Powder filter operation control method under deep peak regulation working condition of thermal power generating unit

Abstract

The invention discloses a powder filter operation control method for a deep peak regulation working condition of a thermal power generating unit, and relates to the technical field of filter operation control, and the method comprises the steps: obtaining the control information of the thermal power generating unit, and carrying out the recognition of a real-time peak regulation working condition; under each peak regulation working condition, influence factors are released, and the powder state is analyzed through a time accumulation analysis model; generating a control instruction according to the analysis result of the powder state; and performing dynamic operation control on the powder filter by taking the control instruction as a control induction signal. The timeliness and the filtering efficiency of dust treatment are remarkably improved, the system blockage probability is reduced, the service life of equipment is prolonged, and the overall operation stability and the environment-friendly safety level of a thermal power generating unit under the conditions of low load and frequent peak regulation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of filter operation control, in particular to a powder filter operation control method for deep peak shaving working condition of thermal power generating unit. BACKGROUND

[0002] Under the background of "double carbon" strategy, in order to improve the renewable energy consumption capacity, thermal power generating unit needs to undertake more and more frequent and larger amplitude peak shaving tasks, especially the deep peak shaving working condition gradually becomes the norm. However, in the process of deep peak shaving operation, the boiler load fluctuates sharply, and the combustion state is unstable, which directly leads to significant changes in dust particle size, concentration, distribution and adhesion performance in flue gas. This complex dynamic disturbance is prone to cause problems such as dust removal system blockage, efficiency reduction and equipment wear and tear increase, especially for powder filter, its blockage, pressure drop rise and operation rhythm disorder become important factors restricting system stability and environmental protection emission.

[0003] The existing powder filter control strategy is mostly based on timing, fixed pressure or static rule triggered control logic, which cannot dynamically respond to the dust disturbance actually generated in the process of peak shaving, especially lacks the modeling and forward-looking control ability of the relationship between peak shaving intensity and dust accumulation. In addition, in the current working condition identification means, it is difficult to accurately describe the evolution mechanism between complex nonlinear peak shaving behavior and powder state, causing filter control lag, frequent misjudgment, affecting operation safety and economy. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a powder filter operation control method for deep peak shaving working condition of thermal power generating unit to solve the problems of complex powder state change, filter control response lag, and inability to predict the risk of blockage in advance, and lack of intelligent adaptation of control strategy in the prior art under deep peak shaving operation.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a powder filter operation control method for deep peak shaving working condition of thermal power generating unit, which comprises: obtaining control information of the thermal power generating unit, identifying the real-time peak shaving working condition; Under each peak shaving working condition, the influence factor is released, and the powder state is analyzed through the time accumulation analysis model; According to the analysis result of the powder state, the control instruction is generated; The control instruction is used as the induction signal of control to dynamically control the powder filter.

[0007] As a preferred embodiment of the powder filter operation control method for deep peak shaving conditions of thermal power units described in this invention, wherein: The control information includes peak-shaving commands received by the thermal power units and operating data of the thermal power units.

[0008] As a preferred embodiment of the powder filter operation control method for deep peak shaving conditions of thermal power units described in this invention, the peak shaving condition identification process includes: through the training of a Bayesian network with preset standard peak shaving conditions, the Bayesian network is able to identify each standard peak shaving condition. Based on the control information, a probability distribution for the peak-shaving condition identification results is generated.

[0009] As a preferred embodiment of the powder filter operation control method for deep peak shaving conditions of thermal power units according to the present invention, wherein: based on the identification result of the peak shaving conditions, the influencing factors of the peak shaving conditions are released: In each standard peak-shaving condition, a corresponding impact factor package is embedded; According to the probability distribution, the impact factor package is released with probability; wherein, the impact factor corresponding to the highest probability in the probability distribution is released with a probability of p×p. Upon release, the additional package of influence factors obtained from historical data training is released according to the peak-shaving condition corresponding to the highest probability in the probability distribution. When the impact factor supplementary package is released, the impact factor is released with probability according to (1-p)×p; where p represents the maximum probability in the probability distribution. The impact factor package is a set containing multiple impact factors, each impact factor in the set being a powder that causes a blockage element; wherein, in a set, there may be multiple powders that cause a blockage element. The set of influence factors in each influence factor package must satisfy the discrimination of the discriminator in the adversarial network; by using the control process of each thermal power unit's deep peak shaving condition and the monitoring data of powder during the control process, the adversarial network is trained to obtain a trained adversarial network, including a generator and a discriminator; when the deep peak shaving condition occurs, the control information is input into the adversarial network to generate the influence factor package; The impact factor supplement package includes n impact factors randomly selected from the database consisting of the training set of the adversarial network under the peak-shaving conditions corresponding to p standards. Where n represents the number of influence factors in the influence factor package corresponding to the highest probability in the probability distribution.

[0010] As a preferred embodiment of the powder filter operation control method for deep peak shaving conditions of thermal power units described in this invention, the time accumulation analysis model includes: establishing a historical peak shaving condition record structure based on a sliding window, storing each influencing factor release event by timestamp, and setting the window length T and sliding step size Δt to extract the influencing factor release sequence within the past T time period; The accumulation of influencing factors is simulated to identify bottleneck elements, and a time-series regression model is used to predict the simulation results. The simulation results of the current congestion factors and the prediction results of the simulation results within the delayed period S are obtained.

[0011] As a preferred embodiment of the powder filter operation control method for deep peak shaving of thermal power units according to the present invention, the powder state analysis includes inputting the simulation results and the predicted results of the simulation results into a pre-trained prior probability distribution model to generate the probability distribution of the powder state at each metering time in the current time T0 and the predicted time period [T1, T2]. Based on the probability distribution, a feature vector is encoded for each measurement time. Using the feature vector, control commands are matched to obtain the control commands for each measurement time.

[0012] As a preferred embodiment of the powder filter operation control method for deep peak shaving of thermal power units according to the present invention, the induced signal includes: constructing a sequence of control commands at each metering time [T0, T2] to obtain a control command sequence; The control command sequence is smoothed to obtain the final control command sequence, and control is performed according to the final control command sequence; during the control process, the final control command sequence is synchronously updated according to the real-time updated control command sequence. During the smoothing process, the control command sequence is processed with the minimum deviation as the objective function.

[0013] Secondly, the present invention provides a powder filter operation control system for deep peak shaving conditions of thermal power units, including a data acquisition unit for acquiring control information of thermal power units and identifying real-time peak shaving conditions. The analysis unit releases influencing factors under each peak-shaving condition and analyzes the powder state through a time-accumulated analysis model. The instruction generation unit generates control instructions based on the analysis results of the powder state. The control unit uses the control commands as induction signals to dynamically control the operation of the powder filter.

[0014] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the powder filter operation control method for deep peak shaving conditions of thermal power units as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the powder filter operation control method for deep peak shaving conditions of thermal power units as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By introducing a mechanism for identifying deep peak-shaving conditions in thermal power units, this invention can accurately determine the peak-shaving type at different operating stages, and construct a probability distribution for peak-shaving identification based on a Bayesian network, achieving real-time perception and classification of peak-shaving behavior. Furthermore, by combining influence factor packets and additional packets generated by adversarial networks, the potential risks of dust generation during peak-shaving are dynamically simulated. By constructing a time-cumulative model, the system can continuously simulate and predict the powder state, and use prior probability distributions to achieve high-confidence prediction of dust blockage trends. Finally, through a smoothed control command sequence, the powder filter is induced to intervene and adjust in advance, significantly improving the timeliness and filtration efficiency of dust treatment, reducing the probability of system blockage, extending equipment lifespan, and enhancing the overall operational stability and environmental safety level of thermal power units under low load and frequent peak-shaving conditions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of the powder filter operation control method for deep peak shaving conditions of thermal power units. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Reference Figure 1 As an embodiment of the present invention, this embodiment provides a powder filter operation control method for deep peak shaving conditions of thermal power units, including the following steps: S1: Acquire control information from thermal power units and identify real-time peak-shaving conditions.

[0023] The control information includes peak-shaving commands received by the thermal power unit and the unit's operating data. Peak-shaving commands include, for example, load change rate commands, peak-shaving time windows, start / stop control signals, or rapid load increase / decrease commands. Operating data includes, for example, the unit's current generating load and its first derivative, boiler flue gas volume, flue gas temperature, flue gas humidity, dust characteristic parameters, and filter status parameters.

[0024] By training a Bayesian network under preset standard peak-shaving conditions, the Bayesian network can identify peak-shaving conditions for each standard; based on the control information, a probability distribution of the peak-shaving condition identification results is generated.

[0025] By collecting multi-dimensional control information, including peak-shaving commands and unit operating data, the system can construct a comprehensive description of the current unit operating status. Peak-shaving commands, such as load change rate, start-stop control, and peak-shaving window, reflect the system's external control requirements, while operating data, such as flue gas parameters, dust proxy indicators, and filter status parameters, reflect internal response behavior. To address the challenges of variable and complex peak-shaving conditions, the invention employs a Bayesian network-based operating condition identification algorithm. Through training and modeling on preset standard operating conditions, the model can capture the evolution trend of operating conditions under multiple factors and output an identification probability distribution with uncertainty expression capabilities. This probability distribution can not only be used for accurate identification of the current peak-shaving state but also serve as the basis for subsequent release of influencing factors and generation of control commands, realizing a closed-loop control strategy from "identification—prediction—response," effectively improving the response efficiency and operational stability of the powder filter under deep peak-shaving conditions.

[0026] S2: Under each peak-shaving condition, the influencing factors are released, and the powder state is analyzed through a time-accumulated analysis model.

[0027] Based on the identification results of the peak-shaving conditions, the impact factors of the peak-shaving conditions are released: The corresponding impact factor package is embedded in each standard peak-shaving operating condition.

[0028] According to the probability distribution, the probability of the influence factors in the influence factor package is released (which is actually a probability extraction process); wherein, the influence factor corresponding to the highest probability in the probability distribution is released with a probability of p×p. By performing a probability extraction process for each working condition in the probability analysis, regardless of special situations that occur during effective simulation operation (simulating the occurrence of special powders).

[0029] Upon release, the additional package of influence factors obtained from historical data training is released based on the peak-shaving condition corresponding to the highest probability in the probability distribution.

[0030] All standard peak-shaving operating conditions This is the m-th standard peak-shaving condition.

[0031] Operating condition identification probability distribution , This represents the probability distribution result for the m-th standard peak-shaving condition.

[0032] Each working condition The corresponding impact factor. This represents the number of impact factors in the impact factor package during the i-th standard peak-shaving operation.

[0033] Operating conditions The set of influencing factors released.

[0034] The final set of all releasing factors.

[0035] For peak-shaving conditions that are not the maximum probability, the release model is as follows: , , in Indicates from the factor set According to proportion A randomly selected subset.

[0036] For the peak-shaving condition corresponding to the highest probability, let k be the index of the peak-shaving condition corresponding to the highest probability: This represents the peak-shaving operating condition with the highest probability.

[0037] Peak shaving conditions The impact factor corresponding to the main package.

[0038] Peak shaving conditions The impact factor of the corresponding sub-package.

[0039] The final set of influencing factors that are released.

[0040] , , , By combining R1 and R2, the final result of the impact factor release is obtained.

[0041] When releasing the additional package of impact factors, the probability of releasing the impact factors is calculated according to (1-p)×p, where p represents the maximum probability in the probability distribution. This ensures that the actual working conditions are released in the most realistic way, while also taking into account historical special cases. The implementation of "releasing the impact factor corresponding to the maximum probability in the probability distribution" through the combination of p×p and (1-p)×p can be simplified to p×p + (1-p)×p = p; it can be seen that this also ensures that the combined probability of the two is p. The sum of the extraction probabilities of impact factors that may come from other working conditions is 1.

[0042] The impact factor package is a set containing multiple impact factors, each of which is a powder that causes a blockage element; wherein, in a set, there may be multiple powders that cause a blockage element.

[0043] The set of influence factors in each influence factor package must satisfy the discrimination of the discriminator in the adversarial network; by using the control process of each thermal power unit's deep peak shaving condition and the monitoring data of powder during the control process, the adversarial network is trained to obtain a trained adversarial network, including a generator and a discriminator; when the deep peak shaving condition occurs, the control information is input into the adversarial network to generate the influence factor package.

[0044] The additional impact factor package includes n impact factors randomly selected from the database consisting of the training set of the adversarial network under the peak-shaving condition corresponding to the p standard. Here, n represents the number of impact factors in the impact factor package corresponding to the highest probability in the probability distribution.

[0045] It is worth mentioning that, when facing different types of standard peak-shaving conditions, the system can flexibly and accurately release key factors affecting dust formation based on the identified probability distribution, thereby constructing a more realistic and reliable powder state simulation environment. By embedding pre-trained influencing factor packages into each peak-shaving condition and releasing a combination of main and secondary packages according to the current identification probability, where the main package is released using p×p and the secondary package is released using (1–p)×p, the system ensures that the main judgment condition dominates the powder generation characteristics, while also taking into account the interference modeling of historical edge or extreme conditions on the control strategy, thereby enhancing the robustness and generalization ability of the overall control system.

[0046] The generator produces potential factor combinations using peak-shaving control parameters, while the discriminator, based on discrimination logic constructed from dust monitoring data, classifies each factor group, effectively screening powder patterns with high representativeness and potential clogging risk. The secondary package extraction mechanism performs compensated sampling based on a database built from historical data, improving simulation coverage. Finally, the fusion release method of R1 and R2 ensures the integrity, rationality, and controllability of the release results, providing precise input support for subsequent time-cumulative analysis models and powder state assessment modules, achieving the adaptive and induced control target of the powder filter.

[0047] Furthermore, the time-accumulation analysis model includes establishing a historical peak-shaving condition record structure based on a sliding window, storing each influencing factor release event by timestamp, and setting the window length T and sliding step size Δt to extract the influencing factor release sequence within the past T time period.

[0048] The cumulative effect of influencing factors is simulated by identifying blockage elements, and a time-series regression model is used to predict the simulation results. It's important to note that directly simulating based on powder size can lead to overfitting. Using an intermediate quantity that reflects the overall effect helps prevent overfitting.

[0049] The simulation results of the current congestion factors and the prediction results of the simulation results within the delayed period S are obtained.

[0050] When simulating clogging factors, the model uses intermediate variables (such as release frequency, intensity, or type indicators) to replace direct physical attributes like powder particle size. This avoids overfitting bias caused by uneven sample dimensionality and drastic particle size distribution, thereby improving the model's generalization ability and robustness under complex operating conditions. Simultaneously, considering the cumulative characteristics of time-series data, the model integrates time-series regression prediction modules with time-delay response structures, such as ARIMA, LSTM, or Gated Recurrent Units (GRUs), to predict clogging risk trends over the next S time period. This prediction result can not only help determine the current filter operating status but also serve as input to the control module, supporting the formulation of more precise induced control strategies, thereby effectively reducing operational efficiency decline and safety risks caused by clogging.

[0051] S3: Based on the analysis results of the powder state, generate control commands.

[0052] Specifically, the simulation results and the predicted results of the simulation results are input into a pre-trained prior probability distribution model to generate the probability distribution of the powder state at each measurement time in the current time T0 and the predicted time period [T1, T2].

[0053] Based on the probability distribution, a feature vector is encoded for each measurement moment. Specifically, by inputting the simulation results of the current moment and the prediction results of future time periods into a pre-trained prior probability distribution model, the uncertainty distribution of the powder state at each measurement moment is obtained, and then a feature vector that can be used for control decision-making is formed by encoding the probability distribution. This approach transforms the changing trend of powder state over multiple time periods into a standardized, computable control input, realizing a structured encoding expression of unstructured prediction results.

[0054] Using the feature vector, control commands are matched to obtain the control command for each measurement time. In this embodiment, a pre-fitted control command model is used as input to the feature vector to generate the corresponding control command (including each control parameter and its value).

[0055] During the control strategy generation process, a pre-fitted control command matching model is introduced. This model can quickly locate the set of control commands that best match the current state based on the aforementioned feature vectors, thereby enabling fine-tuning of the value range of each control parameter. The model is trained based on control effects from historical samples, ensuring that the generated control commands not only possess dynamic adaptability but also perform multi-objective optimization by combining control objectives (such as filtering efficiency, energy consumption minimization, or fault risk suppression). This strategy effectively solves the control response delay problem caused by state change lag or prediction uncertainty, improving the robustness and sensitivity of the powder filter in actual operation and significantly enhancing the system's filtering control capability under complex peak-shaving disturbances.

[0056] S4: Use the control command as a control induction signal to dynamically control the operation of the powder filter.

[0057] The induction signal includes constructing a sequence of control commands at each measurement time [T0, T2] to obtain a control command sequence.

[0058] The control command sequence is smoothed to obtain the final control command sequence, and control is performed according to the final control command sequence; during the control process, the final control command sequence is synchronously updated according to the real-time updated control command sequence.

[0059] In the smoothing process, the control command sequence is processed with the minimum deviation as the objective function. In actual control, to avoid drastic timing jumps in control commands that could cause system instability, this invention introduces the minimum deviation principle for sequence smoothing. Specifically, the control command sequence is fitted using an optimization function to minimize changes in control commands between adjacent time points, thereby generating a final control command sequence that is physically more stable and has a smoother execution strategy. This smoothing mechanism not only improves the stability of the control system but also enhances the adaptability of control behavior to the actual operating state of the physical device. At each metering time, the control commands are dynamically adjusted based on the latest prediction results, ensuring that the powder filter retains its self-learning and self-optimization capabilities even under varying operating conditions. This control strategy based on sequence smoothing and dynamic updates significantly enhances the response sensitivity and control accuracy of the powder filter in thermal power units under deep peak-shaving conditions, ensuring the efficient and safe operation of key components.

[0060] This embodiment also provides a powder filter operation control system for deep peak shaving conditions of thermal power units, including: a data acquisition unit, which acquires control information of thermal power units and identifies real-time peak shaving conditions.

[0061] The analysis unit releases influencing factors under each peak-shaving condition and analyzes the powder state using a time-accumulated analysis model.

[0062] The instruction generation unit generates control instructions based on the analysis results of the powder state.

[0063] The control unit uses the control commands as induction signals to dynamically control the operation of the powder filter.

[0064] This embodiment also provides a computer device applicable to the powder filter operation control method under deep peak shaving conditions of thermal power units, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the powder filter operation control method under deep peak shaving conditions of thermal power units as proposed in the above embodiment.

[0065] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0066] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the powder filter operation control method for achieving deep peak shaving conditions in thermal power units as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0067] In summary, this invention constructs an intelligent operation control method for powder filters under deep peak-shaving conditions in thermal power units by organically integrating operating condition identification, factor release, time series analysis, and dynamic control. First, a Bayesian network is used to jointly model peak-shaving commands and operating data, enabling probabilistic identification of multiple types of peak-shaving conditions. Based on the identification results, a release mechanism for corresponding influencing factor packages is triggered, fusing the main influencing factor and supplementary packages to form a representative release combination.

[0068] Secondly, by constructing a sliding time window and a time-series regression model, we can dynamically simulate the cumulative evolution trend of powder influencing factors over time, which avoids feature overfitting and enhances the ability to make forward-looking judgments on blockage risks.

[0069] Based on this, the powder state is mapped to a feature vector using a prior distribution model, and a corresponding control command sequence is generated. Finally, the sequence smoothing and dynamic update control of the commands are achieved through the minimum deviation optimization criterion. The overall solution is model-driven at its core, taking into account both historical learning and real-time adaptation. It can effectively improve the operating efficiency and risk response capability of powder filters under the background of deep peak shaving in thermal power units, and has significant engineering application value.

[0070] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for operating a powder filter in a deep load modulation condition of a thermal power generating unit, characterized in that: The method comprises: obtaining control information of a thermal power unit; identifying a real-time peak regulation condition; In each peak regulation condition, an influencing factor is released, and a powder state is analyzed through a time accumulation analysis model; According to the analysis result of the powder state, a control instruction is generated; The control instruction is used as an induction signal for control to dynamically control the powder filter.

2. The method for operation control of a powder filter in a deep load modulation condition of a thermal power generating unit according to claim 1, characterized in that: The control information comprises a peak regulation instruction received by the thermal power unit and operation data of the thermal power unit.

3. The method for operation control of a powder filter in a deep load modulation condition of a thermal power generating unit according to claim 2, characterized in that: The identification of the peak regulation condition comprises: enabling a Bayesian network to identify each standard peak regulation condition through a preset standard peak regulation condition and training of the Bayesian network; According to the control information, a probability distribution of the identification result of the peak regulation condition is generated.

4. The method for operation control of a powder filter in a deep load modulation condition of a thermal power generating unit according to claim 3, characterized in that: According to the identification result of the peak regulation condition, an influencing factor of the peak regulation condition is released; In each standard peak regulation condition, a corresponding influencing factor package is embedded; According to the probability distribution, the influencing factor package is subjected to probability release of the influencing factor; wherein, the influencing factor corresponding to the maximum probability in the probability distribution is subjected to p×p probability release; At the time of release, according to the peak regulation condition corresponding to the maximum probability in the probability distribution, an influencing factor additional package trained from historical records is released; At the time of release of the influencing factor additional package, the influencing factor is subjected to (1-p)×p probability release; wherein, p represents the maximum probability in the probability distribution; The influencing factor package is a set containing multiple influencing factors; each influencing factor in the set is a powder causing one clogging element; wherein, in one set, the powders causing one clogging element can be multiple; Each influencing factor set in the influencing factor package needs to satisfy the discrimination of a discriminator in a generative adversarial network; the generative adversarial network is trained by using control processes of each deep peak regulation condition of the thermal power unit and monitoring data of powders in the control processes, to obtain a trained generative adversarial network including a generator and a discriminator; when the deep peak regulation condition occurs, the control information is input into the generative adversarial network to generate the influencing factor package; The influencing factor additional package comprises: n influencing factors randomly extracted from a database constituted by a training set of the generative adversarial network in a p corresponding standard peak regulation condition; Wherein, n represents the number of influencing factors in the influencing factor package corresponding to the maximum probability in the probability distribution.

5. The method for operation control of a powder filter in a deep load modulation condition of a thermal power generating unit according to claim 4, characterized in that: The time accumulation analysis model comprises: establishing a historical peak regulation condition record structure based on a sliding window; storing each influencing factor release event according to a time stamp; and setting a window length T and a sliding step Δt, which are used to extract an influencing factor release sequence in the past T time; The accumulation of the influencing factors is subjected to simulation of clogging elements, and a time series regression model is used to predict the simulation result; The simulation result of the current clogging element and the prediction result of the simulation result for a delayed period S are obtained.

6. The method for control of operation of a powder filter in deep load modulation mode of a thermal power unit according to claim 5, characterized in that: The analysis of the powder state comprises: inputting the simulation result and the prediction result of the simulation result into a pre-trained prior probability distribution model to generate a probability distribution of the powder state at each metering time in the current time T0 and the prediction time period [T1, T2]. According to the probability distribution, a feature vector of each metering moment is coded; Using the feature vector, a matching of control instructions is performed to obtain the control instruction of each metering moment.

7. The method for control of operation of a powder filter in deep load modulation mode of a thermal power unit according to claim 6, characterized in that: The induction signal includes sequence construction of the control instruction of each metering moment in [T0, T2] to obtain a control instruction sequence; The control instruction sequence is smoothed to obtain a final control instruction sequence, and the final control instruction sequence is used for control; during the control, the final control instruction sequence is synchronously updated according to the real-time updated control instruction sequence; During the smoothing process, the control instruction sequence is processed with the minimum deviation as the objective function.

8. A system for controlling the operation of a powder filter in a deep load modulation condition of a thermal power generating unit, based on the method for controlling the operation of a powder filter in a deep load modulation condition of a thermal power generating unit according to any one of claims 1 to 7, characterized in that: It comprises a collection unit, which acquires control information of a thermal power unit and identifies a real-time peak shaving condition; An analysis unit releases an influence factor under each peak shaving condition and analyzes a powder state through a time accumulation analysis model; An instruction generation unit generates a control instruction according to the analysis result of the powder state; A control unit uses the control instruction as an induction signal for control to dynamically control the operation of a powder filter. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to implement the steps of the powder filter operation control method for the deep peak shaving condition of the thermal power unit according to any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the powder filter operation control method for the deep peak shaving condition of the thermal power unit according to any one of claims 1-7.

Images