阅读说明：本技术 基于矿物相与神经网络复合模型预测煤灰熔融温度的方法 (Method for predicting coal ash melting temperature based on mineral phase and neural network composite model ) 是由 叶泽甫 孟献梁 朱竹军 褚睿智 吴国光 宋上 李晓 江晓凤 李啸天 俞时 樊茂洲 于 2021-01-15 设计创作，主要内容包括：本发明公开了一种基于矿物相与神经网络复合模型预测煤灰熔融温度的方法,首先建立煤灰矿物相组成子模型,利用高温下化学组分相互反应产生的吉布斯自由能变化,建立线性规划问题,建立求解指定温度下煤灰矿物相组成的预测模型,并对此模型的一致性进行检验；在矿物相组成子模型的基础上,建立灰熔点预测子模型；建立神经网络模型,对神经网络的各项训练参数进行调校,并采用迭代算法进一步加强预测模型的预测精度,同时加入修正值,用以表示煤灰中次要元素对煤灰熔融性的影响,最后对所建立的模型的精确度和可靠性进行分析,确立预测结果精确性指标,并与支持向量机预测方式的预测结果进行比较。本发明建立的模型具有较高的可靠性。(The invention discloses a method for predicting coal ash fusion temperature based on a mineral phase and neural network composite model, which comprises the steps of firstly establishing a coal ash mineral phase composition sub-model, establishing a linear programming problem by utilizing Gibbs free energy change generated by mutual reaction of chemical components at high temperature, establishing a prediction model for solving coal ash mineral phase composition at specified temperature, and checking the consistency of the model; establishing an ash melting point prediction sub-model on the basis of the mineral phase composition sub-model; establishing a neural network model, adjusting and correcting various training parameters of the neural network, further enhancing the prediction precision of the prediction model by adopting an iterative algorithm, simultaneously adding a correction value for expressing the influence of secondary elements in the coal ash on the coal ash meltability, finally analyzing the precision and reliability of the established model, determining the accuracy index of the prediction result, and comparing the accuracy index with the prediction result of the support vector machine prediction mode. The model established by the invention has higher reliability.)

1. A method for predicting the melting temperature of coal ash based on a mineral phase and neural network composite model is characterized by comprising the following steps:

(1) collecting an XRD (X-ray diffraction) pattern of a certain coal at the temperature T so as to obtain the actual composition of the coal ash mineral phase at the temperature;

(2) taking the contents of five elements of silicon, aluminum, iron, calcium and magnesium in the coal ash and the temperature T as parameters influencing a coal ash-mineral phase prediction sub-model; establishing a linear programming problem by using a thermodynamic method and utilizing Gibbs free energy change generated by mutual reaction of chemical components at high temperature, and establishing a coal ash-mineral phase composition predictor model for solving a specified temperature T by using a Matlab tool;

(3) inputting the contents of silicon, aluminum, iron, calcium and magnesium and the temperature T into the coal ash-mineral phase composition prediction sub-model, and predicting the mineral phase composition at the temperature;

(4) comparing the predicted mineral phase composition with the actual composition, and verifying the accuracy of the coal ash-mineral phase composition prediction submodel;

(5) taking the output data of the coal ash-mineral phase composition predictor model as the input parameters of the next coal ash-ash melting point predictor model, and simultaneously setting the following parameters: maximum value of flow temperature T_maxAnd a minimum value T_minAnd a maximum tolerance temperature difference Δ Tb;

(6) establishing a coal ash-ash melting point prediction sub-model by utilizing a BP neural network;

(7) collecting raw data of coal ash, said raw data comprising K₂O、Na₂O、SiO₂、Al₂O₃、Fe₂O₃、CaO、MgO、SO₃、TiO₂、MnO₂The content of the coal ash, the deformation temperature DT, the softening temperature ST, the hemispherical temperature HT and the flowing temperature FT of the coal ash are used for establishing a coal ash database;

training a coal ash-ash fusion point prediction sub-model by adopting part of real coal ash data, and simulating by using the other part of real coal ash data to verify the accuracy of the model;

(8) introducing a correction value to correct the influence of Ti and Na/K elements on the model, wherein the correction formula is as follows:

in the formula, theta₁、θ₂Respectively represent the mass fractions of Ti and Na + K, w_i(θ_i) Represents the correction value of the specified element to the melting temperature at the temperature T;

then, repeatedly introducing correction values into the training of the neural network to get rid of the prediction influence of the correction values on the neural network, and simultaneously repeatedly solving a new correction value relational expression, thereby enabling the whole prediction model to be accurate;

(9) and analyzing the accuracy of the coal ash-ash fusion point prediction sub-model for predicting the coal ash data, and checking the prediction accuracy of the model.

2. The method for predicting the melting temperature of the coal ash based on the mineral phase and neural network composite model as claimed in claim 1, wherein in the step (6), the input layers of the BP neural network are the following mineral phase content data: wollastonite, pseudo-wollastonite, calcium monoaluminate, calcium dialuminate, tricalcium aluminate, tricalcium silicate, magnesium calcium oxide, diopside, forsterite, akermanite, willemite, anorthite, gehlenite, coleptolite, gehlenite, gayalite, hercynite, fagolite, clinoptilolite, forsterite, gahnite, cordierite, mullite, and quartz for a total of 24 input parameters; the output layer has 4 data of deformation temperature DT, softening temperature ST, hemisphere temperature HT and flow temperature FT.

3. The method for predicting the coal ash fusion temperature based on the mineral phase and neural network composite model as claimed in claim 1, wherein the step (7) of training the coal ash-ash fusion temperature prediction model comprises the following steps: the neural network toolbox in Matlab was chosen to be used, using the Bayesian regularisation algorithm and a hidden layer number of 10 layers, setting the training times to within 500 generations.

4. The method for predicting the melting temperature of the coal ash based on the mineral phase and neural network composite model as claimed in claim 1, wherein in the step (9), the evaluation accuracy indexes are mainly divided into mean absolute percentage error and mean square error, and the calculation formula is as follows:

wherein MAPE is the mean absolute percentage error and MSE is the meanSquare error, x_iAnd y_iRespectively a predicted value and an actual value of the data, and m is the total group number of the data;

meanwhile, linear correlation coefficients are used for expressing the index of the prediction fitting degree, and the formula is as follows:

wherein r represents a linear correlation coefficient, xi and yi are respectively a predicted value and an actual value of data,andthe average predicted value and the average actual value of the data are respectively shown, and m is the total group number of the data.

Technical Field

The invention relates to a method for predicting fusion temperature of coal ash based on a mineral phase and neural network composite model.

Background

China has abundant coal resources, the yield and the consumption of the coal resources are at the top of the world, the yield of raw coal reaches 36.8 hundred million tons in 2015, the consumption of the raw coal reaches 39.65 hundred million tons, and the yield and the consumption of the coal in China respectively account for 72.1 percent and 64 percent of the total energy production and consumption in China. According to the forecast of Chinese engineering institute, the consumption of Chinese coal reaches more than 45 hundred million t by 2030 years according to the calculation of the current energy requirement. According to the resource status of China and the proportion of coal in energy production and consumption structures, the energy structure taking coal as a main body can not be changed for a long time. Therefore, how to more reasonably and efficiently utilize the existing coal resources becomes a very important problem which needs to be solved urgently. Although China has abundant coal resources, the composition difference of the components of the coal ash is large, and the difference of the compositions of the coal ash causes the difference of the melting temperature of the coal ash. In the comprehensive utilization of coal, the melting characteristic of coal ash is an important coal quality index.

When coal is burned, mineral substances therein are converted into ash, and the melting property of coal ash is an important index of coal for power and coal for gasification. Meanwhile, the coal ash meltability is also an important factor influencing the performance of the coal ash. The melting temperature of the coal ash is a direct expression of the melting characteristic of the coal ash and mainly comprises four characteristic temperature values: deformation temperature, softening temperature, hemispherical temperature, and flow temperature. According to different slag discharging modes, the coal gasification process is divided into two main types of solid slag discharging and liquid slag discharging. The solid-state slagging technology requires that the ash melting temperature of the raw material coal is higher than the operation temperature, and ash slag is discharged in a solid state; the slag tapping technique requires that the coal ash is melted at a relatively low temperature and the ash can be discharged in a molten state. Therefore, the coal ash meltability directly determines the selection of slag discharging modes in the coal gasification process, and is an important factor influencing whether the furnace can normally run or not. Therefore, in order to find a method for improving the meltability of the coal ash, it is necessary to intensively study the meltability of the coal ash so as to adapt to combustion and gasification techniques of different slag discharging methods or to expand the range of applicable coal types.

Currently, the main method for industrially measuring the melting temperature of coal ash is laboratory measurement, and the method needs to obtain the content of each element of coal firstly and further obtains the melting temperature by methods such as blending, high-temperature heating and the like. The experimental measurement requires many steps, takes a long time and is expensive. More importantly, when the melting temperature of the coal ash is improved, repeated measurement is needed, and the defect is obvious, so that an accurate and reliable coal ash melting point prediction model is very important in the process of improving the melting property of the coal ash.

Disclosure of Invention

The invention aims to provide a method for predicting the melting temperature of coal ash based on a mineral phase and neural network composite model, which is simple, convenient and feasible and has high accuracy.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows: a method for predicting the melting temperature of coal ash based on a mineral phase and neural network composite model comprises the following steps:

(1) collecting an XRD (X-ray diffraction) pattern of a certain coal at the temperature T so as to obtain the actual composition of the coal ash mineral phase at the temperature;

(2) taking the contents of five elements of silicon, aluminum, iron, calcium and magnesium in the coal ash and the temperature T as parameters influencing a coal ash-mineral phase prediction sub-model; establishing a linear programming problem by using a thermodynamic method and utilizing Gibbs free energy change generated by mutual reaction of chemical components at high temperature, and establishing a coal ash-mineral phase composition predictor model for solving a specified temperature T by using a Matlab tool;

(3) inputting the contents of silicon, aluminum, iron, calcium and magnesium and the temperature T into the coal ash-mineral phase composition prediction sub-model, and predicting the mineral phase composition at the temperature;

(4) comparing the predicted mineral phase composition with the actual composition, and verifying the accuracy of the coal ash-mineral phase composition prediction submodel;

(5) taking the output data of the coal ash-mineral phase composition predictor model as the input parameters of the next coal ash-ash melting point predictor model, and simultaneously setting the following parameters: maximum value of flow temperature T_maxAnd a minimum value T_minAnd a maximum tolerance temperature difference Δ Tb;

(6) establishing a coal ash-ash melting point prediction sub-model by utilizing a BP neural network;

(7) collecting raw data of coal ash, said raw data comprising K₂O、Na₂O、SiO₂、Al₂O₃、Fe₂O₃、CaO、MgO、SO₃、TiO₂、MnO₂The content of the coal ash, the deformation temperature DT, the softening temperature ST, the hemispherical temperature HT and the flowing temperature FT of the coal ash are used for establishing a coal ash database;

training a coal ash-ash fusion point prediction sub-model by adopting part of real coal ash data, and simulating by using the other part of real coal ash data to verify the accuracy of the model;

(8) introducing a correction value to correct the influence of Ti and Na/K elements on the model, wherein the correction formula is as follows:

in the formula, theta₁、θ₂Respectively represent the mass fractions of Ti and Na + K, w_i(θ_i) Represents the correction value of the specified element to the melting temperature at the temperature T;

then, repeatedly introducing correction values into the training of the neural network to get rid of the prediction influence of the correction values on the neural network, and simultaneously repeatedly solving a new correction value relational expression, thereby enabling the whole prediction model to be accurate;

(9) and analyzing the accuracy of the coal ash-ash fusion point prediction sub-model for predicting the coal ash data, and checking the prediction accuracy of the model.

In the step (6), the input layer of the BP neural network is the content data of the following mineral phases: medite (3 CaO.2SiOx)₂) Wollastonite (CaO. SiO)₂) Wollastonite (CaO. SiO)₂) Calcium monoaluminate (CaO. Al)₂O₃) Calcium dialuminate (CaO 2 Al)₂O₃) Tricalcium aluminate (3 CaO. Al)₂O₃) Tricalcium silicate (3 CaO. SiO)₂) Magnesium calcium oxide (CaO. MgO), diopside (CaO. MgO)₂·SiO₂) Calcium forsterite (CaO. MgO. SiO)₂) Akermanite (2 CaO. MgO)₂·SiO₂) MgCao (3 CaO. MgO)₂·SiO₂) Anorthite (CaO. Al)₂O₃·2SiO₂) Caalumite (CaO. Al)₂O₃·SiO₂) Alumino melilite (2 CaO. Al)₂O₃·SiO₂) Calcium aluminum garnet (3 CaO. Al)₂O₃·3SiO₂) Hercynite (FeO. Al)₂O₃) Iron olivine (2 FeO. SiO)₂) And pyroxene (FeO. SiO)₂) Forsterite (2 MgO. SiO)₂) Magnesium aluminate spinel (MgO. Al)₂O₃) Cordierite (2 MgO.2Al)₂O₃·58iO₂) Mullite (3 Al)₂O₃·2SiO₂) And quartz (SiO)₂) A total of 24 input parameters; the output layer has 4 data of deformation temperature DT, softening temperature ST, hemisphere temperature HT and flow temperature FT.

Further, the step (7) of training the coal ash-ash fusion point prediction model comprises the following steps: the neural network toolbox in Matlab was chosen to be used, using the bayesian regularisation algorithm and a hidden layer number of 10 layers, setting the training times to within 500 generations.

Further, in step (9), the index of the evaluation accuracy is mainly divided into an average absolute percentage error and a mean square error, and the calculation formula is as follows:

wherein MAPE is the mean absolute percentage error, MSE is the mean square error, x_iAnd y_iRespectively a predicted value and an actual value of the data, and m is the total group number of the data;

meanwhile, linear correlation coefficients are used for expressing the index of the prediction fitting degree, and the formula is as follows:

wherein r represents a linear correlation coefficient, xi and yi are respectively a predicted value and an actual value of data,andaverage predicted values and the average of the data respectivelyThe mean actual value, m is the total number of sets of data.

Compared with the prior art, the coal ash mineral phase composition and the neural network model are coupled to establish the composite model, and the composite model has higher accuracy and reliability through verification and can be used for predicting the coal ash melting temperature in actual production.

Drawings

FIG. 1 is a flow chart of a method of predicting coal ash fusion temperature according to the present invention;

FIG. 2 is a flow chart of an algorithm for predicting ash fusion point by combining a neural network with a mineral phase model;

FIG. 3 is a flowchart of a correction value reintroduction algorithm;

FIG. 4 is an XRD diffraction pattern of a primary coal ash of the Gemini coal at 1500 ℃.

Detailed Description

The invention is described in further detail below with reference to the figures and specific examples.

The invention provides a method for predicting coal ash fusion temperature based on a mineral phase and neural network composite model, which comprises the steps of firstly establishing a coal ash mineral phase composition sub-model, establishing a linear programming problem by utilizing Gibbs free energy change generated by mutual reaction of chemical components at high temperature, establishing a prediction model for solving the coal ash mineral phase composition at a specified temperature, and checking the consistency of the model; establishing an ash melting point prediction sub-model on the basis of the mineral phase composition sub-model; establishing a neural network model, adjusting and correcting various training parameters of the neural network, further enhancing the prediction precision of the prediction model by adopting an iterative algorithm, simultaneously adding a correction value for expressing the influence of secondary elements in the coal ash on the coal ash meltability, finally analyzing the precision and reliability of the established model, determining the accuracy index of the prediction result, and comparing the accuracy index with the prediction result of the support vector machine prediction mode.

The specific process is shown in fig. 1, and the specific steps are as follows:

(1) the chemical composition of the primary coal ash of the metamorphic coal is collected and is shown in table 1:

TABLE 1 chemical composition of the coal ash of the Gemini coal

Composition (I)	K2O	Na2O	SiO2	Al2O3	Fe2O3	CaO	MgO	SO3	TiO2	MnO2
											Mass fraction/%	0.62	0.30	49.80	35.01	3.65	3.16	0.90	2.40	1.58	0.02

Collecting the XRD pattern of the metamorphic coke, the diaschist coal, at 1500 ℃, as shown in figure 4, so as to obtain the composition of the coal ash mineral phase in the state;

(2) taking the contents of silicon, aluminum, iron, calcium and magnesium and the temperature as input parameters of the coal ash-mineral phase prediction sub-model; a thermodynamic method is used, the Gibbs free energy change generated by the mutual reaction of chemical components at high temperature is utilized, a linear programming problem is established, and the mathematical expression is as follows:

an objective function for a linear programming problem;

n is the reaction number possibly existing in a coal ash molten reaction system;

n_ithe amount of each reaction product in the system is the unit of mol;

t is the temperature at which the reaction occurs and is given in K;

ΔG_T，iis the change in Gibbs free energy at temperature T for reaction i in kJ/mol;

β_ithe amount of the substance taking part in the reaction of each oxide is shown as mol; wherein the oxide involved is SiO₂、Al₂O₃CaO, MgO, and FeO.

The gibbs free energy is calculated using the following formula:

whereinIn order to obtain a standard molar enthalpy of formation of the product,for the standard heat of formation of the reactants, n_bThe amount of the reactant consumed for 1mol of the product formed in each reaction; phi_AverageIs the arithmetic mean of the gibbs free energy.

Compiling a calculation program by using a linear function in a Matlab tool, solving the linear programming problem, compiling the linear programming problem into a DLL dynamic link library, calling by using a Visual Studio 2015 program, and establishing a coal ash-mineral phase prediction submodel for solving the coal ash-mineral phase at the specified temperature;

(3) the contents of silicon, aluminum, iron, calcium, magnesium (see table 2) and the temperature T1500 ℃ were input into the soot-mineral phase predictor model to obtain the mineral phase composition at this temperature, see table 2:

TABLE 2 prediction of mineral phase composition of macerals at 1500 deg.C

Composition (I)	Anorthite	Cordierite	Mullite	Quartz	FeO
						Mass fraction/%	15.68	6.54	37.53	29.08	3.28

(4) Comparing the predicted mineral phase composition with the actual composition, and verifying the accuracy of the coal ash-mineral phase prediction sub-model;

analysis of FIG. 4 reveals that the coal ash of the metamorphic coal, a secondary coal, has a mineral phase consisting primarily of mullite and secondarily of quartz at 1500 ℃. This is consistent with the prediction results of the mineral phase predictor model. The coal ash-mineral phase prediction sub-model has higher reliability.

(5) Taking the output data of the coal ash-mineral phase prediction submodel as the input parameters of the coal ash-ash melting point prediction submodel, and simultaneously setting the following parameters: maximum value of flow temperature T_maxAnd a minimum value T_min(general T)_maxAt 1800 ℃ T_min1200 deg.C), and a maximum tolerated temperature difference Δ Tb;

(6) establishing a coal ash-ash melting point prediction model by using a BP neural network, as shown in figure 2;

the input layer of the BP neural network used in this example is the content data of the following mineral phases: medite (3 CaO.2SiOx)₂) Wollastonite (CaO. SiO)₂) Wollastonite (CaO. SiO)₂) Calcium monoaluminate (CaO. Al)₂O₃) Calcium dialuminate (CaO 2 Al)₂O₃) Tricalcium aluminate (3 CaO. Al)₂O₃) Tricalcium silicate (3 CaO. SiO)₂) Magnesium calcium oxide (CaO. MgO), diopside (CaO. MgO)₂·SiO₂) Calcium forsterite (CaO. MgO. SiO)₂) Akermanite (2 CaO. MgO)₂·SiO₂) MgCao (3 CaO. MgO)₂·SiO₂) Anorthite (CaO. Al)₂O₃·2SiO₂) Caalumite (CaO. Al)₂O₃·SiO₂) Yellow feldspar of calcium aluminium(2CaO·Al₂O₃·SiO₂) Calcium aluminum garnet (3 CaO. Al)₂O₃·3SiO₂) Hercynite (FeO. Al)₂O₃) Iron olivine (2 FeO. SiO)₂) And pyroxene (FeO. SiO)₂) Forsterite (2 MgO. SiO)₂) Magnesium aluminate spinel (MgO. Al)₂O₃) Cordierite (2 MgO.2Al)₂O₃·5SiO₂) Mullite (3 Al)₂O₃·2SiO₂) And quartz (SiO)₂) A total of 24 input parameters; the output layer has 4 data of deformation temperature DT, softening temperature ST, hemisphere temperature HT and flow temperature FT.

(7) Collecting raw data for 335 groups of coal ashes, the data comprising K₂O、Na₂O、SiO₂、Al₂O₃、Fe₂O₃、CaO、MgO、SO₃、TiO₂、MnO₂And the deformation temperature DT, the softening temperature ST, the hemisphere temperature HT and the flowing temperature FT of the coal ash, establishing a mysql coal ash database, and carrying out basic setting of the database.

Table 3 includes the basic data distribution of these data;

TABLE 3355 compositional value variation Range for group coal ash data

In general coal ash, SiO₂Generally 30-70%, Al₂O₃The content is 15-30%. Fe₂O₃The content of (B) is generally 5 to 15%. The content of CaO varies widely, and sometimes can be as high as 30% or more. The MgO content in the coal ash is low, and generally less than 4 percent.

In comparison with Table 3, the coal ash data collected in this example was found to contain substantial amounts of the major components over the range of the corresponding component content in typical coal ash. And the average value of each component is positioned at the position close to the center of the content variation interval. Therefore, the coal ash data collected in the embodiment is reliable, random, uniform in distribution and wide in distribution.

Training a coal ash-ash fusion point prediction model by adopting 70% of real coal ash data, and simulating by using 30% of real coal ash data to verify the accuracy of the model; the training of the coal ash-ash fusion point prediction model comprises the following steps: using a Bayesian Regularization algorithm and a hidden layer number of 10 layers, selecting a neural network toolbox in Matlab R2015b to be used, and setting the training times to be within 500 generations;

(8) introducing a correction value to correct the influence of Ti and Na/K elements on the model, wherein the correction formula is as follows:

in the formula, theta₁、θ₂Respectively represent the mass fractions of Ti and (Na + K), and because the properties of Na and K are relatively close, the two elements can be considered together. w is a_i(T) represents a correction value of the specified element to the melting temperature at the temperature T. And repeatedly introducing a correction value into the training of the neural network so as to get rid of the prediction influence of the correction value on the neural network. At the same time, the new correction value relation is solved repeatedly, thereby refining the whole prediction model, as shown in fig. 3. The final result is:

the corrected relation of the Ti element in the prediction model is as follows:

w_Ti(θ_Ti)＝0.4367θ_Ti

the corrected relation of the Na/K element in the prediction model is as follows:

w_Na/K(θ_Na/K)＝-14.2θ_Na/K-12.57θ_Na/K ²+1.05θ_Na/K ³

it can be seen that the ash melting point of the coal ash increases with the increase of the Ti element, generally speaking, for every 1% increase of TiO₂The ash melting point of the coal ash increased by about 43.7K. From w_Na/K(θ_Na/K) As can be seen from the expression of (A), Na/K has a function of reducing the ash melting point of the coal ash, and generally, when the content of Na/K reaches about 8%, the melting point is reduced to the maximumLow. This is substantially in accordance with the description in the literature.

(9) The accuracy of the coal ash-ash fusion point predictor model for predicting the coal ash data is analyzed, and the indexes for judging the accuracy are mainly average absolute percentage error and mean square error. The calculation formula is as follows:

wherein MAPE is the mean absolute percentage error, MSE is the mean square error, x_iAnd y_iRespectively, the predicted value and the actual value of the data. m is the total number of sets of data.

Meanwhile, linear correlation coefficients are used for expressing the index of the prediction fitting degree, and the formula is as follows:

wherein r represents a linear correlation coefficient, xi and yi are respectively a predicted value and an actual value of data,andthe average predicted value and the average actual value of the data are respectively shown, and m is the total group number of the data.

The ash fusion point of the coal ash predicted by the model is compared with the actual ash fusion point of the coal ash to calculate the error, and the error is collated into table 3. Data above 1500 ℃ were recorded as 1500 ℃ because they were difficult to obtain.

As can be seen from the analysis of Table 4, the absolute error of DT is between-86K and 79K, the absolute error of ST is between-45K and 70K, and the absolute error of FT is between-50K and 60K.

For linear correlation coefficients, generally, 0.7 < r < 0.8 is acceptable, 0.8 < r < 0.9 is preferred, and r > 0.9 is very good.

The correlation coefficients of the coal ash fusion point prediction model of the present invention are r (dt) 0.807, r (st) 0.843, and r (ft) 0.856, respectively, and the predicted linear correlation degrees are relatively good. The MAPE value for DT was 3.32%, the MAPE value for ST was 2.98%, and the MAPE value for FT was 3.01%. The MSE value for DT is 2306, the MSE value for ST is 2164, and the MSE value for FT is 1978.

TABLE 4 prediction of soot fusibility and error Unit (. degree.C.)

(10) Comparison with support vector machine prediction data

Similar to neural networks, support vector machines are learning-based mechanisms, but unlike neural networks, SVMs use mathematical methods and optimization techniques.

The principle of the support vector machine is to project data into a high-dimensional space and model it in this space:

the model is then used to perform regression functions. The prediction function of the support vector machine can be implemented directly using Matlab R2015 b.

To distinguish from this embodiment, the input quantities of the support vector machine are set to several chemical components (Si, Al, Fe, Ca, Mg, Na, K, Ti), and the output quantities are set to DT, ST, and FT.

The exact same 33 sets of data were used for comparison with the model. Table 5 shows the prediction results of the support vector machine, and data at 1500 ℃ or higher are recorded as 1500 ℃ because they are difficult to obtain.

TABLE 5 prediction of soot fusion and error units (. degree.C.) for support vector machine

Analysis of Table 5 reveals that, in the support vector machine prediction evaluation, the MAPE value of DT is 4.62%, the MAPE value of ST is 4.06%, the MAPE value of FT is 4.81%, the absolute error of DT is between about-107K and 89K, the absolute error of ST is between about-60K and 80K, and the absolute error of FT is between about-60K and 85K. The MSE value for DT is 4306, the MSE value for ST is 3164, and the MSE value for FT is 4978.

TABLE 6 comparison of the prediction results of the model of this embodiment with the SVM

According to the comparison results in the table 6, the coal ash fusibility prediction model established by the invention is comprehensively superior to the prediction results of the support vector machine in all aspects. The prediction accuracy of the model has higher reliability.