A primary frequency modulation control method for a thermal power unit

Abstract

This invention belongs to the field of thermal power generation technology and discloses a primary frequency regulation control method for thermal power units. It constructs a global feature matrix through multi-dimensional high-frequency data acquisition, classifies unit operating conditions into five categories and matches them with dynamic benchmark parameters, and simultaneously identifies four types of frequency disturbances and matches them with frequency regulation strategies. For small disturbances, it controls the adjustment amplitude to avoid frequent equipment actions; for large and extreme disturbances, it maximizes the utilization of energy storage for rapid response, adapting to the changing operating conditions of deep peak shaving in thermal power units. This improves the response speed and adjustment accuracy of primary frequency regulation under different operating conditions and disturbances, meeting the requirements of the power grid assessment. It constructs an adaptive quantitative model for boiler energy storage to accurately calculate available energy storage and regulation boundaries, and designs a dual closed-loop control system for turbine and boiler coordination. On the turbine side, PID parameters are optimized to correct valve commands, while on the boiler side, commands are received synchronously to adjust coal, water, and air volumes in advance to supplement energy storage, achieving simultaneous use and replenishment of energy storage.

Description

Technical Field

[0001] This invention belongs to the field of thermal power generation technology, specifically a primary frequency regulation control method for thermal power units. Background Technology

[0002] As the main adjustable inertia and frequency-regulating power source, thermal power units rely on primary frequency regulation as the first line of defense against frequency fluctuations. The core principle is that when the grid frequency deviates from the rated value, the turbine's DEH system adjusts the valve opening according to a preset speed unequal rate, thereby changing the unit's active power output and compensating for the system's active power deficit. Currently, conventional primary frequency regulation uses fixed-parameter PID closed-loop control, implementing adjustment actions based on the static frequency difference-load correspondence. However, this approach presents the following technical problems in practical applications: Conventional primary frequency regulation control parameters are set based on the rated operating conditions of the unit. However, current thermal power units routinely participate in deep peak shaving at 40% of rated load and below. Under this condition, the boiler energy storage level is significantly reduced and the dynamic characteristics of the thermal system change significantly. Fixed parameter control strategies are prone to problems such as frequency regulation response lag, insufficient regulation accuracy, and frequent over- or under-regulation, which cannot meet the grid's requirements for the response speed and regulation accuracy of primary frequency regulation.

[0003] Conventional control strategies focus only on the rapid action of turbine-side control valves, failing to fully couple the large inertia and large lag dynamic response characteristics of the boiler side. When turbine control valves rapidly open and close to respond to frequency regulation demands, it can easily cause large fluctuations in the main steam pressure and temperature of the boiler. This not only leads to insufficient frequency regulation capability of the unit, but also exacerbates fatigue wear on core equipment such as the boiler and turbine, and may even trigger over-temperature and over-pressure protection actions, seriously affecting the safe operation of the unit.

[0004] Conventional control strategies employ uniform control logic for all frequency disturbances without differentiating the characteristics of these disturbances. For small-amplitude, high-frequency fluctuations, over-adjustment is likely to occur, leading to frequent valve operation and significantly shortening equipment lifespan. For large-amplitude, high-rate frequency drops / rises, the strategies cannot provide sufficiently fast response speed and adjustment, making it difficult to curb the risk of frequency exceeding limits. Summary of the Invention

[0005] The purpose of this invention is to provide a primary frequency regulation control method for thermal power units to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a primary frequency regulation control method for thermal power units, comprising the following specific steps: During the operating condition identification phase, key parameters from the power grid side, the unit thermal side, and the equipment safety side are collected simultaneously to construct a full-domain feature matrix of the unit's operating conditions. Clustering algorithms are used to divide the operating condition intervals and match dynamic characteristic benchmark parameters. Operating condition labels and updated benchmark parameters are output in real time. In the disturbance identification stage, based on the power grid frequency data obtained in the operating condition identification stage, multi-dimensional features of frequency disturbances are extracted and measurement noise is removed. A disturbance classification model is constructed to classify disturbance types, match corresponding frequency regulation priorities and regulation targets, and output disturbance classification results and frequency regulation demand levels. During the energy storage assessment phase, based on the operating condition identification results and real-time operating data, an adaptive quantitative model for boiler energy storage is constructed to calculate the total boiler energy storage and the energy storage capacity and regulation duration that can be safely used for frequency regulation, thereby clarifying the safe use boundary of energy storage. In the feedforward generation stage, the disturbance identification results and energy storage assessment boundary are combined to construct a primary frequency regulation adaptive feedforward control model. Differentiated feedforward strategies are matched for different types of disturbances. The amplitude and duration of the feedforward command are adaptively modified according to the energy storage level. The feedforward control command is generated and output and superimposed on the control valve. In the turbine-boiler coordination stage, based on the operating conditions and disturbance levels, a turbine-boiler coordinated adaptive feedback closed-loop control model is constructed, which is divided into two coordinated sub-loops: a turbine-side frequency regulation closed loop and a boiler-side energy storage compensation closed loop. This optimizes control parameters, corrects valve commands, and synchronously replenishes boiler energy storage. During the safety verification phase, a full-boundary safety constraint model for primary frequency regulation of the unit is constructed based on the unit's full-dimensional operating data. The safety constraint boundaries and frequency regulation command amplitude range are updated in real time. The feedforward and feedback commands are verified for safety and amplitude correction is performed, and the final frequency regulation command that meets the safety requirements is output. During the self-learning phase, a multi-dimensional evaluation system for frequency modulation effect is constructed, frequency modulation-related data is recorded and evaluation results are generated, and a learning algorithm is used to perform online parameter iterative optimization of the core control model. After passing the safety verification, the results are updated to the control system.

[0007] Preferably, the operating condition identification stage synchronously collects multi-source operating data through the unit distributed control system, the turbine DEH system, and the grid dispatch energy management system. The collected multi-source operating data covers three dimensions: grid side, unit thermal side, and unit equipment safety side. Grid side data includes real-time frequency, frequency change rate, dispatch primary frequency regulation assessment requirements, and regional grid inertia level. Unit side thermal operating data includes real-time active power, main steam pressure and temperature, reheat steam parameters, drum water level, boiler coal feed, feedwater flow, and valve opening. Unit equipment safety data includes turbine vibration, shaft displacement, bearing temperature, equipment protection settings, and AGC load commands. Based on the collected real-time data, a global feature matrix of unit operating conditions is constructed. With the real-time load rate of the unit as the core dimension, combined with three auxiliary dimensions of boiler combustion status, thermal system energy storage status, and equipment health status, the unit operating conditions are divided into five major categories of operating condition intervals through fuzzy clustering algorithm: rated operating condition, high load operating condition, medium load operating condition, deep peak shaving operating condition, and variable load transition operating condition. Corresponding dynamic characteristic benchmark parameters of the unit are matched for each category. It matches the operating condition range corresponding to the current operating data in real time, outputs the current operating condition label of the unit, and synchronously updates the unit dynamic characteristic benchmark parameters corresponding to the operating condition.

[0008] Preferably, the disturbance identification stage is based on the grid-side frequency data collected in the operating condition identification stage. It calculates multi-dimensional characteristic parameters of frequency disturbance in real time through a sliding time window, including frequency deviation amplitude, frequency change rate, disturbance duration, frequency fluctuation period, and disturbance change trend, so as to achieve full-dimensional extraction of frequency disturbance characteristics. At the same time, frequency measurement noise is removed through a filtering algorithm. Based on the extracted multi-dimensional feature parameters, a frequency disturbance classification decision tree model is constructed, classifying power grid frequency disturbances into four categories: Category I small-amplitude steady-state disturbances, Category II small-amplitude fluctuation disturbances, Category III large-amplitude step disturbances, and Category IV extreme impact disturbances. For each type of disturbance, a corresponding frequency regulation priority and adjustment target are matched. For small-amplitude steady-state disturbances, the focus is on stable unit operation, controlling the amplitude of frequency regulation actions. For small-amplitude fluctuation disturbances, the focus is on both frequency regulation accuracy and unit stability, suppressing continuous frequency fluctuations. For large-amplitude step disturbances, the focus is on frequency regulation response speed, quickly smoothing frequency deviations and preventing frequency exceedances. For extreme impact disturbances, the focus is on the output of extreme frequency regulation capabilities, maximizing the unit's frequency regulation support capability. The classification results of the current frequency disturbance and the corresponding frequency regulation demand level are output in real time.

[0009] Preferably, the energy storage assessment stage is based on the unit's real-time operating condition tags and real-time operating data output from the operating condition identification stage to construct an adaptive quantitative model for boiler energy storage. The model covers two core modules: boiler steam-water system energy storage and combustion system energy storage. It quantitatively calculates the static energy storage of the boiler steam drum, superheater, reheater and other steam-water systems, as well as the dynamic energy storage of the furnace combustion system and flue gas system. Based on the operational constraints of the unit's current operating conditions, the basic energy storage required for the boiler to maintain its stable operation is eliminated. The maximum available energy storage value and sustainable regulation duration that can be safely used for primary frequency regulation under the current operating conditions are calculated. At the same time, the upper limit of the rate of energy storage release and replenishment is quantified, and the safe use boundary of energy storage during frequency regulation is clarified. The quantitative results of the boiler's available energy storage are output in real time, and the energy storage constraint boundary of the frequency regulation amount is generated simultaneously.

[0010] Preferably, the feedforward generation stage constructs a primary frequency regulation adaptive feedforward control model based on the frequency disturbance classification results and frequency regulation demand level output by the disturbance identification stage, combined with the boiler available energy storage constraint boundary output by the energy storage assessment stage. For different levels of frequency disturbances, the corresponding feedforward control strategy is adaptively matched. For large step disturbances and extreme impact disturbances, the boiler energy storage is maximized and feedforward commands for regulating valve opening are quickly output to achieve rapid opening or closing of the turbine regulating valve. For small steady-state disturbances and small fluctuation disturbances, the feedforward compensation ratio is reduced and only small-amplitude feedforward commands are generated. Based on the quantitative results of the available energy storage of the boiler, the amplitude and duration of the feedforward command are adaptively corrected. When the available energy storage is sufficient, the feedforward compensation amplitude is increased, and when the available energy storage is low, the feedforward amplitude is adaptively reduced, while reserving space for energy storage recovery. The primary frequency regulation adaptive feedforward control command is generated and output in real time and directly superimposed on the control valve command of the turbine DEH system.

[0011] Preferably, the turbine-boiler coordination stage is based on the real-time operating condition tag of the unit output from the operating condition identification stage and the frequency regulation demand level output from the disturbance identification stage, and constructs an adaptive PID feedback closed-loop control model for turbine-boiler coordination. The model is divided into two coordinated sub-loops: a turbine-side frequency regulation closed loop and a boiler-side energy storage compensation closed loop. The turbine-side frequency regulation closed loop adaptively optimizes the proportional, integral, and derivative parameters of the PID control using a fuzzy inference algorithm based on the current operating conditions and disturbance levels. It generates a control valve feedback command based on the real-time frequency deviation and the adjustment effect of the feedforward control to correct the frequency regulation deviation caused by the feedforward adjustment. Simultaneously, based on the real-time fluctuation value of the main steam pressure of the boiler, the control valve command is dynamically corrected; the boiler-side energy storage compensation closed loop synchronously receives the feedforward control command and the turbine-side control valve action command, generates the boiler-side collaborative compensation command in advance, and adaptively adjusts the boiler coal feed rate, water feed rate, and air supply rate to replenish the boiler energy storage in advance and make up for the energy storage released by the turbine control valve action. Real-time output of turbine-side valve feedback control commands and boiler-side collaborative compensation commands are sent to the boiler combustion control loops of the DEH system and DCS system, respectively, to achieve collaborative frequency regulation control on both sides of the turbine and boiler.

[0012] Preferably, the safety verification stage is based on the unit's full-dimensional operating data and equipment protection settings collected in the operating condition identification stage to construct a full-boundary safety constraint model for the unit's primary frequency regulation. The model covers three major categories of safety constraint boundaries: thermal parameter safety boundaries, equipment mechanical safety boundaries, and grid dispatch compliance boundaries. The thermal parameter safety boundaries include main steam pressure and temperature, reheat steam temperature, and drum water level. The equipment mechanical safety boundaries include turbine vibration, shaft displacement, and bearing temperature. The grid dispatch compliance boundaries include dispatch assessment requirements such as primary frequency regulation adjustment amplitude, dead zone, and speed unequal rate. Based on real-time unit operation data, the dynamic upper and lower limits of the above three types of safety constraints are updated in real time to generate the safe amplitude range of frequency regulation commands under the current operating conditions. For Class IV extreme disturbance conditions, the short-term frequency regulation command adjustment amplitude range can be adaptively widened without triggering equipment protection actions, maximizing the release of the unit's frequency regulation capability. The feedforward control commands generated in the feedforward generation stage, the feedback control commands generated in the boiler-turbine coordination stage, and the boiler coordination compensation commands are subject to full-dimensional safety verification and dynamic amplitude limiting. If the command exceeds the safe amplitude range, the command is smoothed and limited. At the same time, the ratio of feedforward and feedback commands is adjusted to ensure that the frequency regulation command maximizes the regulation effect within the safety boundary. If the command triggers the protection setting warning, a lockout command is generated. The final frequency modulation control command, after safety verification and amplitude limiting, is output and sent to the corresponding actuator of the unit.

[0013] Preferably, the self-learning phase constructs a multi-dimensional evaluation system for the primary frequency regulation effect, based on two dimensions: grid frequency regulation effect and unit operation stability, and sets quantitative evaluation indicators such as frequency regulation accuracy, primary frequency regulation contribution rate, frequency over-limit suppression effect, maximum fluctuation range of main steam pressure, number of unit control valve actions, and number of equipment parameter over-limits. After each complete frequency regulation cycle, based on real-time operating data, multi-dimensional evaluation indicators for this frequency regulation operation are calculated to generate a comprehensive evaluation result of the frequency regulation effect. At the same time, the operating condition label, disturbance classification, control parameters, and regulation effect data of this frequency regulation are stored in the model self-learning database. Based on the historical data in the self-learning database and the evaluation results of this operation, a deep reinforcement learning algorithm is used to perform online self-learning optimization on the energy storage quantitative model in the energy storage assessment stage and the adaptive PID closed-loop model in the turbine-boiler coordination stage, and iteratively correct the core parameters and control logic of the model. The definition of a complete frequency regulation cycle is based on the grid frequency disturbance as the starting point and the frequency returning to the allowable deviation range of the rated value and stabilizing as the ending point. Specifically, it is determined that the frequency regulation cycle is completed when the grid frequency deviates from the rated value by ±0.03Hz, triggering the frequency regulation action, until the grid frequency falls back to the range of 50±0.03Hz and remains stable for 10 seconds. At the same time, core parameters such as the unit's active power output and main steam pressure return to the steady-state value within ±5% of the value before the frequency regulation action. If the equipment protection lockout is triggered during the frequency regulation process, it is also considered that the frequency regulation cycle has ended, and the indicators are calculated according to the actual regulation effect.

[0014] A safety verification mechanism is set up for model parameter iteration. After all iterative parameters are verified by offline simulation and safety boundary check, they are updated to the real-time control system, and model optimization reports are output periodically.

[0015] The comprehensive evaluation of frequency regulation effect has a full score of 100 points. Each quantitative evaluation indicator is scored according to its weight. Among them, the indicators of power grid frequency regulation effect account for 60% (frequency regulation accuracy 30%, primary frequency regulation contribution rate 20%, frequency over-limit suppression effect 10%), and the indicators of unit operation stability account for 40% (maximum fluctuation of main steam pressure 15%, number of unit valve actions 15%, number of equipment parameter over-limits 10%). Each indicator is scored from 0 to full marks based on the matching degree between the actual value and the standard value. The comprehensive score is the sum of the actual scores of each indicator. 60 points is the passing score. If the score is lower than the passing score, the frequency regulation effect is considered poor and the corresponding data is listed as invalid data.

[0016] The beneficial effects of this invention are as follows: 1. This invention constructs a global feature matrix through multi-dimensional high-frequency data acquisition, classifies unit operating conditions into five categories and matches them with dynamic benchmark parameters, and simultaneously identifies four types of frequency disturbances and matches them with frequency regulation strategies. For small disturbances, the control adjustment amplitude is adjusted to avoid frequent equipment operation, while for large and extreme disturbances, the energy storage is utilized to achieve rapid response. This invention adapts to the operating condition changes of deep peak shaving of thermal power units, improves the response speed and adjustment accuracy of frequency regulation under different operating conditions and disturbances, and meets the requirements of power grid assessment.

[0017] 2. This invention constructs an adaptive quantitative model for boiler energy storage to accurately calculate available energy storage and regulation boundaries. Simultaneously, it designs a dual closed-loop control system for boiler-turbine coordination. On the turbine side, PID parameters are optimized to correct valve commands, while on the boiler side, commands are received synchronously to adjust coal, water, and air volumes in advance to replenish energy storage, achieving simultaneous energy use and replenishment. This avoids significant fluctuations in main steam pressure and temperature caused by rapid valve action, enhances the continuous regulation capability of primary frequency control, reduces fatigue wear on core boiler and turbine equipment, and prevents protection actions triggered by abnormal thermal parameters, thus balancing frequency regulation performance and unit thermal system stability.

[0018] 3. This invention constructs a full-boundary safety constraint model covering thermal, equipment, and grid dispatch, updates the dynamic limit range in real time, verifies and corrects frequency regulation commands, adaptively relaxes the limit under extreme disturbances, and generates a blocking command when an early warning is triggered to avoid unit protection shutdown. At the same time, it builds a multi-dimensional frequency regulation effect evaluation system, combines deep reinforcement learning to iteratively optimize the core model online, and updates the parameters after offline simulation and safety verification. This ensures that the self-learning process does not affect the safe operation of the unit, enables the control model to continuously adapt to changes in operating conditions and disturbances, and provides data support for manual parameter tuning, thereby improving the safety and long-term adaptability of frequency regulation control. Attached Figure Description

[0019] Figure 1 This is a flowchart of the primary frequency regulation control of thermal power units according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] like Figure 1 As shown, this embodiment of the invention provides a primary frequency regulation control method for thermal power units, including the following specific steps: The operating condition identification phase involves synchronously acquiring multi-source operating data through the unit's distributed control system (DCS), turbine DEH system, and grid dispatch energy management system (EMS) using a high-frequency acquisition cycle of ≤10ms. This acquisition cycle can meet the rapid response requirements of primary frequency regulation. The acquired multi-source operating data covers three dimensions: grid side, unit thermal side, and unit equipment safety side. Grid side data includes real-time frequency, frequency change rate, dispatch primary frequency regulation assessment requirements, and regional grid inertia level. Unit side thermal operating data includes real-time active power, main steam pressure and temperature, reheat steam parameters, drum water level, boiler coal feed, feedwater flow, and valve opening. Unit equipment safety data includes turbine vibration, shaft displacement, bearing temperature, equipment protection settings, and AGC load commands. Multi-system data acquisition employs hardware clock synchronization calibration, using the satellite timing clock of the power grid dispatch energy management system as a reference. The synchronization deviation between the acquisition clocks of the DCS system and the DEH system and the reference clock is controlled within ±1ms, achieving precise timestamp alignment of multi-source data. The anomaly judgment criteria for acquired data are: a deviation of more than 20% between a single parameter's single acquisition value and the average of the previous three values ​​is considered a data jump; no data return for three consecutive acquisition cycles is considered data loss. When a data jump occurs, the average of the previous three values ​​is used to replace the abnormal value; when data is lost, linear interpolation is used to complete the data, ensuring the validity and continuity of the acquired data.

[0022] Both the core and auxiliary dimensions of the global feature matrix are configured with dedicated quantitative parameters. The real-time load rate of the unit in the core dimension is the only quantitative parameter, which is the ratio of the unit's current active power output to its rated active power output. The auxiliary dimensions select furnace heat load, flue gas oxygen content, and pulverized coal burnout rate as quantitative parameters for boiler combustion status. The main steam pressure deviation, steam drum water level deviation, and reheat steam temperature deviation are selected as quantitative parameters for thermal system energy storage status. The turbine vibration value, bearing temperature deviation, and shaft displacement are selected as quantitative parameters for equipment health status. The overall status of each dimension is represented by the comprehensive average value of its respective quantitative parameters.

[0023] Based on the collected real-time data, a global feature matrix of unit operating conditions is constructed. With the real-time load rate of the unit as the core dimension, combined with three auxiliary dimensions of boiler combustion status, thermal system energy storage status, and equipment health status, the unit operating conditions are divided into five major categories of operating condition intervals through fuzzy clustering algorithm: rated operating condition, high load operating condition, medium load operating condition, deep peak shaving operating condition, and variable load transition operating condition. Corresponding dynamic characteristic benchmark parameters of the unit are matched for each category. It matches the operating condition range corresponding to the current operating data in real time, outputs the current operating condition label of the unit, and synchronously updates the unit dynamic characteristic benchmark parameters corresponding to the operating condition.

[0024] When constructing the global feature matrix, the collected multi-source operating data is standardized and normalized, and parameters of different dimensions are mapped to the interval [0, 1]. The weight of the unit's real-time load rate is 40%, and the weights of the boiler combustion status, thermal system energy storage status, and equipment health status are 20%, 20%, and 20%, respectively. The fuzzy clustering algorithm adopts fuzzy C-means clustering, the iteration termination threshold of the cluster center is set to 0.001, and the maximum number of iterations is set to 100. Through this parameter setting, the accurate division of unit operating conditions and the stable output of operating condition labels are achieved.

[0025] The triggering conditions for updating the unit's dynamic characteristic reference parameters are: the unit is continuously in a certain operating condition range for ≥5 minutes, or the core thermodynamic parameters (main steam pressure, load rate) under the current operating condition fluctuate by more than 10% of the rated value for 1 minute. The reference parameters adopt an incremental update logic, which is based on the historical reference parameters under the same operating conditions and incrementally corrected by combining the average real-time operating data under the current operating conditions. The correction range does not exceed 5% of the original reference parameters. If the historical data sample size under the same operating conditions is less than 100 sets, the full update logic is adopted, and the reference parameters are recalibrated with the average real-time data under the current operating conditions.

[0026] The disturbance identification stage is based on the grid-side frequency data collected in the operating condition identification stage. It calculates multi-dimensional characteristic parameters of frequency disturbance in real time through a sliding time window, including frequency deviation amplitude, rate of change of frequency (ROCOF), disturbance duration, frequency fluctuation period, and disturbance change trend. This enables full-dimensional extraction of frequency disturbance characteristics. At the same time, a filtering algorithm is used to remove frequency measurement noise to avoid frequency modulation malfunctions caused by measurement noise interference. The sliding time window adopts a fixed step sliding mode with a sliding step size of 50ms. After each sliding, the frequency perturbation feature parameters within the window are updated and calculated, which ensures the continuity of feature extraction and avoids computational redundancy caused by too small a step size and feature lag caused by too large a step size. Based on the extracted multi-dimensional feature parameters, a frequency disturbance classification decision tree model is constructed, classifying power grid frequency disturbances into four categories: Category I small-amplitude steady-state disturbances, Category II small-amplitude fluctuation disturbances, Category III large-amplitude step disturbances, and Category IV extreme impact disturbances. A corresponding frequency regulation priority and adjustment target are matched for each type of disturbance. For small-amplitude steady-state disturbances, the focus is on stable unit operation, controlling the amplitude of frequency regulation actions to avoid frequent equipment adjustments. For small-amplitude fluctuation disturbances, the focus is on balancing frequency regulation accuracy and unit stability, suppressing continuous frequency fluctuations. For large-amplitude step disturbances, the focus is on frequency regulation response speed, quickly smoothing frequency deviations and preventing frequency exceedances. For extreme impact disturbances, the focus is on the output of extreme frequency regulation capabilities, maximizing the unit's frequency regulation support capability and ensuring power grid frequency security. The model outputs the current frequency disturbance classification results and corresponding frequency regulation demand levels in real time.

[0027] The sliding time window duration is set to 200ms, enabling real-time extraction of frequency disturbance features. The filtering algorithm employs a first-order low-pass filter with a filtering time constant of 50ms, effectively eliminating high-frequency noise during frequency measurement. The frequency disturbance classification decision tree model uses frequency deviation amplitude and frequency change rate as core decision nodes. Specifically, for Class I small-amplitude steady-state disturbances, the frequency deviation amplitude is ≤ ±0.03Hz and the frequency change rate is ≤ ±0.02Hz / s; for Class II small-amplitude fluctuation disturbances, the frequency deviation amplitude is ±0.03~±0.08Hz and the frequency change rate is ±0.02~±0.05Hz / s; for Class III large-amplitude step disturbances, the frequency deviation amplitude is ±0.08~±0.2Hz and the frequency change rate is ±0.05~±0.2Hz / s; and for Class IV extreme impact disturbances, the frequency deviation amplitude is > ±0.2Hz and the frequency change rate is > ±0.2Hz / s. Based on these thresholds, accurate identification of the four types of disturbances is achieved.

[0028] The energy storage assessment stage is based on the real-time operating condition tags and real-time operating data of the unit output from the operating condition identification stage. It constructs an adaptive quantitative model for boiler energy storage. The model covers two core modules: boiler steam-water system energy storage and combustion system energy storage. It quantitatively calculates the static energy storage of the boiler steam drum, superheater, reheater and other steam-water systems, as well as the dynamic energy storage of the furnace combustion system and flue gas system, so as to achieve full-dimensional and accurate quantitative calculation of the total boiler energy storage. The total energy storage of the boiler is the result of a weighted coupling calculation of the static energy storage of the steam-water system and the dynamic energy storage of the combustion system. The static energy storage of the steam-water system accounts for 60% of the weight, and the dynamic energy storage of the combustion system accounts for 40%. The total energy storage value is obtained by summing the two types of energy storage values ​​according to this weight. The weight ratio is set based on the boiler's thermal characteristics and the priority of frequency regulation energy storage release.

[0029] Based on the operational constraints of the unit's current operating conditions, the basic energy storage required for the boiler to maintain its stable operation is eliminated. The maximum available energy storage value and sustainable regulation duration that can be safely used for primary frequency regulation under the current operating conditions are calculated. At the same time, the upper limit of the rate of energy storage release and replenishment is quantified, and the safe use boundary of energy storage during frequency regulation is clarified. The quantitative results of the boiler's available energy storage are output in real time, and the energy storage constraint boundary of the frequency regulation amount is generated simultaneously. This ensures that primary frequency regulation can maximize the use of boiler energy storage to improve response speed, while avoiding excessive release of energy storage that leads to large fluctuations in boiler parameters.

[0030] The static energy storage of the steam-water system is quantified based on the main steam pressure, temperature, and volume of the steam-water system. The enthalpy and mass of the medium are calculated by real-time monitoring of the medium parameters of the steam-water system, thus obtaining the static energy storage value. The dynamic energy storage of the combustion system is quantified based on the furnace heat load, flue gas oxygen content, and pulverized coal combustion efficiency. The dynamic energy storage value is calculated by the real-time heat release of the furnace. The proportion of basic energy storage to be removed varies under different operating conditions. The basic energy storage proportions for rated operating conditions, high load operating conditions, medium load operating conditions, deep peak shaving operating conditions, and variable load transition operating conditions are 20%, 25%, 30%, 50%, and 40%, respectively. The upper limit of the energy storage release rate shall not exceed 10% / s of the total available energy storage, and the upper limit of the energy storage replenishment rate shall not exceed 5% / s of the total available energy storage, so as to avoid fluctuations in boiler thermal parameters caused by rapid energy release or replenishment.

[0031] When the unit's operating conditions change, the boiler's basic energy storage ratio is dynamically adjusted linearly and gradually. The transition time is set according to the type of operating condition being switched: 30 seconds for switching between rated / high / medium load conditions; 60 seconds for switching between normal load conditions and deep peak shaving / variable load transition conditions; and 40 seconds for switching between deep peak shaving conditions and variable load transition conditions. During the transition, the basic energy storage ratio transitions linearly from the original operating condition ratio to the new operating condition ratio over time. This can be calculated synchronously in real time using the energy storage value, avoiding abrupt changes in energy storage quantification results caused by operating condition switching.

[0032] The feedforward generation stage is based on the frequency disturbance classification results and frequency regulation demand level output by the disturbance identification stage, combined with the boiler available energy storage constraint boundary output by the energy storage assessment stage, to construct a primary frequency regulation adaptive feedforward control model. For different levels of frequency disturbances, the corresponding feedforward control strategy is adaptively matched. For large step disturbances and extreme impact disturbances, the boiler energy storage is maximized and feedforward commands for regulating valve opening are quickly output to realize the rapid opening or closing of the turbine regulating valve and quickly make up for the system's active power deficit. For small steady-state disturbances and small fluctuation disturbances, the proportion of feedforward compensation is reduced and only small-amplitude feedforward commands are generated to avoid unit parameter fluctuations and frequent equipment actions caused by over-adjustment. Based on the quantitative results of the available energy storage in the boiler, the amplitude and duration of the feedforward command are adaptively corrected. When the available energy storage is sufficient, the feedforward compensation amplitude is increased and the feedforward duration is extended. When the available energy storage is low, the feedforward amplitude is adaptively reduced and the feedforward duration is shortened. At the same time, the space for energy storage recovery is reserved to avoid the active power output reversal in the later stage of frequency regulation caused by excessive energy storage release. The primary frequency regulation adaptive feedforward control command is generated and output in real time, and directly superimposed on the control valve command of the turbine DEH system to improve the primary frequency regulation response speed.

[0033] The threshold for determining sufficient available energy storage in a boiler is that the available energy storage value is ≥ 60% of the total energy storage. At this time, the amplitude of the feedforward command is increased to 1.2 times the benchmark value and the duration of action is extended to 1.1 times the benchmark value. The threshold for determining insufficient available energy storage in a boiler is that the available energy storage value is < 30% of the total energy storage. At this time, the amplitude of the feedforward command is reduced to 0.5 times the benchmark value and the duration of action is shortened to 0.8 times the benchmark value. When the available energy storage value is between 30% and 60%, the amplitude of the feedforward command and the duration of action are adjusted linearly according to the proportion of available energy storage. The feedforward compensation ratio for different disturbances is set as follows: 10%~20% for Class I small steady-state disturbances, 20%~30% for Class II small fluctuation disturbances, 70%~80% for Class III large step disturbances, and 90%~100% for Class IV extreme impact disturbances.

[0034] The feedforward command reference value is calibrated differently according to the unit's operating conditions. The reference value is based on the change in the rated valve opening under the current operating conditions of the unit. The reference value for the feedforward command under the rated operating conditions is ±10% of the rated valve opening, ±8% under high load conditions, ±6% under medium load conditions, ±3% under deep peak shaving conditions, and ±4% under variable load transition conditions. The feedforward command reference value under all types of operating conditions is linearly related to the real-time load rate of the unit. For every 10% decrease in load rate, the reference value is corrected by 90% of the original calibration value to ensure that the reference value matches the actual regulation capacity of the unit.

[0035] The boiler-turbine coordination stage is based on the real-time operating condition tag of the unit output from the operating condition identification stage and the frequency regulation demand level output from the disturbance identification stage. An adaptive PID feedback closed-loop control model for boiler-turbine coordination is constructed. The model is divided into two coordinated sub-loops: a frequency regulation closed loop on the turbine side and an energy storage compensation closed loop on the boiler side. The turbine-side frequency regulation closed loop adaptively optimizes the proportional, integral, and derivative parameters of PID control using a fuzzy inference algorithm based on the current operating conditions and disturbance levels. Based on the real-time frequency deviation and the adjustment effect of feedforward control, it generates a control valve feedback command to correct the frequency regulation deviation caused by feedforward control and ensure the accuracy of frequency regulation. Simultaneously, based on the real-time fluctuation value of the boiler main steam pressure, the control command of the regulating valve is dynamically corrected to avoid a significant drop in main steam pressure caused by rapid valve action; the boiler-side energy storage compensation closed loop synchronously receives the feedforward control command and the turbine-side regulating valve action command, generates the boiler-side collaborative compensation command in advance, adaptively adjusts the boiler coal feed rate, water feed rate, and air supply volume, replenishes the boiler energy storage in advance, compensates for the energy storage released by the turbine regulating valve action, realizes the "use and replenish" of boiler energy storage, and improves the continuous regulation capability of primary frequency regulation; The boiler energy storage compensation is set with differentiated advance action amounts according to the frequency disturbance type. For Class I and Class II small disturbances, the boiler compensation command is executed 80% of the turbine control valve action amount in advance; for Class III large step disturbances, it is executed 100% of the control valve action amount in advance; and for Class IV extreme impact disturbances, it is executed 120% of the control valve action amount in advance. The differentiated advance action amount is adapted to the inertial lag characteristics of the boiler combustion system to ensure the synchronization of energy storage replenishment and release.

[0036] Real-time output of turbine-side valve feedback control commands and boiler-side collaborative compensation commands are sent to the boiler combustion control loops of the DEH system and DCS system, respectively, to achieve collaborative frequency regulation control on both sides of the turbine and boiler, taking into account both frequency regulation performance and the stability of unit thermal parameters.

[0037] The fuzzy domain of the turbine-side fuzzy inference algorithm is frequency deviation [-0.5, 0.5] Hz, frequency change rate [-0.5, 0.5] Hz / s, and PID parameter correction [-0.2, 0.2]. The fuzzy rule base contains 25 core fuzzy rules. Through fuzzification, fuzzy inference, and defuzzification, the adaptive optimization of PID proportional, integral, and derivative parameters is achieved. When the main steam pressure fluctuation is within ±0.2 MPa, the valve command is linearly corrected in the opposite direction to the pressure fluctuation value. For every 0.1 MPa decrease in pressure, the valve opening command decreases by 5%, and for every 0.1 MPa increase in pressure, the valve opening command increases by 3%. The adjustment ratio of coal, water, and air volume on the boiler side is coal: water: air = 1:1.5:2.0. For every 1% / s increase in the energy storage replenishment rate, the coal, water, and air volumes are increased by 5% simultaneously according to the above ratio, achieving precise replenishment of boiler energy storage.

[0038] The turbine control valve's overall control command adopts a synthesis logic of "feedforward command as the basis + feedback command correction + dynamic pressure compensation". The specific superposition order is as follows: first, the adaptive feedforward control command is used as the initial command for the control valve; second, the control valve feedback control command is superimposed to correct the deviation of the initial command, resulting in the corrected control valve command; finally, the corrected control valve command is dynamically compensated based on the main steam pressure fluctuation value to obtain the final overall control valve command. All commands are superimposed linearly, and the amplitude of the superimposed overall command does not exceed the control valve opening limit value under the current operating conditions.

[0039] The boiler and turbine sides adopt a timing rule of "boiler side commands are issued in advance, turbine side commands are issued in real time". The boiler side collaborative compensation command is issued to the DCS system 100ms earlier than the turbine side control valve command to compensate for the large inertial lag characteristics of the boiler combustion system. At the same time, execution delay compensation logic is set. The execution delay of the DEH system control valve command is calibrated at 50ms and the execution delay of the DCS system coal-water-air volume command is calibrated at 200ms. When the command is issued, the timing and amplitude of the command issuance are slightly corrected according to the real-time monitoring feedback of the actuator action to ensure the timing matching of the actions on both sides of the boiler and turbine.

[0040] The safety verification phase, based on the unit's full-dimensional operating data and equipment protection settings collected during the operating condition identification phase, constructs a full-boundary safety constraint model for the unit's primary frequency regulation. This model covers three main categories of safety constraint boundaries: thermal parameter safety boundaries, equipment mechanical safety boundaries, and grid dispatch compliance boundaries. Thermal parameter safety boundaries include main steam pressure and temperature, reheat steam temperature, and drum water level. Equipment mechanical safety boundaries include turbine vibration, shaft displacement, and bearing temperature. Grid dispatch compliance boundaries include dispatch assessment requirements such as primary frequency regulation amplitude, dead zone, and speed unequalization rate. Based on real-time operating data of the unit, the dynamic upper and lower limits of the above three types of safety constraints are updated in real time to generate the safe amplitude range of frequency regulation command under the current operating conditions. For Class IV extreme disturbance conditions, the short-term frequency regulation command adjustment amplitude range can be adaptively relaxed without triggering equipment protection actions, so as to maximize the release of the unit's frequency regulation capability. The feedforward control commands generated during the feedforward generation stage, the feedback control commands generated during the boiler-turbine coordination stage, and the boiler coordination compensation commands are subject to full-dimensional safety verification and dynamic limiting. If a command exceeds the safety limit range, the command is smoothed and limited. At the same time, the ratio of feedforward and feedback commands is adjusted to ensure that the frequency modulation command maximizes the regulation effect within the safety boundary. If a command triggers a protection setting warning, a lockout command is generated to prohibit further expansion of frequency modulation-related regulation actions, thereby avoiding triggering the unit's protection shutdown logic. The deviation between the frequency modulation command and the safety limit range is set with two levels of execution boundaries. When the command deviation is within 10% of the limit range, the command is smoothed by exponential smoothing. When the command deviation exceeds 10% of the limit range, the command is directly subjected to hard limiting, and the command value is corrected to the limit range boundary value. This avoids excessive limiting under small deviations from affecting the frequency modulation effect, and also prevents smoothing correction under large deviations from causing the unit parameters to exceed the limit.

[0041] The final frequency regulation control command, after safety verification and amplitude limiting, is output and sent to the corresponding actuator of the unit to realize the full-process safety closed-loop constraint of primary frequency regulation control.

[0042] Under Class IV extreme impact disturbance conditions, the short-term relaxation time of the frequency modulation command adjustment limit shall not exceed 5 seconds. The relaxation range of the thermal parameter safety boundary is ±5% of the rated value, and the relaxation range of the equipment mechanical safety boundary is ±3% of the rated value. During the relaxation process, the equipment parameters are monitored in real time, and an early warning is issued when the relaxation threshold of 80% is reached. When the frequency modulation command exceeds the safety limit range, the command is corrected by exponential smoothing method with a smoothing coefficient of 0.7 to achieve a smooth transition of the command. The adjustment rule of the feedforward and feedback ratio is as follows: feedforward:feedback = 7:3 when the command does not exceed the limit; feedforward:feedback = 5:5 when the command exceeds the limit by less than 10%; and feedforward:feedback = 3:7 when the command exceeds the limit by more than 10%. The ratio adjustment ensures the frequency modulation effect while controlling the command within the safety boundary.

[0043] The unit's primary frequency regulation early warning is divided into two levels: Level 1 and Level 2. Level 1 is triggered when equipment parameters reach 90% of the protection setpoint, and Level 2 is triggered when equipment parameters reach 95% of the protection setpoint. During Level 1, only a parameter warning is issued, without triggering frequency regulation interlocking. The feedforward and feedback ratios are adjusted simultaneously to reduce the regulation amplitude. During Level 2, a frequency regulation action interlocking command is immediately generated, locking the frequency regulation-related actions of turbine control valve regulation and boiler coal-water-air volume regulation, retaining only the unit's basic operation regulation. The logic for releasing the frequency regulation interlock is as follows: if the equipment parameters that triggered the interlock fall back to below 85% of the protection setpoint and remain stable for 30 seconds, the interlock is automatically released and normal frequency regulation control is restored. After release, the amplitude of the first frequency regulation command is executed at 50% of the reference value to avoid sudden parameter changes.

[0044] The dynamic limit range of the primary frequency regulation safety constraint of the unit adopts a dual update mechanism of "timed update + trigger update". The timed update cycle is set to 200ms. The latest operating data of the unit is synchronized according to this cycle to make basic corrections to the limit range. The trigger update condition is that the single fluctuation of the core operating parameters of the unit (main steam pressure, load rate, turbine vibration value) exceeds 5% of the rated value, or the unit operating condition changes. After triggering, the frequency regulation command verification is immediately suspended, and the limit range is recalculated and updated within 50ms. After the update is completed, the verification work resumes.

[0045] Among them, the self-learning phase constructs a multi-dimensional evaluation system for the primary frequency regulation effect. Based on two dimensions, namely the grid frequency regulation effect and the unit operation stability, it sets quantitative evaluation indicators including frequency regulation accuracy, primary frequency regulation contribution rate, frequency over-limit suppression effect, maximum fluctuation of main steam pressure, number of unit control valve actions, and number of equipment parameter over-limits. After each complete frequency regulation cycle, based on real-time operating data, multi-dimensional evaluation indicators of this frequency regulation action are calculated to generate a comprehensive evaluation result of the frequency regulation effect. At the same time, the operating condition label, disturbance classification, control parameters, and regulation effect data of this frequency regulation are stored in the model self-learning database. Based on the historical data in the self-learning database and the evaluation results of this time, a deep reinforcement learning algorithm is used to perform online self-learning optimization on the energy storage quantification model in the energy storage assessment stage and the adaptive PID closed-loop model in the turbine-boiler coordination stage. The core parameters and control logic of the model are iteratively corrected to continuously improve the model's adaptive adaptability to different operating conditions and different types of frequency disturbances of the unit. The self-learning database adopts a structured storage specification based on "unique encoding of frequency regulation actions." Each encoding corresponds to a complete set of data, including eight core dimensions: operating condition label, disturbance classification, frequency regulation demand level, energy storage conversion result, feedforward / feedback command parameters, boiler collaborative compensation parameters, safety verification result, and frequency regulation effect evaluation index. Each set of data is accompanied by a collection timestamp and operating condition characteristic label. The validity period of the data in the database is one year, and no more than 500 sets of historical data of the same operating condition and disturbance type are retained. The rules for eliminating redundant / invalid data are as follows: invalid data with a frequency regulation effect comprehensive evaluation score of less than 60 points are eliminated first, and valid data is eliminated in order of timestamp from oldest to newest, ensuring the validity and lightweight nature of the database data.

[0046] The online training of deep reinforcement learning adopts a lightweight mode of small batches and short step sizes. The online training batch after each frequency modulation cycle is set to 32 groups, and the iteration step size is set to 0.001. This parameter setting ensures that the model can complete parameter iteration optimization based on the latest frequency modulation data, while avoiding the system computing power occupation caused by large batch and large step size training, and does not affect the real-time control of the unit.

[0047] A safety verification mechanism for model parameter iteration is set up. After all iterations are verified by offline simulation and safety boundary check, the parameters are updated to the real-time control system to ensure that the self-learning process does not affect the safe operation of the unit. At the same time, model optimization reports are output regularly to provide data support for the manual tuning of the unit's primary frequency regulation parameters.

[0048] The deep reinforcement learning algorithm adopts a dual-Q network structure, which includes an input layer, a hidden layer, and an output layer. The input layer is the frequency regulation effect evaluation index, the hidden layer has two layers with 64 neurons per layer, and the output layer is the model parameter correction amount. The reward function is designed based on the grid frequency regulation effect and unit operation stability. The reward value is 1 when the frequency regulation accuracy meets the standard and the unit parameters do not fluctuate, and the reward value is -1 when the frequency regulation deviation exceeds the standard. For other states, the reward value is calculated linearly according to the index compliance rate. The specific operating conditions for offline simulation verification include five types of operating conditions: rated, high load, medium load, deep peak shaving, and variable load transition. The verification index thresholds are: frequency regulation accuracy ≥95%, maximum main steam pressure fluctuation ≤±0.3MPa, and valve action count ≤50% of conventional control. All iterative parameters must meet the verification index thresholds under all five operating conditions to pass the verification. The online optimization cycle of the model parameters is 24 hours / time after each frequency regulation action. The regularly output model optimization report includes three core contents: parameter iteration record, frequency regulation effect comparison, and operating condition adaptability analysis.

[0049] The triggering conditions for the cyclic optimization of the frequency regulation control in this invention are as follows: the comprehensive evaluation score of the single frequency regulation effect is less than 80 points, or the adjustment accuracy deviation of the same type of disturbance exceeds 5% within three consecutive frequency regulation cycles, or the unit operating conditions are switched more than twice consecutively; after any condition is triggered, the system immediately returns to the operating condition identification and disturbance identification stage, re-collects data, updates model parameters, realizes closed-loop cyclic optimization of the control method, and ensures the continuous adaptability of the frequency regulation control.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A primary frequency regulation control method for thermal power units, characterized in that, The specific steps include the following: During the operating condition identification phase, key parameters from the power grid side, the unit thermal side, and the equipment safety side are collected simultaneously to construct a full-domain feature matrix of the unit's operating conditions. Clustering algorithms are used to divide the operating condition intervals and match dynamic characteristic benchmark parameters. Operating condition labels and updated benchmark parameters are output in real time. In the disturbance identification stage, based on the power grid frequency data obtained in the operating condition identification stage, multi-dimensional features of frequency disturbances are extracted and measurement noise is removed. A disturbance classification model is constructed to classify disturbance types, match corresponding frequency regulation priorities and regulation targets, and output disturbance classification results and frequency regulation demand levels. During the energy storage assessment phase, based on the operating condition identification results and real-time operating data, an adaptive quantitative model for boiler energy storage is constructed to calculate the total boiler energy storage and the energy storage capacity and regulation duration that can be safely used for frequency regulation, thereby clarifying the safe use boundary of energy storage. In the feedforward generation stage, the disturbance identification results and energy storage assessment boundary are combined to construct a primary frequency regulation adaptive feedforward control model. Differentiated feedforward strategies are matched for different types of disturbances. The amplitude and duration of the feedforward command are adaptively modified according to the energy storage level. The feedforward control command is generated and output and superimposed on the control valve. In the turbine-boiler coordination stage, based on the operating conditions and disturbance levels, a turbine-boiler coordinated adaptive feedback closed-loop control model is constructed, which is divided into two coordinated sub-loops: a turbine-side frequency regulation closed loop and a boiler-side energy storage compensation closed loop. This optimizes control parameters, corrects valve commands, and synchronously replenishes boiler energy storage. During the safety verification phase, a full-boundary safety constraint model for primary frequency regulation of the unit is constructed based on the unit's full-dimensional operating data. The safety constraint boundaries and frequency regulation command amplitude range are updated in real time. The feedforward and feedback commands are verified for safety and amplitude correction is performed, and the final frequency regulation command that meets the safety requirements is output. During the self-learning phase, a multi-dimensional evaluation system for frequency modulation effect is constructed, frequency modulation-related data is recorded and evaluation results are generated, and a learning algorithm is used to perform online parameter iterative optimization of the core control model. After passing the safety verification, the results are updated to the control system.

2. The primary frequency regulation control method for thermal power units according to claim 1, characterized in that, The operating condition identification phase collects multi-source operating data through the unit distributed control system, the turbine DEH system, and the grid dispatch energy management system. The grid-side data includes real-time frequency, frequency change rate, and grid inertia level. The unit-side thermal operating data includes real-time active power, main steam parameters, and valve opening. The unit equipment safety data includes turbine vibration, bearing temperature, and equipment protection settings. The global feature matrix is ​​constructed with the real-time load rate of the unit as the core dimension, combined with the boiler combustion status, thermal system energy storage status, and equipment health status. The unit operating conditions are divided into five categories: rated operating conditions, high load conditions, medium load conditions, deep peak shaving conditions, and variable load transition conditions through fuzzy clustering algorithm. Each operating condition interval is matched with the corresponding unit dynamic characteristic benchmark parameters; the current operating condition interval is matched in real time and the operating condition label is output, and the benchmark parameters are updated synchronously.

3. The primary frequency regulation control method for thermal power units according to claim 1, characterized in that, The disturbance identification stage extracts core feature parameters of the frequency disturbance, including the frequency deviation amplitude, frequency change rate, and disturbance duration, through a sliding time window. After filtering out measurement noise using a filtering algorithm, the power grid frequency disturbance is divided into four categories through a frequency disturbance classification decision tree model. The four types of disturbances are: Type I small-amplitude steady-state disturbance, Type II small-amplitude fluctuation disturbance, Type III large-amplitude step disturbance, and Type IV extreme impact disturbance. Each type of disturbance corresponds to a different frequency modulation priority and adjustment target, and the disturbance classification results and frequency modulation requirement level are output in real time.

4. The primary frequency regulation control method for thermal power units according to claim 1, characterized in that, The boiler energy storage adaptive quantitative model in the energy storage assessment stage covers two core modules: boiler steam-water system energy storage and combustion system energy storage, and respectively quantifies the static energy storage of the steam-water system and the dynamic energy storage of the combustion system. Based on the current operating constraints of the unit, the basic energy storage required for the boiler to maintain its own stability is eliminated. The maximum available energy storage value that can be safely used for frequency regulation, the sustainable regulation duration, and the upper limit of energy storage release and replenishment rate are calculated. The safe use boundary of energy storage is clarified, and the quantitative results of the boiler's available energy storage and the constraint boundary of frequency regulation are output in real time.

5. The primary frequency regulation control method for thermal power units according to claim 1, characterized in that, The feedforward generation stage matches different feedforward strategies for different types of disturbances: large step disturbances and extreme impact disturbances maximize the use of boiler energy storage and quickly generate valve feedforward commands; small steady-state disturbances and small fluctuation disturbances reduce the proportion of feedforward compensation and generate small-amplitude feedforward commands. Based on the available energy storage of the boiler, the amplitude and duration of the feedforward command are adaptively adjusted, and the energy storage recovery space is reserved. The generated feedforward control command is directly superimposed on the valve control command of the turbine DEH system.

6. The primary frequency regulation control method for thermal power units according to claim 1, characterized in that, The adaptive PID feedback closed-loop control model in the boiler-turbine coordination stage consists of two sub-loops working together: the turbine-side frequency regulation closed loop optimizes the PID control parameters based on the current operating conditions and disturbance level using a fuzzy inference algorithm, generates valve feedback commands based on frequency deviation and feedforward regulation effect, corrects feedforward regulation deviation, and dynamically corrects the overall valve control command based on main steam pressure fluctuations. The boiler-side energy storage compensation closed loop receives feedforward commands and valve action commands, generates boiler collaborative compensation commands, adaptively adjusts boiler coal feed rate, water feed rate and air supply volume, and replenishes boiler energy storage in advance. The two types of commands are respectively sent to the boiler combustion control loop of the DEH system and DCS system to realize boiler-machine collaborative frequency regulation.

7. The primary frequency regulation control method for thermal power units according to claim 1, characterized in that, The full-boundary safety constraint model in the safety verification stage covers three major categories of safety constraint boundaries: thermal parameters, equipment mechanics, and grid dispatch compliance. Based on the real-time operation data of the unit, the dynamic upper and lower limits of various constraints are updated to generate the safety limit range of frequency regulation commands. Under extreme disturbance conditions, the adjustment limit can be relaxed for a short time. The system performs full-dimensional safety verification and amplitude limit correction on feedforward, feedback and boiler collaborative compensation commands. When the command exceeds the limit range, it smoothly corrects the command and adjusts the feedforward and feedback ratio. When the equipment protection setting warning is triggered, it generates frequency modulation action lockout command and finally outputs frequency modulation command that meets safety requirements to the actuator.

8. The primary frequency regulation control method for thermal power units according to claim 1, characterized in that, The multi-dimensional evaluation system in the self-learning phase sets quantitative evaluation indicators based on the grid frequency regulation effect and unit operation stability; after each frequency regulation cycle is completed, the evaluation indicators are calculated and a comprehensive evaluation result is generated, and the frequency regulation-related data is stored in the self-learning database. Using a deep reinforcement learning algorithm, based on historical data from a self-learning database and the evaluation results of this frequency modulation effect, the core parameters of the energy storage model and the adaptive PID closed-loop model are optimized online. After offline simulation and safety verification, the iterated parameters are updated to the real-time control system, and a parameter optimization report of the core control model is output periodically.

Images