlier processing obtained from the processing module 312 into non-dimensional data with a mean of 0 and a variance of 1, in order to obtain the target production data corresponding to the predefined process parameters.

[0133] In addition, the selection unit 34 comprises:

[0134] A first calculation module 341 configured to calculate a maximum information coefficient between each of the process parameters of the set of process parameters and the heat consumption of the clinker sintering:

[0135] A ranking module 342 configured to rank the process parameters of the process parameter set from largest to smallest according to the maximum information coefficient obtained from the first calculation module 341, to obtain a ranking result;

[0136] A second calculation module 343 configured to calculate a coefficient difference between the maximum information coefficients corresponding to two adjacent parameters among the process parameters in the classification result obtained from the classification module 342; and

[0137] A selection module 344 configured to select the target process parameter from the process parameter set based on the coefficient difference obtained from the second calculation module 343.

[0138] Further, an embodiment of the present disclosure also provides a processor for executing a program. The method for predicting the heat consumption of clinker sintering based on the LightGBM model in [Fig.l] to 3 is performed when the program is running.

[0139] Further, an embodiment of the present disclosure also provides a storage medium for storing computer programs. When the computer programs are being executed, a device on which the storage medium is installed is controlled to execute the prediction method. of the heat consumption of clinker sintering based on the LightGBM model in [Fig.l] to 3.

[0140] Among the aforementioned embodiments, the description of the various embodiments may be focused differently, and a part not explained in a certain embodiment may refer to a related description of other embodiments.

[0141] It is understood that cross-references may be made to the relevant features of the above method and apparatus. Furthermore, "first", "second" and so on in the above embodiments are used to differentiate the embodiments but do not represent the superiorities and inferiorities of the embodiments.

[0142] It will be clearly understood by those skilled in the art that for a practical and concise description, a concrete working process of the systems, apparatuses and units described above may refer to a corresponding process of the above embodiments, which is no longer repeated in this document.

[0143] The algorithm and display provided herein are not inherently tied to any particular computer, virtual system, or other devices. Various general systems may also be used with the teaching based on the present disclosure. From the above description, the structure necessary to construct such a system is obvious. Furthermore, the present disclosure does not relate to any particular programming language. It should be understood that various programming languages ​​may be used to implement the disclosed content as described herein, and that the above description of the particular programming language is intended to disclose the best inventive mode of implementation.

[0144] Further, the memory may comprise volatile memory on a computer readable medium, random access memory (RAM) and / or non-volatile memory such as read only memory (ROM) or flash memory (flash RAM), and the memory comprises at least one storage chip.

[0145] Those skilled in the art will be aware that embodiments of the present disclosure may be provided in the form of a method, a system, or a computer program product. Accordingly, the present disclosure may utilize forms of a full hardware embodiment, a full software embodiment, or an embodiment combining software and hardware aspects. Further, the present disclosure may utilize forms of computer program products implemented on one or more computer storage media (including, but not limited to, disk memory magnetic, CD-ROM, optical memory or the like) that carries computer program code.

[0146] The present disclosure is described with reference to flowcharts and / or block diagrams of the method in the embodiments of the present disclosure, the device (system) and the computer program product. It is understood that each flow and / or block of the flowchart and / or block diagram as well as each combination of flows and / or blocks of the flowchart and / or block diagram can be implemented by computer program instructions.These computer program instructions may be provided for a general purpose computer, a special purpose computer, an embedded processor, or processors of other programmable data processing devices to generate a machine, to generate an apparatus configured to implement designated functions in one or more flows of a flowchart and / or one or more blocks of a block diagram by means of instructions executed by a computer or a processor of other programmable data processing devices.

[0147] These computer program instructions may also be stored in computer-readable memory that can guide a computer or other programmable data processing devices to operate in a particular manner, such that the instructions stored in the computer-readable memory generate a manufactured product having an instruction device that implements the designated functions in one or more flows of a flowchart and / or one or more blocks of a block diagram.

[0148] These computer program instructions may also be loaded onto a computer or other programmable data processing devices, to execute a series of operational steps on the computer or other programmable devices to generate computer-implemented processes, such that the instructions executed on the computer or other programmable devices provide steps configured to implement designated functions in one or more flows of a flowchart and / or one or more blocks of a block diagram.

[0149] In a typical configuration, the computing device includes one or more CPUs, input-output interfaces, network interfaces, and memories.

[0150] The memory may include volatile memory in a computer-readable medium, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0151] The computer-readable medium comprises a non-volatile medium, a volatile medium, a movable medium, or a stationary medium, which may implement the storing information by any method or technology. The information may be a computer-readable instruction, a data structure, a module of a program, or other data. Examples of computer storage media include, but are not limited to, phase change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical memory, a magnetic tape cartridge, magnetic tape or magnetic disk memory, or other magnetic storage devices, or any other non-transferable media, which may be configured to store information that can be accessed by a computing device.As defined herein, computer-readable media does not include transient media, e.g., modulated data signals and carriers.

[0152] Furthermore, terms such as "include", "comprise" or other variations thereof are intended to cover a non-exclusive "include" such that a process, method, commodity or device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent in the process, method, commodity or device. In the absence of other restrictions, elements limited by a phrase "include a ..." do not exclude the fact that other identical elements may exist in a process, method, commodity or device of those elements.

[0153] Those skilled in the art will be aware that embodiments of the present disclosure may be provided in the form of a method, a system, or a computer program product. Accordingly, the present disclosure may utilize forms of a full hardware embodiment, a full software embodiment, or an embodiment combining software and hardware aspects. Further, the present disclosure may utilize forms of computer program products implemented on one or more computer storage media (including, but not limited to, magnetic disk memory, CD-ROM, optical memory, or the like) that includes computer program code.

[0154] The embodiments described above are merely illustrative embodiments of the present disclosure, and are not intended to limit the present disclosure. For those skilled in the art, various modifications and variations may be made to the present disclosure. All modifications, substitutions equivalents and improvements made in the spirit and principle of this disclosure fall within the scope of protection of the claims of this disclosure.

Claims

Claims

1. A method for predicting the heat consumption of clinker sintering based on a LightGBM model, comprising: obtaining corresponding target production data based on preset process parameters, wherein the preset process parameters are process parameters meeting preset requirements, selected from process parameters related to the heat consumption of clinker sintering; and outputting a prediction result of the heat consumption of clinker sintering by means of the preset LightGBM model, based on the target production data.

2. The method of claim 1, further comprising: obtaining corresponding historical production data based on predefined process parameters, wherein a marking time of the historical production data is a corresponding initial production time; dividing the historical production data into a training set and a test set according to a predefined ratio; and obtaining the LightGBM model meeting prediction performance requirements based on the training set and the test set, the LightGBM model being configured to predict the heat consumption of clinker sintering.

3. The method of claim 2, wherein obtaining the corresponding historical production data based on predefined process parameters comprises: setting a time series data input window based on a time difference between the initial production time and the generation time of the historical production data; and obtaining the historical production data corresponding to the predefined process parameters by means of the time series data input window.

4. The method of claim 3, wherein obtaining the LightGBM model meeting prediction performance requirements based on the training set and the test set comprises: optimizing hyperparameters of the LightGBM model using a Bayesian optimization algorithm based on the training set to obtain a trained LightGBM model; evaluating whether the prediction performance of the trained LightGBM model meets the prediction performance requirements based on the test set; and determining whether the LightGBM model is configured to predict the heat consumption of clinker sintering when the prediction performance of the trained LightGBM model meets the prediction performance requirements.

5. The method according to any one of claims 1 to 4, wherein obtaining the corresponding target production data based on predefined process parameters comprises: collecting original production data corresponding to the predefined process parameters; performing outlier processing on the original production data based on the Hampel filter; and transforming the original data subjected to the outlier processing into non-dimensional data with a mean of 0 and a variance of 1, based on z-score normalization, to obtain the target production data corresponding to the predefined process parameters.

6. The method of claim 5, further comprising: obtaining process parameters related to the heat consumption of clinker sintering, to form a process parameter set; and selecting the predefined process parameters by means of a maximum information coefficient correlation analysis method based on the process parameter set.

7. The method of claim 6, wherein selecting the predefined process parameters by means of a maximum information coefficient correlation analysis method based on the process parameter set comprises: calculating a maximum information coefficient between each of the process parameters of the process parameter set and the heat consumption of clinker sintering; ranking the process parameters of the process parameter set from largest to smallest according to the maximum information coefficient, in order to obtain a ranking result; calculating a coefficient difference between the maximum information coefficients corresponding to two adjacent parameters among the process parameters in the ranking result; and selecting the predefined process parameters from the process parameter set based on the coefficient difference.

8. An apparatus for predicting the heat consumption of clinker sintering based on a LightGBM model, comprising: a first obtaining unit (31) configured to obtain corresponding target production data based on preset process parameters, wherein the preset process parameters are process parameters meeting preset requirements, selected from process parameters related to the heat consumption of clinker sintering; and a prediction unit (32) configured to output a prediction result of the heat consumption of clinker sintering by means of the preset LightGBM model, based on the target production data.

9. A storage medium comprising a stored program, wherein when the program is being executed, a device on which the storage medium is installed is controlled to execute the method for predicting the heat consumption of clinker sintering based on the LightGBM model according to any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, when the processor executes the program, the method for predicting the heat consumption of clinker sintering based on the LightGBM model according to any one of claims 1 to 7 is implemented.

Images