Coal-fired unit main steam temperature system GPC-PID cascade control method based on data driving

Abstract

The invention discloses a GPC-PID cascade control method for a main steam temperature system of a coal-fired unit based on data driving, which is based on a data driving model and is combined with an implicit generalized predictive control algorithm fused with a stepped thought to realize better main steam temperature control. The method comprises two control processes, in the main control process, a compensation instruction of a main steam temperature set value is obtained by establishing a main steam temperature prediction model; secondly, calculating a desuperheater outlet temperature set value instruction increment according to the desuperheater outlet temperature set value at the previous moment, the main steam temperature and the compensated main steam temperature set value through an implicit generalized predictive control algorithm; in the auxiliary control process, a desuperheater control valve opening degree instruction is obtained through PID calculation according to the obtained desuperheater outlet temperature set value instruction and the desuperheater outlet temperature. According to the method, the main steam temperature control effect can be optimized for a main steam temperature control system with strong nonlinearity and uncertainty, and the method has important significance for improving the rapid load changing capacity of a coal-fired unit.

Description

Technical Field

[0001] The present invention belongs to the technical field of thermal power system control optimization, and in particular relates to a data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit. Background Art

[0002] As my country's energy system undergoes transformation and upgrades, and to accommodate more renewable energy sources like solar and wind power, coal-fired power plants, as the ballast for a secure and stable energy supply, are required to undertake more peak-shaving and frequency-regulating tasks to ensure the safe and stable operation of the power grid. Consequently, coal-fired power plants are required to operate more frequently with rapid load changes and a wide range of loads.

[0003] The quality of main steam temperature control has become a significant factor limiting the ability of coal-fired power generation thermal power systems to further improve their variable load capabilities. During rapid load changes, timely control of main steam temperature and ensuring its stability are critical issues that need to be addressed. Summary of the Invention

[0004] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and provide a data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit, so as to optimize the water spray cooling control of the main steam temperature system of the coal-fired unit and achieve better main steam temperature control.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit is characterized in that: obtaining the desuperheater valve opening instruction of the main steam temperature system of the coal-fired unit includes two parts: a main control process for coarse adjustment and a sub-control process for fine adjustment;

[0007] The main control process generates the desuperheater outlet temperature setpoint command, which is mainly obtained by the previous desuperheater outlet temperature setpoint, main steam temperature, main steam temperature setpoint and main steam temperature prediction value, and is calculated by the generalized predictive control algorithm (GPC):

[0008] Step 1: Calculate the compensation instruction of the main steam temperature setting value;

[0009] The main steam temperature prediction value is used to calculate the variation characteristics of the main steam temperature and generate compensation for the main steam temperature set value;

[0010] The main steam temperature prediction value is obtained by the long short-term memory network prediction model integrated with the attention mechanism. First, data preprocessing is performed to obtain the time series data of the parameters affecting the main steam temperature, and the erroneous values ​​are eliminated. The time series data with non-equal time intervals and default values ​​are resampled using the piecewise cubic spline interpolation method to obtain equal time interval data for modeling. Secondly, feature screening is performed. The data features are screened using the principal component analysis method, and the features with a high impact on the main steam temperature are retained as model inputs. Then, the delay of the input parameters is estimated using the mutual information method, and the parameters are aligned to avoid the reduction in prediction accuracy caused by the delay time. Finally, training is performed and the prediction accuracy of the model is evaluated. The processed input and output are used to establish the long short-term memory network prediction model integrated with the attention mechanism for main steam temperature prediction.

[0011] The compensation command of the main steam temperature set point is obtained by the following formula:

[0012]

[0013] Where: F1(x) is the compensation instruction for the main steam temperature setting value; k1 is the calculation coefficient; ΔT s The change in the predicted value of the main steam temperature, ℃; Δt * is the sampling time interval, s;

[0014] The main steam temperature set value and the compensation instruction of the main steam temperature set value are summed up to serve as the set value of the actual main control process;

[0015] Step 2: Calculate the initial control increment of the desuperheater outlet temperature set point output by the main control process;

[0016] The initial control increment of the desuperheater outlet temperature setpoint is obtained from the previous desuperheater outlet temperature setpoint, the main steam temperature, and the actual main steam temperature setpoint. First, the desuperheater outlet temperature setpoint and the main steam temperature at the previous time are used to identify the parameters online using the recursive least squares method:

[0017]

[0018] Where: is the vector of parameters to be identified at the current moment, g n-1 ,g n-2 ,…,g0 is the identified parameter, f(t) is the initial value predicted at the current moment; It is the vector of parameters to be identified at the previous moment. The initial value is usually K(t) is a constant matrix, usually K(t)=10 6I, I are the unit matrices; y(t) is the actual SCR outlet NOx concentration at the current moment; n is the maximum prediction length; X(tn) is the vector consisting of the ammonia injection amount command increments at the previous n moments; P(t) and P(t-1) are the constant matrices at the current moment and the previous moment, respectively; λ1 is the forgetting factor;

[0019] Then, using the previous desuperheater outlet temperature setpoint, main steam temperature, actual main steam temperature setpoint, and online identified parameters, an implicit generalized predictive control algorithm incorporating a step-by-step approach is used to calculate the initial control increment for the desuperheater outlet temperature setpoint. The calculation of the initial control increment is to solve the following quadratic programming problem:

[0020]

[0021] stCΔu m (t)≤l

[0022] Where: J is the objective function; m is the control length; Δu m (t) is the control increment vector of the control length at the current moment; t is the current moment; j is the sequence number; y(t+j) is the output at t+j; ω(t+j) is the reference trajectory at t+j; λ(j) is the control weighting coefficient at t+j; Δu(t+j-1) is the control increment at t+j-1; C and l are both coefficient matrices; Finally, take the calculated Δu m The first term Δu(t) is the difference between the initial control increment and the desuperheater outlet temperature setting value T at the previous moment. jw-set (t-1) and obtain the desuperheater outlet temperature setting value instruction T output by the main control process at the current moment jw-set (t);

[0023] Finally, the calculated initial control command increment is summed with the desuperheater outlet temperature setting value at the previous moment to obtain the desuperheater outlet temperature setting value command output by the main control process at the current moment;

[0024] The auxiliary control process generates the desuperheater valve opening instruction, and the main control process generates the desuperheater outlet temperature setting value instruction T jw-set (t) and the desuperheater outlet temperature are calculated using the incremental PID control law:

[0025] e jw =T jw-set -T jw

[0026] Δu jw (t) = K p *(e jw (t)-e jw (t-1))+Ki *e jw (t)

[0027] +K d *(e jw (t)-2*e jw (t-1)+e jw (t-2))

[0028] u jw (t)=Δu jw (t)+u jw (t-1)

[0029] Where: e jw is the desuperheater outlet temperature deviation; T jw-set is the desuperheater outlet temperature setting value, ℃; T jw is the outlet temperature of the desuperheater, °C; Δu jw (t) is the desuperheater valve opening control instruction increment at the current moment; K p is the proportional coefficient; e jw (t) is the desuperheater outlet temperature deviation at the current moment; e jw (t-1) is the desuperheater outlet temperature deviation at the previous moment; K i is the integral coefficient; K d is the differential coefficient; e jw (t-2) is the outlet temperature deviation of the desuperheater at the previous moment; u jw (t) is the desuperheater valve opening instruction at the current moment; u jw (t-1) is the desuperheater valve opening instruction at the previous moment.

[0030] Preferably, when establishing a long short-term memory network prediction model that integrates an attention mechanism for main steam temperature prediction, the parameter characteristics involved in the water spray cooling process that directly affects the main steam temperature of the coal-fired unit are considered, including the cooling water volume and the steam temperature at the water spray cooling inlet. The primary air volume, secondary air volume, and coal feed rate parameter characteristics involved in the boiler combustion process that affect the temperature are also considered. Based on these direct or indirect parameter characteristics, relevant operating data are collected, and these data are required to be time series data with equal time intervals. A training data set is constructed based on these data. This method simultaneously collects the water spray cooling parameter characteristics that directly affect the main steam temperature and the combustion process parameter characteristics that indirectly determine the heat source input, ensuring that the training data can both characterize the rapid response characteristics of the cooling water regulation and reflect the lagged effects of the heat load changes on the combustion side, thereby enhancing the generalization ability and engineering interpretability of the prediction model from the source.

[0031] The main control process integrates the step-by-step approach when calculating the output initial control increment and solving the quadratic programming problem. The input form is rewritten as follows:

[0032]

[0033] Where: Δu(t) is the control increment at the current moment; δ is the control increment at the current moment; Δu(t+i) is the control increment at the i-th moment in the future; β is the step factor;

[0034] Therefore, the input sequence of controlled length is transformed into the following vector form:

[0035] Δu m =[1,β1,...,β m-1 ]δ

[0036] This solution constructs the input sequence as a stepped input, which can effectively suppress high-frequency control jitter and enhance system robustness while ensuring dynamic response performance. It can also significantly reduce the complexity of solving problems and the real-time computing burden.

[0037] The initial control increment of the main control process considers the following constraints:

[0038] Δu min ≤Δu≤Δu max

[0039] u min ≤u≤u max

[0040]

[0041] Where: Δu min and Δu max are the minimum and maximum values ​​of the initial control instruction increment of the desuperheater outlet temperature setting value; u min and u max are the minimum and maximum values ​​of the desuperheater outlet temperature setting value instruction respectively; y min and y max are the maximum and minimum main steam temperature respectively; is the main steam temperature prediction value calculated by the implicit generalized predictive control algorithm, which is calculated by the following formula:

[0042]

[0043] Where: is a vector matrix composed of the main steam temperature output prediction values ​​at different times, where and The output prediction values ​​of the main steam temperature corresponding to time t+1, t+2 and t+n respectively; ΔU = [Δu(t), Δu(t+1), …, Δu(t+n-1)] Tis a vector matrix composed of the desuperheater outlet temperature setpoint inputs at different times, where Δu(t), Δu(t+1), and Δu(t+n-1) correspond to the desuperheater outlet temperature setpoint inputs at times t, t+1, and t+n-1, respectively;

[0044] Among them, G is the matrix composed of identification parameters, as shown below:

[0045]

[0046] f is the prediction initial value matrix. The output prediction value at the previous moment is used for feedback correction to obtain the prediction initial value at the current moment, which is calculated by the following formula:

[0047]

[0048] Where: e(t) is the prediction error at the current time t; y(t) is the main steam temperature output at the current time t; represents the output prediction value of the previous time t-1 for the current time t, and so on; f(t), f(t+1) and f(t+n-1) are the initial prediction values ​​of the corresponding time; h1, h2 and h n is the error correction coefficient;

[0049] The above constraints are written into the quadratic programming problem for solving the initial control increment. The constraints of the quadratic programming are specifically:

[0050]

[0051] Where: A is a unit matrix; B is a lower triangular matrix; Y min Y is the vector matrix of the output lower limit value; max The vector matrix of the output upper limit value; Δu min,0 The vector matrix formed by the lower limit of the control increment; Δu max,0 The vector matrix formed by the upper limit of the control increment; Δu min is the vector matrix composed of the lower limit of the control quantity and the control quantity at the previous moment; Δu max is a vector matrix consisting of the upper limit of the control variable and the control variable at the previous moment. This scheme constrains the input and output characteristics of the control system and integrates the constraints into the quadratic programming problem of solving the control increment. This effectively solves the problem of over-limit of the control variable and the controlled variable in the control process and ensures the stability of the control system.

[0052] The desuperheater valve opening command calculated by the secondary control process satisfies the following relationship:

[0053] u jw-min <u jw <u jw-max

[0054] Where: u jw It is the desuperheater valve opening instruction output by the sub-control process; u jw-min is the lower limit of the desuperheater valve opening; u jw-max The upper limit of the desuperheater valve opening.

[0055] Compared with the prior art, the present invention has the following advantages:

[0056] (1) The present invention generates a dynamic compensation instruction for the main steam temperature set point in advance by predicting the main steam temperature. As a feedforward control function, it acts on the set point of the main control process in advance before disturbances such as fuel quantity changes and load fluctuations actually affect the controlled temperature, enabling the system to respond to disturbances more promptly and proactively.

[0057] (2) The present invention solves the time-varying problem of parameters in the main steam temperature control system through online identification;

[0058] (3) The present invention adopts advanced control to deal with the main disturbance and set value tracking problems with large lag and strong coupling in the main control process, and adopts PID control in the secondary control process, which retains the advantages of fast response and rapid suppression of internal disturbances, thereby optimizing the main steam temperature control effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is the logic diagram of the GPC-PID cascade control strategy for the main steam temperature after optimization in the present invention.

[0060] Figure 2 This is the main steam temperature prediction modeling flow chart of the present invention.

[0061] Figure 3 (a) and (b) are comparison results of the average and standard deviation of the main steam temperature under the control of the method proposed by the present invention and the traditional method, respectively. DETAILED DESCRIPTION

[0062] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0063] The present invention proposes a data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit. The specific implementation method is as follows:

[0064] like Figure 1 As shown, the acquisition of the desuperheater valve opening instruction of the main steam temperature system of the coal-fired unit includes two parts, namely the main control process S01 for coarse adjustment and the sub-control process S02 for fine adjustment;

[0065] The main control process S01 generates the desuperheater outlet temperature set value instruction, which is mainly obtained by the previous desuperheater outlet temperature set value, main steam temperature, main steam temperature set value and main steam temperature prediction value, and is calculated by the generalized predictive control algorithm (GPC):

[0066] Step 1: Calculate the compensation instruction for the main steam temperature set point.

[0067] The main steam temperature prediction value is used to calculate the variation characteristics of the main steam temperature and generate compensation for the main steam temperature set value;

[0068] The main steam temperature prediction value is obtained by the long short-term memory network prediction model integrated with the attention mechanism; the modeling process of the prediction model is as follows Figure 2 As shown in the figure, firstly, data preprocessing is performed to obtain the time series data of the parameters affecting the main steam temperature, and the erroneous values ​​are eliminated. The time series data with non-equal time intervals and default values ​​are resampled by piecewise cubic spline interpolation method to obtain equal time interval data for modeling. Secondly, feature screening is performed. The data features are screened by principal component analysis method, and the features with high influence on the main steam temperature are retained as model input. Then, the delay of the input parameters is estimated by mutual information method, and the parameters are aligned to avoid the reduction of prediction accuracy caused by the delay time. Finally, training is performed and the prediction accuracy of the model is evaluated. The processed input and output are used to establish a long short-term memory network prediction model with integrated attention mechanism for the prediction of the main steam temperature.

[0069] The compensation command of the main steam temperature set point is obtained by the following formula:

[0070]

[0071] Where: F1(x) is the compensation instruction for the main steam temperature setting value; k1 is the calculation coefficient; ΔT s The change in the predicted value of the main steam temperature, ℃; Δt * is the sampling time interval, s;

[0072] The main steam temperature setting value and the compensation instruction of the main steam temperature setting value are summed and used as the setting value of the actual main control process.

[0073] Step 2: Calculate the initial control increment of the desuperheater outlet temperature set point output by the main control process.

[0074] The initial control increment of the desuperheater outlet temperature setpoint is obtained from the previous desuperheater outlet temperature setpoint, the main steam temperature, and the actual main steam temperature setpoint. First, the desuperheater outlet temperature setpoint and the main steam temperature at the previous time are used to identify the parameters online using the recursive least squares method (RLS):

[0075]

[0076] Where: is the vector of parameters to be identified at the current moment, g n-1 ,g n-2 ,…,g0 is the identified parameter, f(t) is the initial value predicted at the current moment; It is the vector of parameters to be identified at the previous moment. The initial value is usually K(t) is a constant matrix, usually K(t)=10 6 I, I are the unit matrices; y(t) is the actual SCR outlet NOx concentration at the current moment; n is the maximum prediction length; X(tn) is the vector consisting of the ammonia injection amount command increments at the previous n moments; P(t) and P(t-1) are the constant matrices at the current moment and the previous moment, respectively; λ1 is the forgetting factor;

[0077] Then, using the previous desuperheater outlet temperature setpoint, main steam temperature, actual main steam temperature setpoint, and online identified parameters, an implicit generalized predictive control (GPC) algorithm incorporating a step-by-step approach was used to calculate the initial control increment for the desuperheater outlet temperature setpoint. The calculation of the initial control increment involves solving the following quadratic programming problem:

[0078]

[0079] sC Δ u m (t)≤l

[0080] Where: J is the objective function; m is the control length; Δu m (t) is the control increment vector of the control length at the current moment; t is the current moment; j is the sequence number; y(t+j) is the output at t+j; ω(t+j) is the reference trajectory at t+j; λ(j) is the control weighting coefficient at t+j; Δu(t+j-1) is the control increment at t+j-1; C and l are both coefficient matrices; Finally, take the calculated Δu m The first term Δu(t) is the difference between the initial control increment and the desuperheater outlet temperature setting value T at the previous moment. jw-set (t-1) and obtain the desuperheater outlet temperature setting value instruction T output by the main control process at the current moment jw-set (t);

[0081] Finally, the calculated initial control instruction increment is summed with the desuperheater outlet temperature setting value at the previous moment to obtain the desuperheater outlet temperature setting value instruction output by the main control process S01 at the current moment.

[0082] The auxiliary control process S02 generates the desuperheater valve opening instruction, and the desuperheater outlet temperature setting value instruction T generated by the main control process jw-set (t) and the desuperheater outlet temperature are calculated using the incremental PID control law:

[0083] e jw =T jw-set -T jw

[0084] Δu jw (t) = K p *(e jw (t)-e jw (t-1))+K i *e jw (t)

[0085] +K d *(e jw (t)-2*e jw (t-1)+e jw (t-2))

[0086] u jw (t)=Δu jw (t)+u jw (t-1)

[0087] Where: e jw is the desuperheater outlet temperature deviation; T jw-set is the desuperheater outlet temperature setting value, ℃; T jw is the outlet temperature of the desuperheater, °C; Δu jw (t) is the desuperheater valve opening control instruction increment at the current moment; K p is the proportional coefficient; e jw (t) is the desuperheater outlet temperature deviation at the current moment; e jw (t-1) is the desuperheater outlet temperature deviation at the previous moment; K i is the integral coefficient; K d is the differential coefficient; e jw (t-2) is the outlet temperature deviation of the desuperheater at the previous moment; u jw (t) is the desuperheater valve opening instruction at the current moment; u jw (t-1) is the desuperheater valve opening instruction at the previous moment.

[0088] The present invention provides a detailed calculation method for the control instruction of the desuperheater valve opening of the main steam temperature control system. The main steam temperature prediction value based on the data-driven model is added to the main control loop to compensate the main steam temperature setting value, and a generalized predictive control algorithm integrating the step-by-step thinking is adopted. At the same time, the constraint influence of the input and output is considered to calculate the instruction of the desuperheater outlet temperature setting value output by the main control loop. Then, the desuperheater opening instruction is calculated by the incremental PID algorithm in the sub-control loop to achieve fine adjustment of the main steam temperature. Benefiting from the advantages of advanced control algorithms in processing main steam temperature control systems with large inertia and large delay, and integrating the step-by-step thinking and corresponding constraints, the sudden change of the control instruction is avoided, and the corresponding feedforward is considered. Finally, the control effect of the main steam temperature is optimized. Figure 3 (a) and (b) are comparisons of the statistical results of the average and standard deviation of the main steam temperature under the control of the method proposed in the present invention and the traditional cascade PID method during the load change process. It can be seen that compared with the traditional method, the average value of the method proposed in the present invention is closer to the set value of the main steam temperature, and the standard deviation under the control of the method proposed in the present invention is smaller, and the discrete degree of the corresponding data is smaller, indicating that the main steam temperature fluctuation degree under the method of the present invention is smaller, and the overall control of the main steam temperature is more stable. It also shows that the target tracking performance of the control method mentioned in the present invention is excellent, which can effectively suppress the fluctuation of the main steam temperature of the coal-fired unit under frequent load changes, stably track the set value, and ensure the stability of the main steam temperature, which plays an important role in improving the operation flexibility of the unit.

Claims

1. A data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit, characterized by: The acquisition of the desuperheater valve opening command for the main steam temperature system of a coal-fired unit consists of two parts: the main control process for coarse adjustment and the sub-control process for fine adjustment. The main control process generates the desuperheater outlet temperature set value instruction, which is mainly obtained by combining the previous desuperheater outlet temperature set value, main steam temperature, main steam temperature set value and main steam temperature prediction value with the generalized predictive control algorithm GPC: Step 1: Calculate the compensation instruction of the main steam temperature setting value; The main steam temperature prediction value is used to calculate the variation characteristics of the main steam temperature and generate compensation for the main steam temperature set value; The main steam temperature prediction value is obtained by the long short-term memory network prediction model integrated with the attention mechanism. First, data preprocessing is performed to obtain the time series data of the parameters affecting the main steam temperature, and the erroneous values ​​are eliminated. The time series data with non-equal time intervals and default values ​​are resampled using the piecewise cubic spline interpolation method to obtain equal time interval data for modeling. Secondly, feature screening is performed. The data features are screened using the principal component analysis method, and the features with a high impact on the main steam temperature are retained as model inputs. Then, the delay of the input parameters is estimated using the mutual information method, and the parameters are aligned to avoid the reduction in prediction accuracy caused by the delay time. Finally, training is performed and the prediction accuracy of the model is evaluated. The processed input and output are used to establish the long short-term memory network prediction model integrated with the attention mechanism for main steam temperature prediction. The compensation command of the main steam temperature set point is obtained by the following formula: Where: F1(x) is the compensation instruction for the main steam temperature setting value; k1 is the calculation coefficient; ΔT s The change in the predicted value of the main steam temperature, ℃; Δt * is the sampling time interval, s; The main steam temperature set value and the compensation instruction of the main steam temperature set value are summed up to serve as the set value of the actual main control process; Step 2: Calculate the initial control increment of the desuperheater outlet temperature set point output by the main control process; The initial control increment of the desuperheater outlet temperature setpoint is obtained from the previous desuperheater outlet temperature setpoint, the main steam temperature, and the actual main steam temperature setpoint. First, the desuperheater outlet temperature setpoint and the main steam temperature at the previous time are used to identify the parameters online using the recursive least squares method: Where: is the vector of parameters to be identified at the current moment, g n-1 ,g n-2 ,…,g0 is the identified parameter, f(t) is the initial value predicted at the current moment; It is the vector of parameters to be identified at the previous moment. The initial value is usually K(t) is a constant matrix, usually K(t)=10 6 I, I are the unit matrices; y(t) is the actual SCR outlet NOx concentration at the current moment; n is the maximum prediction length; X(tn) is the vector consisting of the ammonia injection amount command increments at the previous n moments; P(t) and P(t-1) are the constant matrices at the current moment and the previous moment, respectively; λ1 is the forgetting factor; Then, using the previous desuperheater outlet temperature setpoint, main steam temperature, actual main steam temperature setpoint, and online identified parameters, an implicit generalized predictive control algorithm incorporating a step-by-step approach is used to calculate the initial control increment for the desuperheater outlet temperature setpoint. The calculation of the initial control increment is to solve the following quadratic programming problem: Where: J is the objective function; m is the control length; Δu m (t) is the control increment vector of the control length at the current moment; t is the current moment; j is the sequence number; y(t+j) is the output at t+j; ω(t+j) is the reference trajectory at t+j; λ(j) is the control weighting coefficient at t+j; Δu(t+j-1) is the control increment at t+j-1; C and l are both coefficient matrices; Finally, take the calculated Δu m The first term Δu(t) is the difference between the initial control increment and the desuperheater outlet temperature setting value T at the previous moment. jw-set (t-1) and obtain the desuperheater outlet temperature setting value instruction T output by the main control process at the current moment jw-set (t); Finally, the calculated initial control command increment is summed with the desuperheater outlet temperature setting value at the previous moment to obtain the desuperheater outlet temperature setting value command output by the main control process at the current moment; The auxiliary control process generates the desuperheater valve opening instruction, and the main control process generates the desuperheater outlet temperature setting value instruction T jw-set (t) and the desuperheater outlet temperature are calculated using the incremental PID control law: e jw =T jw-set -T jw Δu jw (t)=K p *(e jw (these jw (t-1))+K i *e jw (t)+K d *(e jw (t)-2*e jw (t-1)+e jw (t-2)) u jw (t)=Δu jw (t)+u jw (t-1) Where: e jw is the desuperheater outlet temperature deviation; T jw-set is the desuperheater outlet temperature setting value, ℃; T jw is the outlet temperature of the desuperheater, °C; Δu jw (t) is the desuperheater valve opening control instruction increment at the current moment; K p is the proportional coefficient; e jw (t) is the desuperheater outlet temperature deviation at the current moment; e jw (t-1) is the desuperheater outlet temperature deviation at the previous moment; K i is the integral coefficient; K d is the differential coefficient; e jw (t-2) is the outlet temperature deviation of the desuperheater at the previous moment; u jw (t) is the desuperheater valve opening instruction at the current moment; u jw (t-1) is the desuperheater valve opening instruction at the previous moment.

2. A data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit according to claim 1, characterized in that: When establishing a long short-term memory network prediction model with an integrated attention mechanism for main steam temperature prediction, the parameter characteristics involved in the water spray cooling process that directly affects the main steam temperature of the coal-fired unit are considered, including the cooling water volume and the steam temperature at the water spray cooling inlet. The primary air volume, secondary air volume, and coal feed rate parameter characteristics involved in the boiler combustion process that affect the temperature are also considered. Based on these direct or indirect parameter characteristics, relevant operating data are collected, and these data are required to be time series data with equal time intervals. A training dataset is constructed based on these data.

3. The data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit according to claim 1, characterized in that: The main control process integrates the step-by-step approach when calculating the output initial control increment and solving the quadratic programming problem. The input form is rewritten as follows: Where: Δu(t) is the control increment at the current moment; δ is the control increment at the current moment; Δu(t+i) is the control increment at the i-th moment in the future; β is the step factor; Therefore, the input sequence of controlled length is transformed into the following vector form: Thu m =[1,β1,...,β m-1 ]d.

4. The data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit according to claim 1, characterized in that: The initial control increment of the main control process considers the following constraints: Δu min ≤Δu≤Δu max in min Oh, oh. max Where: Δu min and Δu max are the minimum and maximum values ​​of the initial control instruction increment of the desuperheater outlet temperature setting value; u min and u max are the minimum and maximum values ​​of the desuperheater outlet temperature setting value instruction respectively; y min and y max are the maximum and minimum main steam temperature respectively; is the main steam temperature prediction value calculated by the implicit generalized predictive control algorithm, which is calculated by the following formula: Where: is a vector matrix composed of the main steam temperature output prediction values ​​at different times, where and The output prediction values ​​of the main steam temperature corresponding to time t+1, t+2 and t+n respectively; ΔU = [Δu(t), Δu(t+1), …, Δu(t+n-1)] T is a vector matrix composed of the desuperheater outlet temperature setpoint inputs at different times, where Δu(t), Δu(t+1), and Δu(t+n-1) correspond to the desuperheater outlet temperature setpoint inputs at times t, t+1, and t+n-1, respectively; Among them, G is the matrix composed of identification parameters, as shown below: f is the prediction initial value matrix. The output prediction value at the previous moment is used for feedback correction to obtain the prediction initial value at the current moment, which is calculated by the following formula: Where: e(t) is the prediction error at the current time t; y(t) is the main steam temperature output at the current time t; represents the output prediction value of the previous time t-1 for the current time t, and so on; f(t), f(t+1) and f(t+n-1) are the initial prediction values ​​of the corresponding time; h1, h2 and h n is the error correction coefficient; The above constraints are written into the quadratic programming problem for solving the initial control increment. The constraints of the quadratic programming are specifically: Where: A is a unit matrix; B is a lower triangular matrix; Y min Y is the vector matrix of the output lower limit value; max The vector matrix of the output upper limit value; Δu min,0 The vector matrix formed by the lower limit of the control increment; Δu max,0 The vector matrix formed by the upper limit of the control increment; Δu min is the vector matrix composed of the lower limit of the control quantity and the control quantity at the previous moment; Δu max It is a vector matrix composed of the upper limit of the control amount and the control amount at the previous moment.

5. The data-driven GPC-PID cascade control method for the main steam temperature system of a coal-fired unit according to claim 1, characterized in that: The desuperheater valve opening command calculated by the secondary control process satisfies the following relationship: in jw-min <in jw <in jw-max Where: u jw It is the desuperheater valve opening instruction output by the sub-control process; u jw-min is the lower limit of the desuperheater valve opening; u jw-max The upper limit of the desuperheater valve opening.

Images