An emergency control method and system for power grid voltage stability based on CNN-LSTM combined network

Abstract

This invention relates to the field of power system technology and discloses a method and system for emergency control of grid voltage stability based on a CNN-LSTM combined network. It integrates the advantages of Convolutional Neural Networks (CNN) in feature extraction and Long Short-Term Memory Networks (LSTM) in learning data temporal dependencies, establishing a voltage stability margin assessment model based on the CNN-LSTM combination. This model achieves accurate assessment of grid voltage stability margin under large disturbance faults, with higher prediction accuracy and results closer to the true value compared to classical neural networks. It can accurately characterize the sensitivity of each bus voltage to generator tripping and load shedding control measures, providing strong evidence and guidance for formulating optimal emergency control strategies. This effectively and accurately assesses grid voltage stability emergency control, improves prediction accuracy, and ensures the safe and stable operation of the grid after large disturbance faults.

Description

Technical Field

[0001] This invention relates to the field of power system technology, specifically to an emergency control method and system for grid voltage stability based on a CNN-LSTM combined network. Background Technology

[0002] Against the backdrop of the "dual carbon" target, my country's wind power, photovoltaic, and other new energy sources have experienced leapfrog growth, and high-voltage direct current (HVDC) transmission technology has been widely applied, forming the world's largest and most complex AC / DC hybrid power grid. However, with a large number of new energy sources replacing traditional synchronous generators with large mechanical inertia and strong anti-interference capabilities through power electronic equipment, the inherent weak inertia of these generators will reduce the grid's damping characteristics, significantly decrease its voltage regulation capacity, and weaken its anti-interference ability, posing a significant challenge to the safe and stable operation of the power system. When a serious fault occurs on the AC trunk line, large-scale power flow transfer may lead to grid voltage instability; when a DC line experiences a blocking fault, the delayed disconnection of reactive power compensation equipment at the converter station will cause a brief voltage rise, followed by insufficient reactive power support within the grid, causing the voltage to gradually decrease and eventually leading to voltage instability. If control measures are not taken in time, voltage collapse may occur in severe cases. Therefore, there is an urgent need to develop flexible, rapid, accurate, and reliable emergency control technologies.

[0003] Emergency control refers to the control measures taken by a power system to maintain stable operation and continuous power supply under large disturbance faults. Common methods include generator tripping, load shedding, and DC modulation for safety and stability. In existing technologies, emergency control schemes mainly include pre-planned control based on strategy tables and response-based emergency control.

[0004] Pre-planned control based on strategy tables is currently the most mature, lowest-cost, and most difficult-to-apply emergency control system. Pre-planned control generates a strategy table based on a set of anticipated incidents and stores it in the safety and stability control device. After a fault occurs, the control device searches the strategy table based on the fault type, the current operating state of the power grid, and other conditions to determine and execute corresponding emergency control measures. Because the decision-making process precedes the fault execution, the safety and stability control device only needs to query the strategy table after a fault occurs, resulting in rapid decision-making. However, this method involves hypothetical faults and pre-decision-making, making it difficult to cover all fault types in AC / DC hybrid power grids and carrying the risk of decision-making errors. Fault-matching stability control measures cannot guarantee the safe operation of the power grid under complex, unforeseen fault conditions.

[0005] With the widespread application of Wide Area Measurement Systems (WAMS) based on millisecond-level synchronous measurements in power grids, response-based emergency control methods have developed rapidly. One effective approach is to estimate the subsequent behavior of dynamic components in the system and make emergency decisions based on real-time measurement data collected by WAMS, using direct methods, trajectory fitting, and other techniques. These methods generally do not require transient stability time-domain simulations and have low computational cost, but they commonly suffer from insufficient computational accuracy and conservative strategies. The development of artificial intelligence (AI) has provided a new approach to response-based emergency control. AI methods can often more quickly and effectively assess power grid stability and formulate emergency control strategies accordingly, offering higher prediction accuracy and computational efficiency compared to traditional methods. However, the currently used deep neural network models are relatively classic, and their prediction accuracy and precision need improvement. Furthermore, existing research mainly focuses on addressing power angle stability issues, with relatively few studies addressing emergency control for power grid voltage stability. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the present invention aims to provide a power grid voltage stability emergency control method and system based on CNN-LSTM combined network, so as to solve the technical problems of the inability to accurately evaluate the emergency control of power grid voltage stability and the low prediction accuracy in the existing technology.

[0007] This invention is achieved through the following technical solution:

[0008] An emergency control method for grid voltage stability based on a CNN-LSTM combined network includes the following steps:

[0009] Step 1: Construct a large disturbance voltage stability margin index, and establish a voltage stability margin evaluation model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM) using the large disturbance voltage stability margin index.

[0010] Step 2: Predict the large disturbance voltage stability margin of the key bus after the alternative generator tripping and load shedding point stabilization measures are implemented by using a voltage stability margin assessment model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM). Determine the sensitivity of the response-driven emergency control measures by calculating the increase in grid voltage stability margin.

[0011] Step 3: Based on the sensitivity and constraints of the response-driven emergency control measures, establish an emergency control optimization problem under the actual operating constraints of the power grid, and solve for the optimal emergency control strategy to complete the emergency control work for power grid voltage stability.

[0012] Preferably, in step 1, the formula for constructing the large disturbance voltage stability margin index is as follows:

[0013]

[0014] Among them, u N The voltage is the rated value; u(s) is the actual voltage value; t clr t is the fault clearing time. f This is the simulation end time; u cr For voltage binary meter (u cr , τ cr The voltage threshold in ) ; τ cr For voltage binary meter (u cr , τ cr The duration in ) indicates that the voltage curve remains below u for an extended period. cr The duration should not exceed τ cr .

[0015] Preferably, in step 1, the large disturbance voltage stability margin of the corresponding scenario is calculated based on the large disturbance voltage stability margin index as a label. The CNN-LSTM network is trained with the key bus voltage time series value as input and the voltage stability margin as output. The network structure and parameters are adjusted by evaluating its regression prediction results, and finally a voltage stability margin evaluation model based on the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) with high prediction accuracy is generated.

[0016] Furthermore, when the CNN-LSTM network is used to evaluate the voltage stability margin of the power grid, the evaluation performance is reflected by the prediction accuracy metric, which is calculated as follows:

[0017]

[0018] Among them, A S To determine the prediction accuracy, N represents the number of critical bus nodes; y'(i) and y(i) represent the predicted and actual values ​​of the i-th bus, respectively.

[0019] Preferably, in step 1, the voltage stability margin evaluation model based on the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) includes an input layer, a CNN layer, an LSTM layer, and an output layer, wherein the output of the input layer is connected to the input of the CNN layer, the output of the CNN layer is connected to the input of the LSTM layer, and the output of the LSTM layer is connected to the input of the output layer.

[0020] Preferably, in step 2, when pre-selecting the operating bus, the voltage sequence of the key bus after the pre-selected stability control measures are obtained through simulation based on the actual operating characteristics. The voltage stability margin of the power grid after the stability control measures are taken is predicted using a voltage stability margin evaluation model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM) that has been trained. The sensitivity of control measures for different types and different action points is calculated and determined.

[0021] Preferably, when predicting the voltage stability margin of critical buses after the activation of alternative generator tripping and load shedding point stabilization measures, a voltage stability emergency control measure sensitivity index is constructed based on the stability margin to characterize the sensitivity of each bus voltage to generator tripping and load shedding control measures. The calculation formula for the control measure sensitivity index is as follows:

[0022]

[0023] Where l = 1, 2, ..., n, k = 1, 2, ..., m; m is the total number of grid voltage monitoring nodes, n is the total number of load shedding nodes, and ΔP l Let ξ be the load change at the l-th node. k ξ represents the voltage stability margin at node k after the stabilization measures are implemented. k,0 λ represents the voltage stability margin at node k when the post-fault stability control measures fail to operate. k,l This represents the change in the voltage stability margin of the k-th node caused by the removal of a unit load from the l-th node;

[0024] The impact of the same load measures on the voltage stability margin at different nodes varies, therefore the calculation formula is as follows:

[0025]

[0026] Where, k u,l This represents the impact of removing a unit load at node l on the grid voltage stability margin; λ k,l This represents the change in voltage stability margin at node k caused by the removal of a unit load at node l; m is the total number of voltage monitoring nodes in the power grid.

[0027] Preferably, in step 3, the emergency control optimization problem under the actual operating constraints of the power grid is a dynamic optimization problem, the objective of which is to minimize the total amount of generator tripping and load shedding, as shown in the following formula:

[0028]

[0029] Where, N G N represents the number of switchable generator nodes. L ΔP represents the number of load-shearing nodes. G,i Let ΔP be the number of generators that can be switched over at the i-th generator node; L,j Let u be the load that can be sheared at the j-th load node; i u j The control variables are the generator tripping amount and the load shedding amount, respectively; in engineering practice, generator tripping always involves taking the entire generator out of operation, therefore u i The value is a discrete integer; the load shedding amount can usually be proportionally cut off based on the degree of voltage drop caused by the fault, therefore u jIt is a continuous variable in (0, 1);

[0030] The cutting machine and cutting load should meet the following constraints:

[0031]

[0032] 0≤u j ≤1,j=1,2,...N L

[0033] Emergency control measures for generator and load shedding after a fault improve grid stability. Control strategies are adopted to ensure that the grid stability margin is greater than a set threshold ε. Therefore, the most basic constraints for stable grid operation are as follows:

[0034]

[0035] Where, ξ k,0 N represents the voltage stability margin at node k when the post-fault stability control measures fail to operate; G N represents the number of switchable generator nodes. L ΔP represents the number of load-shearing nodes. G,i Let ΔP be the number of generators that can be switched over at the i-th generator node; L,j Let u be the load that can be sheared at the j-th load node; i The control variable represents the number of machines cut off; u j λ is the control variable, representing the load shedding amount. k,i λ represents the change in voltage stability margin at the k-th node caused by the disconnection of a unit generator at the i-th generator node; k,j This represents the change in voltage stability margin at node k caused by the removal of a unit load at node j; ε is the threshold value for voltage stability margin.

[0036] Consider the following power balance constraints:

[0037]

[0038] Among them, S bmin The lower limit of unbalanced power when considering the frequency regulation capability of generator units in the power grid; S bmax The upper limit of unbalanced power when considering the frequency regulation capability of generator units in the power grid; N G N represents the number of switchable generator nodes. L ΔP represents the number of load-shearing nodes. G,i Let ΔP be the number of generators that can be switched over at the i-th generator node; L,j Let u be the load that can be sheared at the j-th load node; i The control variable represents the number of machines cut off; u j The control variable represents the load shedding amount.

[0039] Preferably, in step 3, the optimal emergency control strategy is obtained by using CPLEX.

[0040] An emergency control system for grid voltage stability based on a CNN-LSTM combined network, including

[0041] The model building module is used to construct a large disturbance voltage stability margin index, and to build a voltage stability margin evaluation model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM) through the large disturbance voltage stability margin index.

[0042] The first data processing module is used to predict the large disturbance voltage stability margin of the key bus after the alternative generator trip and load shedding point stabilization measures are taken through a voltage stability margin assessment model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM), and to determine the sensitivity of response-driven emergency control measures by calculating the increase in grid voltage stability margin.

[0043] The second data processing module is used to establish an emergency control optimization problem based on the sensitivity and constraints of response-driven emergency control measures, and to solve for the optimal emergency control strategy to complete the emergency control work for grid voltage stability.

[0044] Compared with the prior art, the present invention has the following beneficial technical effects:

[0045] This invention provides an emergency control method for power grid voltage stability based on a CNN-LSTM combined network. It integrates the advantages of Convolutional Neural Networks (CNN) in feature extraction and Long Short-Term Memory Networks (LSTM) in learning data temporal dependencies, establishing a voltage stability margin assessment model based on the CNN-LSTM combination. This model achieves accurate assessment of the power grid voltage stability margin under large disturbance faults, with higher prediction accuracy and results closer to the true values ​​compared to classical neural networks. It can accurately characterize the sensitivity of each bus voltage to generator tripping and load shedding control measures, providing strong evidence and guidance for formulating optimal emergency control strategies. This effectively and accurately assesses the emergency control of power grid voltage stability, improves prediction accuracy, and ensures the safe and stable operation of the power grid after large disturbance faults. Attached Figure Description

[0046] Figure 1 This is a flowchart of the emergency control method for grid voltage stability based on CNN-LSTM combined network in this invention;

[0047] Figure 2 This is a schematic diagram of the network structure based on CNN-LSTM combination in this invention;

[0048] Figure 3This is a schematic diagram of the AC / DC hybrid power grid structure in a local area of ​​Northwest China, as described in this embodiment of the invention.

[0049] Figure 4 This is a schematic diagram of the voltage response of the 750kV busbar in Qinghai under the fault scenario of Hewu N-2 in an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of the regression prediction results of the power grid voltage stability margin in an embodiment of the present invention;

[0051] Figure 6 This is a schematic diagram of the voltage response of the 750kV bus in Qinghai after the control measures are activated in an embodiment of the present invention. Detailed Implementation

[0052] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0053] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0054] The present invention will now be described in further detail with reference to the accompanying drawings:

[0055] The purpose of this invention is to provide a method and system for emergency control of grid voltage stability based on a CNN-LSTM combined network, so as to solve the technical problems of the inability to accurately evaluate and predict the emergency control of grid voltage stability in the prior art.

[0056] Specifically, according to Figure 1 As shown, the emergency control method for grid voltage stability based on a CNN-LSTM combined network includes the following steps:

[0057] Step 1: Construct a large disturbance voltage stability margin index, and establish a voltage stability margin evaluation model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM) using the large disturbance voltage stability margin index.

[0058] Specifically, the network structure and parameters are determined by evaluating the regression prediction results, fully exploring the mapping relationship between the voltage of key bus nodes and the stability margin of large disturbance voltages. Power systems generally describe their voltage stability using voltage acceptability. Voltage acceptability is often expressed using a voltage binary table (u... cr , τ cr The form is given, that is, the voltage curve is consistently below u. cr The duration should not exceed τ cr The practical criteria for large disturbance voltage stability in power systems require that, during a dynamic process, the bus voltage at the system's central point drops below 0.75 pu for no more than 1 second, and the bus voltage does not fall below 0.9 pu after the dynamic process subsides.

[0059] The formula for constructing the large disturbance voltage stability margin index is as follows:

[0060]

[0061] Among them, u N The voltage is the rated value; u(s) is the actual voltage value; t clr t is the fault clearing time. f This is the simulation end time; u cr For voltage binary meter (u cr , τ cr The voltage threshold in ) ; τ cr For voltage binary meter (u cr , τ cr The duration in ) indicates that the voltage curve remains below u for an extended period. cr The duration should not exceed τ cr .

[0062] u cr and τ cr Take 0.75 pu and 1 s respectively. In (t clr , t f Within a time interval, using a width of τ cr The voltage curve is scanned within a time window, and the voltage stability margin within each scanning window is calculated throughout the simulation. Large-scale power grids typically select several nodes with the weakest voltage for monitoring. Regarding large disturbance voltage stability, the minimum value of ξ among the observed nodes is taken as the voltage stability margin of the power grid.

[0063] Specifically, the voltage stability margin of the corresponding scenario is calculated based on the voltage stability margin index of the large disturbance as a label. The CNN-LSTM network is trained with the time series value of the key bus voltage as input and the voltage stability margin as output. The network structure and parameters are adjusted by evaluating its regression prediction results, and finally a voltage stability margin evaluation model with high prediction accuracy based on the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) is generated.

[0064] Specifically, when a CNN-LSTM network is used to evaluate the voltage stability margin of a power grid, the prediction accuracy metric reflects the model's evaluation performance. The formula for this metric is as follows:

[0065]

[0066] Among them, A S The prediction accuracy is represented by N, where N is the number of critical bus nodes; y'(i) and y(i) are the predicted and actual values ​​of the i-th bus, respectively. This indicator reflects the overall effectiveness of the power grid stability margin assessment; a higher prediction accuracy indicates a better model prediction performance.

[0067] Specifically, CNN is a deep neural network widely used in deep learning, possessing excellent feature extraction capabilities. LSTM is an optimized structure of recurrent neural networks, adept at processing time series data. By combining the advantages of CNN in feature extraction and LSTM in learning temporal dependencies in data, a CNN-LSTM combined voltage stability margin evaluation model can be built, such as... Figure 2 As shown, the voltage stability margin evaluation model based on the combination of Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) includes an input layer, a CNN layer, an LSTM layer, and an output layer. The output of the input layer is connected to the input of the CNN layer, the output of the CNN layer is connected to the input of the LSTM layer, and the output of the LSTM layer is connected to the input of the output layer.

[0068] The input layer is used to evaluate the voltage stability margin of the power grid after a large disturbance fault occurs, and the voltage timing value of the key bus after the fault is cleared is selected as the input.

[0069] The role of the CNN layer is to extract the complex dynamic features of the input key bus voltage data. The fixed connection method of CNN dictates that the model should not have too many feature extraction layers. Typically, 1-3 feature extraction layers are used for prediction. This model is constructed using a CNN framework consisting of two convolutional layers and two pooling layers.

[0070] The LSTM layer takes the feature vectors extracted by the CNN layer as input, learns the internal variation patterns, and captures the temporal relationships between data. Choosing two LSTM layers allows for the full learning of these temporal relationships from the feature information, while also reducing model training time.

[0071] The output layer takes the implicit state output by the last time step of the LSTM as input and calculates the large disturbance voltage stability margin assessment result of the critical bus through the fully connected layer.

[0072] Step 2: Predict the large disturbance voltage stability margin of the key bus after the alternative generator tripping and load shedding point stabilization measures are implemented by using a voltage stability margin assessment model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM). Determine the sensitivity of the response-driven emergency control measures by calculating the increase in grid voltage stability margin.

[0073] Specifically, to measure the impact of stabilization measures such as generator tripping and load shedding on grid voltage stability, a sensitivity index for emergency voltage stability control measures is constructed based on the stability margin. This index characterizes the sensitivity of each bus voltage to the magnitude of generator tripping and load shedding control measures. The formula for calculating the sensitivity index is as follows:

[0074]

[0075] In the formula, l = 1, 2, ..., n, k = 1, 2, ..., m. Taking load shedding control measures as an example, m is the total number of grid voltage monitoring nodes, n is the total number of load shedding nodes, and ΔP l Let ξ be the load change at the l-th node. k ξ represents the voltage stability margin at node k after the stabilization measures are implemented. k,0 λ represents the voltage stability margin at node k when the post-fault stability control measures fail to operate. k,l This represents the change in voltage stability margin at node k caused by the removal of a unit load at node l. Since the same load shedding measure has different effects on voltage stability margins at different nodes, the following formula is defined to reduce computational complexity:

[0076]

[0077] Where, k u,l This represents the impact of removing a unit load at node l on the grid voltage stability margin; λ k,l This represents the change in voltage stability margin at node k caused by the removal of a unit load at node l; m is the total number of voltage monitoring nodes in the power grid.

[0078] When pre-selecting the operating bus, the voltage sequence of the key bus after the pre-selected stabilization measures are obtained through simulation based on the actual operating characteristics. The voltage stability margin of the power grid after the stabilization measures are predicted by the trained CNN-LSTM combined model. Then, the sensitivity of control measures of different types and different operating points is determined according to Equations (3) and (4).

[0079] Step 3: Based on the sensitivity and constraints of the response-driven emergency control measures, establish an emergency control optimization problem under the actual operating constraints of the power grid, and solve for the optimal emergency control strategy to complete the emergency control work for power grid voltage stability.

[0080] Voltage stability emergency control is a dynamic optimization problem. If only generator tripping or load shedding is performed after a fault, it may cause grid power imbalance, further deteriorating grid stability. Therefore, this invention considers both generator tripping and load shedding strategies. The objective of the control measure optimization problem is to minimize the total amount of generator tripping and load shedding, as shown in the following equation:

[0081]

[0082] In the formula, N G N represents the number of switchable generator nodes. L ΔP represents the number of load-shearing nodes. G,i Let ΔP be the number of generators that can be switched over at the i-th generator node; L,j Let u be the load that can be sheared at the j-th load node; i u j The control variables are the generator tripping amount and the load shedding amount, respectively. Since generator tripping in engineering practice involves taking the entire generator out of operation, u... i The value is a discrete integer; the load shedding amount can usually be proportionally cut off based on the degree of voltage drop caused by the fault, therefore u j For continuous variables within (0, 1). In summary, the machine cutoff and load shedding amounts should satisfy the following constraints:

[0083]

[0084] Considering the actual operation of the power grid, the constraints for optimizing control measures are determined. Emergency control measures such as generator tripping and load shedding after a fault can improve the stability of the power grid. Adopting appropriate control strategies to ensure that the grid stability margin is greater than the set threshold ε is the most basic constraint for guaranteeing the stable operation of the power grid.

[0085]

[0086] Where, ξ k,0 N represents the voltage stability margin at node k when the post-fault stability control measures fail to operate; G N represents the number of switchable generator nodes. LΔP represents the number of load-shearing nodes. G,i Let ΔP be the number of generators that can be switched over at the i-th generator node; L,j Let u be the load that can be sheared at the j-th load node; i The control variable represents the number of machines cut off; u j λ is the control variable, representing the load shedding amount. k,i λ represents the change in voltage stability margin at the k-th node caused by the disconnection of a unit generator at the i-th generator node; k,j ε represents the change in voltage stability margin of the k-th node caused by the removal of a unit load from the j-th load node; ε is the threshold value for voltage stability margin.

[0087] Based on the practical criteria for large disturbance voltage stability, in order to ensure that the power grid can return to a stable operating state after the stabilization control measures are implemented, u(s) is taken as 0.9pu and the duration is 1s. Substituting into the voltage stability margin index calculation formula, the ε in this stability constraint is found to be 0.6.

[0088] In addition, consider the following power balance constraints:

[0089]

[0090] Among them, S bmin The lower limit of unbalanced power when considering the frequency regulation capability of generator units in the power grid; S bmax The upper limit of unbalanced power when considering the frequency regulation capability of generator units in the power grid; N G N represents the number of switchable generator nodes. L ΔP represents the number of load-shearing nodes. G,i Let ΔP be the number of generators that can be switched over at the i-th generator node; L,j Let u be the load that can be sheared at the j-th load node; i The control variable represents the number of machines cut off; u j The control variable represents the load shedding amount.

[0091] Specifically, the optimal emergency control strategy is obtained by using CPLEX. CPLEX is a high-performance mathematical programming problem solver that can quickly and stably solve large-scale linear programming, quadratic programming, and mixed-integer programming problems. The emergency control measure optimization model consisting of the objective function (5) and the constraints (6)-(8) is a mixed-integer linear programming problem, which can be solved directly using the CPLEX solver. Finally, the emergency control strategy that satisfies all constraints and minimizes the total amount of machine cut-off and load shedding is obtained.

[0092] Example

[0093] A schematic diagram of the actual AC / DC hybrid power grid structure in some parts of Northwest China is shown below. Figure 3As shown, the system baseline capacity is 100MW. When an N-2 fault occurs on the Hexi-Wusheng 750kV line of this power grid (any two transmission lines fail and are disconnected), the voltage response curve of the 750kV bus in Qinghai region is as follows. Figure 4 As shown, the results indicate that the power grid suffers from voltage instability after the fault. Therefore, this system is used to verify the accuracy and effectiveness of the proposed method.

[0094] First, an offline large-disturbance voltage stability sample set is generated and input into a CNN-LSTM combined network for training to determine the network structure and parameters. Finally, a grid voltage stability margin assessment model with high prediction accuracy is generated to fully explore the mapping relationship between the voltage of key bus nodes and the large-disturbance voltage stability margin.

[0095] Taking into account factors such as different load levels, renewable energy penetration rates, fault locations, and fault durations, the time-series sampling values ​​of the voltages of nine key 750kV busbars in the region after fault clearance were selected as inputs, and the voltage stability margin index was selected as the output. The simulation software PSASP was used to generate a sample set for large-disturbance voltage stability simulation of the power grid. The output of traditional power plants and renewable energy stations was adjusted at load levels of 95%, 100%, and 105%. A grid N-2 fault was set, with the Hexi-Wusheng 750kV line and the Xining-Riyueshan 750kV line selected. The fault locations ranged from 2% to 92% of the AC lines, with 20 fault points added in 10% increments. The fault clearance time ranged from 5 cycles (0.1s) to 10 cycles (0.2s) after the fault occurred, with 5 durations added in 1-cycle (0.02s) increments. The simulation time was 10s, the system rated frequency was 50Hz, and the total sample size was 5760.

[0096] Eighty percent of the data in the sample set was selected as the training set, and the remaining 20% ​​was used as the test set to train a CNN-LSTM combined network to evaluate the voltage stability margin of the power grid. During training, the determination of the number and size of convolutional kernels in the network parameters is crucial. From the perspective of model structure and computational complexity, using different kernel sizes in different convolutional layers significantly increases computational complexity; therefore, the kernels of both convolutional layers were chosen to have the same size. In classic CNN structures, the kernel size is typically between 3 and 5. Experiments showed that a size of 4 resulted in the best fit between the predicted and actual values; therefore, a kernel size of 4 was selected for both convolutional layers. Based on empirical knowledge, to better extract input features, the number of kernels used in each convolutional layer should be doubled layer by layer to increase network depth. Considering the high dimensionality of the input vector in this example, the final number of kernels for the two convolutional layers was determined to be 32 and 64, respectively, based on experiments.

[0097] The regression prediction results of the proposed CNN-LSTM combined network are compared with those of neural networks such as BP, LSTM, GRU, CNN, and AlexNet. Taking the 750kV busbar in Chaidamu, Qinghai Province as an example, the voltage stability margin results evaluated using different neural network models are as follows: Figure 5 As shown in Table 1, the proposed evaluation model has relatively good regression prediction performance, and the predicted values ​​are closer to the true values. The average prediction accuracy of different neural networks was calculated using evaluation metrics, and the results are shown in Table 1.

[0098] Table 1. Average prediction accuracy of different regression models

[0099]

[0100] As shown in Table 1, the regression prediction accuracy of the proposed model is 99.3912%, which is higher than that of other neural network models. This further proves that the regression prediction effect of the CNN-LSTM combined network is relatively good and has higher reference value in the process of formulating emergency control strategies.

[0101] Secondly, the trained CNN-LSTM combined network is used to predict the voltage stability margin of key buses after the implementation of stabilization and control measures such as alternative generator tripping and load shedding. The increase in grid voltage stability margin is calculated to determine the sensitivity of response-driven emergency control measures.

[0102] Considering specific generator demands and load importance, a selection of generators and load nodes were chosen as emergency control action points. The alternative generator tripping points include the Dunhuang, Mogao, Jiuquan, Hexi, and Wusheng busbars, corresponding to tripping points 1, 2, 3, 4, and 5, respectively. The number of generators that can be cut off and their capacity at each tripping point are shown in Table 2. The alternative load shedding points include the Guanting, Xining, Guolong, Riyueshan, and Haixi busbars, corresponding to load shedding points 1, 2, 3, 4, and 5, respectively. The available shedding capacity at each load shedding point is shown in Table 3.

[0103] Table 2. Alternative Cutting Machine Control Points and Cuttable Capacity

[0104]

[0105] Table 3. Alternative load shearing control points and removable capacity

[0106]

[0107] Stabilization measures are implemented 0.3s after fault clearing. Simulation data of the preset stabilization measures' activation time is obtained through simulation. The CNN-LSTM combined network is used to predict the grid voltage stability margin after the shedding of generators and loads. The increase in grid voltage stability margin k is calculated using formulas (3) and (4). u,lThe action amount for both the generator tripping and load shedding points is set to 500MW. The improvement in grid voltage stability margin is shown in Table 4. Different stabilization measures have different effects on improving the bus voltage stability margin. The sensitivity ranking results are as follows: generator tripping point 3 > load shedding point 4 > load shedding point 2 > load shedding point 3 > generator tripping point 1 > load shedding point 5 > generator tripping point 2 > load shedding point 1 > generator tripping point 4 > generator tripping point 5.

[0108] Table 4. Alternative load control points and removable capacity

[0109]

[0110] Finally, based on the sensitivity of the obtained emergency control measures and related constraints, the optimal emergency control strategy is obtained.

[0111] Simulation tests showed that the unit regulation power coefficient of the system was approximately 9572 MW / Hz. The "Guidelines for Calculation and Analysis of Power Grid Safety and Stability" stipulates that power grid frequency fluctuations should not exceed ±0.1 Hz. Therefore, the upper and lower limits of active power imbalance are +957.2 MW and -957.2 MW, respectively. The S in the per-unit power balance constraint (8) b max and S bmin Take +9.572 and -9.572 respectively.

[0112] The optimization problem was solved using the CPLEX solver, and the optimal emergency control strategy for the power grid was obtained. The results are shown in Table 5.

[0113] Table 5 Optimal Emergency Control Strategies

[0114]

[0115] This strategy is implemented 0.3 seconds after the fault is cleared. The voltage response of the 750kV bus in Qinghai region is as follows: Figure 6 As shown, the results indicate that after the stabilization measures were implemented, the voltage of each key bus reached above 0.9 pu, and the grid voltage returned to stability, verifying the effectiveness of the optimal emergency control strategy.

[0116] The present invention also provides an emergency control system for grid voltage stability based on a CNN-LSTM combined network, which is used to implement an emergency control method for grid voltage stability based on a CNN-LSTM combined network, including a model building module, a first data processing module and a second data processing module;

[0117] The model building module is used to construct a large disturbance voltage stability margin index, and to build a voltage stability margin evaluation model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM) through the large disturbance voltage stability margin index.

[0118] The first data processing module is used to predict the large disturbance voltage stability margin of the key bus after the alternative generator trip and load shedding point stabilization measures are taken through a voltage stability margin assessment model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM), and to determine the sensitivity of response-driven emergency control measures by calculating the increase in grid voltage stability margin.

[0119] The second data processing module is used to establish an emergency control optimization problem based on the sensitivity and constraints of response-driven emergency control measures, and to solve for the optimal emergency control strategy to complete the emergency control work for grid voltage stability.

[0120] In summary, this invention provides a power grid voltage stability emergency control method and system based on a CNN-LSTM combined network. It integrates the advantages of Convolutional Neural Networks (CNN) in feature extraction and Long Short-Term Memory Networks (LSTM) in learning data temporal dependencies, establishing a voltage stability margin assessment model based on the CNN-LSTM combination. This model achieves accurate assessment of the power grid voltage stability margin under large disturbance faults, with higher prediction accuracy and results closer to the true values ​​compared to classical neural networks. Based on the power grid voltage stability margin assessment model, a sensitivity prediction method for voltage stability emergency control measures is proposed. This method accurately characterizes the sensitivity of each bus voltage to generator and load shedding control measures, providing strong evidence and guidance for formulating the optimal emergency control strategy. Considering the sensitivity of emergency control measures and the actual operating constraints of the power grid, a modeling and solution method for the power grid voltage stability emergency control problem is proposed. The optimal emergency control strategy obtained by the proposed method can ensure the safe and stable operation of the power grid after a large disturbance fault.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for emergency control of power grid voltage stability based on a CNN-LSTM combined network, characterized in that, Includes the following steps: Step 1: Construct a large disturbance voltage stability margin index, and establish a voltage stability margin evaluation model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM) using the large disturbance voltage stability margin index. Step 2: Predict the large disturbance voltage stability margin of the key bus after the alternative generator tripping and load shedding point stabilization measures are implemented by using a voltage stability margin assessment model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM). Determine the sensitivity of the response-driven emergency control measures by calculating the increase in grid voltage stability margin. Step 3: Based on the sensitivity and constraints of the response-driven emergency control measures, establish an emergency control optimization problem under the actual operating constraints of the power grid, and solve for the optimal emergency control strategy to complete the emergency control work for power grid voltage stability. Among them, the emergency control optimization problem under the actual operation constraints of the power grid is a dynamic optimization problem, the objective of which is to minimize the total amount of generator tripping and load shedding, and the specific formula is as follows: in, N G This represents the number of switchable generator nodes; N L This represents the number of load nodes that can be shelved. P G,i For the first i The number of generator nodes that can be switched out; P L,j For the first j The load that can be cut off at each load node; u i The control variable represents the number of machines cut off; u j The variable represents the load shedding amount; in engineering practice, the entire generator is taken out of operation when the generator is shedding, therefore... u i The values ​​are discrete integers; the load shedding amount can be proportionally adjusted based on the degree of voltage drop caused by the fault, therefore... u j It is a continuous variable in (0, 1); The cutting machine and cutting load should meet the following constraints: Emergency control measures for generator and load shedding after a fault improve grid stability by employing control strategies to ensure that the grid stability margin exceeds a set threshold. Therefore, the constraints for stable operation of the power grid are as follows: in, ξ k,0 When the stability control measures fail to activate after a fault. k Voltage stability margin; λ k,i Indicates the first i The first generator node disconnection caused by the unit generator set k The change in the voltage stability margin of each node; λ k,j Indicates the first j The first load node shedding a unit load resulted in the k The change in voltage stability margin at each node; This is the threshold value for voltage stability margin; Consider the following power balance constraints: in, S bmin The lower limit of unbalanced power when considering the frequency regulation capability of generator units in the power grid; S bmax The upper limit of unbalanced power when considering the frequency regulation capability of generator sets in the power grid.

2. The emergency control method for power grid voltage stability based on a CNN-LSTM combined network according to claim 1, characterized in that, In step 1, the formula for constructing the large disturbance voltage stability margin index is as follows: in, u N This is the rated voltage. u ( s () represents the actual voltage value; t clr This refers to the fault clearance time. t f This is the simulation end time; u cr For voltage binary meter ( u cr , τ cr The voltage threshold in ); τ cr For voltage binary meter ( u cr , τ cr The duration in () indicates that the voltage curve remains below () u cr The duration should not exceed τ cr .

3. The emergency control method for grid voltage stability based on a CNN-LSTM combined network according to claim 1, characterized in that, In step 1, the voltage stability margin of the corresponding scenario is calculated based on the voltage stability margin index of the large disturbance as a label. The CNN-LSTM network is trained with the key bus voltage time series value as input and the voltage stability margin as output. The network structure and parameters are adjusted by evaluating its regression prediction results, and finally a voltage stability margin evaluation model based on the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) with high prediction accuracy is generated.

4. The emergency control method for power grid voltage stability based on a CNN-LSTM combined network according to claim 3, characterized in that, When a CNN-LSTM network is used to evaluate the voltage stability margin of a power grid, the evaluation performance is reflected by the prediction accuracy metric, which is calculated as follows: in, A S To improve prediction accuracy, N Number of critical busbar nodes; y '( i )and y ( i ) are respectively the first i Predicted and actual values ​​of each busbar.

5. The emergency control method for grid voltage stability based on a CNN-LSTM combined network according to claim 1, characterized in that, In step 1, the voltage stability margin evaluation model based on the combination of Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM) includes an input layer, a CNN layer, an LSTM layer, and an output layer. The output of the input layer is connected to the input of the CNN layer, the output of the CNN layer is connected to the input of the LSTM layer, and the output of the LSTM layer is connected to the input of the output layer.

6. The emergency control method for grid voltage stability based on a CNN-LSTM combined network according to claim 1, characterized in that, In step 2, when pre-selecting the operating bus, the voltage sequence of the key bus after the pre-selected stability control measures are obtained through simulation based on the actual operating characteristics. The voltage stability margin of the power grid after the stability control measures are predicted by using the voltage stability margin evaluation model based on the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) obtained through training, and the sensitivity of control measures of different types and different operating points is calculated and determined.

7. The emergency control method for grid voltage stability based on a CNN-LSTM combined network according to claim 1, characterized in that, When predicting the voltage stability margin of critical buses after the activation of alternative generator tripping and load shedding point stabilization measures, a voltage stability emergency control measure sensitivity index is constructed based on the stability margin. This index characterizes the sensitivity of each bus voltage to generator tripping and load shedding control measures. The calculation formula for the control measure sensitivity index is as follows: in, l= 1,2,…, n , k= 1,2,…, m ; m This represents the total number of voltage monitoring nodes in the power grid. n This represents the total number of load shedding nodes. P l For the first l The load change at each node, ξ k As a node after the implementation of stability control measures k Voltage stability margin, ξ k,0 When the stability control measures fail to activate after a fault. k Voltage stability margin; The impact of the same load measures on the voltage stability margin at different nodes varies, therefore the calculation formula is as follows: ； in, k u,l Indicates in l The impact of node load shedding on grid voltage stability margin; λ k,l Indicates the first l The first node to remove a unit load, resulting in the k The change in the voltage stability margin of each node.

8. The emergency control method for power grid voltage stability based on a CNN-LSTM combined network according to claim 1, characterized in that, In step 3, the optimal emergency control strategy is obtained by using CPLEX.

9. An emergency control system for grid voltage stability based on a CNN-LSTM combined network, characterized in that, An emergency control method for grid voltage stability based on a CNN-LSTM combined network, as described in any one of claims 1-8, includes: The model building module is used to construct a large disturbance voltage stability margin index, and to build a voltage stability margin evaluation model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM) through the large disturbance voltage stability margin index. The first data processing module is used to predict the large disturbance voltage stability margin of the key bus after the alternative generator trip and load shedding point stabilization measures are taken through a voltage stability margin assessment model based on a combination of convolutional neural network (CNN) and long short-term memory network (LSTM), and to determine the sensitivity of response-driven emergency control measures by calculating the increase in grid voltage stability margin. The second data processing module is used to establish an emergency control optimization problem based on the sensitivity and constraints of response-driven emergency control measures, and to solve for the optimal emergency control strategy to complete the emergency control work for grid voltage stability.

Images