Dry quenching circulating air volume online prediction method and system based on AGRU-PLS model

Abstract

The invention discloses a dry quenching circulating air volume online prediction method and system based on an AGRU-PLS model, and belongs to the technical field of dry quenching production process intelligent prediction. According to the method, the production process data of the dry quenching full-process technology is collected, and a corresponding information database is constructed and stored; preprocessing the data, including cleaning and filling missing data, and finally performing standardization processing on the data; performing feature processing on the data in a correlation analysis mode, and finding out variable features with high correlation with the circulating air volume in the whole process; establishing an AGRU-PLS model, establishing a characteristic and circulating air volume covariance maximum model through an outer model, and solving a data collinearity problem in a mode of solving latent variables by iteration; the method comprises the following steps: establishing an Attention-GRU internal model, and mining nonlinear and dynamic relationships among data; the model is updated, and online prediction of the circulating air volume is achieved; and the prediction result of the circulating air volume is converted into a given value of the circulating fan, and the given value is written into UaExpert software to achieve control over the circulating fan.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent prediction of dry coke quenching production process, and specifically relates to an online prediction method and system for dry coke quenching circulating air volume based on an AGRU-PLS model. Background Art

[0002] As an important support for steel production, the coking industry occupies an indispensable position in my country's economic construction. Coke dry quenching technology is an environmentally friendly technology in the coking process, which effectively cools high-temperature coke through an inert gas circulation system. In the coke dry quenching process, precise control of the circulating air volume is crucial to maintaining appropriate coke discharge temperature and boiler inlet temperature, which directly affects production stability and product quality. Specifically, if the circulating air volume is set too low, the coke discharge temperature and boiler inlet temperature will increase abnormally, which may cause the coke discharge belt to burn out and increase equipment maintenance costs; conversely, if the circulating air volume is too large, the boiler inlet temperature will be reduced, which will not only affect steam production, but may also cause unnecessary waste of energy.

[0003] However, in current production practice, the control of circulating air volume mainly relies on the personal experience and manual operation of the operator, which has significant limitations. Due to the different experience levels of each operator and the different judgment criteria for circulating air volume control, it is difficult to ensure the stability of the production process, which affects the overall production efficiency and consistency of product quality. At the same time, the actual production process of CDQ is a closed-loop process, and the production data has multivariable coupling, nonlinearity and time-varying properties, which cannot be fully captured manually. Therefore, exploring and implementing a more intelligent and automated circulating air volume control system to reduce dependence on manual experience and improve the stability and controllability of the production process has become an urgent need for the transformation and upgrading of the coking industry and enhance its competitiveness. Summary of the invention

[0004] Aiming at the problems of inconsistent circulating air volume adjustment parameters, untimely adjustment and unprovable manual experience in the actual production process of dry coke quenching, which lead to unstable production, uncontrollable production process parameters and poor coke quality, the present invention provides an online prediction method and system for circulating air volume of dry coke quenching based on AGRU-PLS model. The present invention fully considers the multi-coupling, nonlinear and time-varying nature of variables in the production process, and realizes real-time automatic control of circulating air volume. It is of great significance to realize the automation and intelligent transformation of dry coke quenching.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An online prediction method for CDQ cycle air volume based on AGRU-PLS model comprises the following steps:

[0007] Step 1, data collection and database construction: collect the whole process data of CDQ from the DCS system, build the corresponding information database, and store the collected data in the database;

[0008] Step 2, data preprocessing: select a period of historical data of the full-process production process, clean the data, remove abnormal data, fill in missing data, and finally standardize the data;

[0009] Step 3, feature engineering: perform feature processing on the data through correlation analysis to find out the variable features with high correlation with the circulating air volume in the entire process;

[0010] Step 4: Establish an AGRU-PLS model: including an external model and an internal model; establish a maximum covariance model of features and circulating air volume through the external model, and solve the collinearity problem of data by iteratively solving latent variables; and explore the nonlinear and dynamic relationships between data by establishing an Attention-GRU internal model;

[0011] Step 5, prediction results: update the model every 3 to 5 hours to achieve online prediction of circulating air volume;

[0012] Step 6, direct control: convert the predicted result of the circulation air volume into a given value of the circulation fan, and write it into the UaExpert software to realize the control of the circulation fan.

[0013] Further, the step 2 specifically includes the following steps:

[0014] Step 2.1, take the production process parameters with missing data as output, and other parameters without missing data as input, select the data with complete features to establish a regression model; then bring the input into the model to predict the output missing data, and complete the filling of missing data;

[0015] Step 2.2, a box plot is performed to visualize the parameters of the whole process in the historical database used to train the model, and data points above the upper whisker and below the lower whisker are considered as abnormal points and are removed;

[0016] In step 2.3, the Z-Score standardization method is used to normalize the full-process process data after outlier processing and missing value supplementation.

[0017] Further, the step 3 specifically includes the following steps:

[0018] Step 3.1, use the Pearson correlation coefficient to obtain the characteristics linearly related to the circulating air volume, as shown in the following formula:

[0019]

[0020] Where x represents the variable in the whole process, and y represents the circulating air volume;

[0021] Step 3.2, use the Spearman correlation coefficient to obtain the characteristics of nonlinear correlation with circulating air volume, as shown in the following formula:

[0022]

[0023] Where x represents the variable in the whole process, and y represents the circulating air volume.

[0024] Furthermore, the external model in step 4 specifically includes the following steps:

[0025] Step 4.1, establish the covariance objective function of the feature and the circulating air volume in the latent variable space:

[0026]

[0027] in, is the circulating air volume, is the selected feature variable, q and w are the projection vectors of Y and X respectively, u=Yq and t=Xw are the latent variables of Y and X respectively; T represents transposition;

[0028] Step 4.2: Use the Attention-GRU model to establish a dynamic nonlinear internal model between latent variables u and t. The estimated value of the internal model is:

[0029]

[0030] Among them, the G(·) function is the prediction model established by Attention-GRU;

[0031] Step 4.3: After extracting the latent variable with the greatest correlation between the feature and the circulating air volume, continue to extract the next most important latent variable from the remaining information, and repeat steps 4.1 to 4.3 until enough latent variables are extracted:

[0032]

[0033] Where p = X T t / t T t and q=Y T u / u T u is the load vector; b represents the regression coefficient of u and t;

[0034] Step 4.4, the characteristics and circulating air volume are finally decomposed into:

[0035]

[0036] in B=diag[b1,b2,…,b L ], E and F are the residual matrices of X and Y respectively; represents the real number space;

[0037] Step 4.5, the regression model established:

[0038] Y=XC=XW(P T W) -1 B (7)

[0039] C represents the regression coefficients of X and Y, and W represents the projection matrix of X in the latent variable space;

[0040] Step 4.6, the final prediction model established is:

[0041]

[0042] in, Represents the prediction results of the model; Represents the variable being predicted.

[0043] Furthermore, the internal model in step 4 is specifically:

[0044] According to the attention mechanism module, the influence coefficient of the variables at the historical moment on the variables at the current moment is obtained, and the influence coefficient is used as the weight α of the GRU-RNN model input t ;

[0045] The input sequence of the GRU-RNN module is {x1,x2,x3,…,x Q}, the hidden state is {h1,h2,h3,…,h Q}, the output layer of the GRU-RNN model at time step s is:

[0046] Y s =h s W hq +b q (9)

[0047] Among them, h t is the hidden state at time s, W hq Represents the weight matrix; the calculation formula for time step t is:

[0048]

[0049] Among them, h s-1 is the hidden state at time s-1, ⊙ represents element multiplication, represents the candidate hidden state; z sThe update gate is used to update the memory and decide how much past information to pass to the future; its expression is:

[0050] z s =σ(W z x s +U z h s-1 ) (11)

[0051] Among them, σ is the logistic sigmoid function, U z and W z represents the weight matrix; candidate hidden state Represents the current memory content, calculated by the following formula:

[0052]

[0053] Among them, r t It is a reset gate, which discards historical information irrelevant to prediction and combines the new input information with the previous memory. The calculation formula is:

[0054] r s =σ(W r x s +U r h s-1 ) (13).

[0055] Furthermore, the variable characteristics include the given circulating air volume, the coke discharge volume, the coke discharge temperature, the boiler inlet temperature, the boiler outlet temperature, the CDQ furnace inlet temperature, the fan outlet pressure, the fan inlet pressure, the CDQ furnace inlet pressure, the boiler feed water pump bearing temperature, the boiler feed water pump flow, the circulating fan current, the boiler feed water flow, the boiler outlet superheated steam flow, and the dust removal fan vibration.

[0056] An online prediction model for CDQ cycle air volume based on AGRU-PLS model, comprising a data acquisition and database storage module, a data preprocessing module, a data feature extraction module, a model training module, an online control variable prediction module, and a control module;

[0057] The data collection and database storage module collects the full process data of the dry coke quenching production process from the OPC server through the Java language and stores the data in the MySQL database;

[0058] The data preprocessing module processes outliers, supplements missing data, and normalizes data zscores in the database;

[0059] The data feature extraction module is to select the variable features with high correlation with the circulating air volume from the collected CDQ process variables, including linear and nonlinear relationships;

[0060] The variable characteristics include: given circulating air volume, coke discharge volume, coke discharge temperature, boiler inlet temperature, boiler outlet temperature, CDQ furnace inlet temperature, fan outlet pressure, fan inlet pressure, CDQ furnace inlet pressure, boiler feed water pump bearing temperature, boiler feed water pump flow, circulating fan current, boiler feed water flow, boiler outlet superheated steam flow, and dust removal fan vibration. ;

[0061] The model training module is used to establish the AGRU-PLS model to predict the circulation air volume.

[0062] The control module is to write the predicted results into UaExpert, and then write the results into the given node of the corresponding circulating fan of PLC through UaExpert.

[0063] Compared with the prior art, the present invention has the following advantages:

[0064] The present invention significantly improves the intelligence level of circulating air volume regulation. The present invention innovatively applies artificial intelligence technology to deeply explore the complex relationship between high-dimensional variables and circulating air volume, thereby realizing automatic and precise control of circulating air volume. This breakthrough not only significantly improves the stability of the production process, but also effectively avoids the uncertainty and error caused by traditional manual experience adjustment. More importantly, the model constructed by the present invention has a strong data processing capability and can solve the complex characteristics such as collinearity, dynamics and nonlinearity that are prevalent among data. This feature enables the model to maintain a high degree of accuracy and reliability when facing a complex and changeable production environment. In addition, the present invention also cleverly introduces an attention mechanism, which can intelligently capture and focus on the key variables that have the most significant impact on the circulating air volume. This precise information screening and weight distribution further ensures the accuracy and robustness of the model in predicting the circulating air volume.

[0065] In summary, the present invention realizes high-precision automatic control of circulating air volume by integrating artificial intelligence, complex data processing technology and attention mechanism, providing strong technical support for improving production efficiency, ensuring product quality and reducing operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of the online prediction method for CDQ cycle air volume based on the AGRU-PLS model proposed in the present invention;

[0067] Figure 2 This is a framework diagram of the online prediction method for CDQ cycle air volume based on the AGRU-PLS model proposed in the present invention;

[0068] Figure 3 A schematic diagram of the Attention-GRU model used in the present invention;

[0069] Figure 4 The prediction results of the online prediction method for CDQ cycle air volume based on the AGRU-PLS model proposed in the present invention;

[0070] Figure 5 The prediction results of the online prediction method for CDQ cycle air volume based on the AGRU-PLS model proposed in the present invention are compared with the prediction results of the LSTM model; DETAILED DESCRIPTION

[0071] In order to gain a deeper understanding of the present invention, we will provide a comprehensive and detailed description of the present invention. However, the present invention has multiple implementations and is not limited to the specific examples listed herein. The presentation of these examples is intended to deepen the comprehensive understanding of the disclosure of the present invention.

[0072] Example 1

[0073] An online prediction model for CDQ cycle air volume based on AGRU-PLS model, comprising a data acquisition and database storage module, a data preprocessing module, a data feature extraction module, a model training module, an online control variable prediction module, and a control module;

[0074] The data collection and database storage module collects the full process data of the dry coke quenching production process from the OPC server through the Java language and stores the data in the MySQL database;

[0075] The data preprocessing module processes outliers, supplements missing data, and normalizes data zscores in the database;

[0076] The data feature extraction module is to select the variable features with high correlation with the circulating air volume from the collected CDQ process variables, including linear and nonlinear relationships;

[0077] The variable characteristics include: given circulating air volume, coke discharge volume, coke discharge temperature, boiler inlet temperature, boiler outlet temperature, CDQ furnace inlet temperature, fan outlet pressure, fan inlet pressure, CDQ furnace inlet pressure, boiler feed water pump bearing temperature, boiler feed water pump flow, circulating fan current, boiler feed water flow, boiler outlet superheated steam flow, dust removal fan vibration;

[0078] The model training module is used to establish the AGRU-PLS model, explore the dynamic nonlinear relationship between the extracted features and the circulating air volume, and realize the prediction of the circulating air volume.

[0079] The control module is to write the predicted results into UaExpert, and then write the results into the given node of the corresponding circulating fan of PLC through UaExpert.

[0080] Example 2

[0081] An online prediction method for CDQ cycle air volume based on AGRU-PLS model comprises the following steps:

[0082] Step 1, data collection and database construction: collect the whole process data of CDQ from the DCS system, build the corresponding information database, and store the collected data in the database;

[0083]

[0084]

[0085] Step 2, data preprocessing: select a period of historical data of the full-process production process, clean the data, remove abnormal data, fill in missing data, and finally standardize the data;

[0086] Step 3, feature engineering: perform feature processing on the data through correlation analysis to find out the variable features with high correlation with the circulating air volume in the entire process;

[0087] Step 4: Establish an AGRU-PLS model: including an external model and an internal model; establish a maximum covariance model of features and circulating air volume through the external model, and solve the collinearity problem of data by iteratively solving latent variables; and explore the nonlinear and dynamic relationships between data by establishing an Attention-GRU internal model;

[0088] Step 5, prediction results: update the model every 3 to 5 hours to achieve online prediction of circulating air volume;

[0089] Step 6, direct control: convert the predicted result of the circulation air volume into a given value of the circulation fan, and write it into the UaExpert software to realize the control of the circulation fan.

[0090] Further, the step 2 specifically includes the following steps:

[0091] Step 2.1, take the production process parameters with missing data as output, and other parameters without missing data as input, select the data with complete features to establish a regression model; then bring the input into the model to predict the output missing data, and complete the filling of missing data;

[0092] Step 2.2, a box plot is performed to visualize the parameters of the whole process in the historical database used to train the model, and data points above the upper whisker and below the lower whisker are considered as abnormal points and are removed;

[0093] In step 2.3, the Z-Score standardization method is used to normalize the full-process process data after outlier processing and missing value supplementation.

[0094] Further, the step 3 specifically includes the following steps:

[0095] Step 3.1, use the Pearson correlation coefficient to obtain the characteristics linearly related to the circulating air volume, as shown in the following formula:

[0096]

[0097] Where x represents the variable in the whole process, and y represents the circulating air volume;

[0098] Step 3.2, use the Spearman correlation coefficient to obtain the characteristics of nonlinear correlation with circulating air volume, as shown in the following formula:

[0099]

[0100] Where x represents the variable in the whole process, and y represents the circulating air volume.

[0101] Furthermore, the external model in step 4 specifically includes the following steps:

[0102] Step 4.1, establish the covariance objective function of the feature and the circulating air volume in the latent variable space:

[0103]

[0104] in, is the circulating air volume, is the selected feature variable, q and w are the projection vectors of Y and X respectively, u=Yq and t=Xw are the latent variables of Y and X respectively; T represents transposition;

[0105] Step 4.2: Use the Attention-GRU model to establish a dynamic nonlinear internal model between latent variables u and t. The estimated value of the internal model is:

[0106]

[0107] Among them, the G(·) function is the prediction model established by Attention-GRU;

[0108] Step 4.3: After extracting the latent variable with the greatest correlation between the feature and the circulating air volume, continue to extract the next most important latent variable from the remaining information, and repeat steps 4.1 to 4.3 until enough latent variables are extracted:

[0109]

[0110] Where p = XT t / t T t and q=Y T u / u T u is the load vector; b represents the regression coefficient of u and t;

[0111] Step 4.4, the characteristics and circulating air volume are finally decomposed into:

[0112]

[0113] in B=diag[b1,b2,…,b L ], E and F are the residual matrices of X and Y respectively; represents the real number space;

[0114] Step 4.5, the regression model established:

[0115] Y=XC=XW(P T W) -1 B (7)

[0116] C represents the regression coefficients of X and Y, and W represents the projection matrix of X in the latent variable space.

[0117] Step 4.6, the final prediction model established is:

[0118]

[0119] in, Represents the prediction results of the model; Denotes the variable being predicted.

[0120] Furthermore, the internal model in step 4 is specifically:

[0121] According to the attention mechanism module, the influence coefficient of the variables at the historical moment on the variables at the current moment is obtained, and the influence coefficient is used as the weight α of the GRU-RNN model input t ;

[0122] The input sequence of the GRU-RNN module is {x1,x2,x3,…,x Q}, the hidden state is {h1,h2,h3,…,h Q}, the output layer of the GRU-RNN model at time step s is:

[0123] Y s =h s W hq +b q (9)

[0124] Among them, h t is the hidden state at time s, Whq Represents the weight matrix; the calculation formula for time step s is:

[0125]

[0126] Among them, h s-1 is the hidden state at time s-1, ⊙ represents element multiplication, represents the candidate hidden state; z s The update gate is used to update the memory and decide how much past information to pass to the future; its expression is:

[0127] z s =σ(W z x s +U z h s-1 ) (11)

[0128] Among them, σ is the logistic sigmoid function, U z and W z represents the weight matrix; candidate hidden state Represents the current memory content, calculated by the following formula:

[0129]

[0130] Among them, r s It is a reset gate, which discards historical information irrelevant to prediction and combines the new input information with the previous memory. The calculation formula is:

[0131] r s =σ(W r x s +U r h s-1 ) (13).

[0132] Example 3

[0133] according to Figure 1 The schematic diagram of the online prediction method of CDQ circulating air volume based on AGRU-PLS model is used to test the model. The AGRU-PLS model training is established using the selected 15 features (circulating air volume given, coke discharge volume, coke discharge temperature, boiler inlet temperature, boiler outlet temperature, CDQ furnace inlet temperature, fan outlet pressure, fan inlet pressure, CDQ furnace inlet pressure, boiler feed water pump bearing temperature, boiler feed water pump flow, circulating fan current, boiler feed water flow, boiler outlet superheated steam flow, dust removal fan vibration). The prediction results are as follows Figure 4 As shown in the figure, the blue curve is the actual value of the circulating air volume, the orange curve is the predicted value of the circulating air volume, and the mean square error mse is 0.0212. The LSTM model is established using the selected 15 features for comparison. The prediction results are as follows Figure 5 As shown in the figure, the blue curve is the actual value of the circulating air volume, the orange curve is the predicted value of the circulating air volume, and the mean square error mse is 0.1053. From the results, it can be seen that the effect of the AGRU-PLS model established by the present invention is significant.

[0134] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in the field. Although the illustrative specific embodiments of the present invention are described above to facilitate the understanding of the present invention by the technical personnel in the field, it should be clear that the present invention is not limited to the scope of the specific embodiments. For the ordinary technical personnel in the field, as long as various changes are within the spirit and scope of the present invention defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. An online prediction method for CDQ cycle air volume based on AGRU-PLS model, characterized by: The following steps are involved: Step 1, data collection and database construction: collect the whole process data of CDQ from the DCS system, build the corresponding information database, and store the collected data in the database; Step 2, data preprocessing: select a period of historical data of the full-process production process, clean the data, remove abnormal data, fill in missing data, and finally standardize the data; Step 3, feature engineering: perform feature processing on the data through correlation analysis to find out the variable features with high correlation with the circulating air volume in the entire process; Step 4: Establish an AGRU-PLS model: including an external model and an internal model; establish a maximum covariance model of features and circulating air volume through the external model, and solve the collinearity problem of data by iteratively solving latent variables; and explore the nonlinear and dynamic relationships between data by establishing an Attention-GRU internal model; Step 5, prediction results: update the model every 3 to 5 hours to achieve online prediction of circulating air volume; Step 6, direct control: convert the predicted result of the circulation air volume into a given value of the circulation fan, and write it into the UaExpert software to realize the control of the circulation fan.

2. The online prediction method for CDQ cycle air volume based on AGRU-PLS model according to claim 1 is characterized by: The step 2 specifically includes the following steps: Step 2.1, take the production process parameters with missing data as output, and other parameters without missing data as input, select the data with complete features to establish a regression model; then bring the input into the model to predict the output missing data, and complete the filling of missing data; Step 2.2, a box plot is performed to visualize the parameters of the whole process in the historical database used to train the model, and data points above the upper whisker and below the lower whisker are considered as abnormal points and are removed; In step 2.3, the Z-Score standardization method is used to normalize the full-process process data after outlier processing and missing value supplementation.

3. The online prediction method for CDQ cycle air volume based on AGRU-PLS model according to claim 1 is characterized in that: The step 3 specifically includes the following steps: Step 3.1, use the Pearson correlation coefficient to obtain the characteristics linearly related to the circulating air volume, as shown in the following formula: Where x represents the variable in the whole process, is the corresponding mean; y represents the circulating air volume, is the corresponding mean, n is the number of samples; Step 3.2, use the Spearman correlation coefficient to obtain the characteristics of nonlinear correlation with circulating air volume, as shown in the following formula: Where x represents the variable in the whole process, y represents the circulating air volume, R(x i ) and R(y i ) are the positions of the i-th x and y, respectively. and are the average ranks of x and y respectively.

4. The online prediction method for CDQ cycle air volume based on AGRU-PLS model according to claim 1 is characterized in that: The external model in step 4 specifically includes the following steps: Step 4.1, establish the covariance objective function of the feature and the circulating air volume in the latent variable space: st ||w||=1 ||q||=1 in, is the circulating air volume, is the selected feature variable, q and w are the projection vectors of Y and X respectively, u=Yq and t=Xw are the latent variables of Y and X respectively; T represents transposition; Step 4.2: Use the Attention-GRU model to establish a dynamic nonlinear internal model between latent variables u and t. The estimated value of the internal model is: Among them, the G(·) function is the prediction model established by Attention-GRU; Step 4.3: After extracting the latent variable with the greatest correlation between the feature and the circulating air volume, continue to extract the next most important latent variable from the remaining information, and repeat steps 4.1 to 4.3 until enough latent variables are extracted: u:=u-btq T Where p = X T t / t T t and q=Y T u / u T u is the load vector; b represents the regression coefficient of u and t; Step 4.4, the characteristics and circulating air volume are finally decomposed into: X=TP T +E Y=TB+F in B=diag[b1,b2,…,b L ], E and F are the residual matrices of X and Y respectively; represents the real number space; Step 4.5, the regression model established: Y=XC=XW(P T W) -1 B C represents the regression coefficients of X and Y, and W represents the projection matrix of X in the latent variable space; Step 4.6, the final prediction model established is: in, Represents the prediction results of the model; Represents the variable being predicted.

5. The online prediction method for CDQ cycle air volume based on AGRU-PLS model according to claim 1 is characterized in that: The internal model in step 4 is specifically: According to the attention mechanism module, the influence coefficient of the variables at the historical moment on the variables at the current moment is obtained, and the influence coefficient is used as the weight α of the GRU-RNN model input s ; The input sequence of the GRU-RNN module is {x1,x2,x3,…,x Q }, the hidden state is {h1,h2,h3,…,h Q }, the output layer of the GRU-RNN model at time step s is: Y s =h s W hq +b q Among them, h t is the hidden state at time s, W hq Represents the weight matrix; the calculation formula for time step s is: Among them, h s-1 is the hidden state at time s-1, ⊙ represents element multiplication, represents the candidate hidden state; z s The update gate is used to update the memory and decide how much past information to pass to the future; its expression is: z s =σ(W z x s +U z h s-1 ) Among them, σ is the logistic sigmoid function, U z and W z represents the weight matrix; candidate hidden state Represents the current memory content, calculated by the following formula: Among them, r s It is a reset gate, which discards historical information irrelevant to prediction and combines the new input information with the previous memory. The calculation formula is: r s =σ(W r x s +U r h s-1 ); Among them, W r and U r represents the weight matrix.

6. The online prediction method for CDQ cycle air volume based on AGRU-PLS model according to claim 1 is characterized by: The variable characteristics include the given circulating air volume, the coke discharge volume, the coke discharge temperature, the boiler inlet temperature, the boiler outlet temperature, the CDQ furnace inlet temperature, the fan outlet pressure, the fan inlet pressure, the CDQ furnace inlet pressure, the boiler feed water pump bearing temperature, the boiler feed water pump flow, the circulating fan current, the boiler feed water flow, the boiler outlet superheated steam flow, and the dust removal fan vibration.

7. An online prediction model for CDQ cycle air volume based on AGRU-PLS model, characterized by: It includes data acquisition and database storage module, data preprocessing module, data feature extraction module, model training module, online control variable prediction module and control module; The data collection and database storage module collects the full process data of the dry coke quenching production process from the OPC server through the Java language and stores the data in the MySQL database; The data preprocessing module processes outliers, supplements missing data, and normalizes data zscores in the database; The data feature extraction module is to select the variable features with high correlation with the circulating air volume from the collected CDQ process variables, including linear and nonlinear relationships; The variable features include: given circulating air volume, coke discharge volume, coke discharge temperature, boiler inlet temperature, boiler outlet temperature, CDQ furnace inlet temperature, fan outlet pressure, fan inlet pressure, CDQ furnace inlet pressure, boiler feed water pump bearing temperature, boiler feed water pump flow, circulating fan current, boiler feed water flow, boiler outlet superheated steam flow, dust removal fan vibration, etc. 15 features; The model training module is used to establish the AGRU-PLS model, explore the dynamic nonlinear relationship between the extracted features and the circulating air volume, and realize the prediction of the circulating air volume. The control module is to write the predicted results into UaExpert, and then write the results into the given node of the corresponding circulating fan of PLC through UaExpert.

Images