Transregional tie line power fluctuation suppression method based on deep learning

Abstract

The invention discloses a cross-regional tie line power fluctuation suppression method based on deep learning, and belongs to the field of deep learning. The method comprises the following steps: designing a tie line power fluctuation suppression controller: driving new energy to quickly adjust power, and dynamically suppressing tie line power fluctuation; calculation of active power modulation quantity distribution: constructing a voting-based decision-making mechanism, and reasonably distributing control measures of the new energy station; the gain value of the total modulation amount is calculated, specifically, the gain value is calculated by using the boosting regression integration algorithm with the minimum active modulation amount as the target. According to the method, the boosting algorithm and the voting mechanism are applied to tie line power control for the first time, compared with a traditional control method, the method has the advantages of being high in response speed, high in control precision and excellent in economical efficiency, and cascading failures caused by the fact that power transmission in the transient process exceeds the static stability limit after a tie line fault occurs are avoided; and the method has important significance for ensuring safe and stable operation of a large-scale interconnection system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of deep learning, especially to the field of deep learning related to power system, and in particular to a method for suppressing power fluctuation of cross-zone tie-line based on deep learning. BACKGROUND

[0002] During the transition period of UHV AC / DC power grid, the Central China-North China power grid is only connected by the single-circuit Changzhi-Nanyang-Jingmen 1000 kV UHV AC line (referred to as Changnan line), forming a typical long-distance and weakly connected two-terminal system. The internal AC fault of the Central China and North China power grid, as well as the in-zone DC fault, may cause large-scale power flow transfer in the Central China and North China power grid, and the peak value of power fluctuation of the Changnan line may exceed the static stability limit, leading to the splitting accident of the Central China-North China power grid (Zhou Xin, Sun Haishun, Zhao Bing, et al. A swing peak suppression strategy of UHV Tie-line power flow based on HVDC emergency control[J]. Proceedings of the CSEE, 2015, 35(10): 2392-2399.). It is difficult for the traditional security and stability control method based on the set of expected faults to cover complex faults such as multi-circuit DC successive commutation failure. On the other hand, simulation results show that under the full-load sending condition of Tianzhong DC, the power flow is transferred to the Changnan line after the DC fault, and the traditional generator tripping and load shedding control measures cannot meet the control requirements of the power grid, and there is a safety risk.

[0003] To cope with the above problems, a flexible AC / DC coordinated control system was built in 2016 in Central China Power Grid. The system can real-time distinguish the stability margin of power grid based on the response information of key branches, online calculate the amount of emergency control measures, and has the coordinated control function of 9 back HVDCs such as Longzhou, Yihua, and Linfeng (Shao Dejun, Xu Youping, Zhao Bing, et al. Application of flexible coordinated AC / DC control technology in central-China power grid [J]. Power System Technology, 2017, 41(4): 1146-1151 (in Chinese)). Based on the power fluctuation of Changnan line collected by WAMS, the system analyzes the stability margin of tie-line online, absorbs or compensates the excess or deficiency power of the connected AC system through HVDC power modulation, and suppresses the random power fluctuation on the AC tie-line (He Jian, Sun Huadong, Guo Jianbo, et al. Suppressing AC tie-line stochastic power fluctuation by HVDC power modulation control [J]. Proceedings of the CSEE, 2013, 33(25): 93-98) (Xu Shiyun, Wu Ping, Zhao Bing, et al. Coordinated control strategy of interconnected grid integrated with UHVDC transmission line from Hami to Zhengzhou [J]. Power System Technology, 2015, 39(7): 1773-1778).

[0004] As an emergency control means after the power grid withstands transient unbalanced power impact, the direct current emergency power modulation has the advantages of strong power controllability and fast climbing speed (hundred milliseconds level). (Yang Weidong, Xue Yusheng, Jing Yong, et al. Emergency DC power support to AC power system in the South China Power Grid [J]. Automation of Electric Power System, 2003, 27(17): 68-72.) The principle of using direct current power to improve the stability of alternating current system is analyzed, and the influence of direct current power control, control time and power change rate on stability is analyzed. (Hu Yi, Teng Yufei, Wang Xiaoru. The steady-state frequency control strategy of power grid with multi-send & multi-infeed HVDCs based on wide-area measurement [J]. Power System Technology, 2018, 42(1): 25-33.) A multi-loop direct current emergency power support coordination control algorithm based on machine learning is designed, which realizes the system frequency stability by using direct current emergency power support and load shedding, but only considers the direct current overload capacity constraint and regional power grid frequency constraint when calculating the control quantity. On this basis, (Xu Tao, Wu Xuelian, Li Zhaowei, et al. Coordinated control strategy of Multi-DC emergency power support to improve frequency stability of power systems [J]. Automation of Electric Power Systems, 2018, 42(22): 69-77, 143.) A multi-direct current power emergency support coordination control strategy is constructed considering the voltage constraint of converter station alternating current bus and the power flow dispersion ability of alternating current power grid.(Peng Long, Tang Yong, Zhao Bing, et al. LCC-HVDC Emergency Power Support Control Considering the DC Real-time Available Capacity[J]. Power System Technology, 2021, 45(11): 4470-4477.), (Suo Zhiwen, Li Hui, Yu Zhao, et al. Research on characteristics and calculation method of HVDC transmission after DC power fast ramp[J]. Power System Technology, 2018, 42(12): 3833-3841.) proposed the calculation method of emergency power regulation capacity considering the constraint of DC commutation failure and the constraint of rectifier trigger angle, respectively, further taking the stability domain of DC equipment itself into constraint, (Suo Zhiwen, Li Hui, Yu Zhao, et al. Research on characteristics and calculation method of HVDC transmission after DC power fast ramp[J]. Power System Technology, 2018, 42(12): 3833-3841.) also pointed out that when the DC power instruction is too large, the inverter turns into current control, the DC power rises slowly, and it is difficult to meet the requirement of fast rise. It can be seen that the power support that DC can provide is not unlimited, but the AC power grid connected at both ends and the stability domain of its own equipment should be considered.

[0005] With the rapid increase in the proportion of new energy access, the active participation of new energy in system stability control has become an inevitable trend. As a core component of new energy equipment, the inverter has millisecond-level fast power control capability. In the short term, if new energy power plants can quickly respond to the system's active power control commands and participate in the transient stability control of the power grid, the safety level of the power system can be significantly improved (Gao Bingtuan, Hu Zhengyang, Wang Weisheng, et al. Review on fastactive power control and frequency support technologies of renewable energy stations[J]. Proceedings of the CSEE, 2024, 44(11): 4335-4352.). At present, the industry has carried out some research on the participation of new energy in transient stability support (Wang Shuchao, Sun Guanghui, Yu Chengsheng, et al. Photovoltaic power generation system level rapid power control technology and its application[J]. Proceedings of the CSEE, 2018, 38(21): 6254-6263.). (Wang Shuchao, Duan Shengpeng, Wang Jian, et al. Research and practice of fast frequency response oriented control optimization technology of PV powerstations[J]. Power System Protection and Control, 2019, 47(14): 59-70.) (YANWei, WANG Shuchao, LIU Xiang, et al. Systemlevel PV fast power control technology and application[J]. Proceedings of the CSEE, 2019, 39(S1): 213-224.) proposed a leapfrog MPPT power execution algorithm for inverters, which improved the power execution speed of photovoltaic inverters. Application tests were carried out in photovoltaic power stations in Inner Mongolia and Tibet. The test results showed that the time taken for photovoltaic inverters to perform power increase / decrease was 15-20ms.(Wang Shuchao, Sun Guanghui, Yu Chengsheng, et al. Photovoltaic power generation system level rapid power control technology and its application[J]. Proceedings of the CSEE, 2018, 38(21): 6254-6263.) A rapid power control system for photovoltaic power plants was designed. The results of its application in the Tibet power grid show that the system can achieve a power response speed of about 30ms for the entire station and about 60ms for the entire network. (Gao Bingtuan, Hu Zhengyang, Wang Weisheng, et al. Review on fast active power control and frequency support technologies of renewable energy stations[J]. Proceedings of the CSEE, 2024, 44(11): 4335-4352.) It is pointed out that wind turbines contain a large number of mechanical inertial components, resulting in a longer fast active control time, with a response time of less than 200ms. The above studies show that, in terms of response speed, photovoltaic power plants have met the requirements for participating in the transient stability control of the system.

[0006] In general, when grid failures such as outages or DC blocking cause significant power imbalances and large fluctuations in tie-line power, the system can implement measures such as emergency DC power boosting, load shedding, disconnecting pumped storage units, and rapid power control of renewable energy sources. However, load shedding strategies have significant negative impacts and should be avoided whenever possible. DC power modulation must consider the grid support capacity at both the sending and receiving ends, as well as the stability domain of the equipment itself. Photovoltaics offer flexible control capabilities and rapid power support during power-limited operation, making them an effective means of controlling power surplus. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide a deep learning-based method for suppressing power fluctuations in cross-regional tie lines. This method modulates the power output of renewable energy generating units to smooth out power surges in tie lines, preventing power delivery during transient processes after a tie line fault from exceeding its static stability limit and triggering a cascading failure.

[0008] The objective of this invention is achieved through the following technical solution: a deep learning-based method for suppressing cross-regional tie-line power fluctuations, comprising the following steps: (1) Based on the tie-line power fluctuation suppression controller, drive the new energy to quickly adjust the power and dynamically suppress the tie-line power fluctuation; (2) Obtain the set of disturbance faults, generate the active power fluctuation curve through the tie line power fluctuation suppression controller, and calculate the active power allocation scheme through the voting-based multi-classification algorithm; (3) Obtain the disturbance fault set, generate the active power fluctuation curve through the tie-line power fluctuation suppression controller, and calculate the modulation total gain value K through the boosting-based ensemble regression algorithm. p This is to optimize the parameters of the tie-line power fluctuation suppression controller.

[0009] Further, step (1) includes the following sub-steps: (1.1) Power modulation stage: to dampen power fluctuations in the tie line, including low-pass filter stage, DC blocking stage, gain / filter stage, lead / lag stage, and limiting stage; (1.2) System Action Trigger / Exit Link: The function of the system action trigger / exit link is to prevent the controller from malfunctioning. It only activates when the cross-zone tie line is operating at high power in the initial operating condition and the power fluctuation reaches a certain threshold after the fault. The system action trigger / hold link includes three structures: power operation level dead zone link, power fluctuation dead zone link, and controller trigger / exit link.

[0010] Further, step (1.1) includes the following sub-steps: (1.1.1) Low-pass filter stage: used to perform low-pass filtering on the active power of cross-regional tie lines; (1.1.2) DC blocking element: used to filter out the DC component of active power in cross-regional interconnection lines; (1.1.3) Gain / Filtering Stage: Used to adjust the controller gain and filter the output of the DC blocking stage; (1.1.4) Lead and lag elements: used to compensate for time delay and phase of the controller output.

[0011] Further, step (1.2) includes the following sub-steps: (1.2.1) Power operation level dead zone: When the cross-zone tie line power is lower than the preset value, the system has a large stability margin and no power modulation is required; by setting a high power threshold, the controller will only start when it detects that the cross-zone tie line power level has reached the high power threshold. (1.2.2) Power fluctuation dead zone: used to avoid natural power fluctuations in cross-regional tie lines; when there are natural power fluctuations in the transmission line, the controller does not need to be adjusted; the power dead zone criterion is to calculate the long-term average power as the steady-state power of the system and set the action dead zone value of the fluctuation amount to avoid natural power fluctuations in the tie line. (1.2.3) Controller trigger / exit mechanism: By setting the trigger time, the device can effectively avoid short-term transient disturbances and ensure that the device does not operate reliably; by setting the hold time, the device is always in an operating state during the transient and dynamic continuous process after a large disturbance in the power grid, continuously controlling the power fluctuation of the tie line and not exiting prematurely.

[0012] Furthermore, step (2) includes the following sub-steps: (2.1) Sample Generation: In the simulation software, a large number of disturbance faults are generated for the case study to form an offline fault set. The active power fluctuation curve of the tie line after the fault is output. The tie line power fluctuation curve is input into the controller, and different faults are obtained by iterating through the data. K p The active power modulation curve under the given value is calculated. The peak value at the abrupt change in the active power modulation curve is taken as the total amount of new energy modulation, Pmod, and the abrupt change time is taken as the modulation time, Tmod. Different active power allocation schemes are set. In the simulation software, different active power allocation schemes corresponding to different faults are traversed, and the active power fluctuation curve of the tie line after modulation is obtained through simulation. The value of the objective function is calculated, and the optimal allocation scheme corresponding to each fault is determined.

[0013] (2.2) Based on the training set obtained in step (2.1), a multi-class ensemble algorithm based on voting is performed. The input of the multi-class ensemble algorithm is the active power fluctuation curve, and the output is the optimal allocation scheme.

[0014] Furthermore, the voting-based multi-class ensemble algorithm specifically involves: integrating different prediction models and assigning different weights to different prediction models based on their prediction accuracy and stability; and after multiple rounds of iteration and tuning, obtaining the weight configuration that optimizes the overall prediction performance of the integrated model.

[0015] Furthermore, step (3) includes the following sub-steps: (3.1) Sample generation: In the simulation software, a large number of disturbance faults are generated for the case study to form an offline fault set. The active power fluctuation curve of the tie line after the fault is output. The tie line power fluctuation curve is input into the controller, and different K values ​​are obtained by iterating through the data. pThe active power modulation curve under the given value is calculated. The peak value at the abrupt change in the active power modulation curve is taken as the total amount of new energy modulation, Pmod, and the abrupt change time is taken as the modulation time, Tmod. Different active power allocation schemes are set. In the simulation software, different active power allocation schemes corresponding to different faults are traversed, and the active power fluctuation curve of the tie line after modulation is obtained through simulation. The value of the objective function is calculated, and the optimal allocation scheme corresponding to each fault is determined.

[0016] (3.2) Based on the training set obtained in step (3.1), a boosting-based regression ensemble algorithm is performed. The input of the regression ensemble algorithm is the active power fluctuation curve, and the output is the optimal K. p value.

[0017] Furthermore, the boosting-based regression ensemble algorithm specifically involves: training multiple basic models sequentially, attempting to correct misclassified samples from the previous training round in each training iteration, and obtaining the weight configuration that optimizes the overall predictive performance of the model after multiple iterations and tuning.

[0018] The present invention also provides an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the deep learning-based cross-regional tie-line power fluctuation suppression method described above.

[0019] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the deep learning-based cross-regional tie-line power fluctuation suppression method described above.

[0020] The beneficial effects of this invention are as follows: 1. The hybrid enhanced intelligent safety and stability control system designed in this paper can effectively avoid natural power fluctuations in cross-regional tie lines. After a large disturbance occurs in the system, it can generate an emergency control strategy for the active power of new energy power plants by using only real-time tie line power response information.

[0021] 2. Both the controller gain and the power allocation scheme affect the suppression of tie-line power fluctuations. A higher gain results in faster recovery from tie-line power fluctuations but may cause oscillations, while a lower gain weakens the modulation effect. Different power allocation schemes affect the modulation effect of the initial swing of tie-line power fluctuations. By selecting a better allocation scheme, the amplitude of the initial swing of power fluctuations can be reduced under the same emergency control parameters.

[0022] 3. Employing voting-based multi-class ensemble and regression ensemble algorithms enables intelligent parameter tuning of the controller gain and prediction of the optimal power allocation scheme. The ensemble algorithm combines multiple prediction models through intelligent weight allocation technology, effectively avoiding the drawbacks of traditional single models and improving overall prediction performance and generalization ability.

[0023] In summary, this invention proposes an emergency control strategy to suppress power fluctuations in cross-regional tie lines with the goal of minimizing the amount of control measures. It establishes an optimal allocation scheme for emergency control measures and designs a tie line head-swing power control system based on hybrid enhanced intelligence. This system enables the smoothing of tie line head-swing power impacts by modulating the power output of new energy generating units, thus preventing the power transmission during the transient process after a tie line fault from exceeding its static stability limit and causing cascading failures. This is of great significance for ensuring the safe and stable operation of large-scale interconnected systems. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a design diagram for a tie-line power fluctuation suppression controller; Figure 2 This is the CEPRI-197 node system diagram; Figure 3 This is a graph showing the active power fluctuation of the tie line after a fault is set in the CEPRI-197 node system. Figure 4 This is a comparison chart of the active power fluctuation curves of the tie line after modulation with the default parameter scheme. Figure 5 This is a comparison chart of the active power modulation curves of new energy sources after using the modulation scheme of this study. Detailed Implementation

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0027] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0028] The present invention will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0029] The core technology of this invention is the design and parameter optimization of a tie-line power fluctuation suppression controller, in order to achieve an emergency control strategy for suppressing cross-regional tie-line power fluctuations with the goal of minimizing the amount of control measures.

[0030] This invention proposes a deep learning-based method for suppressing cross-regional tie line power fluctuations, comprising the following steps: Step 1: Based on the tie-line power fluctuation suppression controller, drive the new energy source to quickly adjust its power and dynamically suppress tie-line power fluctuations. The tie-line power fluctuation suppression controller is as follows: Figure 1 As shown; Step 2: Obtain the disturbance fault set, generate the active power fluctuation curve through the tie-line power fluctuation suppression controller, and calculate the active power allocation scheme through a voting-based multi-classification algorithm; different allocation schemes will have a direct impact on the power modulation effect. By selecting a better allocation scheme, the emergency control quantity can be effectively reduced under the same modulation effect. Step 3: Obtain the disturbance fault set, generate the active power fluctuation curve through the tie-line power fluctuation suppression controller, and calculate the modulation total gain value K using an ensemble regression algorithm based on boosting. p This is to optimize the parameters of the tie-line power fluctuation suppression controller.

[0031] As a preferred embodiment, the tie-line power fluctuation suppression controller described in step one includes the following components: (1.1) Power Modulation Stage: The power modulation stage dampens power fluctuations in the tie line, including a low-pass filter, DC blocking, gain / filtering, lead / lag, and limiting stage. Its transfer function formula is as follows: ; Among them, P mod P represents the total modulation amount. ac K represents the power fluctuation value of the inter-regional tie line. p For regulator gain, T mes T is the time constant of the low-pass filter stage.W T1 is the time constant of the DC blocking stage, T2 is the time constant of the gain / filtering stage, T3 and T4 are the time constants of the first lead-lag stage, and T4 are the time constants of the second lead-lag stage. The unit is seconds.

[0032] (1.2) System action triggering / exit mechanism: The function of the system action triggering / exit mechanism is to prevent the controller from malfunctioning. It only operates when the cross-zone tie line is in high power operation under the initial operating condition and the power fluctuation reaches a certain threshold value after the fault.

[0033] Furthermore, the tie-line power fluctuation suppression controller also includes: a system action triggering / holding circuit comprising three structures: a power operating level dead zone circuit, a power fluctuation dead zone circuit, and a controller triggering / exit circuit, specifically: (1.2.1) Power operation level dead zone When the power of the inter-zone tie line is at a low level, the system has a large stability margin, eliminating the need for power modulation and avoiding wasted resources. By setting a high power threshold, the controller only activates when it detects that the power level of the inter-zone tie line has reached a certain value. The criterion is as follows: if the formula is satisfied, it indicates that the AC tie line power threshold condition is met.

[0034] ; Among them, P minac The minimum power threshold for cross-regional interconnects to be used by the controller.

[0035] (1.2.2) Power fluctuation dead zone Power dead-time elements are used to avoid natural power fluctuations (P) in cross-regional interconnects. 自然 During natural power fluctuations in transmission lines, the controller requires no adjustment. The power dead zone criterion calculates the long-term average power as the system's steady-state power and sets a dead zone value for the fluctuation to avoid natural power fluctuations in the tie line. The formula for calculating the fluctuation difference is as follows: ; The formula for the power dead zone criterion is as follows: ; Among them, △P ac This is the fluctuation difference. P is the length of the time window. minflu The dead zone value for the power fluctuation of the inter-regional tie line is P. 大扰动 >P minflu >P 自然 When the formula is satisfied, it indicates that the AC tie-line power dead zone condition is met.

[0036] (1.2.3) Controller Trigger / Exit Phase The trigger-and-hold mechanism is maintained by setting the trigger time T. trig This ensures the device effectively avoids short-term transient disturbances and reliably prevents it from malfunctioning. This is achieved by setting a hold time T. hold This ensures the device remains operational throughout the transient and dynamic processes following a major power grid disturbance, continuously controlling tie-line power fluctuations and preventing premature shutdown. The controller trigger criteria are as follows: ; in, This refers to the time during which both the power operating level dead zone criterion and the power fluctuation dead zone criterion are simultaneously satisfied. The controller exit criterion is that, after the controller is triggered, the time during which either the power operating level dead zone or the power fluctuation dead zone criterion fails to meet the requirement is greater than the hold time T. hold The controller exits operation.

[0037] In a preferred embodiment, step two further includes the following sub-steps: (2.1) Scheme Selection: Currently, when studying active power allocation schemes for renewable energy power plants, the industry typically adopts average allocation or weighted allocation based on adjustable capacity. No research has yet considered the distance factor between the renewable energy power plant and the fault location. This invention considers the distance factor and power allocation weights of the renewable energy power plant and proposes four different allocation schemes: Option 1: Allocate new energy power stations within the region based on their adjustable capacity using a weighted average method. Option 2: Weighted allocation based on the distance between the new energy power station and the fault point; Option 3: Randomly select new energy power stations within the region and distribute them evenly; Option 4: Randomly select new energy power stations within the region and allocate them according to the weighted distance between the power station and the fault point.

[0038] The first two approaches are based on model-based cognition, while the latter two are based on data-driven cognition. By integrating these heterogeneous strategies, the system can dynamically balance deterministic rules with implicit data patterns, thereby improving the robustness of allocation under different fault scenarios.

[0039] (2.2) Sample Generation: In the simulation software, a large number of disturbance faults are generated for the case study to form an offline fault set. The active power fluctuation curve of the tie line after the fault is output. The tie line power fluctuation curve is input into the controller, and different K values ​​are obtained by iterating through the data. p The active power modulation curve under the given value. The peak value at the moment of abrupt change in the active power modulation curve is taken as the total amount of new energy modulation P. mod The moment of abrupt change in the curve is taken as the modulation time T. modDifferent active power allocation schemes are set up. The different active power allocation schemes corresponding to different faults are traversed in the simulation software to obtain the active power fluctuation curves of the modulated tie line. The objective function value is calculated to determine the optimal allocation scheme for each fault.

[0040] (2.3) Selection of Training Model: The selection of an active power allocation scheme is a multi-class classification problem. Current mainstream algorithms for solving this problem include decision trees, support vector machines, K-Nearest Neighbor (KNN) algorithms, and deep neural network algorithms. The core advantage of decision trees lies in their inherent support for multi-class splitting, requiring no additional strategies. Furthermore, they can intuitively reveal classification rules through feature thresholding, combining efficiency and interpretability, but are prone to overfitting. Support vector machines excel in solving high-dimensional problems and avoiding overfitting, but their disadvantage is that they may perform poorly with a large amount of missing data. KNN is a distance-based classification algorithm that predicts the label of a new data point based on the label information of its "K" nearest neighbors, but it is sensitive to imbalanced data, where minority class samples are easily overwhelmed by the majority class. Deep neural networks can handle large-scale, high-dimensional datasets and can uncover deep-seated structures and patterns in the data, but their model interpretability is poor, making it difficult to analyze the decision-making process.

[0041] Preferably, this invention proposes a multi-class ensemble algorithm based on Voting. This method combines the advantages of multiple models, improving overall prediction performance and generalization ability while ensuring model stability. The logic behind the ensemble method is that multiple prediction models combined often perform better than any single model because hierarchical models allow for cross-validation of results, effectively avoiding the drawbacks of a single model.

[0042] Specifically, the Voting-based multi-class ensemble algorithm employs intelligent weight allocation technology. This means that not all models are assigned the same weight, but rather appropriate weights are given based on their predictive accuracy and stability. By introducing this intelligent allocation strategy, the contribution of each model to the prediction results can be more precisely controlled. During the weight allocation phase, models with better predictive performance are given larger weights. Through multiple rounds of iteration and tuning, the optimal weight configuration that maximizes the overall predictive performance of the model is found. This intelligent weight allocation technology enhances the model's generalization ability under different data distributions and application scenarios, enabling the ensemble model to outperform single algorithms in complex multi-class classification tasks.

[0043] (2.4) Training data: Based on the training set obtained in step (2.2), a multi-class ensemble algorithm based on voting is performed. The input of the multi-class ensemble algorithm is the active power fluctuation curve, and the output is the optimal allocation scheme.

[0044] In a preferred embodiment, step three further includes the following sub-steps: (3.1) Sample generation: In the simulation software, a large number of disturbance faults are generated for the case study to form an offline fault set. The active power fluctuation curve of the tie line after the fault is output. The tie line power fluctuation curve is input into the controller, and different K values ​​are obtained by iterating through the data. p The active power modulation curve under the given value. The peak value at the moment of abrupt change in the active power modulation curve is taken as the total amount of new energy modulation P. mod The moment of abrupt change in the curve is taken as the modulation time T. mod Different active power allocation schemes are set up. The different active power allocation schemes corresponding to different faults are traversed in the simulation software to obtain the active power fluctuation curves of the modulated tie line. The objective function value is calculated to determine the optimal allocation scheme for each fault.

[0045] (3.2) Training Model Selection: Mainstream regression algorithms include LASSO regression, Ridge regression, and deep learning regression. LASSO regression, by introducing an L1 norm penalty term, can not only effectively handle the collinearity problem but also has feature selection capabilities, thereby enhancing the interpretability of the model. However, it suffers from instability in the selection of strongly correlated variables. Ridge regression uses an L2 norm penalty term, which can stably handle highly correlated independent variables, but lacks a variable selection mechanism. Deep learning regression methods, through multi-layer neural network architectures, exhibit powerful nonlinear modeling capabilities, but their "black box" characteristics lead to poor model interpretability.

[0046] Preferably, this invention proposes a boosting-based regression ensemble method. The boosting method trains multiple basic models sequentially, attempting to correct misclassified samples from the previous training round in each iteration. Through multiple rounds of iteration and tuning, the optimal weight configuration for overall model prediction performance is found, thereby gradually improving the model's predictive performance. The root mean square error (RMSE) is used as the criterion for evaluating model performance. The smaller the RMSE value, the smaller the difference between the model's predicted values ​​and the actual observed values, indicating better model performance. The RMSE formula is shown below: ; Where y i Indicates the optimal K p value, K represents the model's prediction. p Values. Boosting-based regression ensemble methods allow for the free selection of the base model based on the data type to be predicted. This paper selects the LASSO regression, ridge regression, and deep learning regression models described above.

[0047] (3.3) Training data: Based on the training set obtained in step (3.1), a regression ensemble algorithm is performed. The input of the regression ensemble algorithm is the active power fluctuation curve, and the output is the optimal K. p value.

[0048] This invention also provides an electronic device, including a memory and a processor, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the deep learning-based cross-regional tie-line power fluctuation suppression method described in any embodiment.

[0049] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the deep learning-based cross-regional tie-line power fluctuation suppression method described in any embodiment.

[0050] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be any data processing device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0051] As a preferred embodiment, the present invention uses the CEPRI-197 node system as an example, and its network structure is as follows: Figure 2 As shown. The simulation tool is PSD power system analysis software. It is set that three photovoltaic units, Gen1, Gen3, and Gen4, disconnect from the grid at 1 second. The power fluctuation curve of the tie line after the fault is shown in the figure. Figure 3 As shown. The tie line initially had an active power of 1279MW, which rose to a maximum of 1390MW after the fault. The controller was set with initial parameters of gain Kp = 0.3 and power distribution scheme of equal modulation distribution to all renewable energy units in the region. After inputting the active power curve into the controller, the modulation amount Pmod was 279.4MW and the modulation time Tmod was 1.27s. A comparison of the active power fluctuations of the tie line before and after modulation is shown below. Figure 4 As shown, the blue line represents areas without control measures, and the black line represents areas where emergency control measures have been implemented. Figure 4It can be seen that the active power fluctuation of the line was improved after modulation, but the effect of suppressing the power fluctuation of the tie line first swing was not obvious.

[0052] The optimal power allocation scheme and the optimized K are predicted using a single learning model and the intelligent model proposed in this paper, respectively. p Value. The active power fluctuation of the tie line under different schemes, for example... Figure 5 As shown in the figure, the black line represents no control measures, the red line represents the initial scheme, the green line represents the single learning model scheme, and the blue line represents the intelligent scheme. The figure shows that the modulation amount predicted by the single learning model or the intelligent model is smaller than that of the original scheme, but the effect of suppressing the initial swing fluctuation is stronger. Compared with the single learning model, although the modulation amount predicted by the intelligent model is the same, the suppression effect of the initial swing is stronger due to the different power allocation scheme.

[0053] In summary, the tie-line power fluctuation suppression controller can effectively suppress tie-line power fluctuations. Both the gain value and the power distribution scheme affect the suppression of tie-line power fluctuations. The proposed multi-class ensemble algorithm and regression ensemble algorithm can effectively enhance the modulation effect through intelligent parameter tuning.

[0054] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process may be rearranged without departing from the scope of this disclosure. The appended method claims provide elements of various steps in an exemplary order and are not intended to limit the scope to the specific order or hierarchy described.

Claims

1. A method for suppressing power fluctuations in cross-regional tie lines based on deep learning, characterized in that, Includes the following steps: (1) Based on the tie-line power fluctuation suppression controller, drive the new energy to quickly adjust the power and dynamically suppress the tie-line power fluctuation; (2) Obtain the set of disturbance faults, generate the active power fluctuation curve through the tie line power fluctuation suppression controller, and calculate the active power allocation scheme through the voting-based multi-classification algorithm; (3) Obtain the disturbance fault set, generate the active power fluctuation curve through the tie-line power fluctuation suppression controller, and calculate the modulation total gain value K through the boosting-based ensemble regression algorithm. p This is to optimize the parameters of the tie-line power fluctuation suppression controller.

2. The method for suppressing cross-regional tie line power fluctuations based on deep learning according to claim 1, characterized in that, Step (1) includes the following sub-steps: (1.1) Power modulation stage: to dampen power fluctuations in the tie line, including low-pass filter stage, DC blocking stage, gain / filter stage, lead / lag stage, and limiting stage; (1.2) System action trigger / exit mechanism: used to prevent the controller from malfunctioning. It only activates when the cross-zone tie line is in high power operation under initial operating conditions and the power fluctuation reaches a preset threshold value after a fault.

3. The method for suppressing cross-regional tie-line power fluctuations based on deep learning according to claim 2, characterized in that, Step (1.1) includes the following sub-steps: (1.1.1) Low-pass filter stage: used to perform low-pass filtering on the active power of cross-regional tie lines; (1.1.2) DC blocking element: used to filter out the DC component of active power in cross-regional interconnection lines; (1.1.3) Gain / Filtering Stage: Used to adjust the controller gain and filter the output of the DC blocking stage; (1.1.4) Lead and lag elements: used to compensate for time delay and phase of the controller output.

4. The method for suppressing cross-regional tie line power fluctuations based on deep learning according to claim 2, characterized in that, Step (1.2) includes the following sub-steps: (1.2.1) Power operation level dead zone: When the cross-zone tie line power is lower than the preset value, the system has a large stability margin and no power modulation is required; by setting a high power threshold, the controller will only start when it detects that the cross-zone tie line power level has reached the high power threshold. (1.2.2) Power fluctuation dead zone: used to avoid natural power fluctuations in cross-regional tie lines; when there are natural power fluctuations in the transmission line, the controller does not need to be adjusted; the power dead zone criterion is to calculate the long-term average power as the steady-state power of the system and set the action dead zone value of the fluctuation amount to avoid natural power fluctuations in the tie line. (1.2.3) Controller trigger / exit mechanism: By setting the trigger time, the device can effectively avoid short-term transient disturbances and ensure that the device does not operate reliably; by setting the hold time, the device is always in an operating state during the transient and dynamic continuous process after a large disturbance in the power grid, continuously controlling the power fluctuation of the tie line and not exiting prematurely.

5. The method for suppressing cross-regional tie line power fluctuations based on deep learning according to claim 1, characterized in that, Step (2) includes the following sub-steps: (2.1) Sample Generation: In the simulation software, a large number of disturbance faults are generated for the case study to form an offline fault set. The active power fluctuation curve of the tie line after the fault is output. The tie line power fluctuation curve is input into the controller, and different faults are obtained by iterating through the data. K p The active power modulation curve under the value; take the peak value at the moment of sudden change of the active power modulation curve as the total amount of new energy modulation Pmod, and take the moment of sudden change of the curve as the modulation time Tmod; set different active power allocation schemes; traverse different active power allocation schemes corresponding to different faults in the simulation software, and simulate to obtain the active power fluctuation curve of the modulated tie line. Calculate the value of the objective function and determine the optimal allocation scheme for each fault; (2.2) Based on the offline fault set, a voting-based multi-class ensemble algorithm is performed. The input of the multi-class ensemble algorithm is the active power fluctuation curve, and the output is the optimal allocation scheme.

6. The method for suppressing cross-regional tie line power fluctuations based on deep learning according to claim 5, characterized in that, The voting-based multi-class ensemble algorithm specifically involves: integrating different prediction models and assigning different weights to different prediction models based on their prediction accuracy and stability. After multiple rounds of iteration and tuning, the weight configuration that optimizes the overall prediction performance of the integrated model is obtained.

7. The method for suppressing cross-regional tie line power fluctuations based on deep learning according to claim 1, characterized in that, Step (3) includes the following sub-steps: (3.1) Sample generation: In the simulation software, a large number of disturbance faults are generated for the case study to form an offline fault set. The active power fluctuation curve of the tie line after the fault is output. The tie line power fluctuation curve is input into the controller, and different K values ​​are obtained by iterating through the data. p The active power modulation curve under the value; take the peak value at the moment of sudden change of the active power modulation curve as the total amount of new energy modulation Pmod, and take the moment of sudden change of the curve as the modulation time Tmod; set different active power allocation schemes; traverse different active power allocation schemes corresponding to different faults in the simulation software, and simulate to obtain the active power fluctuation curve of the modulated tie line. Calculate the value of the objective function and determine the optimal allocation scheme for each fault; (3.2) Based on the offline fault set, a boosting-based regression ensemble algorithm is performed. The input of the regression ensemble algorithm is the active power fluctuation curve, and the output is the optimal K. p .

8. The method for suppressing cross-regional tie line power fluctuations based on deep learning according to claim 7, characterized in that, The boosting-based regression ensemble algorithm is specifically designed as follows: by training multiple basic models sequentially, each training attempt is made to correct the samples that were misclassified in the previous training round. After multiple rounds of iteration and tuning, the weight configuration that optimizes the overall prediction performance of the model is obtained.

9. An electronic device comprising a memory and a processor, characterized in that, The memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the deep learning-based cross-regional tie-line power fluctuation suppression method as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the deep learning-based cross-regional tie-line power fluctuation suppression method as described in any one of claims 1-8.

Images