CN115510638A - Rotary kiln multi-target parameter optimization method based on GRNN model - Google Patents

Abstract

A rotary kiln multi-objective parameter optimization method based on a GRNN model comprises the following steps: acquiring process parameters of a rotary kiln to be optimized, performing numerical simulation through a preset rotary kiln numerical simulation model, and outputting energy efficiency indexes and quality indexes of the process parameters; combining the energy efficiency index and the quality index, taking the coal feeding amount, the secondary wind speed and the secondary wind temperature of the rotary kiln as input parameters of a preset GRNN model, and outputting to obtain parameters to be optimized of the rotary kiln; establishing a multi-objective optimization function according to the parameters to be optimized, taking the parameters to be optimized as an initial population, optimizing the multi-objective optimization function by using a third-generation non-dominated sorting genetic algorithm, and outputting to obtain optimal process parameters; and optimizing the rotary kiln according to the optimal process parameters. The invention uses the data model to replace a numerical simulation model as a data support model, uses NSGA-III to carry out multi-target parameter optimization on the rotary cement kiln, and effectively improves the energy efficiency.

Description

Rotary kiln multi-target parameter optimization method based on GRNN model

Technical Field

The invention belongs to the field of optimization research of rotary kilns, and particularly relates to a rotary kiln multi-target parameter optimization method based on a GRNN model.

Background

In the process flow of cement production, the energy consumption of preheating, decomposing and burning is the largest. In the process, the power consumption and coal consumption of the rotary cement kiln can account for more than 70% of the total energy consumption in the process, and meanwhile, the firing process which most influences the product quality in the cement production also occurs in the rotary cement kiln, so the rotary cement kiln is a thermal device which is most relevant to the influences of energy consumption and product quality in the cement production. In view of large yield, high energy consumption and large lifting space in the cement industry, the research on the energy efficiency optimization of the rotary cement kiln is strongly necessary.

In related optimization researches, the rotary kiln optimization research is a research hotspot of a student in the cement industry, the research method covers the aspects of mechanism modeling, data modeling and simulation analysis, and the research direction covers the aspects of structure optimization, process parameter optimization and the like. Zhan Xinjian, etc. by adjusting the fan, the coal inlet and the vibration frequency of the grate cooler, the purposes of increasing the waste heat recovery amount and enhancing the cooling capacity of the grate cooler are achieved, so that the energy efficiency optimization is achieved, and the effectiveness of the research is verified. TAtmaca is obtained by performing energy analysis on a rotary kiln

Analysis shows that the combustion efficiency is the main factor influencing the system efficiency, so that measures for reducing energy consumption can be started from insulation, heat conduction efficiency improvement and effective sealing. Hu Jing ocean is subjected to heat balance analysis based on data collected on a rotary kiln production site, and on the basis, income and expenditure of energy materials are calculated, so that preliminary consumption reduction optimization is developed. Ditaranto et al characterized and compared several oxy-fuel flames of a rotary kiln based on numerical simulation models and with reference to the flame burning in air, and the experimental variables were the oxidant composition and flow rate of the primary and secondary air, but were not further optimized.

In the numerical simulation research of the rotary kiln, boateng et al establish two heat transfer models of axial one-dimensional and cross-sectional two-dimensional, which reduces the three-dimensional spatial distribution of the temperature field to a certain extent, but the research is still deficient. The mechanism modeling of the rotary kiln can describe the flow field in the kiln to a certain extent, but the modeling difficulty is high, and the data acquisition is limited. The Elattar et al establishes a two-dimensional model of the rotary kiln based on Fluent and researches the influence of different process parameters and geometric parameters of the rotary kiln on flame, thereby providing a correlation of the flame length of the restricted jet flow. The Checky constructs a rotary kiln three-dimensional model based on Fluent and researches the influence rule of kiln skins and reaction heat on the flow field in the kiln on the basis. Zhao Liangxia et al studied the influence relationship between the coal feeding amount and the primary air volume ratio on the concentration and temperature of NOx in the kiln based on Fluent, and considered that the ratio of the coal feeding amount to the primary air volume ratio is 0.6, which is more helpful for obtaining a product with higher quality. Lu Cong establishes a rotary kiln numerical simulation model based on Fluent, and on the basis, a highest temperature soft measurement model is established, and meanwhile, the structure of the reducing rotary kiln is optimized.

In the industrial multi-objective optimization research, sohani et al have certain research progress on the multi-objective optimization of the toluene production industry, the research takes energy consumption, environment and economy as optimization objectives, compared with the past optimization, the considered optimization objectives are more comprehensive, and the obtained optimization effect is proved by experiments to be more valuable than the past optimization results. Chen Jiang, etc. respectively has comprehensive exploration research on the multi-target parameter optimization of the performance of a firing zone of a horseshoe flame glass kiln regenerator and a ceramic roller kiln, and combines numerical simulation, data modeling and an intelligent optimization algorithm to obtain better operation effects on different thermal equipment. Mohanty et al studied on a rotary kiln based on an artificial neural network and a GA optimization algorithm, and the study demonstrated the effectiveness of multi-objective optimization analysis in evaluating and improving the performance of an industrial rotary kiln device, and by combining the method with a data-driven model, the optimization potential of the study path can be further exploited. Hao et al propose a multi-objective collaborative optimization method for cement calcination process. The method takes the coal consumption and the f-CaO content as optimization targets, and provides a TDRM-Jaya algorithm to optimize a model, so that the optimization of the cement calcination process is realized, but the f-CaO content has the problems of long sampling time and calcination time interval and the like on site.

According to the research background, the optimization of the rotary cement kiln is divided into three fields according to different methods: optimizing a rotary kiln, simulating a rotary kiln numerical value and optimizing industrial multiple targets, wherein:

the optimization of the rotary kiln is mainly started from two aspects of structure optimization and technological parameter optimization. The former is more suitable for being carried out in the early stage of equipment design and development, and the structural modification of the currently running equipment is high in cost and difficulty. The latter has the characteristics of low cost, high feasibility and remarkable effect. However, most of the existing rotary kiln process parameter researches are considered singly, and multi-objective optimization researches of comprehensive consideration are lacked;

the rotary kiln numerical simulation is developed from mechanism modeling to the assistance of numerical simulation software, the modeling of the thermal process in the rotary kiln is improved and reliable, and the rotary kiln numerical simulation has the characteristics of low cost, comprehensive data and the like and becomes a reliable research means. However, the problem that no proper quantization index exists in the subsequent optimization research of numerical simulation, and most of the research is only to simply compare and select an experimental group with a relatively good flow field performance as an optimization result. On the other hand, the numerical simulation cannot cover the whole variable space due to relatively long calculation time, and the data requirement of the intelligent optimization strategy is difficult to meet;

the industrial multi-objective optimization is a research thought for carrying out multi-objective optimization by means of a numerical simulation model and a data method, and feasibility and effectiveness of the industrial multi-objective optimization are proved in many industrial fields. The rotary cement kiln has strong commonality with the optimization objects in the industrial fields, so the thought can be used for providing an effective optimization approach with low cost and high efficiency for the cement industry. Compared with multi-objective optimization research in other industrial fields, comprehensive and appropriate quantitative indexes are still not considered in the research of the rotary cement kiln.

Disclosure of Invention

The invention aims to solve the problems that a rotary cement kiln in the prior art needs a large amount of experimental group data, a numerical simulation method is long in calculation time and high in time cost, and a large amount of data support of optimization research is difficult to provide, and the single-target optimization research effect has limitation and multi-target optimization realization difficulty is high.

Specifically, the invention provides a rotary kiln multi-objective parameter optimization method based on a GRNN model, which comprises the following steps:

acquiring process parameters of a rotary kiln to be optimized, performing numerical simulation through a preset rotary kiln numerical simulation model, and outputting energy efficiency indexes and quality indexes of the process parameters;

combining the energy efficiency index and the quality index, and outputting to-be-optimized parameters of the rotary kiln by taking the coal feeding amount, the secondary wind speed and the secondary wind temperature of the rotary kiln as input parameters of a preset GRNN model;

establishing a multi-objective optimization function according to the parameters to be optimized, taking the parameters to be optimized as an initial population, optimizing the multi-objective optimization function by using a third-generation non-dominated sorting genetic algorithm, and outputting to obtain optimal process parameters;

and optimizing the rotary kiln according to the optimal process parameters.

Further, the preset rotary kiln numerical simulation model comprises a turbulence model, a particle orbit model, a pulverized coal combustion model and a radiation model, wherein the turbulence model satisfies the relation (1):

wherein, G _k Being turbulent kinetic energy, alpha _k Is k equation turbulent Plantt number, alpha _ε Turbulent Plantt number, C, of the epsilon equation _1ε Has a value of 1.42,C _2ε Is 1.68;

the particle trajectory model satisfies the relation (2):

wherein, F _D For rotational lift, F _G Is the weight of the particles, u _p Is the particle velocity;

the pulverized coal combustion model comprises a volatilization analysis sub-model and a coke combustion sub-model, wherein the volatilization analysis sub-model and the coke combustion sub-model respectively satisfy relational expressions (3) and (4):

wherein m is _p Is the mass of the coal dust particles, f _v,0 Is the initial mass fraction of volatile components, f _w,0 Is mass fraction of volatile material, m _p,0 Is the initial mass of the coal dust particles, k is the coal dust combustion reaction rate constant, R _c For the coke burning rate, p is the partial pressure of oxygen, K _c As diffusion coefficient, K _d Is the kinetic coefficient;

the radiation model satisfies the relation (5):

wherein, alpha is absorption coefficient, sigma _s For the scattering coefficient, C is the phase function coefficient and G is the incident amplitude.

Further, the parameters to be optimized comprise the maximum sintering temperature, the length of a sintering zone and the thermal efficiency.

Furthermore, in the preset GRNN model, the input parameter is x in the i-th group of working conditions _i The output parameter to be optimized is y _i Then the parameter to be optimized satisfies the following relation (6):

y _i ＝f(x _i )(i＝1,2,…,n) (6)；

wherein x is _i Is particularly expressed as x _i ＝[m _i ,v _i ,t _i ]，m _i 、v _i 、t _i Respectively the coal feeding amount and the coal feeding amount in the ith group of working conditionsSpecific values of three variables of secondary wind speed and secondary wind temperature, y _i Is specifically expressed as y _i ＝[T _i ,L _i ,η _i ]Wherein T is _i 、L _i 、η _i Specific numerical values of the three indexes of the highest firing temperature, the length of the firing zone and the thermal efficiency in the ith working condition group are respectively set;

defining a predicted value f (x) of the parameter to be optimized when the parameter is x, wherein the predicted value f (x) meets the relation (7):

furthermore, the preset GRNN model includes an input layer, a mode layer, a summation layer, and an output layer, in the i-th set of operating conditions, the input variable dimension M is 3, the output variable dimension K is 3, the training data set has N sets, the number of neurons in the input layer is the same as the input variable dimension M, the number of neurons in the mode layer is the same as the training set N, the summation layer includes two kinds of neurons, a numerator unit and a denominator unit, the number of the numerator unit is the same as the output variable dimension K, the number of the denominator unit is 1, and the number of neurons in the output layer is the same as the output variable dimension K.

Further, the activation function of the mode layer satisfies relation (8):

wherein X is an input variable, X _i Is the ith neuron center and σ is the smoothing factor.

Further, the weight of the molecular unit is defined as y _i,j The output of the molecular unit in the jth dimension satisfies the relation (9):

the output of the denominator unit satisfies the relation (10):

further, the neuron output of the output layer satisfies the relation (11):

still further, the multi-objective optimization function satisfies the relation (12):

wherein, f ₁ ＝-T＝k(m,v,t)、f ₂ ＝-L＝g(m,v,t)、f ₃ = η = h (m, v, t) are three objective functions of the maximum firing temperature, the firing zone length, and the thermal efficiency, respectively, and k (m, v, t), g (m, v, t), h (m, v, t) are data models of the maximum firing temperature, the firing zone length, and the thermal efficiency, respectively.

The invention achieves the following beneficial effects:

1. the data model is adopted to replace a numerical simulation model, so that the data model can be effectively used as a data support model for optimization research, the method has the advantages of short calculation time and low time cost, and compared with other data modeling methods, the GRNN model has better prediction capability under the condition of less training data;

2. and (3) performing multi-target parameter optimization on the rotary cement kiln by using a third-generation non-dominated sorting genetic algorithm (NSGA-III), so that a process parameter combination of the optimal operation condition of the rotary cement kiln is obtained, and the energy efficiency of the rotary cement kiln is effectively improved.

Drawings

FIG. 1 is a flow chart of steps of a rotary kiln multi-objective parameter optimization method based on a GRNN model according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of a preset GRNN model according to an embodiment of the present invention;

FIG. 3 is a schematic reference point diagram of the NSGA-III algorithm provided by the embodiment of the present invention;

FIG. 4 is a schematic diagram of the basic flow of the NSGA-III algorithm provided by the embodiment of the present invention;

FIG. 5 is a schematic diagram of the distribution of the temperature field inside the rotary kiln according to an embodiment of the present invention;

FIG. 6 is a schematic view of the distribution of the temperature field inside another rotary kiln according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of the flue gas temperature provided by an embodiment of the present invention;

FIG. 8 is a velocity flow field schematic diagram of the outlet of a four-channel burner of a rotary kiln according to an embodiment of the present invention;

FIG. 9 is a schematic diagram of the distribution of O2 components in a rotary kiln provided by an embodiment of the present invention;

FIG. 10 is a graph showing a comparison of the maximum firing temperatures provided by the embodiments of the present invention;

FIG. 11 is a schematic diagram showing a comparison of the lengths of the fired belts provided by the embodiments of the present invention;

FIG. 12 is a schematic diagram showing a thermal efficiency ratio provided by an embodiment of the present invention;

FIG. 13 is a schematic diagram of a multi-objective optimization problem solution set provided by an embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

Referring to fig. 1, fig. 1 is a flowchart illustrating steps of a rotary kiln multi-objective parameter optimization method based on a GRNN model according to an embodiment of the present invention, which specifically includes the following steps:

s1, acquiring a process parameter of a rotary kiln to be optimized, performing numerical simulation through a preset rotary kiln numerical simulation model, and outputting to obtain an energy efficiency index and a quality index of the process parameter.

Specifically, in the embodiment of the invention, the basic equations related to numerical simulation in the rotary kiln include four major equations of mass conservation, energy conservation, momentum conservation and component conservation, because the thermal process in the kiln not only conforms to the most basic physical process, but also includes the process of chemical reaction between different components of pulverized coal combustion.

The basic equations in the examples of the present invention are as follows:

mass conservation equation:

wherein ρ is density, t is time, x, y and z are coordinates, and u, v and w are velocity components corresponding to the coordinates.

Energy conservation equation:

where T is temperature, λ is thermal conductivity, S _T Representing a viscous dissipation term.

Conservation of momentum equation:

wherein the content of the first and second substances,

mu is dynamic viscosity, p is air flow pressure, S _u 、S _v 、 S _w Is a generalized source term.

Component conservation equation:

wherein, c _m Is volume concentration, D _m Diffusion system of mNumber, S _m Is the productivity of component m.

The preset rotary kiln numerical simulation model comprises a turbulence model, a particle orbit model, a pulverized coal combustion model and a radiation model.

In the numerical simulation process, how to select a turbulence model is mainly based on the characteristics of a flow field, whether fluid can be compressed, calculation precision, computer calculation power and the like, and through the process analysis of a rotary kiln, after primary air is blown out of a combustor, high flow velocity is generated and strong cyclone air is accompanied, and the flow velocity of air flow in the axial direction of the rotary kiln can be rapidly reduced, wherein the embodiment of the invention selects an RNG k-epsilon model as a turbulence equation, and has the advantages of good calculation convergence effect in a simulation calculation example with strong cyclone and large velocity gradient, capability of improving certain calculation precision and reducing certain calculation amount, and the turbulence model satisfies the relation (1):

wherein G is _k Being turbulent kinetic energy, alpha _k Is k equation turbulent Plantt number, alpha _ε Turbulent Plantt number, C, of the epsilon equation _1ε Has a value of 1.42,C _2ε Is 1.68;

the simulation of the combustion process of the coal powder in the rotary kiln is a key part of a numerical simulation model, the movement of the coal powder particles in the rotary kiln is very complex, and the coal powder particles and fluid and the coal powder particles and the wall surface can interact to change the movement. The discrete phase model is a particle tracking algorithm defined on the basis of a Lagrange method, fluid in simulation can be used as a continuous phase, solid particles are used as discrete phases, the model firstly carries out flow field calculation of continuous media, and then carries out motion trail calculation of the particles in the flow field on the basis. In the calculation process, the collision between particles is ignored, and the heat, mass and momentum between the discrete phase and the continuous phase are not exchanged after the particle track is calculated. In the embodiment of the invention, the diameter, the initial speed direction and the initial speed direction of the coal dust particles are assumed to be consistent with the size in the discrete phase model, the coal dust particles are uniformly sprayed out from the coal wind outlet surface of the combustor, the discrete phase model calculation is performed every 40 steps in the numerical simulation calculation process, and the particle orbit model meets the relation formula (2):

wherein, F _D For rotational lift, F _G Is the weight of the particles, u _p Is the particle velocity;

in the rotary kiln, pulverized coal particles are ejected from an air outlet of a four-channel combustor and are mixed with high-temperature air to be combusted, in Fluent, the combustion of the pulverized coal particles is usually simulated by adopting an Eddy Dissipation combustion model in a component transportation model, the pulverized coal combustion model comprises a volatilization analysis sub-model and a coke combustion sub-model, and the volatilization analysis sub-model and the coke combustion sub-model respectively satisfy relational expressions (3) and (4):

wherein m is _p Is the mass of the coal dust particles, f _v,0 Is the initial mass fraction of volatile components, f _w,0 Is mass fraction of volatile material, m _p,0 Is the initial mass of the coal dust particles, k is the coal dust combustion reaction rate constant, R _c For the coke burning rate, p is the partial pressure of oxygen, K _c As diffusion coefficient, K _d Is the kinetic coefficient;

in summary, the embodiment of the present invention obtains a conclusion that the rotary cement kiln uses radiation heat transfer as a main heat transfer mode, therefore, a radiation model needs to be considered in model setting, and in view of the characteristics of the rotary cement kiln that the cylinder has a long length and a large length and diameter, the embodiment of the present invention selects a P1 radiation model as a radiation model for the numerical simulation, and the P1 radiation model is general and has consideration of radiation scattering, is suitable for a research object with a complex structure and related to a combustion process, has the characteristics of high calculation speed and high efficiency, and satisfies the relational expression (5):

wherein, alpha is the absorption coefficient, sigma _s For the scattering coefficient, C is the phase function coefficient and G is the incident amplitude.

Preferably, the embodiment of the invention also sets boundary conditions for simulating a real scene, and the partial boundary conditions of the inlet and the outlet are set as shown in table 1, wherein each air channel of the secondary air channel and the combustor is set as a speed inlet, the kiln outlet is set as a pressure outlet, and the wall surface is set as a non-slip wall surface.

TABLE 1 partial boundary condition setting table for rotary kiln

And S2, combining the energy efficiency index and the quality index, taking the coal feeding amount, the secondary wind speed and the secondary wind temperature of the rotary kiln as input parameters of a preset GRNN model, and outputting to obtain parameters to be optimized of the rotary kiln.

GRNN was originally proposed by Specht in 1991, belongs to an improved form of a radial basis network, has superior nonlinear regression capability and high fault tolerance and robustness compared to other models, and is an effective method for solving the nonlinear regression problem. Therefore, GRNN as a construction method of the data model selected by the embodiment of the invention conforms to the data characteristics, has better prediction capability under the condition of less training data, and can process some unstable data.

The parameters to be optimized comprise the highest sintering temperature, the length of a sintering zone and the thermal efficiency.

Furthermore, in the preset GRNN model, the input parameter is x in the i-th group of working conditions _i The output parameter to be optimized is y _i Then the parameter to be optimized satisfies the following relation (6):

y _i ＝f(x _i )(i＝1,2,…,n) (6)；

wherein x is _i Is particularly expressed as x _i ＝[m _i ,v _i ,t _i ]，m _i 、v _i 、t _i The concrete numerical values of the three variables of the coal feeding quantity, the secondary air speed and the secondary air temperature in the ith group of working conditions, y _i Is particularly shown as y _i ＝[T _i ,L _i ,η _i ]Wherein T is _i 、L _i 、η _i Specific numerical values of the three indexes of the highest firing temperature, the length of the firing zone and the thermal efficiency in the ith group of working conditions are respectively set;

the GRNN regression method is realized based on kernel regression of a density estimation concept, and when an observed value f (x) of a certain point x is required to be obtained, the conditional probability density of Y under the x condition is firstly calculated, and then the model predicted value under the x condition is finally obtained by solving the expectation of continuous variables.

Defining a predicted value f (x) of the parameter to be optimized when the parameter is x, wherein the predicted value f (x) meets the relation (7):

in relation (7), the conditional probability density f _Y (Y | X) can be derived from the joint probability density of X and Y divided by the edge probability density for X, i.e.:

by combining the above equation with relation (7), a regression function of f (x) can be obtained:

the joint probability density of X and Y and the edge probability density of X can be determined by training set data

And a Parzen-Rosenblatt density estimator, namely:

wherein K (x) is a kernel function of the density estimator; h is smoothness, which controls the width of the nucleus. On the basis of equation (14), it can be simplified by a Nadaraya-Watson regression estimator. After normalizing the training set data, a weighting function may be defined:

wherein, for any x, all

The regression equation of f (x) can be obtained by combining the above relations:

the kernel function of GRNN uses a gaussian distribution function, that is:

wherein m is ₀ Is xThe dimension, here 3. Assuming the same width σ is used, σ is mapped on the kernel function identically to the smoothness h, and x _i As kernel function center, i.e.:

substituting the above equation into the regression estimator yields:

if the GRNN model is expressed in a matrix form, a final regression function of the preset GRNN model is obtained;

further, the preset GRNN model includes an input layer, a mode layer, a summation layer, and an output layer, specifically, referring to fig. 2, fig. 2 is a schematic structural diagram of the preset GRNN model provided in the embodiment of the present invention, in the i-th set of operating conditions, the input variable dimension M is 3, the output variable dimension K is 3, and the training data set has N sets;

the input layer of the network is responsible for receiving an input variable set, the number of neurons of the input layer is the same as the dimension M of the input variable, and the input layer in the model is provided with 3 neurons;

the mode layer of the network is responsible for activating data information based on a Gaussian function, and the number of neurons in the mode layer is the same as that of the training set N;

a summation layer of the network is responsible for carrying out weighted summation and arithmetic summation on the neurons of the mode layer respectively, the summation layer comprises two neurons of a numerator unit and a denominator unit, the number of the numerator unit is the same as the dimension K of the output variable, and the number of the denominator unit is 1;

the output layer of the network is responsible for computing two types of neurons of the summation layer, and the number of neurons in the output layer is the same as the dimension K of the output variable.

Further, the activation function of the mode layer satisfies relation (8):

wherein X is an input variable, X _i Is the ith neuron center and σ is the smoothing factor.

Further, defining the weight of the molecular unit as y _i,j The output of the molecular unit in the jth dimension satisfies the relation (9):

the output of the denominator unit satisfies the relation (10):

further, the neuron output of the output layer satisfies the relation (11):

and S3, establishing a multi-objective optimization function according to the parameters to be optimized, taking the parameters to be optimized as an initial population, optimizing the multi-objective optimization function by using a third-generation non-dominated sorting genetic algorithm, and outputting to obtain optimal process parameters.

In the cement production process, energy efficiency and quality are very important standards, and the product quality is ensured, and the energy efficiency during production is also improved as much as possible, so that the production economy is as high as possible on the premise of ensuring the product quality. In other words, in the optimization of the rotary kiln parameters, a group of optimal process parameter combinations needs to be found, even if the highest firing temperature and the length of a firing zone are properly improved and the thermal efficiency is improved as much as possible, the energy efficiency and the quality have a strong coupling relation, and the situation that the optimal process parameters are simultaneously achieved cannot exist. The energy utilization capacity, the maximum firing temperature and the length of a firing zone of the rotary kiln evaluated by the thermal efficiency of the rotary kiln are evaluated to determine whether the working temperature field of the rotary kiln reaches the standard or not.

Still further, the multi-objective optimization function satisfies the relation (12):

wherein, f ₁ ＝-T＝k(m,v,t)、f ₂ ＝-L＝g(m,v,t)、f ₃ = η = h (m, v, t) are three objective functions of the maximum firing temperature, the firing zone length, and the thermal efficiency, respectively, and k (m, v, t), g (m, v, t), h (m, v, t) are data models of the maximum firing temperature, the firing zone length, and the thermal efficiency, respectively.

In the embodiment of the present invention, a third-generation non-dominated sorting genetic algorithm is used to optimize the multi-objective optimization function, where the third-generation non-dominated sorting genetic algorithm (NSGA-III) is an optimization algorithm based on reference point selection, please refer to fig. 3 and fig. 4, fig. 3 is a schematic reference point diagram of the NSGA-III algorithm provided in the embodiment of the present invention, and fig. 4 is a schematic basic flow diagram of the NSGA-III algorithm provided in the embodiment of the present invention.

Specifically, the basic flow of the NSGA-III algorithm is as follows: initially, reference points are constructed and then an initial population P of N individuals is randomly generated _t Generating Q by cross-mutation _t Progeny of the offspring, breeding P _t And Q _t Merging into a new population R with the individual number of 2N _t Then to R _t The 2N individuals in the next generation are subjected to non-dominant sorting, and the individuals which can be selected based on the non-dominant grade are directly selected into a next generation parent population P _t+1 Discarding the individuals which are selected to be dominated based on the non-dominated level, and then selecting the remaining individuals which cannot be selected according to the non-dominated level based on a reference point mechanism until the number of the individuals meets N- (x + y); finally, two different selection mechanisms are combinedCombining the two groups of individuals into P _t+1 And (5) parent population, repeating the process until the ending condition is reached.

And S4, optimizing the rotary kiln according to the optimal process parameters.

Illustratively, in the embodiment of the invention, based on Ansys Fluent 2020R1 numerical solving software, the basic equation, the RNG k-epsilon turbulence model, the discrete phase model, the Eddy Dissipation combustion model and the P1 radiation model are solved by the SIMPLEC algorithm, and the boundary conditions are as shown in the table 1.

The specific solving process is as follows: because the numerical simulation of the rotary kiln is complex, the calculation of simultaneously starting a plurality of models is complex and difficult, the convergence of the calculation result is slow, and the error is easy to report. Therefore, firstly, cold state numerical simulation should be carried out, only the basic equation and the RNG k-epsilon turbulence model are opened in the step, and the simulated flow field is stable when iteration is carried out for 400 steps. Then, after the simulation fluency tends to be stable, starting a discrete phase model, an Eddy distance combustion model and a P1 radiation model, realizing ignition by arranging a small high-temperature area at an air outlet of the combustor, starting to continue model solution calculation, and entering a thermal state numerical simulation stage.

Because the discrete phase model of the embodiment of the invention is set to calculate the particle motion track once every 40 steps, the residual error curve of the numerical simulation fluctuates once every 40 steps, which is a normal phenomenon. And because the combustion process of the pulverized coal in the rotary cement kiln is very complicated, and the calculation convergence residual error of the numerical simulation is larger than that of a common research case, whether the calculation convergence is realized or not is judged by monitoring the numerical changes of a plurality of key position points such as the kiln tail temperature, the burning zone area temperature and the like. After about 1600 steps of calculation, the numerical value and residual curve of each monitoring point tend to be stable, so that the numerical value simulation calculation iteration is considered to be converged, and subsequent data post-processing can be carried out.

In the embodiment of the present invention, the distribution of the temperature field inside the rotary kiln can be obtained after the preset rotary kiln numerical simulation model is solved, as shown in fig. 5 and 6.

According to the temperature distribution condition in the rotary kiln, the temperature in the rotary kiln is increased and then decreased along the length direction of the rotary kiln, the pulverized coal is ignited at a position 10m away from the kiln head after being ejected at a high speed along with primary air, and the heat released by combustion is transferred to flue gas to form a high-temperature region. After the high-temperature area, the temperature of the flue gas gradually decreases along the length direction of the kiln in a gradient manner until the temperature reaches the outlet of the kiln tail because the heat is dissipated to the outside through the kiln wall and is consumed by cement burning. And as can be seen from the cross section temperature cloud chart, the whole temperature distribution of the cross section is uniform, the firing quality is improved, the temperature is reduced in a gradient manner along the radius direction, and the actual conditions of kiln crust heat dissipation and firing heat consumption are met.

Through the temperature data analysis of the temperature field, the highest firing temperature in the kiln is 1855K, the firing temperature is lower, and the condition may cause that the cement firing quality is not high and the product quality is lower, which indicates that a large optimization space exists.

As shown in FIG. 7, the flue gas enters the rotary kiln at an initial temperature of 300K from the kiln head, rapidly rises in temperature starting at about 10m from the kiln head, breaks through 1773K at 15m from the kiln head, and reaches a maximum temperature of 1855K at about 16m from the kiln head. After 16m, the flue gas temperature was stepped down, dropping below 1773K at about 22m, and finally exiting the kiln tail at 1534K. In combination with the explanation of the division method of each area of the rotary cement kiln, the area from 0m to 15.1m to the kiln head can be divided into a cooling zone, the area from 15.1m to 22.2m to the kiln head can be divided into a burning zone, and the area from 22.2m to the kiln tail outlet can be divided into a cooling zone. Therefore, the length of the burning zone under the basic working condition of the rotary kiln is about 7.1m, and the burning zone belongs to a lower numerical value in an interval with qualified length.

The velocity flow field conditions of the outlets of the four-channel combustor of the rotary kiln can be obtained after the preset rotary kiln numerical simulation model is solved, as shown in fig. 8.

As can be seen from the fluid vector diagram, the flow velocity of the primary air is obviously higher than that of the secondary air, especially the rotational flow air and the axial flow air in the primary air. The position of the cyclone wind outlet is between the secondary wind and the central wind, and therefore, the cyclone wind, the secondary wind and the central wind have great speed gradient, and a plurality of backflow regions are formed. The backflow zone is helpful for the adsorption and aggregation of the pulverized coal, so that the pulverized coal is fully combusted, and heat is released more quickly and efficiently. Therefore, the speed flow field issuing condition of the numerical simulation model accords with theoretical reality, the fact that the speed gradient and pulverized coal combustion can be closely related can be found, and the wind speeds of primary wind and secondary wind are required to be included in the consideration range of parameters to be optimized.

The distribution condition of the O2 component in the rotary kiln can be obtained after the rotary kiln numerical simulation model is preset, as shown in FIG. 9.

As can be seen from the O2 component distribution cloud chart, the mass fraction of O2 is relatively stable and is close to the initial content at a position 0 to 5m away from the kiln head, which indicates that in the interval, O2 is not consumed as a combustion improver by coal dust combustion. While at 5 to 15m the mass fraction of O2 decreases rapidly and is concentrated close to the axis, indicating a sufficient combustion of the coal dust in this interval. At 15m to the end of the kiln, the mass fraction of O2 remained essentially constant and 6.6% of the mass fraction of O2 was expelled from the end of the kiln, indicating that at the subsequent stage the coal powder had burnt out and no more O2 was consumed. Compared with a temperature cloud picture, the distribution shape of the mass fraction cloud picture of O2 is slightly similar, but the change trend of the mass fraction cloud picture is more forward, which just accords with the phenomenon that coal powder burns firstly and then releases heat, and the shape of the flame in the mass fraction cloud picture of O2 is more obvious and hammer-shaped, which accords with the theoretical reality.

The preset rotary kiln numerical simulation model used in the embodiment of the invention is verified as follows:

the data of the measured parameters obtained through actual measurement and the flow field variation trend of the numerical simulation result are compared and analyzed to verify the reliability of the numerical simulation model. The comparison between the numerical simulation model provided by the embodiment of the present invention and the actual data is shown in table 2 below.

TABLE 2 model verification table

As can be seen from Table 2, the simulated values of the mass fractions of the highest temperature, the kiln tail temperature, the kiln head temperature and the kiln tail O2 have errors with the actual values, but the maximum relative error is only 8.9%, and the relative error is not more than 10%. By combining the relative error and the convergence of the temperature cloud chart, the speed cloud chart and the O2 cloud chart in the above result analysis with the real distribution state in the rotary kiln, the numerical simulation model constructed by the embodiment of the invention can be proved to have higher reliability.

Meanwhile, the heat efficiency, the highest firing temperature and the firing zone length of the rotary kiln are calculated by combining the energy efficiency and the quality indexes, energy efficiency and quality quantification values under the initial working condition can be obtained, and subsequent optimization research can use the energy efficiency and quality quantification values as the reference to judge whether the optimization effect is good or bad. The specific values of each index of the initial working condition are shown in the following table 3.

TABLE 3 initial working condition index numerical table

Evaluation index	Initial operating condition value
		Thermal efficiency	53.08％
Maximum firing temperature	1855.0K
		Length of fired belt	7.1m

The preset GRNN model used in the embodiments of the present invention is verified as follows:

and (3) constructing a data model of the rotary kiln by taking the secondary air speed, the secondary air temperature and the coal feeding amount as input variables and taking the highest firing temperature, the length of a firing zone and the thermal efficiency as output variables to replace a numerical simulation model with low calculation efficiency. Firstly, 108 groups of experimental data are obtained through a numerical simulation model based on a three-factor nine-level orthogonal experimental table, and the value of each index is calculated, wherein the parameter level is shown in the following table 4. Because the embodiment of the invention aims to find the optimal working condition, the elimination of part of extreme working conditions does not influence the optimization result, the modeling and prediction rate can be improved, and finally 72 groups are determined as a training group and 10 groups are determined as a test group.

TABLE 4 parameter level table

For GRNN, given a set of training variables, its network structure and weight values y _i,j The data model precision can be adjusted only by the smooth factor sigma, the smooth factor sigma plays a deterministic role in the GRNN model precision, if the smooth factor sigma is close to infinity, the prediction result approaches to the average value of output variables in a training set, and if the smooth factor sigma is close to infinity, the prediction result approaches to the output variable value closest to the point, namely overfitting. The embodiment of the invention determines the optimal value of the smoothing factor sigma to be 0.1 through cross validation.

After the data modeling is completed, the accuracy of the built rotary kiln data model is checked to determine whether the accuracy is necessary to replace the numerical simulation model. In the embodiment of the invention, the relative error of the predicted value relative to the experimental value is calculated by comparing the predicted value of the data model with the highest firing temperature, the length of the firing zone and the thermal efficiency in 10 test groups and the numerical simulation experimental value, so as to judge whether the established GRNN data model is accurate and reliable, and the comparison results are shown in fig. 10, fig. 11 and fig. 12.

As can be seen from the comparison results in fig. 10, 11, and 12, the relative errors of the predicted values and the experimental values of the three output dimensions are all controlled within 10%. Therefore, it can be concluded that the error of the preset GRNN model constructed in the embodiment of the present invention is within an acceptable range, and the error can be used for subsequent optimization instead of a numerical simulation model.

To more particularly illustrate the accuracy of the predetermined GRNN model, experimental values, predicted values and their relative errors for the 10 test groups are detailed in table 5.

TABLE 5 relative error between experimental value and predicted value

As can be seen from Table 5, in 10 test groups, the maximum relative errors of the maximum firing temperature, the length of the firing zone and the thermal efficiency were 3.43%, 9.73% and 7.24%, the minimum relative errors were 0.19%, 5.73% and 0.02%, and the average relative errors were 1.87%, 7.68% and 3.20%. The conclusion further verifies that the preset GRNN model is accurate and reliable.

After the accuracy of the GRNN data model is verified, the embodiment of the present invention is compared with a common data modeling method, so as to further verify the advantages of the modeling method of the preset GRNN model in small sample data modeling, as shown in table 6 below.

TABLE 6 GRNN and BPNN regression effect comparison table

In the embodiment of the present invention, the solution example of the multi-objective optimization function is as follows:

the adopted calculation software is Matlab R2020b, a population size is set to be 30, a cross coefficient is 0.5, a variation coefficient is 0.5, the number of iterative calculations is 3000 based on a PlatEMO multi-objective optimization platform, and a pareto frontier solution set is obtained by solving a multi-objective optimization problem through an NSGA-III algorithm as shown in FIG. 13.

As can be seen from fig. 13, the pareto solutions solved by NSGA-III are relatively uniform in distribution, and 30 pareto leading edge solutions in the graph can be further selected according to different emphasis on energy efficiency and quality and different process requirements of different products in practice. In combination with the actual requirement range of the firing zone temperature and the firing zone length, the pareto frontier solution with the highest firing temperature of 2038.1K, the firing zone length of 17.9m and the thermal efficiency of 66.21% is selected as the optimal solution in the embodiment of the present invention, and the corresponding process parameter combinations are that the coal feeding amount is 1.09kg/s, the secondary air speed is 2.77m/s and the secondary air temperature is 1162K. Under the combination of the technological parameters, the thermal efficiency is relatively high, and the maximum firing temperature and the length of a firing zone are appropriate.

After the optimal process parameter combination is selected from the pareto solution set, in order to visually reflect the optimization effect, the energy efficiency and quality index values corresponding to the optimal process parameter combination are compared with the energy efficiency and quality index values of the initial working conditions, and the optimization effect is shown in the following table 7.

TABLE 7 comparison of initial and optimal solutions

As can be seen from the above table, the working condition after optimization is 183.1K higher than the maximum firing temperature of the working condition before optimization, the length of a firing zone is 10.8m higher, and the thermal efficiency is 13.13% higher. The results fully show that the optimization strategy of the invention has effectiveness on the optimization of cement production process parameters, obviously improves the energy efficiency of the rotary kiln, improves the firing quality on the original basis, effectively improves the energy efficiency of the rotary kiln, and ensures the high temperature and reasonable distribution of a temperature field.

The invention achieves the following beneficial effects:

1. the data model is adopted to replace a numerical simulation model, so that the data model can be effectively used as a data support model for optimization research, the method has the advantages of short calculation time and low time cost, and compared with other data modeling methods, the GRNN model has better prediction capability under the condition of less training data;

2. and (3) performing multi-target parameter optimization on the rotary cement kiln by using a third-generation non-dominated sorting genetic algorithm (NSGA-III), so that a process parameter combination of the optimum operation condition of the rotary cement kiln is obtained, and the energy efficiency of the rotary cement kiln is effectively improved.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by a computer program, which can be stored in a computer readable storage medium, and when executed, the computer program may include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like. For example, in a possible implementation manner, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the computer program implements each process and step in the rotary kiln multi-objective parameter optimization method based on the GRNN model provided in the embodiment of the present invention, and can implement the same technical effect, and in order to avoid repetition, details are not described here again.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.

Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.

While the present invention has been described with reference to the preferred embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, which are illustrative, but not restrictive, and that various changes may be made therein by those skilled in the art without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (9)

1. A rotary kiln multi-objective parameter optimization method based on a GRNN model is characterized by comprising the following steps:

acquiring process parameters of a rotary kiln to be optimized, performing numerical simulation through a preset rotary kiln numerical simulation model, and outputting energy efficiency indexes and quality indexes of the process parameters;

combining the energy efficiency index and the quality index, and outputting to-be-optimized parameters of the rotary kiln by taking the coal feeding amount, the secondary wind speed and the secondary wind temperature of the rotary kiln as input parameters of a preset GRNN model;

establishing a multi-objective optimization function according to the parameters to be optimized, taking the parameters to be optimized as an initial population, optimizing the multi-objective optimization function by using a third-generation non-dominated sorting genetic algorithm, and outputting to obtain optimal process parameters;

and optimizing the rotary kiln according to the optimal process parameters.

2. The GRNN model-based rotary kiln multi-objective parameter optimization method of claim 1, wherein the preset rotary kiln numerical simulation model comprises a turbulence model, a particle orbit model, a pulverized coal combustion model, and a radiation model, wherein the turbulence model satisfies a relation (1):

wherein G is _k Being turbulent kinetic energy, alpha _k Is k equation turbulent Plantt number, alpha _ε Turbulent Plantt number, C, of epsilon equation _1ε Has a value of 1.42,C _2ε Is 1.68;

the particle trajectory model satisfies the relation (2):

wherein, F _D For rotational lift, F _G Is the weight of the particles, u _p Is the particle velocity;

the pulverized coal combustion model comprises a volatilization analysis sub-model and a coke combustion sub-model, wherein the volatilization analysis sub-model and the coke combustion sub-model respectively satisfy relational expressions (3) and (4):

wherein m is _p Is the mass of the coal dust particles, f _v,0 Is the initial mass fraction of volatile components, f _w,0 Is mass fraction of volatile material, m _p,0 Is the initial mass of the coal dust particles, k is the coal dust combustion reaction rate constant, R _c For the coke burning rate, p is the partial pressure of oxygen, K _c As diffusion coefficient, K _d Is the kinetic coefficient;

the radiation model satisfies the relation (5):

wherein alpha is the absorption coefficient，σ _s For the scattering coefficient, C is the phase function coefficient and G is the incident amplitude.

3. The GRNN model-based rotary kiln multi-objective parameter optimization method of claim 1, wherein the parameters to be optimized include maximum firing temperature, firing zone length, thermal efficiency.

4. The GRNN model-based rotary kiln multi-objective parameter optimization method of claim 3, wherein the input parameter x is input in the predetermined GRNN model in defining the i-th set of operating conditions _i The output parameter to be optimized is y _i Then the parameter to be optimized satisfies the following relation (6):

y _i ＝f(x _i ) (i＝1,2,…,n) (6)；

wherein x is _i Is particularly expressed as x _i ＝[m _i ,v _i ,t _i ]，m _i 、v _i 、t _i The specific numerical values y of the three variables of the coal feeding amount, the secondary air speed and the secondary air temperature in the ith group of working conditions _i Is particularly shown as y _i ＝[T _i ,L _i ,η _i ]Wherein T is _i 、L _i 、η _i Specific numerical values of the three indexes of the highest firing temperature, the length of the firing zone and the thermal efficiency in the ith working condition group are respectively set;

defining a predicted value f (x) of the parameter to be optimized when the parameter is x, wherein the predicted value f (x) meets the relation (7):

5. the GRNN model-based rotary kiln multi-objective parameter optimization method according to claim 4, wherein the predetermined GRNN model includes an input layer, a mode layer, a summation layer, and an output layer, in the i-th set of operating conditions, an input variable dimension M is 3, an output variable dimension K is 3, N sets of training data are provided, the number of neurons in the input layer is the same as the input variable dimension M, the number of neurons in the mode layer is the same as the training set N, the summation layer includes two kinds of neurons, a numerator unit and a denominator unit, the number of neurons in the numerator unit is the same as the number of neurons in the output variable dimension K, and the number of neurons in the output layer is the same as the number of neurons in the output variable dimension K.

6. The GRNN model-based rotary kiln multi-objective parameter optimization method of claim 5, wherein the activation function of the mode layer satisfies the relation (8):

wherein X is an input variable, X _i Is the ith neuron center and σ is the smoothing factor.

7. The GRNN model-based rotary kiln multi-objective parameter optimization method of claim 5, wherein the weight defining the molecular unit is y _i,j The output of the molecular unit in the jth dimension satisfies the relation (9):

the output of the denominator unit satisfies the relation (10):

8. the GRNN model-based rotary kiln multi-objective parameter optimization method of claim 7, wherein the neuron output of the output layer satisfies the relation (11):

9. the GRNN model-based rotary kiln multi-objective parameter optimization method of claim 4, wherein the multi-objective optimization function satisfies the relation (12):

wherein f is ₁ ＝-T＝k(m,v,t)、f ₂ ＝-L＝g(m,v,t)、f ₃ = η = h (m, v, t) are three objective functions of the maximum firing temperature, the firing zone length, and the thermal efficiency, respectively, and k (m, v, t), g (m, v, t), h (m, v, t) are data models of the maximum firing temperature, the firing zone length, and the thermal efficiency, respectively.

Images