Intelligent adjusting and optimizing method for high-pressure roller mill based on neural network

Abstract

A neural network-based intelligent adjustment and optimization method for a high-pressure roller mill comprises the following steps: respectively obtaining a long-term reference model, a short-term prediction model and an intelligent adjustment and optimization model by using operation data of the high-pressure roller mill, and predicting a future operation state of the high-pressure roller mill through the short-term prediction model. According to whether the data fluctuation range is in a normal interval or not, automatic adjustment and optimization are carried out, when adjustment and optimization need to be carried out, the correlation coefficient of the controlled variable and the operation variable is obtained through the intelligent adjustment and optimization model, so that the operation variable with the maximum correlation is selected, and debugging of the high-pressure roller mill is carried out by adjusting the operation variable; and ending the tuning process until the predicted fluctuation range does not exceed the normal fluctuation range. The method can cope with the complex operation environment of the high-pressure roller mill, and can continuously and efficiently carry out automatic adjustment and optimization of the high-pressure roller mill.

Description

Technical Field

[0001] The invention relates to the field of intelligent control methods for high pressure roller mills, and in particular to an intelligent tuning method for high pressure roller mills based on a neural network. Background Art

[0002] With the development of industrial automation, the management of high-pressure grinding rolls is gradually developing towards intelligence. The traditional maintenance and tuning methods of high-pressure grinding rolls mainly rely on manual experience and fixed rules, lack of real-time and adaptability, and are difficult to cope with complex equipment operating environments. In recent years, neural network technology has performed well in processing large-scale data and complex pattern recognition. Therefore, applying neural networks to the intelligent tuning of high-pressure grinding rolls has become a promising research direction. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent tuning method for a high pressure roller mill based on a neural network, which utilizes the operating data of the high pressure roller mill, predicts the future operating state of the high pressure roller mill through a neural network model, and performs automatic tuning based on the prediction results.

[0004] The technical solution adopted by the present invention to solve the above technical problems is: an intelligent tuning method for a high pressure roller mill based on a neural network, comprising the following steps: Step 1: Build a long-term benchmark model Once every six months, obtain the operation data and production data of the high pressure roller grinding mill in the previous 1-2 years, pre-process the data, and divide the pre-processed data into production variables, operating variables, controlled variables and interference variables; The production variables, operating variables and interference variables are used as the input of the model, and the controlled variables are used as the output of the model to train the initial model of the neural network. After reaching the preset number of iterations, the trained model is stored as a long-term benchmark model. Step 2: Build a short-term forecasting model Once a month, obtain the operation data and production data of the high pressure roller grinding mill in the previous month, pre-process the data, and divide the pre-processed data into production variables, operating variables, controlled variables and interference variables; perform statistics on the controlled variables and obtain the normal fluctuation range of each controlled variable; The production variables, operating variables and interference variables are used as the input of the model, and the controlled variables are used as the output of the model. The long-term benchmark model stored after the latest training is trained again. After reaching the preset number of iterations, the trained model is stored as a short-term prediction model. Step 3: Establish an intelligent tuning model Obtain historical operation data and production data of the high pressure roller grinding mill, pre-process the data, and divide the pre-processed data into production variables, operating variables, controlled variables, and interference variables; The operating variables and controlled variables are used as data sets, and the initial neural network model is trained using the random forest algorithm and the bagging algorithm. After reaching the preset number of iterations, the trained model is stored as an intelligent tuning model, and the correlation coefficients of each controlled variable and all the operating variables are obtained from the intelligent tuning model. Step 4: High Pressure Roller Grinding Machine Tuning During the operation of the high pressure roller mill, the real-time operation data and production data of the high pressure roller mill are read multiple times according to preset time intervals, and the data are pre-processed, and the pre-processed data are divided into production variables, operating variables, controlled variables and interference variables; Input the production variables and operation variables into the short-term prediction model stored after the latest training, output the predicted value of the controlled variable in the future through the short-term prediction model, obtain the predicted fluctuation range of the controlled variable in the future according to the predicted value of the controlled variable, and then compare the predicted fluctuation range with the normal fluctuation range obtained in step 2; When the predicted fluctuation range does not exceed the normal fluctuation range, the prediction process ends; When the predicted fluctuation range of one or more controlled variables exceeds the normal fluctuation range, the intelligent tuning model of the controlled variables is called to compare all the correlation coefficients of the controlled variables whose predicted fluctuation range exceeds the normal fluctuation range, select the operating variable with the greatest correlation with the controlled variable, and then adjust the operating variable according to the preset debugging step; After debugging is completed, the predicted fluctuation range of the controlled variable is obtained again. When the predicted fluctuation range exceeds the normal fluctuation range, debugging is performed again until the predicted fluctuation range does not exceed the normal fluctuation range, and the tuning process is completed.

[0005] Preferably, the preprocessing includes data alignment, removal of null values, removal of duplicate values ​​and data downsampling.

[0006] According to the above technical solution, the beneficial effects of the present invention are: The present invention utilizes the operating data of the high-pressure roller mill to obtain a regularly updated long-term benchmark model and a short-term prediction model, respectively. The future operating state of the high-pressure roller mill is predicted by the short-term prediction model, and automatic optimization is performed according to whether the data fluctuation range is within the normal range. When optimization is required, the correlation coefficient between the controlled variable and the operating variable is obtained by the latest trained optimization model, so as to select the operating variable with the largest correlation. The high-pressure roller mill is debugged by adjusting the operating variable until the predicted fluctuation range does not exceed the normal fluctuation range, and the optimization process ends. The present invention can cope with the complex operating environment of the high-pressure roller mill and can continuously and efficiently perform automatic optimization of the high-pressure roller mill. DETAILED DESCRIPTION

[0007] An intelligent tuning method for a high pressure roller grinding mill based on a neural network comprises the following steps: Step 1: Build a long-term benchmark model Once every six months, the operating data and production data of the high-pressure roller grinding mill in the previous 1-2 years are obtained, and the data are preprocessed. The preprocessing includes data alignment, removal of null values, removal of duplicate values ​​and data downsampling. The preprocessed data are then divided into production variables, operating variables, controlled variables and interference variables.

[0008] The production variables, operating variables and interference variables are used as the input of the model, and the controlled variables are used as the output of the model. The initial model of the neural network is trained. After reaching the preset number of iterations, the trained model is stored as a long-term benchmark model.

[0009] Step 2: Build a short-term forecasting model Once a month, the operation data and production data of the high-pressure roller mill in the previous month are obtained, and the data are preprocessed, including data alignment, removal of null values, removal of duplicate values ​​and data downsampling. The preprocessed data are then divided into production variables, operating variables, controlled variables and interference variables. The controlled variables are then statistically analyzed to obtain the normal fluctuation range of each controlled variable, such as ±3σ, ±1.5IQR, etc.

[0010] Taking production variables, operating variables and interference variables as the input of the model and controlled variables as the output of the model, the long-term benchmark model stored after the latest training is trained again. After reaching the preset number of iterations, the trained model is stored as a short-term prediction model.

[0011] Step 3: Establish an intelligent tuning model The historical operation data and production data of the high pressure roller grinding mill are obtained, and the data are preprocessed, and the preprocessed data are divided into production variables, operating variables, controlled variables and interference variables.

[0012] With the operating variables and controlled variables as data sets, the random forest algorithm and bagging algorithm are used to train the initial model of the neural network. After reaching the preset number of iterations, the trained model is stored as an intelligent tuning model, and the correlation coefficients of each controlled variable and all the operating variables are obtained from the intelligent tuning model.

[0013] Step 4: High Pressure Roller Grinding Machine Tuning During the operation of the high pressure roller mill, the real-time operation data and production data of the high pressure roller mill are read multiple times according to preset time intervals, and the data are preprocessed to divide the preprocessed data into production variables, operating variables, controlled variables and interference variables.

[0014] Input the production variables and operating variables into the short-term prediction model stored after the latest training. The short-term prediction model outputs the predicted value of the controlled variable for a period of time in the future, such as 5-30 minutes. According to the predicted value of the controlled variable, the predicted fluctuation range of the controlled variable in the next 5-30 minutes is obtained, and then the predicted fluctuation range is compared with the normal fluctuation range obtained in step 2.

[0015] When the predicted fluctuation range does not exceed the normal fluctuation range, the prediction process ends.

[0016] When the predicted fluctuation range of one or more controlled variables exceeds the normal fluctuation range, the intelligent tuning model of the controlled variables is called to compare all the correlation coefficients of the controlled variables whose predicted fluctuation range exceeds the normal fluctuation range, and select the operating variable with the greatest correlation with the controlled variable, such as: motor stator temperature = a1*current + a2*pressure + a3*opening size, where a1, a2, and a3 are all correlation coefficients. By comparing the numerical values ​​of a1, a2, and a3, the operating variable corresponding to the largest value is selected, and then the operating variable is adjusted according to the preset debugging step.

[0017] After debugging is completed, the predicted fluctuation range of the controlled variable is obtained again. When the predicted fluctuation range exceeds the normal fluctuation range, debugging is performed again until the predicted fluctuation range does not exceed the normal fluctuation range, and the tuning process is completed.

[0018] After the automatic tuning process is completed, the value of the debugging parameter can be sent to the on-site operator as a recommended value, and the operator will decide whether to perform the operation.

Claims

1. An intelligent tuning method for a high pressure roller mill based on a neural network, characterized in that: The following steps are involved: Step 1: Build a long-term benchmark model Once every six months, obtain the operation data and production data of the high pressure roller grinding mill in the previous 1-2 years, pre-process the data, and divide the pre-processed data into production variables, operating variables, controlled variables and interference variables; The production variables, operating variables and interference variables are used as the input of the model, and the controlled variables are used as the output of the model to train the initial model of the neural network. After reaching the preset number of iterations, the trained model is stored as a long-term benchmark model. Step 2: Build a short-term forecasting model Once a month, obtain the operation data and production data of the high pressure roller grinding mill in the previous month, pre-process the data, and divide the pre-processed data into production variables, operating variables, controlled variables and interference variables; Conduct statistics on the controlled variables and obtain the normal fluctuation range of each controlled variable; The production variables, operating variables and interference variables are used as the input of the model, and the controlled variables are used as the output of the model. The long-term benchmark model stored after the latest training is trained again. After reaching the preset number of iterations, the trained model is stored as a short-term prediction model. Step 3: Establish an intelligent tuning model Obtain historical operation data and production data of the high pressure roller grinding mill, pre-process the data, and divide the pre-processed data into production variables, operating variables, controlled variables, and interference variables; The operating variables and controlled variables are used as data sets, and the initial neural network model is trained using the random forest algorithm and the bagging algorithm. After reaching the preset number of iterations, the trained model is stored as an intelligent tuning model, and the correlation coefficients of each controlled variable and all the operating variables are obtained from the intelligent tuning model. Step 4: High Pressure Roller Grinding Machine Tuning During the operation of the high pressure roller mill, the real-time operation data and production data of the high pressure roller mill are read multiple times according to preset time intervals, and the data are pre-processed, and the pre-processed data are divided into production variables, operating variables, controlled variables and interference variables; Input the production variables and operation variables into the short-term prediction model stored after the latest training, output the predicted value of the controlled variable in the future through the short-term prediction model, obtain the predicted fluctuation range of the controlled variable in the future according to the predicted value of the controlled variable, and then compare the predicted fluctuation range with the normal fluctuation range obtained in step 2; When the predicted fluctuation range does not exceed the normal fluctuation range, the prediction process ends; When the predicted fluctuation range of one or more controlled variables exceeds the normal fluctuation range, the intelligent tuning model of the controlled variables is called to compare all the correlation coefficients of the controlled variables whose predicted fluctuation range exceeds the normal fluctuation range, select the operating variable with the greatest correlation with the controlled variable, and then adjust the operating variable according to the preset debugging step; After debugging is completed, the predicted fluctuation range of the controlled variable is obtained again. When the predicted fluctuation range exceeds the normal fluctuation range, debugging is performed again until the predicted fluctuation range does not exceed the normal fluctuation range, and the tuning process is completed.

2. The intelligent tuning method of a high pressure roller grinding mill based on a neural network according to claim 1, characterized in that: The preprocessing includes data alignment, null value removal, duplicate value removal and data downsampling.