AGC signal intelligent correction and control method, device, equipment and medium

Abstract

The invention provides an AGC signal intelligent correction and control method, device, equipment and medium, and the method comprises the steps: collecting an AGC instruction signal of a power grid dispatching side, an actual power feedback signal of a power plant side and a unit operation state parameter, and constructing multi-source heterogeneous data; inputting the multi-source heterogeneous data into a pre-trained digital twinborn model to obtain a predicted value of the real power of the unit; calculating a real-time deviation between the predicted value of the real power and the actual power feedback signal, and dynamically compensating the AGC instruction signal based on the real-time deviation to generate a corrected AGC instruction signal; and on the basis of the corrected AGC instruction signal, a feedforward control signal is generated, and is superposed to a basic control instruction of a unit distributed control system (DCS) in an increment form, so that full-load section accurate control is realized, and the problems of accurate correction of the AGC signal, accurate prediction of real power and dynamic compensation are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power systems and their automation, and particularly relates to an AGC signal intelligent correction and control method, device, equipment and medium. BACKGROUND

[0002] With the year-by-year expansion of new energy power generation (wind power, photovoltaic) and the scale of external power into the province, the power source structure of the power grid has changed significantly. The traditional thermal power unit gradually changes from the "main power supply" to the "basic power supply for ensuring the stability of the power grid and providing peak regulation and frequency modulation auxiliary services". In response to the frequency imbalance problem caused by new energy output fluctuation and external power failure, the power grid puts forward more stringent requirements on the AGC (automatic generation control) performance of thermal power units through policies such as "two rules". It clearly assesses indicators such as K1 (response speed), K2 (regulation accuracy), K3 (input rate), and Kp (comprehensive performance). If the indicators do not meet the standards, the monthly assessment will be carried out according to the standard of unit rated capacity x percentage of shortage x 1 point / kW, which directly affects the economic benefits of the power plant. At the same time, the unit needs to consider its own safe and stable operation under the premise of meeting the demand of the power grid for rapid response. The contradiction between the two becomes the core challenge of the grid control of thermal power units.

[0003] The current AGC control of thermal power units generally has the technical pain points of "insufficient signal accuracy" and "poor prediction model adaptability": on the one hand, the AGC command signal is derived from the EMS system of the power grid dispatching side, and the actual power feedback signal is derived from the DCS system of the power plant side. The two have different sources of data reference and collection links, resulting in inherent deviation between the AGC regulation quantity (command signal) and the regulated quantity (feedback signal). The existing control scheme does not systematically correct this deviation, which directly affects the regulation accuracy. On the other hand, traditional power prediction relies on linear models or single parameter fitting, and does not fully integrate the coupling relationship of unit operating state parameters (such as main steam pressure, coal supply, load change rate, etc.). It cannot accurately capture the dynamic characteristics of the unit under all operating conditions, resulting in a large deviation between the true power prediction value and the actual feedback value. It is difficult to support accurate adjustment of AGC commands, further exacerbating the contradiction between "power grid response speed" and "unit operation stability".

[0004] The above technical difficulties directly lead to three problems: first, the AGC performance indicators frequently do not meet the standards, and the power plant needs to bear high examination fees, while missing the AGC auxiliary service compensation (such as the "compensation fee and Kp value linearly related" bonus in the later policy) due to excellent indicators; second, blindly adjusting the control parameters for the purpose of response speed, which easily leads to large fluctuations in key parameters such as main steam pressure and coal supply, increases the risk of equipment wear and tear, and even triggers protection actions; third, the cumulative adjustment deviation leads to a lag in the response of the unit to the load demand of the power grid, which cannot effectively compensate for the frequency fluctuations caused by new energy consumption, affecting the stability of the grid and source, and is contrary to the industry goal of "building a new power system and ensuring energy security".

[0005] Therefore, an AGC signal intelligent correction and control method is urgently needed to solve the technical problems of accurate correction of AGC signals, accurate prediction of real power, and dynamic compensation. SUMMARY

[0006] To overcome the problems in the related art, the present disclosure provides an AGC signal intelligent correction and control method, device, equipment and medium to solve the technical problems of accurate correction of AGC signals, accurate prediction of real power, and dynamic compensation in the related art.

[0007] One or more embodiments of the present specification provide an AGC signal intelligent correction and control method, comprising the following steps: Collecting AGC instruction signals on the grid dispatching side and actual power feedback signals and group operating state parameters on the power plant side, and constructing multi-source heterogeneous data; Inputting the multi-source heterogeneous data into a pre-trained digital twin model to obtain a predicted value of the real power of the unit; Calculating the real-time deviation between the predicted value of the real power and the actual power feedback signal, dynamically compensating the AGC instruction signal based on the real-time deviation, and generating a corrected AGC instruction signal; Based on the corrected AGC instruction signal, a feedforward control signal is generated and added to the basic control instruction of the unit distributed control system (DCS) in incremental form to achieve accurate control in the full load range.

[0008] Preferably, the digital twin model adopts an architecture combining long short-term memory network (LSTM) and Transformer attention mechanism, captures the time sequence features of the input parameters through long short-term memory network (LSTM), and dynamically weights the influence weight of different features in the input parameters on power output through the attention mechanism.

[0009] Preferably, the dynamic compensation of the AGC instruction signal based on the real-time deviation specifically includes the following steps: The incremental PID algorithm is used for dynamic compensation, and the output of the incremental PID algorithm is the instruction compensation amount AL, which satisfies:

[0010] wherein e(t) is the real-time deviation, a is a proportional coefficient, and β is an integral coefficient; The proportional coefficient a and the integral coefficient β are dynamically self-adaptively adjusted by a reinforcement learning agent according to the current load change rate of the unit and the main steam pressure deviation; The AGC instruction signal is superimposed with the instruction compensation amount to generate a corrected AGC instruction signal, and the calculation formula of the corrected AGC instruction signal is: L_corrected = L_agc + AL; wherein L_agc is the AGC instruction signal.

[0011] Preferably, the feedforward control signal is generated based on the corrected AGC instruction signal, and specifically includes the following steps: A multivariable feedforward model is constructed based on the corrected AGC instruction signal and the signal change rate to respectively output a turbine main control feedforward amount and a boiler main control feedforward amount; wherein when the turbine main control feedforward amount is generated, a negative feedback of the main steam pressure deviation is introduced for correction; wherein when the boiler main control feedforward amount is generated, a quadrant control strategy is adopted to call different nonlinear functions for calculation according to the load rising direction and the current load interval.

[0012] Preferably, the key parameters in the multivariable feedforward model are optimized online by a deep deterministic policy gradient (DDPG) reinforcement learning algorithm; The reward function R of the DDPG algorithm is designed as: R = w 1 *(1 / |ΔP|) + w 2 *(1 / |ΔP_mainSteam|) - w3*(|d Coal / d t |) ; wherein ΔP is a power deviation, ΔP_mainSteam is a main steam pressure deviation, d Coal / d t is a coal supply rate, w1 , w2 , w3 is a weight coefficient.

[0013] One or more embodiments of the present specification provide an AGC signal intelligent correction and control device, comprising: The signal acquisition module is configured to acquire an AGC instruction signal on the power grid dispatching side, an actual power feedback signal on the power plant side, and group operation state parameters, and construct multi-source heterogeneous data. The prediction module is configured to input the multi-source heterogeneous data into a pre-trained digital twin model to obtain a predicted value of real power of the unit. The deviation calculation module is configured to calculate a real-time deviation between the predicted value of the real power and the actual power feedback signal. The signal correction module is configured to dynamically compensate the AGC instruction signal based on the real-time deviation to generate a corrected AGC instruction signal. The control module is configured to generate a feedforward control signal based on the corrected AGC instruction signal, and superimpose the feedforward control signal in an incremental form to a basic control instruction of a unit decentralized control system (DCS) to achieve precise control in the full load range.

[0014] Preferably, the digital twin model adopts an architecture combining a long short-term memory (LSTM) network and a Transformer attention mechanism, the LSTM network is used to capture time sequence features of input parameters, and the attention mechanism is used to dynamically weight the influence weight of different features in the input parameters on power output.

[0015] Preferably, the signal correction module is specifically configured to: The signal correction module is specifically configured to:

[0016] wherein e(t) is the real-time deviation, and α and β are a proportional coefficient and an integral coefficient, respectively. The proportional coefficient α and the integral coefficient β are dynamically self-adaptively adjusted by a reinforcement learning agent according to a current load change rate of the unit and a main steam pressure deviation. The AGC instruction signal and the instruction compensation amount are superimposed to generate a corrected AGC instruction signal, and a calculation formula of the corrected AGC instruction signal is as follows: L_corrected=L_agc+ΔL; wherein L_agc is the AGC instruction signal.

[0017] One or more embodiments of the present specification provide a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the AGC signal intelligent correction and control method as described above when executing the computer program.

[0018] This specification provides one or more embodiments of a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the AGC signal intelligent correction and control method described above.

[0019] This disclosure provides an intelligent correction and control method, device, equipment, and medium for AGC signals. Its advantages lie in its ability to construct multi-source heterogeneous data by collecting AGC command signals from the power grid dispatching side, actual power feedback signals from the power plant side, and group operating status parameters. This enables comprehensive capture and real-time acquisition of core control data, providing accurate and complete raw data support for subsequent correction, prediction, and control processes. This ensures the reliability and timeliness of data sources throughout the AGC control link, avoiding control deviations caused by data loss or delays. The multi-source heterogeneous data is input into a pre-trained digital twin model to obtain a predicted value of the unit's actual power. This fully utilizes the model's ability to accurately replicate the unit's operating characteristics, achieving high-precision prediction of the unit's actual power. This predicted value accurately reflects the power potential under actual operating conditions, overcoming the limitations of single-parameter prediction and providing a scientific and reliable benchmark for subsequent real-time deviation calculations. The real-time deviation between the predicted value of the actual power and the actual power feedback signal is calculated, accurately capturing the difference between the unit's actual power and the ideal state during AGC command execution. This enables immediate feedback of deviation information, ensuring accurate control of the power generation process. The rapid perception of unit operation fluctuations provides a direct and precise control target for the dynamic compensation process, avoiding the problem of untimely control response caused by the lag in deviation perception. Based on the real-time deviation, dynamic compensation is performed on the AGC command signal to generate a corrected AGC command signal, achieving adaptive optimization of the AGC command. This allows for flexible adjustment of command parameters according to the unit's real-time operating deviation, correcting the mismatch between the original AGC command and the actual operating state of the unit, improving the targeting and executability of the AGC command, and effectively reducing the inherent deviation between the command and actual execution, providing an optimized command foundation for precise control. Based on the corrected AGC command signal, a feedforward control signal is generated and incrementally superimposed on the basic control command of the unit's distributed control system (DCS), achieving precise control across the entire load range. This avoids significant modifications to the existing distributed control system and achieves synergistic linkage between feedforward control and the original basic control, significantly improving the unit's response speed and tracking accuracy to AGC commands, reducing fluctuations during power regulation, optimizing AGC regulation quality, and ensuring stable and economical operation of the unit while meeting grid dispatch requirements. Attached Figure Description

[0020] In order to make one or more embodiments of the present specification or the prior art clearer, the drawings needed to be used in the embodiment or prior art description will be briefly described below. Obviously, the drawings in the following description are only some embodiments of the present specification, and those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0021] Figure 1 A flowchart of an AGC signal intelligent correction and control method provided by one or more embodiments of the present specification; Figure 2 A structural diagram of an AGC signal intelligent correction and control device provided by one or more embodiments of the present specification; Figure 3 A structural diagram of a computer device provided by one or more embodiments of the present specification. DETAILED DESCRIPTION

[0022] In order to make one or more embodiments of the present specification or the prior art clearer, the drawings needed to be used in the embodiment or prior art description will be briefly described below. Obviously, the drawings in the following description are only some embodiments of the present specification, and those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0023] The present application will be described in detail below with reference to specific embodiments and drawings.

[0024] Method embodiments According to the embodiment of the present application, an AGC signal intelligent correction and control method is provided, and an AGC advanced control system integrating intelligent perception, accurate diagnosis, collaborative decision and safe execution is constructed. The core idea is to create an "intelligent driving system" for the generator set which can learn and optimize itself. First, a multi-source heterogeneous data fusion platform is used to collect data from different sources such as power grid dispatch (RTU), unit control system (DCS), primary frequency modulation device, etc. at high frequency and high reliability, and to clean, align and manage the data to provide high-quality, unified time-stamped "data fuel" for the upper intelligent application. On this basis, the system core uses digital twin technology to construct a virtual mirror of AGC signal, and to diagnose and correct the deviation between the instruction and the feedback in real time, thus fundamentally solving the core pain point of "different sources of signals". The corrected accurate instruction is then sent to an intelligent feedforward controller based on reinforcement learning, which can simulate or even surpass the experience of experienced operation experts to dynamically generate the optimal control strategy for coordinating the actions of the boiler and the steam turbine. The whole process is protected by a security fault-tolerant system, which ensures that any single component failure will not affect the original safe operation of the unit. The four links form a complete intelligent closed loop from data to decision to action. Figure 1 As shown in FIG. 1, a flowchart of the AGC signal intelligent correction and control method provided by the embodiment is shown. The AGC signal intelligent correction and control method according to the embodiment of the present application comprises the following steps: S110, a light edge computing node deployed at the power plant side is used to connect RTU, DCS, primary frequency modulation device and other systems through hardwiring and multi-protocol communication to collect AGC instruction signals from the power grid dispatch side, actual power feedback signals from the power plant side, and unit operating state parameters such as load, main steam pressure, load rate, load span, etc. to construct multi-source heterogeneous data.

[0025] S120, after the multi-source heterogeneous data such as AGC instruction signals, actual power feedback signals and unit operating state parameters are preprocessed by cleaning, fusion and storage, etc., the federal learning mechanism is used for alignment and fusion to realize efficient storage and query of high-frequency data and topological relationship. The fused data is input into the pre-trained digital twin model to obtain the predicted value of the actual power of the unit. The digital twin model adopts the architecture combining long short-term memory network LSTM and Transformer attention mechanism. The long short-term memory network LSTM is used to capture time series features of input parameters such as load change trend and pressure fluctuation time sequence. The Transformer attention mechanism is used to dynamically weight the influence weight of different features in the input parameters on the power output, for example, the influence weight of the main steam pressure deviation on the power is dynamically adjusted.

[0026] Based on the long-term operation history data of the unit, through big data analysis and iterative training of model parameters, the model can accurately map the unit operation characteristics to realize high-precision prediction of real power.

[0027] In S130, an incremental signal compensation algorithm is used to calculate the real-time deviation between the predicted value of real power and the actual power feedback signal, and the AGC instruction signal is dynamically compensated based on the real-time deviation to generate a corrected AGC instruction signal, which is the same as the signal of the dispatch EMS system.

[0028] In S140, the corrected AGC instruction signal is used to generate a feedforward control signal, which is added to the basic control instruction of the unit decentralized control system (DCS) in an incremental form to realize accurate control in the full load range. Meanwhile, a fault switching loop is added to the DCS system side. When the AGC instruction signal intelligent correction device fails or abnormally, a switching signal is sent to the DCS through hardwiring. The DCS forces the instruction compensation amount ΔL to zero and automatically switches to the original coordinated control loop, without affecting the normal operation of the unit.

[0029] The method provided by the embodiment realizes comprehensive capture and real-time acquisition of core control data by collecting AGC instruction signals on the power grid dispatching side, actual power feedback signals on the power plant side and group operation state parameters, and constructing multi-source heterogeneous data, provides accurate and complete original data support for subsequent correction, prediction and control links, ensures the reliability and timeliness of the data source of the entire AGC control link, and avoids control deviation caused by data loss or delay; the multi-source heterogeneous data is input into the pre-trained digital twin model to obtain a predicted value of the real power of the unit, the precise copying ability of the model for the operation characteristics of the unit is fully utilized, high-precision prediction of the real power of the unit is realized, the predicted value can accurately reflect the power potential under the actual operation state of the unit, and the limitation of single parameter prediction is overcome, thereby providing a scientific and reliable benchmark basis for subsequent real-time deviation calculation; the real-time deviation between the predicted value of the real power and the actual power feedback signal is calculated, the difference between the actual power of the unit and the ideal state in the AGC instruction execution process is accurately captured, immediate feedback of the deviation information is realized, the rapid perception of the operation fluctuation of the unit is ensured, a direct and accurate regulation target is provided for the dynamic compensation link, and the problem of untimely control response caused by lagging deviation perception is avoided; the AGC instruction signal is dynamically compensated based on the real-time deviation to generate a corrected AGC instruction signal, adaptive optimization of the AGC instruction is realized, the instruction parameters can be flexibly adjusted according to the real-time operation deviation of the unit, the mismatch between the original AGC instruction and the actual operation state of the unit is corrected, the pertinence and executability of the AGC instruction are improved, the inherent deviation between the instruction and the actual execution is effectively reduced, and an optimized instruction basis is provided for accurate control; based on the corrected AGC instruction signal, a feedforward control signal is generated, which is added to the basic control instruction of the unit distributed control system DCS in the form of an increment, accurate control in the full load section is realized, the existing distributed control system is not greatly modified, the feedforward control and the original basic control are cooperatively linked, the response speed and tracking accuracy of the unit to the AGC instruction are significantly improved, the fluctuation amplitude in the power regulation process is reduced, the AGC regulation quality is optimized, and stable and economic operation of the unit is realized while meeting the requirements of the power grid dispatching.

[0030] In one embodiment, the AGC instruction signal is dynamically compensated based on the real-time deviation, and the steps include the following: The real-time deviation e (t) is obtained by comparing the real power predicted value output by the digital twin model with the actual power feedback signal on the power plant side, and the dynamic compensation is performed by using an incremental PID algorithm, and the output of the incremental PID algorithm is an instruction compensation amount ΔL, which satisfies:

[0031] Wherein, e(t) is the real-time deviation, and a is a proportional coefficient, and β is an integral coefficient.

[0032] The proportional coefficient a and the integral coefficient β are dynamically self-adaptively adjusted by the reinforcement learning agent according to the current load change rate of the unit and the main steam pressure deviation, so as to realize accurate compensation under different working conditions.

[0033] The AGC instruction signal is superimposed with the instruction compensation amount to generate a corrected AGC instruction signal, and the calculation formula of the corrected AGC instruction signal is: L_corrected=L_agc+ΔL; Wherein, L_agc is the AGC instruction signal, and the corrected instruction is ensured to be consistent with the dispatch EMS system signal.

[0034] The method provided in the embodiment adopts the incremental PID algorithm of proportional and integral terms to calculate the instruction compensation amount, considers calculation simplification and response real-time performance, can quickly offset the current deviation and eliminate the accumulated deviation, reduces the hysteresis and overshoot; meanwhile, the reinforcement learning agent dynamically adjusts the proportional coefficient a and the integral coefficient β according to the load change rate of the unit and the main steam pressure deviation, solves the working condition adaptability problem of the fixed parameters of the traditional PID, enhances the adaptive ability and anti-interference performance of the compensation, and avoids overcompensation or insufficient compensation; finally, the corrected instruction is generated by superimposing the original AGC instruction with the dynamic compensation amount, the unit operation characteristics are accurately fitted, the inherent deviation between the instruction and the actual response is reduced, and the tracking accuracy and the operation stability of the unit to the AGC dispatch requirement are significantly improved.

[0035] In one embodiment, a feedforward control signal is generated based on the corrected AGC instruction signal, specifically including the following steps: Based on the corrected AGC instruction signal and the signal change rate, a multivariable feedforward model is constructed to respectively output a turbine main control feedforward amount and a boiler main control feedforward amount.

[0036] Wherein, when the turbine main control feedforward amount is generated, a negative feedback of the main steam pressure deviation is introduced to correct the loop, the main steam pressure stability is balanced while the AGC performance index is pursued, and part of the load is sacrificed through dynamic adjustment of the feedforward amount when necessary to ensure the safety of the unit.

[0037] Wherein, when the boiler main control feedforward amount is generated, a quadrant control strategy is adopted, different nonlinear functions are called for calculation according to the load rising or falling direction and the current load interval, and the coal feeding logic is optimized to reduce the unit limited output condition and improve the variable load capacity.

[0038] The turbine main control feedforward control signal and the boiler main control feedforward control signal are superimposed in the form of increment to the unit DCS basic control instruction, so as to realize the optimization and upgrading of the coordinated control loop, and ensure that the unit responds to the demand of the power grid quickly, accurately and stably.

[0039] The method provided by the embodiment converts the corrected AGC instruction and the change rate into steam turbine and boiler main control feedforward quantities through a multivariable feedforward model; the steam turbine feedforward introduces a main steam pressure deviation negative feedback to suppress fluctuation, and the boiler feedforward adopts quadrant control to improve working condition adaptability and precision, thereby finally accelerating AGC instruction response, optimizing power tracking precision, reducing parameter fluctuation, and ensuring economic and stable operation of the unit.

[0040] In one embodiment, key parameters in the multivariable feedforward model are optimized online by a deep deterministic policy gradient (DDPG) reinforcement learning algorithm. The reward function R of the DDPG algorithm is designed as: R = w 1 *(1 / |ΔP|) + w 2 *(1 / |ΔP_mainSteam|) - w3*(|d Coal / d t |) ; Wherein, ΔP is a power deviation, ΔP_mainSteam is a main steam pressure deviation, d Coal / d t is a coal supply change rate, w1 , w2 , w3 is a weight coefficient.

[0041] ΔP: the power deviation, the smaller the reward is higher.

[0042] ΔP_mainSteam: the main steam pressure deviation, the smaller the reward is higher.

[0043] d Coal / d t : the coal supply change rate, the smaller indicates that the combustion is more stable.

[0044] If any single adjustment of K1 / K2 / K3 does not meet the standard, a large negative reward is given.

[0045] Through continuous interaction with the DCS environment, the agent learns to find the optimal balance point between ensuring safety and stability and pursuing rapid response.

[0046] The method provided by the embodiment optimizes key parameters of the multivariable feedforward model online through the DDPG reinforcement learning algorithm, and combines a reward function taking the power deviation, the main steam pressure deviation, and the coal supply change rate as cores to realize self-adaptive working condition dynamic adjustment, thereby improving power tracking precision and main steam pressure stability, suppressing severe coal supply fluctuation, reducing energy consumption and equipment loss, and optimizing AGC regulation quality and operation economy of the unit.

[0047] Device embodiment According to the embodiment of the application, an AGC signal intelligent correction and control device is provided, as shown inFigure 2 Fig. 2 is a structural schematic diagram of an AGC signal intelligent correction and control device provided by the embodiment, and the AGC signal intelligent correction and control device according to the embodiment of the application comprises: A signal acquisition module 21 is configured to acquire an AGC instruction signal on the power grid dispatching side, an actual power feedback signal on the power plant side, and a group operation state parameter, and construct multi-source heterogeneous data.

[0048] A prediction module 22 is configured to input the multi-source heterogeneous data into a pre-trained digital twin model to obtain a predicted value of real power of a unit, wherein the digital twin model adopts an architecture combining a long short-term memory network LSTM and a Transformer attention mechanism, the long short-term memory network LSTM is used to capture time sequence features of input parameters, and the attention mechanism is used to dynamically weight influence weights of different features in the input parameters on power output.

[0049] A signal correction module 23 is configured to calculate a real-time deviation between the predicted value of the real power and the actual power feedback signal, dynamically compensate the AGC instruction signal based on the real-time deviation, and generate a corrected AGC instruction signal.

[0050] A control module 24 is configured to generate a feedforward control signal based on the corrected AGC instruction signal, and superimpose the feedforward control signal on a basic control instruction of a unit decentralized control system DCS in an incremental form to realize precise control in a full load section.

[0051] The device provided by the embodiment collects the AGC instruction signal of the power grid dispatching side, the actual power feedback signal of the power plant side and the group operation state parameter through the signal acquisition module 21, constructs multi-source heterogeneous data, realizes comprehensive capture and real-time acquisition of core control data, provides accurate and complete original data support for subsequent correction, prediction and control links, ensures the reliability and timeliness of the data source of the entire AGC control link, and avoids control deviation caused by data loss or delay; the multi-source heterogeneous data input into the pre-trained digital twin model in the prediction module 22 obtains the predicted value of the real power of the unit, fully utilizes the accurate copying ability of the model to the operation characteristics of the unit, realizes high-precision prediction of the real power of the unit, and the predicted value can accurately reflect the power potential under the actual operation state of the unit, overcoming the limitations of single parameter prediction, and providing a scientific and reliable benchmark basis for subsequent real-time deviation calculation; the signal correction module 23 calculates the real-time deviation between the predicted value of the real power and the actual power feedback signal, accurately captures the difference between the actual power of the unit and the ideal state in the AGC instruction execution process, realizes instant feedback of the deviation information, ensures the rapid perception of the unit operation fluctuation, provides a direct and accurate regulation target for the dynamic compensation link, avoids the problem of untimely control response caused by deviation perception lag, dynamically compensates the AGC instruction signal based on the real-time deviation, generates the corrected AGC instruction signal, realizes adaptive optimization of the AGC instruction, can flexibly adjust the instruction parameters according to the real-time operation deviation of the unit, corrects the mismatch between the original AGC instruction and the actual operation state of the unit, improves the pertinence and executability of the AGC instruction, effectively reduces the inherent deviation between the instruction and the actual execution, and provides an optimized instruction basis for accurate control; the control module 24 generates a feedforward control signal based on the corrected AGC instruction signal, and adds the feedforward control signal to the basic control instruction of the unit distributed control system DCS in the form of increment, realizes accurate control in the full load section, avoids large-scale modification of the existing distributed control system, realizes the cooperative linkage of the feedforward control and the original basic control, significantly improves the response speed and tracking accuracy of the unit to the AGC instruction, reduces the fluctuation amplitude in the power regulation process, optimizes the AGC regulation quality, and ensures stable and economic operation of the unit while meeting the requirements of the power grid dispatching.

[0052] In one embodiment, the signal correction module 23 is specifically configured to: The incremental PID algorithm is used for dynamic compensation, and the output of the incremental PID algorithm is an instruction compensation amount ΔL, which satisfies:

[0053] Wherein, e(t) is the real-time deviation, and a is a proportional coefficient and β is an integral coefficient.

[0054] The proportional coefficient alpha and the integral coefficient beta are dynamically self-adaptively adjusted by the reinforcement learning agent according to the current load change rate of the unit and the main steam pressure deviation.

[0055] The calculation formula of the corrected AGC instruction signal is: L_corrected=L_agc+DeltaL Wherein, L_agc is the AGC instruction signal.

[0056] The device provided in the embodiment adopts the incremental PID algorithm of proportional and integral terms to calculate the instruction compensation amount, considers calculation simplification and response real-time performance, can quickly offset the current deviation and eliminate the accumulated deviation, reduces the hysteresis and overshoot, dynamically adjusts the proportional coefficient alpha and the integral coefficient beta by the reinforcement learning agent according to the unit load change rate and the main steam pressure deviation, solves the working condition adaptability problem of the fixed parameters of the traditional PID, enhances the adaptive ability and anti-interference performance of the compensation, avoids overcompensation or insufficient compensation, finally generates the corrected instruction by superimposing the dynamic compensation amount on the original AGC instruction, accurately fits the unit operation characteristics, reduces the inherent deviation between the instruction and the actual response, and significantly improves the tracking accuracy and the operation stability of the unit to the AGC scheduling requirements.

[0057] The device embodiment of the embodiment of the application corresponds to the method embodiment described above, and the specific operations of the modules can be understood with reference to the description of the method embodiment, which will not be repeated here.

[0058] As shown in Figure 3 The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the AGC signal intelligent correction and control method in the above embodiment, or the computer program is executed by the processor to implement the AGC signal intelligent correction and control method in the above embodiment.

[0059] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium used in each embodiment provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0060] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the device or system embodiment, since it is basically similar to the method embodiment, it is described more simply, and the relevant part can be referred to the part of the method embodiment. The above-described device and system embodiments are only illustrative, and the units described as separate components can be or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0061] It should be pointed out finally that the above embodiments are only used to illustrate the technical solutions of the present application, but not to limit the same; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can still be modified, or some or all of the technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and the contents not described in detail in the specification of the present application belong to the common knowledge of those skilled in the art.

Claims

1. An AGC signal intelligent correction and control method, characterized in that, The method comprises the following steps: Collecting AGC instruction signals on the power grid dispatching side, actual power feedback signals on the power plant side, and group operating state parameters, and constructing multi-source heterogeneous data; Inputting the multi-source heterogeneous data into a pre-trained digital twin model to obtain a predicted value of real power of the unit; Calculating a real-time deviation between the predicted value of the real power and the actual power feedback signal, dynamically compensating the AGC instruction signal based on the real-time deviation, and generating a corrected AGC instruction signal; Based on the corrected AGC instruction signal, a feedforward control signal is generated, which is superimposed on the basic control instruction of the unit decentralized control system (DCS) in an incremental form to realize precise control in the full load range.

2. The AGC signal intelligent correction and control method of claim 1, wherein, The digital twin model adopts an architecture combining a long short-term memory network (LSTM) and a Transformer attention mechanism, captures time sequence features of input parameters through the LSTM, and dynamically weights the influence weight of different features in the input parameters on power output through the attention mechanism.

3. The method of claim 1, wherein the AGC signal intelligent correction and control method is characterized by, The dynamic compensation of the AGC instruction signal based on the real-time deviation comprises the following steps: An incremental PID algorithm is used for dynamic compensation, and the output of the incremental PID algorithm is an instruction compensation amount ΔL, which satisfies: Wherein, e(t) is the real-time deviation, and α is a proportional coefficient, and β is an integral coefficient; The proportional coefficient α and the integral coefficient β are dynamically self-adaptively adjusted by a reinforcement learning agent according to the current load change rate of the unit and the main steam pressure deviation; The AGC instruction signal and the instruction compensation amount are superimposed to generate a corrected AGC instruction signal, and the calculation formula of the corrected AGC instruction signal is: L_corrected=L_agc+ΔL; Wherein, L_agc is the AGC instruction signal.

4. The method of AGC signal intelligent correction and control of claim 1, wherein, The feedforward control signal is generated based on the corrected AGC instruction signal, which comprises the following steps: Based on the corrected AGC instruction signal and the signal change rate, a multivariate feedforward model is constructed to respectively output a turbine main control feedforward amount and a boiler main control feedforward amount; Wherein, when the turbine main control feedforward amount is generated, a negative feedback of the main steam pressure deviation is introduced for correction; Wherein, when the boiler main control feedforward amount is generated, a quadrant control strategy is adopted, and different nonlinear functions are called for calculation according to the load rising direction and the current load interval.

5. The method of AGC signal intelligent correction and control of claim 4, wherein, The key parameters in the multivariate feedforward model are optimized online by a deep deterministic policy gradient (DDPG) reinforcement learning algorithm; The reward function R of the DDPG algorithm is designed as: R=w 1 *(1 / |ΔP|)+w 2 *(1 / |ΔP_mainSteam|)-w3*(|d Coal / d t |) ； wherein ΔP is the power deviation, ΔP_mainSteam is the main steam pressure deviation, d Coal / d t is the coal feed rate change rate, w1 , w2 , w3 is the weight coefficient.

6. An AGC signal intelligent correction and control device, characterized in that, It comprises: A signal acquisition module for acquiring AGC instruction signals on the power grid dispatching side and actual power feedback signals on the power plant side, and constructing multi-source heterogeneous data; A prediction module for inputting the multi-source heterogeneous data into a pre-trained digital twin model to obtain a predicted value of real power of the unit; The signal correction module is configured to calculate a real-time deviation between a predicted value of the real power and the actual power feedback signal, dynamically compensate the AGC instruction signal based on the real-time deviation, and generate a corrected AGC instruction signal; The control module is configured to generate a feedforward control signal based on the corrected AGC instruction signal, and superimpose the feedforward control signal on a basic control instruction of a unit distributed control system (DCS) in an incremental form to realize precise control in a full load range.

7. The AGC signal intelligent correction and control device of claim 6, wherein, The digital twin model adopts an architecture combining a long short-term memory (LSTM) network and a Transformer attention mechanism, captures time sequence features of input parameters through the LSTM network, and dynamically weights influence weights of different features in the input parameters on power output through the attention mechanism.

8. The AGC signal intelligent correction and control device of claim 6, wherein, The signal correction module is specifically configured to: The incremental PID algorithm is adopted for dynamic compensation, and an output of the incremental PID algorithm is an instruction compensation amount ΔL, and the following formula is satisfied: Wherein, e(t) is the real-time deviation, and a is a proportional coefficient, and β is an integral coefficient; The proportional coefficient a and the integral coefficient β are dynamically self-adaptively adjusted by a reinforcement learning agent according to a current load change rate of the unit and a main steam pressure deviation; The AGC instruction signal and the instruction compensation amount are superimposed to generate the corrected AGC instruction signal, and a calculation formula of the corrected AGC instruction signal is as follows: L_corrected=L_agc+ΔL; Wherein, L_agc is the AGC instruction signal.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the AGC signal intelligent correction and control method according to any one of claims 1 to 5.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to realize the steps of the AGC signal intelligent correction and control method according to any one of claims 1 to 5.

Images