A wind power frequency modulation optimization control method based on capsule network prediction

Abstract

The application discloses a wind power frequency modulation optimization control method based on capsule network prediction, and comprises the following steps: S1, randomly selecting different v, d and K for combination, performing frequency modulation simulation to obtain deviation Δf max , Δf and Δt; S2, taking the selected n groups of v, d and K and simulation data as training samples, adopting a capsule network to construct a prediction model, and predicting the values of Δf max , Δf and Δt of the remaining combinations of v, d and K; S3, obtaining d and K under different v; S4, establishing a mapping relationship between v and d and a mapping relationship between v and K; S5, obtaining the optimal load shedding coefficient d and droop coefficient K corresponding to the current v through f1 and f2; S6, calculating wind turbine load shedding power P del and frequency modulation power ΔP, and calculating actual wind turbine frequency modulation power P ref . The capsule neural network is introduced into the wind power frequency modulation optimization control, and the capsule neural network has the advantages of fast training speed, good effect of capturing internal features of data, high prediction result accuracy and the like.

Description

Technical Field

[0001] This invention belongs to the field of wind power frequency regulation control technology, specifically relating to a wind power frequency regulation optimization control method based on capsule network prediction. Background Technology

[0002] Wind power generation technology is currently the fastest-growing new energy power generation technology, and the global installed capacity of wind power is increasing year by year due to the rapid development of wind power. However, wind power is random and volatile, and large-scale integration of wind power into the grid can lead to problems such as power imbalance and frequency fluctuations in the power system. Among these, the frequency regulation problem of the power system is an important factor restricting the large-scale development of wind power.

[0003] On the one hand, large-scale wind power grid connection introduces greater randomness and volatility. Traditional turbines, limited by frequency regulation response speed and ramp-up speed, struggle to track wind power fluctuations, posing a greater threat to system frequency. On the other hand, variable-speed wind turbines are completely decoupled from system frequency, lacking the ability to respond to system frequency changes. Large-scale wind power integration reduces the power system's inertia. Furthermore, because wind turbines typically operate in Maximum Power Point Tracking (MPPT) mode, they lack frequency regulation active power reserve, further reducing the power system's frequency regulation capability. Therefore, researching optimized control methods for wind power frequency regulation is crucial for improving wind power utilization, alleviating the frequency regulation pressure on traditional turbines, and ensuring the safe and stable operation of the power system.

[0004] Overspeed control and pitch angle control can be used to control the load reduction of wind turbines, thereby reserving frequency regulation capacity. Currently, most research uses a fixed load reduction rate; however, wind power is greatly affected by wind speed, and a fixed load reduction rate can lead to increased curtailment or poor frequency regulation. When the power system is disturbed, using droop control with a fixed droop coefficient for wind turbine frequency regulation also suffers from underutilization of wind turbine load reduction capacity and poor frequency regulation performance. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a wind power frequency regulation optimization control method based on capsule network prediction that introduces capsule neural networks into the field of wind power frequency regulation optimization control technology, enabling wind power frequency regulation optimization control with varying load reduction coefficients and varying droop coefficients under different wind speeds, thereby improving the system frequency regulation capability and improving and stabilizing the system frequency.

[0006] The objective of this invention is achieved through the following technical solution: a wind power frequency regulation optimization control method based on capsule network prediction, comprising the following steps:

[0007] S1. Randomly select n different combinations of wind speed v, fan load reduction factor d, and fan droop factor K, then perform frequency modulation simulation to obtain the maximum system frequency deviation Δf. max Steady-state frequency deviation Δf, dynamic frequency modulation response time Δt;

[0008] S2. Using the selected n sets of v, d, K and simulation data as training samples, a capsule network is used to construct a prediction model and predict the system Δf under the remaining combinations of v, d, and K. max The values ​​of Δf and Δt;

[0009] S3. Using equation (1) as the objective function, obtain the fan load reduction coefficient d and fan sag coefficient K under different wind speeds v;

[0010] min∑(aΔf max +bΔf+cΔt)=F(d,K) (1)

[0011] Where, Δf max The coefficients a, b, and c before Δf and Δt represent their corresponding weights;

[0012] S4. In the simulation model, establish the mapping relationship between v and d: d = f1(v) and the mapping relationship between v and K: K = f2(v). Unknown data points are filled using linear interpolation.

[0013] S5. Measure the real-time wind speed v of the fan, and obtain the optimal load reduction coefficient d and sag coefficient K corresponding to the current v through f1 and f2;

[0014] S6. Calculate the fan load reduction power P based on d and K. del And the frequency regulation power ΔP, and then the actual frequency regulation power P of the wind turbine is calculated. ref This completes the fan load reduction and frequency regulation control.

[0015] The beneficial effects of this invention are:

[0016] (1) Capsule neural networks are introduced into the field of wind power frequency regulation optimization control technology. Taking advantage of the advantages of capsule networks such as fast training speed, good effect of capturing internal features of data, and high prediction accuracy, a system is constructed with random combinations of v, d, and K as inputs. The system simulation result Δf max The capsule network prediction model outputs Δf and Δt, and predicts the system Δf under other combinations of v, d, and K. max The values ​​of Δf and Δt are used for subsequent optimization of parameters d and K;

[0017] (2) Taking into full account the shortcomings of wind turbine fixed load reduction coefficient and fixed droop coefficient frequency regulation strategy, such as low wind energy utilization or insufficient frequency regulation capability, with the goal of improving wind power frequency regulation capability, the optimal load reduction coefficient and droop coefficient of wind turbine are found under multiple parameter combinations, so as to realize wind power frequency regulation optimization control with variable load reduction coefficient and variable droop coefficient under different wind speeds, thereby improving the system frequency regulation capability and improving and stabilizing the system frequency. Attached Figure Description

[0018] Figure 1 This is a flowchart of the wind power frequency regulation optimization control method based on capsule network prediction of the present invention;

[0019] Figure 2 Taking the frequency drop of the system of the present invention after being disturbed as an example, the dynamic frequency response curve of the system is shown.

[0020] Figure 3 This is a schematic diagram of the prediction model constructed using capsule networks according to the present invention;

[0021] Figure 4 This is a flowchart of the offline optimization process for wind turbine load reduction and frequency regulation parameters according to the present invention;

[0022] Figure 5 This is a wind turbine load reduction control curve diagram of the present invention;

[0023] Figure 6 Optimize the control block diagram for wind turbine load reduction and frequency regulation. Detailed Implementation

[0024] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0025] like Figure 1 As shown, the wind power frequency regulation optimization control method based on capsule network prediction of the present invention includes the following steps:

[0026] S1. Randomly select n different combinations of wind speed v, fan load reduction factor d, and fan droop factor K, then perform frequency modulation simulation to obtain the maximum system frequency deviation Δf. max Steady-state frequency deviation Δf, dynamic frequency modulation response time Δt;

[0027] v increases or decreases by a factor of 0.1 and v∈(v cut-in ,v cut-out ), v cut-in For the cut-in wind speed of the fan, v cut-out The cutoff wind speed for the fan; d increases or decreases by an increment of 0.01 and d∈[0,d max ], d max K is the maximum allowable load reduction factor; K is an integer and K∈[0,K] max ], K maxThe maximum allowable droop coefficient; n < N, where N represents the total number of combinations of v, d, and K.

[0028] The frequency regulation simulation system is a microgrid system constructed in simulation software, containing wind turbines, traditional generator sets, and electrical loads. System frequency disturbances are simulated through load switching. This embodiment uses the frequency drop after the system is disturbed as an example; the system frequency dynamic response curve is shown below. Figure 2 As shown.

[0029] S2. Using the selected n sets of v, d, K and simulation data as training samples, a capsule network is used to construct a prediction model and predict the system Δf under the remaining combinations of v, d, and K. max The values ​​of Δf and Δt;

[0030] The prediction model constructed using capsule networks consists of convolutional layers, primary capsule layers, digital capsule layers, flattened layers, and fully connected layers, as follows: Figure 3 As shown;

[0031] The training method for the prediction model includes the following steps:

[0032] S21. Process the n sets of v, d, and K selected in S1 into an input vector T. in Form; the maximum frequency deviation Δf of n systems obtained from S1 frequency modulation simulation. max The steady-state frequency deviation Δf and dynamic frequency modulation response time Δt are processed into the true output vector T. out-true Form, then make T in The data is passed through two convolutional layers to extract primary features, which are then fed into a primary capsule layer.

[0033] S22. The primary capsule layer processes the primary feature vectors inside the data to obtain the high-level feature vectors inside the data, and then sends them into the digital capsule layer using a dynamic routing algorithm.

[0034] S23. The digital capsule layer processes the high-level feature vectors inside the data, outputs the prediction vector, and sends it to the Flatten layer for "flattening".

[0035] S24. The Flatten layer flattens the output vector of the digital capsule layer and sends it to the fully connected layer to output the prediction result, thus obtaining the prediction output vector T. out-pre The prediction vector has a dimension of 3*1, and the elements in the prediction vector represent Δf. max Predicted values ​​of Δf and Δt;

[0036] S25, predict the output vector T out-pre With the true output vector T out-true Calculate the loss function and update the parameters of the capsule network until the capsule network converges.

[0037] S3. Using equation (1) as the objective function, obtain the fan load reduction coefficient d and fan droop coefficient K under different wind speeds v, and complete the offline optimization of the fan load reduction coefficient d and fan droop coefficient K.

[0038] min∑(aΔf max +bΔf+cΔt)=F(d,K) (1)

[0039] Where, Δf max The coefficients a, b, and c before Δf and Δt represent their corresponding weights. To improve the transient performance of the frequency, a, b, and c are set to 0.6, 0.3, and 0.1 respectively. The offline optimization flowchart for wind turbine load shedding frequency regulation parameters is shown below. Figure 4 As shown.

[0040] S4. In the simulation model, establish the mapping relationship between v and d: d = f1(v) and the mapping relationship between v and K: K = f2(v). Unknown data points are filled using linear interpolation.

[0041] S5. Measure the real-time wind speed v of the fan, and obtain the optimal load reduction coefficient d and sag coefficient K corresponding to the current v through f1 and f2;

[0042] S6. Calculate the fan load reduction power P based on d and K. del And the frequency regulation power ΔP, and then the actual frequency regulation power P of the wind turbine is calculated. ref The system completes the wind turbine load reduction and frequency regulation control; the mechanical power captured when the wind turbine is operating normally in MPPT mode is:

[0043]

[0044] Where ρ is air density, R is rotor radius, v is wind speed, β is turbine pitch angle, ω is rotor speed; λ is tip speed ratio, a function of v and ω; C p Let be the wind energy utilization coefficient, which is a function of λ and β;

[0045] Wind turbine reduced load actual power P del for:

[0046] P del =(1-d)P opt (3)

[0047] The wind turbine load reduction control curve is shown below. Figure 5 As shown in the figure, ω max This indicates the rated speed of the fan.

[0048] When the wind turbine reduces power load through overspeed load reduction control, the turbine rotor speed ω reaches the rated speed ω. maxIf the power reduction factor d is still not achieved, the pitch angle is increased by controlling the pitch angle to assist the wind turbine in achieving the power reduction factor d.

[0049] When the system frequency changes, the frequency modulation power ΔP is:

[0050] ΔP=-K*(ff n (4)

[0051] Where f is the system frequency; f n The system's nominal frequency is used to obtain the actual power output P of the fan during frequency regulation. ref for:

[0052] P ref =P del +ΔP (5)

[0053] Complete the optimized control of wind turbine load reduction and frequency regulation. The block diagram for optimized wind turbine load reduction and frequency regulation control is shown below. Figure 6 As shown.

[0054] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A wind power frequency regulation optimization control method based on capsule network prediction, characterized in that, Includes the following steps: S1, randomly select n groups of different wind speed v, fan unloading coefficient d, fan droop coefficient K combination, and then carry out frequency modulation simulation to obtain the maximum frequency deviation of the system Δf max , steady-state frequency deviation Δf, dynamic frequency modulation response time Δt; S2. Using the selected n sets of v, d, K and simulation data as training samples, a capsule network is used to construct a prediction model and predict the system Δf under the remaining combinations of v, d, and K. max The values ​​of Δf and Δt; S3. Using equation (1) as the objective function, the fan load reduction coefficient d and fan sag coefficient K are obtained under different wind speeds v. min∑(aΔf max +bΔf+cΔt)=F(d,K) (1) Where, Δf max The coefficients a, b, and c before Δf and Δt represent their corresponding weights; S4. Establish the mapping relationship between v and d in the simulation model: d = f1(v) and the mapping relationship between v and K: K = f2(v). Unknown data points are filled using linear interpolation. S5. Measure the real-time wind speed v of the fan, and obtain the optimal load reduction coefficient d and sag coefficient K corresponding to the current v through f1 and f2; S6. Calculate the fan load reduction power P based on d and K. del And the frequency regulation power ΔP, and then the actual frequency regulation power P of the wind turbine is calculated. ref This completes the fan load reduction and frequency regulation control.

2. The wind power frequency regulation optimization control method based on capsule network prediction according to claim 1, characterized in that, In step S1, v increases or decreases by an increment of 0.1 and v∈(v cut-in ,v cut-out ), v cut-in For the fan's cut-in wind speed, v cut-out The cutoff wind speed for the fan; d increases or decreases by an increment of 0.01 and d∈[0,d max ], d max K is the maximum allowable load reduction factor; K is an integer and K∈[0,K] max ], K max This is the maximum allowable droop coefficient.

3. The wind power frequency regulation optimization control method based on capsule network prediction according to claim 1, characterized in that, In step S2, the prediction model constructed using capsule networks sequentially includes convolutional layers, primary capsule layers, digital capsule layers, flatten layers, and fully connected layers. The training method for the prediction model includes the following steps: S21. Process the n sets of v, d, and K selected in S1 into an input vector T. in Form; the maximum frequency deviation Δf of n systems obtained from S1 frequency modulation simulation. max The steady-state frequency deviation Δf and dynamic frequency modulation response time Δt are processed into the true output vector T. out-true Form, then make T in The data is passed through two convolutional layers to extract primary features, which are then fed into a primary capsule layer. S22. The primary capsule layer processes the primary feature vectors inside the data to obtain the high-level feature vectors inside the data, and then sends them into the digital capsule layer using a dynamic routing algorithm. S23. The digital capsule layer processes the high-level feature vectors inside the data, outputs the prediction vector, and sends it to the Flatten layer for "flattening" processing. S24. The Flatten layer flattens the output vector of the digital capsule layer and sends it to the fully connected layer to output the prediction result, thus obtaining the prediction output vector T. out-pre The prediction vector has a dimension of 3*1, and the elements in the prediction vector represent Δf. max Predicted values ​​of Δf and Δt; S25, predict the output vector T out-pre With the true output vector T out-true Calculate the loss function and update the parameters of the capsule network until the capsule network converges.

4. The wind power frequency regulation optimization control method based on capsule network prediction according to claim 1, characterized in that, In step S6, the mechanical power captured by the wind turbine when it is operating normally in MPPT mode is: C p =C p (β,λ),λ=λ(v,ω) Where ρ is air density, R is rotor radius, v is wind speed, β is turbine pitch angle, ω is rotor speed; λ is tip speed ratio, a function of v and ω; C p Let be the wind energy utilization coefficient, which is a function of λ and β; Wind turbine reduced load actual power P del for: P del =(1-d)P opt When the wind turbine reduces power load through overspeed load reduction control, the turbine rotor speed ω reaches the rated speed ω. max If the power reduction factor d is still not achieved, the pitch angle is increased by controlling the pitch angle to assist the wind turbine in achieving the power reduction factor d. When the system frequency changes, the frequency modulation power ΔP is: ΔP=-K*(f-f n ) Where f is the system frequency; f n The system's nominal frequency is used to obtain the actual power output P of the fan during frequency regulation. ref for: P ref =P del +ΔP。

Images