Emergency control method for transient stability of microgrid based on parameter rolling regulation

Abstract

The microgrid transient stability emergency control method based on parameter rolling regulation first initializes a time window and collects the actual values ​​of , , , and for each VCI (Variable Instability Controller) and the adopted control sequence within the time window. Then, the collected data is input into a power trajectory prediction model based on a spatiotemporal graph convolutional neural network to predict the , , , and for each VCI within the time window. If the predicted , , and satisfy the stability criterion, the emergency control ends; otherwise, based on the predicted , , and stability constraints, the control parameters are rolled and regulated to obtain the optimal control sequence that can stabilize the microgrid. This optimal control sequence is then applied to each VCI controller within the time window. Finally, a new round of judgment and rolling regulation is performed until the stability criterion is met, at which point regulation stops. This method requires no additional measurement equipment and can ensure reliable stability of the microgrid system under any instability conditions.

Description

Technical Field

[0001] This invention relates to the field of microgrid transient stability control technology, and in particular to an emergency control method for microgrid transient stability based on parameter rolling regulation. Background Technology

[0002] Similar to synchronous generators, voltage-controlled inverters (VCIs) face transient stability problems under large disturbances, severely impacting the stable operation of microgrids. Furthermore, microgrids have small radii, low inertia, and limited capacity, with transient instability timescales typically ranging from hundreds of milliseconds to seconds. This places extremely stringent demands on the speed and reliability of emergency transient stability control in microgrids. Currently, few publications or patents propose comprehensive emergency control strategies specifically for the characteristics of microgrids; most research focuses on microgrid transient stability assessment and microgrid stability improvement.

[0003] In the area of ​​online transient stability assessment of microgrids, patent application CN202111018894.3 discloses a transient stability assessment method for VSG multi-machine systems based on artificial neural networks, and patent application CN202111188280.X discloses a microgrid transient stability assessment method based on long short-term memory networks. These patents successfully apply ANN and LSTM to VCI single-machine infinite bus systems. Patent application CN202210828283.3 discloses a microgrid transient power angle trajectory prediction method based on a spatiotemporal graph convolutional neural network with an attention mechanism. It combines graph convolutional neural networks to propose a microgrid power angle trajectory method, providing a foundation for transient stability assessment. However, the above technologies stop at the level of microgrid transient stability assessment and cannot perform emergency control after a fault.

[0004] Regarding the improvement of transient stability of microgrids, references [1] and [2] analyzed the influence of control parameters on virtual synchronous generators (VSGs) and improved VSG stability by flexibly adjusting the virtual inertia of the VSG. However, the improvement capability is limited and cannot ensure that the VSG returns to stability after a fault. References [3]-[6] made different improvements to the active and reactive power control loops of the VSG, but the control coefficients of the newly added control loops are complicated to calculate online and are difficult to adapt to various unknown large disturbances, making it difficult to guarantee reliability. In addition, the existing methods are all based on the current state of the system and do not take into account the future evolution trend of the system, resulting in limited control effect. In summary, the existing stability improvement methods have all improved the stability of the VSG to a certain extent, but the online real-time speed and reliability are difficult to guarantee.

[0005] References:

[0006] [1] M. Li, W. Huang, N. Tai, L. Yang, D. Duan and Z. Ma, “A Dual-Adaptivity Inertia Control Strategy for Virtual Synchronous Generator,” IEEE Trans. Power Sys. , vol. 35, no. 1, pp. 594-604, Jan. 2020;

[0007] [2] P. Ge, C. Tu, F. Xiao, Q. Guo and J. Gao, “Design-OrientedAnalysis and Transient Stability Enhancement Control for a VirtualSynchronous Generator,” IEEE Trans. Ind. Electron. , vol. 70, no. 3, pp. 2675-2684, March 2023;

[0008] [3] M. Choopani, S. H. Hosseinian and B. Vahidi, “New TransientStability and LVRT Improvement of Multi-VSG Grids Using the Frequency of theCenter of Inertia,” IEEE Trans. Power Sys. , vol. 35, no. 1, pp. 527-538, Jan.2020;

[0009] [4] K. Sun, W. Yao, J. Wen and L. Jiang, “ Two-Stage SimultaneousControl Scheme for the Transient Angle Stability of VSG Considering CurrentLimitation and Voltage Support,” IEEE Trans. Power Sys. , vol. 37, no. 3, pp.2137-2150, May 2022;

[0010] [5] X. Xiong, C. Wu and F. Blaabjerg, “An Improved SynchronizationStability Method of Virtual Synchronous Generators Based on FrequencyFeedforward on Reactive Power Control Loop,” IEEE Trans. Power Electron. , vol.36, no. 8, pp. 9136-9148, Aug. 2021;

[0011] [6] X. Xiong, C. Wu, B. Hu, D. Pan and F. Blaabjerg, “TransientDamping Method for Improving the Synchronization Stability of VirtualSynchronous Generators,” IEEE Transactions on Power Electron , vol. 36, no. 7, pp.7820-7831, July 2021. Summary of the Invention

[0012] The technical problem to be solved by the present invention is to provide an emergency control method for transient stability of microgrids based on parameter rolling regulation. This method does not require additional measuring devices and can ensure reliable recovery of microgrid system under any instability conditions.

[0013] To solve the above-mentioned technical problems, the present invention adopts the following technical method: a microgrid transient stability emergency control method based on parameter rolling regulation, comprising:

[0014] Step S1, setting A time window, This is the sequence number of the time window. ,initialization ;

[0015] Step S2, collect Power angle of each VCI in the microgrid within the time window ,Voltage Output active power Output reactive power Actual values ​​and the control sequence used by the VCI controller The control sequence for:

[0016] (1)

[0017] In the formula, These are the virtual inertia, reference active power, and damping coefficient of the VCI, respectively. The number of VCIs;

[0018] Step S3: Input the data collected in step S2 into the power trajectory prediction model based on spatiotemporal graph convolutional neural network to predict... Active power output of each VCI within the time window ;

[0019] Step S4: Determine the predicted output active power. Whether the stabilization criterion is met, the stabilization criterion being "with Compared to the actual value of the output active power within the time window, the predicted value is higher. "The output active power begins to decrease within the time window." If this condition is met, the emergency control ends; otherwise, under the premise of satisfying stability constraints, the control proceeds as follows: 0 pairs By adjusting two control parameters in a rolling manner, an optimal control sequence is obtained that allows the microgrid to return to stability. Within the time window, the optimal control sequence is applied to each VCI controller;

[0020] Step S5, setting Then proceed to step S2 for a new round of judgment and rolling adjustment until the stabilization criterion is met, and then stop the adjustment.

[0021] Furthermore, in step S3, the power trajectory prediction model based on the spatiotemporal graph convolutional neural network predicts... Active power output of each VCI within the time window The process is as follows:

[0022] Step S31, control sequence feature mining;

[0023] For the Control sequence within a time window Perform min-max normalization:

[0024] (2)

[0025] in:

[0026] (3)

[0027] Obtain control sequence Afterwards, Perform feature flattening, then reduce the dimensionality. Input a fully connected layer and use the forward propagation method to mine control sequence features. , The calculation formula is as follows:

[0028] (4)

[0029] In the formula, and For the first Weights and biases of neurons in fully connected layers; This represents the number of neurons in the first fully connected layer.

[0030] Step S32, Time Series Spatiotemporal Dynamic Feature Mining;

[0031] First, determine the number Electrical adjacency matrix within a time window and time series The electrical adjacency matrix for:

[0032] (5)

[0033] The time series , The length of the time series, where any time segment... On The mathematical expression is:

[0034] (6)

[0035] In the formula, , , , VCI in Output active power, output reactive power, voltage, and power angle for a time segment; The number of VCIs;

[0036] Next, two spatiotemporal convolutional modules are stacked to perform deep mining of spatiotemporal dynamic features. The mining process of each spatiotemporal convolutional module is as follows:

[0037] 1) After passing through the spatiotemporal attention layer, the hidden temporal and spatial correlations in the time series are adaptively modeled to obtain the temporal attention matrix and the spatial attention matrix.

[0038] 2) Multiplying the obtained time attention matrix yields a new time series based on time attention. ;

[0039] 3) The new time series Spatial attention matrix and electrical adjacency matrix The inputs are fed into a spatial convolutional layer for spatial dynamic feature mining.

[0040] 4) Input is fed into a temporal convolutional layer for temporal dynamic feature mining;

[0041] The spatiotemporal dynamic features after in-depth analysis are: The calculation formula is as follows:

[0042] (7)

[0043] In the formula, and These are spatiotemporal convolution modules. Temporal convolutional layer kernels and spatial convolutional layer kernels; For graph convolution operations, This is a standard convolution operation;

[0044] Step S33, output power trajectory prediction;

[0045] First, the control sequence features are defined according to equations (8) and (9). Expanded to ;

[0046] (8)

[0047] (9)

[0048] In the formula, Total OK;

[0049] Then, the expanded Spatiotemporal dynamic characteristics Perform feature integration, and define the integrated features as follows: ;

[0050] Finally, the obtained Input a standard two-dimensional convolutional layer to predict The power trajectory values ​​of each VCI within the time window; the power prediction value is defined as follows: The calculation formula is as follows:

[0051] (10)

[0052] In the formula, These are the parameters of the convolution kernel for this convolution module.

[0053] Furthermore, in step S32, the electrical adjacency matrix The calculation process is as follows:

[0054] First, calculate the impedance matrix of each node of the VCI according to the following formula. :

[0055] (11)

[0056] in, For the first The self-impedance of the VCI in Taiwan For the first Taiwan VCI to the first The mutual impedance of the VCI in Taiwan. , , ;

[0057] Next, calculate the electrical distance between any two VCI nodes according to the following formula. ;

[0058] (12)

[0059] Then, the electrical distance is calculated according to the following formula. Perform normalization processing;

[0060] (13)

[0061] This yields the electrical adjacency matrix shown in equation (5). .

[0062] Furthermore, in step S4, the stability constraints include equilibrium point constraints, power angle constraints, and frequency constraints; wherein, the equilibrium point constraint is the VCI power angle curve and the reference active power after fault clearance in the current-limited state. There is a balance point The power angle constraint is the point on the VCI power angle curve corresponding to the limit clearing angle after fault clearance. At the equilibrium point Left side; Frequency constraint is that, provided the VCI can return to stability, the frequency offset does not exceed .

[0063] Furthermore, in step S4, Rolling control is performed using the following formula (14);

[0064] (14)

[0065] In the formula, For VCI in the Output active power within a time window This is the rated active power of the VCI. ;

[0066] After determining each VCI After adjusting the value, to ensure the frequency remains within the constraints, The following formula (15) is used for regulation;

[0067] (15)

[0068] In the formula, This is the power grid reference angular frequency of VCI.

[0069] Preferably, in step S1, each time window is set to 100ms.

[0070] Compared with traditional technologies, the microgrid transient stability emergency control method based on parameter rolling regulation provided by this invention does not require additional measurement devices, can guarantee reliable system recovery under any instability conditions, and has a fast recovery speed and small power fluctuations. This invention mainly combines the advantages of flexible VCI parameter adjustment and the powerful nonlinear mapping capabilities of deep learning, ensuring the proposed method can guarantee reliable and rapid microgrid recovery under any fault conditions from both mechanistic and algorithmic perspectives. Specifically, this invention derives a VCI regression stability criterion considering the current limiter based on the equal area ratio criterion, and proposes a microgrid control parameter rolling regulation strategy considering stability constraints, ensuring the reliability of the proposed method from a mechanistic perspective. Furthermore, this invention proposes an improved power trajectory prediction model based on graph convolutional neural networks, enabling the proposed microgrid control parameter rolling regulation strategy to adjust based on future trends, effectively improving the effectiveness of microgrid emergency control and ensuring the speed of control implementation. Attached Figure Description

[0071] Figure 1 This is a flowchart of the microgrid transient stability emergency control method based on parameter rolling regulation involved in the present invention;

[0072] Figure 2 This is a framework diagram of the microgrid transient stability emergency control method based on parameter rolling regulation involved in this invention;

[0073] Figure 3 The VSG power angle characteristic curve is shown in the figure considering the current limiter;

[0074] Figure 4 A schematic diagram of VSG control parameter adjustment strategies under three fault conditions;

[0075] Figure 5 This is a framework diagram of the power trajectory prediction model based on spatiotemporal graph convolutional neural network involved in the present invention;

[0076] Figure 6This is a schematic diagram of the three constraints in the microgrid transient stability emergency control method based on parameter rolling regulation involved in this invention on the VSG power angle characteristic curve;

[0077] Figure 7 The graph shows the active power output and power angle response characteristics of the VSG under a randomly selected transient scenario in an embodiment of the present invention. Detailed Implementation

[0078] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0079] Before describing the microgrid transient stability emergency control method based on parameter rolling regulation involved in this invention, it is worth noting that in order to achieve online emergency control of the microgrid and improve its stability, this invention needs to first analyze the conditions for VCI stabilization, which is detailed below.

[0080] 1. Determine the stabilization conditions of VCIs

[0081] VCI generally includes droop-controlled inverters and virtual synchronous control inverters. In terms of transient stability, droop-controlled inverters and VSGs are mathematically equivalent. Without loss of generality, this invention uses VSG as an example for analysis. To protect the VSG from being burned out by excessive fault current, a current limiter is required. When the system encounters a fault, the inverter enters a current-limiting state, and its swing equation is shown in equation (16):

[0082] (16)

[0083] In the formula, For VSG virtual inertia, The damping coefficient is... The reference active power of VSG, The output active power of the VSG. The voltage of the VSG. and These are the angular frequencies of the VSG and the power grid reference angular frequency, respectively. This is the equivalent resistance value of the line. For VSG, This is the current limiting value. It is the sum of the current limiting angle and the power angle.

[0084] Based on equation (16), the power angle curve of the VSG considering the current limiter can be plotted, such as... Figure 3As shown, power angle curves I / II / III represent the power angle curves under normal conditions, during a fault (in current-limited state), and after the fault is cleared (in current-limited state), respectively. Therefore, based on the equal area rule, the criterion for VSG regression stability can be derived as follows:

[0085] (17)

[0086] in, The power angle acceleration area of ​​the VSG during the fault. This represents the power angle reduction area of ​​the VSG after the fault is cleared. and These are the virtual inertia values ​​of VSG during the fault and after the fault is cleared, respectively.

[0087] It can be seen that as long as the power angle curve III is cleared at the time of fault clearance, There exists an equilibrium point; adjustments can be made after the fault is cleared. If the value is 0, then no deceleration area (VSG) is needed to return to stability. Therefore, , This is a key factor in ensuring the stability of VSG regression. In emergency control, it can be achieved through adjustment... and This allows VSG to return to stability.

[0088] 2. Analyze VCI stabilization strategies under various fault conditions.

[0089] Analysis revealed three possible scenarios for the VSG after the fault was cleared. The corresponding emergency control measures for each scenario are as follows:

[0090] 1) Case 1, the work angle curve III and There is an intersection (there is an equilibrium point) ), and the limit resection angle Within the equilibrium point. At this point, it is only necessary to address the issue after the fault. Adjusting it to 0 will change the running point from Run along the arrow on the line to Point, at The surrounding area has stabilized after oscillations. Figure 4 As shown in (a).

[0091] 2) Case 2, the work angle curve III and There is an intersection (there is an equilibrium point) ),but Not within the equilibrium point; at this time, After adjusting to 0, you need to set... Less than the fault clearing time .like Figure 4 As shown in (b), the run point will be determined by... Run along the arrow on the line to Point, at The area has stabilized after oscillations.

[0092] 3) Case 3, the work angle curve III and There is no intersection. At this point, After adjusting to 0, you still need to set it. Less than the fault clearing time However, if the active power reference value is adjusted too much at this time, it may cause the frequency fluctuation range to exceed [the specified range]. Then adjustments are needed. It is worth mentioning that, Increasing the value of VSG is more conducive to stabilization. For example... Figure 4 As shown in (c), the run point will be determined by... Run along the arrow on the line to Point, at The area has stabilized after oscillations.

[0093] It is evident that regulation and To achieve VSG stabilization, it is also necessary to be able to predict future time windows in real time. Continuous trajectory. Within the time window where future time can be obtained. Under the premise of trajectory, to ensure that the operating point returns to the intersection of power angle curve III and power angle curve I during the control period, a comprehensive control strategy of rolling control parameters can be adopted, rolling the control parameters once every certain period of time. In addition, since under any fault condition, if the VSG is to return to stability, its output power must first increase and then decrease, therefore, in the next time window... Once the reduction begins, regulation can be stopped.

[0094] Based on the above analysis, this invention proposes a microgrid transient stability emergency control method based on parameter rolling regulation, such as... Figure 1 and Figure 2 As shown, the method includes:

[0095] Step S1, setting There are several time windows, each lasting 100ms. This is the sequence number of the time window. ,initialization .

[0096] Step S2, collect Power angle of each VSG in the microgrid within the time window ,Voltage Output active power Output reactive power Actual values ​​and the control sequence used by the VSG controller The control sequence for:

[0097] (1)

[0098] In the formula, These are the virtual inertia, reference active power, and damping coefficient of the VSG, respectively. The number of VSGs.

[0099] Note: The electrical values ​​for time window 0 are the electrical quantities during the fault period, and the control sequence for time window 0 is the original control parameters of VSG.

[0100] Step S3: Input the data collected in step S2 into the power trajectory prediction model based on spatiotemporal graph convolutional neural network to predict... Active power output of each VSG within the time window .

[0101] As mentioned above, one of the keys to ensuring the effective execution of rolling control is the ability to accurately predict the next control time window. The trajectory. In this step, the present invention proposes a power trajectory prediction model based on a spatiotemporal graph convolutional neural network. This model can transfer the main computational burden of online emergency control to offline training, greatly reducing the online execution time of emergency control. Figure 5 As shown, the model consists of three parts: control sequence feature mining, time series spatiotemporal dynamic feature mining, and output power trajectory prediction. The processing procedures for each part are as follows.

[0102] 1. Control Sequence Feature Mining Section

[0103] This section The input feature within each time window is the control sequence. As shown in equation (1).

[0104] Regarding the first Control sequence within a time window Perform min-max normalization:

[0105] (2)

[0106] in:

[0107] (3)

[0108] Obtain control sequence Afterwards, Perform feature flattening, then reduce the dimensionality. Input a fully connected layer and use the forward propagation method to mine control sequence features. , The calculation formula is as follows:

[0109] (4)

[0110] In the formula, and For the first Weights and biases of neurons in fully connected layers; This represents the number of neurons in the first fully connected layer.

[0111] 2. Time Series Spatiotemporal Dynamic Feature Mining

[0112] In the Within a time window, the input features for this part include the electrical adjacency matrix. and composite time series Two parts. The electrical adjacency matrix is ​​one of them. The calculation process is as follows: First, calculate the impedance matrix of each node of the VSG according to the following formula. :

[0113] (11)

[0114] in, For the first The self-impedance of the VSG, For the first Taiwan VSG to the first The mutual impedance of the VSG, , , ;

[0115] Next, calculate the electrical distance between any two VSG nodes according to the following formula. ;

[0116] (12)

[0117] Then, the electrical distance is calculated according to the following formula. Perform normalization processing;

[0118] (13)

[0119] The electrical adjacency matrix is ​​obtained as shown in the following formula. .

[0120] (5)

[0121] In addition, the time series is , The length of the time series, where any time segment... On The mathematical expression is:

[0122] (6)

[0123] In the formula, , , , VSG in Output active power, output reactive power, voltage, and power angle for a time segment; The number of VSGs;

[0124] like Figure 5 As shown, in this step, two spatiotemporal convolutional modules are stacked to perform deep mining of spatiotemporal dynamic features. The mining process of each spatiotemporal convolutional module is as follows:

[0125] 1) After passing through the spatiotemporal attention layer, the hidden temporal and spatial correlations in the time series are adaptively modeled to obtain the temporal attention matrix and the spatial attention matrix.

[0126] 2) Multiplying the obtained time attention matrix yields a new time series based on time attention. .

[0127] 3) The new time series Spatial attention matrix and electrical adjacency matrix The inputs are fed into a spatial convolutional layer for spatial dynamic feature mining.

[0128] 4) Input into the temporal convolutional layer for temporal dynamic feature mining.

[0129] like Figure 5 As shown, the spatiotemporal convolution module also includes a residual module, which ensures model training performance and prevents gradient descent. The spatiotemporal dynamic features after deep mining are... The calculation formula is as follows:

[0130] (7)

[0131] In the formula, and These are spatiotemporal convolution modules. Temporal convolutional layer kernels and spatial convolutional layer kernels; For graph convolution operations, This is a standard convolution operation.

[0132] 3. Output power trajectory prediction section

[0133] Due to the aforementioned control sequence characteristics Size is spatiotemporal dynamic characteristics Size is , The number of filters in the temporal convolutional layer is not consistent with the size of the two features. Therefore, the control sequence features are determined here according to equations (8) and (9). Expanded to ;

[0134] (8)

[0135] (9)

[0136] In the formula, Total OK;

[0137] Then, the expanded Spatiotemporal dynamic characteristics Perform feature integration, and define the integrated features as follows: Finally, the obtained Input a standard two-dimensional convolutional layer to predict The power trajectory values ​​of each VSG within the time window; the power prediction value is defined as follows: The calculation formula is as follows:

[0138] (10)

[0139] In the formula, These are the parameters of the convolution kernel for this convolution module.

[0140] The training objective of the power trajectory prediction model is to make the predicted power trajectory prediction model more accurate. The problem involves minimizing the error from the actual value, which is a regression problem. Therefore, this invention selects MSE as the loss function for model training, and uses RMSE, MAE, and MAPE as evaluation metrics. These are commonly used evaluation metrics for regression problems, and their mathematical formulas will not be elaborated in this method. Once the model performance test results are satisfactory, the model is directly deployed for application.

[0141] Step S4: Determine the predicted output active power. Whether the stabilization criterion is met, the stabilization criterion being "with Compared to the actual value of the output active power within the time window, the predicted value is higher. "The output active power begins to decrease within the time window." If this condition is met, emergency control ends; otherwise, it continues. 0 pairs By rolling adjustment of two control parameters, all control sequences that can bring the microgrid back to stability can be obtained. This is because within each control time window... Within, the control sequence that satisfies rolling control , and The combinations are numerous, constituting an infinite set problem. To transform this infinite set problem into a finite set problem, this invention designs three types of constraints to obtain the optimal control sequence. For example... Figure 2 and Figure 6 As shown, the three constraints ensuring VSG regression stability are the equilibrium point constraint, the power angle constraint, and the frequency constraint. The equilibrium point constraint is the VSG power angle curve and the reference active power after fault clearance in the current-limited state. There is a balance point The power angle constraint is the point C on the VCI power angle curve corresponding to the limit cut-off angle after fault clearance, which is at the equilibrium point. Left side; Frequency constraint is that, provided the VSG can return to stability, the frequency deviation does not exceed To satisfy stability constraints and minimize the total amount of control inputs, the specific design of each control parameter within any control time window is as follows:

[0142] (15)

[0143] In the formula, For VSG in the Output active power within a time window This is the rated active power of the VCI. .

[0144] After determining each VSG After adjusting the value, the corresponding value can be determined. The control value is designed to ensure that the frequency is within the constraint range. as follows:

[0145] (16)

[0146] In the formula, This is the power grid reference angular frequency of the VSG.

[0147] Thus, the optimal control sequence that satisfies the constraints can be obtained. k Within the time window, the optimal control sequence is applied to each VSG controller.

[0148] Step S5, setting Then proceed to step S2 for a new round of judgment and rolling adjustment until the stabilization criterion is met, and then stop the adjustment.

[0149] To evaluate the effectiveness of the proposed microgrid transient stability emergency control method based on parameter rolling regulation, this embodiment conducts a case study verification on the improved Dong'ao Island 10-unit, 16-bus microgrid system. First, a power trajectory prediction model based on a spatiotemporal graph convolutional neural network is obtained for this system. Then, according to... Figure 2 The control framework shown is Figure 1 The illustrated process demonstrates the effectiveness of online emergency control applications. A randomly selected transient scenario is used as an example to showcase its effectiveness. In this transient scenario, the load level is 75%, the fault location is bus 5, and the fault duration is 0.1 seconds. Figure 7 The effectiveness of the proposed control strategy is demonstrated, proving that the proposed method can guarantee the microgrid's recovery to stability after a fault. Specifically, Figure 7 (a) shows the system power and power angle response curves when emergency control is not used. It can be seen that VSG3 and VSG4 will lose stability in this transient scenario. Figure 7 (b) The response curves of system power and power angle under the proposed emergency control method are shown. It can be seen that if the proposed emergency control is implemented, each VSG will return to stability after the fault is cleared, which proves the effectiveness of the emergency control method proposed in this invention.

[0150] The above embodiments are preferred implementations of the present invention. In addition, the present invention can be implemented in other ways. Any obvious substitutions without departing from the concept of the present technical solution are within the protection scope of the present invention.

[0151] To facilitate understanding by those skilled in the art of the improvements of this invention over the prior art, some of the accompanying drawings and descriptions have been simplified, and for clarity, some other elements have been omitted from this application. Those skilled in the art should realize that these omitted elements may also constitute the content of this invention.

Claims

1. A method for micro-grid transient stability emergency control based on parameter rolling regulation, characterized in that, include: Step S1: Set K time windows, where k is the sequence number of the time window, k=1, 2...K, and initialize k=1; Step S2: Collect the power angle δ, voltage V, and output active power P of each voltage-controlled inverter (VCI) in the microgrid within the k-1 time window. em Output reactive power Q em The actual value and the control sequence CS used by the VCI controller k-1 The control sequence CS k-1 for: (1) In the formula, J, P0, D p respectively, the virtual inertia, reference active power, damping coefficient of VCI; N is the number of VCIs; Step S3: Input the data collected in step S2 into the power trajectory prediction model based on spatiotemporal graph convolutional neural network. The power trajectory prediction model includes a control sequence feature mining part, a time series spatiotemporal dynamic feature mining part, and an output power trajectory prediction part, predicting the output active power P of each VCI within k time windows. em ; Step S4: Determine the predicted output active power P em Whether the stabilization criterion is met, wherein the stabilization criterion is "the output active power P within the k-1 time window". em Compared to the actual value, the predicted output active power P within the k-time window is... em "Start decreasing", if satisfied, then end emergency control; if not satisfied, then, under the premise of satisfying stability constraints, set J to 0 for P0 and D. p Two control parameters are adjusted in a rolling manner to obtain the optimal control sequence that can bring the microgrid back to stability. This optimal control sequence is then applied to each VCI controller within the k-time window. Stability constraints include equilibrium point constraints, power angle constraints, and frequency constraints. The equilibrium point constraint requires that, after fault clearance in current-limited conditions, the VCI power angle curve and the reference active power P0 have an equilibrium point D. The power angle constraint requires that the point C corresponding to the limiting cut-off angle on the VCI power angle curve after fault clearance is to the left of equilibrium point D. The frequency constraint requires that, provided the VCI can return to stability, the frequency deviation does not exceed [a certain value]. ; P0 is controlled using the following formula (14); (14) In the formula, Let VCI be the output active power during the k-th time window. P is the rated active power of this VCI. n =P0; After the P0 control value of each VCI is determined, in order to ensure that the frequency is within the constraint range, the D p The following formula (15) is used for control: (15) In the formula, is the grid reference angular frequency for the VCI; Step S5: Set k=k+1, then proceed to step S2 for a new round of judgment and rolling control until the stabilization criterion is met, and then stop the control.

2. The method of claim 1, wherein the method further comprises: In the step S3, the power trajectory prediction model based on the spatio-temporal graph convolutional neural network predicts the output active power P of each VCI in the k time window em The process is as follows: Step S31, control sequence feature mining part; Control sequence CS for the k-1th time window k-1 min-max normalization is performed: (2) in: (3) Obtain control sequence Afterwards, Perform feature flattening, then reduce the dimensionality. Input a fully connected layer and use the forward propagation method to mine control sequence features. , The calculation formula is as follows: (4) In the formula, and Let m1 be the weights and biases of the neurons in the i-th fully connected layer; m1 is the number of neurons in the first fully connected layer. Step S32, Time Series Spatiotemporal Dynamic Feature Mining Part; First, determine the electrical adjacency matrix A and the time series X within the k-1th time window k-1 , the electrical adjacency matrix A is: (5) In the formula, For electrical distance D ij The normalization result; The time sequence , is the time sequence length, wherein the mathematical expression of any one time segment t on (6) In the formula, , , , These represent the output active power, output reactive power, voltage, and power angle of the VCI during time segment t, respectively; N is the number of VCIs. Next, two spatiotemporal convolutional modules are stacked to perform deep mining of spatiotemporal dynamic features. The mining process of each spatiotemporal convolutional module is as follows: 1) to The hidden temporal correlation and spatial correlation in the time series are adaptively modeled through the spatiotemporal attention layer to obtain a temporal attention matrix and a spatial attention matrix. 2) multiplying the obtained time attention matrix with the time series to obtain a new time series based on time attention ;​ 3) new time series The spatial attention matrix is input into a spatial convolution layer together with the electrical adjacency matrix A to mine spatial dynamic features. 4) Input is fed into a temporal convolutional layer for temporal dynamic feature mining; The spatiotemporal dynamic features after in-depth analysis are: The calculation formula is as follows: (7) In the formula, and These are the temporal convolutional layer kernel and the spatial convolutional layer kernel in the spatiotemporal convolutional module i, respectively; For graph convolution operations, This is a standard convolution operation; Step S33, Output power trajectory prediction section; First, the control sequence feature H1 is expanded according to the following equations (8) and (9) ; (8) (9) In the formulae, Common Row; Then, the extended characteristics H2, the integrated characteristics are defined as ; Finally, the obtained The input two-dimensional standard convolution layer predicts the power trajectory value of each VCI in the k time window; the power prediction value is defined as The calculation formula is: (10) In the formula, These are the parameters of the convolution kernel in the convolution module.

3. The microgrid transient stability emergency control method based on parameter rolling regulation according to claim 2, characterized in that: In step S32, the calculation process of the electrical adjacency matrix A is as follows: First, the impedance matrix Z of each node of the VCI is calculated according to the following equation VCI : (11) Among them, Z ii Z is the self-impedance of the i-th VCI. ij Let be the mutual impedance between the i-th VCI and the j-th VCI, where 1 ≤ i ≤ N, 1 ≤ j ≤ N. ; Next, the electrical distance D between any two VCI nodes is calculated according to the following formula ij ; (12) Then, the electrical distance D is calculated according to the following formula ij normalization is performed; (13) This yields the electrical adjacency matrix A as shown in equation (5).

4. The method of claim 3, wherein the method further comprises: In step S1, each time window is set to 100ms.

Images