Small powder bin and coal mill cooperative coal feeding control method based on dynamic load response

Abstract

The invention provides a small powder bin and coal mill cooperative coal feeding control method based on dynamic load response, and relates to the technical field of thermal power generating unit fire coal control. Firstly, operating parameters of a unit are collected in real time, and a neural network is used for predicting the short-time load change trend; then, state information of all the small powder bins and the coal mill is obtained, and a pulverized coal supply capacity model is constructed; further based on load prediction and equipment constraint, a multi-objective optimization model is established, and the optimal coal output proportion is solved by adopting an alternating direction multiplier method (ADMM); and 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 in combination with boiler operation feedback. The method has the advantages of being high in response speed, high in system coordination, free of hardware transformation and the like, and the fuel supply efficiency and the combustion stability of the ultra-supercritical unit under the working conditions of deep peak regulation and rapid climbing are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of coal combustion control technology for thermal power units, specifically to a method for coordinated coal feeding control between a small pulverized coal bin and a coal mill based on dynamic load response. Background Technology

[0002] With the advancement of "dual-carbon" goals and the large-scale integration of new energy sources, ultra-supercritical coal-fired power units are playing an increasingly prominent role in system regulation and deep peak shaving. Compared to conventional subcritical or supercritical units, ultra-supercritical units have higher steam parameters, greater thermal inertia, and stronger load regulation capabilities. However, in scenarios of rapid load increases, the response speed and coordination of their fuel supply chain still have significant shortcomings, especially in the pulverized coal supply link between the coal mill and the small pulverized coal silos. In traditional operating strategies, the small pulverized coal silos, as buffer devices between the coal mill and the combustion system, mainly function to stabilize the pulverized coal flow and absorb short-term fluctuations in the pulverizing process. However, current control methods generally use fixed feeding ratios or simple logical judgments to adjust the coal output rate of each small pulverized coal silo, lacking deep linkage with the operating status of the coal mill and the trend of load changes. This results in a lag in pulverized coal supply response under high load fluctuations, severely restricting the unit's load increase speed and combustion stability.

[0003] Especially during periods of high load shifts or deep peak shaving response, units often need to ramp up loads of tens of megawatts within minutes, placing higher demands on the dynamic regulation of the fuel system. During this process, the coal mill needs to rapidly increase the raw coal feed rate to improve pulverizing capacity; however, if the small pulverized coal silos, as intermediate storage units, have insufficient pulverized coal reserves or an improperly set coal output ratio, coal supply interruptions can occur, leading to adverse operating conditions such as fluctuations in boiler main steam pressure, burner instability, and even significant changes in oxygen levels. This lack of coordinated control not only affects combustion efficiency but may also cause secondary oscillations in the thermal control system. Furthermore, when multiple small pulverized coal silos operate in parallel, their coal output ratios are typically set based on empirical values ​​or static parameters, failing to optimize matching according to the load capacity and dynamic response capabilities of each coal mill, further amplifying the fuel supply and demand imbalance during load regulation.

[0004] Although some studies have attempted to achieve closed-loop control of small pulverized coal silos by introducing level sensors or feedback from feed valves, most remain limited to localized adjustments within a single silo and mill, failing to address the issues of coordinated response and pulverized coal distribution at the system level. Some power plants have also tried switching strategies such as "feeding first, then milling" or "milling first, then feeding" to improve rapid load response capabilities. However, due to a lack of in-depth modeling of the coupling mechanism between real-time load forecasting and coal output capacity, their effectiveness is often significantly affected by factors such as coal quality changes, valve hysteresis, and boiler disturbances, resulting in insufficient adaptability. Current mainstream control strategies still rely primarily on manual intervention and static settings, lacking a dynamic coordinated control method that integrates real-time load forecasting, silo inventory assessment, and coal output capacity analysis. This leads to slow system response, low coal blending efficiency, and large operational fluctuations during load increases.

[0005] Therefore, there is an urgent need to propose a coordinated coal feeding method for the operating characteristics of ultra-supercritical units, which can achieve dynamic coordination between the coal output ratio of the small pulverized coal bin and the pulverizing capacity of the coal mill under load increase conditions, release the potential of the pulverized coal bin in advance, and realize the synchronization of rapid coal replenishment and output response. This will break through the response bottleneck of the traditional pulverized coal supply chain in the scenario of rapid load change, and provide technical support for improving the unit's flexible adjustment capability and combustion stability. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to propose a method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response, comprising:

[0007] Real-time acquisition of operating parameters of the ultra-supercritical unit; based on the operating parameters of the ultra-supercritical unit, obtaining the predicted load sequence at time H in the future. Then, the future average load increase rate is calculated to determine whether it is less than the set load increase response threshold. If the average load increase rate in the future is less than the set load increase response threshold At the same time, the coal feeding control method remains unchanged;

[0008] The future average load increase rate is greater than or equal to the set load increase response threshold. Simultaneously, the state variables of the small pulverized coal silo and the pulverizer are collected in real time to construct a pulverized coal supply state vector. ;

[0009] Initialize the coal output ratio allocation vector of the small powder silo based on the predicted load sequence at time H in the future. and coal powder supply state vector The objective function and operational constraints are constructed, and the Alternating Direction Multiplier Method (ADMM) is used to process the objective function and operational constraints to obtain the optimal coal output ratio allocation vector. ;

[0010] Based on the optimal coal output ratio allocation vector Generate a control signal for the material feeding valve. If the control signal of the feed valve does not meet the amplitude limit protection, or does not meet the soft start and stop constraint, or the state variable of the small powder silo meets the failure tolerance, return to execution: real-time acquisition of the operating parameters of the ultra-supercritical unit;

[0011] Under the condition that the feed valve control signal meets the amplitude limiting protection and soft start / stop constraints, and the state variables of the small powder silo meet the failure tolerance, determine the actual coal output rate monitored in real time. and target coal output rate deviation Does it exceed the preset threshold, in the deviation If the preset threshold is not exceeded, return to execution: Real-time collection of operating parameters of the ultra-supercritical unit;

[0012] In deviation When the preset threshold is exceeded, obtain State feature input vector at time step Input the state features into the vector Input the trained neural network function model to obtain the correction coefficient for the coal output ratio of the small powder silo. ;

[0013] Deviation over m time intervals When the fluctuation does not exceed the preset threshold and the correction coefficient for the proportion of coal output from the small powder silo meets the correction range limit, then... The initial coal output ratio allocation vector of the small pulverized coal silo at each time step is corrected to obtain the corrected coal output ratio allocation vector of the small pulverized coal silo. The process then returns to the execution: real-time acquisition of the operating parameters of the ultra-supercritical unit; deviation within m time steps. If the fluctuation exceeds the preset threshold, or if the correction coefficient of the coal output ratio of the small powder silo does not meet the correction range limit, the correction will be suspended and an alarm mechanism will be triggered.

[0014] Optionally, the operating parameters of the ultra-supercritical unit are collected in real time, and based on the operating parameters of the ultra-supercritical unit, the predicted load sequence at time H is obtained. Then, calculate the future average load increase rate, including:

[0015] The operating parameters of the ultra-supercritical unit are collected in real time by the distributed control system (DCS) and a system state vector is constructed. The system state vector is represented as follows:

[0016] ;

[0017] in, Indicates time The system state vector, Set the load for the current dispatcher; For the actual output electrical load, Main steam pressure This refers to the coal quantity control command or the current coal feed rate.

[0018] Collect system state vectors at n time points to form a historical data sequence. Historical data sequence Represented as:

[0019] ;

[0020] in, Indicates time The system state vector, Indicates time The system state vector;

[0021] Historical data sequences Input the time series forecasting model to obtain the predicted load sequence at time H in the future. , is represented as:

[0022] ;

[0023] in, Indicates time Forecasted load, Indicates time Forecasted load, Indicates time Forecasted load;

[0024] The time-series prediction model is obtained by training a long short-term memory network model using multiple training samples. The training samples include input samples and output samples. The input samples are data sequences, and the output samples are the actual output load sequences at future time H corresponding to the data sequences. The input samples are input into the long short-term memory network model to obtain the predicted load sequences at future time H of the data sequences. The prediction error loss function is calculated based on the actual output load sequences and the predicted load sequences. The parameters in the long short-term memory network model are modified. The parameters in the long short-term memory network model are iteratively modified through training samples to minimize the prediction error loss function, thus obtaining the trained time-series prediction model.

[0025] The prediction error loss function is calculated based on the actual output load sequence and the predicted load sequence, and is achieved through the following formula:

[0026] ;

[0027] in, This represents the value of the error loss function. Represents the time in the actual output load sequence Actual output load, Represents the time in the predicted load sequence Forecasted load;

[0028] Based on historical data sequences and predicted load sequence The future average load increase rate is calculated using the following formula:

[0029] ;

[0030] in, Indicates the future average load increase rate. Indicates time Forecasted load, Indicates time Actual output electrical load.

[0031] Optionally, the state variables of the small pulverized coal silo and the pulverizer can be collected in real time to construct a pulverized coal supply state vector. ,include:

[0032] Real-time collection of status variables from N small powder bins. The state variables of the small powder hopper i at any given time include the current material level height. Feeding rate Maximum allowable feeding rate Effective bottom area of ​​small pink bin i and coal powder density ;

[0033] according to The state variable of the small pink warehouse i at time i is calculated. The current remaining available time of the Pink Warehouse i Specifically, this is achieved through the following formula:

[0034] ;

[0035] Where ε is a constant to prevent division by zero;

[0036] Real-time acquisition of the state variables of M coal mills. The state variables of the coal mill j at time j include the current coal feed rate. Maximum allowable coal feeding rate Coal mill current and start / stop status The start / stop status indicates whether the coal mill is running;

[0037] according to The state variables of the coal mill j at time j are calculated. The load rate of the coal mill j at any given time Specifically, this is achieved through the following formula:

[0038] ;

[0039] Construct a pulverized coal supply state vector based on the state variables of N small pulverized coal silos and the state vectors of M coal mills. , is represented as:

[0040] ;

[0041] in, express The current material level in the small powder silo 1 is displayed. express The current material level of the small powder silo N is displayed. express The feeding rate of the small powder hopper 1 at all times. express The feeding rate of the small powder hopper N at all times. express The load rate of coal mill 1 at all times. express The load rate of the coal mill M at any given time.

[0042] Optionally, initialize the coal output ratio allocation vector of the small powder silo based on the predicted load sequence at time H in the future. and coal powder supply state vector The objective function and operational constraints are constructed, and the Alternating Direction Multiplier Method (ADMM) is used to process the objective function and operational constraints to obtain the optimal coal output ratio allocation vector. ,include:

[0043] The initialization vector for the coal output ratio of the small powder silo is expressed as:

[0044] ;

[0045] in, This indicates the proportion of coal output per unit time from the small powder silo. This indicates the proportion of coal output from the small powder silo per unit time. N represents the proportion of coal output per unit time in the small powder silo, and T represents the matrix transpose;

[0046] in, , Let i represent the proportion of coal output from small powder silo i per unit time, and satisfy the following condition: ;

[0047] Based on the predicted load sequence at time H in the future ,calculate Future control cycle predicted at any time Total coal supply required Specifically, it is calculated using the following formula:

[0048] ;

[0049] in, The conversion coefficient between load and coal is obtained by looking up a table, and the future control period includes the future time H.

[0050] Define the vector of maximum instantaneous coal output capacity of the small coal silo. , is represented as: ,in, This indicates the maximum instantaneous coal output capacity of small powder silo 1. This indicates the maximum instantaneous coal output capacity of small coal silo 2. This represents the maximum instantaneous coal output capacity of the small coal silo N;

[0051] Define the objective function objective function Represented as:

[0052] ;

[0053] in, This represents the error between the actual total coal output rate and the predicted demand. A measure of the balance of coal output ratio. This indicates the difference between the actual and expected load values ​​of each coal mill. For adjustable weighting parameters, where, , and Calculated using the following formula:

[0054] ;

[0055] ;

[0056] ;

[0057] ;

[0058] in, This represents the average coal output ratio per unit time across all small coal silos. This indicates the load rating for optimal economic operation of the coal mill;

[0059] Set operational constraints, including coal output rate constraints, pulverizer silo reserve constraints, and pulverizing capacity limits of the coal mill.

[0060] The coal output rate constraint is expressed as follows:

[0061] ;

[0062] in, Indicates the future control period;

[0063] The powder silo balance constraint is expressed as follows:

[0064] ;

[0065] The coal mill pulverizing capacity limitation is expressed as follows:

[0066] ;

[0067] in, This indicates the amount of coal the coal mill needs to supply to the small pulverizer. This represents the direct pulverized coal supply of the j-th coal mill boiler. This indicates the maximum pulverization rate of the coal mill within its future control cycle. , and Calculated using the following formula:

[0068] ;

[0069] ;

[0070] ;

[0071] in, The parameters in the mapping matrix, Indicates whether the small powder silo i is connected to the coal mill j;

[0072] Introducing auxiliary variables Construct an optimized structure, represented as:

[0073] ;

[0074] in, Let be the objective function. The constraint functions are: the coal output ratio allocation vector in the small pulverizer silo, the state variables of the small pulverizer silo and the coal mill, and the future control period. Total coal supply required and the vector of maximum instantaneous coal output capacity of the small powder silo Under the condition that the operating constraints are met, the condition constraint function is 0, which includes the coal output ratio allocation vector of the small powder silo, the state variables of the small powder silo and the coal mill, and the future control cycle. Total coal supply required and the vector of maximum instantaneous coal output capacity of the small powder silo If the operational constraints are not met, the constraint function is +∞;

[0075] Based on the Alternating Direction Multiplier Method (ADMM), the coal output ratio allocation vector, auxiliary variables, and Lagrange multipliers of the small powder bin are iteratively updated, specifically through the following formula:

[0076] ;

[0077] ;

[0078] ;

[0079] in, Let represent the coal output ratio allocation vector of the small powder bin in the (o+1)th iteration, and ρ be the penalty parameter in ADMM that controls the strength of consistency constraints. Denotes the auxiliary variable for the o-th iteration. Denotes the Lagrange multiplier in the o-th iteration. This represents the auxiliary variable for the (o+1)th iteration. This indicates projection onto the constraint set C, where the constraint set C is the union of the runtime constraints. Represents the Lagrange multiplier in the (o+1)th iteration;

[0080] Based on the updated coal output ratio allocation vector of the small powder silo and auxiliary variables, calculate Then, the coal output ratio allocation vector, auxiliary variables, and Lagrange multipliers of the small powder silo are iteratively updated. After multiple iterations, in all From the values, obtain The minimum value is obtained, and the corresponding small coal silo coal output ratio allocation vector is taken as the optimal coal output ratio allocation vector. , is represented as:

[0081] ;

[0082] in, This indicates the optimal coal output ratio for small powder bin 1. This indicates the optimal coal output ratio for small powder bin 2. This represents the optimal coal output ratio of the small coal silo N.

[0083] Optionally, allocate vectors based on the optimal coal output ratio. Generate a control signal for the material feeding valve. ,include:

[0084] Based on the optimal coal output ratio allocation vector Calculate the target coal output rate from the powder silo. Specifically, this is achieved through the following formula:

[0085] ;

[0086] DCS system will Mapped to feed valve control signal Specifically, this is achieved through the following formula:

[0087] ;

[0088] ;

[0089] in, for and The inverse function of the nonlinear mapping, where a1, a2, a3 are nonlinear fitting parameters;

[0090] Control signal for the feeding valve The actual coal output is adjusted by sending data from the DCS system to the PLCs of each small coal silo.

[0091] Optionally, the limiting protection is expressed by the following formula:

[0092] ;

[0093] in, This indicates the minimum allowable value for the valve opening control signal. This indicates the maximum allowable value for the valve opening control signal;

[0094] The soft start-stop constraint is expressed by the following formula:

[0095] ;

[0096] in, express The constant-time control signal for the material feeding valve. Indicates the valve opening threshold;

[0097] Among them, the failure tolerance is that the state variables of the small pink warehouse do not change within a preset time.

[0098] Optional, obtain State feature input vector at time step Input the state features into the vector Input the trained neural network function model to obtain the correction coefficient for the coal output ratio of the small powder silo. ,include:

[0099] The system collects feedback indicators, including main steam pressure error, coal output deviation, coal mill current fluctuation, furnace oxygen content, and flame morphology.

[0100] Among them, the main steam pressure error Calculated using the following formula:

[0101] ;

[0102] in, This represents the actual value of the main steam pressure. This indicates the set value of the main steam pressure;

[0103] Among them, coal output deviation Calculated using the following formula:

[0104] ;

[0105] Get Optimal coal output ratio allocation vector at time 1 and calculate Target coal output rate at any time Specifically, this is achieved through the following formula:

[0106] ;

[0107] in, express The optimal coal output ratio for the small powder silo i at any given time. express Future control cycle predicted at any time Total coal supply required;

[0108] collection Actual coal output rate at any time Calculate the coal deviation Specifically, this is achieved through the following formula:

[0109] ;

[0110] Get State feature input vector at time step The state feature input vector includes system state, scheduling instructions, valve control instructions, and boiler operation feedback. The system state includes pulverized coal silo level, small pulverized coal silo discharge rate, coal mill load rate, and scheduling instructions. The optimal coal output ratio allocation vector at any given time; valve control commands are used to control valve opening; boiler operation feedback includes the actual value of main steam pressure, furnace oxygen content, and flame offset coefficient.

[0111] Will Input the trained neural network function model, and obtain Correction coefficient for the proportion of coal output from the small powder silo at any given time Specifically, it includes:

[0112] Will Input the first hidden layer and get the output of the first hidden layer. Specifically, it is expressed by the following formula:

[0113] ;

[0114] in, This represents the activation function. This represents the weight matrix of the first hidden layer. d represents the bias of the first hidden layer. h express The quantity dimension, dx represents Feature dimensions;

[0115] The output of the first hidden layer The input is fed into the second hidden layer, and the output of the second hidden layer is obtained. Specifically, it is expressed by the following formula:

[0116] ;

[0117] in, This represents the weight matrix of the second hidden layer. Indicates the bias of the second hidden layer;

[0118] Output the second hidden layer The input is fed into the attention mechanism layer, and the output of the attention mechanism layer is obtained. Specifically, it is expressed by the following formula:

[0119] ;

[0120] in, For attention weight vectors, For activation function, This represents the weight matrix of the attention mechanism layer. ⊙ represents the bias of the attention mechanism layer, and ⊙ represents the corresponding element-wise multiplication operation;

[0121] Output of the attention mechanism layer The input is fed into the output layer to obtain the correction coefficient for the coal output ratio of the small powder silo, which is specifically expressed by the following formula:

[0122] ;

[0123] in, This represents the weight matrix of the output layer. This indicates the bias of the output layer.

[0124] Optionally, the neural network function model is trained in the following way:

[0125] By inputting the state feature vector from historical moments into the neural network function model, the correction coefficient for the coal output ratio of the small powder silo at each historical moment is obtained. Calculate the loss function Specifically, this is achieved through the following formula:

[0126] ;

[0127] According to the loss function The Adam optimizer is used to update the parameters of the neural network function model, specifically through the following formula:

[0128] ;

[0129] in, These are the network parameters before the neural network function model is updated. Here, η represents the network parameters after the neural network function model is updated, and η is the learning rate.

[0130] The network parameters of the neural network function model are updated multiple times to obtain the trained neural network function model.

[0131] Optional, for The initial coal output ratio allocation vector of the small powder silo at time t is corrected to obtain the corrected coal output ratio allocation vector of the small powder silo, which is achieved through the following formula:

[0132] ;

[0133] in, This is the corrected coal output ratio allocation vector for the small powder silo. for The optimal coal output ratio for the small powder silo i at any given time. for Correction coefficient for the coal output ratio of small powder silo i at any given time;

[0134] Optional, correction range limitation conditions:

[0135] ;

[0136] in, This is the maximum adjustment limit for a single ratio. This is the limit for the total adjustment range.

[0137] The beneficial effects of adopting the above technical solution are as follows:

[0138] This invention proposes a collaborative coal feeding control method for small pulverized coal silos and coal mills based on dynamic load response. This method enables dynamic adjustment and precise control of the pulverized coal discharge ratio, overcoming the limitations of traditional fixed-ratio or static adjustment methods in terms of response speed and coordination. Secondly, this method fully considers the coupling of multi-source information, including the pulverized coal reserve in the small silo, the real-time pulverizing capacity of the coal mill, and the changing load demands of the boiler, constructing a more adaptive and forward-looking coal feeding logic that effectively alleviates the fuel supply lag problem during rapid load increases. Thirdly, this invention has good system compatibility and engineering feasibility, and can be directly embedded into existing control systems without additional hardware investment, achieving collaborative optimization control of multi-silo and multi-coal mill systems and significantly improving the overall operating efficiency of the fuel supply chain. Finally, this invention can enhance the load response capability and combustion stability of units under complex operating conditions such as deep peak shaving and frequent start-stop operations, and has significant engineering significance and application prospects for promoting the flexible transformation of thermal power units and the optimization of high-parameter unit operation. Attached Figure Description

[0139] Figure 1 This is a flowchart illustrating the method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response in an embodiment of the present invention.

[0140] Figure 2 This is a diagram of the LSTM prediction network structure in an embodiment of the present invention;

[0141] Figure 3 This is a comparison chart of the main steam load response curves under different control methods in the embodiments of the present invention;

[0142] Figure 4 This is a timing diagram of the actual valve position execution response of powder bins 1–4 in an embodiment of the present invention;

[0143] Figure 5 The figures show the comparison curves of main steam pressure and load response under the three methods in the embodiments of the present invention. Detailed Implementation

[0144] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0145] To address the issues of fuel response lag, coal output imbalance, and insufficient coordination of the pulverized coal supply chain during rapid load increases in ultra-supercritical units, this invention provides a dynamic load response-based method for coordinated coal feeding control between the pulverized coal silo and the coal mill. By combining load change trends, silo storage status, and mill operating capacity, the method achieves dynamic adjustment and predictive scheduling of the silo's coal output ratio, effectively improving the rapid response capability and coordination level of the fuel supply system. This method requires no additional hardware and possesses good engineering feasibility. It can significantly improve the load ramp-up speed and boiler combustion stability of ultra-supercritical units under complex operating conditions such as deep peak shaving and flexible operation, and has significant engineering practical value and promotional significance for promoting the intelligent operation and flexible regulation capabilities of coal-fired power units.

[0146] Based on this, the present invention provides a method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response, combined with... Figure 1 This may include the following steps:

[0147] Collect operating parameters and predict load change trends. Real-time acquisition of key operating parameters of the ultra-supercritical unit, including main steam pressure, load command, current output, and coal feed rate. Simultaneously, a neural network prediction model trained on historical operating data is constructed to identify the current load trend and predict short-term load change trends. When the prediction indicates an imminent rapid load increase phase, the coordinated coal feeding control process is triggered, as detailed in step 1.

[0148] Step 1: Real-time acquisition of operating parameters of the ultra-supercritical unit; based on the operating parameters of the ultra-supercritical unit, obtain the predicted load sequence at time H in the future. Then, the future average load increase rate is calculated to determine whether it is less than the set load increase response threshold. If the average load increase rate in the future is less than the set load increase response threshold At that time, without changing the coal feeding control method, if the future average load increase rate is greater than or equal to the set load increase response threshold... Then, proceed to step 2;

[0149] Step 1.1: The operating parameters of the ultra-supercritical unit are collected in real time through the Distributed Control System (DCS), and a system state vector is constructed. The system state vector is represented as follows:

[0150] ;

[0151] in, Indicates time The system state vector, Set the load (MW) for the current dispatcher; This represents the actual output electrical load (MW). Main steam pressure (MPa). This refers to the coal quantity control command or the current coal feed rate (t / h).

[0152] The state vector is updated every time step Δt. The system state vectors at n time points are collected to form a historical data sequence. Historical data sequence Represented as:

[0153] ;

[0154] in, Indicates time The system state vector, Indicates time The system state vector;

[0155] Step 1.2: Extract historical data sequences Input the time series forecasting model to obtain the predicted load sequence at time H in the future. , is represented as:

[0156] ;

[0157] in, Indicates time Forecasted load, Indicates time Forecasted load, Indicates time Forecasted load;

[0158] The time series prediction model consists of an input layer, a hidden layer, and an output layer, with the input layer having a dimension of [missing information]. Input historical state sequence The hidden layer consists of a combination of several recurrent layers and fully connected layers, and the output layer outputs the predicted load sequence for the next H steps.

[0159] The time-series prediction model is obtained by training a long short-term memory network model using multiple training samples. The training samples include input samples and output samples. The input samples are data sequences, and the output samples are the actual output load sequences at future time H corresponding to the data sequences. The input samples are input into the long short-term memory network model to obtain the predicted load sequences at future time H of the data sequences. The prediction error loss function is calculated based on the actual output load sequences and the predicted load sequences. The parameters in the long short-term memory network model are modified. The parameters in the long short-term memory network model are iteratively modified through training samples to minimize the prediction error loss function, thus obtaining the trained time-series prediction model.

[0160] The prediction error loss function is calculated based on the actual output load sequence and the predicted load sequence, and is achieved through the following formula:

[0161] ;

[0162] in, This represents the value of the error loss function. Represents the time in the actual output load sequence Actual output load, Represents the time in the predicted load sequence Forecasted load;

[0163] Step 1.3: Based on historical data sequences and predicted load sequence The future average load increase rate is calculated using the following formula:

[0164] ;

[0165] in, Indicates the future average load increase rate. Indicates time Forecasted load, Indicates time Actual output electrical load;

[0166] Step 1.4: Determine whether the future average load increase rate is greater than or equal to the set load increase response threshold. If the average future load increase rate is greater than or equal to the set load increase response threshold If the future average load increase rate is less than the set load increase response threshold, proceed to step 2. At this time, the system has not entered the rapid load increase mode and the coal feeding control method has not been changed;

[0167] Step 2: Obtain the operating status of small pulverized coal silos and coal mills. Collect real-time information on pulverized coal level, feeding rate, and upper limit of feeding capacity for all small pulverized coal silos, and obtain parameters such as coal feed rate, current load, start / stop status, and pulverizing capacity boundary for each coal mill. Construct a complete description of pulverized coal supply capacity by fusing the status data, providing an input basis for subsequent optimization. Specifically, collect the state variables of the small pulverized coal silos and the coal mills in real time to construct a pulverized coal supply state vector. ;

[0168] Step 2.1: Real-time collection of state variables from N small powder bins. The state variables of the small powder hopper i at any given time include the current material level height. (Unit: m) Feeding rate (Unit: t / h) Maximum allowable feeding rate Effective bottom area of ​​small pink bin i (Unit: m) 2 ) and coal powder density (Unit: t / m³).

[0169] according to The state variable of the small pink warehouse i at time i is calculated. The current remaining available time of the Pink Warehouse i This refers to how long the coal supply can continue without additional coal replenishment, which is determined by the following formula:

[0170] ;

[0171] Where ε is a small constant to prevent division by zero;

[0172] Step 2.2: Real-time acquisition of the state variables of M coal mills. The state variables of the coal mill j at time j include the current coal feed rate. (Unit: t / h) Maximum allowable coal feeding rate Coal mill current (Reflecting load) and start / stop status The start / stop status indicates whether the coal mill is running;

[0173] according to The state variables of the coal mill j at time j are calculated. The load rate of the coal mill j at any given time Specifically, this is achieved through the following formula:

[0174] ;

[0175] The load rate is used to evaluate the current operating intensity of the coal mill, serving as an important reference for allocating tasks to downstream pulverizer silos.

[0176] Step 2.3: Construct the pulverized coal supply state vector based on the state variables of N small pulverized coal silos and the state vectors of M coal mills. , is represented as:

[0177] ;

[0178] in, express The current material level in the small powder silo 1 is displayed. express The current material level of the small powder silo N is displayed. express The feeding rate of the small powder hopper 1 at all times. express The feeding rate of the small powder hopper N at all times. express The load rate of coal mill 1 at all times. express The load rate of the coal mill M at any given time;

[0179] Among them, the pulverized coal supply state vector is used as the input of the subsequent optimization scheduling model to realize the coordinated calculation and dynamic adjustment of coal output from the small pulverized coal bin and the coal mill.

[0180] A coal powder supply and demand matching optimization model is constructed. Based on the load forecast demand in step 1 and the system status data obtained in step 2, a multi-objective optimization model is constructed. This model uses minimum load response delay, optimal small coal silo utilization rate, and coal mill load balance as objective functions, while also considering constraints such as small coal silo feeding capacity and coal mill operating status. To achieve real-time optimization, the ADMM algorithm is used to solve for the optimal coal output ratio, which is explained in detail in step 3.

[0181] Step 3: Initialize the coal output ratio allocation vector of the small powder silo, based on the predicted load sequence at time H in the future. and coal powder supply state vector The objective function and operational constraints are constructed, and the Alternating Direction Multiplier Method (ADMM) is used to process the objective function and operational constraints to obtain the optimal coal output ratio allocation vector. ;

[0182] Step 3.1: Initialize the coal output ratio allocation vector of the small powder silo, represented as:

[0183] ;

[0184] in, This indicates the proportion of coal output per unit time from the small powder silo. This indicates the proportion of coal output from the small powder silo per unit time. N represents the proportion of coal output per unit time in the small powder silo, and T represents the matrix transpose;

[0185] in, , Let i represent the proportion of coal output from small powder silo i per unit time, and satisfy the following condition: ;

[0186] Based on the predicted load sequence at time H in the future ,calculate Future control cycle predicted at any time Total coal supply required The unit is tons (t), and it is calculated using the following formula:

[0187] ;

[0188] in, The conversion coefficient between load and coal is obtained by looking up a table, and the future control period includes the future time H.

[0189] Define the vector of maximum instantaneous coal output capacity of the small coal silo. , is represented as: ,in, This indicates the maximum instantaneous coal output capacity of small powder silo 1. This indicates the maximum instantaneous coal output capacity of small coal silo 2. This represents the maximum instantaneous coal output capacity of the small powder silo N, in tons (t).

[0190] Define the objective function objective function Represented as:

[0191] ;

[0192] in, This represents the error between the actual total coal output rate and the predicted demand. A measure of the balance of coal output ratio. This indicates the difference between the actual and expected load values ​​of each coal mill. These are adjustable weighting parameters used to balance response speed, uniformity, and coal mill load regulation. , and Calculated using the following formula:

[0193] ;

[0194] ;

[0195] ;

[0196] ;

[0197] in, This represents the average coal output ratio per unit time across all small coal silos. This indicates the load rating for optimal economic operation of the coal mill;

[0198] Step 3.2: Set operational constraints, including coal output rate constraints, pulverizer silo reserve constraints, and pulverizing capacity limits of the coal mill;

[0199] Among them, the coal output rate constraint is used to prevent exceeding the equipment capacity, and the coal output rate constraint is expressed as:

[0200] ;

[0201] in, Indicates the future control period, in hours (h).

[0202] Among them, the pulverized coal silo reserve constraint is used to prevent pulverized coal from being emptied, and the pulverized coal silo reserve constraint is expressed as:

[0203] ;

[0204] The coal mill pulverizing capacity limit is used to prevent coal mill overload, and the coal mill pulverizing capacity limit is expressed as follows:

[0205] ;

[0206] in, This indicates the amount of coal the coal mill needs to supply to the small pulverizer. This represents the direct pulverized coal supply to the j-th coal mill boiler, which is the coal output of the coal mill minus the amount of coal the coal mill needs to supply to the small pulverized coal bin during this period. This indicates the maximum pulverization rate of the coal mill within its future control cycle. , and Calculated using the following formula:

[0207] ;

[0208] ;

[0209] ;

[0210] in, The parameters in the mapping matrix, This characterizes whether small powder silo i is connected to coal mill j. If small powder silo i is connected to coal mill j, then... Otherwise, it is 0;

[0211] To achieve the optimal pulverized coal allocation control strategy at the system level, this invention constructs a multi-objective constrained optimization model in step 3.1. The objective is to dynamically determine the allocation of each small pulverized coal silo within the current control cycle based on the main steam load variation trend. Optimal coal extraction ratio within Since the coal output behavior of each silo is not only constrained by its own physical limitations but also coupled with the state of its respective coal mill, the overall problem is characterized by strong coupling and high dimensionality. Therefore, to improve the parallelism, scalability, and industrial deployment efficiency of the algorithm, this invention introduces the Alternating Direction Multiplier Method (ADMM) for distributed solution of this scheduling problem, and introduces auxiliary variables. Specifically, this is achieved through step 3.3.

[0212] Step 3.3: Introduce auxiliary variables Construct an optimized structure, represented as:

[0213] ;

[0214] in, Let be the objective function. The constraint functions are: the coal output ratio allocation vector in the small pulverizer silo, the state variables of the small pulverizer silo and the coal mill, and the future control period. Total coal supply required and the vector of maximum instantaneous coal output capacity of the small powder silo Under the condition that the operating constraints are met, the condition constraint function is 0, which includes the coal output ratio allocation vector of the small powder silo, the state variables of the small powder silo and the coal mill, and the future control cycle. Total coal supply required and the vector of maximum instantaneous coal output capacity of the small powder silo If the operational constraints are not met, the constraint function is +∞;

[0215] Based on the Alternating Direction Multiplier Method (ADMM), the coal output ratio allocation vector, auxiliary variables, and Lagrange multipliers of the small powder bin are iteratively updated, specifically through the following formula:

[0216] ;

[0217] ;

[0218] ;

[0219] in, This represents the coal output ratio allocation vector of the small powder bin in the (o+1)th iteration, where ρ is the penalty parameter in ADMM that controls the strength of consistency constraints, used to ensure... and Asymptotic consistency, and affects the algorithm's convergence behavior; Denotes the auxiliary variable for the o-th iteration. Denotes the Lagrange multiplier in the o-th iteration. This represents the auxiliary variable for the (o+1)th iteration. This indicates projection onto the constraint set C, where the constraint set C is the union of the runtime constraints. This represents the Lagrange multiplier in the (o+1)th iteration. The Lagrange multiplier update is used to adjust the deviation between the main variable and the auxiliary variable to be consistent, ensuring eventual convergence.

[0220] Based on the updated coal output ratio allocation vector of the small powder silo and auxiliary variables, calculate Then, the coal output ratio allocation vector, auxiliary variables, and Lagrange multipliers of the small powder silo are iteratively updated. After multiple iterations, in all From the values, obtain The minimum value is obtained, and the corresponding small coal silo coal output ratio allocation vector is taken as the optimal coal output ratio allocation vector. , is represented as:

[0221] ;

[0222] in, This indicates the optimal coal output ratio for small powder bin 1. This indicates the optimal coal output ratio for small powder bin 2. This represents the optimal coal output ratio of the small powder silo N;

[0223] Step 4: Generate dynamic coal output ratio and issue control commands. The coal output ratio of each small silo obtained from the optimization algorithm is used as the control target to generate corresponding control commands, adjusting the opening of the feed valve or regulating the logic. The control commands are executed periodically in the DCS system, forming a dynamic coal output mechanism that tracks load changes in real time. Specifically, a vector is allocated according to the optimal coal output ratio. Generate a control signal for the material feeding valve. If the control signal of the feeding valve does not meet the amplitude limiting protection, or does not meet the soft start and stop constraint, or the state variable of the small powder silo meets the failure fault tolerance, return to step 1.

[0224] Under the condition that the feed valve control signal meets the amplitude limiting protection and soft start / stop constraints, and the state variables of the small powder silo meet the failure tolerance, determine the actual coal output rate monitored in real time. and target coal output rate deviation Does it exceed the preset threshold, in the deviation If the value does not exceed the preset threshold, return to step 1; if the value deviates... If the preset threshold is exceeded, proceed to step 5;

[0225] Step 4.1: Allocate vectors according to the optimal coal output ratio Calculate the target coal output rate from the powder silo. Specifically, this is achieved through the following formula:

[0226] ;

[0227] Step 4.2: The DCS system will Mapped to feed valve control signal The mapping relationship is obtained from on-site calibration and can be expressed as the inverse form of a nonlinear function, specifically achieved through the following formula:

[0228] ;

[0229] ;

[0230] in, for and The inverse function of the nonlinear mapping, where a1, a2, and a3 are nonlinear fitting parameters that can be adjusted and set according to valve model, installation conditions, and pulverized coal properties.

[0231] Control signal for the feeding valve By using the DCS system to continuously send commands to the PLCs of each small coal bin in a cycle τ, the continuity and real-time nature of the commands are ensured, and the actual coal output is adjusted. To ensure execution safety, this invention designs an execution protection mechanism.

[0232] Step 4.3: Set up an execution guarantee mechanism, which includes amplitude limiting protection, soft start and stop constraints, failure tolerance, and actual rate verification mechanism;

[0233] The amplitude limiting protection is expressed by the following formula:

[0234] ;

[0235] in, This indicates the minimum allowable value for the valve opening control signal. This indicates the maximum allowable value for the valve opening control signal;

[0236] Among them, the soft start / stop constraint requires that the valve opening degree must not change more than a threshold within one cycle. It can be expressed by the following formula:

[0237] ;

[0238] in, express The constant-time control signal for the material feeding valve. Indicates the valve opening threshold;

[0239] Among them, the failure tolerance is that the state variables of the small pink warehouse do not change within a preset time.

[0240] If the control signal of the feeding valve does not meet the amplitude limiting protection, or does not meet the soft start and stop constraint, or if the state variable of the small powder silo meets the failure tolerance, return to step 1.

[0241] Under the condition that the feed valve control signal meets the amplitude limiting protection and soft start / stop constraints, and the state variables of the small powder silo meet the failure tolerance, determine the actual coal output rate monitored in real time. Does it meet the actual rate verification mechanism? Specifically, it determines the actual coal output rate monitored in real time. and target coal output rate deviation Does it exceed the preset threshold, in the deviation If the threshold is not exceeded, return to step 1.

[0242] In deviation If the preset threshold is exceeded, proceed to step 5.

[0243] The system provides real-time feedback on control performance and performs adaptive corrections. During control execution, the system continuously collects operational indicators such as boiler main steam pressure response, burner flame stability, and actual pulverized coal flow rate to evaluate the control effect. If the actual response deviates significantly from the prediction, a deviation learning model constructed using a neural network is used for correction, and parameters or control logic are automatically updated and optimized to enhance the system's adaptability and robustness. This is explained in detail in step 5.

[0244] Step 5: Obtain State feature input vector at time step Input the state features into the vector Input the trained neural network function model to obtain the correction coefficient for the coal output ratio of the small powder silo. Deviation over m time intervals When the fluctuation does not exceed the preset threshold and the correction coefficient for the proportion of coal output from the small powder silo meets the correction range limit, then... The initial coal output ratio allocation vector of the small powder silo at time step m is corrected, and step 1 is returned to be executed; the deviation within m time steps is... When the fluctuation exceeds the preset threshold, or when the correction coefficient for the coal output ratio of the small powder silo does not meet the correction range limit, the correction is suspended and an alarm mechanism is triggered.

[0245] Step 5.1: Collect feedback indicators, including main steam pressure error, coal output deviation, coal mill current fluctuation, furnace oxygen content, and flame morphology;

[0246] Among them, the main steam pressure error Calculated using the following formula:

[0247] ;

[0248] in, This represents the actual value of the main steam pressure. This indicates the set value of the main steam pressure;

[0249] Among them, coal output deviation Calculated using the following formula:

[0250] ;

[0251] Step 5.2: This step aims to establish a learning and correction mechanism for dynamic coal output deviation in small coal silos. This mechanism predicts the deviation between actual coal output and dispatch instructions caused by coal quality disturbances, equipment response lags, or uneven mill loads, and intelligently corrects the next round of coal output ratio accordingly. Specifically, it obtains... Optimal coal output ratio allocation vector at time 1 and calculate Target coal output rate at any time Specifically, this is achieved through the following formula:

[0252] ;

[0253] in, express The optimal coal output ratio for the small powder silo i at any given time. express Future control cycle predicted at any time Total coal supply required;

[0254] collection Actual coal output rate at any time Calculate the coal deviation Specifically, this is achieved through the following formula:

[0255] ;

[0256] Step 5.3: Obtain State feature input vector at time step The state feature input vector includes system state, scheduling instructions, valve control instructions, and boiler operation feedback. The system state includes pulverized coal silo level, small pulverized coal silo discharge rate, coal mill load rate, and scheduling instructions. The optimal coal output ratio allocation vector at any given time; valve control commands are used to control valve opening; boiler operation feedback includes the actual value of main steam pressure, furnace oxygen content, and flame offset coefficient.

[0257] Will Input the trained neural network function model, and obtain Correction coefficient for the proportion of coal output from the small powder silo at any given time Specifically, it includes:

[0258] Will Input the first hidden layer (encoding layer) and obtain the output of the first hidden layer. Specifically, it is expressed by the following formula:

[0259] ;

[0260] in, This represents the activation function. This represents the weight matrix of the first hidden layer. d represents the bias of the first hidden layer. h express The quantity dimension, dx represents Feature dimensions;

[0261] The output of the first hidden layer The input is fed into the second hidden layer (higher-order abstraction layer), and the output of the second hidden layer is obtained. Specifically, it is expressed by the following formula:

[0262] ;

[0263] in, This represents the weight matrix of the second hidden layer. Indicates the bias of the second hidden layer;

[0264] Output the second hidden layer The input is fed into the attention mechanism layer, and the output of the attention mechanism layer is obtained. Specifically, it is expressed by the following formula:

[0265] ;

[0266] in, For attention weight vectors, For activation function, This represents the weight matrix of the attention mechanism layer. ⊙ represents the bias of the attention mechanism layer, and ⊙ represents the corresponding element-wise multiplication operation;

[0267] Output of the attention mechanism layer The input is fed into the output layer to obtain the correction coefficient for the coal output ratio of the small powder silo, which is specifically expressed by the following formula:

[0268] ;

[0269] in, This represents the weight matrix of the output layer. Indicates the bias of the output layer;

[0270] To ensure the model maintains good adaptability under different operating conditions, coal quality fluctuations, equipment aging, and other factors, this invention designs a dual-mode mechanism combining offline training and online fine-tuning. The neural network function model is trained in the following way:

[0271] By inputting the state feature vector from historical moments into the neural network function model, the correction coefficient for the coal output ratio of the small powder silo at each historical moment is obtained. Calculate the loss function Specifically, this is achieved through the following formula:

[0272] ;

[0273] This loss function measures the mean square error between the model correction term and the actual scheduling deviation, and has good differentiability and engineering interpretability.

[0274] According to the loss function The Adam optimizer is used to update the parameters of the neural network function model, specifically through the following formula:

[0275] ;

[0276] in, These are the network parameters before the neural network function model is updated. Here, η represents the network parameters after the neural network function model is updated, and η is the learning rate.

[0277] The network parameters of the neural network function model are updated multiple times to obtain the trained neural network function model.

[0278] Step 5.4: Deviation over m time intervals When the fluctuation does not exceed the preset threshold and the correction coefficient for the coal output ratio of the small powder silo meets the correction range limit, based on Correction coefficient for the proportion of coal output from the small powder silo at any given time ,right The initial coal output ratio allocation vector of the small powder bin at each time step is corrected to accelerate the algorithm's convergence speed and improve the response accuracy to system operating condition disturbances. This enables intelligent linkage closed-loop control of scheduling optimization and feedback learning, resulting in the corrected coal output ratio allocation vector of the small powder bin. Specifically, this is achieved through the following formula:

[0279] ;

[0280] Return to step 1:

[0281] Deviation over m time intervals If the fluctuation exceeds the preset threshold, or if the correction coefficient of the coal output ratio of the small powder silo does not meet the correction range limit, the correction will be suspended and an alarm mechanism will be triggered.

[0282] Among them, the correction range limitation conditions are as follows:

[0283] ;

[0284] in, This is the maximum adjustment limit for a single ratio. This is a limit on the total adjustment range;

[0285] Steps 1 to 5 are executed at a fixed period τ, forming a rolling optimization control closed loop. This continuously updates the prediction, optimization, execution, and feedback modules, achieving rapid load response and optimal coal blending coordination. It also enables intelligent sensing and self-correction of disturbances and deviations, and system-level coupled scheduling of multiple pulverized coal silos and multiple coal mills. The final system architecture resembles an intelligent pulverized coal supply system driven by data, with state awareness, intelligent decision-making, execution feedback, and adaptive learning, comprehensively improving the fuel response efficiency and system stability of ultra-supercritical units under flexible operating scenarios.

[0286] Example 1: Single-unit dual-compartment rapid load ramp-up control case

[0287] (1) Scene setting and initial parameter description

[0288] This embodiment selects some actual operating conditions of a 600MW ultra-supercritical coal-fired power generating unit, setting up a scenario of coordinated coal feeding control between the small pulverizer silo and the coal mill under rapid load increase conditions. Assuming that the current unit load is stable at 300MW, the system receives a dispatch command and needs to rapidly increase to 360MW within 5 minutes (300 seconds), with a load change rate of 12MW / min per unit time, which is a high-response demand condition.

[0289] The controlled object is a single operating coal mill and its two corresponding small pulverized coal bins (bin A and bin B). The system has real-time data acquisition and execution control capabilities. The control cycle Δt is set to 10 seconds, and the system performs full-process prediction, optimization, and control every 10 seconds. At the initial moment (t = 0), the key operating parameters acquired are as follows:

[0290] Table 1 Key Operating Parameters

[0291]

[0292] (2) Load forecasting and load increase trend identification

[0293] Based on the initial operational data collection completed in step 1, the system enters the load trend prediction stage. This embodiment uses a Long Short-Term Memory (LSTM) neural network as the short-term load change trend prediction model. Its training dataset consists of the unit's historical operational data, collected every 10 seconds, covering a 60-day operational cycle, including multiple typical load increase and decrease conditions, with a total of approximately 43,200 training samples. The prediction model structure is as follows... Figure 2 As shown, it includes the following main modules:

[0294] Input layer: The input dimension is n×d, where n=30 is the time step length and d=4 is the input feature dimension, including scheduling load command, current load, main steam pressure, current coal feeding rate, etc.

[0295] LSTM layer: Assume 2 stacked LSTM layers, with 64 hidden units in each layer and tanh activation function;

[0296] Fully connected layer: connects the hidden state of the last time step and outputs the predicted value;

[0297] Output layer: Outputs the electrical load prediction sequence within a future prediction window H=6 steps. Each step has a 10-second interval.

[0298] The model is trained using supervised learning, and the loss function is the weighted MSE error.

[0299] ;

[0300] in, This is used to emphasize the importance of long-term predictions. The optimizer is Adam, the initial learning rate is set to 0.001, and an early stopping mechanism is used to prevent overfitting. The prediction accuracy on the test set is as follows:

[0301] Table 2 Comparison of Forecast Values ​​and Actual Load

[0302] After the predicted sequence is generated, the system calculates the load growth rate:

[0303] ;

[0304] Where Δt = 10s, H = 6, if If this is the case, then the system is considered to be in a phase of rapid load increase. This embodiment sets a threshold. If the average increase per minute is not less than 4.8MW, the subsequent collaborative scheduling strategy (steps 3-6) will be triggered, and the small powder silo-coal mill intelligent coordinated control process will begin.

[0305] (3) Construct a coal supply optimization model and solve for the optimal ratio.

[0306] To achieve precise coal blending during rapid load increases, this embodiment employs the aforementioned ADMM algorithm for rolling optimization of the coal output ratio. Based on the load growth trend predicted in step 2, and within the control cycle, a distributed pulverized coal supply optimization model is constructed with the objectives of improving the main steam response speed, balancing the extraction intensity of the small pulverized coal silos, and avoiding coal supply interruptions. This model is then solved using high-frequency iterative methods on a simulation platform. The specific process is as follows:

[0307] Control object settings: Powder bin A and powder bin B are respectively connected to the same coal mill, allowing dynamic adjustment of the coal output ratio;

[0308] Input parameters include: current main steam load trend forecast, powder silo level, and discharge capacity boundary;

[0309] Optimization results: The ADMM algorithm converges quickly within 5 rounds, and the final output ratio vector is: target ratio of pink bin A: 0.42, target ratio of pink bin B: 0.58.

[0310] The optimization results are shown in Table 3. These results indicate that, considering the current low coal level in silo A, the system automatically tends to increase the coal supply load in silo B, while ensuring that the overall coal supply matches the load demand. Compared to a fixed 1:1 ratio scheme, this scheme has significant advantages in terms of residual pulverized coal control and boiler load response speed.

[0311] Table 3 ADMM Iterative Optimization Results

[0312]

[0313] (4) Opening instruction generation and execution logic

[0314] In step 3, the system obtains the optimal coal output ratio for each small coal bin under the current scheduling cycle using the ADMM algorithm. This step transforms the optimization results into valve opening control commands that can be actually executed by the DCS system, thus realizing the engineering implementation of the coal supply control intent.

[0315] Let the total coal demand corresponding to the load forecast for this period be... Combined with ADMM output ratio results Then the target coal output of powder silos A and B is:

[0316] ;

[0317] This rate serves as the basis for quantification of subsequent valve control logic.

[0318] In actual control, coal discharge from the powder silo is achieved by adjusting the opening of the feeding valve. The system uses a valve opening function to control the coal discharge rate. Mapped to valve opening control signal Based on the on-site calibration results, a cubic function fitting model was adopted:

[0319] ;

[0320] in, These are the empirical regression coefficients. Substituting them into the target rate, we calculate: Valve opening of powder silo A: approximately 25.7%, Valve opening of powder silo B: approximately 34.9%.

[0321] This signal is updated periodically at intervals of τ (10 seconds) and is sent from the central controller to the PLC systems of each powder silo. To enhance operational safety and field adaptability, the following protection strategies are designed in this embodiment:

[0322] 1. Bandwidth Limiting Protection Mechanism

[0323] To prevent abrupt changes in the valve within a single cycle, the maximum range of variation is limited to:

[0324] ;

[0325] 2. Soft start-stop control

[0326] If the powder silo did not operate in the previous cycle or the material level was below the threshold, a slow start logic is adopted, gradually raising it to the target value within 3 cycles.

[0327] 3. Fault tolerance rollback mechanism

[0328] If the feedback sensor detects that the valve is stuck or unresponsive, the system will temporarily switch to an average proportion coal blending strategy (e.g., 50%:50%) and trigger an alarm to prompt manual inspection.

[0329] 4. Actual speed verification mechanism

[0330] The system provides real-time comparison and feedback on coal output rate. With target rate When the error exceeds the set threshold (±5%) for three consecutive cycles, the process automatically switches to the deviation learning and correction process in step 5.

[0331] Table 4 shows the comparison between the target coal extraction rate and the actual rate in this experiment.

[0332] Table 4 Comparison of Target Coal Production Rate and Actual Rate

[0333]

[0334] The above control strategy ensures that the system can maintain high coal supply stability and operational safety even under conditions of coal quality changes, equipment aging, or measurement noise interference.

[0335] (5) Deviation feedback correction mechanism and model update

[0336] During the execution of valve control commands, the system monitors the actual coal output rate of each pulverizer silo in real time and compares it with the expected value to assess the deviation. In this embodiment, a neural network deviation learning model based on an attention mechanism is introduced to predict and correct coal output errors, thereby improving the robustness and adaptability of the scheduling strategy.

[0337] The actual coal output of each silo is collected during each control cycle. And the target coal output in the optimized output For comparison, the error term is defined as:

[0338] ;

[0339] When the deviation continues to exceed the threshold If this continues for more than two cycles, a correction process will be triggered.

[0340] This embodiment employs a multi-layer feedforward neural network with an attention weighting mechanism, the input of which includes:

[0341] Current system state vector : Powder silo level, coal mill load rate, boiler pressure, etc.;

[0342] Previous cycle control instructions;

[0343] Historical sequence of actual coal output deviation.

[0344] The output is a scaling correction factor. Used to update the scheduling ratio for the next cycle:

[0345] ;

[0346] The model updates training data through a sliding window, automatically performing incremental training every 10 cycles. It uses mean squared error (MSE) as the loss function, Adam as the optimizer, and a learning rate of 0.001. Table 5 shows a comparison of the uncorrected and corrected coal extraction errors (RMSE and MAE).

[0347] Table 5 Comparison of Coal Output Errors Before and After Deviation Correction

[0348]

[0349] The results show that after introducing the deviation correction mechanism, the coal supply error of each powder bin decreased significantly, and the system stability and response accuracy were greatly improved.

[0350] (6) Closed-loop rolling execution

[0351] This embodiment executes steps 1 to 5 in a rolling fashion over 300 seconds, with each cycle lasting 10 seconds, for a total of 30 control cycles. This achieves closed-loop control of LSTM prediction, ADMM optimization, command issuance, feedback acquisition, and model correction. The overall system performance is shown in Table 6. Figure 3 The boiler response curves under different control methods are shown.

[0352] Table 6 Overall System Performance

[0353]

[0354] This embodiment fully verifies the control performance and engineering applicability of the "cooperative coal feeding control method for small pulverized coal silos and coal mills based on neural network prediction and ADMM optimization" proposed in this invention under rapid load increase conditions. By applying this method during the dynamic load response process from 300MW to 360MW, the unit successfully completed the load ramp-up within 5 minutes, shortening the response time by approximately 8% compared to the traditional fixed-proportion strategy. Simultaneously, the accuracy of coal output flow rate was significantly improved, with RMSE decreasing by over 63%, effectively reducing coal feeding deviation. The main steam pressure fluctuation range was significantly narrowed, enhancing the stability and safety margin of the combustion system. Particularly noteworthy is that, through the linkage of dynamic coal output ratio scheduling and feedback correction mechanisms, the system completely suppressed the risk of pulverized coal silo emptying, significantly improving the resilience and reliability of the pulverized coal supply chain. Furthermore, the deviation learning neural network effectively compensated for deviations caused by equipment nonlinearity and coal quality disturbances, achieving adaptive updates and rolling optimization of the control strategy. In summary, this cooperative control method not only has good industrial implementation feasibility but also significantly improves the coal supply response speed and overall system operating efficiency of ultra-supercritical units under scenarios of rapid load changes.

[0355] In Example 1, this invention verifies the core logic and execution flow of the proposed "coordinated coal feeding control method for small pulverizer-pulverizer based on neural network prediction and ADMM optimization" under a typical rapid load increase condition of a single coal mill with a dual-pulverizer system. This example fully demonstrates the effectiveness of the closed-loop mechanism of load prediction, dynamic matching of coal powder supply and demand, proportional scheduling optimization, and feedback correction within the control cycle. The results show that compared with the traditional fixed proportional control strategy, this method exhibits significant advantages in terms of coal output response speed, flow accuracy, and main steam pressure stability.

[0356] However, in actual engineering, ultra-supercritical units are often equipped with multiple pulverized coal silos and multiple coal mills, and face more complex operating conditions and load change scenarios. For example, some silos may have insufficient pulverized coal reserves, valve response may be sluggish, or the coal mills may be operating under non-full-condition conditions. The system has strong coupling and high uncertainty in boundary constraints. Against this background, the control mechanism verified in Example 1 still needs to be extended and verified under more complex system structures and scheduling conditions to comprehensively evaluate the distributed coordination capability and engineering robustness of the method of the present invention.

[0357] Therefore, Example 2 further designed a complex scenario involving multiple powder bins and multiple coal mills, and conducted collaborative scheduling tests under typical abnormal operating conditions to verify the practicality and promotional value of the method of the present invention in terms of group coordination and robust optimization. Next, the steps of system modeling, predictive analysis, and optimized control will be carried out sequentially.

[0358] Example 2: Group Coordinated Scheduling Control Case Study under Multiple Powder Silos and Multiple Coal Mills

[0359] This embodiment, based on the control method proposed in Embodiment 1, further extends to a multi-pulverizer-multi-coal mill collaborative control scenario with stronger system coupling. This example aims to verify the scheduling adaptability and control robustness of the present invention in dealing with complex operating conditions, differences in equipment boundary conditions, and inconsistencies in pulverized coal supply status.

[0360] (1) System parameter acquisition and setting

[0361] Consider an ultra-supercritical unit that needs to increase its capacity from 360MW to 440MW (an 80MW increase) within 10 minutes. The unit's coal supply subsystem consists of four small pulverized coal silos and two coal mills, with each silo interleaved with one of the two mills. Initially, the material level in silo 2 is below the set lower limit (only 25%), and coal mill 1 is operating at half load. The control cycle is set to 10 seconds, and the prediction window is H = 6. Table 7 shows the initial configuration and boundary constraints of the system equipment.

[0362] Table 7 Initial Configuration and Boundary Constraints of System Equipment

[0363]

[0364] The data in this table indicates that the current available coal output capacity of pulverizer B is limited, and pulverizer M1 does not have the capacity to supply pulverized coal at full load, which places higher demands on the scheduling strategy.

[0365] (2) Multivariate prediction and state coding modeling

[0366] This embodiment employs a MIMO-structured neural network to jointly predict load curves and coal consumption changes. Based on the scheduling time window in step 1, it identifies load growth rates and pulverized coal supply gaps in advance. Simultaneously, the system encodes the status of each pulverized coal silo and constructs a scoring mechanism based on margin, combining scores for remaining pulverized coal quantity, valve opening response capability, and historical stability to support adaptive weight updates in the ADMM scheduling algorithm. Table 8 shows the specific prediction data.

[0367] Table 8 Combined Forecast Data of Load and Coal Consumption

[0368]

[0369] (3) Distributed cooperative optimization scheduling modeling

[0370] Based on the predicted load growth trend and powder silo status scores from step 2, a distributed optimization scheduling model is constructed. The ADMM algorithm is employed, decomposing the problem into a local coal output ratio sub-problem and a global coordination and matching problem in each iteration, dynamically adjusting the target coal output ratio of the powder silos. Furthermore, to address the risk of coal output interruption caused by the "low material level" state of powder silo 2, the scheduling model introduces a constraint weight adjustment mechanism in the form of a penalty function, prioritizing the scheduling of powder silos with sufficient material levels and high stability for coal supply. Table 9 shows the optimized scheduling results.

[0371] Table 9 Scheduling Optimization Results

[0372]

[0373] (4) Opening degree execution mechanism and scheduling fault tolerance design

[0374] Based on the target coal output ratio obtained in step 3, the system generates corresponding valve opening control commands. Each powder silo valve has a maximum opening limit, a response rate limit (Δu ≤ 5%), and a response failure protection mechanism. When the actual coal output rate of powder silo 2 deviates from the target value due to insufficient material level, the system automatically triggers the redundancy scheduling mechanism to distribute the remaining load demand among powder silos A, C, and D. Table 10 shows a comparison between the target opening and the actual coal output flow rate. Figure 4 The timing diagram of the actual valve position execution response of powder bins 1–4 is shown.

[0375] Table 10 Comparison of Target Opening Degree and Actual Coal Output

[0376]

[0377] (5) Rolling control and model update

[0378] The control system executes load forecasting and optimized scheduling processes on a rolling basis with a 10-second cycle. The deviation learning model parameters are updated every 30 cycles (approximately 5 minutes), using weighted corrections based on the results of the last three rounds of sampling. Under conditions of significant feedback error, the system dynamically adjusts the priority score of the powder silos to prevent low-quality powder silos from interfering with overall control performance.

[0379] To demonstrate the effectiveness of this invention, three control strategies were designed for comparative experiments. The experimental results are shown in Table 11 and... Figure 5 As shown:

[0380] Option A: Fixed ratio control (does not change with state);

[0381] Option B: Manual experience-based proportional scheduling (based on manual intervention in coal quantity);

[0382] Option C: The control method of this invention (neural network + ADMM);

[0383] Table 11 Performance Comparison Results

[0384]

[0385] The comparison results show that Example 2 further verifies the applicability of the method of the present invention in a multi-pulverizer-multi-grinder system. Its collaborative scheduling strategy can effectively avoid the risks caused by abnormal operating conditions and improve the overall response speed and robustness of the system. It is significantly superior to traditional methods in terms of coal output accuracy, load ramp-up speed, and boiler pressure control, and has strong engineering implementation capabilities.

[0386] In summary, the "coordinated coal feeding control method for small pulverized coal silos and coal mills based on neural network prediction and ADMM optimization" proposed in this invention not only demonstrates advantages in rapid response and high-precision control under a single-unit dual-silo structure, but also exhibits good scalability and scheduling robustness in more complex multi-silo-multi-coal-mill scenarios. System verification through Examples 1 and 2, covering typical load variation processes under normal operation and abnormal disturbance conditions, fully demonstrates the comprehensive effectiveness of this method in improving coal output response speed, optimizing load matching, enhancing main steam pressure stability, and preventing silo cavitation. This method achieves a breakthrough in traditional fixed-ratio scheduling logic through a soft strategy approach, possessing strong engineering feasibility and industrial promotion potential, and can significantly improve the dynamic fuel supply capacity of ultra-supercritical coal-fired units in flexible adjustment and deep peak-shaving tasks.

[0387] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response, characterized in that: include: Real-time acquisition of operating parameters of the ultra-supercritical unit; based on the operating parameters of the ultra-supercritical unit, obtaining the predicted load sequence at time H in the future. Then, the future average load increase rate is calculated to determine whether it is less than the set load increase response threshold. If the average load increase rate in the future is less than the set load increase response threshold At the same time, the coal feeding control method remains unchanged; The future average load increase rate is greater than or equal to the set load increase response threshold. Simultaneously, the state variables of the small pulverized coal silo and the pulverizer are collected in real time to construct a pulverized coal supply state vector. ; Initialize the coal output ratio allocation vector of the small powder silo based on the predicted load sequence at time H in the future. and coal powder supply state vector The objective function and operational constraints are constructed, and the Alternating Direction Multiplier Method (ADMM) is used to process the objective function and operational constraints to obtain the optimal coal output ratio allocation vector. ; Based on the optimal coal output ratio allocation vector Generate a control signal for the material feeding valve. If the control signal of the feed valve does not meet the amplitude limit protection, or does not meet the soft start and stop constraint, or the state variable of the small powder silo meets the failure tolerance, return to execution: real-time acquisition of the operating parameters of the ultra-supercritical unit; Under the condition that the feed valve control signal meets the amplitude limiting protection and soft start / stop constraints, and the state variables of the small powder silo meet the failure tolerance, determine the actual coal output rate monitored in real time. and target coal output rate deviation Does it exceed the preset threshold, in the deviation If the preset threshold is not exceeded, return to execution: Real-time collection of operating parameters of the ultra-supercritical unit; In deviation When the preset threshold is exceeded, obtain State feature input vector at time step Input the state features into the vector Input the trained neural network function model to obtain the correction coefficient for the coal output ratio of the small powder silo. ; Deviation over m time intervals When the fluctuation does not exceed the preset threshold and the correction coefficient for the proportion of coal output from the small powder silo meets the correction range limit, then... The initial coal output ratio allocation vector of the small pulverized coal silo at each time step is corrected to obtain the corrected coal output ratio allocation vector of the small pulverized coal silo. The process then returns to the execution: real-time acquisition of the operating parameters of the ultra-supercritical unit; deviation within m time steps. If the fluctuation exceeds the preset threshold, or if the correction coefficient of the coal output ratio of the small powder silo does not meet the correction range limit, the correction will be suspended and an alarm mechanism will be triggered.

2. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 1, characterized in that, Real-time acquisition of operating parameters of the ultra-supercritical unit; based on the operating parameters of the ultra-supercritical unit, obtaining the predicted load sequence at time H in the future. Then, calculate the future average load increase rate, including: The operating parameters of the ultra-supercritical unit are collected in real time by the distributed control system (DCS) and a system state vector is constructed. The system state vector is represented as follows: ； in, Indicates time The system state vector, Set the load for the current dispatcher; For the actual output electrical load, Main steam pressure This refers to the coal quantity control command or the current coal feed rate. Collect system state vectors at n time points to form a historical data sequence. Historical data sequence Represented as: ； in, Indicates time The system state vector, Indicates time The system state vector; Historical data sequences Input the time series forecasting model to obtain the predicted load sequence at time H in the future. , is represented as: ； in, Indicates time Forecasted load, Indicates time Forecasted load, Indicates time Forecasted load; The time-series prediction model is obtained by training a long short-term memory network model using multiple training samples. The training samples include input samples and output samples. The input samples are data sequences, and the output samples are the actual output load sequences at future time H corresponding to the data sequences. The input samples are input into the long short-term memory network model to obtain the predicted load sequences at future time H of the data sequences. The prediction error loss function is calculated based on the actual output load sequences and the predicted load sequences. The parameters in the long short-term memory network model are modified. The parameters in the long short-term memory network model are iteratively modified through training samples to minimize the prediction error loss function, thus obtaining the trained time-series prediction model. The prediction error loss function is calculated based on the actual output load sequence and the predicted load sequence, and is achieved through the following formula: ； in, This represents the value of the error loss function. Represents the time in the actual output load sequence Actual output load, Represents the time in the predicted load sequence Forecasted load; Based on historical data sequences and predicted load sequence The future average load increase rate is calculated using the following formula: ； in, Indicates the future average load increase rate. Indicates time Forecasted load, Indicates time Actual output electrical load.

3. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 1, characterized in that, Real-time acquisition of state variables from the small pulverized coal silo and the coal mill to construct a pulverized coal supply state vector. ,include: Real-time collection of status variables from N small powder bins. The state variables of the small powder hopper i at any given time include the current material level height. Feeding rate Maximum allowable feeding rate Effective bottom area of ​​small pink bin i and coal powder density ; according to The state variable of the small pink warehouse i at time i is calculated. The current remaining available time of the Pink Warehouse i Specifically, this is achieved through the following formula: ； Where ε is a constant to prevent division by zero; Real-time acquisition of the state variables of M coal mills. The state variables of the coal mill j at time j include the current coal feed rate. Maximum allowable coal feeding rate Coal mill current and start / stop status The start / stop status indicates whether the coal mill is running; according to The state variables of the coal mill j at time j are calculated. The load rate of the coal mill j at any given time Specifically, this is achieved through the following formula: ； Construct a pulverized coal supply state vector based on the state variables of N small pulverized coal silos and the state vectors of M coal mills. , is represented as: ； in, express The current material level in the small powder silo 1 is displayed. express The current material level of the small powder silo N is displayed. express The feeding rate of the small powder hopper 1 at all times. express The feeding rate of the small powder hopper N at all times. express The load rate of coal mill 1 at all times. express The load rate of the coal mill M at any given time.

4. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 1, characterized in that, Initialize the coal output ratio allocation vector of the small powder silo based on the predicted load sequence at time H in the future. and coal powder supply state vector The objective function and operational constraints are constructed, and the Alternating Direction Multiplier Method (ADMM) is used to process the objective function and operational constraints to obtain the optimal coal output ratio allocation vector. ,include: The initialization vector for the coal output ratio of the small powder silo is expressed as: ； in, This indicates the proportion of coal output from the small powder silo per unit time. This indicates the proportion of coal output from the small powder silo per unit time. N represents the proportion of coal output per unit time in the small powder silo, and T represents the matrix transpose; in, , Let i represent the proportion of coal output per unit time in small powder silo i, and satisfy the following condition: ; Based on the predicted load sequence at time H in the future ,calculate Future control cycle predicted at any time Total coal supply required Specifically, it is calculated using the following formula: ； in, The conversion coefficient between load and coal is obtained by looking up a table, and the future control period includes the future time H. Define the vector of maximum instantaneous coal output capacity of the small coal silo. , is represented as: ,in, This indicates the maximum instantaneous coal output capacity of small powder silo 1. This indicates the maximum instantaneous coal output capacity of small coal silo 2. This represents the maximum instantaneous coal output capacity of the small coal silo N; Define the objective function objective function Represented as: ； in, This represents the error between the actual total coal output rate and the predicted demand. A measure of the balance of coal output ratio. This indicates the difference between the actual and expected load values ​​of each coal mill. For adjustable weighting parameters, where, , and Calculated using the following formula: ； ； ； ； in, This represents the average coal output ratio per unit time across all small coal silos. This indicates the load rating for optimal economic operation of the coal mill; Set operational constraints, including coal output rate constraints, pulverizer silo reserve constraints, and pulverizing capacity limits of the coal mill. The coal output rate constraint is expressed as follows: ； in, Indicates the future control period; The powder silo balance constraint is expressed as follows: ； The coal mill pulverizing capacity limitation is expressed as follows: ； in, This indicates the amount of coal the coal mill needs to supply to the small pulverizer. This represents the direct pulverized coal supply of the j-th coal mill boiler. This indicates the maximum pulverization rate of the coal mill within its future control cycle. , and Calculated using the following formula: ； ； ； in, The parameters in the mapping matrix, Indicates whether the small powder silo i is connected to the coal mill j; Introducing auxiliary variables Construct an optimized structure, represented as: ； in, Let be the objective function. The constraint functions are: the coal output ratio allocation vector in the small pulverizer silo, the state variables of the small pulverizer silo and the coal mill, and the future control period. Total coal supply required and the vector of maximum instantaneous coal output capacity of the small powder silo Under the condition that the operating constraints are met, the condition constraint function is 0, which includes the coal output ratio allocation vector of the small powder silo, the state variables of the small powder silo and the coal mill, and the future control cycle. Total coal supply required and the vector of maximum instantaneous coal output capacity of the small powder silo If the operational constraints are not met, the constraint function is +∞; Based on the Alternating Direction Multiplier Method (ADMM), the coal output ratio allocation vector, auxiliary variables, and Lagrange multipliers of the small powder bin are iteratively updated, specifically through the following formula: ； ； ； in, Let represent the coal output ratio allocation vector of the small powder bin in the (o+1)th iteration, and ρ be the penalty parameter in ADMM that controls the strength of consistency constraints. Denotes the auxiliary variable for the o-th iteration. Denotes the Lagrange multiplier in the o-th iteration. This represents the auxiliary variable for the (o+1)th iteration. This indicates projection onto the constraint set C, where the constraint set C is the union of the runtime constraints. Represents the Lagrange multiplier in the (o+1)th iteration; Based on the updated coal output ratio allocation vector of the small powder silo and auxiliary variables, calculate Then, the coal output ratio allocation vector, auxiliary variables, and Lagrange multipliers of the small powder silo are iteratively updated. After multiple iterations, in all From the values, obtain The minimum value is obtained, and the corresponding small coal silo coal output ratio allocation vector is taken as the optimal coal output ratio allocation vector. , is represented as: ； in, This indicates the optimal coal output ratio for small powder bin 1. This indicates the optimal coal output ratio for small powder bin 2. This represents the optimal coal output ratio of the small coal silo N.

5. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 1, characterized in that, Based on the optimal coal output ratio allocation vector Generate a control signal for the material feeding valve. ,include: Based on the optimal coal output ratio allocation vector Calculate the target coal output rate from the powder silo. Specifically, this is achieved through the following formula: ； DCS system will Mapped to feed valve control signal Specifically, this is achieved through the following formula: ； ； in, for and The inverse function of the nonlinear mapping, where a1, a2, a3 are nonlinear fitting parameters; Control signal for the feeding valve The actual coal output is adjusted by sending data from the DCS system to the PLCs of each small coal silo.

6. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 1, characterized in that, The amplitude limiting protection is expressed by the following formula: ； in, This indicates the minimum allowable value for the valve opening control signal. This indicates the maximum allowable value for the valve opening control signal; The soft start-stop constraint is expressed by the following formula: ； in, express The constant-time control signal for the material feeding valve. Indicates the valve opening threshold; Among them, the failure tolerance is that the state variables of the small pink warehouse do not change within a preset time.

7. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 1, characterized in that, Get State feature input vector at time step Input the state features into the vector Input the trained neural network function model to obtain the correction coefficient for the coal output ratio of the small powder silo. ,include: The feedback indicators include main steam pressure error, coal output deviation, coal mill current fluctuation, furnace oxygen content, and flame pattern. Among them, the main steam pressure error Calculated using the following formula: ； in, This represents the actual value of the main steam pressure. This indicates the set value of the main steam pressure; Among them, coal output deviation Calculated using the following formula: ； Get Optimal coal output ratio allocation vector at time 1 and calculate Target coal output rate at any time Specifically, this is achieved through the following formula: ； in, express The optimal coal output ratio for the small powder silo i at any given time. express Future control cycle predicted at any time Total coal supply required; collection Actual coal output rate at any time Calculate the coal deviation Specifically, this is achieved through the following formula: ； Get State feature input vector at time step The state feature input vector includes system state, scheduling instructions, valve control instructions, and boiler operation feedback. The system state includes pulverized coal silo level, small pulverized coal silo discharge rate, pulverizer load rate, and scheduling instructions. The optimal coal output ratio allocation vector at any given time; valve control commands are used to control valve opening; boiler operation feedback includes the actual value of main steam pressure, furnace oxygen content, and flame offset coefficient. Will Input the trained neural network function model, and obtain Correction coefficient for the proportion of coal output from the small powder silo at any given time Specifically, it includes: Will Input the first hidden layer and get the output of the first hidden layer. Specifically, it is expressed by the following formula: ； in, This represents the activation function. This represents the weight matrix of the first hidden layer. d represents the bias of the first hidden layer. h express The quantity dimension, dx represents Feature dimensions; The output of the first hidden layer The input is fed into the second hidden layer, and the output of the second hidden layer is obtained. Specifically, it is expressed by the following formula: ； in, This represents the weight matrix of the second hidden layer. Indicates the bias of the second hidden layer; Output the second hidden layer The input is fed into the attention mechanism layer, and the output of the attention mechanism layer is obtained. Specifically, it is expressed by the following formula: ； in, For attention weight vectors, For activation function, This represents the weight matrix of the attention mechanism layer. ⊙ represents the bias of the attention mechanism layer, and ⊙ represents the corresponding element-wise multiplication operation; Output of the attention mechanism layer The input is fed into the output layer to obtain the correction coefficient for the coal output ratio of the small powder silo, which is specifically expressed by the following formula: ； in, This represents the weight matrix of the output layer. This indicates the bias of the output layer.

8. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 7, characterized in that, The neural network function model was trained in the following way: By inputting the state feature vectors from historical moments into the neural network function model, the correction coefficient for the coal output ratio of the small powder silo at each historical moment is obtained. Calculate the loss function Specifically, this is achieved through the following formula: ； According to the loss function The Adam optimizer is used to update the parameters of the neural network function model, specifically through the following formula: ； in, These are the network parameters before the neural network function model is updated. Here, η represents the network parameters after the neural network function model is updated, and η is the learning rate. The network parameters of the neural network function model are updated multiple times to obtain the trained neural network function model.

9. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 1, characterized in that, right The initial coal output ratio allocation vector of the small powder silo at time t is corrected to obtain the corrected coal output ratio allocation vector of the small powder silo, which is achieved through the following formula: ； in, This is the corrected coal output ratio allocation vector for the small powder silo. for The optimal coal output ratio for the small powder silo i at any given time. for Correction coefficient for the coal output ratio of small powder silo i at any given time.

10. The method for coordinated coal feeding control between a small pulverized coal silo and a coal mill based on dynamic load response as described in claim 1, characterized in that, Correction range limitations: ； in, This is the maximum adjustment limit for a single ratio. This is a limit on the total adjustment range.

Images