Method for predicting process parameters of rotary cement kiln based on mechanism and data collaboration

Abstract

The invention relates to the field of rotary cement kiln process parameter prediction, and particularly discloses a rotary cement kiln process parameter prediction method based on mechanism and data collaboration, which comprises the following steps: selecting production-related parameters from production data of a rotary cement kiln according to a cement clinker firing mechanism, and recording the parameters as process parameters of the rotary cement kiln; sampling data of each time of the process parameters are collected, preprocessing of missing value filling, abnormal value elimination, mean filtering, time matching and homogenization is carried out, and a preprocessed data set of the process parameters of the rotary cement kiln is obtained; selecting a target process parameter, evaluating the correlation degree between other process parameters and the target process parameter, screening a key variable of the target process parameter, and training a prediction model of the target process parameter in combination with an LSTM model; the prediction model is constructed by pertinently selecting the variables which are highly correlated with the target process parameters, so that the prediction model accurately captures the change rule of the target process parameters, and the prediction precision is improved.

Description

Technical Field

[0001] The invention relates to the field of cement rotary kiln process parameter prediction, and in particular to a cement rotary kiln process parameter prediction method based on mechanism and data collaboration. Background Art

[0002] As a basic material in the construction industry, the stable and high-quality production of cement is crucial to ensure the smooth progress of infrastructure construction. The cement rotary kiln is the core equipment in the cement production process, and its operating status directly affects the quality and output of cement clinker.

[0003] In the actual production process, the interior of the cement rotary kiln is an extremely complex physical and chemical reaction system, involving multiple processes such as fuel combustion, material transmission, heat exchange and chemical reaction. These processes are intertwined, making the process parameters extremely susceptible to fluctuations due to multiple factors. If the process parameters can be accurately predicted in advance, production personnel can adjust production operations in a timely manner based on the prediction results to prevent problems such as product quality degradation and equipment failure caused by abnormal parameters, thereby effectively improving production efficiency and economic benefits.

[0004] At present, there are obvious defects in the research and application of prediction of cement rotary kiln process parameters.

[0005] For example, the existing Chinese patent with publication number CN117216498A discloses a prediction system and prediction method for the operating condition of a cement rotary kiln, the prediction method comprising obtaining characteristic data characterizing the operating condition of the cement rotary kiln, the characteristic data comprising training data and test data; constructing an operating condition prediction model, the operating condition prediction model comprising an LSTM model, wherein the LSTM model comprises a forget gate, an input gate and an output gate, and at least one of the forget gate, the input gate and the output gate exposes the cell information of the previous moment to the cell information of the current moment during operation; the LSTM model is trained using the training data to obtain a trained operating condition prediction model. The test data is predicted using the trained operating condition prediction model, and the prediction result is output. In this way, the cell information of the previous moment is exposed to the cell information of the current moment, which improves the accuracy of the prediction result output by the operating condition prediction model.

[0006] For example, the existing Chinese patent with publication number CN115392125A discloses a temperature prediction method for a cement rotary kiln, including the following steps: S1, obtaining multiple groups of historical process parameter data of different time periods in the cement rotary kiln, the real value of the kiln head temperature, and the preset process parameter data when prediction is required; S2, constructing an initial kiln head temperature prediction model through the historical process parameter data of a part of the groups, an integrated residual neural network, and a bidirectional new gated cyclic unit network; S3, inputting the historical process parameter data of the remaining groups and the real value of the kiln head temperature into the initial kiln head temperature prediction model for training to obtain a mature kiln head temperature prediction model; S4, inputting the preset process parameter data into the mature kiln head temperature prediction model for prediction to obtain the final kiln head temperature prediction value. This invention maximizes the use of process parameter data, improves prediction accuracy, reduces the phenomenon of hysteresis in recurrent neural network prediction, and meets the temperature prediction needs of small sample cement kilns.

[0007] The above patent selects fixed variables or all process parameters in the cement rotary kiln production as input variables when predicting the operating conditions or process parameters of the cement rotary kiln. It does not systematically analyze the correlation between various variables and process parameters in the cement rotary kiln production based on the mechanism of burning cement clinker, and thus cannot specifically select variables with a high correlation with process parameters to construct a prediction model, resulting in the prediction model being unable to accurately capture the changing patterns of process parameters, and the prediction accuracy is low, making it difficult to meet the demand for high-precision prediction in actual production, which seriously restricts the intelligent and refined development of the cement production process. Summary of the invention

[0008] In response to the above problems, the present invention proposes a method for predicting cement rotary kiln process parameters based on mechanism and data collaboration. The specific technical scheme is as follows: A method for predicting cement rotary kiln process parameters based on mechanism and data collaboration, comprising the following steps: Step 1: According to the mechanism of burning cement clinker, parameters related to the production status are selected from the temperature data, pressure data, flow-related data, equipment operation data, and gas composition data produced by the cement rotary kiln, and recorded as the process parameters of the cement rotary kiln.

[0009] Step 2: Monitor the data of each sampling of process parameters in cement rotary kiln production, and perform preprocessing such as missing value filling, outlier removal, mean filtering, time matching, and homogenization to obtain a data set of cement rotary kiln process parameters after preprocessing.

[0010] Step 3: Select the target process parameters of the cement rotary kiln, evaluate the correlation between other process parameters and the target process parameters based on the data set of the cement rotary kiln process parameters after pretreatment, and screen the key variables of the target process parameters.

[0011] Step 4: Use the data set of the target process parameters and their key variables as the data set of the target process parameter prediction model, and divide it into a training set and a test set. Input the training set into the LSTM model to train the target process parameter prediction model.

[0012] Step 5: Based on the test set of the target process parameter prediction model, the accuracy of the target process parameter prediction model is evaluated.

[0013] Compared with the prior art, the method for predicting cement rotary kiln process parameters based on mechanism and data collaboration described in the present invention has the following beneficial effects: 1. The present invention preprocesses the data used to predict the target process parameters of the cement rotary kiln by filling in missing values, eliminating outliers, filtering by mean, time matching, and homogenization in sequence, so that the data can be better applied to the prediction of cement rotary kiln process parameters, thereby improving the accuracy of the prediction.

[0014] 2. The present invention preliminarily selects the process parameters of the cement rotary kiln according to the mechanism of burning cement clinker, further evaluates the correlation between other process parameters and the target process parameters, screens the key variables of the target process parameters, and trains the prediction model of the target process parameters in combination with the LSTM model. The prediction model is constructed by selectively selecting variables with a high correlation with the target process parameters, so that the prediction model can accurately capture the changing patterns of the target process parameters and improve the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0016] Figure 1 It is a schematic diagram of the method flow of the present invention.

[0017] Figure 2 The present invention is a schematic diagram of the process parameters pretreatment process of the cement rotary kiln.

[0018] Figure 3 It is a schematic diagram of the process of screening key variables of target process parameters of the present invention.

[0019] Figure 4 It is a flow chart of the present invention's target process parameter prediction model based on LSTM model training. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] See also Figure 1 As shown, the present invention provides a cement rotary kiln process parameter prediction method based on mechanism and data collaboration, comprising the following steps: Step 1: According to the mechanism of burning cement clinker, parameters related to the production status are selected from the temperature data, pressure data, flow-related data, equipment operation data, and gas composition data produced by the cement rotary kiln, and recorded as the process parameters of the cement rotary kiln.

[0022] As a preferred solution, the specific analysis process of step one is as follows: S1: according to the mechanism of firing cement clinker, parameters related to the production status are selected from various parameters of the cement rotary kiln production temperature data, including smoke chamber temperature, kiln head hood temperature, tertiary air temperature, decomposition furnace temperature, electric dust collector inlet temperature, C5A and C5B outlet temperature and feed pipe temperature, kiln shell surface temperature, and cooler outlet clinker temperature.

[0023] It should be noted that the mechanism of burning cement clinker is to mix and grind raw materials such as limestone, clay, iron ore and coal in a certain proportion, and then undergo a series of complex physical and chemical changes under high temperature to eventually form cement clinker with calcium silicate as the main component. The main processes include: raw material crushing and pre-homogenization, raw material grinding, raw material homogenization, preheating and decomposition, clinker burning, clinker cooling, and waste gas treatment.

[0024] It should be noted that the tertiary air temperature refers to the temperature of the air that enters the tail of the rotary kiln from the grate cooler and then enters the decomposition furnace during the production process of the cement rotary kiln.

[0025] It should be noted that C5A and C5B represent two cyclones in the five-stage preheater. The C5A outlet temperature and the C5B outlet temperature refer to the gas temperature measurement values ​​at the outlets of the C5A cyclone and the C5B cyclone; the C5A feed pipe temperature and the C5B feed pipe temperature refer to the temperature of the materials in the C5A cyclone feed pipe and the C5B cyclone feed pipe.

[0026] S2: Select parameters related to the production status from various parameters of the cement rotary kiln production pressure data, including smoke chamber pressure, tertiary air pressure, kiln head negative pressure, electrostatic precipitator inlet pressure, high-temperature fan inlet pressure, high-temperature fan outlet pressure, coal injection negative pressure at the kiln head and kiln tail, pressure at the kiln tail seal, inlet pressure and outlet pressure of C5A and C5B.

[0027] It should be noted that the tertiary air pressure refers to the air pressure when the hot air from the grate cooler is introduced into the decomposition furnace through the tertiary air duct.

[0028] S3: Select parameters related to the production status from various parameters of cement rotary kiln production flow-related data, including the amount of raw material fed into the kiln, the amount of coal fed at the kiln tail and kiln head, the frequency and current of the kiln head exhaust fan, the high-temperature fan current, the cooling air flow, and the coal powder conveying gas flow.

[0029] S4: Select parameters related to the production status from various parameters of the cement rotary kiln production equipment operation data, including the kiln main machine current, the kiln main motor speed, the grate pressure of the grate cooler, the coal injection load at the kiln tail and kiln head, the bearing temperature of the rotary kiln, the oil temperature and vibration value of the reducer.

[0030] S5: Select parameters related to the production status from various parameters of cement rotary kiln production gas composition data, including kiln nitrogen oxide concentration, kiln oxygen content, smoke chamber CO concentration, kiln tail exhaust SO2 content, and kiln reducing gas content.

[0031] S6: Record the parameters related to the production status of the cement rotary kiln in S1-S5 as process parameters of the cement rotary kiln, and summarize them to obtain a set of process parameters of the cement rotary kiln.

[0032] It should be noted that the parameters related to the production status are selected from the temperature data, pressure data, flow-related data, equipment operation data, and gas composition data of the cement rotary kiln production. In addition to being based on the mechanism of burning cement clinker, it can also be based on equipment characteristics, production experience, data analysis, etc.

[0033] Step 2: Monitor the data of each sampling of process parameters in cement rotary kiln production, and perform preprocessing such as missing value filling, outlier removal, mean filtering, time matching, and homogenization to obtain a data set of cement rotary kiln process parameters after preprocessing.

[0034] As a preferred option, see Figure 2 As shown, the specific analysis process of step 2 is as follows: F1: Collecting data: Setting the time period for monitoring cement rotary kiln production, and setting the sampling time interval and sampling times.

[0035] Obtain the data of each sampling of each process parameter of the cement rotary kiln during cement rotary kiln production.

[0036] F2: Missing value filling: Determine whether there are missing data in each sampling of the process parameters of the cement rotary kiln. If a process parameter is missing in a certain sampling, the mean of the corresponding data of the adjacent previous sampling and the adjacent next sampling of the process parameter in this sampling will be used as the data of the process parameter in this sampling and filled.

[0037] It should be noted that there are many methods for data filling, and you can also fill in the data by median filling, mode filling, filling based on similar samples, using machine learning algorithms, etc. Reasonable estimation and filling of missing values ​​according to the characteristics of the data and specific needs can reduce the impact of missing data on subsequent data analysis and modeling.

[0038] F3: Outlier elimination: According to the Laida criterion, outliers in the sampling data of each process parameter of the cement rotary kiln are eliminated.

[0039] F4: Mean filtering: Take the data of each sampling of each process parameter of the cement rotary kiln as the center, set the neighborhood range, record the average value of the data points in the neighborhood range as the neighborhood mean corresponding to the center point data, and replace the center point data with the neighborhood mean, so as to obtain the neighborhood mean corresponding to the sampling data of each process parameter of the cement rotary kiln and replace it.

[0040] It should be noted that, for one-dimensional data, several data points before and after the current data point can be selected as the neighborhood.

[0041] F5: Time Matching: Extract the residence time of the same batch of materials in the cement rotary kiln production at each stage of cement clinker burning stored in the database, and combine the sampling time points corresponding to each sampling data of each process parameter of the cement rotary kiln to determine whether the sampling data of different process parameters of the cement rotary kiln are the data collected when producing the same batch of materials and classify them, and then time align the sampling data of each process parameter of the cement rotary kiln.

[0042] In a specific embodiment, the residence time of the same batch of materials in various stages of cement clinker burning in cement rotary kiln production is shown in Table 1.

[0043] Table 1. Material stay schedule at each stage

[0044]

[0045] It should be noted that cement production is a continuous process. The same batch of materials stays in different equipment for different times, and various process parameters and test parameters will be generated at different stages. Therefore, it is necessary to match the time of each process parameter of the cement rotary kiln before it can be applied to the prediction of the process parameters of the cement rotary kiln. The test parameters of the laboratory are not detected online and also need to be time-matched with the process parameters.

[0046] It should be noted that the various process parameters of the cement rotary kiln are generated at different times. In order to accurately predict the process parameters of the cement rotary kiln in the future, it is necessary to reasonably correspond the relevant process parameters in time. That is, it is necessary to integrate all the relevant process parameters generated by the same batch of materials at different times and locations according to the actual residence time sequence of the materials in each equipment, so that they have an accurate correspondence on the time axis, so that they can be used in subsequent analysis and prediction models.

[0047] F6: Homogenization: Perform homogenization processing on the data of each sampling of each process parameter of cement rotary kiln according to the homogenization formula.

[0048] F7: Based on the data of each sampling of each process parameter of the cement rotary kiln after pretreatment, a data set of each process parameter of the cement rotary kiln after pretreatment is constructed.

[0049] As a preferred solution, the specific analysis process of step F3 is as follows: F31: Record the data of each sampling of each process parameter of the cement rotary kiln as , Indicates The number of the process parameter, , Indicates The number of subsamples, .

[0050] F32: By calculating the formula Obtain the mean value of each process parameter of cement rotary kiln ,in Indicates the sampling times of the process parameter.

[0051] It should be noted that the sampling times of the process parameters are the number of data of the process parameters.

[0052] F33: By calculating the formula Calculate the deviation between the sampling data of each process parameter of cement rotary kiln and its process parameter mean value .

[0053] F34: By calculating the formula Calculate the standard deviation between the sampling data of each process parameter of the cement rotary kiln and its process parameter mean .

[0054] F35: By calculating the formula Analyze the abnormal index of each sampling data of each process parameter of cement rotary kiln If the abnormal index of a certain sampling data of a process parameter of a cement rotary kiln is 1, the sampling data of the process parameter is judged to be an abnormal value and is eliminated.

[0055] As a preferred solution, the specific analysis process of step F6 is as follows: F61: by calculating formula Calculate the variance between the sampling data of each process parameter of cement rotary kiln and its process parameter mean .

[0056] F62: The average value of each process parameter of cement rotary kiln and the variance between each sampling data of each process parameter and its process parameter mean Substitute into the calculation formula Obtain the homogenized data of each sampling of each process parameter of the cement rotary kiln.

[0057] In this embodiment, the present invention preprocesses the data used to predict the target process parameters of the cement rotary kiln by filling missing values, eliminating outliers, filtering the mean, matching time, and homogenizing, so that the data can be better applied to the prediction of the process parameters of the cement rotary kiln and the accuracy of the prediction is improved.

[0058] Step 3: Select the target process parameters of the cement rotary kiln, evaluate the correlation between other process parameters and the target process parameters based on the data set of the cement rotary kiln process parameters after pretreatment, and screen the key variables of the target process parameters.

[0059] As a preferred option, see Figure 3 As shown, the specific analysis process of step three is: D1: Determine the analysis sequence: Select the process parameter to be predicted from the various process parameters of the cement rotary kiln, record it as the target process parameter of the cement rotary kiln, and record the process parameters except the target process parameter as other process parameters.

[0060] It should be noted that the target process parameters of the cement rotary kiln are artificially selected according to the actual application requirements. There can be multiple target process parameters of the cement rotary kiln, and the prediction principle is exactly the same. In one specific embodiment, the target process parameter is the kiln head hood temperature; in another specific embodiment, the target process parameters include the kiln main machine current, kiln head hood temperature, kiln tail smoke chamber temperature, tertiary air temperature, kiln head negative pressure, kiln tail negative pressure, carbon monoxide concentration, oxygen concentration, and nitrogen oxide concentration.

[0061] According to the data set of cement rotary kiln process parameters after pretreatment, the data set of target process parameters is screened and merged to construct a parent sequence, and the data sets of other process parameters are obtained and merged to construct subsequences corresponding to other process parameters, which are recorded as subsequences.

[0062] It should be noted that the present invention uses a grey correlation analysis method to obtain key variables of target process parameters. Grey correlation analysis is a multi-factor statistical analysis method used to determine the degree of correlation between factors, thereby selecting key variables.

[0063] It should be noted that the parent sequence is also called the reference sequence, and the child sequence is also called the comparison sequence. The reference sequence is a variable sequence that reflects the behavior characteristics of the system and is the target variable that researchers are concerned about. The comparison sequence is a variable sequence of various influencing factors that are associated with the reference sequence.

[0064] D2: Data preprocessing: Calculate column means and normalize data for the parent sequence and each subsequence.

[0065] It should be noted that the parent sequence and each subsequence are preprocessed by column mean calculation and data normalization to eliminate the influence of dimension and order of magnitude.

[0066] It should be noted that column mean calculation and data normalization in data preprocessing are relatively mature methods and means, which will not be elaborated here.

[0067] D3: Calculate the correlation coefficient: Calculate the correlation coefficient between each subsequence and the parent sequence at each sampling according to the correlation coefficient formula , Indicates The number of the subsequence, which also indicates the number of the other The number of the process parameter, .

[0068] D4: Calculate the compensation amount of the correlation: Obtain the compensation amount of the correlation between each subsequence and the parent sequence through regression analysis method, and record it as .

[0069] D5: Calculate the correlation: by calculating the formula Analyze the correlation between each subsequence and the parent sequence, where Indicates The degree of association between a subsequence and the parent sequence.

[0070] It should be noted that the correlation coefficient reflects the degree of correlation between the subsequence and the parent sequence at each sampling, but in order to obtain a comprehensive correlation index, the correlation coefficient needs to be weighted averaged.

[0071] It should be noted that the compensation amount of the correlation obtained by regression analysis in the above correlation calculation formula is used to correct the correlation error caused by system deviation or nonlinear relationship. For example, when there is a lag effect or nonlinear dependence between the parent sequence and the child sequence, the regression residual can quantify such deviations; the formula comprehensively evaluates the final correlation by combining the average correlation coefficient and the compensation amount to comprehensively measure the overall correlation between the child sequence and the parent sequence.

[0072] It should be noted that the above correlation calculation formula uses mean calculation, which can eliminate random interference and extract the overall correlation trend; the use of regression analysis compensation can capture the systematic deviation ignored by traditional correlation calculation; the use of a comprehensive formula of mean and compensation can take into account both local correlation strength and global system deviation, thereby improving the accuracy of the correlation.

[0073] It should be noted that a feasible simulation process assumes that the parent sequence of the target process parameters is , the first other process parameter corresponding subsequence is , the second other process parameter corresponds to the subsequence The preset resolution coefficient is set manually at the beginning of the system operation. ; and The correlation coefficient is: , the mean correlation coefficient is 0.96; and The correlation coefficient is: , the mean correlation coefficient is 0.67; the first other process parameter corresponds to the compensation amount of the correlation between the subsequence and the parent sequence. , the second other process parameter corresponds to the compensation amount of the correlation between the subsequence and the parent sequence: , we can get the simulation results, please refer to Table 2 for details.

[0074] Table 2. Correlation calculation results

[0075]

[0076] Through the above simulation calculation, the subsequence The correlation with the parent sequence is greater than that with the child sequence The degree of association with the parent sequence indicates the subsequence Stronger correlation with the parent sequence.

[0077] D6: Association ranking: Sort the associations between each subsequence and the parent sequence from high to low to obtain the association rankings between other process parameters and the target process parameter.

[0078] It should be noted that the greater the correlation, the stronger the correlation between the subsequence and the parent sequence, and the more likely it is to be a key variable.

[0079] D7: Verification of resolution coefficient: If the difference between the correlation between a subsequence and the parent sequence and the correlation between another subsequence and the parent sequence within the set correlation ranking is less than the preset difference threshold, the resolution coefficient needs to be adjusted, and alternative values ​​of the resolution coefficient are extracted from the database for iteration.

[0080] In a specific embodiment, the ranking is set to the top ten.

[0081] In one specific embodiment, the candidate values ​​of the resolution coefficient are , .

[0082] It should be noted that the selection of the resolution coefficient in the grey relational analysis is an important task before the analysis. Selecting a suitable resolution coefficient is conducive to better distinguishing the correlation between the target process parameter and other process parameters. Therefore, in order to better select other process parameters that are strongly correlated with the target process parameter, the resolution coefficient is modified so that the grey relational analysis can select process parameters with obvious mutual correlation to better promote the prediction of the target process parameter.

[0083] In one specific embodiment, the target process parameter is the kiln hood temperature, based on , , The three resolution coefficients are used to calculate the correlation between some other process parameters and the kiln head hood temperature. Please refer to Table 3, Table 4 and Table 5 for details.

[0084] Table 3. Correlation results when resolution coefficient ρ=0.5

[0085]

[0086] In Table 3, the top three correlations with the kiln head hood temperature are tertiary wind temperature, smoke chamber temperature, and kiln head discharge frequency, but the correlation values ​​are very similar and there is no distinguishing effect.

[0087] Table 4. Correlation results when resolution coefficient ρ = 0.1

[0088]

[0089] Table 5. Correlation results when resolution coefficient ρ=0.05

[0090]

[0091] Comprehensively from Table 3, Table 4, and Table 5, when the resolution coefficient is 0.5, other process parameters related to the kiln head hood temperature cannot be effectively screened out. When it is 0.1, the results after correlation analysis are also not clearly distinguished. When it is 0.05, the correlation between other process parameters and the kiln head hood temperature can be clearly analyzed. Therefore, the resolution coefficient is adjusted to 0.05 to better achieve gray correlation degree distinction.

[0092] D8: Recalculation of correlation: Based on the adjusted resolution coefficient, execute D3-D6 to recalculate the correlation between other process parameters and the target process parameter.

[0093] D9: Key variable selection: other process parameters corresponding to the correlation degree greater than or equal to the preset correlation degree threshold are recorded as key variables of the target process parameters, and the key variables of the target process parameters are counted.

[0094] In a specific embodiment, the preset correlation threshold is 0.85.

[0095] It should be noted that the key variables are determined based on the correlation ranking results, combined with the background of the actual problem and the research purpose. Other process parameters with higher correlation are generally considered to be key variables that have a greater impact on the target process parameters, but in actual selection, factors such as the operability and economic significance of the variables also need to be considered.

[0096] In a specific embodiment, the target process parameter is the kiln head hood temperature, and the key variables of the target process parameter are tertiary air temperature, raw material feeding amount into the kiln, kiln main machine speed, kiln head coal injection negative pressure, kiln head coal feeding amount, decomposition furnace temperature, kiln nitrogen oxide concentration, kiln main machine current, tertiary air pressure, kiln head negative pressure, and kiln oxygen content.

[0097] As a preferred solution, the specific analysis process of step D3 is: by calculating the formula Analyze the correlation coefficient between each subsequence and the parent sequence at each sampling, where Indicates The subsequence and the parent sequence are in The correlation coefficient on the subsampling, Indicates that the parent sequence is The subsampled data, Indicates The subsequence is The subsampled data, Indicates the preset resolution factor, , represents the minimum difference between the two levels, Indicates the maximum difference between two levels.

[0098] As a preferred solution, the specific analysis process of step D4 is: taking the subsequence as the independent variable and the parent sequence as the dependent variable, establishing a regression model between each subsequence and the parent sequence according to each subsequence and the parent sequence, using the least squares method to analyze the parameters in the regression model, and calculating the determination coefficient of the regression model between each subsequence and the parent sequence.

[0099] The relationship comparison table between the regression model determination coefficient and the association degree compensation amount stored in the database is extracted to match and obtain the compensation amount of the association degree between each subsequence and the parent sequence.

[0100] It should be noted that the regression model includes but is not limited to linear regression, polynomial regression, etc.

[0101] It should be noted that the value of the determination coefficient of the regression model is between 0 and 1. The closer it is to 1, the better the model fits the data, that is, the independent variable can better explain the changes in the dependent variable.

[0102] It should be noted that the method for obtaining the determination coefficient of the regression model in regression analysis is an existing relatively mature technology and will not be elaborated here.

[0103] Step 4: Use the data set of the target process parameters and their key variables as the data set of the target process parameter prediction model, and divide it into a training set and a test set. Input the training set into the LSTM model to train the target process parameter prediction model.

[0104] As a preferred option, see Figure 4 As shown, the specific analysis process of step four is: taking the data set of the target process parameters and their key variables as the data set of the target process parameter prediction model, dividing the data set according to the set ratio between the training set and the test set, and obtaining the training set and test set of the target process parameter prediction model.

[0105] The target process parameters are taken as output and the key variables of the target process parameters are taken as input. The training set of the target process parameter prediction model is input into the LSTM time series prediction model to train the target process parameter prediction model.

[0106] In a specific embodiment, the data set is divided into a training set and a test set in a ratio of 9:1.

[0107] It should be noted that LSTM is a long short-term memory network, which is a special recurrent neural network. It is the result of the improvement of the traditional recurrent neural network. LSTM introduces the concepts of "gate" and "cell state". The gate structure can control the input and output of information, the cell state can store the state for a long time, and LSTM can control the cell state of the main line through the "gate" structure. In LSTM, the function of the gate is a way to selectively pass information. It is composed of Sigmoid neural network and matrix point-by-point multiplication operation. There are forget gate, input gate and output gate in LSTM. These three gates control the flow of information and determine which information needs to be forgotten, which information needs to be added, and which information needs to be output. The main advantage of LSTM is that it can effectively process long sequence data. Through the introduced gating mechanism and cell state, it can maintain information over a long span and better adapt to the time dynamic characteristics of the data center.

[0108] It should be noted that the production data of cement rotary kiln is a kind of time series data, which involves the changes of multiple parameters over time, such as temperature, pressure, material composition, etc. There are complex dependencies between these parameters, and this relationship may change over time. LSTM neural network is a model that is particularly suitable for processing such time-dependent data.

[0109] It should be noted that in the process of training the target process parameter prediction model, the hyperparameters of the LSTM time series prediction model can be reasonably set to optimize the model performance. The hyperparameters include the number of LSTM layers, activation function, number of neurons, learning rate, etc. In a specific embodiment, the hyperparameter settings of the LSTM time series prediction model are shown in Table 6.

[0110] Table 6. Hyperparameter settings

[0111]

[0112] Step 5: Based on the test set of the target process parameter prediction model, the accuracy of the target process parameter prediction model is evaluated.

[0113] As a preferred solution, the specific analysis process of step five is: substituting the data of each sampling of each key variable of the target process parameter in the test set into the target process parameter prediction model to obtain the predicted value of each sampling of the target process parameter in the test set, and recording the data of each sampling of the target process parameter in the test set as the true value of each sampling of the target process parameter.

[0114] The difference between the predicted value and the true value of each sampling of the target process parameter in the test set is recorded as the error of each prediction of the target process parameter in the test set, and the average value is calculated to obtain the mean prediction error of the target process parameter. The relationship comparison table between the mean prediction error and the model accuracy stored in the database is extracted to match the accuracy of the target process parameter prediction model.

[0115] It should be noted that when the accuracy of the target process parameter prediction model is lower than the required accuracy, the test set is included in the training set to obtain a new training set, and the target process parameter prediction model is retrained based on the new training set. The target process parameter prediction model is optimized by expanding the data set to improve the generalization ability of the model.

[0116] It should be noted that the target process parameter prediction model obtained by selecting the input quantity through correlation analysis, that is, through dynamic input quantity selection, is more accurate than the target process parameter prediction model obtained by selecting the input quantity without correlation analysis, that is, through fixed input quantity selection, and can better realize the trend judgment. In a specific embodiment, referring to Table 7, the prediction effect of the dynamic selected input quantity training prediction model and the fixed selected input quantity training prediction model of some process parameters of the cement rotary kiln are compared from the three indicators of the regression model's fit metric R2, root mean square error, and mean absolute error.

[0117] Table 7. Comparison of the effects of some time series models

[0118]

[0119] In this embodiment, the present invention preliminarily selects the process parameters of the cement rotary kiln according to the mechanism of burning cement clinker, further evaluates the correlation between other process parameters and the target process parameters, screens the key variables of the target process parameters, and combines the LSTM model to train the prediction model of the target process parameters. The prediction model is constructed by selectively selecting variables with a high correlation with the target process parameters, so that the prediction model can accurately capture the changing patterns of the target process parameters and improve the prediction accuracy.

[0120] The above contents are merely examples and explanations of the concept of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they should all fall within the protection scope of the present invention.

Claims

1. A cement rotary kiln process parameter prediction method based on mechanism and data collaboration, characterized in that: The steps include: Step 1: According to the mechanism of firing cement clinker, parameters related to the production status are selected from the temperature data, pressure data, flow-related data, equipment operation data, and gas composition data produced by the cement rotary kiln, and recorded as the process parameters of the cement rotary kiln; Step 2: Monitor the data of each sampling of process parameters in cement rotary kiln production, and perform preprocessing such as missing value filling, outlier removal, mean filtering, time matching, and homogenization to obtain a data set of cement rotary kiln process parameters after preprocessing; Step 3: Select the target process parameters of the cement rotary kiln, evaluate the correlation between other process parameters and the target process parameters based on the pre-processed cement rotary kiln process parameter data set, and screen the key variables of the target process parameters; Step 4: Use the data set of the target process parameters and their key variables as the data set of the target process parameter prediction model, and divide it into a training set and a test set. Input the training set into the LSTM model to train the target process parameter prediction model. Step 5: Based on the test set of the target process parameter prediction model, the accuracy of the target process parameter prediction model is evaluated.

2. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 1 is characterized in that: The specific analysis process of step 1 is as follows: S1: According to the mechanism of firing cement clinker, parameters related to the production status are selected from various parameters of the cement rotary kiln production temperature data, including smoke chamber temperature, kiln head hood temperature, tertiary air temperature, decomposition furnace temperature, electric dust collector inlet temperature, C5A and C5B outlet temperature and feed pipe temperature, kiln shell surface temperature, cooler outlet clinker temperature; S2: Select parameters related to the production status from various parameters of cement rotary kiln production pressure data, including smoke chamber pressure, tertiary air pressure, kiln head negative pressure, electric dust collector inlet pressure, high temperature fan inlet pressure, high temperature fan outlet pressure, kiln head and kiln tail coal injection negative pressure, kiln tail sealing pressure, C5A and C5B inlet pressure and outlet pressure; S3: Select parameters related to the production status from various parameters of cement rotary kiln production flow-related data, including the amount of raw material fed into the kiln, the amount of coal fed to the kiln tail and kiln head, the frequency and current of the kiln head exhaust fan, the current of the high-temperature fan, the cooling air flow, and the coal powder conveying gas flow; S4: Selecting parameters related to the production status from various parameters of the cement rotary kiln production equipment operation data, including kiln main machine current, kiln main motor speed, grate pressure of the grate cooler, coal injection load at the kiln tail and kiln head, rotary kiln support wheel bearing temperature, reducer oil temperature and vibration value; S5: Select parameters related to the production status from various parameters of cement rotary kiln production gas composition data, including kiln nitrogen oxide concentration, kiln oxygen content, smoke chamber CO concentration, kiln tail exhaust SO2 content, and kiln reducing gas content; S6: Record the parameters related to the production status of the cement rotary kiln in S1-S5 as process parameters of the cement rotary kiln, and summarize them to obtain a set of process parameters of the cement rotary kiln.

3. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 1 is characterized in that: The specific analysis process of step 2 is as follows: F1: Data collection: Set the time period for monitoring cement rotary kiln production, and set the sampling time interval and sampling times; Obtain the data of each sampling of each process parameter of the cement rotary kiln during cement rotary kiln production; F2: Missing value filling: Determine whether there are missing data in each sampling of the process parameters of the cement rotary kiln. If a process parameter is missing in a certain sampling, the mean of the corresponding data of the adjacent previous sampling and the adjacent next sampling of the process parameter in this sampling is used as the data of the process parameter in this sampling and filled; F3: Outlier elimination: According to the Laida criterion, outliers in the sampling data of each process parameter of the cement rotary kiln are eliminated; F4: Mean filtering: Take the data of each process parameter of the cement rotary kiln sampled at each time as the center, set the neighborhood range, record the average value of the data points in the neighborhood range as the neighborhood mean corresponding to the center point data, and replace the center point data with the neighborhood mean, so as to obtain the neighborhood mean corresponding to the sampled data of each process parameter of the cement rotary kiln and replace it; F5: Time matching: Extract the residence time of the same batch of materials in the cement rotary kiln production at each stage of cement clinker burning stored in the database, and combine the sampling time points corresponding to each sampling data of each process parameter of the cement rotary kiln to determine whether the sampling data of different process parameters of the cement rotary kiln are the data collected when the same batch of materials is produced and classify them, and then time align the sampling data of each process parameter of the cement rotary kiln; F6: Homogenization: Perform homogenization processing on the data of each sampling of each process parameter of cement rotary kiln according to the homogenization formula; F7: Based on the data of each sampling of each process parameter of the cement rotary kiln after pretreatment, a data set of each process parameter of the cement rotary kiln after pretreatment is constructed.

4. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 3 is characterized in that: The specific analysis process of step F3 is as follows: F31: Record the sampling data of each process parameter of cement rotary kiln as , Indicates The number of the process parameter, , Indicates The number of subsamples, ; F32: By calculating the formula Obtain the mean value of each process parameter of cement rotary kiln ,in Indicates the sampling times of process parameters; F33: By calculating the formula Calculate the deviation between the sampling data of each process parameter of cement rotary kiln and its process parameter mean value ; F34: By calculating formula Calculate the standard deviation between the sampling data of each process parameter of the cement rotary kiln and its process parameter mean ; F35: By calculating the formula Analyze the abnormal index of each sampling data of each process parameter of cement rotary kiln If the abnormal index of a certain sampling data of a process parameter of a cement rotary kiln is 1, the sampling data of the process parameter is judged to be an abnormal value and is eliminated.

5. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 4 is characterized in that: The specific analysis process of step F6 is as follows: F61: By calculating the formula Calculate the variance between the sampling data of each process parameter of cement rotary kiln and its process parameter mean ; F62: The average value of each process parameter of cement rotary kiln and the variance between each sampling data of each process parameter and its process parameter mean Substitute into the calculation formula Obtain the homogenized data of each sampling of each process parameter of the cement rotary kiln.

6. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 4 is characterized in that: The specific analysis process of step three is as follows: D1: Determine the analysis sequence: select the process parameter to be predicted from the various process parameters of the cement rotary kiln, record it as the target process parameter of the cement rotary kiln, and record the various process parameters except the target process parameter as other process parameters; According to the data set of cement rotary kiln process parameters after pretreatment, the data set of target process parameters is screened and merged to construct a parent sequence, and the data sets of other process parameters are obtained and merged to construct subsequences corresponding to other process parameters, which are recorded as subsequences; D2: Data preprocessing: Calculate column mean and normalize data for the parent sequence and each subsequence; D3: Calculate the correlation coefficient: Calculate the correlation coefficient between each subsequence and the parent sequence at each sampling according to the correlation coefficient formula , Indicates The number of the subsequence, which also indicates the number of the other The number of the process parameter, ; D4: Calculate the compensation amount of the correlation: Obtain the compensation amount of the correlation between each subsequence and the parent sequence through regression analysis method, and record it as ; D5: Calculate the correlation: by calculating the formula Analyze the correlation between each subsequence and the parent sequence, where Indicates The degree of association between the subsequences and the parent sequence; D6: Association ranking: Sort the associations between each subsequence and the parent sequence from high to low to obtain the association ranking between other process parameters and the target process parameter; D7: Verification of resolution coefficient: If the difference between the correlation between a subsequence and the parent sequence and the correlation between another subsequence and the parent sequence in the set correlation ranking is less than the preset difference threshold, the resolution coefficient needs to be adjusted, and the candidate value of the resolution coefficient is extracted from the database for iteration; D8: Recalculation of correlation: According to the adjusted resolution coefficient, execute D3-D6 to recalculate the correlation between other process parameters and the target process parameter; D9: Key variable selection: other process parameters corresponding to the correlation degree greater than or equal to the preset correlation degree threshold are recorded as key variables of the target process parameters, and the key variables of the target process parameters are counted.

7. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 6 is characterized in that: The specific analysis process of step D3 is as follows: By calculating the formula Analyze the correlation coefficient between each subsequence and the parent sequence at each sampling, where Indicates The subsequence and the parent sequence are in The correlation coefficient on the subsampling, Indicates that the parent sequence is The subsampled data, Indicates The subsequence is The subsampled data, Indicates the preset resolution factor, , represents the minimum difference between the two levels, Indicates the maximum difference between two levels.

8. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 6 is characterized in that: The specific analysis process of step D4 is as follows: Taking the subsequence as the independent variable and the parent sequence as the dependent variable, a regression model between each subsequence and the parent sequence is established, the least squares method is used to analyze the parameters in the regression model, and the determination coefficient of the regression model between each subsequence and the parent sequence is calculated; The relationship comparison table between the regression model determination coefficient and the association degree compensation amount stored in the database is extracted to match and obtain the compensation amount of the association degree between each subsequence and the parent sequence.

9. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 1, characterized in that: The specific analysis process of step 4 is as follows: The data set of the target process parameter and its key variables is used as the data set of the target process parameter prediction model, and the data set is divided according to the set ratio between the training set and the test set to obtain the training set and the test set of the target process parameter prediction model; The target process parameters are taken as output and the key variables of the target process parameters are taken as input. The training set of the target process parameter prediction model is input into the LSTM time series prediction model to train the target process parameter prediction model.

10. The method for predicting cement rotary kiln process parameters based on mechanism and data collaboration according to claim 1, characterized in that: The specific analysis process of step 5 is as follows: Substitute the data of each sampling of each key variable of the target process parameter in the test set into the target process parameter prediction model to obtain the predicted value of each sampling of the target process parameter in the test set, and record the data of each sampling of the target process parameter in the test set as the true value of each sampling of the target process parameter; The difference between the predicted value and the true value of each sampling of the target process parameter in the test set is recorded as the error of each prediction of the target process parameter in the test set, and the average value is calculated to obtain the mean prediction error of the target process parameter. The relationship comparison table between the mean prediction error and the model accuracy stored in the database is extracted to match the accuracy of the target process parameter prediction model.

Images