CN-121995824-A - Small powder bin and coal mill collaborative coal feeding control method based on dynamic load response

Abstract

The invention provides a small powder bin and coal mill collaborative coal feeding control method based on dynamic load response, which relates to the technical field of coal burning control of thermal power generating units, and comprises the steps of firstly collecting unit operation parameters in real time and predicting short-time load change trend by utilizing a neural network; the method comprises the steps of obtaining state information of each small powder bin and a coal mill, constructing a coal powder supply capacity model, further establishing a multi-objective optimization model based on load prediction and equipment constraint, solving the optimal coal yield ratio by adopting an Alternating Direction Multiplier Method (ADMM), dynamically generating a control instruction according to an optimization result, issuing and executing the control instruction, and realizing parameter self-adaptive correction through a deviation learning model by combining boiler operation feedback. The method has the advantages of high response speed, strong system coordination, no need of hardware transformation and the like, and remarkably improves the fuel supply efficiency and combustion stability of the ultra-supercritical unit under the working conditions of deep peak shaving and rapid climbing.

Inventors

• WANG YINGCHUN
• LIU FAZHI
• TENG FEI
• JIANG ZHIHAO
• WU MINGLIANG
• Liu Xingce
• YUAN FENGYI
• SHI XIAORAN
• SUN SHUYU
• ZHOU ZHONGRUI
• LIU YANG
• YANG DONGSHENG
• YANG YINGHUA
• DING YUMING
• Cheng Weikun
• ZHANG PENGWEI
• HE JIANLE
• HUANG YONGCHEN

Assignees

• 东北大学
• 华电电力科学研究院有限公司

Dates

Publication Date

20260508

Application Date

20260127

Claims (10)

1. 1. The method for controlling the collaborative coal feeding of the small powder bin and the coal mill based on dynamic load response is characterized by comprising the following steps: The method comprises the steps of collecting operation parameters of a super-supercritical unit in real time, and obtaining a predicted load sequence at the future time H based on the operation parameters of the super-supercritical unit Further, the future average load rising rate is calculated, and whether the future average load rising rate is smaller than the set rising load response threshold value is judged In the future the average load increase rate is less than the set load increase response threshold When the method is used, the coal feeding control mode is not changed; average load ramp rate in the future is greater than or equal to a set ramp load response threshold During the process, the state variable of the small powder bin and the state variable of the coal mill are collected in real time to construct a pulverized coal supply state vector ; Initializing a small powder bin coal outlet proportion distribution vector, and predicting a load sequence according to the future H moment And a pulverized coal supply state vector Constructing an objective function and an operation constraint condition, and processing the objective function and the operation constraint condition by adopting an alternate direction multiplier method ADMM to obtain an optimal coal yield proportion distribution vector ; Distributing vectors according to the optimal coal-out proportion Generating a blanking valve control signal Under the condition that the blanking valve control signal does not meet the limiting protection or the soft start-stop constraint or the state variable of the small powder bin meets the failure fault tolerance, returning to execute the process of collecting the operation parameters of the ultra-supercritical unit in real time; Under the condition that the blanking valve control signal meets the limit protection and soft start-stop constraint and the state variable of the small powder bin meets the failure fault tolerance, judging the actual coal output rate monitored in real time And a target coal rate Deviation of (2) Whether or not the preset threshold is exceeded, at the deviation When the operation parameters of the ultra-supercritical unit are not more than a preset threshold value, returning to execute the operation; At the deviation of When the preset threshold is exceeded, acquiring Time-of-day state feature input vector Inputting state characteristics into vectors Inputting the trained neural network function model to obtain a correction coefficient of the coal outlet proportion of the small powder bin ; Deviation in m times When the fluctuation of the coal ratio does not exceed a preset threshold value and the correction coefficient of the coal ratio of the small powder bin meets the correction amplitude limiting condition, the method is characterized in that Correcting the initial small powder bin coal-out proportion distribution vector at moment to obtain a corrected small powder bin coal-out proportion distribution vector, and returning to execute the process of collecting the operation parameters of the ultra-supercritical unit in real time and deviating in m moments When the fluctuation of the ratio exceeds a preset threshold or the correction coefficient of the coal outlet ratio of the small powder bin does not meet the correction amplitude limiting condition, the correction is suspended, and an alarm mechanism is triggered.
2. 2. The method for controlling the collaborative coal feeding of a small powder bin and a coal mill based on dynamic load response according to claim 1, wherein the method is characterized in that the operation parameters of a super-supercritical unit are collected in real time, and a predicted load sequence at the future time H is obtained based on the operation parameters of the super-supercritical unit Further, calculating a future average load increase rate includes: the method comprises the steps of collecting operation parameters of the ultra-supercritical unit in real time through a distributed control system DCS system, and constructing a system state vector, wherein the system state vector is expressed as follows: ; Wherein, the Indicating time of day Is a function of the system state vector of (a), Setting a load for current dispatching personnel; in order to actually output the electrical load, Is the pressure of the main steam and is equal to the pressure of the main steam, The coal quantity control instruction or the current coal supply quantity; collecting system state vectors at n times to form a historical data sequence Historical data sequence Expressed as: ; Wherein, the Indicating time of day Is a function of the system state vector of (a), Indicating time of day Is a system state vector of (1); sequence historical data Inputting a time sequence prediction model to obtain a predicted load sequence at the future H moment Expressed as: ; Wherein, the Indicating time of day Is used for the prediction of the load of (1), Indicating time of day Is used for the prediction of the load of (1), Indicating time of day Is a predicted load of (2); The time sequence prediction model is obtained by performing model training on a long-short-term memory network model based on a plurality of training samples, the training samples comprise an input sample and an output sample, the input sample is a data sequence, the output sample is an actual output load sequence of a future H moment corresponding to the data sequence, the input sample is input into the long-short-term memory network model to obtain a predicted load sequence of the future H moment of the data sequence, a predicted error loss function is calculated based on the actual output load sequence and the predicted load sequence, parameters in the long-short-term memory network model are modified, and the parameters in the long-short-term memory network model are iteratively modified through the training samples, so that the predicted error loss function is minimum, and a time sequence prediction model with completed training is obtained; The prediction error loss function is calculated based on the actual output load sequence and the predicted load sequence, and is realized by the following formula: ; Wherein, the The value of the error loss function is indicated, Representing time instants in the actual output load sequence The actual output load is applied to the load, Representing moments in a predicted load sequence Is a predicted load of (2); based on historical data sequences And predicting the load sequence The future average load rising rate is calculated, which is specifically realized by the following formula: ; Wherein, the Indicating the average load rise rate in the future, Indicating time of day Is used for the prediction of the load of (1), Indicating time of day The electrical load is actually output.
3. 3. The method for controlling the collaborative coal feeding of a small powder bin and a coal mill based on dynamic load response according to claim 1, wherein the state variables of the small powder bin and the state variables of the coal mill are collected in real time to construct a coal powder supply state vector Comprising: Collecting state variables of N small powder bins in real time, The state variable of the small powder bin i at the moment comprises the current material level height Discharge rate Maximum allowable blanking rate Effective bottom area of small powder bin i And coal dust density ; According to Calculating state variable of small powder bin i at moment The current remaining available time length of the instant small powder bin i The method is realized by the following specific formula: ; wherein ε is a constant that prevents divide by zero; The state variables of M coal mills are collected in real time, The state variables of coal mill j at the moment include the current coal feed rate Maximum allowable coal feed rate Coal mill current And start-stop state The start-stop state represents whether the coal mill is in operation or not; According to Calculating state variables of the moment coal mill j Load factor of coal mill j at moment The method is realized by the following specific formula: ; According to the state variables of the N small powder bins and the state vectors of the M coal mills, constructing a coal powder supply state vector Expressed as: ; Wherein, the Representation of The current level of the small powder bin 1 at the moment, Representation of The current level of the small powder bin N at the moment, Representation of The blanking rate of the small powder bin 1 at the moment, Representation of The blanking rate of the small powder bin N at the moment, Representation of The load factor of the coal mill 1 at the moment, Representation of Load factor of coal mill M at the moment.
4. 4. The method for controlling the collaborative coal feeding of a small powder bin and a coal mill based on dynamic load response according to claim 1, wherein a small powder bin coal-out proportion distribution vector is initialized, and a predicted load sequence at the future time H is obtained And a pulverized coal supply state vector Constructing an objective function and an operation constraint condition, and processing the objective function and the operation constraint condition by adopting an alternate direction multiplier method ADMM to obtain an optimal coal yield proportion distribution vector Comprising: initializing a small powder bin coal outlet proportion distribution vector, which is expressed as: ; Wherein, the The coal outlet ratio of the small powder bin 1 in unit time is shown, The coal outlet ratio of the small powder bin 2 in unit time is shown, The coal outlet ratio of the small powder bin N in unit time is represented, and T represents matrix transposition; Wherein, the , The coal outlet ratio of the small powder bin i in unit time is expressed and meets the requirements of ; Predicted load sequence based on future H time instant Calculation of Future control period predicted by time Total coal feeding amount Specifically, the method is calculated by the following formula: ; Wherein, the The conversion coefficient of the load and the coal is obtained through a table lookup method, and the future control period comprises the future time H; Defining the maximum instantaneous coal output capacity vector of small powder bin Expressed as: , wherein, Indicating the maximum instantaneous coal-discharging capacity of the small powder bin 1, Indicating the maximum instantaneous coal-out capacity of the small powder bin 2, Representing the maximum instantaneous coal outlet capacity of the small powder bin N; Defining an objective function Objective function Expressed as: ; Wherein, the Representing the error between the actual total coal rate and the predicted demand, A measure of the equilibrium of the coal proportions is shown, Indicating the difference between the actual and desired values of the load of each mill, Is an adjustable weight parameter, wherein, 、 And Calculated by the following formula: ; ; ; ; Wherein, the Represents the average value of the coal output proportion of the single-position time of all small powder bins, Load ratings indicative of optimal economic operation of the coal mill; Setting operation constraint conditions, wherein the operation constraint conditions comprise coal outlet rate constraint, powder bin allowance constraint and coal mill pulverizing capacity limitation; wherein, the coal-out rate constraint is expressed as: ; Wherein, the Representing a future control period; Wherein, the powder bin allowance constraint is expressed as: ; Wherein, coal mill pulverizing capacity restriction expresses as: ; Wherein, the The coal quantity of the coal mill which needs to be supplied to the small powder bin is represented, The direct powder supply quantity of the j-th coal mill boiler is shown, Indicating the maximum pulverizing quantity in future control period of the coal mill, 、 And Calculated by the following formula: ; ; ; Wherein, the In order to map the parameters in the matrix, Whether the small powder bin i is connected with a coal mill j or not is represented; Introducing auxiliary variables Constructing an optimized structure, expressed as: ; Wherein, the As a function of the object to be processed, As a conditional constraint function, distributing vectors, state variables of the small powder bin, state variables of a coal mill and future control period in the coal outlet proportion of the small powder bin Total coal feeding amount And the maximum instantaneous coal output capacity vector of the small powder bin Under the condition that the operation constraint condition is met, the constraint function is 0, and the coal outlet proportion distribution vector of the small powder bin, the state variable of the coal mill and the future control period are controlled Total coal feeding amount And the maximum instantaneous coal output capacity vector of the small powder bin Under the condition that the operation constraint condition is not met, the condition constraint function is + -infinity; based on an alternating direction multiplier method ADMM, carrying out iterative updating on the distribution vector of the coal outlet proportion of the small powder bin, the auxiliary variable and the Lagrangian multiplier, and specifically realizing the method by the following formula: ; ; ; Wherein, the Represents the coal proportion distribution vector of the (o+1) th iteration small powder bin, ρ is a penalty parameter for controlling the consistency constraint intensity in the ADMM, The auxiliary variable representing the o-th iteration, The lagrangian multiplier for the o-th iteration, The auxiliary variable representing the o +1 iteration, The representation is projected onto constraint set C, which is a collection of operating constraints, A lagrangian multiplier representing the (o+1) th iteration; Based on the updated small powder bin coal outlet proportion distribution vector and auxiliary variables, calculating Further, the distribution vector of the coal outlet proportion of the small powder bin, the auxiliary variable and the Lagrange multiplier are iteratively updated, and after multiple iterations, all the small powder bin is subjected to In the numerical value of (2), obtain Minimum, and obtain the corresponding small powder bin coal-out proportion distribution vector as the optimal coal-out proportion distribution vector Expressed as: ; Wherein, the Represents the optimal coal outlet proportion of the small powder bin 1, Indicating the optimal coal yield of the small powder bin 2, Indicating the optimal coal yield of the small powder bin N.
5. 5. The method for controlling the collaborative coal feeding of a small powder bin and a coal mill based on dynamic load response according to claim 1, wherein the vector is distributed according to the optimal coal yield ratio Generating a blanking valve control signal Comprising: distributing vectors according to the optimal coal-out proportion Calculating the target coal output rate of the powder bin The method is realized by the following specific formula: ; DCS system will Mapping to a blanking valve control signal The method is realized by the following specific formula: ; ; Wherein, the Is that And Is the inverse of the nonlinear mapping of a 1 ,a 2 ,a 3 , a is the nonlinear fitting parameter; Control signal of blanking valve The actual coal outlet adjustment is realized through the PLC issued to each small powder bin by the DCS system.
6. 6. The dynamic load response based small powder bin and coal mill collaborative coal feeding control method according to claim 1, wherein the clipping protection is expressed by the following formula: ; Wherein, the Indicating the minimum value allowed by the valve opening control signal, Representing the maximum value allowed by the valve opening control signal; Wherein the soft start-stop constraint is expressed by the following formula: ; Wherein, the Representation of A control signal of a blanking valve at the moment, Representing a valve opening threshold; the failure fault tolerance is that the state variable of the small powder bin is not changed within preset time.
7. 7. The dynamic load response-based small powder bin and coal mill collaborative coal feeding control method according to claim 1, wherein the method is characterized in that the method comprises the following steps of Time-of-day state feature input vector Inputting state characteristics into vectors Inputting the trained neural network function model to obtain a correction coefficient of the coal outlet proportion of the small powder bin Comprising: collecting feedback indexes, wherein the feedback indexes comprise main steam pressure errors, coal outlet deviation, coal mill current fluctuation, hearth oxygen and flame morphology; Wherein, the main steam pressure error Calculated by the following formula: ; Wherein, the Representing the actual value of the main vapour pressure, A set point representing the main vapor pressure; wherein, the coal outlet deviation Calculated by the following formula: ; Acquisition of Time optimal coal proportion distribution vector And calculate Target coal output rate at time The method is realized by the following specific formula: ; Wherein, the Representation of Optimal coal outlet proportion of the small powder bin i at the moment, Representation of Future control period predicted by time The total coal feeding amount is required; Acquisition of Time actual coal output rate Calculating the coal deviation The method is realized by the following specific formula: ; Acquisition of Time-of-day state feature input vector The state characteristic input vector comprises a system state, a scheduling instruction, a valve control instruction and boiler operation feedback, wherein the system state comprises a powder bin material level, a small powder bin blanking rate, a coal mill load rate and a scheduling instruction, and the scheduling instruction is that The optimal coal proportion distribution vector at moment, the valve control instruction is used for controlling the opening of the valve, and the boiler operation feedback comprises the actual value of the main steam pressure, the hearth oxygen and the flame offset coefficient; Will be Inputting the trained neural network function model to obtain Correction coefficient of coal outlet proportion of small powder bin at moment The method specifically comprises the following steps: Will be Inputting the first hidden layer to obtain the output of the first hidden layer Specifically, the method is represented by the following formula: ; Wherein, the The activation function is represented as a function of the activation, A weight matrix representing the first hidden layer, Representing the bias of the first hidden layer, d h represents Is represented by dx Is a feature dimension of (1); Output of the first hidden layer Input to the second hidden layer to obtain a second hidden layer output Specifically, the method is represented by the following formula: ; Wherein, the A weight matrix representing the second hidden layer, Representing the bias of the second hidden layer; Outputting the second hidden layer Input to the attention mechanism layer to obtain the output of the attention mechanism layer Specifically, the method is represented by the following formula: ; Wherein, the In order to pay attention to the weight vector, In order to activate the function, A weight matrix representing the attention mechanism layer, Indicating the bias of the attention mechanism layer, by indicating the corresponding element product operation; outputting the attention mechanism layer Inputting the correction coefficient into an output layer to obtain the correction coefficient of the coal output proportion of the small powder bin, wherein the correction coefficient is specifically expressed by the following formula: ; Wherein, the Representing a weight matrix of the output layer, Representing the bias of the output layer.
8. 8. The method for controlling the collaborative coal feeding of a small powder bin and a coal mill based on dynamic load response according to claim 7, wherein the neural network function model is trained by the following ways: Inputting the state characteristics of the historical moment into a neural network function model to obtain a correction coefficient of the coal outlet proportion of the small powder bin at the historical moment Calculating a loss function The method is realized by the following specific formula: ; according to the loss function Updating parameters of the neural network function model by adopting an Adam optimizer, and specifically realizing the parameters by the following formula: ; Wherein, the Network parameters prior to updating the neural network function model, Network parameters updated for the neural network function model, wherein eta is the learning rate; And updating the network parameters of the neural network function model for a plurality of times to obtain the trained neural network function model.
9. 9. The method for controlling the collaborative coal feeding of the small powder bin and the coal mill based on dynamic load response according to claim 1, wherein the method comprises the following steps of The initial small powder bin coal outlet proportion distribution vector at the moment is corrected to obtain the corrected small powder bin coal outlet proportion distribution vector, and the correction is realized by the following formula: ; Wherein, the Distributing vectors for the corrected coal outlet proportion of the small powder bin, Is that Optimal coal outlet proportion of the small powder bin i at the moment, Is that Correction coefficient of coal outlet proportion of the small powder bin i at the moment.
10. 10. The dynamic load response based small powder bin and coal mill collaborative coal feeding control method according to claim 1, wherein the amplitude limiting condition is modified: ; wherein, the The limit value is adjusted for the single ratio maximum, The amplitude limit is adjusted for the total.

Description

Small powder bin and coal mill collaborative coal feeding control method based on dynamic load response Technical Field The invention relates to the technical field of coal-fired control of thermal power generating units, in particular to a method for controlling coal feeding of a small powder bin and a coal mill in a coordinated manner based on dynamic load response. Background Along with the promotion of the 'double carbon' target and the large-scale access of new energy, the ultra-supercritical coal-fired unit has increasingly prominent roles in bearing the tasks of system adjustment and deep peak regulation. Compared with a conventional subcritical or supercritical unit, the supercritical unit has higher steam parameters, larger thermal inertia and stronger load adjustment capability. However, in a fast load-raising scenario, there is still a significantly short plate for the response speed and coordination of the fuel supply chain, particularly in the coal dust supply link between the coal mill and the small powder bin. In the traditional operation strategy, the small powder bin is used as a buffer device between the coal mill and the combustion system, and the small powder bin is mainly used for stabilizing the flow of coal powder and absorbing short-time fluctuation in the pulverizing process. However, the current control mode generally adopts fixed blanking proportion or simple logic judgment to adjust the coal outlet rate of each small powder bin, and lacks deep linkage with the running state and the load change trend of the coal mill, so that the coal powder supply response is delayed under the condition of high load fluctuation, and the load rising speed and the combustion stability of the unit are seriously restricted. In particular, in the process of large load jump or deep peak regulation response, the unit often needs to complete load climbing of tens of megawatts in a few minutes, and higher dynamic call regulation requirements are set for the fuel system. In the process, the coal mill needs to rapidly increase the raw coal feeding amount to improve the pulverizing capacity, and if the small powder bin is used as an intermediate storage unit and the pulverized coal allowance is insufficient or the coal outlet proportion is unreasonably arranged, coal feeding break points can occur, so that bad working conditions such as main steam pressure fluctuation of a boiler, instability of a combustor, even great change of oxygen amount and the like are caused. This lack of coordinated control not only affects combustion efficiency, but may also cause secondary oscillations of the thermal control system. In addition, when a plurality of small powder bins run in parallel, the coal outlet proportion is usually set according to an empirical value or a static parameter, and optimization matching cannot be carried out according to the load capacity and the dynamic response capacity of each coal mill, so that the contradiction between fuel supply and demand in the load adjusting process is further amplified. Although there have been studies attempting to achieve closed loop control of the fine powder bin by introducing level sensor or feed valve feedback, most are still limited to single bin-single mill local regulation, and it is difficult to solve the problems of synergistic response and pulverized coal distribution from the system level. Some power plants also try to adopt switching strategies such as 'before-feed grinding' or 'before-grinding before-feed', so as to improve the rapid load response capability, but due to lack of deep modeling of a coupling mechanism of real-time load prediction and coal output capability, the effect is often influenced by factors such as coal quality change, valve hysteresis, boiler disturbance and the like, and the adaptability is insufficient. The current main stream control strategy is still mainly set by manual intervention and static state, and lacks a dynamic cooperative regulation and control method capable of integrating real-time load prediction, bin memory assessment and coal discharge capacity analysis, so that the system has slow reaction, low coal blending efficiency and large operation fluctuation in the load rising process. Therefore, it is highly desirable to provide a collaborative coal feeding method for the operation characteristics of the ultra-supercritical unit, which can realize dynamic coordination between the coal outlet proportion of the small coal bin and the pulverizing capacity of the coal mill under the load-lifting working condition, release the potential of the coal bin in advance, realize the synchronization of rapid coal feeding and output response, break through the response bottleneck of the traditional coal powder supply chain under the rapid load-changing scene, and provide technical support for improving the flexible adjustment capacity and the combustion stability of the unit. Disclosure of Invention Aiming at the defects of the prio