A deep learning-based reheating steam temperature control method for thermal power generating units

Abstract

The application discloses a kind of deep learning-based thermal power unit reheat steam temperature control method, comprising the following steps: S1, the historical operation data under wide load of unit is collected, and steady-state operating condition data in the historical operation data under wide load of unit is selected;S2, unit steady-state operating condition data is input into LSTM neural network, and neural network model 1 is obtained by offline training;S3, dynamic historical data under wide load operation of unit is collected, input into LSTM neural network, and neural network model 2 is obtained by offline training;S4, according to the output data of neural network model 1 and neural network model 2, combine unit real-time operation data to construct reheat steam temperature control system under wide load of unit.The application solves the problem that existing reheat steam temperature control method requires thermal power unit to have wide load operation capability, and control strategy is difficult to consider high and low load operation characteristics.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control technology for power generation processes, and specifically relates to a deep learning-based method for controlling reheat steam temperature in thermal power units. Background Technology

[0002] Reheat steam temperature is a crucial parameter directly affecting the thermal efficiency of thermal power units, and improving the accuracy of automatic reheat steam temperature control is an important means to enhance the unit's economic efficiency. The reheat steam temperature control object is a process with large inertia, large delay, and strong nonlinearity. The characteristics of this process are mainly related to the unit load, and are simultaneously affected by changes in coal quality, desuperheating water volume, inlet steam temperature and pressure, and inlet flue gas flow rate and temperature, making it highly susceptible to disturbances and exhibiting nonlinear and time-varying characteristics. Current regulation methods are difficult to directly achieve their intended effect; therefore, reheat steam temperature damper control systems face problems such as difficulty in automatic activation and poor regulation performance.

[0003] Chinese invention patent application number CN202111340430.4 discloses a method for controlling reheat steam temperature in a secondary reheat boiler based on flue gas damper adjustment; Chinese invention patent application number CN202311407747.4 discloses a method, device, storage medium, and equipment for optimizing reheat steam temperature control in a secondary reheat unit; Chinese invention patent application number CN202311044755.7 discloses a method, system, and equipment for controlling reheat steam temperature based on a dual closed-loop control structure; and Chinese invention patent application number CN201911311466.2 discloses a method for controlling reheat steam temperature based on a neural network inverse model.

[0004] The drawback of the aforementioned patents is that they all employ a reheat steam temperature control method based on a cascade control structure. However, current thermal power units are required to have wide load operation capabilities. Under wide loads, the controlled characteristics of the reheat steam temperature object vary greatly, and existing control strategies are difficult to take into account both high and low load operation characteristics. Summary of the Invention

[0005] To address the problem that existing reheat steam temperature control methods for thermal power units require wide-load operation capabilities and that control strategies struggle to accommodate both high and low load operating characteristics, this invention provides a deep learning-based reheat steam temperature control method for thermal power units.

[0006] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows:

[0007] A deep learning-based method for reheat steam temperature control in thermal power units, comprising the following steps:

[0008] S1. Collect historical operating data of the unit under wide load conditions, and select steady-state operating data from the historical operating data of the unit under wide load conditions;

[0009] S2. Input the unit's steady-state operating data into the LSTM neural network and train it offline to obtain neural network model 1;

[0010] S3. Collect dynamic historical data under wide load operation of the unit, input it into the LSTM neural network, and train it offline to obtain neural network model 2;

[0011] S4. Based on the output data of neural network model 1 and neural network model 2, and combined with the real-time operation data of the unit, construct a reheat steam temperature control system under wide load conditions.

[0012] Furthermore, the criteria for selecting steady-state operating data from the historical operating data of the unit under wide load conditions are as follows:

[0013]

[0014] in X represents the average load of this section of the unit. i This represents the unit load at a specific moment in the historical operating data. σ represents the deviation threshold, which can be set to 5%, 1%, etc., depending on the requirements. The data segment length is set to 30 minutes. The above rules filter out 100 or more steady-state operating points.

[0015] Furthermore, the historical operating data of the unit under wide load conditions includes unit load, reheater inlet flue gas temperature, reheater inlet steam temperature, coal feed rate, lower heating value of coal, number of coal mills in operation, and damper opening data.

[0016] The unit load, reheater inlet flue gas temperature, reheater inlet steam temperature, coal feed rate, lower heating value of coal, and number of coal mills in operation are used as inputs to neural network model 1, and the damper opening data value is used as the output of neural network model 1.

[0017] Furthermore, the dynamic historical data under wide load operation of the unit includes flue gas damper opening, reheater inlet flue gas temperature, reheater inlet steam temperature, unit load change rate, unit load, number of coal mills in operation, and reheater outlet steam temperature.

[0018] The flue gas damper opening, reheater inlet flue gas temperature, reheater inlet steam temperature, unit load change rate, unit load, and number of coal mills in operation are the inputs to neural network model 2, and the reheater outlet steam temperature is the output of neural network model 2.

[0019] Furthermore, the reheat steam temperature control system under wide load conditions includes a feedback loop module, a controller input module, a controller module, and a feedforward signal module.

[0020] Furthermore, the feedback loop module includes feedback loop module 1 and feedback loop module 2;

[0021] Feedback loop module 1 adopts the formula: Let z2=K*u, where K is a constant and u is the flue gas damper opening control command;

[0022] Feedback loop module 2 uses the formula: Let z1 = y1 - y2, where y2 is the output of neural network model 2 and y1 is the real-time measured value of the reheater outlet steam temperature.

[0023] Furthermore, the controller input module uses the formula: r2=r-z1-z2, where r is the set value of the reheater outlet steam temperature.

[0024] Furthermore, the feedforward control signal formula used by the feedforward signal module is: u = v1 + v2, where v1 is the output of neural network model 1 and v2 is the controller output.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] The nonlinear characteristics of the reheat steam temperature controlled object under wide load are fitted by a neural network. At the same time, the nonlinear characteristics under wide load are compensated in the control strategy to construct a system suitable for wide load operation of the reheat steam temperature control system. In addition, the compensation effect of the large delay and large inertia link is improved by the neural network model 2, so as to realize the automatic and precise control of the reheat steam temperature under wide load. Attached Figure Description

[0027] Figure 1 This is an overall flowchart of a deep learning-based reheat steam temperature control method for thermal power units in an embodiment of the present invention.

[0028] Figure 2 This is a schematic diagram of a deep learning-based reheat steam temperature control method for thermal power units in an embodiment of the present invention. Detailed Implementation

[0029] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0030] like Figure 1 As shown, this embodiment provides a deep learning-based reheat steam temperature control method for thermal power units, including the following steps:

[0031] S1. Collect historical operating data of the unit under wide load conditions, and select steady-state operating data from the historical operating data of the unit under wide load conditions;

[0032] S2. Input the unit's steady-state operating data into the LSTM neural network and train it offline to obtain neural network model 1;

[0033] S3. Collect dynamic historical data under wide load operation of the unit, input it into the LSTM neural network, and train it offline to obtain neural network model 2;

[0034] S4. Based on the output data of neural network model 1 and neural network model 2, and combined with the real-time operation data of the unit, construct a reheat steam temperature control system under wide load conditions.

[0035] The criteria for selecting steady-state data from historical operating data under wide load conditions are as follows:

[0036]

[0037] in X represents the average load of this section of the unit. i This represents the unit load at a specific moment in the historical operating data. σ represents the deviation threshold, which can be set to 5%, 1%, etc., depending on the requirements. The data segment length is set to 30 minutes. The above rules filter out 100 or more steady-state operating points.

[0038] Historical operating data of the unit under wide load conditions includes unit load, reheater inlet flue gas temperature, reheater inlet steam temperature, coal feed rate, lower heating value of coal, number of coal mills in operation, and damper opening data.

[0039] The unit load, reheater inlet flue gas temperature, reheater inlet steam temperature, coal feed rate, lower heating value of coal, and number of coal mills in operation are used as inputs to neural network model 1, and the damper opening data value is used as the output of neural network model 1.

[0040] The dynamic historical data of the unit under wide load operation includes flue gas damper opening, reheater inlet flue gas temperature, reheater inlet steam temperature, unit load change rate, unit load, number of coal mills in operation, and reheater outlet steam temperature.

[0041] The flue gas damper opening, reheater inlet flue gas temperature, reheater inlet steam temperature, unit load change rate, unit load, and number of coal mills in operation are the inputs to neural network model 2, and the reheater outlet steam temperature is the output of neural network model 2.

[0042] The reheat steam temperature control system under wide load conditions includes a feedback loop module, a controller input module, a controller module, and a feedforward signal module.

[0043] The feedback loop module includes feedback loop module 1 and feedback loop module 2;

[0044] Feedback loop module 1 adopts the formula: Let z2=K*u, where K is a constant and u is the flue gas damper opening control command;

[0045] Feedback loop module 2 uses the formula: Let z1 = y1 - y2, where y2 is the output of neural network model 2 and y1 is the real-time measured value of the reheater outlet steam temperature.

[0046] The controller input module uses the formula: r2=r-z1-z2, where r is the set value of the reheater outlet steam temperature.

[0047] The feedforward control signal formula used by the feedforward signal module is: u = v1 + v2, where v1 is the output of neural network model 1 and v2 is the controller output.

[0048] Compared with the prior art, the present invention has the following advantages:

[0049] The nonlinear characteristics of the reheat steam temperature controlled object under wide load are fitted by a neural network. At the same time, the nonlinear characteristics under wide load are compensated in the control strategy to construct a system suitable for wide load operation of the reheat steam temperature control system. In addition, the compensation effect of the large delay and large inertia link is improved by the neural network model 2, so as to realize the automatic and precise control of the reheat steam temperature under wide load.

[0050] The above provides a detailed description of a deep learning-based reheat steam temperature control method for thermal power units. The specific embodiments described are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the scope of protection of the claims.

Claims

1. A deep learning-based method for controlling reheat steam temperature in thermal power units, characterized in that, Including the following steps: S1. Collect historical operating data of the unit under wide load conditions, and select steady-state operating data from the historical operating data of the unit under wide load conditions; S2. Input the unit's steady-state operating data into the LSTM neural network and train it offline to obtain neural network model 1; S3. Collect dynamic historical data under wide load operation of the unit, input it into the LSTM neural network, and train it offline to obtain neural network model 2; S4. Construct a reheat steam temperature control system for the unit under wide load conditions by combining the output data of neural network model 1 and neural network model 2 with the real-time operation data of the unit. The criteria for selecting steady-state data from historical operating data under wide load conditions are as follows: ， Where X is the average load of this section of the unit. i This represents the unit load at a specific moment in the historical operating data, where σ represents the deviation threshold, and the data segment length is set to 30 minutes. The reheat steam temperature control system under wide load conditions includes a feedback loop module, a controller input module, a controller module, and a feedforward signal module. The feedback loop module includes feedback loop module 1 and feedback loop module 2; Feedback loop module 1 adopts the formula: Let z2=K*u, where K is a constant and u is the flue gas damper opening control command; Feedback loop module 2 adopts the formula: Let z1=y1-y2, where y2 is the output of neural network model 2 and y1 is the real-time measured value of reheater outlet steam temperature; The controller input module uses the formula: r2=r-z1-z2, where r is the set value of the reheater outlet steam temperature; The feedforward control signal formula used by the feedforward signal module is: u=v1+v2, where v1 is the output of neural network model 1 and v2 is the controller output; Historical operating data of the unit under wide load conditions includes unit load, reheater inlet flue gas temperature, reheater inlet steam temperature, coal feed rate, lower heating value of coal, number of coal mills in operation, and damper opening data. The unit load, reheater inlet flue gas temperature, reheater inlet steam temperature, coal feed rate, lower heating value of coal, and number of coal mills in operation are used as inputs to neural network model 1, and the damper opening data value is used as outputs to neural network model 1. The dynamic historical data of the unit under wide load operation includes flue gas damper opening, reheater inlet flue gas temperature, reheater inlet steam temperature, unit load change rate, unit load, number of coal mills in operation, and reheater outlet steam temperature. The flue gas damper opening, reheater inlet flue gas temperature, reheater inlet steam temperature, unit load change rate, unit load, and number of coal mills in operation are the inputs to neural network model 2, and the reheater outlet steam temperature is the output of neural network model 2.

Images