A PGNN feedforward dynamic compensation-multi-model predictive control method

Abstract

The present invention discloses a multi-model predictive control method based on multi-model gap measurement PGNN feedforward dynamic compensation, which relates to the technical field related to coordinated control of generator sets. The predictive model estimates the future state of the system and guides the controller to make adjustments in advance to solve the deviation and overshoot problems that are easy to occur in the main steam temperature when the unit is running at a wide load, and realizes accurate and rapid response to load changes through feedforward compensation. This solution integrates technologies such as PGNN predictive model, feedforward dynamic compensation and multi-model predictive control to improve the flexibility of the control system. Through accurate prediction of the predictive model and optimization of valve opening feedforward dynamic compensation, the control system can better respond to changes in the main steam temperature, improve the performance of the control system, eliminate deviations, maintain stability, and enable the main steam temperature to more accurately follow the set value, thereby achieving better control effects and improving control quality.

Description

Technical Field

[0001] The present invention relates to the technical field related to coordinated control of generator sets, and in particular to a PGNN feedforward dynamic compensation-multi-model predictive control method. Background Art

[0002] With the development of clean energy technologies such as solar and wind power, new energy sources are being integrated into the power system on a large scale. However, renewable energy sources are characterized by volatility, randomness, and intermittency, which can easily cause wide fluctuations in grid load. In the steam-water system of coal-fired units, precise control of main steam temperature is crucial for safe and reliable unit operation, extended equipment life, and economic benefits. Therefore, under conditions of frequent fluctuations in unit operating load, quickly and stably maintaining the main steam temperature within a reasonable range is a key factor in maintaining stable unit operation, ensuring power plant operational safety, and improving unit thermal efficiency.

[0003] In the main steam temperature control system of coal-fired units, superheated primary and secondary water sprays are typically used as a means of regulation, jointly achieving control and regulation functions. Currently, the main steam temperature control of most coal-fired units in my country adopts a PID cascade control system, which controls the main steam temperature within the rated range of the set value through the combined action of proportional, differential, and integral links. However, during peak operation of the unit, due to the influence of unstable factors such as large-scale load fluctuations, the PID controller has difficulty in timely and accurate correction of the dynamic deviations generated by the control system in a short period of time, resulting in dynamic overshoot of the superheated steam temperature. The main steam temperature fluctuates violently and can only be adjusted manually by the operator, which not only increases the workload of the operator but also increases human error.

[0004] Feedforward dynamic compensation of the spray water desuperheater valve opening is of great significance. It helps improve control performance, adapt to the system's dynamic characteristics, reduce manual intervention, and better cope with load fluctuations, thereby enhancing the performance and reliability of superheated steam temperature control systems in coal-fired units. A physically guided neural network (PGNN) was introduced to improve superheated steam temperature control. This effectively copes with load fluctuations, reduces the impact of disturbances on the main steam temperature, improves the main steam temperature's ability to follow the setpoint, and enhances control accuracy and stability. Furthermore, it adapts to the system's dynamic characteristics and load variations, achieving adaptive adjustments and improving control effectiveness. Furthermore, optimizing the spray water desuperheater valve opening through feedforward dynamic compensation reduces human error, improves automation, and enhances system stability and reliability.

[0005] In actual power plants, various operating parameters are affected by multiple factors, making their control more complex. Main steam, a critical parameter in the power system, is influenced by multiple subsystems. Predictive control using a multi-model strategy can overcome model mismatch and avoid disturbances introduced by multiple controller switching. Summary of the Invention

[0006] The present invention provides a PGNN feedforward dynamic compensation-multi-model predictive control method, whose technical purpose is to estimate the future state of the system through the predictive model, guide the controller to make adjustments in advance to reduce the main steam temperature deviation and overshoot, and achieve accurate and rapid response to load changes through feedforward compensation.

[0007] The above technical objectives of the present invention are achieved through the following technical solutions:

[0008] The first step is to establish a feedforward dynamic compensation logic module based on PGNN, including:

[0009] S1: Construct a main steam temperature prediction model based on PGNN. Use the physics-guided neural network modeling method to predict the main steam temperature trend in the future.

[0010] S2: Introduce the valve opening feedforward dynamic compensation module. The desired main steam temperature setting value is y set , the predicted main steam temperature is y t (k), the deviation between them is set as the compensation of valve opening, and the predicted deviation change trend is introduced into the output of the PID controller through feedforward dynamic compensation, so that the control structure can respond quickly to input changes, reduce tracking error and improve the control quality of superheated steam temperature.

[0011] The second step is to build a PGNN feedforward dynamic compensation-multi-model predictive control system based on the above modules, which specifically includes:

[0012] S3: A predictive controller based on a multi-model R / S switching strategy is adopted in the control system, and the system disturbance is applied to the water spray valve in advance by combining the PGNN feedforward dynamic compensation optimization model to reduce the system hysteresis.

[0013] Furthermore, the step P3 can be subdivided into:

[0014] S31: The predictive controller control link based on the multi-model R / S switching strategy overcomes model mismatch and avoids the interference introduced to the system by model switching, especially multi-controller switching.

[0015] S32: The first-stage superheater outlet temperature is structurally connected with other disturbance variables in the pilot zone, such as load and coal quantity, and then used as the input of the PGNN model in the pilot zone of the second-stage superheated steam temperature. The same control method is adopted to improve the regulation accuracy and control quality of the main steam temperature.

[0016] S33: Establish simulation models of the primary superheated steam temperature control system and the secondary control system. By optimizing the logic and the controller, eliminate the deviations caused by internal and external disturbances in the control system, maintain the stability of the control system, and enable the main steam temperature to more accurately follow its set value.

[0017] Finally, the performance of the PGNN feedforward dynamic compensation-multi-model predictive control method was tested. The optimization performance of the PGNN feedforward dynamic compensation-multi-model predictive control scheme in the present invention was tested by comparing the effects of PID cascade control and PGNN feedforward dynamic compensation-multi-model predictive control. The absolute deviation of the method used in the present invention is smaller than that of traditional PID cascade control, indicating that PGNN feedforward dynamic compensation-multi-model predictive control can more accurately track the main steam temperature to the set value, improving control stability.

[0018] The beneficial effects of the present invention are:

[0019] This invention utilizes a PGNN-based feedforward dynamic compensation and multi-model predictive control method, specifically designed for superheated steam temperature control systems. This method improves control system performance by utilizing PGNN predictive models, feedforward dynamic compensation, and multi-model predictive control. By accurately predicting the predictive model and optimizing valve opening feedforward dynamic compensation, the control system can better respond to changes in main steam temperature, thereby improving control quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a structural diagram of the PGNN feedforward dynamic compensation-multi-model predictive control system in the present invention;

[0021] Figure 2 This is a comparison chart of the PGNN feedforward dynamic compensation-multi-model predictive control and cascade PID control effects in the present invention. DETAILED DESCRIPTION

[0022] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific examples described herein are only used to explain the present invention and are not intended to limit the present invention.

[0023] like Figure 1As shown, the present invention provides a multi-model predictive control system based on multi-model gap measurement PGNN feedforward dynamic compensation. By combining the PGNN prediction model, feedforward dynamic compensation and multi-model predictive control technologies, the control system can better respond to changes in the main steam temperature, thereby improving the performance of the control system and achieving the purpose of improving control quality. set Ⅰ 、y set Ⅱ They are the set values ​​of the first-stage superheated steam outlet temperature and the second-stage superheated steam outlet temperature; PI2 Ⅰ and PI2 Ⅱ They are the controller functions of the primary superheater and the secondary superheater control loop respectively. The establishment of the predictive control system includes the following steps:

[0024] Step 1: Build a main steam temperature prediction model

[0025] A prediction model for main steam temperature is constructed using the PGNN neural network modeling method. A physics-guided neural network modeling method is used to predict the main steam temperature trend over a period of time.

[0026] Step 2: Introduce the valve opening feedforward dynamic compensation module

[0027] (1) The deviation between the main steam temperature prediction result obtained by the prediction model and the set value is used as compensation for the valve opening and introduced into the output of the PID controller, that is, the valve opening of the water spray desuperheater, through feedforward dynamic compensation. This achieves the effect of reducing tracking error.

[0028] (2) Use the main steam temperature prediction model and the expected main steam temperature setting value to calculate the feedforward dynamic compensation amount in order to take advance action on changes in factors such as load. The expected main steam temperature setting value is y set , the predicted main steam temperature is y t (k), the deviation between them is expressed as:

[0029] e F (k) = y set -y t (k)

[0030] (3) The feedforward dynamic compensation of the valve opening is calculated using the deviation to quickly respond to the change in main steam temperature caused by the change in unit load. The feedforward dynamic compensation of the valve opening Δu F (k) is expressed as:

[0031]

[0032] Where, and They are the proportional gain and integral gain of the valve opening feedforward dynamic compensation module, which need to be adjusted and optimized according to the actual situation of the system.

[0033] Step 3: Design of multi-model predictive controller based on R / S switching strategy

[0034] Predictive Functional Control (PFC) is a predictive control algorithm that emphasizes the impact of the controller's input structure on control system performance. By properly designing the controller's input structure, the system's characteristics and constraints can be fully utilized to improve the control system's response speed, stability, and robustness. The specific steps for implementing the algorithm are as follows:

[0035] (1) The prediction model of the predictive function controller is a first-order inertia plus pure lag link model. The model is represented by the following transfer function:

[0036]

[0037] Where K m is the gain of the transfer function; T m is the time delay constant; T d is the lag time constant; s is the complex frequency domain variable.

[0038] (2) For a first-order link without lag, the lag time constant T d Equal to 0, the continuous time first-order inertia link is discretized through the zero-order holder:

[0039] y m (k+1)=α m y m (k)+K m (1-α m )u(k)

[0040] Where y m (k+1) is the output at time k; u(k) is the output of the controller at time k; α m is the sampling period T r and the prediction function controls the time function of the reference trajectory.

[0041] (3) In predictive function control, the following discrete model can be used to establish a predictive model and predict the future response of the system. The mathematical expression is:

[0042]

[0043] Where P is the prediction time length.

[0044] (4) Over-regulating the control signal so that the error between the process output and the reference trajectory is minimized in the optimization time domain, and the minimization error function is expressed as:

[0045]

[0046] Where P1 and P2 are the upper and lower limits of the control time domain.

[0047] (5) Order The optimal control quantity can be obtained, that is, the upper and lower limits of the control time domain are equal P1=P2=P, and the control quantity u(k) is obtained, which is expressed as:

[0048]

[0049] Where c(k+P) is the set value of the main steam temperature system,

[0050] (6) When the lag time T d When it is not zero, the Smith predictor control concept can be used to improve the performance of the control system by correcting the system model output, which can be expressed as:

[0051] y pav (k) = y(k) + y m (k)-y m (kD)

[0052] Where D = T d / T m .

[0053]

[0054] (7) Replace y(k) in equation (5) with y pav (k) is replaced by the output u(k) of the PFC with a lag link.

[0055]

[0056] This is the output u(k) of the optimized controller at time k, which can minimize the error of the system in the future prediction time domain and make the output of the system as close as possible to the reference trajectory at future times.

[0057] Finally, through the above steps, a multi-model predictive control system based on multi-model gap measurement PGNN feedforward dynamic compensation can be established. Comparative analysis of the main steam temperature following of PGNN feedforward dynamic compensation-multi-model predictive control and traditional cascade PID control shows that PGNN feedforward dynamic compensation-multi-model predictive control has better performance in main steam temperature control than PID cascade control. Figure 2It can be seen that compared with the control effect of the traditional PID of the original control system, the combined effect of the valve opening feedforward dynamic compensation logic and the multi-model predictive controller can improve the performance of the control system, eliminate deviations, maintain stability, and enable the main steam temperature to follow the set value more accurately, thereby achieving better control effects.

[0058] The above is only a preferred embodiment of the present invention, and other effective embodiments are all possible. For other technicians in the same technical field, if effective improvements are proposed based on the present invention, these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A PGNN feedforward dynamic compensation-multi-model predictive control method, characterized in that: The following steps are involved: Step 1: Establish a feedforward dynamic compensation logic module based on PGNN; Step 2: Construct a PGNN feedforward dynamic compensation-multi-model predictive control system based on the above modules; The first step comprises: S1: Construct a main steam temperature prediction model based on PGNN; use the physics-guided neural network modeling method to predict the main steam temperature change trend in the future; S2: Introduce the valve opening feedforward dynamic compensation module; the desired main steam temperature setting value is y set , the predicted main steam temperature is y t (k) The deviation between them is set as the compensation of valve opening, and the predicted deviation change trend is introduced into the output of the PID controller through feedforward dynamic compensation, so that the control structure can respond quickly to input changes, reduce tracking error and improve the control quality of superheated steam temperature; The step S2 specifically includes: (1) The deviation between the main steam temperature prediction result obtained by the prediction model and the set value is used as the compensation for the valve opening and introduced into the output of the PID controller, i.e., the valve opening of the water spray desuperheater, through the feedforward dynamic compensation method, to achieve the effect of reducing the tracking error; (2) The feedforward dynamic compensation is calculated using the main steam temperature prediction model and the expected main steam temperature set value, in order to take advance action on the change of load factors; the expected main steam temperature set value is y set , the predicted main steam temperature is y t (k), the deviation between them is expressed as: and F (k)=y set -and t (k) (3) The feedforward dynamic compensation of the valve opening is calculated using the deviation to quickly respond to the change in main steam temperature caused by the change in unit load. The feedforward dynamic compensation of the valve opening Δu F (k) is expressed as: Where, and They are proportional gain and integral gain of valve opening feedforward dynamic compensation module respectively; The second step includes: S3: adopting a predictive controller control link based on a multi-model R / S switching strategy in the control system, and combining the PGNN feedforward dynamic compensation optimization model to act on the water spray valve in advance to reduce the hysteresis of the system; said step S3 includes: S31: A predictive controller based on a multi-model R / S switching strategy overcomes model mismatch and avoids interference introduced by multi-controller switching to the system. S32: The first-stage superheater outlet temperature is structurally connected with other disturbance variables in the pilot zone, including load and coal quantity, and then used as the input of the PGNN model in the pilot zone of the second-stage superheated steam temperature. The same control method is adopted to improve the regulation accuracy and control quality of the main steam temperature. S33: Establish simulation models of the primary superheated steam temperature control system and the secondary control system. By optimizing the logic and the controller, eliminate the deviations caused by internal and external disturbances in the control system, maintain the stability of the control system, and enable the main steam temperature to more accurately follow its set value.

2. A PGNN feedforward dynamic compensation-multi-model predictive control method according to claim 1, characterized in that: The step S3 specifically includes: (1) The prediction model of the predictive function controller is a first-order inertia plus pure lag link model; the model is represented by the following transfer function: Where K m is the gain of the transfer function; T m is the time delay constant; T d is the lag time constant; s is the complex frequency domain variable; (2) For a first-order link without lag, the lag time constant T d Equal to 0, the continuous time first-order inertia link is discretized through the zero-order holder: y m (k+1)=a m y m (k)+K m (1-a m )u(k) Where y m (k+1) is the output at time k; u(k) is the output of the controller at time k; α m is the sampling period T r and the prediction function controls the time function of the reference trajectory; (3) In predictive function control, the following discrete model is used to establish a predictive model and predict the future response of the system. The mathematical expression is: Where P is the prediction time length; (4) Over-regulating the control signal so that the error between the process output and the reference trajectory is minimized in the optimization time domain, and the minimization error function is expressed as: Where P1 and P2 are the upper and lower limits of the control time domain; (5) Order The optimal control quantity is obtained, that is, the upper and lower limits of the control time domain are equal to P1=P2=P, and the control quantity u(k) is obtained, which is expressed as: Where c(k+P) is the set value of the main steam temperature system, (6) When the lag time T d When it is not zero, the Smith predictor control idea is adopted to improve the performance of the control system by correcting the system model output, which is expressed as: and pav (k)=y(k)+y m (k)-y m (kD) Where D = T d / T m ; (7) Replace y(k) in (5) with y pav (k) is replaced by the output u(k) of the PFC with a lag link; This is to optimize the output u(k) of the controller at time k, so that the system can minimize the error in the future prediction time domain and make the output of the system as close as possible to the reference trajectory at future times.

Images