Intelligent control method for fly ash coal oxygen-enriched melting based on multi-modal coupling prediction

Abstract

The embodiment of the application discloses a fly ash coal oxygen-enriched melting intelligent control method based on multi-modal coupling prediction, and belongs to the cross technical field of solid waste resource utilization and intelligent control. The method comprises the following steps: S101, acquiring multi-source data collected by a sensor; S102, obtaining control parameters through an intelligent control algorithm according to the multi-source data; S103, realizing oxygen-enriched lance, slurry outlet and material ratio control according to the control parameters; wherein S102 comprises: for temperature T, furnace negative pressure P, flue gas oxygen concentration C O and slurry yield Q p , a multivariate decoupling prediction model is used for decoupling prediction; for the furnace negative pressure P, a fuzzy self-adaptive PID controller is used for fuzzy control; and a multi-objective optimization algorithm is used to calculate the control parameters. The control method has small melting temperature fluctuation, strong furnace negative pressure stability, high quartz sand utilization rate, low ton slurry energy consumption and high qualified rate of thermal insulation cotton.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of solid waste resource utilization and intelligent control, in particular to a fly ash coal oxygen-enriched melting intelligent control method based on multi-modal coupling prediction. BACKGROUND

[0002] Fly ash is mainly derived from industrial facilities such as coal-fired power plants and waste incineration plants, and contains a large amount of fine particulate matter and harmful substances such as heavy metals and dioxins. If not properly treated, it can pose a serious threat to the environment and human health. Fly ash melting treatment is an effective waste treatment technology that converts fly ash into stable glassy products through high-temperature melting, achieving waste harmlessness and resource utilization, not only reducing harmful substances in fly ash, but also converting them into valuable building materials (such as insulation cotton, etc.).

[0003] In the existing fly ash coal oxygen-enriched melting system for preparing insulation cotton, traditional segmented PID control methods and single-target optimization control methods are usually used, wherein:

[0004] In the traditional segmented PID control method, the control strategy is to use segmented PID control to adjust single parameters such as melting temperature and furnace negative pressure (for example, setting PID controllers to control temperature and pressure respectively); the implementation is to use a PID regulator with fixed parameters to rely on the experience of operators to set PID parameters (such as proportional coefficient and integral time) and to perform simple closed-loop control on real-time data fed back by sensors; and the applicable scenario is to maintain basic temperature and pressure control under conditions of stable fly ash composition and small changes in working conditions.

[0005] The disadvantage of this control method is that there is a problem of multi-variable strong coupling: no coupling model is established for fuel quantity, oxygen concentration, material ratio, temperature, furnace negative pressure, and molten slurry yield, and dynamic interaction between variables (such as changes in fuel quantity affecting temperature and negative pressure) leads to serious regulation lag (pure temperature response lag of 16-28s) and easy risk of negative pressure exceeding the limit (>200Pa).

[0006] In the single-target optimization control method, the control target is to optimize a single index (such as energy consumption or yield), for example, to minimize unit energy consumption by adjusting fuel quantity and oxygen flow; the implementation is to adjust operating parameters through a single-target optimization algorithm (such as gradient descent method) based on a fixed mechanism model (such as heat transfer equation Q=kAΔT) or empirical formula; and the applicable scenario is to achieve local optimization under conditions of relatively stable working conditions and single target.

[0007] The control method has the following disadvantages: (1) time-varying nonlinear system: fly ash chlorine content fluctuation (±15%) causes furnace slagging, changes the heat transfer coefficient k (fluctuation range ±20%), and the conventional model Q=kAΔT cannot be corrected in real time, resulting in temperature control deviation (±50℃); (2) multi-dimensional control target conflict: the molten slurry yield Q p ≥2t / d, insulation cotton fiber diameter (1-4μm, qualified rate ≥90%), furnace negative pressure P∈(-200, 20Pa), and traditional single-target control is difficult to balance. SUMMARY

[0008] Therefore, the embodiment of the present application provides a fly ash coal oxygen-enriched melting intelligent control method based on multi-modal coupling prediction to solve the problems of multi-variable strong coupling, time-varying nonlinear characteristics and multi-dimensional control target conflict in the process of preparing insulation cotton by oxygen-enriched melting of fly ash from waste incineration.

[0009] A fly ash coal oxygen-enriched melting intelligent control method based on multi-modal coupling prediction, comprising:

[0010] Step S101: acquiring multi-source data collected by a sensor;

[0011] Step S102: obtaining control parameters by an intelligent control algorithm according to the multi-source data;

[0012] Step S103: realizing oxygen-enriched lance, slurry outlet and material ratio control according to the control parameters;

[0013] In the step S102, the following steps are included:

[0014] Step S1021: for temperature T, furnace negative pressure P, flue gas oxygen concentration C O and molten slurry yield Q p , a multi-variable decoupling prediction model is used for decoupling prediction;

[0015] Step S1022: for furnace negative pressure P, a fuzzy adaptive PID controller is used for fuzzy control;

[0016] Step S1023: a multi-objective optimization algorithm is used to calculate the control parameters.

[0017] Preferably, in the step S101, the multi-source data collected by the sensor includes:

[0018] Temperature data collected by K-type thermocouples arranged at the bottom, top and outlet of the incinerator;

[0019] Furnace negative pressure data collected by a pressure transmitter arranged at the waist center of the incinerator furnace;

[0020] The flue gas oxygen concentration data collected by the zirconia oxygen analyzer deployed at the outlet of the flue gas cooler;

[0021] The frequency data collected by the frequency converter of the screw conveyor deployed at the coal / fly ash / quartz sand / alumina feeding port.

[0022] Preferably, in the step S1021, the multivariate decoupling prediction model describes the melting process using the following state space equation:

[0023] ;

[0024] wherein X=[T, P, C O , Q p ] T : state vector; T: temperature in the melting furnace; P: negative pressure in the furnace; C O : oxygen concentration of flue gas; Q p : slurry yield;

[0025] U=[m c , V O , r] T : control vector; m c : fuel coal feeding amount; V O : oxygen flow rate; r: material ratio, i.e., mass ratio of fly ash: quartz sand: alumina;

[0026] Y: output vector, corresponding to sensor measurement; A: system matrix, describing the natural dynamic relationship between state variables; B: input matrix, representing the coupling effect of control variables on state variables; C: output matrix, mapping state variables to sensor measurement; W: process noise; V: measurement noise;

[0027] In the input matrix B, the coupling coefficient is defined based on the mechanism model, the LSTM is used to process the nonlinear and time-varying coupling characteristics in the historical data, to capture the dynamic mapping relationship between the control vector element u j and the state vector element x i , to replace the fixed coupling coefficient matrix in the traditional mechanism model, and to output the dynamic value b ij under each working condition.

[0028] Preferably, the step S1021 comprises:

[0029] The multi-step prediction is performed once every 5s, the future 30s state is output by the LSTM model, and m c is adjusted in advance through rolling optimization.

[0030] And / or, the prediction error E is calculated, if E>5%, the model online update is triggered.

[0031] Preferably, in the step S1022, the fuzzy adaptive PID controller is a two-dimensional fuzzy controller, the input is the furnace negative pressure deviation e=P set -P real and the change rate of deviation ec=Δe / Δt, and the output is ΔK p , ΔK i , ΔK d .

[0032] The membership function adopts a Gaussian type:

[0033] .

[0034] wherein e: furnace negative pressure deviation; c i : domain center; σ i : Gaussian function standard deviation, ΔK p : proportional coefficient increment, ΔK i : integral coefficient increment, ΔK d : differential coefficient increment, P set : furnace negative pressure set value, P real : real-time measured value of furnace negative pressure.

[0035] Preferably, the step S1022 comprises:

[0036] When |e|>50Pa, the fuzzy control is started:

[0037] (1) Calculate e(t)=P set -P(t), ec(t)=[e(t)-e(t-1)] / Δt;

[0038] (2) The fuzzy reasoning obtains ΔK p , ΔK i , ΔK d , and the PID parameters are updated: K p (t)=K p (t-1)+ΔK p , K i (t)=K i (t-1)+ΔK i , K d (t)=K d (t-1)+ΔK d .

[0039] (3) The control output u(t)=K p e+K i ∑eΔt+K d Δe / Δt, and it is ensured that the negative pressure fluctuation is ≤±10Pa.

[0040] Preferably, in the step S1023, a double objective function is defined:

[0041] Objective function 1 for minimizing energy consumption:

[0042] ;

[0043] Wherein, m c : fuel coal feed rate; LHV: low heat value; V O : oxygen flow rate; C cost : oxygen cost unit price; Q p : slurry production;

[0044] Objective function 2 for maximizing component qualification rate:

[0045] ;

[0046] Wherein, r: actual material ratio; r opt : optimal material ratio; k: S-shaped function slope parameter.

[0047] Preferably, the step S1023 comprises:

[0048] Run the optimization algorithm every 10 min:

[0049] (1) Initialize population: 50 individuals, encoded as [m c , V O , r];

[0050] (2) Calculate fitness f1, f2, perform non-dominated sorting and crowding degree calculation;

[0051] (3) Selection, crossover, mutation to generate offspring;

[0052] (4) Keep the top 20% Pareto optimal solution, output the optimal operating point ( , , ).

[0053] Preferably, a hydraulic push rod system is provided at the slurry outlet, the hydraulic push rod system is linked with the slurry outlet opening degree α, and satisfies the flow equation Q p =k v αΔp 0.5 , wherein k v is the flow coefficient; Δp is the pressure difference between the upstream and downstream of the slurry outlet.

[0054] Preferably, the step S103 comprises:

[0055] Step S104: Establish a three-dimensional thermodynamic model of the melting furnace, real-time map the temperature field and pressure distribution, and set a three-level alarm mechanism.

[0056] The object of the present application is to solve the technical bottlenecks of the process of preparing thermal insulation cotton from waste incineration fly ash by oxygen-rich melting, such as multi-variable strong coupling, time-varying nonlinear characteristics and multi-dimensional control target conflicts, etc., by multi-source data fusion and intelligent algorithm coupling, to provide a full-automatic control method with high efficiency (production capacity improvement), intelligence (full-automatic control), environmental protection (low consumption and low emission), to solve the core defects of traditional control technology and realize the improvement of key performance indicators. The fly ash coal oxygen-rich melting intelligent control method based on multi-modal coupling prediction of the present application has small melting temperature fluctuation, strong furnace negative pressure stability, high quartz sand utilization rate, low ton of molten slurry energy consumption and high thermal insulation cotton qualified rate. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0058] Figure 1 The flowchart of the fly ash coal oxygen-rich melting intelligent control method based on multi-modal coupling prediction of the present application is shown in the figure.

[0059] Figure 2 The system architecture diagram of the fly ash coal oxygen-rich melting intelligent control method based on multi-modal coupling prediction of the present application is shown in the figure, which shows the data flow and control flow of four-layer architecture, and highlights the closed-loop logic of “data acquisition→algorithm processing→execution control→monitoring feedback”;

[0060] Figure 3 The fuzzy control rule reasoning flowchart in the method of the present application (for furnace negative pressure control) is shown in the figure.

[0061] Figure 4 The NSGA-Ⅱ multi-objective optimization flowchart in the method of the present application (energy consumption and qualified rate collaborative optimization) is shown in the figure. DETAILED DESCRIPTION

[0062] The embodiments of the present application will be described in detail below with reference to the drawings.

[0063] It should be clear that the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0064] The application relates to a full-automatic control method applied to a 'fly ash coal oxygen-enriched smelting preparation system of thermal insulation cotton', aiming at the high-temperature smelting process (1300-1500 DEG C) of garbage incineration fly ash (treatment scale 1.80t / d) and auxiliary materials (quartz sand 0.65t / d, aluminum oxide 0.12t / d), through multi-source data fusion and intelligent algorithm coupling, the complex nonlinear control problems of oxygen-enriched combustion (oxygen flow 250Nm3 / h), material ratio (coal 2.45t / d), smelting slurry conditioning (slurry output >=2t / d) are solved, and the efficient, intelligent and environmental protection of the preparation of thermal insulation cotton are realized.

[0065] The embodiment of the application provides a fly ash coal oxygen-enriched smelting (preparation of thermal insulation cotton) intelligent control method based on multi-modal coupling prediction, as shown in Figure 1 The embodiment of the application provides a fly ash coal oxygen-enriched smelting (preparation of thermal insulation cotton) intelligent control method based on multi-modal coupling prediction, as shown in

[0066] Step S101: acquiring multi-source data collected by a sensor;

[0067] This step corresponds to the (multi-source) data acquisition layer in Figure 2

[0068] As an optional embodiment, in the step S101, the multi-source data collected by the sensor comprises:

[0069] Temperature data collected by K-type thermocouples arranged at the bottom, top and outlet of the incinerator;

[0070] Furnace negative pressure data collected by a pressure transmitter arranged at the waist center of the furnace chamber of the incinerator;

[0071] Flue gas oxygen concentration data collected by a zirconia oxygen meter arranged at the outlet of the flue gas cooler;

[0072] Frequency data collected by a frequency converter of a screw conveyor arranged at the coal / fly ash / quartz sand / alumina feeding port.

[0073] In specific implementation, the hardware configuration of this step can be as shown in Table 1:

[0074] Table 1

[0075]

[0076] Step S102: obtaining control parameters through an intelligent control algorithm according to the multi-source data;

[0077] This step corresponds to the intelligent control algorithm layer in Figure 2

[0078] The step S102 comprises:

[0079] Step S1021: for temperature T, furnace negative pressure P, flue gas oxygen concentration C​​O and slurry yield Q p , a multivariate decoupling prediction model is used for decoupling prediction;

[0080] In the process of fly ash coal oxygen-enriched smelting, there is a complex multivariate coupling relationship between temperature T, furnace negative pressure P, flue gas oxygen concentration C O and slurry yield Q p . Traditional PID control is difficult to cope with such strong coupling and nonlinear dynamic system. The core goal of the multivariate decoupling prediction model is to decouple the interaction between variables by constructing a state space equation combined with mechanism model and data-driven method, and to realize accurate prediction and control. As an optional embodiment, the multivariate decoupling prediction model uses the following state space equation to describe the smelting process:

[0081] ;

[0082] Where:

[0083] X=[T,P,C O ,Q p ] T : state vector, containing four core parameters.

[0084] U=[m c ,V O ,r] T : control vector, representing fuel coal feed rate, oxygen flow, material ratio respectively.

[0085] Y: output vector, corresponding to sensor measurement values (such as temperature, pressure, etc.).

[0086] A: system matrix, describing the natural dynamic relationship between state variables (such as natural attenuation of temperature with time).

[0087] B: input matrix, representing the coupling effect of control variables on state variables (key decoupling parameters).

[0088] C: output matrix, mapping state variables to sensor measurement values.

[0089] W: process noise (such as fly ash composition fluctuation, environmental disturbance).

[0090] V: measurement noise (sensor error).

[0091] For state vector X=[T,P,C O ,Q p ] T

[0092] T: temperature in the smelting furnace (core control parameter), temperature of the molten pool and the upper flue gas, designed to run in the range of 1300℃~1500℃;

[0093] P: Furnace negative pressure (key parameter to prevent harmful gas from escaping), real-time value of the center pressure measurement point at the waist of the furnace, design control range -2000 Pa ~ 200 Pa;

[0094] C O : Oxygen concentration of flue gas (oxygen-enriched combustion efficiency index), zirconia oxygen content detector detection value at the outlet of the flue gas cooler, reflecting the utilization rate of oxygen-enriched air;

[0095] Q p : Slurry yield (core capacity index), daily discharge of molten slurry from the melting furnace, design scale ≥ 2 t / d;

[0096] For the control vector U = [m c , V O , r] T

[0097] m c : Fuel coal feeding amount, controlled by coal conveying screw, rated feeding amount 102 kg / h (2.45 t / d);

[0098] V O : Oxygen flow (core parameter of oxygen-enriched combustion system), outlet flow of oxygen generation system, design capacity 250 Nm³ / h;

[0099] r: Material ratio (fly ash: quartz sand: alumina), mass ratio, design value 1.80 t / d: 0.65 t / d: 0.12 t / d ≈ 15:5.4:1;

[0100] For the input matrix B

[0101] The coupling coefficient based on the mechanism model is defined (b ij , that is, each element in the input matrix B), wherein x i is the i-th element of the state vector X (i = 1, 2, 3, 4, corresponding to T, P, C O , Q p in turn), and u j is the j-th element of the control vector U (j = 1, 2, 3, corresponding to m c , V O , r in turn). The input matrix B is a 4 × 3 matrix, and the i-th row and j-th column element b ij represents the coupling coefficient of the control vector element u j to the state vector element x i . The LSTM (model) is used to process the nonlinear and time-varying coupling characteristics (such as the difference in the influence amplitude of fuel amount on temperature under different chlorine contents) in historical data, to capture the relationship between the control vector element u j and the state vector element xi The dynamic mapping relationship between the variables is replaced by the fixed coupling coefficient matrix in the traditional mechanism model, and the b ij dynamic values are output under each working condition.

[0102] The complete structure of the input matrix B is as follows:

[0103] ;

[0104] Each row in the matrix corresponds to a state variable, and each column corresponds to a control variable. For example:

[0105] The first row [b T-mc , b T-VO , b T-r ] represents the coupling effect of fuel quantity, oxygen flow, and material ratio on temperature;

[0106] The second column [ b T-VO , b P-VO , b CO-VO , b Qp-VO ] represents the coupling effect of oxygen flow on temperature, negative pressure, oxygen concentration, and yield.

[0107] For example, when the fly ash chlorine content increases from 2.5% to 5.8%, the LSTM learns through time series data that b T-mc (b T-mc is the influence rate of unit fuel coal feeding quantity change on melting temperature) increases from 3.0℃ / (t / d) to 4.2℃ / (t / d), and b T-mc is updated adaptively to avoid the insufficient correction of the constant model due to the fixed k.

[0108] For the output matrix C (which corresponds to the sensor mapping relationship)

[0109] The output matrix maps the state variables to the sensor measurement values:

[0110] ;

[0111] Ensure that each state variable corresponds to the real-time monitoring data of a unique sensor.

[0112] Model advantages and decoupling mechanism

[0113] Dynamic decoupling: Through matrix B, the coupling relationship between variables is quantified, and the multivariable system is decomposed into approximately independent subsystems. For example, when adjusting m c , the model can predict its combined effect on T and P, and compensate for the coupling effect through the control algorithm.

[0114] Rolling optimization: multi-step prediction (future 30s state) is performed every 5s, combined with real-time data to correct prediction error (if error > 5%, trigger model update), improve dynamic response speed.

[0115] Data-driven enhancement: LSTM training complements the shortcomings of mechanism models, adapts to complex working conditions such as fluctuations in fly ash composition (e.g. ±15% chlorine content).

[0116] As another alternative embodiment, the step S1021 comprises:

[0117] Multi-step prediction is performed every 5s, and the future 30s state is output by the LSTM model, and m c is adjusted in advance through rolling optimization.

[0118] And / or, calculate the prediction error E, if E > 5%, trigger online model update.

[0119] In specific implementation, the coupled model prediction (rolling optimization) performs prediction every 5s:

[0120] Input current U(t) = [m c (t), V O (t), r(t)]

[0121] 1. The LSTM model outputs the future 30s state X(t+1), …, X(t+6) (time step 5s), and m c is adjusted in advance through rolling optimization (e.g. if it is predicted that the temperature will overshoot, reduce the fuel quantity 10s in advance), and the response time of traditional PID is shortened from > 16s to within 15s.

[0122] 2. Calculate the prediction error , if E > 5%, trigger online model update.

[0123] Step S1022: for the furnace negative pressure P, a fuzzy adaptive PID controller is used for fuzzy control;

[0124] This step uses a fuzzy adaptive PID controller (for furnace negative pressure control).

[0125] As an alternative embodiment, a two-dimensional fuzzy controller is designed, with input being the negative pressure deviation e = P set -P real and the rate of change of the deviation ec = Δe / Δt, and output being ΔK p , ΔK i , ΔK d .

[0126] ΔK p : proportional gain increment, used to dynamically adjust the proportional gain of the PID controller, and improve the response speed of the system.

[0127] ΔK i : integral coefficient increment, used to adjust the strength of integral action, eliminate static error;

[0128] ΔK d : differential coefficient increment, used to adjust the strength of differential action, suppress system overshoot.

[0129] The membership function adopts Gaussian type: ;

[0130] Where:

[0131] e: furnace negative pressure deviation, e=P set -P real , set value is the target value in the range of-200~20Pa;

[0132] P set : furnace negative pressure set value (control target value);

[0133] P real : real-time measurement value of furnace negative pressure (acquired by pressure transmitter).

[0134] c i : domain center (fuzzy subset core value), take NB (negative large) =-200Pa, NM (negative medium) =-100Pa, ZO (zero) =0Pa, PM (positive medium) =100Pa, PB (positive large) =200Pa;

[0135] σ i : Gaussian function standard deviation (determines the width of membership degree distribution), take 50Pa, ensure the adjacent fuzzy subset overlap degree about 30%, improve control smoothness;

[0136] Fuzzy rule matrix (part of example), can be shown in Table 2:

[0137] Table 2

[0138]

[0139] As another alternative embodiment, for fuzzy PID dynamic adjustment, when |e|>50Pa, start fuzzy control:

[0140] 1, calculate e(t)=P set -P(t), ec(t)=[e(t)-e(t-1)] / Δt;

[0141] 2, fuzzy reasoning gets ΔK p , ΔK i, ΔK d , update PID parameters: K p (t)=Kp (t-1)+ΔK p , K i (t)=K i (t-1)+ΔK i , K d (t)=K d (t-1)+ΔK d ;

[0142] 3、Control output u(t)=K p e+K i ∑eΔt+K d Δe / Δt, ensure negative pressure fluctuations ≤±10Pa.

[0143] The execution process of this step can be shown as Figure 3 , Figure 3 The closed-loop process from negative pressure collection to PID parameter adjustment is fully presented, and the core steps of fuzzy logic "input fuzzification → rule inference → solution fuzzification" are highlighted. The core is the application of two-dimensional fuzzy rule matrix and Gaussian membership function.

[0144] Step S1023: Calculate the control parameters by using a multi-objective optimization algorithm.

[0145] This step can specifically use the NSGA-II multi-objective optimization algorithm, which aims to optimize energy consumption and qualified rate.

[0146] As an optional embodiment, in this step, a double-objective function is defined:

[0147] Objective function 1 (minimize energy consumption):

[0148] ;

[0149] Numerator: Total daily fuel cost (coal heat consumption + oxygen purchase cost);

[0150] Denominator: Daily slurry output;

[0151] Objective: Minimize unit slurry energy consumption cost;

[0152] Wherein,

[0153] m c : Fuel coal feeding amount (defined as state vector), unit: t / d;

[0154] LHV=25MJ / kg (industry conventional value, coal low heat value corresponds to 25000kJ / kg);

[0155] V O : Oxygen flow (defined as state vector), unit: Nm³ / h;

[0156] C cost : Oxygen cost unit price, estimated at 0.8 yuan / Nm³ according to the industrial liquid oxygen market price;

[0157] Q p : Molten slag yield (same as state vector definition), unit: t / d;

[0158] Objective function 2 (maximize component pass rate):

[0159] ;

[0160] When the actual ratio r approaches the optimal ratio r opt , f2 approaches 1 (100% pass rate);

[0161] The nonlinear mapping relationship reflects the sensitive characteristics of the insulation cotton components (such as the Al2O3 / SiO2 ratio) to the ratio fluctuations;

[0162] Where:

[0163] r: Actual material ratio (fly ash: quartz sand: alumina mass ratio) real-time calculation value, determined dynamically by the material feeding rate of each material;

[0164] r opt : Optimal material ratio (test calibration value), fly ash: quartz sand: alumina = 15:5.4:1;

[0165] k: S-shaped function slope parameter (empirical value k=10), used to amplify the influence of ratio deviation on pass rate and improve optimization sensitivity.

[0166] As another alternative embodiment, for multi-objective optimization solution (NSGA-II implementation), the optimization algorithm is run once every 10 minutes:

[0167] 1. Initialize the population (50 individuals, encoded as [m c , V O , r]);

[0168] 2. Calculate the fitness f1, f2, perform non-dominated sorting and congestion calculation;

[0169] 3. Selection, crossover (single-point crossover, probability 0.8), mutation (Gaussian mutation, standard deviation 0.05) to generate offspring;

[0170] 4. Keep the top 20% Pareto optimal solution, output the optimal operating point( , , ), fuel efficiency is improved, and ton of molten slag energy consumption is reduced, where:

[0171] : Optimal fuel coal feed quantity (unit: t / d), the optimal value of fuel quantity obtained by NSGA-Ⅱ algorithm optimization;

[0172] : Optimal oxygen flow (unit: Nm³ / h), the optimal gas supply rate of the oxygen-enriched combustion system;

[0173] : Optimal material ratio (fly ash: quartz sand: alumina mass ratio).

[0174] The execution process of this step can be as shown in Figure 4 , Figure 4 demonstrates how the non-dominated sorting genetic algorithm (NSGA-Ⅱ) solves the Pareto optimal solution of the fuel quantity m c , oxygen flow V O , and material ratio r through selection, crossover, and mutation operations, corresponding to the dual objectives of "minimizing unit smelt energy consumption cost" and "maximizing component qualification rate".

[0175] Step S103: Realize oxygen-enriched lance, slurry outlet, and material ratio control according to the control parameters;

[0176] This step corresponds to the execution mechanism layer in Figure 2 to realize precise control.

[0177] As an optional embodiment, the control in this step mainly includes:

[0178] 8 sets of oxygen-enriched lance regulating valves (DN20): pneumatic regulating valves with flow control accuracy ±1% and response time <1s;

[0179] Hydraulic push rod system: stroke 1600mm, linked with slurry outlet opening degree α, satisfying the incompressible fluid flow equation Q p =k v αΔp 0.5 (k v is the flow coefficient; represents the flow characteristics of the valve or slurry outlet, obtained by field calibration; Δp is the pressure difference between the upstream and downstream of the slurry outlet, i.e. the driving force for smelt flow);

[0180] Screw conveyor frequency converter: speed regulation range 0-50Hz, realizing quantitative feeding of coal (102kg / h±15%) and fly ash (75kg / h±10%).

[0181] As another optional embodiment, the step S103 is followed by:

[0182] Step S104: Establish a three-dimensional thermodynamic model of the smelting furnace, real-time map the temperature field and pressure distribution, and set a three-level alarm mechanism.

[0183] This step corresponds to the remote monitoring layer (digital twin) in Figure 2 .

[0184] In specific implementation, a three-dimensional thermodynamic model of the melting furnace is established to real-time map the temperature field (resolution ± 50 mm) and pressure distribution (accuracy ± 5 Pa);

[0185] A three-level alarm mechanism is set:

[0186] Early warning (deviation ≤ 10%): acoustic-optical prompt, automatic adjustment of control parameters;

[0187] Exceeding limit (deviation 10%-30%): triggering of a backup spray gun and starting of an emergency feeding mode;

[0188] Fault (deviation > 30%): linkage with the DCS system to shut down, and recording of fault codes (based on Bayesian network diagnosis, accuracy ≥ 95%).

[0189] The use effect of the method of the application is shown in Table 3:

[0190] Table 3

[0191]

[0192] In summary, the application aims to solve the technical bottlenecks of multi-variable strong coupling, time-varying nonlinear characteristics, and multi-dimensional control target conflicts in the process of preparing thermal insulation cotton from waste incineration fly ash through oxygen-enriched melting, and provides a full-automatic control method with high efficiency (production capacity improvement), intelligence (full-automatic control), and environmental protection (low consumption and low emission) by coupling multi-source data fusion and intelligent algorithms, to solve the core defects of traditional control technology and achieve the improvement of key performance indicators. The fly ash coal oxygen-enriched melting intelligent control method based on multi-modal coupling prediction of the application has small melting temperature fluctuation, strong furnace negative pressure stability, high quartz sand utilization rate, low ton of melting slurry energy consumption, and high thermal insulation cotton qualification rate.

[0193] Implementation case:

[0194] Project background (a waste incineration plant in Guangxi)

[0195] Core equipment: oxygen-enriched melting furnace (melting pool size 1500×800×600 mm), four-roll centrifuge (power 30 kW);

[0196] Control difficulties: fly ash chlorine content fluctuation (average 4.2%, range 2.5%-5.8%), air humidity > 85% in rainy season affecting material flowability;

[0197] Key formula field verification:

[0198] 1). Multi-variable decoupling model parameter calibration

[0199] Coupling coefficient is obtained by step response test:

[0200] Fuel quantity step +10%(2.45t / d→2.695t / d), melting temperature rises 35℃(1300℃→1335℃), b T-mc =3.5℃ / (t / d);

[0201] Oxygen flow step +5%(250Nm³ / h→262.5Nm³ / h), furnace negative pressure drops 20Pa, b P-VO =-4Pa / (Nm³ / h);

[0202] 2). Fuzzy PID control parameter tuning

[0203] According to the field commissioning to determine the initial PID parameters:

[0204] K p 0=1.2, K i 0=0.05, K d 0=0.8;

[0205] Fuzzy adjustment range: K p ∈[0.8,1.5], K i ∈[0.03,0.1], K d ∈[0.5,1.2];

[0206] Typical working condition: When the air leakage at the inlet causes the negative pressure to drop-150Pa, the fuzzy controller stabilizes the negative pressure to-10Pa within 15s, which is 25s faster than the conventional PID;

[0207] 3). Multi-objective optimization calculation example

[0208] Daily fly ash composition: chlorine 3.8%, heavy metals (Pb+Cd) 0.08%, set Q p =2.1t / d;

[0209] Initial parameters: m c =2.5t / d, V O =240Nm³ / h, r=2.8:1:0.15;

[0210] Objective function value: f1=12.5yuan / t, f2=0.82;

[0211] After NSGA-Ⅱ optimization: m c =2.38t / d, V O =255Nm³ / h, r=2.7:1:0.18;

[0212] Optimization effect: f1=10.2 yuan / t (down 18.4%), f2=0.91 (up 10.9%), while the furnace negative pressure fluctuation is reduced from ±80Pa to ±15Pa, and the qualified rate of heat preservation cotton is increased from 82% to 95%.

[0213] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An intelligent control method for fly ash coal oxygen-enriched smelting based on multi-modal coupling prediction, characterized in that, The application relates to a method for realizing intelligent control of oxygen-enriched lance, slurry outlet and material proportioning, and belongs to the field of intelligent control of industrial furnaces. Step S101: acquiring multi-source data collected by a sensor; Step S102: obtaining control parameters through an intelligent control algorithm according to the multi-source data; Step S103: realizing control of an oxygen-enriched lance, a slurry outlet and material proportioning according to the control parameters; The step S102 comprises: Step S1021: for temperature T, furnace negative pressure P, flue gas oxygen concentration C O and smelt slurry yield Q p , decoupling prediction is performed by using a multivariate decoupling prediction model; Step S1022: adopting a fuzzy self-adaptive PID controller to realize fuzzy control for the furnace negative pressure P; Step S1023: calculating the control parameters through a multi-objective optimization algorithm; In the step S1021, the multi-variable decoupling prediction model adopts the following state space equation to describe the melting process: ； wherein, : state vector; T: temperature in the smelting furnace; P: negative pressure in the hearth O : oxygen concentration of the flue gas; Q p : smelt output : control vector; m c : fuel coal feed rate; V O : oxygen flow rate; r: material ratio, i.e. fly ash: quartz sand: alumina, mass ratio Y: output vector, corresponding to the sensor measurement value; A: system matrix, describing the natural dynamic relationship between state variables; B: input matrix, representing the coupling influence of the control variable on the state variable; C: output matrix, mapping the state variable to the sensor measurement value; W: process noise; V: measurement noise; The coupling coefficients in the input matrix B are defined based on a mechanism model The LSTM is used to process the nonlinear and time-varying coupling characteristics in the historical data, capture the dynamic mapping relationship between the control vector elements u j and the state vector elements x i , replace the fixed coupling coefficient matrix in the traditional mechanism model, and output the dynamic values of b ij under various working conditions.

2. The multi-modal coupling prediction based intelligent control method of fly ash coal oxygen-enriched melting according to claim 1, characterized in that, In the step S101, the multi-source data collected by the sensor comprises: Temperature data collected by K-type thermocouples arranged at the bottom, top and outlet of the incinerator; Furnace negative pressure data collected by a pressure transmitter arranged at the waist center of the incinerator; Flue gas oxygen concentration data collected by a zirconia oxygen analyzer arranged at the outlet of the flue gas cooler; Frequency data collected by frequency converters of screw conveyors arranged at the coal / fly ash / quartz sand / alumina feeding ports.

3. The multi-modal coupling prediction based intelligent control method of fly ash coal oxygen-enriched melting according to claim 1, characterized in that, The step S1021 comprises: Multi-step prediction is performed every 5s, and the future 30s state is output by the LSTM model, and m is adjusted in advance through rolling optimization c ; And / or, the prediction error E is calculated, and if E>5%, the model is triggered for online updating.

4. The multi-modal coupling prediction based intelligent control method of fly ash coal oxygen-enriched melting according to claim 1, characterized in that, In the step S1022, the fuzzy adaptive PID controller is a two-dimensional fuzzy controller, the input is the furnace negative pressure deviation e=P set -P real and the deviation change rate ec=Δe / Δt, and the output is ΔK p , ΔK i , ΔK d ; The membership function adopts a Gaussian type: ； Wherein, e: furnace negative pressure deviation; c i : domain center; s i : Gaussian function standard deviation, AK p : proportional coefficient increment, AK i : integral coefficient increment, AK d : differential coefficient increment, P set : furnace negative pressure set value, P real : furnace negative pressure real-time measurement value.

5. The multi-modal coupling prediction based intelligent control method of fly ash coal oxygen-enriched melting according to claim 4, characterized in that, The step S1022 comprises: When |e|>50Pa, the fuzzy control is started: (1) Calculate e(t) = P set P(t), ec(t) = [e(t) - e(t-1)] / Δt; (2) ΔK is obtained by fuzzy inference p , ΔK i , ΔK d , PID parameter is updated: K p (t) = K p (t-1) + ΔK p , K i (t) = K i (t-1) + ΔK i , K d (t) = K d (t-1) + ΔK d ; (3) Control output u(t) = K p e + K i ∑eΔt + K d Δe / Δt, ensure negative pressure fluctuations ≤ ± 10 Pa.

6. The multi-modal coupling prediction based intelligent control method of fly ash coal oxygen-enriched melting according to claim 1, characterized in that, In the step S1023, a double-objective function is defined: Objective function 1 for calculating energy consumption minimization: ； wherein, m c : fuel coal feed quantity; LHV: lower heating value; V O : oxygen flow rate; C cost : oxygen cost unit price; Q p : slurry production Objective function 2 for calculating component qualification rate maximization: ； Where, r: actual material ratio; r opt : optimal material ratio; k: S-shaped function slope parameter.

7. The multi-modal coupling prediction based intelligent control method of fly ash coal oxygen-enriched melting according to claim 6, characterized in that, The step S1023 comprises: The optimization algorithm is run once every 10 minutes: (1) Initialization of population: 50 individuals, encoding as [m c ,V O ,r]; (2) The fitness f1 and f2 are calculated, and non-dominated sorting and congestion degree calculation are performed; (3) Selection, crossover and mutation are performed to generate offspring; (4) Keep the top 20% Pareto optimal solutions, output the optimal operating point , , ).

8. The multi-modal coupling prediction based intelligent control method of fly ash coal oxygen-enriched melting according to claim 1, characterized in that, A hydraulic push rod system is arranged at the pulp outlet, which is linked with the opening degree of the pulp outlet α, and satisfies the flow equation Q of incompressible fluid p =k v αΔp 0.5 , wherein k v is the flow coefficient; and Δp is the pressure difference between the upstream and downstream of the pulp outlet.

9. The multi-modal coupling prediction based intelligent control method of fly ash coal oxygen-enriched melting according to claim 1, characterized in that, The step S103 comprises: Step S104: establishing a three-dimensional thermodynamic model of the melting furnace, realizing real-time mapping of a temperature field and pressure distribution, and setting a three-level alarm mechanism.

Images