CN115755779A

CN115755779A - Control-oriented data-driven urban solid waste incineration full-process modeling method

Info

Publication number: CN115755779A
Application number: CN202211035911.9A
Authority: CN
Inventors: 汤健; 王天峥; 夏恒; 乔俊飞
Original assignee: Beijing University of Technology
Current assignee: Beijing University of Technology
Priority date: 2022-08-27
Filing date: 2022-08-27
Publication date: 2023-03-07

Abstract

The invention provides a control-oriented data-driven urban solid waste incineration full-flow modeling method which comprises the steps of firstly, selecting input characteristics of an in-furnace incineration process model based on experience cognition, then, constructing an in-furnace incineration process series model by utilizing XGboost, then, selecting input characteristics of a flue gas treatment process model based on mutual information analysis results, carrying out related modeling, and finally, training in a echelon to provide a full-flow series-parallel model. The control-oriented data-driven urban solid waste incineration full-flow modeling method provided by the invention can realize full-flow model building, guides the existing running factories and provides reference for the research of the subsequent MSWI process advanced intelligent optimization control algorithm.

Description

Control-oriented data-driven urban solid waste incineration full-process modeling method

Technical Field

The invention relates to the technical field of municipal solid waste treatment, in particular to a control-oriented data-driven municipal solid waste incineration full-process modeling method.

Background

The MSW is a major factor causing a serious urban pollution problem in our country, and along with the gradual emphasis of the country on garbage classification, the calorific value of MSW is gradually increased, so that the MSW incineration (MSWI) technology becomes an important means for treating MSW. Compared with other treatment means, the MSWI technology with the characteristics of harmlessness, reduction, recycling and the like can save a large amount of land resources, and has obvious social benefits, economic benefits and environmental benefits. However, the domestic MSWI technology is mostly introduced abroad, so that an Automatic Combustion Control (ACC) system of a domestic MSWI power plant is difficult to stably operate, and at present, manual maintenance is mostly depended on field experts. The field experts control the existing problems according to the expert experience and by combining the MSWI process operation monitoring condition, thereby ensuring the safe and stable operation of the MSWI power plant. However, according to different expert experiences, control means and timeliness for solving the problems are difficult to guarantee, so that the MSWI process is low in control efficiency, the pollutant emission possibly exceeds the standard in a short period and the like, and therefore the problem of researching MSWI process intelligent control is mainly solved.

Due to the complex mechanism of the MSWI process full-flow modeling, the MSWI process full-flow modeling is researched by adopting a Computational Fluid Dynamics (CFD) software mode. However, research is conducted on the interior of a hearth to perform research modeling, and most CFD software is applied to the front of a factory building and is less instructive to the existing operating factory. Therefore, it is necessary to design a control-oriented data-driven urban solid waste incineration full-process modeling method.

Disclosure of Invention

The invention aims to provide a control-oriented data-driven urban solid waste incineration full-flow modeling method, which can realize full-flow model building, guide the existing running factories and provide reference for the research of the subsequent MSWI process advanced intelligent optimization control algorithm.

In order to achieve the purpose, the invention provides the following scheme:

a control-oriented data-driven urban solid waste incineration full-process modeling method comprises the following steps:

step 1: the method comprises the steps of building an MSWI full-flow model, wherein the MSWI full-flow model comprises an in-furnace combustion process model input selection module, an in-furnace combustion process model building module and a flue gas treatment process model building module, the in-furnace combustion process model input selection module is connected with the in-furnace combustion process model building module, and the in-furnace combustion process model building module is connected with the flue gas treatment process model building module;

step 2: and training the MSWI full-process model based on the echelon series-parallel training mechanism.

Optionally, in step 1, a MSWI full-process model is built, specifically:

building an in-furnace combustion process model input selection module based on experience cognition, and selecting the input of the in-furnace combustion process model according to the experience cognition of field experts, wherein the model comprises a fire grate speed, combustion air flow and an SNCR (selective non-catalytic reduction) system;

the speed of the grate is controlled by controlling the speed of the material pusher and the stage speed of drying the grate, the input of a combustion process model in the furnace is simplified, and the speed u of the material pusher is controlled _FAS And velocity u of drying grate stage _DGAS Taking an average value to obtain:

the combustion air flow consists of primary air flow and secondary air flow, the opening of a primary air baffle is set to be a fixed proportion according to the experience of field experts, and the primary air flow u is set _PriAir And secondary air flow u _SecAir As an air volume input feature;

the amount of the ammonia water fed into the furnace is controlled to control the SNCR systemAccording to the actual spraying position of the SNCR system in the MSWI process, the ammonia water is added into the furnace

The hearth temperature model added into the combustion process in the furnace is used as an input characteristic, and the basic input characteristic of the finally obtained combustion process model in the furnace is as follows:

optionally, in step 1, a MSWI full-flow model is built, specifically:

building an in-furnace combustion process model building module based on XGboost, wherein the in-furnace combustion process model building module comprises a hearth temperature model, a boiler steam flow model and a G1 flue gas oxygen content model, the output of the hearth temperature model is connected with the boiler steam flow model and the G1 flue gas oxygen content model, the output of the boiler steam flow model is connected with the G1 flue gas oxygen content model, and the outputs of the hearth temperature model, the boiler steam flow model and the G1 flue gas oxygen content model are connected with a flue gas treatment process model building module;

building a hearth temperature model, specifically: determining an FT model based on basic input features of the in-furnace combustion process model as follows:

combining with an FT model, building a hearth temperature model based on XGboost, and controlling the complexity of the model by introducing a regular term into a loss function of the XGboost, wherein the loss function of the XGboost can be expressed as:

in the formula, c ^XGBoost Representing the predicted value of the next base learner in the XGBoost model,

expressing the predicted value of the jth base learner in the XGboost model, carrying out Taylor expansion on the loss function, and decomposing to obtain:

simplifying it, and simplifying the first derivative and the second derivative as G and H, to obtain:

according to the simplified formula, the following results are obtained:

deleting entries without output values, resulting in:

and the derivative is obtained, the derivative of the above formula is equal to 0, and the optimal solution is obtained as follows:

repeating the operation until the number of the base learners reaches a set value, and determining that the hearth temperature model is as follows:

building a boiler steam flow model, specifically: adding on the basis of a hearth temperature model

As model inputs, the boiler steam flow model is determined as follows:

a G1 flue gas oxygen content model is built, and the method specifically comprises the following steps: adding on the basis of a boiler steam flow model

As model input, determining a G1 flue gas oxygen content model as follows:

optionally, in step 1, a MSWI full-process model is built, specifically:

a flue gas treatment process model building module based on MI and XGboost is built, and the input characteristics of the flue gas treatment process model are expressed as follows:

wherein u is _FGCP ＝{u _C ,u _Ca(OH)2 And (4) deleting the constant value to obtain the input characteristics of the flue gas treatment process model as follows:

and (3) screening input characteristics by combining mutual information and an MSWI process mechanism, and finally determining a flue gas treatment process model as follows:

optionally, in step 2, training the MSWI full-process model based on the echelon series-parallel training mechanism specifically includes:

performing echelon serial training on the combustion process model in the boiler, namely inputting the output of the hearth temperature model to a boiler steam flow model for training after the training of the hearth temperature model is finished, and inputting the output of the boiler steam flow model to a G1 flue gas oxygen content model for training after the training is finished to finish the training;

after the training of the combustion process model in the furnace is finished, the output of the combustion process model is subjected to feature selection, the flue gas treatment process model is subjected to parallel training, and after the training is finished, a final MSWI full-flow model based on a echelon series-parallel training mechanism is determined.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects: the invention provides a control-oriented data-driven urban solid waste incineration full-flow modeling method which comprises the steps of firstly, selecting input characteristics of an incineration process model in a furnace based on experience cognition; then, constructing a series model of the incineration process in the furnace by utilizing XGboost; then, selecting the input characteristics of the flue gas treatment process model based on the mutual information analysis result, and performing related modeling; finally, training the proposed full-flow series-parallel model in a echelon mode; the effectiveness of the model is verified through actual operation data of the factory, and reference is provided for the research of the advanced intelligent optimization control algorithm in the subsequent MSWI process.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

FIG. 1 is a schematic flow chart of a control-oriented data-driven urban solid waste incineration full-process modeling method in an embodiment of the invention;

FIG. 2 is a MSWI power plant process flow diagram;

FIG. 3a is a graph of the correlation coefficient between the operating variable and the hearth temperature during combustion in the furnace;

FIG. 3b is a graph of the correlation coefficient between the manipulated variable and the boiler steam flow during combustion in the boiler;

FIG. 3c is a graph showing the correlation coefficient between the manipulated variable and the oxygen content of G1 portion of flue gas during combustion in the furnace;

FIG. 4a is a graph of the correlation coefficient between the manipulated variable and the oxygen content of G3 portion of flue gas during flue gas treatment;

FIG. 4b is a graph of correlation coefficients between manipulated variables and the amount of NOx in the G3 portion of the flue gas treatment process;

FIG. 4c is a graph of the correlation coefficient between the manipulated variables and the CO content of the G3 portion of the flue gas treatment process;

FIG. 4d is a graph of the correlation coefficient between the manipulated variable and the CO2 content of the G3 portion of the flue gas treatment process;

FIG. 4e is a graph of the correlation coefficient between the manipulated variable and the acid gas content of the G3 portion of the flue gas treatment process;

FIG. 5 is a schematic diagram of the MSWI full flow model architecture;

FIG. 6 is a schematic diagram of a MSWI full-process model training process;

FIG. 7a is a schematic diagram showing the results of MI evaluation of manipulated variables and the oxygen content of G3 portion of flue gas during flue gas treatment;

FIG. 7b is a graphical illustration of the MI evaluation of manipulated variables versus NOx content of the G3 portion of the flue gas treatment process;

FIG. 7c is a graphical representation of the MI evaluation of manipulated variables versus CO content of the G3 portion of the flue gas treatment process;

FIG. 7d is a graphical illustration of the MI evaluation of manipulated variables versus CO2 content of the G3 portion of the flue gas treatment process;

FIG. 7e is a schematic illustration of the MI evaluation results of the manipulated variables versus the acid gas content of the G3 portion of the flue gas treatment process;

FIG. 8a is a diagram illustrating the results of a furnace temperature test;

FIG. 8b is a schematic diagram showing the results of a steam flow test of a boiler;

FIG. 8c is a graph showing the results of the oxygen content test of G1 flue gas;

FIG. 8d is a graph showing the results of the oxygen content test on part G3;

FIG. 8e is a graph showing the results of a partial G3NOx test;

FIG. 8f is a graph showing the results of the CO content test in part G3;

FIG. 8G is a graph showing the results of the CO2 content test on part G3;

FIG. 8h is a graph showing the results of the acid gas content test for part G3;

FIG. 9 is a MSWI process mutual information result diagram;

FIG. 10 is a schematic diagram of the MSWI process key controlled variable full flow control process.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

As shown in fig. 1, the control-oriented data-driven urban solid waste incineration full-process modeling method provided by the embodiment of the invention comprises the following steps:

and 2, step: and training the MSWI full-process model based on the echelon series-parallel training mechanism.

The technological process of certain MSWI power plant is shown in figure 2, MSW is transported by vehicles, is discharged into a solid waste pool after being weighed by weighbridge, and is put into a hopper by a grab bucket after biological fermentation and dehydration for 3-7 days; then, the feeder pushes MSW to the fire grate, after three stages of drying, burning and burning out, the burning process ensures that harmful substances in the high-temperature flue gas are fully decomposed and burnt, the temperature of the flue gas is controlled to be above 850 ℃, the flue gas stays for more than 2 seconds and has enough flue gas turbulence; then, the high-temperature flue gas enters a waste heat boiler, and high-temperature steam generated by heat exchange drives a steam turbine generator unit to generate electricity; then, the flue gas mixed with lime and activated carbon enters a deacidification reactor to perform neutralization reaction, and DXN and heavy metals in the flue gas are adsorbed; next, removing flue gas particles, neutralizing reactants and an activated carbon adsorbent in a bag type dust collector; and finally, discharging waste gas containing other substances such as dust, CO, NOx, SO2, HCL, HF, hg, cd, DXN and the like into the atmosphere through a chimney, conveying ash slag generated by burning to a slag pit through a slag dragging machine, and then conveying the ash slag to a specified place for landfill by a vehicle.

In order to realize the purpose of minimum harmless risk in the MSW treatment process, the existing MSWI process control mainly follows the principle of '3T + E', namely, the temperature of a hearth is more than 850 ℃, the retention time of flue gas is more than 2 seconds, certain flue gas turbulence intensity and sufficient air quantity. The principle of '3T + E' can effectively ensure that harmful substances generated in the combustion process are fully decomposed, and the generation of acid gas and DXN is controlled from the source. The method is limited by the difference of MSWI power plants at home and abroad and the uncertainty of MSW components, the ACC system at home is difficult to operate normally at present, and the method mainly depends on a manual control mode of field experts, as shown in FIG. 10.

Wherein:

γ _G3 ＝{γ _NOx ,γ _CO ,γ _CO2 ,γ _Acid } (1)

the legend means:

is a flow rate measuring instrument, and is characterized in that,

is a steam detection instrument and is characterized in that,

is an oxygen content detecting instrument and a method for detecting oxygen content,

is a smoke detection instrument, which is a smoke detection instrument,

is a temperature measuring instrument, and is characterized in that,

in order to be a dose controller for a patient,

is a frequency converter controller, and is characterized in that,

is a fire grate speed controller, which is a fire grate speed controller,

is a baffle opening controller;

the physical meaning of the symbols is: r is _max To set the target upper limit, r _min To set the target lower limit, r ^* To set the target value, w _Speed For the set value of the grate velocity, w _Baffle Is a set value of the opening degree of the primary air baffle plate, w _PriAir Is a primary air flow set value, w _SecAir Is a set value of the flow rate of the secondary air,

is the set values of the left inner speed, the left outer speed, the right inner speed and the right outer speed of the feeder,

is the set values of the left inner speed, the left outer speed, the right inner speed and the right outer speed of the drying grate,

setting values of the left inner speed, the left outer speed, the right inner speed and the right outer speed of the 1-section fire grate,

setting values of the left inner speed, the left outer speed, the right inner speed and the right outer speed of the 2-section fire grate,

the set values of the left inner speed and the right inner speed of the burnout fire grate are set,

is the opening set value of the left and right air inlet pipeline baffles of the drying grate 1 section,

is the opening set value of the left and right air inlet pipeline baffles of the drying grate 2 section,

is a set value of the opening degree of the left and right air inlet pipeline baffles of the 1-1 section of the combustion grate,

is the opening set value of the left and right air inlet pipeline baffles of the 1-2 sections of the combustion grate,

is the opening set value of the left and right air inlet pipeline baffles of the 2-1 section of the combustion grate,

is the opening set value of the left and right air inlet pipeline baffles of the 2-2 sections of the combustion grate,

is the opening set value w of the left and right air inlet pipeline baffles of the burnout grate _PriAir Is a primary air flow set value, w _SecAir Is a set value of the secondary air flow, w _NH3·H2O Set value for the amount of ammonia injected, w _C Set value for activated carbon, w _Ca(OH)2 The setting value of the slaked lime is set,

the control quantity of the left inner, left outer, right inner and right outer speed control mechanism of the feeder,

the control quantity of the left inner, left outer, right inner and right outer speed control mechanism of the drying grate,

the control quantity of the speed control mechanism for burning the left inner, the left outer, the right inner and the right outer of the 1-section fire grate,

the control quantity of the speed control mechanism for controlling the left inner, left outer, right inner and right outer speeds of the 2-section fire grate is combusted,

the control quantity of the left inner speed control mechanism and the right inner speed control mechanism of the burnout fire grate,

the control quantity of the opening control mechanism for the left and right air inlet pipeline baffles of the drying grate 1 section is controlled,

the control quantity of the opening control mechanism for the left and right air inlet pipeline baffles of the drying grate 2 section,

the control quantity of the baffle opening control mechanism of the left and right air inlet pipelines at the 1-1 section of the combustion grate,

the control quantity of the baffle opening control mechanism of the left and right air inlet pipelines at the 1-2 sections of the combustion grate,

the control quantity of the baffle opening control mechanism of the left and right air inlet pipelines at the 2-1 section of the combustion grate,

the control quantity of the baffle opening control mechanism of the left and right air inlet pipelines at the 2-2 sections of the combustion grate,

the control quantity u of the opening control mechanism for the left and right air pipe baffles of the burnout grate _PriAir For the primary air flow control mechanism control quantity u _SecAir Is twoControl quantity of secondary air flow control mechanism, u _NH3·H2O The ammonia water control mechanism control quantity u _C Control of mechanism control for activated carbon, u _Ca(OH)2 The control quantity of the slaked lime control mechanism is controlled,

is a detected value of the temperature of the fire grate,

the temperature detection values of the left inner, the left outer, the right inner and the right outer of the drying grate,

for burning the left inner, left outer, right inner and right outer temperature detection values of the 1-section fire grate,

the left inner, left outer, right inner and right outer temperature detection values of the 2-section fire grate are burnt,

the left inner temperature and the right inner temperature of the burn-out grate are detected,

the opening detection values of the left and right air inlet pipeline baffles of the drying grate 1 section are obtained,

the opening detection values of the left and right air inlet pipeline baffles of the drying grate 2 section are obtained,

the opening detection values of the left and right air inlet pipeline baffles of the 1-1 section of the combustion grate are obtained,

the opening detection values of the left and right air inlet pipeline baffles of the 1-2 sections of the combustion grate,

the opening detection values of the left and right air inlet pipeline baffles at the 2-1 section of the combustion grate,

the opening detection values of the left and right air inlet pipeline baffles of the 2-2 sections of the combustion grate are obtained,

the opening detection values of the left and right air inlet pipeline baffles of the burn-out grate are y _PriAir As primary air flow detection value, y _SecAir As a measure of the secondary air flow, y _FT As a value of the furnace temperature, y _BSF Is a detected value of steam flow of the boiler, y _G1OX Is a detection value of oxygen content in G1 part of smoke, y _G3OX Is a detection value of oxygen content of G3 part of smoke, gamma _G3 Is the partial contaminant concentration of G3, γ _NOx Is the G3 partial NOx content, γ _CO Is the CO content of the G3 fraction, γ _CO2 As part of CO G3 ₂ Content, gamma _Acid Is the G3 partial acid gas content.

As can be seen from fig. 10, the inputs to the MSWI process can be expressed as:

wherein u is _FCP And u _FGCP Respectively representing the control input of the combustion process and the flue gas treatment process in the furnace, and concretely comprising the following steps:

u _FGCP ＝{u _C ,u _Ca(OH)2 } (7)

the domain expert manual control strategy is as follows:

step 1: according to production requirements and pollution emission indexes, field experts set key controlled variables in the MSWI process according to abundant expert experiences;

and 2, step: according to the target set value and range set by experience, each loop control mechanism carries out calculation to obtain the control quantity of the corresponding actuating mechanism, and the control quantity is input into the MSWI process to achieve the purpose of stable operation;

and step 3: according to the detection result of the physical detection equipment on the controlled variable, various detection information is uploaded to a large screen of a monitoring room through a Distributed Control System (DCS) System to realize real-time online monitoring, and functions of generating a report, displaying data trend, alarming and the like are realized

And 4, step 4: according to the real-time online monitoring result, the field expert combines the expert experience to judge and decide the current operation condition, and the step 1 to the step 3 are circulated, so that the stable operation of the MSWI process is ensured.

From the above, the process perceived by the domain expert can be described as:

it can be seen that the incineration operation control depending on the domain experts has the problems of great randomness, hysteresis and the like, thereby causing the MSWI process control to be inefficient. Because of the safety of the actual industrial field and the complexity of the MSWI process, the intelligent optimization control algorithm is difficult to be directly applied to the actual industrial field, so the MSWI full-flow model is provided for researching the control algorithm of the controlled variable of the combustion process in the furnace and the optimization of the pollutant emission index of the smoke treatment process.

According to the MSWI process, the method is divided into two parts: a furnace combustion process and a flue gas treatment process. The combustion process in the Furnace comprises hearth temperature (FT), boiler Steam Flow (BSF) and Oxygen content of G1 part of flue gas (G1 OX); the flue gas treatment process comprises the Oxygen content of the G3 flue gas (Oxygen content of gas in G3 part, G3 OX), the NOx content of the G3 part (G3 NOx), the CO content of the G3 part (G3 CO), the CO2 content of the G3 part (G3 CO 2) and the Acid gas content of the G3 part (G3 Acid).

1) Combustion process in furnace

The relationship between the operating variables and the key controlled variables of the combustion process in the furnace is subjected to factor analysis by adopting Pearson Correlation Coefficient (PCC), and the specific result is shown in FIG. 3. Wherein the blue bar column represents positive correlation, the red bar column represents negative correlation, and the absolute value represents correlation strength;

as can be seen from FIG. 3, different positive and negative correlation relationships exist between each operating variable and the key controlled variables of the combustion process in the furnace, and the positive and negative correlation of the same operating variable in different key controlled variables are different, for example, the overfire air has a positive correlation with FT and has a negative correlation with BSF and G1 OX; the correlation between manipulated variables and FT, BSF is stronger than G1OX.

2) Flue gas treatment process

The correlation between the key controlled variables and the manipulated variables of the flue gas treatment process is shown in figure 4. Wherein the blue bar column represents positive correlation, the red bar column represents negative correlation, and the absolute value represents correlation strength;

as can be seen from fig. 4, compared with the furnace incineration process, the correlation between the key controlled variable/pollutant emission concentration and the operation variable in the G3 part is weak, the PCC absolute value is generally within 0.2, and there are also cases where the same operation variable contributes to different key controlled variable/pollutant emission concentrations in different ways; but the operating variables have a stronger effect on NOx levels than other key controlled variables/pollutant emission concentrations;

in summary, the relationship between the key controlled variable/pollutant emission concentration and the operation variable of the MSWI whole process is complex, and it is difficult to simply select the model input features according to the process sequence, so the invention distinguishes and screens the key controlled variable/pollutant emission concentration model input features of different parts of the MSWI process.

As shown in fig. 5, according to a certain MSWI power plant process flow in beijing, a control-oriented data-driven urban solid waste incineration full-flow modeling method provided by the invention is adopted to establish a control-oriented data-driven MSWI full-flow model, which comprises an in-furnace incineration process model input selection module based on experience cognition, an in-furnace combustion process model construction module based on XGBoost and a flue gas treatment process model construction module based on MI and XGBoost, wherein ammonia water, activated carbon and slaked lime are used as harmful substance removal agents, and an input model input characteristic is added according to an actual spraying position;

wherein u is _FAS And u _DGAS Respectively showing the uniform speed control quantity u of the feeder and the uniform speed control quantity u of the drying grate _NH3·H2O Represents the amount of ammonia injected, u _C Denotes the amount of activated carbon injected, u _Ca(OH)2 The injection amount of the slaked lime is expressed;

and

represents FT, BSF, G1OX and G3OX model outputs, respectively;

and

respectively representing the output of each pollutant model; the model input characterized by the solid end arrow is

The model input represented by the dashed arrow is

Based on analysis of influence factors of the operating variables and the controlled variables of the combustion process in the furnace, a combustion process model is further selected and input according to experience cognition of field experts. In order to meet the production requirements and meet the emission standards of pollution, the MSWI process stability control is the main target of field expert operation. Generally, after considering daily operation plan scheduling, operation condition recording, operation work order during operation, MSWI process real-time monitoring information, MSWI operation workshop equipment state and other information, a field expert adjusts the control quantity of an execution mechanism in each control loop according to a current target set value so as to achieve the purpose of stable operation. Therefore, the MSWI process considers the strong coupling of the above-mentioned many factors to effectively achieve stable control;

under the conventional operation condition, the operation conditions of unconventional conditions such as furnace starting, furnace stopping, low-efficiency operation maintenance and the like of the MSWI factory are not considered, and the input selection of the combustion process model in the furnace is analyzed through the actual case of the MSWI operation process. The in-furnace combustion process includes grate speed control, combustion air flow control and SNCR system control with the objective of achieving a stable combustion process, uniform steam mass, complete combustion of MSW and minimal harmful gas production. FT, BSF and G1OX have significant follow-up as controlled variables of the combustion process, for example: high FT gives large BSF and small G1 OX;

in actual operation, when abnormal fluctuations occur in FT, the grate speed control and the combustion air flow rate control are the operations that are first considered by experts in the field. The increasing and decreasing of the grate speed can be used for controlling the thickness of the MSW to achieve the purposes of accelerating and delaying combustion, experts in the field at present mainly control the speed of a pusher and the speed of a drying grate stage to achieve the purposes, and the speeds of grates in other sections keep larger pushing speed to maintain the daily processing speed of the MSW. Determining pusher velocity u to reduce model input features _FAS And drying grate stage velocity u _DGAS Carrying out variable averaging processing on the similar operation as follows:

in fig. 10, the combustion air flow control consists of primary air flow and secondary air flow, where the primary air control contains 14 branches. The primary air quantity is mainly used for drying and supporting combustion of MSW, the secondary air quantity is mainly used for promoting further combustion of combustible components in flue gas and increasing turbulence of flue gas, and the distribution control of primary air and secondary air is also usedThe excess air factor of the furnace is determined. The field expert mainly controls the total primary air amount (the total primary air amount is distributed by 14 branches in a fixed proportion) and the secondary air auxiliary control strategy to realize combustion control. Therefore, the opening of the primary air baffle is set to be a fixed proportion based on field expert experience, and is not used as an input characteristic for model training; primary air flow u _PriAir And secondary air flow u _SecAir As an air volume input feature.

In order to minimize the generation of NOx, the SNCR control system sprays ammonia water into the furnace to realize the standard emission of NOx, and field experts mainly realize the aim by controlling the charging amount of the ammonia water. Therefore, according to the actual spraying position in the MSWI process, the furnace temperature model of adding the ammonia water into the furnace in the combustion process in the furnace is used as an input characteristic.

In conclusion, the basic input characteristics of the in-furnace incineration process model are determined as follows:

an XGboost-based in-furnace combustion process model construction module comprises:

1) Hearth temperature model

Firstly, inputting the output result of the selection module based on the experience-recognized in-furnace incineration model, and determining that the FT model can be expressed as follows:

the following modeling process for constructing the XGboost in combination with the FT model:

the XGboost model is improved on the basis of a Gradient Boosting Decision Tree (GBDT) algorithm, a new base learner is generated by correcting the residual error of the trained learner, and the prediction results of all the base learners are summed to form a prediction result.

Taking the ith model as an example, describing the XGboost model construction process, specifically as follows:

the XGBoost controls the complexity of the model and prevents overfitting by introducing a regularization term into the penalty function, which can be expressed as:

simplifying equation (34) and simplifying the first and second derivatives as G and H yields:

from the simplified formula, formula (27) is expressed as:

next, to minimize the loss function, the term without output value in equation (36) is deleted as follows:

the method comprises the following steps of building a boiler steam flow model, specifically: adding on the basis of a hearth temperature model

As model inputs, the boiler steam flow model is determined as follows:

detecting G1OX at the position of a hearth outlet, and building a G1 flue gas oxygen content model, specifically: adding on the basis of a boiler steam flow model

As model input, determining a G1 flue gas oxygen content model as follows:

considering u of the current MSWI procedure _FGCP ＝{u _C ,u _Ca(OH)2 And (4) deleting the constant value to obtain the input characteristics of the flue gas treatment process model as follows:

the input features are screened by combining Mutual Information (MI) and MSWI process mechanism, taking G3OX as an example, the k-th input feature and the Mutual Information calculation formula between the k-th input feature and the k-th input feature are as follows:

wherein the content of the first and second substances,

and ρ _prob (y _G3OX ) To represent

And y _G3OX The boundary probability density of (a) is,

representing the joint probability density, H (y) _G3OX ) The entropy of the information is represented and,

representing a conditional entropy;

the MI value in the rejection characteristic input selection module is less than a set threshold value theta _MI Finally determining a flue gas treatment process model and recording as follows:

wherein the content of the first and second substances,

representation based on theta _MI In that

Medium G3 flue gas oxygen content model f _G3OX Input of (i) selection, i.e. presence

Similarly, the NOx model, the CO2 model, and the acid gas model may be expressed as: :

wherein the content of the first and second substances,

and

respectively represent a base on theta _MI At x _FGCP In the NOx model, CO ₂ Model and acid model selection as model inputs, i.e.

The MSWI full-flow model training process sequence is specifically shown in FIG. 6, wherein the input and output of each model are shown in FIG. 5, as can be seen from FIG. 6, according to the MSWI full-flow processing process, because the burning process in the furnace is performed stage by stage, the echelon serial training of the combustion process model in the furnace is determined, that is, after the training of the previous model is completed, the output of the previous model is added to the input characteristic of the next model for training until all the burning process models in the furnace are trained; meanwhile, harmful substances are removed uniformly in the stage of the flue gas treatment process, so that a parallel training mode is adopted for a flue gas treatment process model, namely after the model of the incineration process in the furnace is trained, the output of the model is subjected to characteristic selection, the flue gas treatment process model is trained in parallel, and until all models are trained, the MSWI full-flow model based on the echelon series-parallel training mechanism is determined.

To verify the accuracy of the model, the invention adopts the following steps of 3 months, 19 days and 8 days of 2021 year of a certain MSWI power plant in Beijing: 00 to 24:00 total 16 hours of continuous operation. And (3) removing abnormal values after taking 60s of mean data of the process data, and finally totaling 857 groups of data, and adopting interval sampling, wherein 50% of the data are used for a training set, 25% of the data are used for a test set, and 25% of the data are used for a verification set.

The relationship between the manipulated variable and the key controlled variable/pollutant emission concentration in the flue gas treatment process was evaluated by MI, and the specific results are shown in fig. 7.

Wherein the characteristic sequence is as follows: primary air flow, secondary air flow, uniform feeder speed, uniform drying grate speed, ammonia water, FT, BSF, G1OX, lime and activated carbon.

As can be seen from fig. 7, the key controlled variables/pollutant emission concentration influencing factors of the operation variables in different flue gas treatment processes are different, so that the characteristics with MI values smaller than 0.3 are removed for relevant model modeling according to the MI evaluation results.

Correspondingly, the input-output relationship of the flue gas treatment process model can be expressed as;

1) G3 flue gas oxygen content model

The input of the G3 flue gas oxygen content model is the uniform speed of a feeder, the uniform speed of a drying grate, the primary air flow, the secondary air flow, the ammonia injection amount, the FT prediction output, the BSF prediction output and the G1OX prediction output, the output is the G3 flue gas oxygen content which can be expressed as:

2) NOx model

The model inputs are feeder uniform speed, drying grate uniform speed, primary air flow, secondary air flow, ammonia water injection amount, FT prediction output, BSF prediction output and G1OX prediction output, the output is the content of partial NOx of G3, and can be expressed as:

3) CO model

The model inputs primary air flow, secondary air flow, ammonia water injection amount, FT prediction output, BSF prediction output and G1OX prediction output, the output is the CO content of a G3 part, and can be expressed as:

4)CO ₂ model (model)

The model inputs include uniform feeder speed, uniform drying grate speed, primary air flow, secondary air flow, ammonia water injection amount, FT prediction output, BSF prediction output and G1OX prediction output, and the output is G3 flue gas CO ₂ The amount, can be expressed as:

5) Acid gas model

The model inputs primary air flow, secondary air flow, ammonia water injection amount, FT prediction output, BSF prediction output and G1OX prediction output, the output is the content of G3 part of acid gas, and can be expressed as:

for the above 8 models, a echelon series-parallel training mechanism is adopted, and the parameter settings are shown in table 3. The RMSE values are shown in Table 4. The effect of the fit is shown in figure 8.

TABLE 3 model parameter settings

TABLE 4 model RMSE

As can be seen from table 4 and fig. 8, the proposed MSWI full-flow tandem model has good approximation capability, but in the case of a true value with large fluctuation, the predicted value is low.

For comparison with the method provided by the present invention, the relationship between the manipulated variable and the key controlled variable/pollutant emission concentration is evaluated by using MI, and the specific result is shown in fig. 9;

as can be seen from fig. 9, the influence relationship between the primary air flow, the secondary air flow, the left inner feeder speed, the left outer feeder speed, the right inner feeder speed, the right outer feeder speed, the left inner drying grate speed, the left outer drying grate speed, the right inner drying grate speed, the right outer drying grate speed and the key controlled variable/pollutant emission concentration is higher than the correlation coefficient of other manipulated variables; at the same time, the impact of different key controlled variables/pollutant emission concentrations is different. The MI result is basically consistent with the analysis based on experience cognition in the invention, so that the input characteristics of the incineration process model in the furnace can be further determined as follows: { u _FAS ,u _DGAS ,u _PriAir ,u _SecAir ,u _NH3·H2O }。

A typical complex industrial process MSWI process has the characteristics of multiple input, multiple output and multiple coupling, and the modeling of a single controlled object is difficult to characterize the whole process and the process characteristics of the single controlled object. Aiming at the problem, the invention constructs a control-oriented data-driven MSWI full-process model, and the contribution is as follows: a full-process series-parallel model structure is designed by combining the MSWI process flow, and reference is provided for full-process modeling research based on data driving; and determining input and output by using a process mechanism and expert experience according to the process model by model, and performing model training according to the flow processing sequence in a gradient series-parallel connection mode. The MSWI full-flow model fitting process mechanism provided by the invention can lay a foundation for a subsequent advanced intelligent optimization control algorithm.

The invention provides a control-oriented data-driven urban solid waste incineration full-flow modeling method which comprises the steps of firstly, selecting input characteristics of an incineration process model in a furnace based on experience cognition; then, constructing a series model of the incineration process in the furnace by utilizing XGboost; then, selecting the input characteristics of the flue gas treatment process model based on the mutual information analysis result, and performing related modeling; finally, training the proposed full-flow series-parallel model in a echelon mode; the effectiveness of the model is verified through actual operation data of the plant, and reference is provided for the research of the advanced intelligent optimization control algorithm in the subsequent MSWI process.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims

1. A control-oriented data-driven urban solid waste incineration full-process modeling method is characterized by comprising the following steps:

2. The control-oriented data-driven urban solid waste incineration full-process modeling method for the control-oriented data-driven urban solid waste incineration according to claim 1, wherein in the step 1, an MSWI full-process model is built, and specifically:

building an input selection module of a furnace combustion process model based on experience cognition, and selecting the input of the furnace combustion process model according to the experience cognition of field experts, wherein the input selection module comprises a fire grate speed, a combustion air flow and an SNCR (selective non-catalytic reduction) system;

the grate speed is controlled by controlling the speed of the pusher and the stage speed of drying the grate, the input of a combustion process model in the furnace is simplified, and the speed u of the pusher is controlled _FAS And velocity u of drying grate stage _DGAS Taking an average value to obtain:

the combustion air flow consists of primary air flow and secondary air flow, the opening of a primary air baffle is set to be a fixed proportion according to the experience of a field expert, and the primary air flow u is set _PriAir And secondary air flow u _SecAir As an air volume input feature;

controlling the amount of the ammonia water to enter the furnace to control the SNCR system, and controlling the amount of the ammonia water to enter the furnace according to the actual spraying position of the SNCR system in the MSWI process

3. the control-oriented data-driven urban solid waste incineration full-process modeling method for the control-oriented data-driven urban solid waste incineration according to claim 2, wherein in the step 1, an MSWI full-process model is built, specifically:

the method comprises the following steps of (1) building a hearth temperature model, specifically: based on the basic input features of the in-furnace combustion process model, determining an FT model as follows:

and (3) representing the predicted value of the jth base learning device in the XGboost model, carrying out Taylor expansion on the loss function, and decomposing to obtain:

according to the simplified formula, the following results are obtained:

deleting entries without output values, resulting in:

As model inputs, the boiler steam flow model is determined as follows:

As model input, determining a G1 flue gas oxygen content model as follows:

4. the control-oriented data-driven urban solid waste incineration full-process modeling method for the control-oriented data-driven urban solid waste incineration according to claim 3, wherein in the step 1, an MSWI full-process model is built, specifically:

wherein the content of the first and second substances,

and (3) deleting the constant value to obtain the input characteristics of the flue gas treatment process model as follows:

5. the control-oriented data-driven urban solid waste incineration full-process modeling method for the control-oriented system according to claim 4, wherein in the step 2, the MSWI full-process model is trained based on a echelon series-parallel training mechanism, and specifically comprises the following steps:

performing echelon serial training on the combustion process model in the boiler, namely inputting the output of the hearth temperature model to a boiler steam flow model for training after the training of the hearth temperature model is finished, and inputting the output of the boiler steam flow model to a G1 flue gas oxygen content model for training to finish the training after the training is finished;