An intelligent prediction and control method for the dust concentration at the inlet of a dust removal system

Abstract

The present invention discloses an intelligent prediction and control method for the dust concentration at the inlet of a dust removal system in the technical field of dust removal systems, comprising the following steps: S1: Establishing N groups of training samples for a SA-RBF hybrid neural network model predictor; S2: Optimizing the number of hidden layer neurons of the RBF hybrid neural network through a particle swarm parameter optimization algorithm, selecting the parameters of the neural network, and initializing the weights of the neural network; S3: Training the prediction model and the controller with the DPSO training algorithm to obtain the weight parameters of the neural network; S4: Determining the lag time of the system, using the input parameters and the actual output parameters as the initial values of the controller, calculating the prediction error, and adjusting the controller parameters through the error; S5: Calculating the output of the neural network prediction model, using the obtained prediction output as the control output to act on the controlled object, achieving accurate prediction of the dust concentration at the inlet of the dust collector under different working conditions and optimizing the operating parameters, and improving the reliability and economy of the system operation.

Description

Technical Field

[0001] The present invention relates to the technical field of dust removal systems, and particularly to an intelligent prediction and control method for the dust concentration at the inlet of a dust removal system. Background Art

[0002] During the production process of thermal power plants, a large amount of coal is burned, and the coal often contains a large amount of ash. After combustion, soot particles are formed. If not treated, it will cause great harm to the environment. Therefore, controlling and reducing the soot pollutants from coal-fired power plants is crucial for environmental governance.

[0003] Currently, electrostatic dust removal is mainly used in thermal power plants for dust removal. Due to the relatively high dust concentration at the inlet of the electrostatic precipitator and frequent failures of the measuring instruments, many power plants do not install dust measuring devices at the inlet of the electrostatic precipitator, and thus cannot provide real-time feedback on the dust concentration at the inlet of the precipitator.

[0004] Currently, due to the tight coal market, the quality of the coal fed into the furnace of power plants fluctuates greatly. Without accurate dust concentration, when the dust concentration at the inlet of the precipitator increases, due to the untimely control of the electrostatic precipitator, it is easy to cause excessive emissions of soot. Or when the dust concentration at the inlet of the precipitator is low, the voltage and current of the precipitator are controlled too high, resulting in high operating energy consumption and waste of system energy consumption.

[0005] In view of the above-related technologies, the inventor provides an intelligent prediction and control method for the dust concentration at the inlet of a dust removal system. Summary of the Invention

[0006] In order to improve the problem that without accurate dust concentration, when the dust concentration at the inlet of the precipitator increases, due to the untimely control of the electrostatic precipitator, it is easy to cause excessive emissions of soot, or when the dust concentration at the inlet of the precipitator is low, the voltage and current of the precipitator are controlled too high, resulting in high operating energy consumption and waste of system energy consumption, the present invention provides an intelligent prediction and control method for the dust concentration at the inlet of a dust removal system to more accurately adjust the voltage and current of the precipitator, reduce system energy consumption, and avoid excessive emissions of dust concentration.

[0007] The present invention provides an intelligent prediction and control method for the dust concentration at the inlet of a dust removal system, adopting the following technical solutions, including the following steps:

[0008] S1: Establish N groups of training samples for the SA-RBF hybrid neural network model predictor;

[0009] S2: Optimize the number of hidden layer neurons of the RBF hybrid neural network through the particle swarm parameter optimization algorithm, select the parameters of the neural network, and initialize the weights of the neural network;

[0010] S3: Train the prediction model and controller of the SA-RBF hybrid neural network model predictor using the DPSO training algorithm to obtain the neural network weight parameters;

[0011] S4: Determine the lag time of the system, use the input parameters and actual output parameters as the initial values of the controller, calculate the prediction error, and adjust the controller parameters through the error;

[0012] S5: Calculate the output of the established intelligent SOA-SOFNN prediction model for soot concentration, obtain the predicted output of the controlled object at future moments, and use the obtained predicted output as the control output to act on the controlled object.

[0013] Optionally, the training samples in S1 include unit load, main steam flow rate, total air volume, total coal feed amount, primary air fan valve opening, secondary air fan valve opening, coal feed amounts of mills A to E, and secondary air openings of mills A to E.

[0014] Optionally, the principal component analysis method for determining the training samples includes calculating the covariance matrix, then calculating the eigenvalues and eigenvectors of the covariance matrix, sorting the eigenvalues and eigenvectors of the covariance matrix and obtaining the contribution rate, obtaining the load matrix of the eigenvectors with a contribution rate greater than 85% and the original eigenvectors, and reducing the dimension of the input parameters according to the load matrix.

[0015] Optionally, in the particle swarm parameter optimization algorithm in S2, first randomly initialize the positions and velocities of the particles, then define the fitness function, track the individual optimal solution and the global optimal solution, and update their positions and velocities each time. During this process, the fitness value is calculated for each iteration, and after multiple iterations, the set target fitness value is reached, thereby obtaining the optimal solution.

[0016] Optionally, for each sample Xi in each group of training sample sets X in the DPSO training algorithm in S3, train the training samples except Xi to obtain the model Mi; use the sorting improvement method and use the model Mi to calculate the gradient estimate of the sample Xi; use the new model to score the sample Xi again to form a weak learner; perform weighted processing on all weak learners to obtain the final strong classifier.

[0017] Optionally, in S4, based on the extracted input parameters, collect and analyze the relevant operation data of the formation of boiler soot in the coal-fired power unit continuously for 30 days, collect once per minute, organize the unit load, main steam flow rate, total air volume, total coal feed amount, primary air fan valve opening, secondary air fan valve opening, coal feed amounts of mills A to E, and secondary air openings of mills A to E input by each burner as the input parameters of the SOA-SOFNN prediction model, and at the same time use the soot concentration measured at each time point as the output parameter for model training.

[0018] Optionally, in S5, the real-time operation data of the influencing factors of the dust concentration at the dust collector inlet are input into the established intelligent SOA-SOFNN prediction model for dust concentration, and the dust concentration at the dust collector inlet is obtained. The voltage and current of the dust collector are adjusted according to the inlet dust concentration.

[0019] Optionally, the method for adjusting the voltage and current of the dust collector according to the inlet dust concentration is as follows: when the predicted dust concentration is higher than the designed concentration, the voltage and current should be adjusted to the maximum input and output current; when the dust concentration is lower than the designed concentration, the input and output current of the dust collector should be reduced.

[0020] Optionally, the intelligent prediction and control method for the dust concentration at the inlet of the dust removal system adopts an independent external platform and is connected to the DCS through a communication method.

[0021] In summary, the present invention has at least one of the following beneficial effects: The present invention realizes accurate prediction of the dust concentration at the inlet of the dust collector under different working conditions and optimization of operation parameters, improves the reliability and economy of system operation, can effectively improve the stability of the dust removal system, and at the same time improves the automatic operation level and system economy of the dust removal system of thermal power units, and has good technical and application value. Detailed Embodiment

[0022] The following further details the present invention.

[0023] The present invention provides a method for predicting and controlling the dust at the inlet of a dust collector, mainly using a self-organizing fuzzy neural network (SOA-SOFNN) prediction model. The SA-RBF hybrid neural network model needs to be realized by the (SOA-SOFNN) prediction model and the RBF hybrid neural network. This method not only considers the self-organization of the structure but also improves the parameter adjustment method. It has a fast training speed and meets the requirements of online prediction. Its main analysis process is as follows:

[0024] S1: Establish N groups of training samples for the SA-RBF hybrid neural network model predictor;

[0025] According to the combustion principle of coal-fired units and the mechanism of the dust removal system, determine the influencing factors of the dust concentration at the dust collector inlet and the input variables of the (SOA-SOFNN) prediction model; mainly including parameters such as unit load, main steam flow, total air volume, total coal supply, primary fan valve opening, secondary fan valve opening, coal supply of A-E mills, and secondary air opening of A-E mills. Select a common network topology: a three-layer neural network for simulation calculation, with a total of 18 input nodes corresponding to 18 input parameters related to boiler combustion.

[0026] The principal component analysis method for determining the training samples includes calculating the covariance matrix, then calculating the eigenvalues and eigenvectors of the covariance matrix, sorting the eigenvalues and eigenvectors of the covariance matrix and obtaining the contribution rate, obtaining the load matrix of the eigenvectors with a contribution rate greater than 85% and the original eigenvectors, and reducing the dimension of the input parameters according to the load matrix.

[0027] S2: Optimize the number of hidden layer neurons of the RBF hybrid neural network through the particle swarm parameter optimization algorithm, select the parameters of the neural network, and initialize the weights of the neural network;

[0028] The particle swarm parameter optimization algorithm first randomly initializes the positions and velocities of the particles, then defines the fitness function, tracks the individual optimal solution and the global optimal solution, and updates its own position and velocity each time. During this process, the fitness value is calculated for each iteration. After multiple iterations, the set target fitness value is reached, and thus the optimal solution is obtained.

[0029] S3: Train the prediction model and the controller of the SA-RBF hybrid neural network model predictor using the DPSO training algorithm to obtain the weight parameters of the neural network;

[0030] For each sample Xi in each group of training sample sets X in the DPSO training algorithm, train the training samples except Xi to obtain the model Mi; use the sorting boosting method and use the model Mi to calculate the gradient estimate of the sample Xi; use the new model to score the sample Xi again to form a weak learner; perform weighted processing on all weak learners to obtain the final strong classifier.

[0031] S4: Determine the lag time of the system, use the input parameters and the actual output parameters as the initial values of the controller, calculate the prediction error, and adjust the controller parameters through the error;

[0032] Based on the extracted input parameters, relevant operation data on the formation of boiler soot in coal-fired units are collected and analyzed continuously for 30 days, once per minute. Parameters such as the unit load, main steam flow rate, total air volume, total coal feed amount, primary fan valve opening, secondary fan valve opening, coal feed amounts of mills A - E, and secondary air openings of mills A - E input by each burner are sorted out as the input parameters of the SOA-SOFNN prediction model. At the same time, the soot concentration measured at each time point is used as the output parameter for model training.

[0033] S5: Calculate the output of the neural network prediction model to obtain the predicted output of the controlled object at future moments, and use the obtained predicted output as the control output to act on the controlled object.

[0034] Input the real-time operation data of the influencing factors of the dust concentration at the inlet of the dust collector into the established intelligent SOA-SOFNN prediction model for dust concentration to obtain the dust concentration at the dust collector inlet, and adjust the voltage and current of the dust collector according to the inlet dust concentration. The method for adjusting the voltage and current of the dust collector according to the inlet dust concentration is as follows: when the predicted dust concentration is higher than the designed value concentration, the voltage and current should be adjusted to the maximum input value; when the dust concentration is lower than the designed concentration, the input value of the voltage and current should be reduced. Taking the actual operation of a certain power plant as an example, according to the SOA-SOFNN prediction model, the dust concentration at the inlet of the dust collector is 25.2 g / Nm 3 , lower than the designed concentration of 32 g / Nm 3 . At this time, the input voltage and input current of the dust collector should be adjusted from 70 kV and 539.82 mA to 58 kV and 420 mA.

[0035] The intelligent prediction and control method for the dust concentration at the inlet of the dust removal system of the present invention adopts an independent external platform and is connected to the DCS through a communication method, and can accurately and reliably control the dust removal system.

[0036] In the embodiment of the present invention, the method of the present invention is applied to the control of the dust removal system of a certain power plant to verify the effect of the present invention. The deviation between the predicted dust removal system and the actual is 2.3%. The actual test shows that the dust concentration at the inlet of the dust removal system is 25.2 g / Nm 3 , and the predicted dust concentration at the inlet of the dust removal system is 24.6 g / Nm 3 . At this time, the input voltage and input current of the dust collector should be adjusted from 70 kV and 539.82 mA to 58 kV and 420 mA, and the overall energy consumption is reduced by 20%. It can be seen that the method of the present invention can effectively improve the stability of the dust removal system, and at the same time improve the automatic operation level and system economy of the dust removal system of thermal power units.

[0037] The above are all the preferred embodiments of the present invention, and the protection scope of the present invention is not limited by this. Therefore, all equivalent changes made according to the structure, shape, and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. An intelligent prediction and control method for the dust concentration at the inlet of a dust removal system, characterized in that: It includes the following steps: S1: Establish N groups of training samples for the SA-RBF hybrid neural network model predictor; S2: Optimize the number of hidden layer neurons of the RBF hybrid neural network through the particle swarm parameter optimization algorithm, select the parameters of the neural network, and initialize the neural network weights; S3: Use the DPSO training algorithm to train the prediction model and controller of the SA-RBF hybrid neural network model predictor to obtain the neural network weight parameters; S4: Determine the lag time of the system, use the input parameters and actual output parameters as the initial values of the controller, calculate the prediction error, and adjust the controller parameters through the error; S5: Calculate the output of the established intelligent SOA-SOFNN prediction model for the dust concentration, obtain the predicted output at the future moment of the controlled object, and use the obtained predicted output as the control output to act on the controlled object; In the DPSO training algorithm in S3, for each sample Xi in each group of training sample sets X, train the training samples except Xi to obtain the model Mi; use the sorting boosting method and use the model Mi to calculate the gradient estimate of the sample Xi; use the new model to score the sample Xi again to form a weak learner; perform weighted processing on all weak learners to obtain the final strong classifier; In S4, based on the extracted input parameters, continuously collect and analyze the relevant operation data of the dust formation in the coal-fired unit boiler for 30 consecutive days, collect once per minute, and organize the unit load, main steam flow, total air volume, total coal feed, primary fan valve opening, secondary fan valve opening, coal feed of A-E mills, and secondary air opening of A-E mills input by each burner as the input parameters of the SOA-SOFNN prediction model. At the same time, use the measured dust concentration at each time point as the output parameter for model training.

2. An intelligent prediction and control method for the dust concentration at the inlet of a dust removal system according to claim 1, characterized in that: The training samples in S1 include unit load, main steam flow, total air volume, total coal feed, primary fan valve opening, secondary fan valve opening, coal feed of A-E mills, and secondary air opening of A-E mills.

3. An intelligent prediction and control method for the dust concentration at the inlet of a dust removal system according to claim 2, characterized in that: The method for determining the principal component analysis of the training samples includes calculating the covariance matrix, then calculating the eigenvalues and eigenvectors of the covariance matrix, sorting the eigenvalues and eigenvectors of the covariance matrix and calculating the contribution rate, obtaining the load matrix of the eigenvectors with a contribution rate greater than 85% and the original eigenvectors, and reducing the dimension of the input parameters according to the load matrix.

4. An intelligent prediction and control method for the dust concentration at the inlet of a dust removal system according to claim 1, characterized in that: In the particle swarm parameter optimization algorithm in S2, the positions and velocities of the particles are first randomly initialized, and then the fitness function is defined to track the individual optimal solution and the global optimal solution, and update their own positions and velocities each time. During this process, the fitness value is calculated in each iteration, and after multiple iterations, the set target fitness value is reached, thereby obtaining the optimal solution.

5. An intelligent prediction and control method for the inlet dust concentration of a dust removal system according to claim 1, characterized in that: In S5, the real-time operation data of the influencing factors of the dust concentration at the inlet of the dust collector are input into the established intelligent SOA-SOFNN prediction model of the dust concentration to obtain the dust concentration at the inlet of the dust removal, and the voltage and current of the dust collector are adjusted according to the inlet dust concentration.

6. An intelligent prediction and control method for the inlet dust concentration of a dust removal system according to claim 5, characterized in that: The method for adjusting the voltage and current of the dust collector according to the inlet dust concentration is as follows: when the predicted dust concentration is higher than the designed concentration, the voltage and current should be adjusted to the maximum input and output currents; when the dust concentration is lower than the designed concentration, the input and output currents of the dust collector should be reduced.

7. An intelligent prediction and control method for the inlet dust concentration of a dust removal system according to claim 1, characterized in that: The intelligent prediction and control method for the inlet dust concentration of the dust removal system adopts an independent external platform and is connected to the DCS through a communication method.