Exhaust gas powder carrying amount change model in starting and stopping process of coal pulverizing system and powder feeding control method

Abstract

The invention relates to the technical field of thermoelectric production, and discloses an exhaust gas powder carrying amount change model and a powder feeding control method in the start-stop process of a pulverizing system, and the method comprises the following steps: building a reference working condition data set based on pulverizing reference working condition historical data, building a milling starting working condition data set based on pulverizing starting working condition historical data, and utilizing the reference working condition data set to carry out powder feeding control. The method comprises the following steps: constructing a reference exhaust gas powder carrying amount model, calculating model reference parameters, obtaining a coal pulverizing system start-stop signal, obtaining boiler real-time operation data, carrying out data preprocessing, constructing a real-time working condition data set, constructing an exhaust gas powder carrying amount change model by utilizing the model reference parameters and the real-time working condition data set, and calculating the real-time exhaust gas powder carrying amount in the start-stop process. According to the method, through the exhaust gas powder carrying amount change model in the starting and stopping process of the coal pulverizing system, the exhaust gas powder carrying amount change in the starting and stopping process is quantitatively calculated, the adjustment operation of the powder feeder is output, combustion fluctuation in the starting and stopping process is reduced, and the stability and economical efficiency of a unit are improved.

Description

Technical Field

[0001] The present invention relates to the field of thermoelectricity production, and more particularly to a powder feeding control method and a model for variation of exhaust gas powder carrying amount during the start-up and shutdown process of a powder making system. Background Art

[0002] The exhaust gas generated by the central storage pulverized coal furnace pulverizing system carries a large amount of pulverized coal, which is fed into the furnace through primary or tertiary air flow for combustion. To maintain an appropriate pulverized coal bunker level, the central storage pulverizing system requires frequent startup and shutdown during operation. The rapid changes in the exhaust gas pulverized coal content during startup and shutdown significantly disrupt the combustion and heat exchange within the furnace, causing significant fluctuations in parameters such as main steam pressure, nitrogen oxides, and wall temperature, severely impacting the unit's combustion efficiency, environmental emissions, and steam quality.

[0003] Therefore, a model of the change of exhaust gas powder carrying amount and a powder feeding control method during the start-up and shutdown process of the pulverizing system are proposed, so that the smooth control of the unit during the start-up and shutdown process of the pulverizing system has effective data support. Summary of the Invention

[0004] The present invention provides a model for the change of exhaust gas powder carrying amount during the start-up and shutdown process of the pulverizing system and a powder feeding control method, which solves the problem that the related technologies cannot measure the change of exhaust gas powder carrying amount during the start-up and shutdown process of the pulverizing system and cannot effectively improve the control problem caused by the powder amount disturbance during the start-up and shutdown process of the pulverizing system.

[0005] The present invention provides a model for the change of the amount of powder carried by exhaust gas during the start-up and shutdown process of a pulverizing system and a pulverizing control method, which includes the following steps:

[0006] S100, obtaining historical data of milling benchmark working conditions, performing data preprocessing and sample construction, and establishing a benchmark working condition data set;

[0007] S200, obtaining historical data of milling startup conditions, performing data preprocessing and sample construction, and establishing a milling startup condition data set;

[0008] S300: Using the benchmark operating condition data set, a benchmark exhaust gas powder carrying capacity model is constructed and the model benchmark parameters are calculated;

[0009] S400: Obtain the start and stop signals of the pulverizing system, obtain the real-time operation data of the boiler, perform data preprocessing, and construct a real-time operating condition data set;

[0010] S500: Using the model baseline parameters and real-time operating condition data set, a model for the variation of exhaust gas powder load is constructed to calculate the real-time exhaust gas powder load during the start-up and shutdown process.

[0011] S600: Using the grinding start-up condition data set and model benchmark parameters, a powder quantity correction model is constructed to calculate the powder quantity correction coefficient.

[0012] S700 uses the real-time exhaust gas powder carrying amount and powder amount correction coefficient during the start-up and shutdown process to calculate the powder feeder adjustment operation during the start-up and shutdown process of the pulverizing system;

[0013] S800 determines the availability of the exhaust gas powder load change model in real time and regularly updates the model baseline parameters and powder load correction coefficient.

[0014] Furthermore, in S100, the following steps are specifically included:

[0015] S110, extracting the benchmark operating data of the pulverizing system from the historical database;

[0016] S120, establish the baseline working condition screening standard;

[0017] S130, performing a quality check on the screened data to remove outliers and missing data;

[0018] S140, grouping the data according to the operating status combination of the coal mill to establish data subsets of different operating conditions;

[0019] S150, performing standardization processing on the grouped data to eliminate the dimension effect;

[0020] S170, performing statistical analysis and validity verification on the constructed benchmark operating condition data set.

[0021] Furthermore, in S200, the following steps are specifically included:

[0022] S210, identifying and extracting a complete data sequence of the milling system start-up process from the historical operation data;

[0023] S220, extracting key operating parameters during the grinding process and establishing time series data;

[0024] S230, determining a baseline parameter value at the start-up time as a reference point for subsequent change calculations;

[0025] S240, calculating a sequence of changes in each key parameter relative to a baseline value at the start-up time;

[0026] S250, performing quality assessment on the extracted grinding start-up condition data;

[0027] S260, dividing the grinding process into different stages and extracting characteristic parameters of each stage;

[0028] S270 integrates data from multiple grinding start-up events to build a standardized grinding start-up condition dataset.

[0029] Furthermore, in S300, the following steps are specifically included:

[0030] S310, collecting measurement data of the pulverized gas volume of each coal mill exhaust gas pipeline under the baseline operating conditions, and establishing a corresponding relationship between the pulverized gas volume and the operating parameters;

[0031] S320, analyzing the correlation between various operating parameters and exhaust gas powder carrying capacity in the benchmark operating condition data set, and screening key characteristic parameters;

[0032] S330, establishing a sub-model for predicting the amount of pulverized gas carried by a single coal mill based on its operating parameters;

[0033] S340 integrates the individual models of each coal mill, considers the mutual influence between the coal mills, and establishes a system-wide dust load model;

[0034] S350, optimizing and training the model parameters using the benchmark operating condition data set to determine the optimal parameter combination;

[0035] S360, validates and evaluates the performance of the trained baseline airborne particle load model using an independent validation dataset;

[0036] S370, Model Applicability Analysis and Parameter Sensitivity Assessment: Analyze the applicability of the model under different operating conditions and evaluate the sensitivity of key parameters to the prediction results.

[0037] Furthermore, the calculation formula of the exhaust gas powder carrying capacity prediction sub-model is as follows:

[0038] Hidden layer output:

[0039] Output layer calculation:

[0040] Among them, w in,i,n 、w out,n are the weight parameters of the hidden layer and output layer of the neural network, b h,n 、b out are the bias parameters of the hidden layer and output layer of the neural network, respectively, f act is the activation function, N input 、N hidden are the number of neurons in the input layer and hidden layer respectively, X input,i is the i-th input feature, h n,j is the output of the nth hidden layer neuron of the neural network of the jth coal mill, M dust,j represents the amount of dust carried by exhaust gas of the j-th coal mill;

[0041] The calculation formula of the powder carrying capacity model of the whole system is as follows:

[0042] Calculation of total powder carrying capacity of the system:

[0043] Interaction effects between coal mills:

[0044]

[0045] Load correction factor:

[0046]

[0047] Corrected system powder carrying capacity: M dust,system =K load ·M dust,total ;

[0048] Among them, M dust,total is the total powder carrying capacity of the system, M interaction is the interaction term between coal mills, β j,k is the interaction coefficient between the jth coal mill and k, L boiler,rated is the rated load of the boiler, γ1 and γ2 are load correction coefficients, K load is the load correction factor, M dust,system is the total powder carrying capacity of the system after correction, S mill,j is the operating status of the j-th coal mill, I mill,j is the operating current of the j-th coal mill.

[0049] Furthermore, in S400, the following steps are specifically included:

[0050] S410, collecting start / stop signals and operating status information of each coal mill in the pulverizing system;

[0051] S420, collecting real-time operating parameters of the boiler and pulverizing system;

[0052] S430, performing quality inspection and exception handling on the collected real-time data;

[0053] S440, constructing a working condition data set based on the collected real-time data;

[0054] S450, smoothes the real-time data to eliminate the influence of noise.

[0055] Furthermore, in S500, the following steps are specifically included:

[0056] S510, establishing a state space model describing the start-up and shutdown process of the pulverizing system, and determining state variables and state transition equations;

[0057] S520: Based on the state space model and in combination with the system-wide powder carrying capacity model established in S340, a prediction model for the exhaust gas powder carrying capacity is constructed to describe the changes in the exhaust gas powder carrying capacity during the start-up and shutdown processes.

[0058] The calculation formula of the exhaust gas powder carrying capacity prediction model is as follows:

[0059] M dust,system(t+1)=g(x(t),u(t),w(t))

[0060] x(t)=[L boiler (t),F total (t),I mill,1 (t),...,I mill,M (t),ΔP mill,1 (t),...,ΔP mill,M (t)] T

[0061] Among them, M dust,system (t+1) is the predicted value of the exhaust gas dust carrying capacity at time t+1, g(·) is the dust carrying capacity model of the entire system, x(t) represents the system state vector at time t, u(t) is the system control input, including the coal feed rate and the coal feeder speed, and w(t) is the system disturbance input, including the ambient temperature and humidity, and the fuel quality.

[0062] S530, analyzing the influence of various state variables and control inputs on the change of exhaust gas powder carrying amount during the start-up and shutdown process;

[0063] In step S540 , the exhaust gas powder carrying capacity prediction model is optimized for parameter training and model structure optimization based on the real-time operating condition data set processed in step S400 .

[0064] Furthermore, in S600, the following steps are specifically included:

[0065] S610, calculating the amount of powder that needs to be corrected during the start-stop process based on the prediction result of the exhaust gas powder amount prediction model established in S520;

[0066] S620: Establish a dynamic coal feed rate adjustment model based on the correction demand and determine the coal feed rate adjustment strategy for each coal mill;

[0067] S630: Establish a coordinated control model between the speed of the coal feeder and the coal feeding amount to ensure that the corrected coal feeding amount can be accurately implemented;

[0068] S640 optimizes the output distribution of each coal mill while meeting the pulverized coal quantity correction requirements;

[0069] S650: Establish a real-time feedback mechanism for correction effects and a model adaptive adjustment mechanism to improve correction accuracy;

[0070] S660, formulates corresponding correction strategy switching mechanism according to different start-stop operating conditions.

[0071] Furthermore, in S700, the following steps are specifically included:

[0072] S710, calculating the total coal feed adjustment amount of the entire pulverizing system according to the pulverized coal amount correction model constructed in S600;

[0073] S720, calculating the speed adjustment amount of each coal mill feeder according to the coal feeder speed coordinated control model in S630;

[0074] S730, calculating the output adjustment amount of each coal mill according to the coal mill output balance optimization model in S640;

[0075] S740, allocating the total coal feed rate adjustment, the coal feeder speed adjustment, and the coal mill output adjustment calculated in S710-S730 to specific actuators;

[0076] S750: Allocate the calculated control quantity to the corresponding actuator and establish a real-time feedback mechanism.

[0077] Furthermore, in S800, the following steps are specifically included:

[0078] S810, based on the powder feeding control amount distribution and real-time feedback information in S750, online identification and update of the powder making system model parameters;

[0079] S820 removes noise and drift in the model parameters updated online by performing offline optimization regularly.

[0080] The beneficial effects of the present invention are:

[0081] The present invention establishes a model for the change of exhaust gas powder carrying amount during the start-up and shutdown process of the pulverizing system, quantitatively calculates the change of exhaust gas powder carrying amount during the start-up and shutdown process of the mill, outputs the corresponding adjustment operation of the powder feeder, reduces the combustion fluctuation during the start-up and shutdown process of the pulverizing system, improves the stability and economy of the unit, and does not require additional hardware investment; establishes a powder amount correction model to adjust the powder feeder operation amount online, further improves the control effect during the start-up and shutdown of the mill, and provides effective support for the health management of the pulverizing system, environmental protection control optimization, and equipment failure early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 This is a general flow chart of the exhaust gas powder carrying amount change model and powder feeding control method during the start-up and shutdown process of the pulverizing system proposed by the present invention. DETAILED DESCRIPTION

[0083] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. In addition, features described with respect to some examples may also be combined in other examples.

[0084] like Figure 1 As shown in FIG, the exhaust gas powder carrying amount change model and powder feeding control method during the start-up and shutdown process of the pulverizing system include the following steps:

[0085] S100, obtaining historical data of milling benchmark working conditions, performing data preprocessing and sample construction, and establishing a benchmark working condition data set;

[0086] In one embodiment of the present invention, the following steps are specifically included:

[0087] S110, historical data collection and preliminary screening: Extract the benchmark operating data of the milling system from the historical database. The time span is recommended to be no less than 6 months.

[0088] The acquisition parameters include:

[0089] Boiler operating parameters: boiler load L boiler , the current power generation load of the boiler; the main steam pressure P steam , boiler main steam pressure; powder feeder total frequency F total , the sum of the operating frequencies of all powder feeders.

[0090] Coal mill operating parameters: Coal mill operating status S mill,j , the operating status of the j-th coal mill (0-stop, 1-run); coal mill current I mill,j , the operating current of the jth coal mill; the pressure difference between the inlet and outlet of the coal mill ΔP mill,j , the pressure difference between the inlet and outlet of the j-th coal mill.

[0091] Other related parameters: primary air volume Q air,j , the primary air volume of the jth coal mill; the coal mill outlet temperature T out,j , the outlet temperature of the jth coal mill; where j is the number of the coal mill, j = 1, 2,…, M, and M is the total number of coal mills.

[0092] S120, Definition of Baseline Condition Screening Criteria: Establish baseline condition screening criteria to ensure data quality and representativeness;

[0093] The screening conditions for the benchmark operating conditions include the stable operating condition determination conditions, load range limitation conditions and the effectiveness of the coal mill's operating status;

[0094] The conditions for determining stable working conditions are as follows:

[0095]

[0096] in, is the standard deviation of boiler load in the time window, is the standard deviation of the main steam pressure in the time window, σ L,threshold is the preset load stability threshold, σ P,threshold is the preset pressure stability threshold;

[0097] The load range is limited as follows: L min ≤L boiler ≤L max

[0098] Among them, L min is the minimum allowable load value, L max is the maximum allowable load value, L boiler is the current boiler load;

[0099] The operating status effectiveness of the coal mill is as follows:

[0100] S mill,j ∈{0,1}

[0101]

[0102] Among them, N min The minimum number of coal mills in operation is required;

[0103] S130, data quality inspection and cleaning: perform quality inspection on the screened data and remove outliers and missing data;

[0104] Outlier detection formula:

[0105] 3σ criterion detection: |X k -μ X |≤3σ X ;

[0106] Interquartile range method:

[0107] Q1-1.5×IQR≤X k ≤Q3+1.5×IQR

[0108] IQR=Q3-Q1

[0109] Among them, X k is the actual value of the data point to be detected, μ X is the mean of the data series, σ Xis the standard deviation of the data series, Q1 is the first quartile of the data series, Q3 is the third quartile of the data series, and IQR is the interquartile range;

[0110] S140, grouping by coal mill operation combination: grouping the data according to the coal mill operation state combination to establish data subsets of different operation conditions;

[0111] Group identification formula:

[0112]

[0113] Among them, G combo A unique identification code for the coal mill operation combination;

[0114] Grouping condition: N group,i ≥N sample,min ;

[0115] Among them, N group,i is the number of samples in the i-th group, N sample,min is the minimum sample size requirement, i is the group number;

[0116] S150, data standardization and normalization: standardize the grouped data to eliminate the dimension effect;

[0117] Z-scre normalization:

[0118]

[0119] Min-Max Normalization:

[0120]

[0121] Among them, X norm,k is the normalized data point value, X scaled,k is the normalized data point value, X k is the original data point value, μ group is the mean of the data within the group, σ group is the standard deviation of the data within the group, X min is the minimum value of the data in the group, X max The maximum value of the data in the group

[0122] S160, constructing training data pairs: constructing the processed data into input-output data pairs for subsequent model training;

[0123] Data pair construction:

[0124] DataPair k ={Input k ,Target k}

[0125] Input k =[L boiler,k ,P steam,k ,F total,k ,S mill,1,k ,...,S mill,M,k ]

[0126] Target k =[I mill,1,k ,...,I mill,M,k ,ΔP mill,1,k ,...,ΔP mill,M,k ]

[0127] Among them, DataPair k is the kth training data pair, Input k is the input feature vector, Target k is the target output vector, k is the data pair number, L boiler,k is the kth boiler load value, P steam,k is the kth main steam pressure value, F total,k is the total frequency value of the kth powder feeder, S mill,1,k is the operating status of the first coal mill in the kth coal mill, S mill,M,k is the operating status of the Mth coal mill in the kth coal mill, I mill,1,k is the current value of the first coal mill in the kth coal mill, I mill,M,k is the current value of the Mth coal mill in the kth mill, ΔP mill,1,k is the pressure difference value of the first coal mill in the kth coal mill, ΔP mill,M,k is the pressure difference value of the Mth coal mill in the kth one;

[0128] S200, obtaining historical data of milling startup conditions, performing data preprocessing and sample construction, and establishing a milling startup condition data set;

[0129] In one embodiment of the present invention, the following steps are specifically included:

[0130] S210, Grinding process data collection and identification: Identify and extract the complete data sequence of the grinding process of the pulverizing system from the historical operation data;

[0131] Coal mill start-up judgment condition: ΔS mill,j (t) = S mill,j (t)-S mill,j (t-1)=1;

[0132] Start time determination: t start,j =argmin t {ΔS mill,j (t) = 1};

[0133] Time window definition: T window =[t start,j -T pre ,t start,j +T post ];

[0134] Where, ΔS mill,j (t) is the change in the start-stop state of the j-th coal mill at time t, S mill,j (t) is the operating status of the j-th coal mill at time t, t start,j is the starting time of the jth coal mill, T pre is the observation time before start-up, T post is the observation time after startup, T window is the data collection time window;

[0135] S220, extracting key parameters of the grinding start-up condition: extracting key operating parameters during the grinding start-up process and establishing time series data;

[0136] Boiler operating parameters:

[0137]

[0138] Coal mill parameters:

[0139]

[0140] Among them, L boiler (t) is the boiler load at time t, P steam (t) is the main steam pressure at time t, F total (t) is the total frequency of the powder feeder at time t, I mill,j (t) is the current of the j-th coal mill at time t, ΔP mill,j (t) is the pressure difference of the j-th coal mill at time t, Δt is the sampling time interval, K pre is the number of sampling points before startup, K post is the number of sampling points after startup, is the boiler load value of k sampling intervals before and after the start time, is the main steam pressure value of k sampling intervals before and after the start time, is the total frequency value of the powder feeder in k sampling intervals before and after the start time, is the current value of the jth coal mill in k sampling intervals before and after the start time, is the pressure difference value of the jth coal mill in k sampling intervals before and after the startup time, k represents the time series index, -K pre to K post , t start,j +kΔt is the k sampling intervals before and after the start-up time of the j-th coal mill;

[0141] S230, determining a reference value at the start time: determining a reference parameter value at the start time as a reference point for subsequent change calculations;

[0142] Boiler load reference value at startup:

[0143]

[0144] Main steam pressure reference value at startup:

[0145]

[0146] The total frequency reference value of the powder feeder at the start time:

[0147]

[0148] Among them, L boiler,ref is the boiler load reference value, P steam,ref is the main steam pressure reference value, F total,ref is the total frequency reference value of the powder feeder, N avg The number of samples for the average calculation;

[0149] S240, parameter change sequence calculation: calculating the change sequence of each key parameter relative to the baseline value at the start time;

[0150] Boiler parameter changes:

[0151] ΔL boiler (t) = L boiler (t)-L boiler,ref

[0152] ΔP steam (t) = P steam (t)-P steam,ref

[0153] ΔF total (t) = F total (t)-F total,ref

[0154] Coal mill parameter changes:

[0155] ΔI mill,j (t) = I mill,j (t)-I mill,j,idle

[0156] ΔΔP mill,j (t) = ΔP mill,j (t)-ΔP mill,j,idle

[0157] Where, ΔL boiler(t) is the change in boiler load at time t, ΔP steam (t) is the change in main steam pressure at time t, ΔF total (t) is the total frequency change of the powder feeder at time t, ΔI mill,j (t) is the current change of the j-th coal mill at time t, ΔΔP mill,j (t) is the pressure difference change of the j-th coal mill at time t, I mill,j,idle is the no-load current of the j-th coal mill, ΔP mill,j,idle is the no-load pressure difference of the j-th coal mill;

[0158] S250, Grinding start-up condition data quality assessment: Perform quality assessment on the extracted grinding start-up condition data to ensure data integrity and validity;

[0159] Data integrity check:

[0160] Determination of the effectiveness of the startup process:

[0161] Coal mill current rise: max(I mill,j (t))-I mill,j,idle ≥ΔI threshold ;

[0162] Pressure difference change: max(ΔP mill,j (t))-ΔP mill,j,idle ≥ΔP threshold ;

[0163] Among them, R complete is the data integrity rate, N valid is the number of valid data points, N total is the total number of data points, R threshold is the completeness threshold, ΔI threshold is the current change judgment threshold, ΔP threshold is the pressure difference change judgment threshold, max(·) is the maximum value function; I mill,j,idle is the no-load current value of the j-th coal mill, ΔP mill,j,idle is the no-load pressure difference value of the j-th coal mill;

[0164] S260, grinding start-up stage division and feature extraction: dividing the grinding start-up process into different stages and extracting feature parameters of each stage;

[0165] Pre-startup phase: Phase pre =[t start,j -T pre ,t start,j ];

[0166] Startup transition phase: Phase trans =[t start,j ,tstart,j +T trans ];

[0167] Stable operation phase: Phase stable =[t start,j +T trans ,t start,j +T post ];

[0168] Maximum rate of change:

[0169] Response time constant: τ response =t 63％ -t start,j ;

[0170] Among them, Phase pre Phase is the time interval of the pre-start phase. trans Phase is the time interval for starting the transition phase. stable is the time interval of the stable operation phase, T trans is the duration of the transition phase, is the maximum rate of change of the parameter, τ response is the system response time constant, t 63％ is the moment when the parameter reaches 63% of the steady-state value;

[0171] S270, Multi-condition Dataset Construction and Standardization: Integrate multiple grinding event data to construct a standardized grinding condition dataset;

[0172] The classification of working condition types includes load level classification and coal mill combination classification;

[0173] Load level classification:

[0174]

[0175] Coal mill combination classification:

[0176] The grinding start-up condition data set is constructed as follows:

[0177]

[0178] Time series normalization:

[0179] Among them, Class load For load level classification, L low is the low load threshold, L medium is the medium load threshold, G combo,start N is the coal mill combination identification code, startup is the total number of effective grinding events, ΔX norm,i(t) is the normalized parameter change, μ ΔX is the mean value of the variation under similar working conditions, σ ΔX is the standard deviation of the variation under the same working conditions, SDS (StartupDataSet) represents the grinding start-up condition data set;

[0180] S300: Using the benchmark operating condition data set, a benchmark exhaust gas powder carrying capacity model is constructed and the model benchmark parameters are calculated;

[0181] In one embodiment of the present invention, the following steps are specifically included:

[0182] S310, exhaust gas dust entrainment measurement data collection and processing: Collect dust entrainment measurement data of each coal mill exhaust gas pipeline under the baseline operating conditions, and establish a corresponding relationship between dust entrainment and operating parameters;

[0183] Measurement parameters of exhaust gas powder content:

[0184]

[0185] Calculation of powder carrying capacity per unit time: M dust,j (t) = C dust,j (t)×Q exhaust,j (t)×ρ air ;

[0186] Among them, C dust,j (t) is the dust concentration in the exhaust gas pipeline of the j-th coal mill, Q exhaust,j (t) is the exhaust gas volume of the j-th coal mill, M dust,j (t) is the amount of powder carried per unit time by the j-th coal mill, ρ air is the air density, N sample is the number of sampling points, t k is the kth sampling moment, is the dust concentration in the exhaust gas pipeline of the j-th coal mill at the k-th sampling moment, is the exhaust gas volume of the j-th coal mill at the k-th sampling moment;

[0187] S320, selection and correlation analysis of baseline operating condition characteristic parameters: Analyze the correlation between various operating parameters in the baseline operating condition data set and the exhaust gas powder carrying capacity, and select key characteristic parameters;

[0188] Characteristic parameter matrix construction:

[0189] X base =[L boiler ,P steam ,F total ,I mill,1 ,...,I mill,M ,ΔP mill,1 ,...,ΔPmill,M ] T

[0190] Correlation matrix calculation:

[0191]

[0192] Feature Importance Assessment:

[0193]

[0194] Among them, X base is the characteristic parameter matrix of the benchmark working condition, X i,k is the kth sampling value of the i-th characteristic parameter, is the mean value of the i-th characteristic parameter, is the average value of the powder carrying capacity of the j-th coal mill, N total is the total number of sampling points, N feature is the total number of characteristic parameters, is the correlation coefficient between the ith characteristic parameter and the pulverized dust load of the jth coal mill, w feature,i is the importance weight of the i-th feature parameter;

[0195] S330, establishing a sub-model for predicting the amount of pulverized gas carried by dust: establishing a sub-model for predicting the amount of pulverized gas carried by dust based on the operating parameters of a single coal mill;

[0196] The formula of the exhaust gas powder load prediction sub-model is as follows:

[0197] Hidden layer output:

[0198] Output layer calculation:

[0199] Among them, w in,i,n 、w out,n are the weight parameters of the hidden layer and output layer of the neural network, b h,n 、b out are the bias parameters of the hidden layer and output layer of the neural network, respectively, f act is the activation function, N input 、N hidden are the number of neurons in the input layer and hidden layer respectively, X input,i is the i-th input feature, h n,j is the output of the nth hidden layer neuron of the neural network of the jth coal mill, M dust,j represents the amount of dust carried by exhaust gas of the j-th coal mill;

[0200] S340, Construction of the system-wide dust entrainment model: Integrate the individual models of each coal mill, consider the mutual influence between coal mills, and establish a system-wide dust entrainment model;

[0201] The formula for the whole system powder carrying capacity model is as follows:

[0202] Calculation of total powder carrying capacity of the system:

[0203]

[0204] Interaction effects between coal mills:

[0205]

[0206] Load correction factor:

[0207]

[0208] Corrected system powder carrying capacity: M dust,system =K load ·M dust,total

[0209] Among them, M dust,total is the total powder carrying capacity of the system, M interaction is the interaction term between coal mills, β j,k is the interaction coefficient between the jth coal mill and k, L boiler,rated is the rated load of the boiler, γ1 and γ2 are load correction coefficients, K load is the load correction factor, M dust,system is the total powder carrying capacity of the system after correction, I mill,j is the operating current of the j-th coal mill;

[0210] S350, model parameter optimization and training: Use the benchmark operating condition data set to optimize the model parameters and determine the optimal parameter combination;

[0211] Mean squared error loss:

[0212] Mean absolute percentage error:

[0213] Comprehensive loss function: L total =λ1·L MSE +λ2·L MAPE +λ3·L regularization ;

[0214] Regularization term:

[0215] Among them, N train is the number of training samples, M dust,system,i The model predicts the pollen carrying capacity of the i-th sample, M dust,measured,i is the measured pollen load of the i-th sample, λ1, λ2, λ3 are the mean square, mean absolute percentage and weight coefficient of the loss function, and w pis the model weight parameter, L MSE is the mean square error loss, L MAPE is the mean absolute percentage error, L regularization is the regularization term, L total is the comprehensive loss function;

[0216] S360, Model Validation and Performance Evaluation: Validate and evaluate the performance of the trained baseline airborne particle load model using an independent validation dataset;

[0217] Coefficient of determination:

[0218]

[0219] Root mean square error:

[0220]

[0221] Forecast error distribution test:

[0222] e i =M dust,system,i -M dust,measured,i

[0223]

[0224] Among them, N test is the number of test samples, is the average value of the measured powder carrying capacity, R 2 is the coefficient of determination, RMSE is the root mean square error, e i is the prediction error of the i-th sample, is the mean prediction error, σ e is the standard deviation of the prediction error, M dust,system,i The model predicts the pollen carrying capacity of the i-th sample, M dust,measured,i is the measured pollen carrying capacity of the i-th sample;

[0225] S400: Obtain the start and stop signals of the pulverizing system, obtain the real-time operation data of the boiler, perform data preprocessing, and construct a real-time operating condition data set;

[0226] In one embodiment of the present invention, the following steps are specifically included:

[0227] S410, pulverizing system start / stop signal collection: collects start / stop signals and operating status information of each coal mill in the pulverizing system;

[0228] Start-stop signal collection:

[0229]

[0230] Start-stop state change detection: ΔS mill,j (t) = Smill,j (t)-S mill,j (t-Δt);

[0231] Among them, S mill,j (t) is the operating status of the j-th coal mill at time t, ΔS mill,j (t) is the state change of the j-th coal mill at time t, Δt is the sampling time interval, j is the coal mill number, j = 1, 2, ..., M;

[0232] S420, real-time boiler operation data collection: collect real-time operating parameters of the boiler and pulverizing system;

[0233] Boiler parameter collection: X boiler (t)=[L boiler (t),P steam (t),F total (t)] T ;

[0234] Coal mill parameter collection: X mill,j (t)=[I mill,j (t),ΔP mill,j (t),Q air,j (t),T out,j (t)] T ;

[0235] Among them, L boiler (t) is the boiler load at time t, P steam (t) is the main steam pressure at time t, F total (t) is the total frequency of the powder feeder at time t, I mill,j (t) is the current of the jth coal mill at time t, ΔP mill,j (t) is the pressure difference of the j-th coal mill at time t, Q air,j (t) is the primary air volume of the j-th coal mill at time t, T out,j (t) is the outlet temperature of the jth coal mill at time t, X boiler (t) is the boiler operating parameter vector at time t, X mill,j (t) is the operating parameter vector of the j-th coal mill at time t, j is the coal mill number, j = 1, 2, ..., M;

[0236] S430, data quality inspection and processing: perform quality inspection and exception processing on the collected real-time data;

[0237] Data validity test:

[0238]

[0239] Data continuity test:

[0240]

[0241] Among them, X min,i is the minimum allowed value of the i-th variable, X max,i is the maximum allowed value of the i-th variable, ΔX max,i is the maximum allowed change of the i-th variable, X i (t) is the measured value of the i-th variable at time t, Valid(X i (t)) is the validity flag of the i-th variable at time t, Continuous(X i (t)) is the continuity flag of the i-th variable at time t;

[0242] S440, constructing a real-time working condition data set: constructing a working condition data set based on the collected real-time data;

[0243] Working condition data vector construction: X real (t) = [X boiler (t) T ,X mill,1 (t) T ,...,X mill,M (t) T ] T ;

[0244] Working condition identification calculation:

[0245] Among them, X real (t) is the real-time working condition data vector at time t, G combo (t) is the coal mill combination identification code at time t, T Represents a transpose operation;

[0246] S450, data smoothing and filtering: smoothing real-time data to eliminate noise effects;

[0247] Sliding average filter:

[0248]

[0249] Exponential smoothing filter:

[0250]

[0251] in, is the sliding mean of the i-th variable, is the exponential smoothing value of the i-th variable, N window is the sliding window size, α is the exponential smoothing coefficient (0<α≤1);

[0252] S500: Using the model baseline parameters and real-time operating condition data set, a model for the variation of exhaust gas powder load is constructed to calculate the real-time exhaust gas powder load during the start-up and shutdown process.

[0253] In one embodiment of the present invention, the following steps are specifically included:

[0254] S510, state space modeling of the start-stop process: Establish a state space model that describes the start-stop process of the milling system, and determine the state variables and state transition equations;

[0255] State variable definition:

[0256] x(t)=[L boiler (t),F total (t),I mill,1 (t),...,I mill,M (t),ΔP mill,1 (t),...,ΔP mill,M (t)] T

[0257] Among them, L boiler (t) is the boiler load at time t, F total (t) is the total frequency of the powder feeder at time t, I mill,j (t) is the current of the jth coal mill at time t, ΔP mill,j (t) is the pressure difference of the j-th coal mill at time t, and x(t) represents the system state vector at time t;

[0258] State transition equation: x(t+1)=f(x(t),u(t),w(t));

[0259] Where x(t+1) is the system state vector at the next moment, u(t) is the system control input (such as coal feed rate, coal feeder speed, etc.), w(t) is the system disturbance input (such as ambient temperature and humidity, fuel quality, etc.), and f(·) is the state transfer function.

[0260] S520, exhaust gas powder carrying capacity prediction model construction: Based on the state space model and combined with the full system powder carrying capacity model established in S340, an exhaust gas powder carrying capacity prediction model is constructed to describe the exhaust gas powder carrying capacity changes during the start-up and shutdown process;

[0261] The exhaust gas powder carrying capacity prediction model is: M dust,system (t+1)=g(x(t),u(t),w(t));

[0262] Among them, M dust,system (t+1) is the predicted value of the system exhaust gas powder carrying capacity at time t+1, and g(·) is the powder carrying capacity model of the entire system;

[0263] S530, analysis of factors affecting changes in exhaust gas powder load: Analyze the influence of various state variables and control inputs on changes in exhaust gas powder load during the start-up and shutdown processes;

[0264] Sensitivity analysis:

[0265]

[0266] in, is the state variable x i Sensitivity coefficient to pollen load, is the control input u j Sensitivity coefficient to pollen load, x i is the i-th state variable, u j is the jth control input

[0267] Identification of key factors:

[0268]

[0269] in, is the correlation coefficient between the i-th state variable and the system powder carrying capacity, w key is the importance weight of the i-th state variable, N total is the total number of samples, is the mean value of the i-th state variable, is the average amount of powder carried by the system, N state is the number of state variables, N control To control the input quantity;

[0270] S540, exhaust gas powder load prediction model optimization: Combined with the real-time operating condition data set processed in S400, the exhaust gas powder load prediction model is trained with parameters and optimized in terms of model structure.

[0271] Grid Search: Theta optimal =argmin θ∈Θ L total (θ);

[0272] Among them, θ optimal is the optimal model parameter set, θ is the model parameter set, Θ is the parameter search space, L total is the total loss function;

[0273] Stochastic Gradient Descent:

[0274] Among them, θ i+1 is the model parameter of the i+1th iteration, θ i is the model parameter of the i-th iteration, η is the learning rate, is the parameter gradient operator;

[0275] S600: Using the grinding start-up condition data set and model benchmark parameters, a powder quantity correction model is constructed to calculate the powder quantity correction coefficient.

[0276] In one embodiment of the present invention, the following steps are specifically included:

[0277] S610, calculation of the corrected required quantity: based on the prediction results of the exhaust gas powder quantity prediction model established in S500, the powder quantity that needs to be corrected during the start-up and shutdown process is calculated;

[0278] Real-time correction demand: ΔM correct (t) = M dust,target (t)-M dust,system (t);

[0279] Cumulative modified demand:

[0280] Modify the priority weight:

[0281] Among them, M dust,target (t) is the target powder carrying capacity, M dust,system (t) is the actual powder carrying capacity of the system, ΔM correct (t) is the real-time correction demand, M correct,acc (t) is the cumulative correction demand, Δt is the sampling time interval, ∈ correct To avoid small amounts of division by zero, t start is the start time of the start-stop process, w correct (t) is the modified priority weight;

[0282] S620, coal feed rate dynamic adjustment model: establish a coal feed rate dynamic adjustment model based on the correction demand and determine the coal feed rate adjustment strategy of each coal mill;

[0283] Correction of coal feeding quantity of single coal mill: F coal,j,correct (t) = F coal,j,base (t)+ΔF coal,j (t);

[0284] Coal feed quantity correction distribution strategy:

[0285] Among them, F coal,j,correct (t) is the corrected coal feed rate of the j-th coal mill, F coal,j,base (t) is the reference coal feed rate of the j-th coal mill, ΔF coal,j (t) is the adjustment amount of coal feeding to the j-th coal mill, α j is the distribution coefficient of the j-th coal mill, C dust,j is the pulverized coal carrying efficiency coefficient of the jth coal mill, S mill,j (t) is the operating status of the j-th coal mill;

[0286] Partition coefficient constraints:

[0287] S630, Coal feeder speed coordinated control model: Establish a coordinated control model between the coal feeder speed and the coal feed rate to ensure that the corrected coal feed rate can be accurately implemented;

[0288] Coal feeder speed-coal feeding amount relationship: F coal,j,correct (t) = K speed,j ·N feeder,j (t)·ρ coal ;

[0289] Speed correction calculation:

[0290]

[0291] Speed change rate constraint:

[0292]

[0293] Among them, N feeder,j (t) is the speed of the j-th coal feeder, K speed,j is the speed-coal feeding coefficient of the j-th coal feeder, ρ coal is the density of pulverized coal, R speed,max,j is the maximum speed change rate of the j-th coal feeder;

[0294] S640, coal mill output balance optimization model: Optimizes the output distribution of each coal mill while meeting the pulverized coal quantity correction requirements to ensure the stability and economy of system operation;

[0295] Output balance objective function:

[0296]

[0297] The constraints include load constraints, powder carrying capacity constraints and coal mill operation constraints;

[0298] Load constraints:

[0299]

[0300] Powder carrying capacity constraint: ||M dust,system (t)-M dust,target (t)||≤∈ dust,allow ;

[0301] Coal mill operation constraints:

[0302] Among them, J balance is the output balance objective function value, w balance,j is the balance weight of the j-th coal mill, I mill,j(t) is the actual current of the j-th coal mill, I mill,j,ref is the reference current of the j-th coal mill, Q coal,j is the calorific value of coal for the jth coal mill, L boiler (t) is the boiler load, ∈ dust,allow To allow for deviation in powder carrying capacity, F coal,j,min is the minimum coal feeding amount of the j-th coal mill, F coal,j,max is the maximum coal feeding amount of the j-th coal mill;

[0303] S650, Correction Effect Feedback and Model Adaptation: Establishes a real-time feedback mechanism for correction effects and a model adaptive adjustment mechanism to improve correction accuracy;

[0304] Evaluation of the effect of the revision:

[0305]

[0306] Adaptive update of model parameters:

[0307]

[0308] K speed,j (t+1)=K speed,j (t)+β adapt ·(F coal,j,actual (t)-F coal,j,correct (t))

[0309] Sliding window correction:

[0310]

[0311] Forecast Revisions:

[0312]

[0313] Among them, E correct (t) is the correction effect evaluation index, γ adapt is the adaptive learning rate of the distribution coefficient, β adapt is the adaptive learning rate of the speed coefficient, F coal,j,actual (t) is the actual coal feeding amount of the j-th coal mill, W window is the sliding window size (number of sampling points), K predict is the prediction correction coefficient, is the average corrected demand of the sliding window, M dust,system,corrected (t) is the corrected predicted value of the system's pollen carrying capacity;

[0314] S660, multi-operating condition correction strategy switching: formulate corresponding correction strategy switching mechanism according to different start-stop operating conditions;

[0315] Working condition identification and strategy selection:

[0316]

[0317] Dynamic adjustment of strategy parameters:

[0318]

[0319] Among them, Strategy correct is the currently selected correction strategy, Strategy1, Strategy2, and Strategy3 are correction strategies under different working conditions, and Class load is the load level, G combo,start It is the coal mill start-stop combination code. G1, G2 and G3 are different coal mill combination categories. is the allocation coefficient after strategy adjustment, is the base distribution coefficient, is the speed coefficient after strategy adjustment, is the reference speed coefficient, K strategy,j is the allocation coefficient strategy adjustment coefficient, ξ strategy,j It is the speed coefficient strategy adjustment coefficient;

[0320] S700 uses the real-time exhaust gas powder carrying amount and powder amount correction coefficient during the start-up and shutdown process to calculate the powder feeder adjustment operation during the start-up and shutdown process of the pulverizing system;

[0321] In one embodiment of the present invention, the following steps are specifically included:

[0322] S710, calculation of total coal feed adjustment: Calculation of the total coal feed adjustment for the entire pulverizing system based on the pulverized coal amount correction model constructed in S600;

[0323] Total coal feed correction:

[0324]

[0325] Among them, F coal,total,correct (t) is the total coal feeding amount of the pulverizing system after correction, F coal,j,correct (t) is the corrected coal feed rate of the j-th coal mill, and M is the total number of coal mills;

[0326] Total coal feed adjustment: ΔF coal,total (t) = F coal,total,correct (t)-F coal,total,base (t);

[0327] Among them, F coal,total,base (t) is the benchmark total coal feed rate of the pulverizing system, ΔF coal,total (t) is the total coal feed adjustment amount;

[0328] S720, calculating the speed adjustment amount of the coal feeder: calculating the speed adjustment amount of each coal mill feeder according to the coal feeder speed coordinated control model in S630;

[0329] Coal feeder speed adjustment: ΔN feeder,j (t) = N feeder,j,correct (t)-N feeder,j,base (t);

[0330] Among them, N feeder,j,base (t) is the reference speed of the j-th coal feeder, N feeder,j,correct (t) is the corrected speed of the j-th coal feeder, ΔN feeder,j (t) is the speed adjustment of the j-th coal feeder;

[0331] Coal feeder speed change rate constraint:

[0332]

[0333] Among them, R speed,max,j is the maximum speed change rate allowed for the j-th coal feeder, and Δt is the control period;

[0334] S730, coal mill output adjustment calculation: Calculate the output adjustment of each coal mill according to the coal mill output balance optimization model in S640;

[0335] Calculation of coal mill output adjustment: ΔI mill,j (t) = I mill,j,correct (t)-I mill,j,base (t);

[0336] Among them, I mill,j,base (t) is the base output of the j-th coal mill, I mill,j,correct (t) is the corrected output of the j-th coal mill, ΔI mill,j (t) is the output adjustment of the j-th coal mill;

[0337] Coal mill output change rate constraint:

[0338]

[0339] Among them, R power,max,j is the maximum output change rate of the j-th coal mill;

[0340] S740, allocating control quantities to actuators: allocating the total coal feed rate adjustment quantity, the coal feeder speed adjustment quantity, and the coal mill output adjustment quantity calculated in S710-S730 to specific actuators;

[0341] Coal supply distribution: F coal,j,adjust (t) = F coal,j,correct (t)-Fcoal,j,base (t);

[0342] Among them, F coal,j,adjust (t) is the coal feed adjustment amount of the j-th coal mill, F coal,j,correct (t) is the corrected coal feed rate of the j-th coal mill, F coal,j,base (t) is the base coal feed rate of the j-th coal mill;

[0343] Coal feeder speed distribution: N feeder,j,adjust (t) = N feeder,j,correct (t)-N feeder,j,base (t);

[0344] Among them, N feeder,j,adjust (t) is the speed adjustment of the j-th coal feeder, N feeder,j,correct (t) is the corrected speed of the j-th coal feeder, N feeder,j,base (t) is the reference speed of the j-th coal feeder

[0345] Coal mill output distribution: I mill,j,adjust (t) = I mill,j,correct (t)-I mill,j,base (t);

[0346] Among them, I mill,j,adjust (t) is the output adjustment of the j-th coal mill, I mill,j,correct (t) is the corrected output of the j-th coal mill, I mill,j,base (t) is the benchmark output of the j-th coal mill;

[0347] S750, powder feeding control amount distribution and real-time feedback: the calculated control amount is distributed to the corresponding actuator and a real-time feedback mechanism is established;

[0348] The pulverized coal feed control execution includes adjusting the coal feed rate, feeder speed and mill output respectively;

[0349] Real-time feedback: monitor and provide feedback on the actual system response for subsequent control strategy optimization;

[0350] M dust,system,actual (t) = Measurement(t)

[0351] Among them, M dust,system,actual (t) is the exhaust gas powder carrying capacity actually measured by the system;

[0352]

[0353] Among them, E control (t) is the control error, M dust,target (t) is the target amount of powder carried by the deflated air;

[0354] S800: Real-time assessment of the availability of the exhaust gas powder load variation model, and regular updates of the model baseline parameters and powder load correction coefficients;

[0355] In one embodiment of the present invention, the following steps are specifically included:

[0356] S810, online identification of model parameters: Based on the powder feeding control amount distribution and real-time feedback information in S750, the powder making system model parameters are identified and updated online;

[0357] Online update of coal mill powder carrying efficiency coefficient:

[0358] in, is the estimated value of the powder carrying efficiency coefficient of the j-th coal mill at time t, γ C is the learning rate of the powder carrying efficiency coefficient, is the gradient of the control error to the powder carrying efficiency coefficient;

[0359] Coal feeder speed-coal feed rate coefficient online update:

[0360]

[0361] in, is the estimated value of the speed-coal feeding coefficient of the j-th coal feeder at time t, γ K is the learning rate of the speed-coal feed rate coefficient, is the gradient of the control error to the speed-coal feed rate coefficient;

[0362] Online update of working condition related adjustment coefficients:

[0363]

[0364] in, is the estimated value of the operating adjustment coefficient of the j-th coal mill at time t, is the estimated value of the operating condition correction coefficient of the j-th coal mill at time t, γ Ks is the learning rate of the working condition adjustment coefficient, γ ξ is the learning rate of the working condition correction coefficient, To control the gradient of the error to the working condition adjustment coefficient, is the gradient of the control error to the working condition correction coefficient;

[0365] S820, offline optimization of model parameters: removing noise and drift in the model parameters updated online by performing offline optimization regularly;

[0366] Offline parameter optimization problem definition:

[0367]

[0368] Among them, Θ is the set of model parameters that need to be optimized, t start is the starting time of the optimization time window, t end To optimize the end moment of the time window;

[0369] Parameter optimization algorithm: Use Bayesian optimization, genetic algorithm and other methods for offline parameter optimization;

[0370] Optimization results update:

[0371]

[0372] in, is the optimal value of the powder carrying efficiency coefficient after optimization of the j-th coal mill, is the optimal value of the speed-coal feeding coefficient after optimization of the j-th coal feeder, is the optimal value of the operating condition adjustment coefficient after optimization of the j-th coal mill, is the optimal value of the operating condition correction coefficient after optimization of the j-th coal mill.

[0373] The above describes the embodiments of the present invention, but the present invention is not limited to the above specific implementation methods. The above specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make many forms under the guidance of the present invention, all of which are protected by the present invention.

Claims

1. The exhaust gas powder carrying amount change model and powder feeding control method during the start-up and shutdown process of the pulverizing system are characterized by: The following steps are involved: S100, obtaining historical data of milling benchmark working conditions, performing data preprocessing and sample construction, and establishing a benchmark working condition data set; S200, obtaining historical data of milling startup conditions, performing data preprocessing and sample construction, and establishing a milling startup condition data set; S300: Using the benchmark operating condition data set, a benchmark exhaust gas powder carrying capacity model is constructed and the model benchmark parameters are calculated; S400: Obtain the start and stop signals of the pulverizing system, obtain the real-time operation data of the boiler, perform data preprocessing, and construct a real-time operating condition data set; S500: Using the model baseline parameters and real-time operating condition data set, a model for the variation of exhaust gas powder load is constructed to calculate the real-time exhaust gas powder load during the start-up and shutdown process. S600: Using the grinding start-up condition data set and model benchmark parameters, a powder quantity correction model is constructed to calculate the powder quantity correction coefficient. S700 uses the real-time exhaust gas powder carrying amount and powder amount correction coefficient during the start-up and shutdown process to calculate the powder feeder adjustment operation during the start-up and shutdown process of the pulverizing system; S800 determines the availability of the exhaust gas powder load change model in real time and regularly updates the model baseline parameters and powder load correction coefficient.

2. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 1 is characterized in that: In S100, the following steps are specifically included: S110, historical data collection and preliminary screening: extracting the benchmark operating data of the milling system from the historical database; S120, Definition of Baseline Working Condition Screening Conditions: Establishing Baseline Working Condition Screening Standards; S130, data quality inspection and cleaning: perform quality inspection on the screened data and remove outliers and missing data; S140, grouping by coal mill operation combination: grouping the data according to the coal mill operation state combination to establish data subsets of different operation conditions; S150, data standardization and normalization: standardize the grouped data to eliminate the dimension effect.

3. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 2 is characterized in that: In S200, the following steps are specifically included: S210, Grinding process data collection and identification: Identify and extract the complete data sequence of the grinding process of the pulverizing system from the historical operation data; S220, extracting key parameters of the grinding start-up condition: extracting key operating parameters during the grinding start-up process and establishing time series data; S230, determining a reference value at the start time: determining a reference parameter value at the start time as a reference point for subsequent change calculations; S240, parameter change sequence calculation: calculating the change sequence of each key parameter relative to the baseline value at the start time; S250, quality assessment of grinding start-up condition data: quality assessment of the extracted grinding start-up condition data; S260, grinding start-up stage division and feature extraction: dividing the grinding start-up process into different stages and extracting feature parameters of each stage; S270, Multi-condition Dataset Construction and Standardization: Integrate multiple grinding start-up event data to construct a standardized grinding start-up condition dataset.

4. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 3 is characterized in that: In S300, the following steps are specifically included: S310, exhaust gas dust entrainment measurement data collection and processing: Collect dust entrainment measurement data of each coal mill exhaust gas pipeline under the baseline operating conditions, and establish a corresponding relationship between dust entrainment and operating parameters; S320, selection and correlation analysis of baseline operating condition characteristic parameters: Analyze the correlation between various operating parameters in the baseline operating condition data set and the exhaust gas powder carrying capacity, and select key characteristic parameters; S330, establishing a sub-model for predicting the amount of pulverized gas carried by dust: establishing a sub-model for predicting the amount of pulverized gas carried by dust based on the operating parameters of a single coal mill; S340, Construction of the system-wide dust entrainment model: Integrate the individual models of each coal mill, consider the mutual influence between coal mills, and establish a system-wide dust entrainment model; S350, model parameter optimization and training: Use the benchmark operating condition data set to optimize the model parameters and determine the optimal parameter combination; S360, Model Validation and Performance Evaluation: Validate and evaluate the performance of the trained baseline airborne particle load model using an independent validation dataset.

5. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 4 is characterized in that: The calculation formula of the exhaust gas powder load prediction sub-model is as follows: Hidden layer output: Output layer calculation: Among them, w in,i,n 、w out,n are the weight parameters of the hidden layer and output layer of the neural network, b h,n 、b out are the bias parameters of the hidden layer and output layer of the neural network, respectively, f act is the activation function, N input 、N hidden are the number of neurons in the input layer and hidden layer respectively, X input,i is the i-th input feature, h n,j is the output of the nth hidden layer neuron of the neural network of the jth coal mill, M dust,j represents the amount of dust carried by exhaust gas of the j-th coal mill; The calculation formula of the powder carrying capacity model of the whole system is as follows: Calculation of total powder carrying capacity of the system: Interaction effects between coal mills: Load correction factor: Corrected system powder carrying capacity: M dust,system =K load ·M dust,total Among them, M dust,total is the total powder carrying capacity of the system, M interaction is the interaction term between coal mills, β j,k is the interaction coefficient between the jth coal mill and k, L boiler,rated is the rated load of the boiler, γ1 and γ2 are load correction coefficients, K load is the load correction factor, M dust,system is the total powder carrying capacity of the system after correction, S mill,j is the operating status of the j-th coal mill, I mill,j is the operating current of the j-th coal mill.

6. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 5 is characterized in that: In S400, the following steps are specifically included: S410, pulverizing system start / stop signal collection: collects start / stop signals and operating status information of each coal mill in the pulverizing system; S420, real-time boiler operation data collection: collect real-time operating parameters of the boiler and pulverizing system; S430, data quality inspection and processing: perform quality inspection and exception processing on the collected real-time data; S440, constructing a real-time working condition data set: constructing a working condition data set based on the collected real-time data; S450, data smoothing and filtering: Smoothing real-time data to eliminate noise.

7. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 6 is characterized in that: In S500, the following steps are specifically included: S510, state space modeling of the start-stop process: Establish a state space model that describes the start-stop process of the milling system, and determine the state variables and state transition equations; S520, exhaust gas powder carrying capacity prediction model construction: Based on the state space model and combined with the full system powder carrying capacity model established in S340, an exhaust gas powder carrying capacity prediction model is constructed to describe the exhaust gas powder carrying capacity changes during the start-up and shutdown process; The calculation formula of the exhaust gas powder carrying capacity prediction model is as follows: M dust,system (t+1)=g(x(t),u(t),w(t)) x(t)=[L boiler (t),F total (t),I mill,1 (t),...,I mill,M (t),ΔP mill,1 (t),...,ΔP mill,M (t)] T Among them, M dust,system (t+1) is the predicted value of the exhaust gas dust carrying capacity at time t+1, g(·) is the dust carrying capacity model of the whole system, x(t) represents the system state vector at time t, u(t) is the system control input, including coal feeding amount and coal feeder speed, w(t) is the system disturbance input, including ambient temperature and humidity, fuel quality, L boiler (t) is the boiler load at time t, F total (t) is the total frequency of the powder feeder at time t, I mill,j (t) is the current of the j-th coal mill at time t, I mill,1 (t) is the current of the first coal mill at time t, I mill,M (t) is the current of the last coal mill at time t, ΔP mill,j (t) is the pressure difference of the jth coal mill at time t, ΔP mill,1 (t) is the pressure difference of the first coal mill at time t, ΔP mill,M (t) is the pressure difference of the last coal mill at time t; S530, analysis of factors affecting changes in exhaust gas powder load: Analyze the influence of various state variables and control inputs on changes in exhaust gas powder load during the start-up and shutdown processes; S540, exhaust gas powder carrying capacity prediction model optimization: Combined with the real-time operating condition data set processed in S400, the exhaust gas powder carrying capacity prediction model is optimized for parameter training and model structure.

8. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 7 is characterized in that: In S600, the following steps are specifically included: S610, calculation of the corrected demand: based on the prediction results of the exhaust gas powder carrying amount prediction model constructed in S520, the amount of powder that needs to be corrected during the start-up and shutdown process is calculated; S620, coal feed rate dynamic adjustment model: establish a coal feed rate dynamic adjustment model based on the correction demand and determine the coal feed rate adjustment strategy of each coal mill; S630, Coal feeder speed coordinated control model: Establish a coordinated control model between the coal feeder speed and the coal feed rate to ensure that the corrected coal feed rate can be accurately implemented; S640, coal mill output balance optimization model: optimizes the output distribution of each coal mill while meeting the pulverized coal quantity correction requirements; S650, Correction Effect Feedback and Model Adaptation: Establishes a real-time feedback mechanism for correction effects and a model adaptive adjustment mechanism to improve correction accuracy; S660, multi-operating condition correction strategy switching: formulate corresponding correction strategy switching mechanism according to different start-stop operating condition types.

9. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 8 is characterized in that: In S700, the following steps are specifically included: S710, calculation of total coal feed adjustment: Calculation of the total coal feed adjustment for the entire pulverizing system based on the pulverized coal amount correction model constructed in S600; S720, calculating the speed adjustment amount of the coal feeder: calculating the speed adjustment amount of each coal mill feeder according to the coal feeder speed coordinated control model in S630; S730, coal mill output adjustment calculation: Calculate the output adjustment of each coal mill according to the coal mill output balance optimization model in S640; S740, allocating control quantities to actuators: allocating the total coal feed rate adjustment quantity, the coal feeder speed adjustment quantity, and the coal mill output adjustment quantity calculated in S710-S730 to specific actuators; S750, powder feeding control amount distribution and real-time feedback: the calculated control amount is distributed to the corresponding actuator, and a real-time feedback mechanism is established.

10. The exhaust gas powder carrying amount variation model and powder feeding control method during the start-up and shutdown process of the pulverizing system according to claim 9 is characterized in that: In S800, the following steps are specifically included: S810, online identification of model parameters: Based on the powder feeding control amount distribution and real-time feedback information in S750, the powder making system model parameters are identified and updated online; S820, offline optimization of model parameters: By performing offline optimization regularly, noise and drift that may exist in the model parameters updated online are removed.

Images