Storage and supply system for enhancing flexible operation of coal power unit in pulverized coal bunker

Abstract

The invention discloses a storage and supply system for enhancing flexible operation of a coal power unit by a pulverized coal bunker, particularly relates to the field of thermal power generation, and is used for solving the problem that fuel conveying and combustion are asynchronous during rapid load adjustment of the coal power unit. Coordination control is carried out with the unified fuel demand as the center, the lag effect of pulverized coal storage and release is accurately predicted and counteracted before pulverized coal enters a boiler, and the pulverized coal conveying rhythm is closely aligned with the actual coal use demand. In this way, the oxygen amount and temperature fluctuation in the boiler combustion process are greatly reduced, and the problems that due to the fact that fuel supply is fast and slow, combustion in the boiler is not stable, and pollutant emission is increased are solved; meanwhile, the whole control method can automatically learn and adjust, frequent human intervention is not needed, operation is simpler, more convenient and safer, the capacity of a power plant for rapidly adapting to power grid load changes is remarkably improved, and the flexible advantage of pulverized coal storage and supply is fully played.

Description

Technical Field

[0001] The present invention relates to the field of thermal power generation, and more particularly to a coal powder bunker storage and supply system for enhancing the flexibility of coal-fired power units. Background Art

[0002] After a high proportion of renewable energy was connected, the dispatch center frequently issued rapid load increase commands. To improve response speed, thermal power plants installed sealed pulverized coal silos between the pulverizers and the furnace. Automatic control logic simultaneously adjusted the pulverized coal discharge from the pulverizers and the silo's discharge valves, striving for instant matching of fuel flow and air supply flow. The pulverized coal was then temporarily retained in the silo before entering the burner through the discharge valve. The air supply side then increased air flow based on the empirical air-to-pulverized coal ratio. Because the empirical model fails to account for the residence time of the pulverized coal in the silo and the inertia of the discharge, the rapid pulverization of the pulverizer is decoupled from the hysteresis on the discharge side. The oxygen content in the furnace first increased and then decreased, the flame center shifted back and forth, and the flue gas temperature formed overlapping peaks. This phenomenon is difficult to observe in conventional direct-feed pulverizing processes. It is only when the pulverized coal silo is used for flexible ramping that the dynamic misalignment characteristic is exposed.

[0003] Existing combustion control algorithms use static empirical coefficients to predict the air-to-pulverized coal ratio, ignoring the retention effect within the pulverized coal silo and the inertia of the discharge valve. When the load suddenly increases, the pulverizer rapidly produces pulverized coal, while the discharge end responds with a lag. First, a fuel shortage causes an increase in oxygen levels, followed by a sharp increase in emissions, which causes a sudden drop in oxygen levels, forming a double-peak waveform of "underfeed followed by overfeed." Although this waveform is short-lived, it is sufficient to trigger flue gas temperature fluctuations, causing a surge in nitrogen oxide emissions and exacerbating thermal fatigue of the water-cooled wall. The problem stems from a dynamic misalignment between the control model and the actual fuel flow, and does not involve the performance of the equipment hardware. Therefore, it is difficult to detect through routine inspections or resolve with simple adjustments.

[0004] In order to solve the above problems, a technical solution is now provided. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a coal powder silo to enhance the flexible operation storage and supply system of coal-fired power units. By coordinating the dynamic states of coal grinding, coal powder temporary storage and discharge in real time, coordinated control is carried out with unified fuel demand as the center, and the lag effect of coal powder storage and release is accurately predicted and offset before the coal powder enters the boiler, so that the coal powder delivery rhythm is closely aligned with the actual coal demand; in this way, the oxygen content and temperature fluctuations in the boiler combustion process are greatly reduced, avoiding the problems of unstable combustion in the furnace and increased pollution emissions caused by fast and slow fuel supply, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions: Load deconstruction module: Receives scheduling load instructions, mill shaft power, pulverized coal bin level, and furnace oxygen content, calculates the fuel demand vector, and writes it into the demand cache; Power simulation control module: Based on the fuel demand vector, polynomial fitting technology is used and slope constraints are imposed to construct the mill power increase curve, which is mapped to the torque set value and written into the coordination buffer together with the expected powder production rate; Flow timing module: This module uses the pulverized coal production rate prediction and pulverized coal silo retention distribution in the coordination buffer to generate a discharge valve opening sequence using a modified autoregressive sliding filter, synchronizing the estimated discharge flow with the fuel demand vector on the time axis. Oxygen difference adjustment module: monitors the oxygen content deviation in the furnace in real time during the discharge period, uses a small pulse modulation mechanism to correct the discharge valve opening sequence online, and writes the correction results into the combustion balance table; Learning feedback module: After the load instruction is completed, the oxygen waveform characteristics are extracted from the combustion balance table, and the buffer zone and the emission valve opening sequence are jointly coordinated for labeled learning, and the autoregressive sliding filter parameters are updated.

[0007] In a preferred embodiment, the load deconstruction module includes the following: Receive the dispatching load instruction, mill shaft power, pulverized coal silo level, and furnace oxygen content; calculate the load change rate, defined as the difference between the dispatching load instruction and the current generated power divided by the time step; and construct the system state vector, which includes the mill shaft power, pulverized coal silo level, and furnace oxygen content.

[0008] In a preferred embodiment, the load deconstruction module further includes the following: Using the dynamic programming prediction model, combined with the load change rate and the system state vector, the coal powder demand per minute in the future period is calculated to form a fuel demand vector; the fuel demand vector is written into the demand cache in time series.

[0009] In a preferred embodiment, the power control module includes the following contents: Based on the fuel demand vector, a third-order polynomial fitting technique is used to construct the mill power ramp-up curve to ensure that the mill pulverized coal output rate smoothly and efficiently matches the fuel demand. A slope constraint is imposed on the mill power ramp-up curve to limit the power change rate to the maximum allowable value to protect the equipment.

[0010] In a preferred embodiment, the power control module further includes the following: The mill power rise curve is mapped to a torque set value to control the coal mill drive motor; the expected pulverization rate is calculated based on the mill power rise curve and the calibration relationship; the mill power rise curve, torque set value and expected pulverization rate are written into the coordination buffer in a time series.

[0011] In a preferred embodiment, the flow timing module includes the following: The expected pulverized coal production rate and pulverized coal silo retention distribution data are read from the coordination buffer zone. The material level radar is used to intercept echo thresholds at different heights to divide the pulverized coal silo into three sections: upper, middle, and lower. The mass proportion of each section is calculated, and then the skewness is calculated and normalized to obtain the retention stratification rate. An autocorrelation analysis is performed on the historical valve opening timing sequence, and the second-order difference energy of the autocorrelation function is calculated to obtain the discharge inertia.

[0012] In a preferred embodiment, the flow timing module further includes the following: With the stagnation stratification rate and emission inertia as input, the gradient boosting tree model is used to calculate the output emission alignment coefficient; the emission alignment coefficient is used to correct the time constant of the autoregressive sliding filter; the estimated pulverized dust production rate is processed with the corrected autoregressive sliding filter to generate the emission valve opening sequence, thereby achieving the synchronization of the estimated emission flow and the fuel demand vector on the time axis.

[0013] In a preferred embodiment, the oxygen difference adjustment module includes the following contents: The oxygen content data of the furnace is collected every second through the oxygen sensor, and the difference between the furnace oxygen content data and the target oxygen content value is calculated to obtain the oxygen content deviation; an adjustment signal is generated based on the oxygen content deviation; the original discharge valve opening is added to the adjustment signal to calculate the corrected discharge valve opening; the corrected discharge valve opening is applied to the discharge valve actuator, and the oxygen content deviation and adjustment signal are recorded in the combustion balance table; the amplitude of the adjustment signal is limited. When the absolute value of the adjustment signal exceeds the preset maximum allowable adjustment amplitude, the adjustment signal is adjusted to the maximum allowable adjustment amplitude to prevent system instability.

[0014] In a preferred embodiment, the oxygen difference adjustment module further includes the following: The magnitude of the adjustment signal is the product of the oxygen content deviation and a pre-calibrated proportional coefficient, and the direction of the adjustment signal is opposite to the oxygen content deviation.

[0015] In a preferred embodiment, the learning back injection module includes the following contents: The time series data of furnace oxygen deviation are extracted from the combustion balance table, and the maximum oxygen deviation, oxygen fluctuation frequency and oxygen deviation duration are calculated to form the oxygen waveform feature vector; the expected powder production rate and the actual discharge valve opening in the coordination buffer are associated with the oxygen waveform feature vector to construct a training data set; a long short-term memory network model is used to train the model with the expected powder production rate and the actual discharge valve opening as input and the oxygen waveform feature vector as the label to predict the oxygen fluctuation characteristics; based on the comparison between the predicted results and the actual values, the time constant and model order of the autoregressive sliding filter are adjusted; the above steps are repeated to continuously optimize the parameters until the maximum oxygen deviation and the oxygen fluctuation frequency meet the predetermined standards.

[0016] The technical effects and advantages of the pulverized coal silo of the present invention to enhance the flexibility of coal-fired power units are as follows: The present invention coordinates the dynamic states of coal grinding, temporary storage of pulverized coal, and discharge in real time, and performs coordinated control centered on unified fuel demand. It accurately predicts and offsets the lag effect of pulverized coal storage and release before the pulverized coal enters the boiler, so that the pulverized coal delivery rhythm is closely aligned with the actual coal demand. In this way, the oxygen content and temperature fluctuations during the boiler combustion process are greatly reduced, avoiding the problems of unstable combustion in the furnace and increased pollution emissions caused by fast and slow fuel supply. At the same time, the entire control method can automatically learn and adjust without frequent human intervention, making the operation simpler and safer, significantly improving the ability of power plants to quickly adapt to changes in grid load, and giving full play to the flexible advantages of pulverized coal storage and supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a structural schematic diagram of a coal powder silo-enhanced storage and supply system for coal-fired power units according to the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] Example 1: Figure 1 The present invention provides a pulverized coal silo to enhance the flexibility of coal-fired power units and a storage and supply system, comprising: Load deconstruction module: receives the scheduling load instruction, mill shaft power, coal powder bin level and furnace oxygen content, calculates the fuel demand vector, and writes it into the demand cache.

[0020] Power simulation control module: Based on the fuel demand vector, polynomial fitting technology is used and slope constraints are imposed to construct the mill power increase curve, which is mapped to the torque set value and written into the coordination buffer together with the expected powder production rate.

[0021] Flow timing module: calls the pulverized coal production rate prediction and pulverized coal bin retention distribution in the coordination buffer, and uses the modified autoregressive sliding filter to generate the discharge valve opening sequence, so that the estimated discharge flow is synchronized with the fuel demand vector on the time axis.

[0022] Oxygen difference adjustment module: monitors the oxygen content deviation in the furnace in real time during the emission period, uses a small pulse modulation mechanism to correct the emission valve opening sequence online, and writes the correction results into the combustion balance table.

[0023] Learning feedback module: After the load instruction is completed, the oxygen waveform characteristics are extracted from the combustion balance table, and the buffer zone and the emission valve opening sequence are jointly coordinated for labeled learning, and the autoregressive sliding filter parameters are updated.

[0024] In scenarios where a high proportion of renewable energy is connected to the grid, thermal power plants need to frequently respond to rapid load increase commands issued by the dispatch center to maintain grid stability. To improve response speed, thermal power plants install sealed pulverized coal silos between the pulverizers and the furnace. Automatic control logic is used to adjust the pulverized coal discharge rate of the pulverized coal silo and the opening of the pulverized coal silo discharge valve, striving to achieve instant matching of fuel flow and air flow. However, traditional control methods do not fully consider the residence time of pulverized coal in the silo and the inertia of the discharge valve, resulting in a dynamic mismatch between fuel supply and demand when the load suddenly increases. Specifically, after the pulverized coal is quickly added to the pulverized coal silo, the discharge of the pulverized coal silo lags behind, the oxygen content in the furnace first increases and then decreases, and the flue gas temperature experiences superimposed peaks, which affects combustion efficiency and increases emission fluctuations.

[0025] This paper proposes a pulverized coal silo storage and supply system that enhances the flexibility of coal-fired power units. By coordinating the dynamic states of coal grinding, temporary coal storage, and discharge in real time, and implementing coordinated control centered on unified fuel demand, it aims to accurately predict and offset lag effects, closely aligning the pulverized coal delivery rhythm with actual demand. As the initial step in the solution, this load decomposition module is responsible for receiving key operating parameters and calculating the fuel demand vector, providing accurate basic data for subsequent steps.

[0026] The load deconstruction module's processing goal is to receive the dispatch load instruction, mill shaft power, pulverized coal bin level, and furnace oxygen content, calculate the fuel demand vector, and write it into the demand cache. The following is the specific technical logic, which is developed in stages: S1.1, data reception; First, four key operating parameters are received to ensure the accuracy and real-time nature of subsequent calculations. These parameters are the dispatching load command, mill shaft power, pulverized coal silo level, and furnace oxygen content. The dispatching load command is the target power generation power issued by the dispatching center, expressed in megawatts, reflecting the load level the thermal power plant needs to achieve over a specific period of time, for example, an increase from 300 MW to 350 MW. The mill shaft power is the real-time operating power of the mill, expressed in kilowatts, indicating the current pulverized coal production capacity of the mill. For example, 500 kilowatts corresponds to medium output. The pulverized coal silo level is the real-time filling height of the pulverized coal silo, expressed in percentage. For example, 60% indicates that the silo is 60% full, reflecting the pulverized coal storage status and release capacity. The furnace oxygen content is the real-time percentage of oxygen content in the furnace. For example, 4% indicates that combustion is within the normal range and provides feedback on the oxygen supply during the combustion process. The real-time collection of these parameters provides a comprehensive input foundation for the subsequent dynamic calculation of fuel demand, enabling the system to promptly respond to load changes and optimize fuel supply.

[0027] S1.2, calculation of load change rate; Before calculating the fuel demand vector, it is necessary to determine the load change rate to capture the dynamic trend of the dispatch load instruction. The load change rate refers to the change in the difference between the dispatch load instruction and the current power generation power per unit time, and is measured in megawatts per minute. The calculation process is as follows: first obtain the value of the dispatch load instruction, such as 350 megawatts, and then subtract the value of the current power generation power, such as 300 megawatts, to obtain a difference of 50 megawatts; then divide this difference by the time step, such as 1 minute, and finally obtain a load change rate of 50 megawatts per minute. This calculation method can reflect the rate of sudden increase or decrease in load, such as a rapid change from 300 megawatts to 350 megawatts, ensuring that the fuel demand forecast can keep up with the dynamic adjustment of the dispatch instruction, thereby achieving accurate matching of fuel supply.

[0028] S1.3, construction of system state vector; To fully describe the current operating status, the mill shaft power, pulverized coal bin level, and furnace oxygen content need to be integrated into a system state vector. The construction process is as follows: directly take the real-time value of the mill shaft power, such as 500 kilowatts; take the real-time percentage of the pulverized coal bin level, such as 60%; take the real-time percentage of the furnace oxygen content, such as 4%; then combine these three values ​​in sequence to form a state description containing three contents, such as a combined state of 500 kilowatts, 60%, and 4%. This combination fully records real-time information on the mill's pulverized coal production capacity, pulverized coal storage conditions, and combustion oxygen levels, providing a multi-dimensional basis for subsequent predictions, allowing fuel demand calculations to comprehensively consider the actual status of equipment operation and the combustion process.

[0029] S1.4, calculation of fuel demand vector; The calculation of the fuel demand vector aims to predict the amount of pulverized coal required at each point in time over a future period to match the dispatch load instructions. The calculation process is based on a dynamic programming prediction model, utilizing the load change rate and the system state vector. The specific steps are as follows: First, two coefficient matrices are obtained through historical data training: one reflects the impact of system state on fuel demand, and the other reflects the impact of load changes on fuel demand. The values ​​of the mill shaft power, pulverized coal bin level, and furnace oxygen content in the system state vector are then multiplied by the corresponding state coefficients and summed to obtain the state-affected portion. Next, the load change rate is multiplied by the corresponding load change coefficient to obtain the load-affected portion. Finally, the state-affected and load-affected portions are added together to generate the minute-by-minute pulverized coal demand for the future period. For example, the demand for the next five minutes would be 1.12 tons / minute, 1.15 tons / minute, 1.18 tons / minute, 1.20 tons / minute, and 1.22 tons / minute, respectively. This method, supported by historical data and comprehensive consideration of multiple factors, ensures that the prediction results can accurately reflect the changing trend of fuel demand under sudden load changes, thereby effectively addressing the dynamic dislocation problems caused by coal powder retention and emission inertia.

[0030] The dynamic programming prediction model is constructed as follows: first, historical data including scheduling load instructions, mill shaft power, pulverized coal bin level, furnace oxygen content, and actual pulverized coal demand are collected and divided into a training set and a validation set; then, a state transition equation is defined, in which the system state is composed of the mill shaft power, pulverized coal bin level, and furnace oxygen content, the action is determined by the load change rate, and the reward function is set as the error between the predicted pulverized coal demand and the actual demand; then, the model is trained using the training set data, and the prediction error is minimized by optimizing the parameters in the state transition equation; finally, the model performance is evaluated on the validation set to ensure prediction accuracy, and the pulverized coal demand at each future time point is iteratively calculated to generate a fuel demand vector.

[0031] S1.5, write to demand cache; The calculated fuel demand vectors are stored in a demand cache in a time series format so they can be directly accessed in subsequent steps. This storage process involves recording the predicted minute-by-minute pulverized coal demand in chronological order, for example, storing 1.12 tons per minute at the first minute, 1.15 tons per minute at the second minute, and finally 1.22 tons per minute at the fifth minute, forming an ordered demand sequence. This storage method ensures the traceability and real-time availability of fuel demand data, enabling subsequent mill power regulation and discharge valve control to be coordinated based on accurate forecast data, thereby improving the overall system's responsiveness and combustion efficiency.

[0032] The load deconstruction module has calculated the fuel demand vector based on the dispatching load instruction, mill shaft power, pulverized coal bin level and furnace oxygen content, and written it into the demand cache, providing a data basis for subsequent control. However, traditional control methods do not fully consider the residence time of pulverized coal in the bin and the inertia of the discharge valve, resulting in a dynamic mismatch between fuel supply and demand when the load suddenly increases, manifested as fluctuations in furnace oxygen content and superimposed peaks in flue gas temperature. To solve this problem, the power simulation control module needs to construct a mill power increase curve based on the fuel demand vector to ensure that the mill increases the pulverized coal output rate as needed, while avoiding equipment loss and energy efficiency reduction caused by power mutations, providing a basis for the subsequent coordinated control of the discharge valve.

[0033] The power simulation module aims to construct a mill power ramp-up curve based on the fuel demand vector, map the incremental slope constraint to a torque setpoint, and write it into the coordination buffer along with the expected pulverized coal production rate. The following is a detailed process, which is carried out in stages: S2.1, read the fuel demand vector from the demand cache; First, the fuel demand vector calculated by the load deconstruction module needs to be extracted from the demand cache. The fuel demand vector is a time series that represents the amount of pulverized coal required at each time point in the future, in tons per minute. For example, assuming that the fuel demand in the next five minutes is 1.12 tons per minute, 1.15 tons per minute, 1.18 tons per minute, 1.20 tons per minute, and 1.22 tons per minute, these values ​​will be read as an ordered sequence. When extracting, ensure that the time resolution is consistent with subsequent calculations, for example, one data point per minute. By reading this sequence, the power simulation control module obtains the input data required to construct the mill power increase curve, providing the necessary basis for subsequent power planning.

[0034] S2.2, determine the mill power increase curve based on the fuel demand vector; After obtaining the fuel demand vector, it is necessary to plan the mill power ramp-up path, namely, construct a mill power ramp-up curve, to ensure that the mill's pulverized coal output rate meets the pulverized coal quantity specified in the fuel demand vector. Three principles must be adhered to during this planning process: power changes must be smooth and gradual to avoid mechanical shock to the mill; power ramp-up must be timely so that the pulverized coal output rate reaches the target value within the specified timeframe; and the power path must be optimized to reduce unnecessary energy consumption. To achieve this, a third-order polynomial fitting technique is used to describe the mill power ramp-up curve. Specifically, the mill power ramp-up curve is defined as a function of time consisting of four components: the cubic term, the square term, the linear term, and the constant term, each controlled by four undetermined coefficients. It is assumed that there is a calibrated functional relationship between the mill pulverized coal output rate and power. For example, the pulverized coal output rate is equal to the power multiplied by a conversion coefficient plus an offset, where the conversion coefficient and offset are pre-measured constants. The goal is to ensure that the calculated pulverized coal output rate is as close as possible to the value of the fuel demand vector. To this end, an error function is defined, representing the sum of the squares of the differences between the calculated pulverized dust rate and the fuel demand vector at all time points. Through numerical optimization methods, such as gradually adjusting the coefficients to minimize the error, the optimal values ​​of the four coefficients are determined, thereby determining the mill power ramp-up curve. A smooth power curve effectively matches demand while protecting equipment and improving energy efficiency.

[0035] S2.3, impose an increasing slope constraint on the mill power increase curve; To ensure a smooth and controllable power ramp-up process within the mill power ramp-up curve, its rate of change needs to be limited to prevent excessive power changes from adversely affecting the coal mill. A maximum allowable power ramp rate, for example, 50 kilowatts per minute, is set as a constraint. The rate of change of the mill power ramp-up curve is determined by its time derivative: the squared term multiplied by three times the first coefficient, the time term multiplied by two times the second coefficient, and the sum of the three coefficients. This is then compared to the maximum allowable power ramp rate. This constraint is incorporated into the optimization process when solving the four coefficients of the mill power ramp-up curve. Specifically, while adjusting the coefficients to reduce error, the rate of change at each time point is checked to see if it exceeds the maximum allowable power ramp rate. If so, the coefficients are adjusted and refitted until the constraint is satisfied. This ensures that the mill power ramp-up curve maintains a smooth ramp over the entire time range, avoiding sudden changes. Implementing this constraint significantly reduces mechanical stress and extends equipment life.

[0036] S2.4, mapping the mill power increase curve to the torque setting value; After determining the mill power increase curve, it needs to be converted into a torque set value to control the operation of the mill drive motor. The relationship between power, torque and speed shows that power is equal to the result of torque multiplied by speed. Assuming that the mill speed is a known constant, such as the rated speed, the torque can be calculated by dividing the power by the speed. In specific implementation, for each time point on the mill power increase curve, its power value is calculated separately, and then the power value is divided by the speed constant to obtain the torque set value at the corresponding time point. By calculating at each time point, a complete sequence of torque set values ​​is generated. This sequence can be directly used for motor control to ensure that the mill operates according to the planned power curve. The precise mapping of the torque setting provides a reliable control signal for subsequent equipment operations, ensuring the actual implementation effect of the power increase.

[0037] S2.5, calculate the expected powder production rate based on the mill power increase curve; After determining the mill power ramp-up curve, it is necessary to calculate the expected pulverization rate to provide reference data for the coordinated control of the subsequent discharge valves. The calculation of the expected pulverization rate is based on the calibrated relationship between the mill's pulverization rate and power. For example, assume that the pulverization rate is equal to the power multiplied by a conversion coefficient plus an offset, where the conversion coefficient is expressed in tons per minute per kilowatt and the offset is expressed in tons per minute, both of which are pre-calibrated constants. For each time point on the mill power ramp-up curve, take its power value and substitute it into the above relationship to calculate the expected pulverization rate at the corresponding time point. By calculating at each time point, a time series of the expected pulverization rate is formed. This series reflects the expected output capacity of the mill under the planned power and provides an important basis for the fuel supply adjustment in the subsequent steps. The calculated expected pulverization rate can enhance the predictability of the system and improve the coordination of the overall control.

[0038] S2.6, write the calculation results into the coordination buffer; After completing the calculation of the mill power rise curve, torque setting value and expected powder production rate, these data need to be written into the coordination buffer in a time series so that they can be called by the subsequent flow timing module. Specifically, for each time point in the next five minutes, the mill power rise value, torque setting value and expected powder production rate at the corresponding moment are stored as a data group. For example, the data group for the first minute contains the power value, torque value and expected powder production rate for the first minute, the data group for the second minute contains the corresponding values ​​for the second minute, and so on, until the fifth minute. When storing, ensure that the data is arranged in chronological order and the units are consistent to form a structured time series. This storage method is convenient for subsequent steps to read and use, ensuring the continuity and accuracy of the control process. After the data is written to the coordination buffer, the system can seamlessly connect to subsequent control links and improve overall operating efficiency.

[0039] The power simulation control module reads the fuel demand vector from the demand buffer, uses a third-order polynomial fit, and imposes a change rate constraint to construct a mill power ramp-up curve, ensuring that the mill pulverized coal output rate smoothly and efficiently matches the fuel demand. Subsequently, the power curve is converted into a torque setpoint, and the expected pulverized coal production rate is calculated, with all results ultimately written to the coordination buffer. In scenarios where a high proportion of renewable energy is connected to the grid, this process effectively balances the mill's response speed and equipment stability, optimizes the pulverized coal silo's fuel supply capacity, and provides technical support for thermal power plants to quickly respond to load commands.

[0040] The power simulation module constructs a mill power ramp-up curve based on the fuel demand vector and writes the torque setpoint and expected pulverized coal production rate into the coordination buffer. However, traditional control methods fail to fully account for the residence time of pulverized coal in the silo and the inertia of the discharge valve. This leads to a dynamic misalignment between fuel supply and demand during sudden load increases, manifesting as fluctuating furnace oxygen levels and superimposed flue gas temperature peaks. To address this issue, the flow synchronization module utilizes data from the coordination buffer to generate a discharge valve opening sequence synchronized with the fuel demand vector, eliminating hysteresis and improving the flexible storage and supply capabilities of the pulverized coal silo.

[0041] The flow synchronization module aims to synchronize the estimated pulverized coal production rate and pulverized coal silo retention distribution in the coordination buffer with the estimated pulverized coal production rate and the pulverized coal silo retention distribution. Using a modified autoregressive sliding filter, the module generates a sequence of discharge valve openings, synchronizing the estimated discharge flow with the fuel demand vector on the time axis. The following is a detailed process, which is carried out in stages: S3.1, read the expected pulverized coal production rate and pulverized coal silo retention distribution data from the coordination buffer; Before generating the discharge valve opening sequence, key input information needs to be extracted from the coordination buffer. The coordination buffer stores the expected pulverization rate and pulverized coal bin retention distribution data. The expected pulverization rate is a time series that represents the predicted pulverization value of the pulverizer at each time point in the future, such as the predicted pulverization value per minute in the next five minutes. The pulverized coal bin retention distribution data is measured by the material level radar to measure the spatial distribution of pulverized coal in the bin, which is specifically expressed as the mass proportion of the upper, middle and lower sections, and the sum of the proportions of these three sections is 1. When reading this data, ensure that the time resolution is consistent with the subsequent calculation process, for example, record one data point per minute. By extracting the expected pulverization rate and pulverized coal bin retention distribution data, the flow synchronization module obtains the input information required to generate the discharge valve opening sequence.

[0042] S3.2, calculate the retention stratification rate to quantify the vertical accumulation tendency of the pulverized coal layer in the pulverized coal silo; To analyze the impact of coal accumulation within a pulverized coal silo on the emission time series, an indicator is needed to quantify the vertical distribution characteristics of the pulverized coal layer. Therefore, the retention stratification rate is calculated. The retention stratification rate is determined by analyzing the mass distribution at different heights within the pulverized coal silo. The specific calculation process is as follows: First, using the echo threshold interception of the material level radar at different heights, the pulverized coal layer within the silo is divided into three sections: upper, middle, and lower. The mass contribution of each section is calculated. Next, the skewness is calculated by subtracting the mass contribution of the lower section from the mass contribution of the upper section and dividing the result by the mass contribution of the middle section to reflect the difference in mass distribution between the upper and lower sections. Finally, the skewness is compared with the minimum and maximum skewness values ​​in historical data to calculate its normalized value. This is calculated by subtracting the minimum skewness from the skewness and dividing it by the difference between the maximum and minimum skewness values ​​to obtain the retention stratification rate. For example, if the mass proportions of the upper, middle, and lower sections are 0.3, 0.4, and 0.3, respectively, the skewness is 0. If the skewness in the historical data ranges from -1 to 1, the retention stratification ratio is 0.5, indicating a uniform distribution of the pulverized coal layer. The retention stratification ratio reveals the accumulation characteristics of pulverized coal within the silo, providing an accurate basis for adjusting the discharge rhythm, thereby improving control accuracy.

[0043] S3.3, calculate the discharge inertia to evaluate the hysteresis and inertia of the discharge valve operation; To quantify the response characteristics and inertia effects of the discharge valve during operation, it is necessary to analyze its dynamic behavior using historical data. Therefore, the discharge inertia is calculated. The discharge inertia is determined by analyzing the autocorrelation of the valve opening time series data. The specific calculation process is as follows: First, historical time series data of valve opening is collected and the autocorrelation function values ​​at different time delays are calculated to reflect the temporal correlation of valve opening. Then, the changes in the autocorrelation function values ​​at adjacent time delays are double-differencing. First, the difference between the autocorrelation function values ​​for adjacent delays is calculated, and then the adjacent differences between these differences are calculated again to obtain the second-order differences. Finally, the squares of all second-order differences are summed, and the resulting sum is used as an indicator of the discharge inertia. For example, by collecting historical valve opening data, calculating the autocorrelation function values ​​for delays of 1 to 10 steps, and then calculating the sum of the squares of the second-order differences, the discharge inertia is calculated to be 0.02. A smaller discharge inertia value indicates a smaller valve inertia and a faster response. In this way, the discharge inertia effectively measures the dynamic characteristics of the valve, providing data support for optimizing emission control and improving system adaptability.

[0044] S3.4, calculate the emission alignment coefficient using the gradient boosting tree model to adjust the emission sequence timing; To ensure that the discharge flow rate is aligned with the expected pulverized coal production rate, the discharge rhythm needs to be adjusted based on the pulverized coal silo status and valve characteristics. Therefore, a discharge alignment coefficient is introduced and calculated using a gradient boosting tree model. The discharge alignment coefficient is used to adjust the temporal position of the discharge valve opening sequence, and its value ranges from 0.5 to 2.0. The specific calculation process involves constructing a gradient boosting tree model using the stagnation stratification rate and discharge compliance as input features. Training on historical data predicts the optimal discharge alignment coefficient under current conditions. For example, if the input stagnation stratification rate is 0.5 and the discharge compliance is 0.02, the model outputs a discharge alignment coefficient of 1.2. When the discharge alignment coefficient is greater than 1, the time position of the discharge sequence needs to be slightly advanced to match the expected pulverized coal production rate; when it is less than 1, it indicates that the time position needs to be delayed. This calculation method allows the discharge alignment coefficient to dynamically adjust the discharge rhythm to synchronize the discharge flow rate with the pulverized coal production rate, thereby improving control flexibility and accuracy and adapting to dynamic changes under different operating conditions.

[0045] S3.5, using the emission alignment coefficient to modify the time constant of the autoregressive sliding filter; When generating the discharge valve opening sequence, a filter is required to smooth the data and optimize its response characteristics. Therefore, the discharge alignment coefficient is used to modify the time constant of the autoregressive sliding filter. The autoregressive sliding filter is used to generate a smoothed discharge valve opening sequence, and its time constant determines the filter's smoothness and response speed. The specific modification process involves first setting an original time constant, such as 1 minute; then multiplying the original time constant by the discharge alignment coefficient to obtain a modified time constant. For example, if the discharge alignment coefficient is 1.2, the modified time constant is 1.2 minutes. Increasing the modified time constant slows the filter's response, shifting the time position of the discharge sequence forward; conversely, it shifts the time position backward. This modification allows the autoregressive sliding filter to adjust the rate of change of the discharge valve opening according to actual needs, ensuring that the discharge flow rate is aligned with the expected pulverized coal production rate. This effectively offsets the time lag caused by pulverized coal retention and valve inertia, improving the real-time and stability of control.

[0046] S3.6, generating a discharge valve opening sequence using a modified autoregressive sliding filter; After completing the parameter correction of the autoregressive sliding filter, the final discharge valve opening sequence needs to be generated to accurately control the discharge flow rate from the pulverized coal silo. The specific generation process is as follows: the expected pulverized coal production rate sequence is used as input data, processed through the modified autoregressive sliding filter, and the discharge valve opening sequence is output. The filter uses an autoregressive integral moving average model, whose smoothing and prediction parameters are determined by the modified time constant. This ensures that the generated discharge valve opening sequence can keep the estimated discharge flow rate as close as possible to the expected pulverized coal production rate. For example, the expected pulverized coal production rate sequence is input to generate a discharge valve opening sequence for the next five minutes, and the temporal match between the estimated discharge flow rate and the expected pulverized coal production rate is verified. In this way, the discharge valve opening sequence can precisely control the release rhythm of the pulverized coal, avoiding the emission deviation caused by timing misalignment in traditional control methods, thereby achieving precise matching of fuel supply, improving combustion efficiency and system stability.

[0047] The flow synchronization module reads the estimated pulverized coal production rate and pulverized coal silo retention distribution data from the coordination buffer, calculates the retention stratification rate to quantify the pulverized coal accumulation characteristics, calculates the discharge inertia to evaluate the valve dynamic characteristics, uses the gradient boosting tree model to generate the discharge alignment coefficient to adjust the timing, and modifies the time constant of the autoregressive sliding filter to optimize the response, ultimately generating a discharge valve opening sequence synchronized with the estimated pulverized coal production rate. In scenarios where a high proportion of renewable energy is integrated into the grid and causes a sudden increase in load, this process can effectively eliminate timing deviations caused by pulverized coal retention and discharge valve inertia, stabilize furnace oxygen content and flue gas temperature, and improve the storage and supply flexibility of the pulverized coal silo during rapid load changes, providing technical support for the flexible operation of thermal power plants.

[0048] The load deconstruction module to the flow timing module have respectively completed the calculation of the fuel demand vector, the construction of the mill power increase curve, and the generation of the emission valve opening sequence. The estimated emission flow and the fuel demand vector are synchronized on the time axis through the modified autoregressive sliding filter, which significantly suppresses the furnace oxygen peak and flue gas temperature swing. However, due to the dynamic changes in the distribution of pulverized coal retention in the pulverized coal bin and the inertia of the emission valve action, there may still be a slight deviation between the estimated emission flow and the actual combustion demand, resulting in fluctuations in the furnace oxygen content, which in turn affects the combustion stability. The oxygen difference adjustment module aims to further eliminate this deviation through real-time monitoring and online correction methods to ensure the dynamic consistency of fuel supply and combustion demand.

[0049] The oxygen difference adjustment module aims to accurately match the fuel flow and air supply flow during pulverized coal discharge by monitoring the furnace oxygen deviation in real time and using a small pulse modulation mechanism to online correct the discharge valve opening sequence, generate the corrected valve opening, and record the correction results in the combustion balance table. The following is the specific processing process, which is carried out in stages: S4.1, real-time monitoring of furnace oxygen deviation to assess the matching degree between fuel supply and combustion demand; During pulverized coal discharge, the oxygen content in the furnace directly reflects whether the fuel-air ratio is optimal. Therefore, continuous monitoring is necessary to determine whether the fuel supply is dynamically aligned with combustion demand. The monitoring process involves collecting furnace oxygen data once per second using an oxygen sensor installed in the furnace, expressed as the percentage of oxygen in air. A target oxygen value is also pre-set based on combustion process requirements, typically between 4% and 6%, for example, 5%. To calculate the oxygen deviation, the sensor's real-time oxygen content is subtracted from the preset target oxygen value to produce a difference. For example, if the furnace oxygen content measured at a given moment is 5.3% and the target oxygen content is 5%, the oxygen deviation is 5.3% minus 5%, or 0.3%. If the measured oxygen content is 4.7%, the oxygen deviation is 4.7% minus 5%, or -0.3%. A positive difference indicates that the furnace oxygen content is above the target, suggesting a fuel shortage; a negative difference indicates that the furnace oxygen content is below the target, suggesting an oversupply. This real-time monitoring method can quickly capture the changing trend of the oxygen content in the furnace, provide an accurate basis for subsequent adjustments, and ensure that the stability of the combustion process is effectively maintained.

[0050] S4.2, using a small pulse modulation mechanism to dynamically adjust the discharge valve opening to offset oxygen deviation; Oxygen deviation during the combustion process can lead to an imbalance between fuel supply and combustion demand. Therefore, dynamic adjustment of the exhaust valve opening is necessary to promptly correct this deviation and restore fuel ratio balance. A small-amplitude pulse modulation mechanism generates an adjustment signal at regular intervals to achieve this goal. Specifically, the adjustment process involves determining an adjustment signal every second based on the real-time calculated oxygen deviation. The signal is proportional to the deviation and inversely proportional to it. The adjustment signal is calculated by multiplying the oxygen deviation by a pre-calibrated proportionality factor to obtain the adjustment amount. The proportionality factor represents the system's sensitivity to oxygen deviation, and is set to 0.15, for example, and is expressed in percentages of oxygen per unit of valve opening. After calculation, the adjustment signal is added to the original exhaust valve opening to obtain the corrected valve opening. For example, if the oxygen deviation is 0.3% and the proportionality factor is 0.15, the adjustment signal is 0.3% multiplied by 0.15, resulting in -0.045. If the original valve opening is 60%, 60% plus -0.045 equals 59.955%, approximately equal to 60%. If the oxygen deviation is -0.3%, the adjustment signal is -0.3% multiplied by 0.15, resulting in 0.045. The corrected valve opening is 60% plus 0.045, resulting in 60.045%. This method allows precise calibration of the fuel supply through small adjustments, quickly stabilizing the furnace oxygen level and reducing fluctuations during the combustion process.

[0051] S4.3, update the discharge valve opening online and record the correction data to achieve real-time control and data traceability; To ensure that the adjusted exhaust valve opening is immediately applied to the combustion system and that operational records are preserved, the calculation results must be applied in real time and the relevant data stored. The implementation process involves sending the corrected valve opening value to the exhaust valve actuator, causing the valve to operate at the new opening. Simultaneously, the oxygen deviation and adjustment signal for each adjustment are stored chronologically in a combustion balance table database. For example, if at one moment the oxygen deviation is 0.3% and the adjustment signal is -0.045, this set of data is recorded. The next moment, the oxygen deviation changes to 0.5% and the adjustment signal is -0.075, and the next set of data is recorded. The combustion balance table stores all records in a time series format, ensuring that data is arranged in the order of occurrence. This approach enables real-time control of fuel supply while providing comprehensive data support for subsequent analysis, improving the accuracy and intelligence of combustion management.

[0052] S4.4, limit the amplitude of the adjustment signal to prevent system instability; Excessive fluctuations in the adjustment signal can cause frequent emission valve actuation or system oscillation. Therefore, the amplitude of the adjustment signal needs to be constrained to maintain smooth operation. The specific limiting method is to pre-set a maximum allowable adjustment amplitude, such as 0.1, in units of valve opening. After calculating the adjustment signal, check whether its absolute value exceeds this limit. If so, adjust it to the maximum allowable amplitude. For example, if the oxygen deviation is 1% and the proportionality factor is 0.15, the adjustment signal is 1% multiplied by 0.15, resulting in -0.15. Since its absolute value, 0.15, is greater than 0.1, the adjustment signal is constrained to -0.1. If the original valve opening is 60%, the corrected valve opening is 60% plus -0.1, resulting in 59.9%. If the oxygen deviation is 0.3%, the adjustment signal is -0.045. Since its absolute value is less than 0.1, no constraint is required and -0.045 is used directly. This limiting mechanism maintains control accuracy while avoiding equipment wear or system instability caused by large adjustments, ensuring long-term combustion process reliability.

[0053] The oxygen differential adjustment module ensures a dynamic match between fuel supply and combustion demand by monitoring furnace oxygen deviations in real time, applying a small pulse modulation mechanism to adjust the emission valve opening, updating the valve opening online and recording the correction data, and limiting the adjustment signal amplitude. This process effectively eliminates oxygen deviations, stabilizes furnace oxygen levels, and suppresses flue gas temperature fluctuations and NOx emission peaks in scenarios with high renewable energy integration, providing technical support for flexible operation of thermal power plants.

[0054] The load deconstruction module to the oxygen difference adjustment module has completed the calculation of the fuel demand vector, the construction of the mill power increase curve, and the generation and online correction of the discharge valve opening sequence. This ensures the immediate matching of the fuel flow and the air flow, and alleviates the fluctuation of the furnace oxygen content caused by the pulverized coal bin retention effect and the inertia of the discharge valve. However, the long-term adaptability of the control model needs to cope with the influence of factors such as changes in coal quality and equipment aging, and continuously optimize parameters based on historical data. The learning reinjection module uses the combustion balance table data after the load instruction is completed, combined with the coordination buffer and the discharge valve opening sequence, to update the autoregressive sliding filter parameters to improve the synchronization accuracy of the next round of scheduling.

[0055] The goal of the learning re-injection module is to extract oxygen waveform features from the combustion balance table, combine them with the estimated pulverized coal production rate and emission valve opening sequence in the coordination buffer, and update the parameters of the autoregressive sliding filter through labeled learning to improve the long-term accuracy of the control model. The following is the specific processing process, which is carried out in stages: S5.1, extract oxygen waveform features to quantify the key characteristics of furnace oxygen fluctuations; After the load instruction is executed, the fluctuation characteristics of the furnace oxygen content can directly reflect the degree of match between fuel supply and combustion demand. Therefore, it is necessary to extract key features from the combustion balance table to quantify this fluctuation. The specific processing process is as follows: First, the furnace oxygen content deviation data during the execution of the load instruction is read from the combustion balance table. This data is recorded in chronological order, with one value recorded per second, and the unit is the difference between the percentage of oxygen in air and the target oxygen content; then, the entire furnace oxygen content deviation data sequence is traversed, the absolute value of all deviation values ​​is found, and the largest one is selected as the maximum oxygen content deviation, which is used to characterize the peak amplitude of the oxygen content fluctuation; next, the furnace oxygen content deviation data sequence is processed by fast Fourier transform, and the transformation results are analyzed to determine the periodic frequency with the most significant fluctuation, that is, the main frequency, which is used to characterize the periodic characteristics of the oxygen content fluctuation; finally, the total length of the time period with non-zero values ​​in the furnace oxygen content deviation data sequence is counted to characterize the duration of the oxygen content fluctuation. Taking a five-minute load instruction execution period as an example, assuming the calculated maximum oxygen deviation is 0.5%, the oxygen fluctuation frequency is 0.1 cycles per second, and the oxygen deviation duration is 180 seconds, the resulting oxygen waveform feature vector is composed of three elements: maximum oxygen deviation, oxygen fluctuation frequency, and oxygen deviation duration. The specific values ​​are 0.5%, 0.1 cycles per second, and 180 seconds. This quantification method comprehensively describes the amplitude, frequency, and persistence of furnace oxygen fluctuations through multi-dimensional feature extraction, providing accurate data support for subsequent control model optimization.

[0056] S5.2, jointly coordinate the buffer and discharge valve opening sequences to construct a training dataset; In order to use machine learning models to predict the characteristics of furnace oxygen fluctuations, it is necessary to associate the expected powder production rate and actual discharge valve opening data in the coordination buffer with the oxygen waveform characteristics to construct a structured training dataset. The specific construction process is as follows: First, extract the time series data of the expected powder production rate and actual discharge valve opening from the coordination buffer, ensuring that the time points of these data correspond one-to-one with the time points of the furnace oxygen deviation data; then, combine the expected powder production rate and actual discharge valve opening per second into an input feature vector. For example, if the expected powder production rate at a certain second is 10 kilograms per second and the actual discharge valve opening is 0.7 (indicating the opening ratio), then the input feature vector for that second consists of two values: 10 kilograms per second and 0.7; finally, the oxygen waveform feature vector during the entire load instruction execution period is used as a unified label for all input feature vectors. Taking a five-minute, 300-second load instruction execution period as an example, an input feature vector consisting of the expected pulverized powder production rate and the actual discharge valve opening is generated every second, totaling 300 samples. All samples are labeled with the oxygen content waveform feature vector, with specific values ​​of 0.5%, 0.1 cycles per second, and 180 seconds. This construction method ensures a correspondence between the input data and the oxygen content fluctuation characteristics through time alignment and feature association, enabling the machine learning model to accurately learn the impact of the expected pulverized powder production rate and actual discharge valve opening on furnace oxygen content fluctuations.

[0057] S5.3, perform labeled learning to predict oxygen waveform features; In order to predict the characteristics of the furnace oxygen waveform and optimize the control parameters through machine learning models, the input feature vector needs to be trained to generate prediction results. The specific learning process is as follows: a long short-term memory network is used as the prediction model, the input is a time series input feature vector consisting of the expected powder production rate and the actual discharge valve opening, and the output is the predicted oxygen waveform feature vector; the goal of model training is to minimize the average error between the predicted value and the actual value by adjusting the parameters, where the average error is calculated by first subtracting the maximum oxygen deviation, oxygen fluctuation frequency, and oxygen deviation duration in the predicted oxygen waveform feature vector from the actual value to obtain three differences, then square each difference and add them together, and then divide the total by three to obtain the average error. As an example, assuming the model predicts oxygen waveform feature vectors of 0.52%, 0.11 cycles per second, and 175 seconds, while the actual values ​​are 0.5%, 0.1 cycles per second, and 180 seconds, the calculation process is as follows: the difference in maximum oxygen deviation is 0.52% minus 0.5% = 0.02%, the difference in oxygen fluctuation frequency is 0.11 cycles per second minus 0.1 cycles per second = 0.01 cycles per second, and the difference in oxygen deviation duration is 175 seconds minus 180 seconds = -5 seconds. These differences are then squared to obtain 0.02% squared, 0.01 cycles per second squared, and -5 seconds squared, respectively. The sum is then divided by three to obtain the average error. This average error is gradually reduced by iteratively adjusting the model parameters. After training, the model can accurately predict the key characteristics of furnace oxygen fluctuations based on the input sequence of expected pulverized coal production rate and actual discharge valve opening. This approach fully utilizes the dynamic characteristics of time series, improving prediction accuracy and the adaptability of the control model.

[0058] S5.4, back-transfers updated filter parameters to optimize the performance of the autoregressive sliding filter; Adjusting the parameters of the autoregressive sliding filter based on the oxygen content waveform characteristics predicted by the machine learning model can improve its synchronization accuracy during the next round of load command execution. The specific adjustment process is as follows: First, compare the maximum oxygen content deviation predicted by the model with the actual value. If the predicted value is greater than the actual value, increase the filter's time constant. The specific calculation method is to multiply the current time constant by one plus 0.1 times the difference between the predicted and actual values. For example, if the predicted value is 0.52% and the actual value is 0.5%, the difference is 0.02%, and the time constant is adjusted to the current value multiplied by one plus 0.1 times 0.02%, that is, the current value multiplied by 1.002. Second, compare the predicted oxygen content fluctuation frequency with the actual value. If the predicted value is greater than the actual value, reduce the filter model order by one, but ensure that the order is not less than one. For example, if the current order is three, the predicted frequency is 0.11 cycles per second, and the actual value is 0.1 cycles per second, the new order is three minus one equals two. This adjustment method dynamically optimizes the filter parameters according to the deviation between the predicted results and the actual values, so that it can more effectively smooth the furnace oxygen content data and improve the response speed, thereby enhancing the stability of the control system under complex working conditions.

[0059] S5.5, Validation and iteration to continuously optimize the control model; After each load command is executed, the control model parameters are continuously optimized by repeating the aforementioned steps, improving overall system performance. The specific verification and iteration process is as follows: After each load command is executed, furnace oxygen deviation data is re-extracted from the combustion balance table and the oxygen waveform feature vector is calculated. A training dataset is then constructed by combining the predicted pulverized coal production rate and the actual discharge valve opening sequence in the coordination buffer. A long short-term memory network is then trained to predict the oxygen waveform features. Finally, the autoregressive sliding filter parameters are adjusted based on the predicted results. Simultaneously, the maximum oxygen deviation and oxygen fluctuation frequency are continuously monitored. When the maximum oxygen deviation gradually decreases to within 0.1% and the oxygen fluctuation frequency stabilizes below 0.05 cycles per second, control accuracy is significantly improved and system operation is more stable. This continuous optimization approach, by repeatedly leveraging historical data to improve the model, enables the control system to adapt to factors such as coal quality changes and equipment aging, thereby improving the flexibility and reliability of the pulverized coal storage and supply system and providing technical support for the rapid response of thermal power plants in scenarios with high renewable energy integration.

[0060] The above steps extract oxygen waveform features from the combustion balance table, combine them with coordination buffer and emission valve opening data for training, use a long-short-term memory network to predict fluctuation characteristics, and dynamically adjust filter parameters to achieve self-optimization of the control model. Under scenarios with frequent load fluctuations, this process effectively suppresses furnace oxygen fluctuations, reduces flue gas temperature swings, and enhances the long-term stability and response accuracy of the pulverized coal storage and supply system, providing technical support for the flexible operation of thermal power plants.

[0061] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0062] It should be noted that the system of the present invention can be deployed on the device itself to realize embedded applications, and can also be run on a PC or other terminal with a user interface, thereby meeting a variety of hardware environments and usage requirements.

[0063] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

[0064] It should be noted that, in this document, if there are relational terms such as first and second, etc., they are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0065] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A pulverized coal silo to enhance the flexibility of coal-fired power units storage and supply system, characterized by: Including steps: Load deconstruction module: Receives scheduling load instructions, mill shaft power, pulverized coal bin level, and furnace oxygen content, calculates the fuel demand vector, and writes it into the demand cache; Power simulation control module: Based on the fuel demand vector, polynomial fitting technology is used and slope constraints are imposed to construct the mill power increase curve, which is mapped to the torque set value and written into the coordination buffer together with the expected powder production rate; Flow timing module: This module uses the pulverized coal production rate prediction and pulverized coal silo retention distribution in the coordination buffer to generate a discharge valve opening sequence using a modified autoregressive sliding filter, synchronizing the estimated discharge flow with the fuel demand vector on the time axis. Oxygen difference adjustment module: monitors the oxygen content deviation in the furnace in real time during the discharge period, uses a small pulse modulation mechanism to correct the discharge valve opening sequence online, and writes the correction results into the combustion balance table; Learning feedback module: After the load instruction is completed, the oxygen waveform characteristics are extracted from the combustion balance table, and the buffer zone and the emission valve opening sequence are jointly coordinated for labeled learning, and the autoregressive sliding filter parameters are updated.

2. A pulverized coal silo enhanced coal-fired power unit flexibility operation storage and supply system according to claim 1, characterized in that: The load deconstruction module includes the following: Receive the dispatching load instruction, mill shaft power, pulverized coal silo level, and furnace oxygen content; calculate the load change rate, defined as the difference between the dispatching load instruction and the current generated power divided by the time step; and construct the system state vector, which includes the mill shaft power, pulverized coal silo level, and furnace oxygen content.

3. A pulverized coal silo enhanced coal-fired power unit flexibility operation storage and supply system according to claim 2, characterized in that: The load deconstruction module also includes the following: Using the dynamic programming prediction model, combined with the load change rate and the system state vector, the coal powder demand per minute in the future period is calculated to form a fuel demand vector; the fuel demand vector is written into the demand cache in time series.

4. A pulverized coal silo enhanced coal-fired power unit flexibility operation storage and supply system according to claim 3, characterized in that: The power control module includes the following: Based on the fuel demand vector, a third-order polynomial fitting technique is used to construct the mill power ramp-up curve to ensure that the mill pulverized coal output rate smoothly and efficiently matches the fuel demand. A slope constraint is imposed on the mill power ramp-up curve to limit the power change rate to the maximum allowable value to protect the equipment.

5. A pulverized coal silo enhanced coal-fired power unit flexibility operation storage and supply system according to claim 4, characterized in that: The power control module also includes the following: The mill power rise curve is mapped to a torque set value to control the coal mill drive motor; the expected pulverization rate is calculated based on the mill power rise curve and the calibration relationship; the mill power rise curve, torque set value and expected pulverization rate are written into the coordination buffer in a time series.

6. A pulverized coal silo enhanced coal-fired power unit flexibility operation storage and supply system according to claim 5, characterized in that: The traffic timing module includes the following: The expected pulverized coal production rate and pulverized coal silo retention distribution data are read from the coordination buffer zone. The material level radar is used to intercept echo thresholds at different heights to divide the pulverized coal silo into three sections: upper, middle, and lower. The mass proportion of each section is calculated, and then the skewness is calculated and normalized to obtain the retention stratification rate. An autocorrelation analysis is performed on the historical valve opening timing sequence, and the second-order difference energy of the autocorrelation function is calculated to obtain the discharge inertia.

7. A pulverized coal silo enhanced coal-fired power unit flexibility operation storage and supply system according to claim 6, characterized in that: The traffic timing module also includes the following: Taking the stagnation stratification rate and emission inertia as input, the gradient boosting tree model is used to calculate the output emission alignment coefficient; The emission alignment coefficient is used to correct the time constant of the autoregressive sliding filter. The estimated pulverized powder production rate is processed with the corrected autoregressive sliding filter to generate the emission valve opening sequence, thus achieving synchronization between the estimated emission flow and the fuel demand vector on the time axis.

8. A pulverized coal silo-enhanced coal-fired power unit flexibility storage and supply system according to claim 7, characterized in that: The oxygen difference adjustment module includes the following: The oxygen sensor collects furnace oxygen data every second, calculates the difference between the furnace oxygen data and the target oxygen value to obtain the oxygen deviation; and generates an adjustment signal based on the oxygen deviation. The original discharge valve opening is added to the adjustment signal to calculate the corrected discharge valve opening; the corrected discharge valve opening is applied to the discharge valve actuator, and the oxygen deviation and adjustment signal are recorded in the combustion balance table; the amplitude of the adjustment signal is limited. When the absolute value of the adjustment signal exceeds the preset maximum allowable adjustment amplitude, the adjustment signal is adjusted to the maximum allowable adjustment amplitude to prevent system instability.

9. A pulverized coal silo-enhanced coal-fired power unit flexible operation storage and supply system according to claim 8, characterized in that: The oxygen difference adjustment module also includes the following: The magnitude of the adjustment signal is the product of the oxygen content deviation and a pre-calibrated proportional coefficient, and the direction of the adjustment signal is opposite to the oxygen content deviation.

10. A pulverized coal silo-enhanced coal-fired power unit flexibility storage and supply system according to claim 8, characterized in that: The learning back injection module includes the following: The time series data of furnace oxygen deviation are extracted from the combustion balance table, and the maximum oxygen deviation, oxygen fluctuation frequency and oxygen deviation duration are calculated to form the oxygen waveform feature vector; the expected powder production rate and the actual discharge valve opening in the coordination buffer are associated with the oxygen waveform feature vector to construct a training data set; a long short-term memory network model is used to train the model with the expected powder production rate and the actual discharge valve opening as input and the oxygen waveform feature vector as the label to predict the oxygen fluctuation characteristics; based on the comparison between the predicted results and the actual values, the time constant and model order of the autoregressive sliding filter are adjusted; the above steps are repeated to continuously optimize the parameters until the maximum oxygen deviation and the oxygen fluctuation frequency meet the predetermined standards.

Images