CN111612212A - An online optimization model updating method for pulverized coal fineness of coal mill - Google Patents

Abstract

The invention relates to an on-line optimization model updating method for the fineness of pulverized coal of a coal mill. The method of the invention first collects the prediction data of the original model of the coal mill, and judges whether the original model needs to be updated, and then selects the data that meets the requirements of the model update condition from the original model prediction database, and combines the judgment data to form the model update sample data, On this basis, a data-based silent algorithm is used to establish a new model of coal mill operation optimization to achieve model update. The method ensures the prediction accuracy and generalization ability of the model through data selection, and has high reliability and feasibility.

Description

An online optimization model updating method for pulverized coal fineness of coal mill

technical field

The invention belongs to the technical field of information control, relates to machine learning self-adaptive technology, and in particular relates to an online optimization model updating method of pulverized coal fineness of a coal mill.

Background technique

The optimization of coal mill operation is an important technical means to control the combustion and power consumption of boilers. Boiler combustion and coal mill power consumption are in good condition to maximize production benefits. Different operating parameters such as coal supply, coal quality parameters and air supply conditions have a direct impact on the distribution of pulverized coal fineness. In the case of disturbance, the fineness distribution of pulverized coal is more unstable. For certain production conditions and product requirements, for the ideal operating state required by the coal mill, there is an optimal configuration scheme for each operating parameter, which can optimize the characteristic indicators of the corresponding combustion state. However, the fineness of pulverized coal There is a very complex coupling relationship between the distribution and the operating parameters of each coal mill, and it is not easy to find the optimal configuration of the operating parameters of each burner.

Through data mining, in a large number of different production and operation parameter combinations, the machine learning method is applied to excavate the relationship model between the operating parameters of each coal mill and the fineness distribution of pulverized coal, and then combined with the optimization algorithm to run the coal mill. Optimization is a very promising approach.

Since the operating characteristics of coal mill equipment will change with time or due to maintenance, that is, the relationship between the distribution of coal fineness and the operating parameters of each coal mill is time-varying, how to ensure that the model can be quickly and efficiently Updating to accommodate new situations becomes the key issue with this approach. This problem has a lot to do with modeling methods, sample data selection and update strategies.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose a model updating method that takes into account historical data and new changes in view of the bottleneck problem in the operation optimization of coal mills, which can effectively make the model follow the distribution of coal fineness and the operating parameters of each coal mill. updated as the relationship changes.

The steps of the method of the present invention include:

Step (1) Establish the prediction database of the original model. Collect various operating parameters and corresponding pulverized coal fineness measurement data and prediction data of the original model during the production process of the coal mill, and establish the prediction database of the original model; the specific operating parameters include: industrial analysis data of coal quality, Coal volume, coal mill inlet air volume and inlet air temperature, coal mill current, separator speed. The operating parameters of the coal mill can be obtained through the data monitoring and control system in the production process of the coal mill, or directly through the measurement of samples collected by instruments and equipment. The corresponding fineness of pulverized coal can be obtained by collecting samples for analysis and measurement, which is an industry standard technology. The corresponding prediction data of the pulverized coal fineness of the original model is obtained through the prediction of the original pulverized coal fineness prediction model by using the operating parameters of the pulverizer as the input vector. Store the above data in the prediction database of the original model for model update.

Step (2) Determine the settings of the original model that need to be updated. Set the model's prediction allowable error limit θ, which can be set according to the specific equipment and process operation conditions, generally recommended to take 3%; set the number of times m that the continuous prediction error of the model exceeds the allowable error, which can be collected according to the coal fineness data The difficulty and period setting are generally recommended to be 3; when the model prediction error exceeds the allowable error limit θ for m consecutive times, it is determined that the model needs to be updated, and the data of m times is entered into the prediction database of the original model.

Step (3) Data selection. When the model needs to be updated, select the data in the original model prediction database that meets the following conditions as the model update data: 1. The Euclidean distance between the coal mill operating parameter vector and the m coal mill operating parameter vectors in step (2). They are greater than the set value L respectively. L can be set according to the data in the database and the distribution of operating conditions. Generally, it is recommended to be 2; T time, T can be set according to the data collection time and quantity distribution in the database, and it is generally recommended to be within half a year. According to the above conditions, no less than 30 sets of data are selected. If there are less than 30 sets of data that meet the above conditions, the time limit of condition 2 can be appropriately relaxed, or new data can be collected to supplement the 30 sets of data that meet the conditions. The above 30 sets of data plus the m data in step (2) constitute model update data.

其中x_i表示第i组作为输入数据的磨煤机运行参数，包括：煤质的工业分析数据、给煤量，磨煤机进风量和进风温度，磨煤机电流量，分离器转速。y_i表示第i组作为输出参数的磨煤机煤粉细度，N为样本数量，在训练样本数据基础上，建立新的磨煤机运行参数与对应煤粉细度间的模型。Step (4) Update the model. Taking the data selected in step (3) as a sample, a data-based modeling algorithm is used for modeling, for example, a support vector machine algorithm, a neural network algorithm, etc., the input parameters and output parameters of the samples used for modeling are expressed as:

Where x _i represents the coal mill operating parameters of the i group as input data, including: industrial analysis data of coal quality, coal feed, coal mill inlet air volume and inlet air temperature, coal mill electromechanical flow, and separator speed. y _i represents the pulverized coal fineness of the pulverizer in the i-th group as the output parameter, and N is the number of samples. Based on the training sample data, a new model between the operating parameters of the pulverizer and the corresponding pulverized coal fineness is established.

In the case of modeling data samples, it is a mature and popular way to use data modeling method to establish a data-based prediction model, which will not be repeated here, and the prediction error of the established model should be controlled within 2%. Realize the update of the online optimization model update of the pulverized coal fineness of the coal mill.

The model updating method proposed by the invention makes full use of the existing historical data, greatly reduces the workload of model updating, improves the efficiency of model updating, satisfies the actual requirements of the operation optimization of the coal mill, and ensures the operation optimization of the coal mill. real-time and accurate.

Detailed ways

An online optimization model updating method for pulverized coal fineness of a coal mill, the specific steps are:

(1) Establish the prediction database of the original model. Collect various operating parameters in the production process of the coal mill, the corresponding coal fineness measurement data and the prediction data of the original model, and establish the prediction database of the original model; the specific operating parameters include: industrial analysis data of coal quality, Coal volume, coal mill inlet air volume and inlet air temperature, coal mill current, separator speed. The operating parameters of the coal mill can be obtained through the data monitoring and control system in the production process of the coal mill, or directly through the measurement of samples collected by instruments and equipment. The corresponding fineness of pulverized coal can be obtained by collecting samples for analysis and measurement, which is an industry standard technology. The corresponding prediction data of the pulverized coal fineness of the original model is obtained through the prediction of the original pulverized coal fineness prediction model by using the operating parameters of the pulverizer as the input vector. Store the above data in the prediction database of the original model for model update.

(2) Determine the settings that the original model needs to be updated. Set the model's prediction allowable error limit θ, which can be set according to the specific equipment and process operation conditions, and is generally recommended to be 3%; set the number of times m that the model's continuous prediction error exceeds the allowable error, which can be collected according to the coal fineness data. The difficulty and cycle setting are generally recommended to be 3; when the model prediction error exceeds the allowable error limit θ for m consecutive times, it is determined that the model needs to be updated, and the m times of data are entered into the prediction database of the original model.

(3) Data selection. When the model needs to be updated, select the data in the original model prediction database that meets the following conditions as the model update data: 1. The Euclidean distance between the coal mill operating parameter vector and the m coal mill operating parameter vectors in step (2). are greater than the set value L, L can be set according to the data in the database and the distribution of operating conditions, and is generally recommended to be 2; 2, the data collection time is no more than the earliest time point in the m groups of data in step (2). T time, T can be set according to the data collection time and quantity distribution in the database, and it is generally recommended to be within half a year. Select no less than 30 sets of data according to the above conditions. If there are less than 30 sets of data that meet the above conditions, the time limit of Condition 2 can be appropriately relaxed, or new data can be collected to supplement the 30 sets of data that meet the conditions. The above 30 sets of data plus the m data in step (2) constitute model update data.

其中x_i表示第i组作为输入数据的磨煤机运行参数，包括：煤质的工业分析数据、给煤量，磨煤机进风量和进风温度，磨煤机电流量，分离器转速。y_i表示第i组作为输出参数的磨煤机煤粉细度，N为样本数量，在训练样本数据基础上，建立新的磨煤机运行参数与对应煤粉细度间的模型。(4) Update the model. Taking the data selected in step (3) as a sample, a data-based modeling algorithm is used for modeling, for example, a support vector machine algorithm, a neural network algorithm, etc., the input parameters and output parameters of the samples used for modeling are expressed as:

Where x _i represents the coal mill operating parameters of the i group as input data, including: industrial analysis data of coal quality, coal feed, coal mill inlet air volume and inlet air temperature, coal mill electromechanical flow, and separator speed. y _i represents the pulverized coal fineness of the pulverizer in the i-th group as the output parameter, and N is the number of samples. Based on the training sample data, a new model between the operating parameters of the pulverizer and the corresponding pulverized coal fineness is established.

In the case of modeling data samples, it is a mature and popular way to use data modeling method to establish a data-based prediction model, which will not be repeated here, and the prediction error of the established model should be controlled within 2%. Realize the update of the online optimization model update of the pulverized coal fineness of the coal mill.

Claims (6)

1. An on-line optimization model updating method for coal powder fineness of a coal mill is characterized by comprising the following steps:

step (1), establishing a prediction database of an original model

Collecting each operation parameter, corresponding coal powder fineness measurement data and prediction data of an original model in the production process of the coal mill, and establishing a prediction database of the original model;

step (2) determining the settings of the original model that need to be updated

Setting a prediction allowable error limit theta of the model, and setting the number m of times that the model continuous prediction error exceeds the allowable error; when the prediction error of the model exceeds the allowable error limit theta for m times continuously, judging that the model needs to be updated, and recording the data of m times into a prediction database of the original model;

step (3) data selection

When the model needs to be updated, selecting data meeting the following conditions in the original model prediction database as model updating data:

firstly, enabling Euclidean distances between the coal mill operation parameter vectors and the m coal mill operation parameter vectors in the step (2) to be respectively larger than a set value L;

secondly, the data acquisition time is not more than T time from the earliest time point in the m groups of data in the step (2);

selecting no less than 30 groups of data according to the conditions, if the data meeting the conditions is less than 30 groups, relaxing the time limit of the condition II, or collecting new data to complement the 30 groups of data meeting the conditions; adding the 30 groups of data to the m data in the step (2) to form model updating data;

step (4) updating the model

Taking the data selected in the step (3) as a sample, modeling by adopting a data-based modeling algorithm, and expressing input parameters and output parameters of the sample for modeling as

Wherein x_iRepresenting the ith group of coal mill operating parameters as input data, comprising: the method comprises the following steps of (1) industrial analysis data of coal quality, coal feeding quantity, coal mill air inlet quantity and air inlet temperature, coal mill current quantity and separator rotating speed; y is_iRepresenting the coal powder fineness of the coal mill of the ith group as an output parameter, wherein N is the number of samples, and establishing a model between a new coal mill operation parameter and the corresponding coal powder fineness on the basis of training sample data;

under the condition of modeling data samples, the prediction error of a model established by establishing a prediction model based on data by using a data modeling method is controlled within 2 percent; and updating of the on-line optimization model of the coal powder fineness of the coal mill is realized.

2. The on-line optimization model updating method for coal pulverizer coal powder fineness according to claim 1, characterized by comprising the following steps: the specific operation parameters in the first step comprise: the method comprises the following steps of (1) industrial analysis data of coal quality, coal feeding quantity, coal mill air inlet quantity and air inlet temperature, coal mill current and separator rotating speed; the coal mill operation parameters are obtained through a data monitoring control system in the production process of the coal mill or directly obtained through sample collection and measurement of instrument equipment.

3. The on-line optimization model updating method for coal pulverizer coal powder fineness according to claim 1, characterized by comprising the following steps: the corresponding coal powder fineness in the step one is obtained by collecting samples, analyzing and measuring, and the technology is an industry standard technology.

4. The on-line optimization model updating method for coal pulverizer coal powder fineness according to claim 1, characterized by comprising the following steps: and (4) corresponding coal powder fineness raw model prediction data in the step one is obtained by taking the coal mill operation parameters as input vectors and predicting the coal powder fineness prediction model of the original coal mill.

5. The on-line optimization model updating method for coal pulverizer coal powder fineness according to claim 1, characterized by comprising the following steps: and in the second step, the prediction allowable error limit of the model is 3%, and the number of times that the model continuous prediction error exceeds the allowable error is 3.

6. The on-line optimization model updating method for coal pulverizer coal powder fineness according to claim 1, characterized by comprising the following steps: and in the fourth step, modeling is carried out by adopting a data-based modeling algorithm, wherein the modeling algorithm is a support vector machine algorithm and a neural network algorithm.