Prediction method for fineness of pulverized coal of coal mill
By adopting LSSVM and Bayesian optimization methods in coal fineness prediction, the problems of complex calculation and low generalization ability in the existing technology are solved, and efficient and accurate coal fineness prediction is achieved, which improves combustion efficiency and Environmentally friendly effect.
Patent Information
- Application Number
- CN202410986535.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing coal fineness prediction methods are complex in calculations, have poor real-time performance, low model generalization ability, poor parameter search, and difficult to effectively predict coal fineness, which affects combustion efficiency and environmental protection.
Using the coal fineness prediction method based on LSSVM and Bayesian optimization, the least squares support vector machine model is constructed through data preprocessing, and the hyperparameters are optimized using Bayesian optimization algorithm to deal with nonlinear data and high-dimensional optimization problems. , simplify the calculation process and improve prediction accuracy.
Accurate prediction of coal fineness is achieved, detection cost is reduced, generalization ability and convergence speed of prediction model are improved, local optimal solutions are avoided, and combustion efficiency and environmental protection effect are improved.
Smart Images

Figure CN118940132A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of coal grinding, in particular to a method for predicting the fineness of coal powder in a coal grinding mill. Background Art
[0002] Pulverized coal fineness refers to the particle size of pulverized coal after it passes through a pulverizer. It is a key indicator for coal-fired boilers, and its value is related to factors such as the speed of the pulverizer's dynamic separator, the coal feed rate, and the air temperature and volume at the inlet and outlet. As a key indicator of coal combustion in power plants, pulverized coal fineness has a significant impact on improving combustion efficiency, reducing the carbon content of fly ash, promoting environmental protection, and ensuring safe boiler operation. The impact of pulverized coal fineness on combustion efficiency is primarily reflected in the carbon content of fly ash. Reducing pulverized coal fineness significantly reduces the carbon content of fly ash and reduces incomplete combustion losses, thereby significantly improving combustion efficiency. Increasing pulverized coal fineness increases the mass of fly ash particles, which significantly increases erosion and wear on the heating surfaces of the flue. Furthermore, larger pulverized coal particles lead to longer flames, making the pulverized coal more difficult to burn out. Therefore, controlling the coal powder fineness within a reasonable range is of great significance for saving energy, maintaining economical combustion of boilers, and achieving safe, efficient and low-pollution emissions. However, most of the existing research on coal powder fineness is based on detection and control. The method of first detecting and then controlling the coal powder fineness is not convenient enough. If the coal powder fineness can be predicted first, it will greatly facilitate the control and adjustment of the coal powder fineness and improve the boiler operation efficiency.
[0003] A search for coal fineness prediction revealed the following related patents: Patent application number: CN202010278772.7, which discloses a modeling method for predicting coal fineness in a coal mill. This patent uses a support vector machine and particle swarm optimization algorithm to establish a prediction model, achieving coal fineness prediction and resolving the traditional problem of coal fineness detection by on-site workers. While possessing certain predictive and generalization capabilities, this method still suffers from computational complexity and limited real-time performance. The PSO algorithm has a slow convergence rate, is prone to falling into local optimal solutions, and is ineffective for multimodal problems. Therefore, it is necessary to design corresponding technical solutions to address these existing technical issues. Summary of the Invention
[0004] In view of the fact that most existing research focuses on controlling and adjusting the coal mill system after detecting the fineness of coal powder, and that a small number of studies have problems with the coal powder fineness prediction models, such as complex model establishment calculations, low model generalization ability, and poor parameter optimization, the present invention provides a coal powder fineness prediction method based on LSSVM and Bayesian optimization. The present invention can process nonlinear data, while traditional algorithms such as linear regression may not be able to accurately process nonlinear relationships. LSSVM uses kernel functions to map data to high-dimensional feature space, and Bayesian optimization can handle nonlinear and high-dimensional optimization problems and can automatically select the optimal search direction.
[0005] To achieve the above object, the present invention provides the following technical solution: comprising the following steps:
[0006] Step 1: Obtain historical operating data of the coal mill and data on coal powder fineness, clean and standardize the collected data to eliminate the influence of outliers and dimensions, and perform data preprocessing;
[0007] Step 2: Parameters affecting coal pulverization fineness in coal mill are used as input variables, coal pulverization fineness in coal mill is used as output variable, and a least squares support vector machine prediction model is constructed;
[0008] Step 3: Use the Bayesian optimization algorithm to optimize the hyperparameters in the LSSVM model. Select the next sampling point through the acquisition function, update the posterior distribution, and repeat the process until the stopping condition is met to obtain the optimal parameters.
[0009] Step 4: Use an independent test set to verify the prediction performance of the Bayesian optimized LSSVM model;
[0010] The test set data was used to obtain the prediction results of the model and compared them with the actual measurement results. Finally, the root mean square error (RMSE) was used to evaluate the effectiveness of the coal powder fineness prediction model.
[0011] As a preferred embodiment of the present invention, the operating parameters affecting the fineness of the coal powder in the pulverizer in step 1 include: the rotation speed of the pulverizer dynamic separator, the coal feed rate, the air temperature and volume at the inlet and outlet, and the type of coal.
[0012] As a preferred embodiment of the present invention, a statistical method is used to identify and eliminate outliers, and then the data is normalized to eliminate the influence of different dimensions. The parameter data is normalized to [0, 1], and PCA is used to extract the most important features to reduce the number of features while retaining most of the data information.
[0013] As a preferred embodiment of the present invention, the specific steps of constructing the least squares support vector machine prediction model in step 2 are as follows: the parameters used for modeling are expressed as where x i The parameters that affect the fineness of coal powder in the pulverizer are used as input, y i The coal powder fineness is expressed as the output, N is the number of samples, and the nonlinear function φ(x) is used to map the input space to the feature space, where φ(x) is a nonlinear function. The functional form of the prediction model is as follows: f(x) = w T φ(x)+b, where w is the weight vector and b is the bias term, is an optimization problem based on the structured risk minimization principle:
[0014]
[0015] where e i represents the regression error between the actual value of the output and the predicted value, and γ is the regularization parameter.
[0016] Using the Lagrange multiplier method, introduce the multiplier α i , construct the corresponding Lagrangian function:
[0017]
[0018] where α i is the Lagrange multiplier, which is obtained by i , α i The derivative of is set to zero, and the optimal solution condition of the problem can be obtained:
[0019]
[0020] By solving the above equation, we can get α i and e i The value of α is solved using i and e i , calculate the normal vector w and bias b of the hyperplane, and get b and α i After that, the final LSSVM prediction model can be expressed as:
[0021]
[0022] Where K(x i ,x j ) is a kernel function used to reduce the computational complexity of high-dimensional space. The present invention uses the radial basis function kernel as follows:
[0023] where σ is the bandwidth of the kernel function.
[0024] As a preferred embodiment of the present invention, in step 3, the Bayesian optimization algorithm is used to optimize the hyperparameter regularization parameter γ and kernel function bandwidth σ of the LSSVM model. The specific steps are as follows:
[0025] Choose a prior probability distribution to represent the prior knowledge of hyperparameters θ = (γ, σ), and evaluate an initial design point set D = {(γ1, σ1, y1), ..., (γ n ,σ n ,y n )}, using Gaussian process as prior to model the target function f(θ):
[0026] f(θ)~GP(m(θ),k(θ,θ′))
[0027] Where m(θ) is the mean function, k(θ,θ′) is the covariance function (kernel function), and the posterior distribution p(f|θ,D) is updated based on the initial design point set D. The posterior distribution can be calculated using the following formula:
[0028]
[0029] Among them, p(y|f,D) is the likelihood function, p(f|θ) is the prior distribution, and p(y|D) is the marginal likelihood.
[0030] Using Expected Improvement (EI) as the acquisition function, the formula is:
[0031] EI(θ)=(μ(θ)-f best )·Φ(Z)+σ(θ)·φ(Z)
[0032] Where μ(θ) and σ(θ) are the mean and standard deviation of the posterior distribution, and f best is the current optimal objective function value, Z is a random variable with standard normal distribution, and then optimize the acquisition function to find the parameter value θ that maximizes the acquisition function in the hyperparameter space. * , then evaluate the new point, repeatedly update the posterior distribution, select the acquisition function, optimize the acquisition function and evaluate the new point until the stopping condition is met.
[0033] As a preferred embodiment of the present invention, in step 4, the RMSE expression for evaluating the effectiveness of the coal fineness prediction model is:
[0034]
[0035] Where: y pre is the predicted coal powder fineness, y real is the measured coal powder fineness, n is the number of sample tests, and the smaller the RMSE value, the more accurate the model prediction.
[0036] Compared with the prior art, the present invention has the following beneficial effects:
[0037] 1. In view of the defects of traditional coal powder fineness detection methods such as high cost and complicated process, the present invention adopts a method of first prediction and then control, which greatly reduces the complex links of coal powder fineness detection and saves detection costs. In addition, the prediction model LSSVM adopted in this scheme is a sparse algorithm that can process nonlinear data, simplify the calculation process, and has high generalization ability and can adapt to new data sets. In addition, the Bayesian optimization algorithm adopted in this scheme has a faster convergence speed than the particle swarm optimization (PSO) algorithm, is not easy to fall into the local optimal solution, can automatically select the optimal parameters, and has higher accuracy.
[0038] 2. The coal powder fineness prediction method provided by the present invention is based on LSSVM and Bayesian, and can process nonlinear data. Traditional algorithms such as linear regression may not be able to accurately process nonlinear relationships. LSSVM uses kernel functions to map data to high-dimensional feature space, and Bayesian optimization can handle nonlinear and high-dimensional optimization problems and can automatically select the optimal search direction. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 Schematic diagram of the steps of the prediction model of the present invention;
[0040] Figure 2 This is a structural diagram of the prediction model and optimization of the present invention. DETAILED DESCRIPTION
[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0042] Combine Figure 1-2 A method for predicting coal powder fineness of a coal mill according to this embodiment includes the following steps:
[0043] Step 1: Obtain historical operating data of the coal mill and data on coal fineness. Clean and standardize the collected data to eliminate outliers and the influence of dimensions for data preprocessing. The operating parameters that affect coal fineness of the coal mill include: the speed of the coal mill dynamic separator, the coal feed rate, the air temperature and volume at the inlet and outlet, and the type of coal. Statistical methods are used to identify and eliminate outliers. The data is then normalized to eliminate the influence of different dimensions. The parameter data is normalized to [0, 1]. PCA is used to extract the most important features, reducing the number of features while retaining most of the data information.
[0044] Step 2: Take the parameters that affect the coal fineness of the coal mill as input variables and the coal fineness of the coal mill as output variables to construct a least squares support vector machine prediction model. The specific steps of constructing the least squares support vector machine prediction model are as follows: The parameters used for modeling are expressed as where x i The parameters that affect the fineness of coal powder in the pulverizer are used as input, y i The coal powder fineness is expressed as the output, N is the number of samples, and the nonlinear function φ(x) is used to map the input space to the feature space, where φ(x) is a nonlinear function. The functional form of the prediction model is as follows: f(x) = w Tφ(x)+b, where w is the weight vector and b is the bias term, is an optimization problem based on the structured risk minimization principle:
[0045]
[0046] where e i Represents the regression error between the actual output value and the predicted value, γ is the regularization parameter, and the Lagrange multiplier method is used to introduce the multiplier α i , construct the corresponding Lagrangian function:
[0047]
[0048] By combining w, b, e i , α i The derivative of is set to zero, and the optimal solution condition of the problem can be obtained:
[0049]
[0050] By solving the above equation, we can get α i and e i The value of α is solved using i and e i , we can calculate the normal vector w and bias b of the hyperplane and get b and α i After that, the final LSSVM prediction model can be expressed as:
[0051]
[0052] Where K(x i ,x j ) is a kernel function used to reduce the computational complexity of high-dimensional space. The present invention uses the radial basis function (RBF) kernel as follows:
[0053] Where σ is the bandwidth of the kernel function;
[0054] Step 3: Use the Bayesian optimization algorithm to optimize the hyperparameters in the LSSVM model. Select the next sampling point through the acquisition function, update the posterior distribution, and repeat the process until the stopping condition is met to obtain the optimal parameters. The specific steps of using the Bayesian optimization algorithm to optimize the hyperparameter regularization parameter γ and kernel function bandwidth σ of the LSSVM model are as follows:
[0055] Choose a prior probability distribution P(θ) to represent the prior knowledge of hyperparameters θ = (γ, σ), and evaluate an initial design point set D = {(γ1, σ1, y1), ..., (γ n ,σ n ,y n)}, using Gaussian process (GP) as a prior to model the objective function f(θ):
[0056] f(θ)~GP(m(θ),k(θ,θ′))
[0057] Where m(θ) is the mean function, k(θ,θ′) is the covariance function, and the posterior distribution p(f|θ,D) is updated based on the initial design point set D. The posterior distribution can be calculated using the following formula:
[0058]
[0059] Among them, p(y|f,D) is the likelihood function, p(f|θ) is the prior distribution, and p(y|D) is the marginal likelihood.
[0060] Using expected improvement (EI) as the acquisition function, the formula is
[0061] EI(θ)=(μ(θ)-f best )·Φ(Z)+σ(θ)·φ(Z)
[0062] Where μ(θ) and σ(θ) are the mean and standard deviation of the posterior distribution, and f best is the current optimal objective function value, Z is a random variable with standard normal distribution, and then optimize the acquisition function to find the parameter value θ that maximizes the acquisition function in the hyperparameter space. * , then evaluate the new point, repeatedly update the posterior distribution, select the acquisition function, optimize the acquisition function and evaluate the new point until the stopping condition is met;
[0063] Step 4: Use the test set to verify the prediction performance of the LSSVM model after Bayesian optimization. Use the test set data to obtain the prediction results of the model and compare them with the actual measurement results. Finally, the root mean square error (RMSE) is used to evaluate the effectiveness of the coal fineness prediction model. In step 4, the RMSE expression for evaluating the effectiveness of the coal fineness prediction model is:
[0064]
[0065] Where: y pre is the predicted coal powder fineness, y real is the measured coal powder fineness, n is the number of sample tests, and the smaller the RMSE value, the more accurate the model prediction.
[0066] The method for predicting coal powder fineness of the present invention solves the steps of coal powder extraction and detection on industrial sites, realizes online prediction of coal powder fineness, and has good prediction effect. Compared with the existing PSO-SVM prediction method, the LSSVM of the present invention is a sparse algorithm that can process nonlinear data, can simplify the calculation process, and has high generalization ability and can adapt to new data sets. Compared with the particle swarm optimization algorithm, the Bayesian optimization algorithm has a faster convergence speed, is not prone to falling into local optimal solutions, can automatically select optimal parameters, and has higher accuracy.
[0067] The above is a schematic description of the present invention and its embodiments, which is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs a structure and embodiment similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.
Claims
1. A method for predicting coal powder fineness of a coal mill, characterized in that: The following steps are involved: Step 1: Obtain the historical operation data of the coal mill and the data of coal powder fineness, and clean and standardize the collected data to eliminate the influence of outliers and dimensions; Step 2: Taking the parameters that affect the coal powder fineness of the coal mill as input variables and the coal powder fineness of the coal mill as output variables, a least squares support vector machine prediction model is constructed; Step 3: Use the Bayesian optimization algorithm to optimize the hyperparameters in the LSSVM model, select the next sampling point by using the expected improvement as the acquisition function, update the posterior distribution, and repeat the process until the stopping condition is met to obtain the optimal parameters; Step 4: Use the test set to verify the prediction performance of the LSSVM model after Bayesian optimization, use the test set data to obtain the prediction results of the model, and compare them with the actual measurement results. Finally, RMSE is used to evaluate the effectiveness of the coal powder fineness prediction model.
2. The method for predicting the fineness of coal powder in a coal mill according to claim 1, characterized in that: The operating parameters that affect the coal powder fineness of the coal mill in step 1 include: the rotation speed of the coal mill dynamic separator, the coal feed rate, the air temperature and air volume at the inlet and outlet, and the type of coal.
3. The method for predicting the fineness of coal powder in a coal mill according to claim 1, characterized in that: Statistical methods are used to identify and eliminate outliers, and then the data is normalized to eliminate the influence of different dimensions. The parameter data is normalized to [0,1], and PCA is used to extract the most important features to reduce the number of features while maintaining the data information.
4. The method for predicting the fineness of coal powder in a coal mill according to claim 1, characterized in that: In step 2, the least squares support vector machine prediction model is constructed. The specific steps are as follows: The parameters used for modeling are expressed as where x i The parameter that affects the coal powder fineness of the pulverizer is used as the input quantity, y i The coal powder fineness is taken as the output, N is the number of samples, and the nonlinear function φ(x) is used to map the input space to the feature space, where φ(x) is a nonlinear function. The function form of the prediction model is as follows: f(x) = w T φ(x)+b, where w is the weight vector and b is the bias term. Based on the structured risk minimization principle, the problem is described as an optimization problem: where e i Represents the regression error between the actual output value and the predicted value, γ is the regularization parameter, and the Lagrange multiplier method is used to introduce the multiplier α i , construct the corresponding Lagrangian function: By combining w, b, e i , α i The derivative of is set to zero, and the optimal solution condition of the problem can be obtained: By solving the above equation, we can get α i and e i The value of α is solved using i and e i , calculate the normal vector w and bias b of the hyperplane, and get b and α i After that, the final LSSVM prediction model can be expressed as: Where K(x i ,x j ) is a kernel function, which is used to reduce the computational complexity of high-dimensional space. The present invention uses the radial basis function kernel as follows: Where σ is the bandwidth of the kernel function.
5. The method for predicting the fineness of coal powder in a coal mill according to claim 1, characterized in that: In step 3, the Bayesian optimization algorithm is used to optimize the hyperparameter regularization parameter γ and kernel function bandwidth σ of the LSSVM model. The specific steps are as follows: Choose a prior probability distribution P(θ) to represent the prior knowledge of the hyperparameters θ = (γ, σ) and evaluate an initial design point set D = {(γ1, σ1, y1), ..., (γ n ,σ n ,y n )}, and use the Gaussian process as a prior to model the objective function f(θ): f(θ)~GP(m(θ),k(θ,θ′)) Where m(θ) is the mean function, k(θ,θ′) is the covariance function, and the posterior distribution p(f|θ,D) is updated according to the initial design point set D. The posterior distribution can be calculated by the following formula: Among them, p(y|f,D) is the likelihood function, p(f|θ) is the prior distribution, and p(y|D) is the marginal likelihood. Using Expected Improvement (EI) as the acquisition function, the formula is: EI(θ)=(μ(θ)-f best )·Φ(Z)+σ(θ)·φ(Z) Where μ(θ) and σ(θ) are the mean and standard deviation of the posterior distribution, and f best is the current optimal objective function value, Z is a random variable with standard normal distribution, and then optimizes the acquisition function to find the parameter value θ that maximizes the acquisition function in the hyperparameter space. * , then evaluate the new point, repeat updating the posterior distribution, selecting the acquisition function, optimizing the acquisition function and evaluating the new point until the stopping condition is met. After the optimization process is completed, the best hyperparameter combination and its corresponding objective function value are obtained from the optimization object.
6. The method for predicting the fineness of coal powder in a coal mill according to claim 5, characterized in that: The optimization ranges of the hyperparameter regularization parameter γ and kernel function bandwidth σ of the LSSVM model are [10 -2 ,10 2 ] and [10 -3 ,10 3 ].
Citation Information
Patent Citations
Steel cold rolling and finishing production raw material distributing method based on multi-class support vector machine
CN110918651A
Default user probability prediction method based on particle swarm optimization LSSVM
CN112308288A
Prediction method for fineness of pulverized coal of coal mill
CN117290752A
