Adaptive frequency modulation method and system for variable-parameter power system

Abstract

The invention discloses a variable parameter power system adaptive frequency modulation method and system. The method comprises the steps that according to a variable-parameter thermal power generating unit model, inertia changes of a power system are considered, and a wide-load frequency modulation model under flexible operation of a thermal power generating unit is constructed; constructing a simulation data set according to the wide-load frequency modulation model, training a long-short term memory network Attention-LSTM combined with an attention mechanism, and constructing an Attention-LSTM-based frequency modulation model; and according to the Attention-LSTM-based frequency modulation model, predicting power grid frequency dynamic response characteristics, taking the power grid frequency dynamic response characteristics as an MPC prediction model, designing an optimal objective function, and solving an optimization problem by adopting a differential evolution algorithm to realize rolling optimization. According to the method, the problem of model mismatch caused by model parameter change under wide-load operation of the thermal power generating unit is solved, and the reliability and robustness in the power grid frequency modulation process are enhanced; by combining inertia prediction, the precision of frequency prediction is improved, and the response speed of a power system to frequency change is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for adaptive frequency modulation of a variable parameter power system based on machine learning model predictive control. Background Art

[0002] As the demand for thermal power units to participate in the deep peak regulation of the power system increases, thermal power units are not limited to high-load operation, but also need to operate stably under low-load conditions. The wide load operation of thermal power units causes changes in model parameters. It is difficult to accurately evaluate the system frequency regulation capability using traditional fixed parameter models, which reduces the system frequency regulation performance. At the same time, with the increase in the proportion of renewable energy, the proportion of traditional synchronous generators decreases, resulting in a decrease in the overall inertia of the system. Wind and solar energy have strong volatility and unpredictability, which makes the power system face greater challenges in the frequency regulation process.

[0003] To solve the above problems, existing studies have adopted a composite modeling method that combines mechanism modeling and data identification to establish a nonlinear control model of a subcritical unit, which has improved the system frequency regulation performance to a certain extent. However, the composite modeling method usually involves a combination of multiple models, which increases the complexity of the system. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a variable parameter power system adaptive frequency regulation method and system based on machine learning model predictive control to reduce the impact of model parameter changes caused by wide load operation of thermal power units and system inertia changes caused by uncertainty in new energy penetration on frequency regulation performance.

[0005] In order to solve the above problems, in a first aspect, the present invention provides a variable parameter power system adaptive frequency regulation method based on machine learning model predictive control, comprising:

[0006] Based on the variable parameter thermal power unit model, considering the change of power system inertia, a wide load frequency regulation model under flexible operation of thermal power units is constructed;

[0007] A simulation data set is constructed according to the wide load frequency modulation model, a long short-term memory network Attention-LSTM combined with an attention mechanism is trained, and a frequency modulation model based on Attention-LSTM is constructed;

[0008] The frequency modulation model based on Attention-LSTM is used to predict the dynamic response characteristics of the power grid frequency. This model is used as the MPC prediction model. The controlled output is considered to be close to the reference input, while the control action does not change too much. The optimal objective function is designed, and the differential evolution algorithm is used to solve the optimization problem to achieve rolling optimization.

[0009] Furthermore, the input and output relationship of the frequency modulation model based on Attention-LSTM is as follows:

[0010] [ΔX g (k+1);ΔP r (k+1); ΔP g (k+1);Δf(k+1)]=

[0011] f(Δu(k),P g (k),ΔX g (k),ΔP r (k),ΔP g (k),Δf(k))

[0012] Where ΔX g (k) is the change in the governor valve position at time k, ΔP r (k) is the output power change of the reheat boiler at time k, ΔP g (k) is the change in output power of the steam turbine at time k, P g (k) is the real-time output power of the steam turbine, Δf(k) is the frequency deviation change, Δu(k) is the control signal increment at time k, ΔX g (k+1),ΔP r (k+1),ΔP g (k+1) and Δf(k+1) are the changes of the corresponding variables at time k+1.

[0013] Furthermore, based on the constructed Attention-LSTM-based frequency modulation model, it is used as the MPC prediction model as follows:

[0014] [ΔX g (k+i),ΔP r (k+i),ΔP g (k+i),Δf(k+i)]=f(Δu(k+i-1),P g (k+i-1),ΔX g (k+i-1),

[0015] ΔP r (k+i-1),ΔP g (k+i-1),Δf(k+i-1))

[0016] Where i = 1, 2, ... p, p is the prediction time domain, and:

[0017] P g (k+j)=P g (k+j-1)+ΔP g (k+j)

[0018] Wherein, j = 1, 2, ..., p-1;

[0019] Then, the frequency deviation prediction values ​​f(k+1), f(k+2), ..., f(k+p) at time k+1, k+2, ..., k+p are obtained as follows:

[0020] f(k+i)=f(k+i-1)+Δf(k+i).

[0021] Furthermore, for the setting of the objective function, on the one hand, it is necessary to consider making the system frequency deviation close to the reference value, and on the other hand, it is necessary to make the control signal change not too large. Therefore, the optimal objective function is designed as follows:

[0022]

[0023] In the formula, f ref is the system frequency deviation reference value, q i and w i They are the frequency deviation weighting coefficient and the control signal increment weighting coefficient respectively.

[0024] Furthermore, the steps of the differential evolution algorithm to solve the optimization problem are as follows:

[0025] In order to reduce the difficulty of solving the objective function, it is assumed that the control quantity outside the control domain remains unchanged, that is:

[0026] Δu i (k+i)=0,i=m,m+1,…,p-1

[0027] Where m is the control time domain;

[0028] The optimal control signal increment sequence ΔU(k) ​​is obtained as follows:

[0029] ΔU(k)=[Δu(k),...,Δu(k+m),...Δu(k+p)]

[0030] Take the first element Δu(k) in ΔU(k) ​​plus the controller output u(k) at the previous moment as the controller output. At this point, the control action at one sampling moment is completed. At the next sampling moment, its control time domain moves forward one moment, and ΔU(k) ​​is recalculated. This cycle is repeated to achieve rolling optimization.

[0031] In a second aspect, the present invention provides a variable parameter power system adaptive frequency regulation system, comprising:

[0032] Wide load frequency regulation model construction unit: Based on the variable parameter thermal power unit model and considering the change of power system inertia, a wide load frequency regulation model under the flexible operation of thermal power units is constructed;

[0033] A frequency modulation model construction unit based on Attention-LSTM: constructing a simulation data set according to the wide load frequency modulation model, training a long short-term memory network Attention-LSTM combined with an attention mechanism, and constructing a frequency modulation model based on Attention-LSTM;

[0034] Frequency regulation controller based on Attention-LSTM-MPC: The frequency regulation model based on Attention-LSTM is used to predict the dynamic response characteristics of the power grid frequency, and this is used as the MPC prediction model. Consider making the controlled output close to the reference input while keeping the control action change not too large, design the optimal objective function, and use the differential evolution algorithm to solve the optimization problem to achieve rolling optimization.

[0035] The beneficial effects of the present invention are as follows: The present invention combines inertia prediction to construct a wide load frequency regulation model under the flexible operation of thermal power units, and then uses the Attention-LSTM model to fit and reconstruct the power grid AGC model (automatic generation control model), solving the model mismatch problem caused by the change of model parameters under the wide load operation of thermal power units, and enhancing the reliability and robustness of the power grid frequency regulation process. Due to the change in inertia caused by the uncertainty of the penetration rate of new energy, the accuracy of frequency prediction is improved by combining inertia prediction, and the response speed of the power system to frequency changes is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0037] Figure 1 A flow chart of a variable parameter power system adaptive frequency modulation method according to an embodiment of the present invention;

[0038] Figure 2 A single-area control block diagram of a variable-parameter power system adaptive frequency modulation method according to an embodiment of the present invention;

[0039] Figure 3 In a variable parameter power system adaptive frequency modulation method according to an embodiment of the present invention, T r 、T t Follow P g ;

[0040] Figure 4 The linearized T in a variable parameter power system adaptive frequency modulation method according to an embodiment of the present invention is r 、T t Follow P g ;

[0041] Figure 5This is a structural diagram of a long short-term memory network combined with an attention mechanism in a variable parameter power system adaptive frequency modulation method according to an embodiment of the present invention;

[0042] Figure 6 A model predictive control block diagram of a variable parameter power system adaptive frequency modulation method according to an embodiment of the present invention;

[0043] Figure 7 A comparison diagram of frequency deviation responses of step disturbances of different strategies under different initial loads in a variable parameter power system adaptive frequency modulation method according to an embodiment of the present invention;

[0044] Figure 8 A comparison diagram of power responses of thermal power units to step disturbances of different strategies under different initial loads in a variable parameter power system adaptive frequency modulation method according to an embodiment of the present invention;

[0045] Fig. 9 A comparison diagram of random disturbance frequency deviation responses of different strategies under different initial loads in a variable parameter power system adaptive frequency regulation method according to an embodiment of the present invention;

[0046] Fig.10 A comparison diagram of power responses of thermal power units subjected to random disturbances of different strategies under different initial loads in a variable parameter power system adaptive frequency modulation method according to an embodiment of the present invention;

[0047] In the figure, PID: proportional integral derivative frequency modulation;

[0048] Conventional MPC: Conventional Model Predictive Control;

[0049] Attention-LSTM-MPC: Predictive control based on machine learning models. DETAILED DESCRIPTION

[0050] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0051] In the description of the present invention, the consecutive numbering of the method steps is for the convenience of review and understanding. Combined with the overall technical scheme of the present invention and the logical relationship between the various steps, adjusting the implementation order between the steps will not affect the technical effect achieved by the technical scheme of the present invention.

[0052] This embodiment provides a variable parameter power system adaptive frequency regulation method based on machine learning model predictive control, the method includes steps S100-S300, and the details of each step are as follows:

[0053] S100. Based on the variable parameter thermal power unit model and considering the change of power system inertia, a wide load frequency regulation model under flexible operation of thermal power units is constructed;

[0054] S200, constructing a simulation data set according to the frequency modulation model, training a long short-term memory network (Attention-LSTM) combined with an attention mechanism, and constructing an Attention-LSTM frequency modulation model;

[0055] S300. Predict the dynamic response characteristics of the power grid frequency according to the Attention-LSTM frequency modulation model, and use this as the MPC prediction model. Consider making the controlled output close to the reference input while keeping the control action change not too large, design the optimal objective function, use the differential evolution algorithm to solve the optimization problem, and realize rolling optimization.

[0056] The variable parameter power system adaptive frequency regulation method of the present invention is described in detail below.

[0057] 1) Construct a single-region frequency regulation model with variable-parameter thermal power units

[0058] Please refer to Figure 2 , Figure 2 The single-region frequency regulation model of the thermal power unit with variable parameters shown in FIG. 1 is as follows:

[0059]

[0060] Where: ΔP c , ΔX g , ΔP r , ΔP g are the control signal change, governor, reheat boiler and turbine output change respectively; Δf is the frequency deviation change; T g , K r are the governor response time constant and the reheat boiler gain coefficient respectively; T r 、T t are the response time constant of the reheat boiler and the governor respectively; R is the governor droop coefficient of the thermal power unit. Among them, in the variable parameter thermal power unit model, T r 、T t Will change with the turbine output P g changes with the change of T r 、T t Follow P g Changes such as Figure 3 shown.

[0061] After linearization, we get T r 、T t Follow P g Changes such as Figure 4 As shown, the expression is as follows:

[0062]

[0063] In a single-region power system, the unbalanced active power will cause frequency fluctuations, and the process can be expressed as:

[0064]

[0065] Where: M and D are the system equivalent inertia coefficient and damping constant respectively; ΔP g , ΔP L They are respectively the change in turbine output power and the change in load.

[0066] In summary, the single-area frequency regulation model of the power system can be expressed as:

[0067]

[0068] Where: A, B, C are the state matrix, input matrix, and output matrix of the single-region power system frequency regulation model, respectively; x is the single-region state vector, and the point above it represents the derivative; regional input u = ΔP c , regional output y = Δf.

[0069]

[0070] C = [0 0 0 1],

[0071] Based on the model of formula (7), the MPC frequency controller is designed and the model is discretized using the zero-order hold method to obtain:

[0072]

[0073] In the formula, A d =e AT , Δx(k)=x(k)-x(k-1), Δu(k)=u(k)-u(k-1), T is the system sampling period, T=0.01s.

[0074] 2) Build Attention-LSTM frequency modulation model

[0075] Based on the frequency regulation model of thermal power units in formula (8), a simulation data set is constructed, a neural network is trained, and then an Attention-LSTM model is constructed. The input and output relationship of the model can be expressed by formula (9).

[0076]

[0077] Where ΔX g (k) is the change in the governor valve position at time k, ΔP r (k) is the output power change of the reheat boiler at time k, ΔP g (k) is the change in output power of the steam turbine at time k, P g (k) is the real-time output power of the steam turbine, Δf(k) is the frequency deviation change, Δu(k) is the control signal increment at time k, ΔX g (k+1),ΔP r (k+1),ΔP g (k+1) and Δf(k+1) are the changes of the corresponding variables at time k+1.

[0078] The above data are all generated by Matlab, and the Z-score standardization method is used to standardize the data. The Z-score standardization method is shown in formula (10).

[0079]

[0080] Where: is the mean of the data x, σ is the standard deviation of the data, x * Indicates the data after normalization.

[0081] By dividing the data into training set and test set, the training set containing the input and output running status data is input into the Attention-LSTM model for training, and the frequency modulation model based on Attention-LSTM is obtained.

[0082] Input the test set data into the trained Attention-LSTM model to run the simulation and obtain the running results, Δx i (k) represents the simulated value of the i-th sample at time k, and the true value of the corresponding sample is The evaluation indicators Mean Absolute Error (MAE) and correlation coefficient (R) are used to quantitatively evaluate the characterization accuracy of the BP neural network model. The calculation formulas of MAE and R are shown in Equations (11) and (12).

[0083]

[0084] Where i∈[1, m], MAE ranges from 0 to infinity, representing the average of the absolute values ​​of the differences between the actual values ​​and the predicted values. It is a non-negative value and 0 indicates a perfect prediction, that is, all predicted values ​​are exactly equal to the actual values. The regression value R represents the correlation between the predicted output and the target output. The closer the R value is to 1, the closer the relationship between the predicted and output data is, and the better the fitting effect is.

[0085] 3) Power grid frequency regulation strategy based on Attention-LSTM-MPC

[0086] Based on the constructed Attention-LSTM-based frequency modulation model, it is used as the MPC prediction model. The Attention-LSTM network structure is as follows Figure 5 As shown in the figure, the constructed Attention-LSTM has 7 input layer nodes and 4 output layer nodes, adopts a 4-head attention mechanism, and the number of neurons in the LSTM hidden layer is 20.

[0087] Given the control signal increment Δu(k) at time k and the output P of the thermal power unit g (k) Speed ​​governor output change ΔX g , Reheat boiler output change ΔP r (k), turbine output change ΔP g (k) and the frequency deviation change Δf are used as the neural network inputs to predict the governor output change, reheat boiler output change, turbine output change, and frequency deviation change at the next p moments, as shown in Formula (13).

[0088]

[0089] Where i = 1, 2, ... p, p is the prediction time domain, and:

[0090] P g (k+j)=P g (k+j-1)+ΔP g (k+j)

[0091] Wherein, j = 1, 2, ..., p-1;

[0092] From formula (13), the predicted values ​​of the frequency deviation change at time k+1, k+2, ..., k+p are Δf(k+1), Δf(k+2), ..., Δf(k+p), and further, the predicted values ​​of the frequency deviation at time k+1, k+2, ..., k+p are f(k+1), f(k+2), ..., f(k+p), as shown in formula (14).

[0093] f(k+i)=f(k+i-1)+Δf(k+i) (13)

[0094] When setting the controller objective function, on the one hand, we need to consider making the system frequency deviation close to the reference value, and on the other hand, we need to make the control signal change not too large. Therefore, the following objective function can be set:

[0095]

[0096] In the formula, f refis the system frequency deviation reference value, q i and w i They are the frequency deviation weighting coefficient and the control signal increment weighting coefficient respectively.

[0097] The differential evolution algorithm has been proven to be an efficient intelligent optimization algorithm and has been widely used in artificial neural networks, power, robotics, signal processing and other fields. Therefore, the differential evolution algorithm is used to solve the above optimization problem.

[0098] In order to reduce the difficulty of solving the objective function, it is assumed that the control quantity remains unchanged outside the control time domain, that is:

[0099] Δu i (k+i)=0,i=m,m+1,…,p-1 (15)

[0100] The optimal control signal increment sequence ΔU(k) ​​is obtained as follows:

[0101] ΔU(k)=[Δu(k),...,Δu(k+m),...Δu(k+p)] (16)

[0102] Take the first element Δu(k) in ΔU(k) ​​plus the controller output u(k) at the previous moment as the controller output. At this point, the control action of one sampling moment is completed. The Attention-LSTM-MPC control block diagram is as follows: Figure 6 As shown, at the next sampling moment, the control time domain moves forward one moment, and ΔU(k) ​​is recalculated, and this cycle is repeated to achieve rolling optimization.

[0103] This embodiment combines inertia prediction to build a wide load frequency regulation model under the flexible operation of thermal power units, and then uses the Attention-LSTM model to fit and reconstruct the power grid AGC model, solving the model mismatch problem caused by the change of model parameters under wide load operation of thermal power units, and enhancing the reliability and robustness of the power grid frequency regulation process. By combining inertia prediction, the accuracy of frequency prediction is improved, and the response speed of the power system to frequency changes is improved.

[0104] In order to have a more intuitive understanding of the technical effect of this embodiment, this embodiment builds a system model in MATLAB / Simulink for simulation, wherein the parameters of the system model and related controllers are shown in Table 1:

[0105] Table 1

[0106] Thermal power unit parameters <![CDATA[T g =0.2,F=0.3,R=2%]]> Single-area power system parameters D＝1 Controller Parameters p=60,q=10

[0107] In order to test the anti-disturbance capability of the variable parameter power system adaptive frequency regulation method based on machine learning model predictive control method (Attention-LSTM-MPC) proposed in this embodiment, step load and random load interference are introduced respectively, and the results are compared with those of proportional integral differential (PID) control and conventional model predictive control (MPC) control method.

[0108] Table 2

[0109] Control strategy Maximum frequency deviation / pu IAE ITAE PID 0.0018 0.0043 0.0053 Conventional MPC 0.0068 0.0522 0.0454 Attention-LSTM-MPC 0.0010 <![CDATA[8.6234×10 -4 ]]> <![CDATA[5.2573×10 -4 ]]>

[0110] Table 3

[0111]

[0112] Table 4

[0113] Control strategy Maximum frequency deviation / pu IAE ITAE PID 0.0012 0.0031 0.0073 Conventional MPC 0.0010 <![CDATA[6.9878e -4 ]]> <![CDATA[5.0222e -4 ]]> Attention-LSTM-MPC <![CDATA[8.8313×10 -4 ]]> <![CDATA[5.9132×10 -4 ]]> <![CDATA[3.1337e×10 -4 ]]>

[0114] Please refer to Figure 7 and Tables 2, 3, and 4, Figure 7 The comparison of the step disturbance frequency deviation response of the variable parameter power system adaptive frequency regulation method using PID control, conventional MPC control and machine learning model predictive control under different initial loads is described. Tables 1, 2 and 3 respectively describe the step disturbance frequency deviation response results of the variable parameter power system adaptive frequency regulation method using PID control, conventional MPC control and machine learning model predictive control under different initial loads. Figure 7 (a), (b), and (c) show the frequency deviation response curves of each method when the initial load of the thermal power unit is 0.2pu, 0.5pu, and 0.8pu, respectively. Taking the initial load of the thermal power unit of 0.8pu as an example, the maximum frequency deviation of the system under the variable parameter power system adaptive frequency regulation method based on machine learning model predictive control is 8.8313×10 -4pu, the adjustment time is 1.26s, while the maximum frequency deviation of the system under PID and conventional MPC strategies is 0.0012pu and 0.0010pu, and the adjustment time is 4.67s and 1.88s, respectively. Compared with PID and conventional MPC, the maximum frequency deviation of the variable parameter power system adaptive frequency regulation method based on machine learning model predictive control is reduced by 26.40% and 11.68%, respectively, and the adjustment time is shortened by 73.02% and 32.97%, respectively. It can be seen that the frequency regulation performance of the variable parameter power system adaptive frequency regulation method based on machine learning model predictive control is better than PID and conventional MPC, which effectively improves the frequency regulation capability of the system. The same conclusion can be drawn by comparing the maximum frequency deviation and adjustment time of the system under the initial load of 0.2pu and 0.5pu thermal power units. In addition, when the initial load of the thermal power unit is 0.2pu, due to the large changes in the parameters of the thermal power unit, the system frequency deviation under the conventional MPC strategy showed oscillation. However, under the variable parameter power system adaptive frequency regulation method based on machine learning model predictive control in this embodiment, the system still showed good frequency regulation performance, indicating the effectiveness of the method of the present invention.

[0115] In addition, the output power response of the system thermal power units under step disturbances of different strategies is as follows: Figure 8 As shown, Figure 8 (a), (b), and (c) respectively show the output power response curves of each method when the initial load of the thermal power unit is 0.2pu, 0.5pu, and 0.8pu. From the comparison in the figures, it can be seen that, taking the initial load of the thermal power unit as 0.8pu, the power overshoot and adjustment time of the thermal power unit of the system under the adaptive frequency modulation method of the variable parameter power system based on machine learning model predictive control are 7.61% and 1.56s, the power overshoot and adjustment time of the thermal power unit of the system under the PID strategy are 1.39% and 3.93s, and the power overshoot and adjustment time of the thermal power unit of the system under the conventional MPC strategy are 8.93% and 2.28s. Compared with conventional MPC, the power overshoot of the thermal power unit of the system under the adaptive frequency modulation method of the variable parameter power system based on machine learning model predictive control is reduced by 14.78%, and the adjustment time is shortened by 31.57%. It can be seen that the adaptive frequency modulation method of the variable parameter power system based on machine learning model predictive control has a smaller adjustment amplitude and a faster response speed. Compared with PID, although the power overshoot of the thermal power units in the system under the variable parameter power system adaptive frequency regulation method based on machine learning model predictive control has increased, the adjustment time has been shortened by 60.30%, and the response speed is faster. The same conclusion can be drawn by comparing the power overshoot and adjustment time of the thermal power units under the initial load of 0.2pu and 0.5pu thermal power units.

[0116] Table 5

[0117] Control strategy Frequency deviation fluctuation range / pu IAE ITAE PID -0.0014～0.0022 0.0079 0.0100 Conventional MPC -0.0088～0.0082 0.0730 0.0514 Attention-LSTM-MPC <![CDATA[-6.9039×10 -4 ~6.6983×10 -4 ]]> 0.0033 0.0038

[0118] Table 6

[0119]

[0120] Table 7

[0121]

[0122] Please refer to Fig. 9 and Tables 5, 6, and 7, Fig. 9 (a), (b), (c) respectively describe the comparison of the random disturbance frequency deviation response of the variable parameter power system adaptive frequency modulation method under different initial loads using PID control, conventional MPC control and machine learning model predictive control. Tables 1, 2 and 3 respectively describe the random disturbance frequency deviation response results of the variable parameter power system adaptive frequency modulation method under different initial loads using PID control, conventional MPC control and machine learning model predictive control. Taking the initial load of the thermal power unit 0.8pu as an example, the system frequency deviation fluctuation range under the variable parameter power system adaptive frequency modulation method based on machine learning model predictive control is -4.4203×10 -4 ~6.5172×10 -4 pu, IAE and ITAE indicators are 0.0027 and 0.0030 respectively, while the system frequency deviation fluctuation range under PID strategy is -0.0011~0.0016pu, and ITAE indicators are 0.0062 and 0.0391 respectively, and the system frequency deviation fluctuation range under conventional MPC strategy is -8.9505×10 -4 ~8.2144×10 -4 pu, IAE and ITAE indicators are 0.0042 and 0.0057 respectively. Compared with PID and conventional MPC, the frequency deviation fluctuation range of the system under the adaptive frequency modulation method of variable parameter power system based on machine learning model predictive control is reduced by 59.49% and 36.28% respectively, the IAE indicators are reduced by 56.46% and 35.71% respectively, and the ITAE indicators are reduced by 92.32% and 47.36% respectively. It can be seen that the adaptive frequency modulation method of variable parameter power system based on machine learning model predictive control can more effectively suppress the fluctuation of system frequency. Comparing the frequency fluctuation range, IAE and ITAE indicators of the system under the initial load of 0.2pu and 0.5pu thermal power units, the same conclusion can be drawn.

[0123] In addition, the output power response of the system thermal power units under random disturbances of different strategies is as follows: Fig.10 As shown, Fig.10(a), (b), and (c) respectively show the output power response curves of each method when the initial load of the thermal power unit is 0.2pu, 0.5pu, and 0.8pu. Taking the initial load of the thermal power unit of 0.8pu as an example, compared with the conventional MPC, the power curve of the thermal power unit of the system under the adaptive frequency regulation method of the variable parameter power system based on machine learning model predictive control is smoother, which reduces the adjustment amplitude of the thermal power unit and prolongs the service life of the thermal power unit. Compared with PID, although the adjustment amplitude of the thermal power unit is increased, the system response speed is faster, which is more conducive to suppressing the system frequency fluctuation. Comparing the power curves of the thermal power units of the system under the initial loads of 0.2pu and 0.5pu thermal power units, the same conclusion can be drawn.

[0124] This embodiment also provides a variable parameter power system adaptive frequency regulation system, which includes:

[0125] Wide load frequency regulation model construction unit: Based on the variable parameter thermal power unit model and considering the change of power system inertia, a wide load frequency regulation model under the flexible operation of thermal power units is constructed;

[0126] A frequency modulation model construction unit based on Attention-LSTM: constructing a simulation data set according to the wide load frequency modulation model, training a long short-term memory network Attention-LSTM combined with an attention mechanism, and constructing a frequency modulation model based on Attention-LSTM;

[0127] Frequency regulation controller based on Attention-LSTM-MPC: The frequency regulation model based on Attention-LSTM is used to predict the dynamic response characteristics of the power grid frequency, which is used as the MPC prediction model to design the optimal objective function, and the differential evolution algorithm is used to solve the optimization problem to achieve rolling optimization.

[0128] It should be noted that each unit in the above-mentioned variable parameter power system adaptive frequency modulation system can be fully or partially implemented by software, hardware and a combination thereof. The above-mentioned units can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of a variable parameter power system adaptive frequency modulation system, please refer to the definition of a variable parameter power system adaptive frequency modulation method above. The two have the same functions and effects, which will not be repeated here.

[0129] The above description of the embodiments is to facilitate the understanding and application of the present invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made to the present invention by those skilled in the art based on the disclosure of the present invention should be within the scope of protection of the present invention.

Claims

1. A variable parameter power system adaptive frequency modulation method, characterized in that: include: Based on the variable parameter thermal power unit model, considering the change of power system inertia, a wide load frequency regulation model under flexible operation of thermal power units is constructed; A simulation data set is constructed according to the wide load frequency modulation model, a long short-term memory network Attention-LSTM combined with an attention mechanism is trained, and a frequency modulation model based on Attention-LSTM is constructed; The frequency modulation model based on Attention-LSTM is used to predict the dynamic response characteristics of the power grid frequency. This is used as the MPC prediction model, the optimal objective function is designed, and the differential evolution algorithm is used to solve the optimization problem to achieve rolling optimization.

2. The variable parameter power system adaptive frequency modulation method according to claim 1, characterized in that: The input and output relationship of the FM model based on Attention-LSTM is as follows: [ΔX g (k+1);ΔP r (k+1);ΔP g (k+1);Δf(k+1)]= f(Δu(k),P g (k),ΔX g (k),ΔP r (k),ΔP g (k),Δf(k)) Where ΔX g (k) is the change in the governor valve position at time k, ΔP r (k) is the output power change of the reheat boiler at time k, ΔP g (k) is the change in output power of the steam turbine at time k, P g (k) is the real-time output power of the steam turbine, Δf(k) is the frequency deviation change, Δu(k) is the control signal increment at time k, ΔX g (k+1),ΔP r (k+1),ΔP g (k+1) and Δf(k+1) are the changes of the corresponding variables at time k+1.

3. The variable parameter power system adaptive frequency modulation method according to claim 2, characterized in that: Based on the constructed Attention-LSTM-based frequency modulation model, it is used as the MPC prediction model as follows: [ΔX g (k+i),ΔP r (k+i),ΔP g (k+i),Δf(k+i)]=f(Δu(k+i-1),P g (k+i-1),ΔX g (k+i-1),ΔP r (k+i-1),ΔP g (k+i-1),Δf(k+i-1)) where i=1, 2, ...p, p is the prediction time domain, and: P g (k+j)=P g (k+j-1)+ΔP g (k+j) Wherein, j = 1, 2, ..., p-1; Then, the frequency deviation prediction values ​​f(k+1), f(k+2), ..., f(k+p) at time k+1, k+2, ..., k+p are obtained as follows: f(k+i)=f(k+i-1)+Δf(k+i).

4. The variable parameter power system adaptive frequency modulation method according to claim 3, characterized in that: The optimal objective function is designed as follows: In the formula, f ref is the system frequency deviation reference value, q i and w i They are the frequency deviation weighting coefficient and the control signal increment weighting coefficient respectively.

5. The variable parameter power system adaptive frequency modulation method according to claim 4, characterized in that: The steps of differential evolution algorithm to solve optimization problems are as follows: In order to reduce the difficulty of solving the objective function, it is assumed that the control quantity outside the control domain remains unchanged, that is: Δu i (k+i)=0,i=m,m+1,…,p-1 Where m is the control time domain; The optimal control signal increment sequence ΔU(k) ​​is obtained as follows: ΔU(k)=[Δu(k),...,Δu(k+m),...Δu(k+p)] Take the first element Δu(k) in ΔU(k) ​​plus the controller output u(k) at the previous moment as the controller output. At this point, the control action at one sampling moment is completed. At the next sampling moment, its control time domain moves forward one moment, and ΔU(k) ​​is recalculated. This cycle is repeated to achieve rolling optimization.

6. A variable parameter power system adaptive frequency modulation system, characterized in that: include: Wide load frequency regulation model construction unit: Based on the variable parameter thermal power unit model and considering the change of power system inertia, a wide load frequency regulation model under the flexible operation of thermal power units is constructed; A frequency modulation model construction unit based on Attention-LSTM: constructing a simulation data set according to the wide load frequency modulation model, training a long short-term memory network Attention-LSTM combined with an attention mechanism, and constructing a frequency modulation model based on Attention-LSTM; Frequency regulation controller based on Attention-LSTM-MPC: The frequency regulation model based on Attention-LSTM is used to predict the dynamic response characteristics of the power grid frequency, which is used as the MPC prediction model to design the optimal objective function, and the differential evolution algorithm is used to solve the optimization problem to achieve rolling optimization.

7. The variable parameter power system adaptive frequency modulation system according to claim 6, characterized in that: The input and output relationship of the FM model based on Attention-LSTM is as follows: [ΔX g (k+1);ΔP r (k+1);ΔP g (k+1);Δf(k+1)]= f(Δu(k),P g (k),ΔX g (k),ΔP r (k),ΔP g (k),Δf(k)) In the formula, ΔX g (k) is the change in the governor valve position at time k, ΔP r (k) is the output power change of the reheat boiler at time k, ΔP g (k) is the change in output power of the steam turbine at time k, P g (k) is the real-time output power of the steam turbine, Δf(k) is the frequency deviation change, Δu(k) is the control signal increment at time k, ΔX g (k+1),ΔP r (k+1),ΔP g (k+1) and Δf(k+1) are the changes of the corresponding variables at time k+1.

8. The variable parameter power system adaptive frequency modulation system according to claim 7, characterized in that: Based on the constructed Attention-LSTM-based frequency modulation model, it is used as the MPC prediction model as follows: [ΔX g (k+i),ΔP r (k+i),ΔP g (k+i),Δf(k+i)]=f(Δu(k+i-1),P g (k+i-1),ΔX g (k+i-1),ΔP r (k+i-1),ΔP g (k+i-1),Δf(k+i-1)) Where i = 1, 2, ... p, p is the prediction time domain, and: P g (k+j)=P g (k+j-1)+ΔP g (k+j) Wherein, j = 1, 2, ..., p-1; Then, the frequency deviation prediction values ​​f(k+1), f(k+2), ..., f(k+p) at time k+1, k+2, ..., k+p are obtained as follows: f(k+i)=f(k+i-1)+Δf(k+i).

9. The variable parameter power system adaptive frequency modulation system according to claim 8, characterized in that: In the frequency modulation controller based on Attention-LSTM-MPC, the optimal objective function is designed as follows: In the formula, f ref is the system frequency deviation reference value, q i and w i They are the frequency deviation weighting coefficient and the control signal increment weighting coefficient respectively.

10. The variable parameter power system adaptive frequency modulation system according to claim 9, characterized in that: In the frequency modulation controller based on Attention-LSTM-MPC, the steps of solving the optimization problem by the differential evolution algorithm are as follows: In order to reduce the difficulty of solving the objective function, it is assumed that the control quantity outside the control domain remains unchanged, that is: Δu i (k+i)=0,i=m,m+1,…,p-1 Where m is the control time domain; The optimal control signal increment sequence ΔU(k) ​​is obtained as follows: ΔU(k)=[Δu(k),...,Δu(k+m),...Δu(k+p)] Take the first element Δu(k) in ΔU(k) ​​plus the controller output u(k) at the previous moment as the controller output. At this point, the control action at one sampling moment is completed. At the next sampling moment, its control time domain moves forward one moment, and ΔU(k) ​​is recalculated. This cycle is repeated to achieve rolling optimization.

Images